repo
stringlengths
1
152
file
stringlengths
14
221
code
stringlengths
501
25k
file_length
int64
501
25k
avg_line_length
float64
20
99.5
max_line_length
int64
21
134
extension_type
stringclasses
2 values
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
4,615
36.836066
124
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x160.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x160( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 160 * sizeof(float); batch -= 160 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); const __m512 vx2 = _mm512_loadu_ps(input + 32); const __m512 vx3 = _mm512_loadu_ps(input + 48); const __m512 vx4 = _mm512_loadu_ps(input + 64); const __m512 vx5 = _mm512_loadu_ps(input + 80); const __m512 vx6 = _mm512_loadu_ps(input + 96); const __m512 vx7 = _mm512_loadu_ps(input + 112); const __m512 vx8 = _mm512_loadu_ps(input + 128); const __m512 vx9 = _mm512_loadu_ps(input + 144); input += 160; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); const __m512 vz2 = _mm512_range_ps(vsat_cutoff, vx2, 0xA); const __m512 vz3 = _mm512_range_ps(vsat_cutoff, vx3, 0xA); const __m512 vz4 = _mm512_range_ps(vsat_cutoff, vx4, 0xA); const __m512 vz5 = _mm512_range_ps(vsat_cutoff, vx5, 0xA); const __m512 vz6 = _mm512_range_ps(vsat_cutoff, vx6, 0xA); const __m512 vz7 = _mm512_range_ps(vsat_cutoff, vx7, 0xA); const __m512 vz8 = _mm512_range_ps(vsat_cutoff, vx8, 0xA); const __m512 vz9 = _mm512_range_ps(vsat_cutoff, vx9, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); __m512 vn2 = _mm512_fmadd_ps(vz2, vminus_log2e, vmagic_bias); __m512 vn3 = _mm512_fmadd_ps(vz3, vminus_log2e, vmagic_bias); __m512 vn4 = _mm512_fmadd_ps(vz4, vminus_log2e, vmagic_bias); __m512 vn5 = _mm512_fmadd_ps(vz5, vminus_log2e, vmagic_bias); __m512 vn6 = _mm512_fmadd_ps(vz6, vminus_log2e, vmagic_bias); __m512 vn7 = _mm512_fmadd_ps(vz7, vminus_log2e, vmagic_bias); __m512 vn8 = _mm512_fmadd_ps(vz8, vminus_log2e, vmagic_bias); __m512 vn9 = _mm512_fmadd_ps(vz9, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vs2 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn2), 23)); vn2 = _mm512_sub_ps(vn2, vmagic_bias); const __m512 vs3 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn3), 23)); vn3 = _mm512_sub_ps(vn3, vmagic_bias); const __m512 vs4 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn4), 23)); vn4 = _mm512_sub_ps(vn4, vmagic_bias); const __m512 vs5 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn5), 23)); vn5 = _mm512_sub_ps(vn5, vmagic_bias); const __m512 vs6 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn6), 23)); vn6 = _mm512_sub_ps(vn6, vmagic_bias); const __m512 vs7 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn7), 23)); vn7 = _mm512_sub_ps(vn7, vmagic_bias); const __m512 vs8 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn8), 23)); vn8 = _mm512_sub_ps(vn8, vmagic_bias); const __m512 vs9 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn9), 23)); vn9 = _mm512_sub_ps(vn9, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); const __m512 vt2 = _mm512_fmadd_ps(vn2, vln2, vz2); const __m512 vt3 = _mm512_fmadd_ps(vn3, vln2, vz3); const __m512 vt4 = _mm512_fmadd_ps(vn4, vln2, vz4); const __m512 vt5 = _mm512_fmadd_ps(vn5, vln2, vz5); const __m512 vt6 = _mm512_fmadd_ps(vn6, vln2, vz6); const __m512 vt7 = _mm512_fmadd_ps(vn7, vln2, vz7); const __m512 vt8 = _mm512_fmadd_ps(vn8, vln2, vz8); const __m512 vt9 = _mm512_fmadd_ps(vn9, vln2, vz9); __m512 vp0 = vc6; __m512 vp1 = vc6; __m512 vp2 = vc6; __m512 vp3 = vc6; __m512 vp4 = vc6; __m512 vp5 = vc6; __m512 vp6 = vc6; __m512 vp7 = vc6; __m512 vp8 = vc6; __m512 vp9 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp2 = _mm512_fmadd_ps(vp2, vt2, vc5); vp3 = _mm512_fmadd_ps(vp3, vt3, vc5); vp4 = _mm512_fmadd_ps(vp4, vt4, vc5); vp5 = _mm512_fmadd_ps(vp5, vt5, vc5); vp6 = _mm512_fmadd_ps(vp6, vt6, vc5); vp7 = _mm512_fmadd_ps(vp7, vt7, vc5); vp8 = _mm512_fmadd_ps(vp8, vt8, vc5); vp9 = _mm512_fmadd_ps(vp9, vt9, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp2 = _mm512_fmadd_ps(vp2, vt2, vc4); vp3 = _mm512_fmadd_ps(vp3, vt3, vc4); vp4 = _mm512_fmadd_ps(vp4, vt4, vc4); vp5 = _mm512_fmadd_ps(vp5, vt5, vc4); vp6 = _mm512_fmadd_ps(vp6, vt6, vc4); vp7 = _mm512_fmadd_ps(vp7, vt7, vc4); vp8 = _mm512_fmadd_ps(vp8, vt8, vc4); vp9 = _mm512_fmadd_ps(vp9, vt9, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp2 = _mm512_fmadd_ps(vp2, vt2, vc3); vp3 = _mm512_fmadd_ps(vp3, vt3, vc3); vp4 = _mm512_fmadd_ps(vp4, vt4, vc3); vp5 = _mm512_fmadd_ps(vp5, vt5, vc3); vp6 = _mm512_fmadd_ps(vp6, vt6, vc3); vp7 = _mm512_fmadd_ps(vp7, vt7, vc3); vp8 = _mm512_fmadd_ps(vp8, vt8, vc3); vp9 = _mm512_fmadd_ps(vp9, vt9, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp2 = _mm512_fmadd_ps(vp2, vt2, vc2); vp3 = _mm512_fmadd_ps(vp3, vt3, vc2); vp4 = _mm512_fmadd_ps(vp4, vt4, vc2); vp5 = _mm512_fmadd_ps(vp5, vt5, vc2); vp6 = _mm512_fmadd_ps(vp6, vt6, vc2); vp7 = _mm512_fmadd_ps(vp7, vt7, vc2); vp8 = _mm512_fmadd_ps(vp8, vt8, vc2); vp9 = _mm512_fmadd_ps(vp9, vt9, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); vp2 = _mm512_fmadd_ps(vp2, vt2, vminus_two); vp3 = _mm512_fmadd_ps(vp3, vt3, vminus_two); vp4 = _mm512_fmadd_ps(vp4, vt4, vminus_two); vp5 = _mm512_fmadd_ps(vp5, vt5, vminus_two); vp6 = _mm512_fmadd_ps(vp6, vt6, vminus_two); vp7 = _mm512_fmadd_ps(vp7, vt7, vminus_two); vp8 = _mm512_fmadd_ps(vp8, vt8, vminus_two); vp9 = _mm512_fmadd_ps(vp9, vt9, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vts2 = _mm512_mul_ps(vt2, vs2); const __m512 vsmo2 = _mm512_sub_ps(vs2, vone); const __m512 vts3 = _mm512_mul_ps(vt3, vs3); const __m512 vsmo3 = _mm512_sub_ps(vs3, vone); const __m512 vts4 = _mm512_mul_ps(vt4, vs4); const __m512 vsmo4 = _mm512_sub_ps(vs4, vone); const __m512 vts5 = _mm512_mul_ps(vt5, vs5); const __m512 vsmo5 = _mm512_sub_ps(vs5, vone); const __m512 vts6 = _mm512_mul_ps(vt6, vs6); const __m512 vsmo6 = _mm512_sub_ps(vs6, vone); const __m512 vts7 = _mm512_mul_ps(vt7, vs7); const __m512 vsmo7 = _mm512_sub_ps(vs7, vone); const __m512 vts8 = _mm512_mul_ps(vt8, vs8); const __m512 vsmo8 = _mm512_sub_ps(vs8, vone); const __m512 vts9 = _mm512_mul_ps(vt9, vs9); const __m512 vsmo9 = _mm512_sub_ps(vs9, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vemo2 = _mm512_fmadd_ps(vp2, vts2, vsmo2); const __m512 vemo3 = _mm512_fmadd_ps(vp3, vts3, vsmo3); const __m512 vemo4 = _mm512_fmadd_ps(vp4, vts4, vsmo4); const __m512 vemo5 = _mm512_fmadd_ps(vp5, vts5, vsmo5); const __m512 vemo6 = _mm512_fmadd_ps(vp6, vts6, vsmo6); const __m512 vemo7 = _mm512_fmadd_ps(vp7, vts7, vsmo7); const __m512 vemo8 = _mm512_fmadd_ps(vp8, vts8, vsmo8); const __m512 vemo9 = _mm512_fmadd_ps(vp9, vts9, vsmo9); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); const __m512 vepo2 = _mm512_sub_ps(vemo2, vminus_two); const __m512 vepo3 = _mm512_sub_ps(vemo3, vminus_two); const __m512 vepo4 = _mm512_sub_ps(vemo4, vminus_two); const __m512 vepo5 = _mm512_sub_ps(vemo5, vminus_two); const __m512 vepo6 = _mm512_sub_ps(vemo6, vminus_two); const __m512 vepo7 = _mm512_sub_ps(vemo7, vminus_two); const __m512 vepo8 = _mm512_sub_ps(vemo8, vminus_two); const __m512 vepo9 = _mm512_sub_ps(vemo9, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); __m512 vrepo2 = _mm512_rcp14_ps(vepo2); __m512 vrepo3 = _mm512_rcp14_ps(vepo3); __m512 vrepo4 = _mm512_rcp14_ps(vepo4); __m512 vrepo5 = _mm512_rcp14_ps(vepo5); __m512 vrepo6 = _mm512_rcp14_ps(vepo6); __m512 vrepo7 = _mm512_rcp14_ps(vepo7); __m512 vrepo8 = _mm512_rcp14_ps(vepo8); __m512 vrepo9 = _mm512_rcp14_ps(vepo9); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); const __m512 verepo2 = _mm512_fnmadd_ps(vrepo2, vepo2, vone); const __m512 verepo3 = _mm512_fnmadd_ps(vrepo3, vepo3, vone); const __m512 verepo4 = _mm512_fnmadd_ps(vrepo4, vepo4, vone); const __m512 verepo5 = _mm512_fnmadd_ps(vrepo5, vepo5, vone); const __m512 verepo6 = _mm512_fnmadd_ps(vrepo6, vepo6, vone); const __m512 verepo7 = _mm512_fnmadd_ps(vrepo7, vepo7, vone); const __m512 verepo8 = _mm512_fnmadd_ps(vrepo8, vepo8, vone); const __m512 verepo9 = _mm512_fnmadd_ps(vrepo9, vepo9, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm512_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm512_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm512_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm512_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm512_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm512_fmadd_ps(verepo7, vrepo7, vrepo7); vrepo8 = _mm512_fmadd_ps(verepo8, vrepo8, vrepo8); vrepo9 = _mm512_fmadd_ps(verepo9, vrepo9, vrepo9); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); __m512 vy2 = _mm512_mul_ps(vemo2, vrepo2); __m512 vy3 = _mm512_mul_ps(vemo3, vrepo3); __m512 vy4 = _mm512_mul_ps(vemo4, vrepo4); __m512 vy5 = _mm512_mul_ps(vemo5, vrepo5); __m512 vy6 = _mm512_mul_ps(vemo6, vrepo6); __m512 vy7 = _mm512_mul_ps(vemo7, vrepo7); __m512 vy8 = _mm512_mul_ps(vemo8, vrepo8); __m512 vy9 = _mm512_mul_ps(vemo9, vrepo9); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); vy2 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy2), _mm512_castps_si512(vx2), vsign_mask, 0xD8)); vy3 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy3), _mm512_castps_si512(vx3), vsign_mask, 0xD8)); vy4 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy4), _mm512_castps_si512(vx4), vsign_mask, 0xD8)); vy5 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy5), _mm512_castps_si512(vx5), vsign_mask, 0xD8)); vy6 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy6), _mm512_castps_si512(vx6), vsign_mask, 0xD8)); vy7 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy7), _mm512_castps_si512(vx7), vsign_mask, 0xD8)); vy8 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy8), _mm512_castps_si512(vx8), vsign_mask, 0xD8)); vy9 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy9), _mm512_castps_si512(vx9), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); _mm512_storeu_ps(output + 32, vy2); _mm512_storeu_ps(output + 48, vy3); _mm512_storeu_ps(output + 64, vy4); _mm512_storeu_ps(output + 80, vy5); _mm512_storeu_ps(output + 96, vy6); _mm512_storeu_ps(output + 112, vy7); _mm512_storeu_ps(output + 128, vy8); _mm512_storeu_ps(output + 144, vy9); output += 160; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
16,842
46.579096
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x32.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x32( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 32 * sizeof(float); batch -= 32 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); input += 32; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); __m512 vp0 = vc6; __m512 vp1 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); output += 32; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
7,143
39.134831
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x48.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x48( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 48 * sizeof(float); batch -= 48 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); const __m512 vx2 = _mm512_loadu_ps(input + 32); input += 48; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); const __m512 vz2 = _mm512_range_ps(vsat_cutoff, vx2, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); __m512 vn2 = _mm512_fmadd_ps(vz2, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vs2 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn2), 23)); vn2 = _mm512_sub_ps(vn2, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); const __m512 vt2 = _mm512_fmadd_ps(vn2, vln2, vz2); __m512 vp0 = vc6; __m512 vp1 = vc6; __m512 vp2 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp2 = _mm512_fmadd_ps(vp2, vt2, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp2 = _mm512_fmadd_ps(vp2, vt2, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp2 = _mm512_fmadd_ps(vp2, vt2, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp2 = _mm512_fmadd_ps(vp2, vt2, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); vp2 = _mm512_fmadd_ps(vp2, vt2, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vts2 = _mm512_mul_ps(vt2, vs2); const __m512 vsmo2 = _mm512_sub_ps(vs2, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vemo2 = _mm512_fmadd_ps(vp2, vts2, vsmo2); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); const __m512 vepo2 = _mm512_sub_ps(vemo2, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); __m512 vrepo2 = _mm512_rcp14_ps(vepo2); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); const __m512 verepo2 = _mm512_fnmadd_ps(vrepo2, vepo2, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm512_fmadd_ps(verepo2, vrepo2, vrepo2); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); __m512 vy2 = _mm512_mul_ps(vemo2, vrepo2); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); vy2 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy2), _mm512_castps_si512(vx2), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); _mm512_storeu_ps(output + 32, vy2); output += 48; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
8,354
40.775
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x64.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x64( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 64 * sizeof(float); batch -= 64 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); const __m512 vx2 = _mm512_loadu_ps(input + 32); const __m512 vx3 = _mm512_loadu_ps(input + 48); input += 64; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); const __m512 vz2 = _mm512_range_ps(vsat_cutoff, vx2, 0xA); const __m512 vz3 = _mm512_range_ps(vsat_cutoff, vx3, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); __m512 vn2 = _mm512_fmadd_ps(vz2, vminus_log2e, vmagic_bias); __m512 vn3 = _mm512_fmadd_ps(vz3, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vs2 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn2), 23)); vn2 = _mm512_sub_ps(vn2, vmagic_bias); const __m512 vs3 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn3), 23)); vn3 = _mm512_sub_ps(vn3, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); const __m512 vt2 = _mm512_fmadd_ps(vn2, vln2, vz2); const __m512 vt3 = _mm512_fmadd_ps(vn3, vln2, vz3); __m512 vp0 = vc6; __m512 vp1 = vc6; __m512 vp2 = vc6; __m512 vp3 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp2 = _mm512_fmadd_ps(vp2, vt2, vc5); vp3 = _mm512_fmadd_ps(vp3, vt3, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp2 = _mm512_fmadd_ps(vp2, vt2, vc4); vp3 = _mm512_fmadd_ps(vp3, vt3, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp2 = _mm512_fmadd_ps(vp2, vt2, vc3); vp3 = _mm512_fmadd_ps(vp3, vt3, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp2 = _mm512_fmadd_ps(vp2, vt2, vc2); vp3 = _mm512_fmadd_ps(vp3, vt3, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); vp2 = _mm512_fmadd_ps(vp2, vt2, vminus_two); vp3 = _mm512_fmadd_ps(vp3, vt3, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vts2 = _mm512_mul_ps(vt2, vs2); const __m512 vsmo2 = _mm512_sub_ps(vs2, vone); const __m512 vts3 = _mm512_mul_ps(vt3, vs3); const __m512 vsmo3 = _mm512_sub_ps(vs3, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vemo2 = _mm512_fmadd_ps(vp2, vts2, vsmo2); const __m512 vemo3 = _mm512_fmadd_ps(vp3, vts3, vsmo3); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); const __m512 vepo2 = _mm512_sub_ps(vemo2, vminus_two); const __m512 vepo3 = _mm512_sub_ps(vemo3, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); __m512 vrepo2 = _mm512_rcp14_ps(vepo2); __m512 vrepo3 = _mm512_rcp14_ps(vepo3); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); const __m512 verepo2 = _mm512_fnmadd_ps(vrepo2, vepo2, vone); const __m512 verepo3 = _mm512_fnmadd_ps(vrepo3, vepo3, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm512_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm512_fmadd_ps(verepo3, vrepo3, vrepo3); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); __m512 vy2 = _mm512_mul_ps(vemo2, vrepo2); __m512 vy3 = _mm512_mul_ps(vemo3, vrepo3); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); vy2 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy2), _mm512_castps_si512(vx2), vsign_mask, 0xD8)); vy3 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy3), _mm512_castps_si512(vx3), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); _mm512_storeu_ps(output + 32, vy2); _mm512_storeu_ps(output + 48, vy3); output += 64; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
9,565
42.09009
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x80.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x80( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 80 * sizeof(float); batch -= 80 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); const __m512 vx2 = _mm512_loadu_ps(input + 32); const __m512 vx3 = _mm512_loadu_ps(input + 48); const __m512 vx4 = _mm512_loadu_ps(input + 64); input += 80; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); const __m512 vz2 = _mm512_range_ps(vsat_cutoff, vx2, 0xA); const __m512 vz3 = _mm512_range_ps(vsat_cutoff, vx3, 0xA); const __m512 vz4 = _mm512_range_ps(vsat_cutoff, vx4, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); __m512 vn2 = _mm512_fmadd_ps(vz2, vminus_log2e, vmagic_bias); __m512 vn3 = _mm512_fmadd_ps(vz3, vminus_log2e, vmagic_bias); __m512 vn4 = _mm512_fmadd_ps(vz4, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vs2 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn2), 23)); vn2 = _mm512_sub_ps(vn2, vmagic_bias); const __m512 vs3 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn3), 23)); vn3 = _mm512_sub_ps(vn3, vmagic_bias); const __m512 vs4 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn4), 23)); vn4 = _mm512_sub_ps(vn4, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); const __m512 vt2 = _mm512_fmadd_ps(vn2, vln2, vz2); const __m512 vt3 = _mm512_fmadd_ps(vn3, vln2, vz3); const __m512 vt4 = _mm512_fmadd_ps(vn4, vln2, vz4); __m512 vp0 = vc6; __m512 vp1 = vc6; __m512 vp2 = vc6; __m512 vp3 = vc6; __m512 vp4 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp2 = _mm512_fmadd_ps(vp2, vt2, vc5); vp3 = _mm512_fmadd_ps(vp3, vt3, vc5); vp4 = _mm512_fmadd_ps(vp4, vt4, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp2 = _mm512_fmadd_ps(vp2, vt2, vc4); vp3 = _mm512_fmadd_ps(vp3, vt3, vc4); vp4 = _mm512_fmadd_ps(vp4, vt4, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp2 = _mm512_fmadd_ps(vp2, vt2, vc3); vp3 = _mm512_fmadd_ps(vp3, vt3, vc3); vp4 = _mm512_fmadd_ps(vp4, vt4, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp2 = _mm512_fmadd_ps(vp2, vt2, vc2); vp3 = _mm512_fmadd_ps(vp3, vt3, vc2); vp4 = _mm512_fmadd_ps(vp4, vt4, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); vp2 = _mm512_fmadd_ps(vp2, vt2, vminus_two); vp3 = _mm512_fmadd_ps(vp3, vt3, vminus_two); vp4 = _mm512_fmadd_ps(vp4, vt4, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vts2 = _mm512_mul_ps(vt2, vs2); const __m512 vsmo2 = _mm512_sub_ps(vs2, vone); const __m512 vts3 = _mm512_mul_ps(vt3, vs3); const __m512 vsmo3 = _mm512_sub_ps(vs3, vone); const __m512 vts4 = _mm512_mul_ps(vt4, vs4); const __m512 vsmo4 = _mm512_sub_ps(vs4, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vemo2 = _mm512_fmadd_ps(vp2, vts2, vsmo2); const __m512 vemo3 = _mm512_fmadd_ps(vp3, vts3, vsmo3); const __m512 vemo4 = _mm512_fmadd_ps(vp4, vts4, vsmo4); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); const __m512 vepo2 = _mm512_sub_ps(vemo2, vminus_two); const __m512 vepo3 = _mm512_sub_ps(vemo3, vminus_two); const __m512 vepo4 = _mm512_sub_ps(vemo4, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); __m512 vrepo2 = _mm512_rcp14_ps(vepo2); __m512 vrepo3 = _mm512_rcp14_ps(vepo3); __m512 vrepo4 = _mm512_rcp14_ps(vepo4); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); const __m512 verepo2 = _mm512_fnmadd_ps(vrepo2, vepo2, vone); const __m512 verepo3 = _mm512_fnmadd_ps(vrepo3, vepo3, vone); const __m512 verepo4 = _mm512_fnmadd_ps(vrepo4, vepo4, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm512_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm512_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm512_fmadd_ps(verepo4, vrepo4, vrepo4); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); __m512 vy2 = _mm512_mul_ps(vemo2, vrepo2); __m512 vy3 = _mm512_mul_ps(vemo3, vrepo3); __m512 vy4 = _mm512_mul_ps(vemo4, vrepo4); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); vy2 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy2), _mm512_castps_si512(vx2), vsign_mask, 0xD8)); vy3 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy3), _mm512_castps_si512(vx3), vsign_mask, 0xD8)); vy4 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy4), _mm512_castps_si512(vx4), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); _mm512_storeu_ps(output + 32, vy2); _mm512_storeu_ps(output + 48, vy3); _mm512_storeu_ps(output + 64, vy4); output += 80; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
10,776
43.168033
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-avx512skx-expm1minus-rr1-p6h5ts-nr1-x96.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx512skx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/intrinsics-polyfill.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__avx512skx_expm1minus_rr1_p6h5ts_nr1_x96( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m512 vsat_cutoff = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.sat_cutoff); const __m512 vminus_log2e = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_log2e); const __m512 vmagic_bias = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.magic_bias); const __m512 vln2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.ln2); const __m512 vc6 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c6); const __m512 vc5 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c5); const __m512 vc4 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c4); const __m512 vc3 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c3); const __m512 vc2 = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.c2); const __m512 vminus_two = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.minus_two); const __m512 vone = _mm512_set1_ps(params->avx512_expm1minus_rr1_p6h5.one); const __m512i vsign_mask = _mm512_set1_epi32((int) params->avx512_expm1minus_rr1_p6h5.sign_mask); for (; batch >= 96 * sizeof(float); batch -= 96 * sizeof(float)) { const __m512 vx0 = _mm512_loadu_ps(input); const __m512 vx1 = _mm512_loadu_ps(input + 16); const __m512 vx2 = _mm512_loadu_ps(input + 32); const __m512 vx3 = _mm512_loadu_ps(input + 48); const __m512 vx4 = _mm512_loadu_ps(input + 64); const __m512 vx5 = _mm512_loadu_ps(input + 80); input += 96; const __m512 vz0 = _mm512_range_ps(vsat_cutoff, vx0, 0xA); const __m512 vz1 = _mm512_range_ps(vsat_cutoff, vx1, 0xA); const __m512 vz2 = _mm512_range_ps(vsat_cutoff, vx2, 0xA); const __m512 vz3 = _mm512_range_ps(vsat_cutoff, vx3, 0xA); const __m512 vz4 = _mm512_range_ps(vsat_cutoff, vx4, 0xA); const __m512 vz5 = _mm512_range_ps(vsat_cutoff, vx5, 0xA); __m512 vn0 = _mm512_fmadd_ps(vz0, vminus_log2e, vmagic_bias); __m512 vn1 = _mm512_fmadd_ps(vz1, vminus_log2e, vmagic_bias); __m512 vn2 = _mm512_fmadd_ps(vz2, vminus_log2e, vmagic_bias); __m512 vn3 = _mm512_fmadd_ps(vz3, vminus_log2e, vmagic_bias); __m512 vn4 = _mm512_fmadd_ps(vz4, vminus_log2e, vmagic_bias); __m512 vn5 = _mm512_fmadd_ps(vz5, vminus_log2e, vmagic_bias); const __m512 vs0 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn0), 23)); vn0 = _mm512_sub_ps(vn0, vmagic_bias); const __m512 vs1 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn1), 23)); vn1 = _mm512_sub_ps(vn1, vmagic_bias); const __m512 vs2 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn2), 23)); vn2 = _mm512_sub_ps(vn2, vmagic_bias); const __m512 vs3 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn3), 23)); vn3 = _mm512_sub_ps(vn3, vmagic_bias); const __m512 vs4 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn4), 23)); vn4 = _mm512_sub_ps(vn4, vmagic_bias); const __m512 vs5 = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn5), 23)); vn5 = _mm512_sub_ps(vn5, vmagic_bias); const __m512 vt0 = _mm512_fmadd_ps(vn0, vln2, vz0); const __m512 vt1 = _mm512_fmadd_ps(vn1, vln2, vz1); const __m512 vt2 = _mm512_fmadd_ps(vn2, vln2, vz2); const __m512 vt3 = _mm512_fmadd_ps(vn3, vln2, vz3); const __m512 vt4 = _mm512_fmadd_ps(vn4, vln2, vz4); const __m512 vt5 = _mm512_fmadd_ps(vn5, vln2, vz5); __m512 vp0 = vc6; __m512 vp1 = vc6; __m512 vp2 = vc6; __m512 vp3 = vc6; __m512 vp4 = vc6; __m512 vp5 = vc6; vp0 = _mm512_fmadd_ps(vp0, vt0, vc5); vp1 = _mm512_fmadd_ps(vp1, vt1, vc5); vp2 = _mm512_fmadd_ps(vp2, vt2, vc5); vp3 = _mm512_fmadd_ps(vp3, vt3, vc5); vp4 = _mm512_fmadd_ps(vp4, vt4, vc5); vp5 = _mm512_fmadd_ps(vp5, vt5, vc5); vp0 = _mm512_fmadd_ps(vp0, vt0, vc4); vp1 = _mm512_fmadd_ps(vp1, vt1, vc4); vp2 = _mm512_fmadd_ps(vp2, vt2, vc4); vp3 = _mm512_fmadd_ps(vp3, vt3, vc4); vp4 = _mm512_fmadd_ps(vp4, vt4, vc4); vp5 = _mm512_fmadd_ps(vp5, vt5, vc4); vp0 = _mm512_fmadd_ps(vp0, vt0, vc3); vp1 = _mm512_fmadd_ps(vp1, vt1, vc3); vp2 = _mm512_fmadd_ps(vp2, vt2, vc3); vp3 = _mm512_fmadd_ps(vp3, vt3, vc3); vp4 = _mm512_fmadd_ps(vp4, vt4, vc3); vp5 = _mm512_fmadd_ps(vp5, vt5, vc3); vp0 = _mm512_fmadd_ps(vp0, vt0, vc2); vp1 = _mm512_fmadd_ps(vp1, vt1, vc2); vp2 = _mm512_fmadd_ps(vp2, vt2, vc2); vp3 = _mm512_fmadd_ps(vp3, vt3, vc2); vp4 = _mm512_fmadd_ps(vp4, vt4, vc2); vp5 = _mm512_fmadd_ps(vp5, vt5, vc2); vp0 = _mm512_fmadd_ps(vp0, vt0, vminus_two); vp1 = _mm512_fmadd_ps(vp1, vt1, vminus_two); vp2 = _mm512_fmadd_ps(vp2, vt2, vminus_two); vp3 = _mm512_fmadd_ps(vp3, vt3, vminus_two); vp4 = _mm512_fmadd_ps(vp4, vt4, vminus_two); vp5 = _mm512_fmadd_ps(vp5, vt5, vminus_two); const __m512 vts0 = _mm512_mul_ps(vt0, vs0); const __m512 vsmo0 = _mm512_sub_ps(vs0, vone); const __m512 vts1 = _mm512_mul_ps(vt1, vs1); const __m512 vsmo1 = _mm512_sub_ps(vs1, vone); const __m512 vts2 = _mm512_mul_ps(vt2, vs2); const __m512 vsmo2 = _mm512_sub_ps(vs2, vone); const __m512 vts3 = _mm512_mul_ps(vt3, vs3); const __m512 vsmo3 = _mm512_sub_ps(vs3, vone); const __m512 vts4 = _mm512_mul_ps(vt4, vs4); const __m512 vsmo4 = _mm512_sub_ps(vs4, vone); const __m512 vts5 = _mm512_mul_ps(vt5, vs5); const __m512 vsmo5 = _mm512_sub_ps(vs5, vone); const __m512 vemo0 = _mm512_fmadd_ps(vp0, vts0, vsmo0); const __m512 vemo1 = _mm512_fmadd_ps(vp1, vts1, vsmo1); const __m512 vemo2 = _mm512_fmadd_ps(vp2, vts2, vsmo2); const __m512 vemo3 = _mm512_fmadd_ps(vp3, vts3, vsmo3); const __m512 vemo4 = _mm512_fmadd_ps(vp4, vts4, vsmo4); const __m512 vemo5 = _mm512_fmadd_ps(vp5, vts5, vsmo5); const __m512 vepo0 = _mm512_sub_ps(vemo0, vminus_two); const __m512 vepo1 = _mm512_sub_ps(vemo1, vminus_two); const __m512 vepo2 = _mm512_sub_ps(vemo2, vminus_two); const __m512 vepo3 = _mm512_sub_ps(vemo3, vminus_two); const __m512 vepo4 = _mm512_sub_ps(vemo4, vminus_two); const __m512 vepo5 = _mm512_sub_ps(vemo5, vminus_two); __m512 vrepo0 = _mm512_rcp14_ps(vepo0); __m512 vrepo1 = _mm512_rcp14_ps(vepo1); __m512 vrepo2 = _mm512_rcp14_ps(vepo2); __m512 vrepo3 = _mm512_rcp14_ps(vepo3); __m512 vrepo4 = _mm512_rcp14_ps(vepo4); __m512 vrepo5 = _mm512_rcp14_ps(vepo5); const __m512 verepo0 = _mm512_fnmadd_ps(vrepo0, vepo0, vone); const __m512 verepo1 = _mm512_fnmadd_ps(vrepo1, vepo1, vone); const __m512 verepo2 = _mm512_fnmadd_ps(vrepo2, vepo2, vone); const __m512 verepo3 = _mm512_fnmadd_ps(vrepo3, vepo3, vone); const __m512 verepo4 = _mm512_fnmadd_ps(vrepo4, vepo4, vone); const __m512 verepo5 = _mm512_fnmadd_ps(vrepo5, vepo5, vone); vrepo0 = _mm512_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm512_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm512_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm512_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm512_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm512_fmadd_ps(verepo5, vrepo5, vrepo5); __m512 vy0 = _mm512_mul_ps(vemo0, vrepo0); __m512 vy1 = _mm512_mul_ps(vemo1, vrepo1); __m512 vy2 = _mm512_mul_ps(vemo2, vrepo2); __m512 vy3 = _mm512_mul_ps(vemo3, vrepo3); __m512 vy4 = _mm512_mul_ps(vemo4, vrepo4); __m512 vy5 = _mm512_mul_ps(vemo5, vrepo5); vy0 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy0), _mm512_castps_si512(vx0), vsign_mask, 0xD8)); vy1 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy1), _mm512_castps_si512(vx1), vsign_mask, 0xD8)); vy2 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy2), _mm512_castps_si512(vx2), vsign_mask, 0xD8)); vy3 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy3), _mm512_castps_si512(vx3), vsign_mask, 0xD8)); vy4 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy4), _mm512_castps_si512(vx4), vsign_mask, 0xD8)); vy5 = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy5), _mm512_castps_si512(vx5), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy0); _mm512_storeu_ps(output + 16, vy1); _mm512_storeu_ps(output + 32, vy2); _mm512_storeu_ps(output + 48, vy3); _mm512_storeu_ps(output + 64, vy4); _mm512_storeu_ps(output + 80, vy5); output += 96; } for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m512 vx = _mm512_loadu_ps(input); input += 16; const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_storeu_ps(output, vy); output += 16; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 15 * sizeof(float)); // Prepare mask for valid 32-bit elements (depends on batch). batch >>= XNN_LOG2_SIZEOF_FLOAT; const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << batch) - UINT32_C(1))); const __m512 vx = _mm512_maskz_loadu_ps(vmask, input); const __m512 vz = _mm512_range_ps(vsat_cutoff, vx, 0xA); __m512 vn = _mm512_fmadd_ps(vz, vminus_log2e, vmagic_bias); const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); vn = _mm512_sub_ps(vn, vmagic_bias); const __m512 vt = _mm512_fmadd_ps(vn, vln2, vz); __m512 vp = vc6; vp = _mm512_fmadd_ps(vp, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_fmadd_ps(vp, vt, vminus_two); const __m512 vts = _mm512_mul_ps(vt, vs); const __m512 vsmo = _mm512_sub_ps(vs, vone); const __m512 vemo = _mm512_fmadd_ps(vp, vts, vsmo); const __m512 vepo = _mm512_sub_ps(vemo, vminus_two); __m512 vrepo = _mm512_rcp14_ps(vepo); const __m512 verepo = _mm512_fnmadd_ps(vrepo, vepo, vone); vrepo = _mm512_fmadd_ps(verepo, vrepo, vrepo); __m512 vy = _mm512_mul_ps(vemo, vrepo); vy = _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(vy), _mm512_castps_si512(vx), vsign_mask, 0xD8)); _mm512_mask_storeu_ps(output, vmask, vy); } }
11,987
44.067669
127
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-lut8-p4h3ts-div-x1.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_lut8_p4h3ts_div_x1( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc4, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); }
2,417
27.785714
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-lut8-p4h3ts-div-x2.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_lut8_p4h3ts_div_x2( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; for (; batch >= 2 * sizeof(float); batch -= 2 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; input += 2; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); float vn0 = fmaf(vz0, vminus_log2e, vmagic_bias); float vn1 = fmaf(vz1, vminus_log2e, vmagic_bias); const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vidx0 = vb0 & vindex_mask; const uint32_t vidx1 = vb1 & vindex_mask; const uint32_t vl0 = xnn_table_exp2minus_k_over_8[vidx0]; uint32_t ve0 = vb0 << 20; const uint32_t vl1 = xnn_table_exp2minus_k_over_8[vidx1]; uint32_t ve1 = vb1 << 20; ve0 += vl0; ve1 += vl1; const float vt0 = fmaf(vn0, vln2, vz0); const float vs0 = uint32_as_float(ve0); const float vt1 = fmaf(vn1, vln2, vz1); const float vs1 = uint32_as_float(ve1); float vp0 = fmaf(vc4, vt0, vc3); float vp1 = fmaf(vc4, vt1, vc3); vp0 = fmaf(vp0, vt0, vc2); vp1 = fmaf(vp1, vt1, vc2); vp0 = fmaf(vp0, vt0, vminus_two); vp1 = fmaf(vp1, vt1, vminus_two); const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vemo0 = fmaf(vp0, vts0, vsmo0); const float vemo1 = fmaf(vp1, vts1, vsmo1); const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); output[0] = vy0; output[1] = vy1; output += 2; } if XNN_UNLIKELY(batch != 0) { const float vx = *input; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc4, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output = vy; } }
4,150
27.627586
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-lut8-p4h3ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_lut8_p4h3ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; const float vx2 = input[2]; const float vx3 = input[3]; input += 4; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); float vz2 = fabsf(vx2); float vz3 = fabsf(vx3); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); vz2 = math_pmin_f32(vz2, vsat_cutoff); vz3 = math_pmin_f32(vz3, vsat_cutoff); float vn0 = fmaf(vz0, vminus_log2e, vmagic_bias); float vn1 = fmaf(vz1, vminus_log2e, vmagic_bias); float vn2 = fmaf(vz2, vminus_log2e, vmagic_bias); float vn3 = fmaf(vz3, vminus_log2e, vmagic_bias); const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vb2 = float_as_uint32(vn2); vn2 -= vmagic_bias; const uint32_t vb3 = float_as_uint32(vn3); vn3 -= vmagic_bias; const uint32_t vidx0 = vb0 & vindex_mask; const uint32_t vidx1 = vb1 & vindex_mask; const uint32_t vidx2 = vb2 & vindex_mask; const uint32_t vidx3 = vb3 & vindex_mask; const uint32_t vl0 = xnn_table_exp2minus_k_over_8[vidx0]; uint32_t ve0 = vb0 << 20; const uint32_t vl1 = xnn_table_exp2minus_k_over_8[vidx1]; uint32_t ve1 = vb1 << 20; const uint32_t vl2 = xnn_table_exp2minus_k_over_8[vidx2]; uint32_t ve2 = vb2 << 20; const uint32_t vl3 = xnn_table_exp2minus_k_over_8[vidx3]; uint32_t ve3 = vb3 << 20; ve0 += vl0; ve1 += vl1; ve2 += vl2; ve3 += vl3; const float vt0 = fmaf(vn0, vln2, vz0); const float vs0 = uint32_as_float(ve0); const float vt1 = fmaf(vn1, vln2, vz1); const float vs1 = uint32_as_float(ve1); const float vt2 = fmaf(vn2, vln2, vz2); const float vs2 = uint32_as_float(ve2); const float vt3 = fmaf(vn3, vln2, vz3); const float vs3 = uint32_as_float(ve3); float vp0 = fmaf(vc4, vt0, vc3); float vp1 = fmaf(vc4, vt1, vc3); float vp2 = fmaf(vc4, vt2, vc3); float vp3 = fmaf(vc4, vt3, vc3); vp0 = fmaf(vp0, vt0, vc2); vp1 = fmaf(vp1, vt1, vc2); vp2 = fmaf(vp2, vt2, vc2); vp3 = fmaf(vp3, vt3, vc2); vp0 = fmaf(vp0, vt0, vminus_two); vp1 = fmaf(vp1, vt1, vminus_two); vp2 = fmaf(vp2, vt2, vminus_two); vp3 = fmaf(vp3, vt3, vminus_two); const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vts2 = vt2 * vs2; const float vsmo2 = vs2 - vone; const float vts3 = vt3 * vs3; const float vsmo3 = vs3 - vone; const float vemo0 = fmaf(vp0, vts0, vsmo0); const float vemo1 = fmaf(vp1, vts1, vsmo1); const float vemo2 = fmaf(vp2, vts2, vsmo2); const float vemo3 = fmaf(vp3, vts3, vsmo3); const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; const float vepo2 = vemo2 - vminus_two; const float vepo3 = vemo3 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; float vy2 = vemo2 / vepo2; float vy3 = vemo3 / vepo3; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); vy2 = copysignf(vy2, vx2); vy3 = copysignf(vy3, vx3); output[0] = vy0; output[1] = vy1; output[2] = vy2; output[3] = vy3; output += 4; } if XNN_UNLIKELY(batch != 0) { do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc4, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); } }
5,906
29.606218
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-p6h5ts-div-x1.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_p6h5ts_div_x1( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc6, vt, vc5); vp = fmaf(vp, vt, vc4); vp = fmaf(vp, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); }
2,314
27.231707
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-p6h5ts-div-x2.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_p6h5ts_div_x2( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; for (; batch >= 2 * sizeof(float); batch -= 2 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; input += 2; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); float vn0 = fmaf(vz0, vminus_log2e, vmagic_bias); float vn1 = fmaf(vz1, vminus_log2e, vmagic_bias); const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t ve0 = vb0 << 23; const uint32_t ve1 = vb1 << 23; const float vt0 = fmaf(vn0, vln2, vz0); const float vs0 = uint32_as_float(ve0); const float vt1 = fmaf(vn1, vln2, vz1); const float vs1 = uint32_as_float(ve1); float vp0 = fmaf(vc6, vt0, vc5); float vp1 = fmaf(vc6, vt1, vc5); vp0 = fmaf(vp0, vt0, vc4); vp1 = fmaf(vp1, vt1, vc4); vp0 = fmaf(vp0, vt0, vc3); vp1 = fmaf(vp1, vt1, vc3); vp0 = fmaf(vp0, vt0, vc2); vp1 = fmaf(vp1, vt1, vc2); vp0 = fmaf(vp0, vt0, vminus_two); vp1 = fmaf(vp1, vt1, vminus_two); const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vemo0 = fmaf(vp0, vts0, vsmo0); const float vemo1 = fmaf(vp1, vts1, vsmo1); const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); output[0] = vy0; output[1] = vy1; output += 2; } if XNN_UNLIKELY(batch != 0) { const float vx = *input; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc6, vt, vc5); vp = fmaf(vp, vt, vc4); vp = fmaf(vp, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output = vy; } }
3,933
27.302158
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma-expm1minus-rr1-p6h5ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma_expm1minus_rr1_p6h5ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; const float vx2 = input[2]; const float vx3 = input[3]; input += 4; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); float vz2 = fabsf(vx2); float vz3 = fabsf(vx3); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); vz2 = math_pmin_f32(vz2, vsat_cutoff); vz3 = math_pmin_f32(vz3, vsat_cutoff); float vn0 = fmaf(vz0, vminus_log2e, vmagic_bias); float vn1 = fmaf(vz1, vminus_log2e, vmagic_bias); float vn2 = fmaf(vz2, vminus_log2e, vmagic_bias); float vn3 = fmaf(vz3, vminus_log2e, vmagic_bias); const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vb2 = float_as_uint32(vn2); vn2 -= vmagic_bias; const uint32_t vb3 = float_as_uint32(vn3); vn3 -= vmagic_bias; const uint32_t ve0 = vb0 << 23; const uint32_t ve1 = vb1 << 23; const uint32_t ve2 = vb2 << 23; const uint32_t ve3 = vb3 << 23; const float vt0 = fmaf(vn0, vln2, vz0); const float vs0 = uint32_as_float(ve0); const float vt1 = fmaf(vn1, vln2, vz1); const float vs1 = uint32_as_float(ve1); const float vt2 = fmaf(vn2, vln2, vz2); const float vs2 = uint32_as_float(ve2); const float vt3 = fmaf(vn3, vln2, vz3); const float vs3 = uint32_as_float(ve3); float vp0 = fmaf(vc6, vt0, vc5); float vp1 = fmaf(vc6, vt1, vc5); float vp2 = fmaf(vc6, vt2, vc5); float vp3 = fmaf(vc6, vt3, vc5); vp0 = fmaf(vp0, vt0, vc4); vp1 = fmaf(vp1, vt1, vc4); vp2 = fmaf(vp2, vt2, vc4); vp3 = fmaf(vp3, vt3, vc4); vp0 = fmaf(vp0, vt0, vc3); vp1 = fmaf(vp1, vt1, vc3); vp2 = fmaf(vp2, vt2, vc3); vp3 = fmaf(vp3, vt3, vc3); vp0 = fmaf(vp0, vt0, vc2); vp1 = fmaf(vp1, vt1, vc2); vp2 = fmaf(vp2, vt2, vc2); vp3 = fmaf(vp3, vt3, vc2); vp0 = fmaf(vp0, vt0, vminus_two); vp1 = fmaf(vp1, vt1, vminus_two); vp2 = fmaf(vp2, vt2, vminus_two); vp3 = fmaf(vp3, vt3, vminus_two); const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vts2 = vt2 * vs2; const float vsmo2 = vs2 - vone; const float vts3 = vt3 * vs3; const float vsmo3 = vs3 - vone; const float vemo0 = fmaf(vp0, vts0, vsmo0); const float vemo1 = fmaf(vp1, vts1, vsmo1); const float vemo2 = fmaf(vp2, vts2, vsmo2); const float vemo3 = fmaf(vp3, vts3, vsmo3); const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; const float vepo2 = vemo2 - vminus_two; const float vepo3 = vemo3 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; float vy2 = vemo2 / vepo2; float vy3 = vemo3 / vepo3; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); vy2 = copysignf(vy2, vx2); vy3 = copysignf(vy3, vx3); output[0] = vy0; output[1] = vy1; output[2] = vy2; output[3] = vy3; output += 4; } if XNN_UNLIKELY(batch != 0) { do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = fmaf(vz, vminus_log2e, vmagic_bias); const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = fmaf(vn, vln2, vz); float vp = fmaf(vc6, vt, vc5); vp = fmaf(vp, vt, vc4); vp = fmaf(vp, vt, vc3); vp = fmaf(vp, vt, vc2); vp = fmaf(vp, vt, vminus_two); const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = fmaf(vp, vts, vsmo); const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); } }
5,575
29.140541
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); input += 16; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); __m256 vp0 = vc6; __m256 vp1 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); output += 16; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
7,550
36.014706
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x24.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x24( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); input += 24; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); output += 24; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
8,814
37.662281
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x32.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x32( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 32 * sizeof(float); batch -= 32 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); input += 32; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); output += 32; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
10,078
38.996032
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x40.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x40( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 40 * sizeof(float); batch -= 40 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); input += 40; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); output += 40; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
11,342
40.097826
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x48.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x48( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 48 * sizeof(float); batch -= 48 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); input += 48; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); __m256 vy5 = _mm256_div_ps(vemo5, vepo5); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); output += 48; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
12,606
41.023333
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x56.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x56( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 56 * sizeof(float); batch -= 56 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); input += 56; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); __m256 vy5 = _mm256_div_ps(vemo5, vepo5); __m256 vy6 = _mm256_div_ps(vemo6, vepo6); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); output += 56; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
13,870
41.811728
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x64.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x64( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 64 * sizeof(float); batch -= 64 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); input += 64; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); __m256 vy5 = _mm256_div_ps(vemo5, vepo5); __m256 vy6 = _mm256_div_ps(vemo6, vepo6); __m256 vy7 = _mm256_div_ps(vemo7, vepo7); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); output += 64; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
15,134
42.491379
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x72.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x72( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 72 * sizeof(float); batch -= 72 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); input += 72; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); vz8 = _mm256_max_ps(vsat_cutoff, vz8); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); __m256 vy5 = _mm256_div_ps(vemo5, vepo5); __m256 vy6 = _mm256_div_ps(vemo6, vepo6); __m256 vy7 = _mm256_div_ps(vemo7, vepo7); __m256 vy8 = _mm256_div_ps(vemo8, vepo8); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); output += 72; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
16,398
43.083333
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
4,911
34.594203
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-div-x80.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_div_x80( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 80 * sizeof(float); batch -= 80 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); const __m256 vx9 = _mm256_loadu_ps(input + 72); input += 80; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); __m256 vz9 = _mm256_or_ps(vx9, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); vz8 = _mm256_max_ps(vsat_cutoff, vz8); const __m256 vinvsignx9 = _mm256_xor_ps(vx9, vz9); vz9 = _mm256_max_ps(vsat_cutoff, vz9); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); __m256 vn9 = _mm256_fmadd_ps(vz9, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); const __m128 vn9_hi = _mm256_extractf128_ps(vn9, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); __m256 vs9 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn9)), 23))); const __m128 vs9_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn9_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vs9 = _mm256_insertf128_ps(vs9, vs9_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); vn9 = _mm256_sub_ps(vn9, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); const __m256 vt9 = _mm256_fmadd_ps(vn9, vminus_ln2, vz9); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; __m256 vp9 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp9 = _mm256_fmadd_ps(vp9, vt9, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp9 = _mm256_fmadd_ps(vp9, vt9, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp9 = _mm256_fmadd_ps(vp9, vt9, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp9 = _mm256_fmadd_ps(vp9, vt9, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); vp9 = _mm256_fmadd_ps(vp9, vt9, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vts9 = _mm256_mul_ps(vt9, vs9); const __m256 vsmo9 = _mm256_add_ps(vs9, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vemo9 = _mm256_fmadd_ps(vp9, vts9, vsmo9); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); const __m256 vepo9 = _mm256_add_ps(vemo9, vtwo); __m256 vy0 = _mm256_div_ps(vemo0, vepo0); __m256 vy1 = _mm256_div_ps(vemo1, vepo1); __m256 vy2 = _mm256_div_ps(vemo2, vepo2); __m256 vy3 = _mm256_div_ps(vemo3, vepo3); __m256 vy4 = _mm256_div_ps(vemo4, vepo4); __m256 vy5 = _mm256_div_ps(vemo5, vepo5); __m256 vy6 = _mm256_div_ps(vemo6, vepo6); __m256 vy7 = _mm256_div_ps(vemo7, vepo7); __m256 vy8 = _mm256_div_ps(vemo8, vepo8); __m256 vy9 = _mm256_div_ps(vemo9, vepo9); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); vy9 = _mm256_xor_ps(vy9, vinvsignx9); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); _mm256_storeu_ps(output + 72, vy9); output += 80; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vy = _mm256_div_ps(vemo, vepo); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
17,662
43.603535
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); input += 16; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); __m256 vp0 = vc6; __m256 vp1 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); output += 16; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
8,536
36.774336
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x24.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x24( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); input += 24; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); output += 24; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
10,051
38.574803
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x32.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x32( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 32 * sizeof(float); batch -= 32 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); input += 32; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); output += 32; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
11,566
40.01773
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x40.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x40( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 40 * sizeof(float); batch -= 40 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); input += 40; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); output += 40; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
13,081
41.2
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x48.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x48( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 48 * sizeof(float); batch -= 48 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); input += 48; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); const __m256 vm5 = _mm256_cmp_ps(vz5, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy5 = _mm256_blendv_ps(vy5, vminus_one, vm5); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); output += 48; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
14,596
42.186391
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x56.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x56( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 56 * sizeof(float); batch -= 56 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); input += 56; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); const __m256 vm5 = _mm256_cmp_ps(vz5, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); const __m256 vm6 = _mm256_cmp_ps(vz6, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy5 = _mm256_blendv_ps(vy5, vminus_one, vm5); vy6 = _mm256_blendv_ps(vy6, vminus_one, vm6); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); output += 56; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
16,111
43.021858
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x64.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x64( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 64 * sizeof(float); batch -= 64 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); input += 64; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); const __m256 vm5 = _mm256_cmp_ps(vz5, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); const __m256 vm6 = _mm256_cmp_ps(vz6, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); const __m256 vm7 = _mm256_cmp_ps(vz7, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy5 = _mm256_blendv_ps(vy5, vminus_one, vm5); vy6 = _mm256_blendv_ps(vy6, vminus_one, vm6); vy7 = _mm256_blendv_ps(vy7, vminus_one, vm7); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); output += 64; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
17,626
43.738579
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x72.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x72( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 72 * sizeof(float); batch -= 72 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); input += 72; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); const __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); const __m256 vm5 = _mm256_cmp_ps(vz5, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); const __m256 vm6 = _mm256_cmp_ps(vz6, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); const __m256 vm7 = _mm256_cmp_ps(vz7, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); const __m256 vm8 = _mm256_cmp_ps(vz8, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); __m256 vrepo8 = _mm256_rcp_ps(vepo8); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); const __m256 verepo8 = _mm256_fnmsub_ps(vrepo8, vepo8, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); vrepo8 = _mm256_fmadd_ps(verepo8, vrepo8, vrepo8); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); __m256 vy8 = _mm256_mul_ps(vemo8, vrepo8); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy5 = _mm256_blendv_ps(vy5, vminus_one, vm5); vy6 = _mm256_blendv_ps(vy6, vminus_one, vm6); vy7 = _mm256_blendv_ps(vy7, vminus_one, vm7); vy8 = _mm256_blendv_ps(vy8, vminus_one, vm8); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); output += 72; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
19,141
44.36019
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
5,393
34.96
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1-x80.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1_x80( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 80 * sizeof(float); batch -= 80 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); const __m256 vx9 = _mm256_loadu_ps(input + 72); input += 80; const __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); const __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); const __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); const __m256 vz9 = _mm256_or_ps(vx9, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); const __m256 vm0 = _mm256_cmp_ps(vz0, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); const __m256 vm1 = _mm256_cmp_ps(vz1, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); const __m256 vm2 = _mm256_cmp_ps(vz2, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); const __m256 vm3 = _mm256_cmp_ps(vz3, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); const __m256 vm4 = _mm256_cmp_ps(vz4, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); const __m256 vm5 = _mm256_cmp_ps(vz5, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); const __m256 vm6 = _mm256_cmp_ps(vz6, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); const __m256 vm7 = _mm256_cmp_ps(vz7, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); const __m256 vm8 = _mm256_cmp_ps(vz8, vsat_cutoff, _CMP_LE_OS); const __m256 vinvsignx9 = _mm256_xor_ps(vx9, vz9); const __m256 vm9 = _mm256_cmp_ps(vz9, vsat_cutoff, _CMP_LE_OS); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); __m256 vn9 = _mm256_fmadd_ps(vz9, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); const __m128 vn9_hi = _mm256_extractf128_ps(vn9, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); __m256 vs9 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn9)), 23))); const __m128 vs9_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn9_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vs9 = _mm256_insertf128_ps(vs9, vs9_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); vn9 = _mm256_sub_ps(vn9, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); const __m256 vt9 = _mm256_fmadd_ps(vn9, vminus_ln2, vz9); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; __m256 vp9 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp9 = _mm256_fmadd_ps(vp9, vt9, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp9 = _mm256_fmadd_ps(vp9, vt9, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp9 = _mm256_fmadd_ps(vp9, vt9, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp9 = _mm256_fmadd_ps(vp9, vt9, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); vp9 = _mm256_fmadd_ps(vp9, vt9, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vts9 = _mm256_mul_ps(vt9, vs9); const __m256 vsmo9 = _mm256_add_ps(vs9, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vemo9 = _mm256_fmadd_ps(vp9, vts9, vsmo9); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); const __m256 vepo9 = _mm256_add_ps(vemo9, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); __m256 vrepo8 = _mm256_rcp_ps(vepo8); __m256 vrepo9 = _mm256_rcp_ps(vepo9); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); const __m256 verepo8 = _mm256_fnmsub_ps(vrepo8, vepo8, vminus_one); const __m256 verepo9 = _mm256_fnmsub_ps(vrepo9, vepo9, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); vrepo8 = _mm256_fmadd_ps(verepo8, vrepo8, vrepo8); vrepo9 = _mm256_fmadd_ps(verepo9, vrepo9, vrepo9); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); __m256 vy8 = _mm256_mul_ps(vemo8, vrepo8); __m256 vy9 = _mm256_mul_ps(vemo9, vrepo9); vy0 = _mm256_blendv_ps(vy0, vminus_one, vm0); vy1 = _mm256_blendv_ps(vy1, vminus_one, vm1); vy2 = _mm256_blendv_ps(vy2, vminus_one, vm2); vy3 = _mm256_blendv_ps(vy3, vminus_one, vm3); vy4 = _mm256_blendv_ps(vy4, vminus_one, vm4); vy5 = _mm256_blendv_ps(vy5, vminus_one, vm5); vy6 = _mm256_blendv_ps(vy6, vminus_one, vm6); vy7 = _mm256_blendv_ps(vy7, vminus_one, vm7); vy8 = _mm256_blendv_ps(vy8, vminus_one, vm8); vy9 = _mm256_blendv_ps(vy9, vminus_one, vm9); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); vy9 = _mm256_xor_ps(vy9, vinvsignx9); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); _mm256_storeu_ps(output + 72, vy9); output += 80; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); const __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); vy = _mm256_blendv_ps(vy, vminus_one, vm); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
20,656
44.904444
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); input += 16; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); __m256 vp0 = vc6; __m256 vp1 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); output += 16; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
8,633
36.53913
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x24.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x24( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); input += 24; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); output += 24; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
10,174
38.285714
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x32.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x32( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 32 * sizeof(float); batch -= 32 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); input += 32; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); output += 32; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
11,715
39.680556
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x40.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x40( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 40 * sizeof(float); batch -= 40 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); input += 40; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); output += 40; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
13,256
40.820189
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x48.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x48( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 48 * sizeof(float); batch -= 48 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); input += 48; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); const __m256 vey5 = _mm256_fnmadd_ps(vy5, vepo5, vemo5); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy5 = _mm256_fmadd_ps(vey5, vrepo5, vy5); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); output += 48; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
14,797
41.768786
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x56.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x56( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 56 * sizeof(float); batch -= 56 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); input += 56; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); const __m256 vey5 = _mm256_fnmadd_ps(vy5, vepo5, vemo5); const __m256 vey6 = _mm256_fnmadd_ps(vy6, vepo6, vemo6); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy5 = _mm256_fmadd_ps(vey5, vrepo5, vy5); vy6 = _mm256_fmadd_ps(vey6, vrepo6, vy6); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); output += 56; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
16,338
42.570667
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x64.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x64( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 64 * sizeof(float); batch -= 64 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); input += 64; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); const __m256 vey5 = _mm256_fnmadd_ps(vy5, vepo5, vemo5); const __m256 vey6 = _mm256_fnmadd_ps(vy6, vepo6, vemo6); const __m256 vey7 = _mm256_fnmadd_ps(vy7, vepo7, vemo7); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy5 = _mm256_fmadd_ps(vey5, vrepo5, vy5); vy6 = _mm256_fmadd_ps(vey6, vrepo6, vy6); vy7 = _mm256_fmadd_ps(vey7, vrepo7, vy7); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); output += 64; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
17,879
43.257426
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x72.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x72( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 72 * sizeof(float); batch -= 72 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); input += 72; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); vz8 = _mm256_max_ps(vsat_cutoff, vz8); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); __m256 vrepo8 = _mm256_rcp_ps(vepo8); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); const __m256 verepo8 = _mm256_fnmsub_ps(vrepo8, vepo8, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); vrepo8 = _mm256_fmadd_ps(verepo8, vrepo8, vrepo8); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); __m256 vy8 = _mm256_mul_ps(vemo8, vrepo8); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); const __m256 vey5 = _mm256_fnmadd_ps(vy5, vepo5, vemo5); const __m256 vey6 = _mm256_fnmadd_ps(vy6, vepo6, vemo6); const __m256 vey7 = _mm256_fnmadd_ps(vy7, vepo7, vemo7); const __m256 vey8 = _mm256_fnmadd_ps(vy8, vepo8, vemo8); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy5 = _mm256_fmadd_ps(vey5, vrepo5, vy5); vy6 = _mm256_fmadd_ps(vey6, vrepo6, vy6); vy7 = _mm256_fmadd_ps(vey7, vrepo7, vy7); vy8 = _mm256_fmadd_ps(vey8, vrepo8, vy8); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); output += 72; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
19,420
43.852194
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
5,438
34.782895
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-fma3-expm1minus-rr1-p6h5ts-nr1adj-x80.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/avx-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <immintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__fma3_expm1minus_rr1_p6h5ts_nr1adj_x80( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m256 vsign_mask = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sign_mask); const __m256 vsat_cutoff = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.sat_cutoff); const __m256 vlog2e = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.log2e); const __m256 vmagic_bias = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.magic_bias); const __m256 vminus_ln2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_ln2); const __m256 vc6 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c6); const __m256 vc5 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c5); const __m256 vc4 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c4); const __m256 vc3 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c3); const __m256 vc2 = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.c2); const __m256 vtwo = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.two); const __m256 vminus_one = _mm256_load_ps(params->avx_expm1minus_rr1_p6h5.minus_one); for (; batch >= 80 * sizeof(float); batch -= 80 * sizeof(float)) { const __m256 vx0 = _mm256_loadu_ps(input); const __m256 vx1 = _mm256_loadu_ps(input + 8); const __m256 vx2 = _mm256_loadu_ps(input + 16); const __m256 vx3 = _mm256_loadu_ps(input + 24); const __m256 vx4 = _mm256_loadu_ps(input + 32); const __m256 vx5 = _mm256_loadu_ps(input + 40); const __m256 vx6 = _mm256_loadu_ps(input + 48); const __m256 vx7 = _mm256_loadu_ps(input + 56); const __m256 vx8 = _mm256_loadu_ps(input + 64); const __m256 vx9 = _mm256_loadu_ps(input + 72); input += 80; __m256 vz0 = _mm256_or_ps(vx0, vsign_mask); __m256 vz1 = _mm256_or_ps(vx1, vsign_mask); __m256 vz2 = _mm256_or_ps(vx2, vsign_mask); __m256 vz3 = _mm256_or_ps(vx3, vsign_mask); __m256 vz4 = _mm256_or_ps(vx4, vsign_mask); __m256 vz5 = _mm256_or_ps(vx5, vsign_mask); __m256 vz6 = _mm256_or_ps(vx6, vsign_mask); __m256 vz7 = _mm256_or_ps(vx7, vsign_mask); __m256 vz8 = _mm256_or_ps(vx8, vsign_mask); __m256 vz9 = _mm256_or_ps(vx9, vsign_mask); const __m256 vinvsignx0 = _mm256_xor_ps(vx0, vz0); vz0 = _mm256_max_ps(vsat_cutoff, vz0); const __m256 vinvsignx1 = _mm256_xor_ps(vx1, vz1); vz1 = _mm256_max_ps(vsat_cutoff, vz1); const __m256 vinvsignx2 = _mm256_xor_ps(vx2, vz2); vz2 = _mm256_max_ps(vsat_cutoff, vz2); const __m256 vinvsignx3 = _mm256_xor_ps(vx3, vz3); vz3 = _mm256_max_ps(vsat_cutoff, vz3); const __m256 vinvsignx4 = _mm256_xor_ps(vx4, vz4); vz4 = _mm256_max_ps(vsat_cutoff, vz4); const __m256 vinvsignx5 = _mm256_xor_ps(vx5, vz5); vz5 = _mm256_max_ps(vsat_cutoff, vz5); const __m256 vinvsignx6 = _mm256_xor_ps(vx6, vz6); vz6 = _mm256_max_ps(vsat_cutoff, vz6); const __m256 vinvsignx7 = _mm256_xor_ps(vx7, vz7); vz7 = _mm256_max_ps(vsat_cutoff, vz7); const __m256 vinvsignx8 = _mm256_xor_ps(vx8, vz8); vz8 = _mm256_max_ps(vsat_cutoff, vz8); const __m256 vinvsignx9 = _mm256_xor_ps(vx9, vz9); vz9 = _mm256_max_ps(vsat_cutoff, vz9); __m256 vn0 = _mm256_fmadd_ps(vz0, vlog2e, vmagic_bias); __m256 vn1 = _mm256_fmadd_ps(vz1, vlog2e, vmagic_bias); __m256 vn2 = _mm256_fmadd_ps(vz2, vlog2e, vmagic_bias); __m256 vn3 = _mm256_fmadd_ps(vz3, vlog2e, vmagic_bias); __m256 vn4 = _mm256_fmadd_ps(vz4, vlog2e, vmagic_bias); __m256 vn5 = _mm256_fmadd_ps(vz5, vlog2e, vmagic_bias); __m256 vn6 = _mm256_fmadd_ps(vz6, vlog2e, vmagic_bias); __m256 vn7 = _mm256_fmadd_ps(vz7, vlog2e, vmagic_bias); __m256 vn8 = _mm256_fmadd_ps(vz8, vlog2e, vmagic_bias); __m256 vn9 = _mm256_fmadd_ps(vz9, vlog2e, vmagic_bias); const __m128 vn0_hi = _mm256_extractf128_ps(vn0, 1); const __m128 vn1_hi = _mm256_extractf128_ps(vn1, 1); const __m128 vn2_hi = _mm256_extractf128_ps(vn2, 1); const __m128 vn3_hi = _mm256_extractf128_ps(vn3, 1); const __m128 vn4_hi = _mm256_extractf128_ps(vn4, 1); const __m128 vn5_hi = _mm256_extractf128_ps(vn5, 1); const __m128 vn6_hi = _mm256_extractf128_ps(vn6, 1); const __m128 vn7_hi = _mm256_extractf128_ps(vn7, 1); const __m128 vn8_hi = _mm256_extractf128_ps(vn8, 1); const __m128 vn9_hi = _mm256_extractf128_ps(vn9, 1); __m256 vs0 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn0)), 23))); const __m128 vs0_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0_hi), 23)); __m256 vs1 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn1)), 23))); const __m128 vs1_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn1_hi), 23)); __m256 vs2 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn2)), 23))); const __m128 vs2_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn2_hi), 23)); __m256 vs3 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn3)), 23))); const __m128 vs3_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn3_hi), 23)); __m256 vs4 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn4)), 23))); const __m128 vs4_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4_hi), 23)); __m256 vs5 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn5)), 23))); const __m128 vs5_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn5_hi), 23)); __m256 vs6 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn6)), 23))); const __m128 vs6_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn6_hi), 23)); __m256 vs7 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn7)), 23))); const __m128 vs7_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn7_hi), 23)); __m256 vs8 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn8)), 23))); const __m128 vs8_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn8_hi), 23)); __m256 vs9 = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn9)), 23))); const __m128 vs9_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn9_hi), 23)); vs0 = _mm256_insertf128_ps(vs0, vs0_hi, 1); vs1 = _mm256_insertf128_ps(vs1, vs1_hi, 1); vs2 = _mm256_insertf128_ps(vs2, vs2_hi, 1); vs3 = _mm256_insertf128_ps(vs3, vs3_hi, 1); vs4 = _mm256_insertf128_ps(vs4, vs4_hi, 1); vs5 = _mm256_insertf128_ps(vs5, vs5_hi, 1); vs6 = _mm256_insertf128_ps(vs6, vs6_hi, 1); vs7 = _mm256_insertf128_ps(vs7, vs7_hi, 1); vs8 = _mm256_insertf128_ps(vs8, vs8_hi, 1); vs9 = _mm256_insertf128_ps(vs9, vs9_hi, 1); vn0 = _mm256_sub_ps(vn0, vmagic_bias); vn1 = _mm256_sub_ps(vn1, vmagic_bias); vn2 = _mm256_sub_ps(vn2, vmagic_bias); vn3 = _mm256_sub_ps(vn3, vmagic_bias); vn4 = _mm256_sub_ps(vn4, vmagic_bias); vn5 = _mm256_sub_ps(vn5, vmagic_bias); vn6 = _mm256_sub_ps(vn6, vmagic_bias); vn7 = _mm256_sub_ps(vn7, vmagic_bias); vn8 = _mm256_sub_ps(vn8, vmagic_bias); vn9 = _mm256_sub_ps(vn9, vmagic_bias); const __m256 vt0 = _mm256_fmadd_ps(vn0, vminus_ln2, vz0); const __m256 vt1 = _mm256_fmadd_ps(vn1, vminus_ln2, vz1); const __m256 vt2 = _mm256_fmadd_ps(vn2, vminus_ln2, vz2); const __m256 vt3 = _mm256_fmadd_ps(vn3, vminus_ln2, vz3); const __m256 vt4 = _mm256_fmadd_ps(vn4, vminus_ln2, vz4); const __m256 vt5 = _mm256_fmadd_ps(vn5, vminus_ln2, vz5); const __m256 vt6 = _mm256_fmadd_ps(vn6, vminus_ln2, vz6); const __m256 vt7 = _mm256_fmadd_ps(vn7, vminus_ln2, vz7); const __m256 vt8 = _mm256_fmadd_ps(vn8, vminus_ln2, vz8); const __m256 vt9 = _mm256_fmadd_ps(vn9, vminus_ln2, vz9); __m256 vp0 = vc6; __m256 vp1 = vc6; __m256 vp2 = vc6; __m256 vp3 = vc6; __m256 vp4 = vc6; __m256 vp5 = vc6; __m256 vp6 = vc6; __m256 vp7 = vc6; __m256 vp8 = vc6; __m256 vp9 = vc6; vp0 = _mm256_fmadd_ps(vp0, vt0, vc5); vp1 = _mm256_fmadd_ps(vp1, vt1, vc5); vp2 = _mm256_fmadd_ps(vp2, vt2, vc5); vp3 = _mm256_fmadd_ps(vp3, vt3, vc5); vp4 = _mm256_fmadd_ps(vp4, vt4, vc5); vp5 = _mm256_fmadd_ps(vp5, vt5, vc5); vp6 = _mm256_fmadd_ps(vp6, vt6, vc5); vp7 = _mm256_fmadd_ps(vp7, vt7, vc5); vp8 = _mm256_fmadd_ps(vp8, vt8, vc5); vp9 = _mm256_fmadd_ps(vp9, vt9, vc5); vp0 = _mm256_fmadd_ps(vp0, vt0, vc4); vp1 = _mm256_fmadd_ps(vp1, vt1, vc4); vp2 = _mm256_fmadd_ps(vp2, vt2, vc4); vp3 = _mm256_fmadd_ps(vp3, vt3, vc4); vp4 = _mm256_fmadd_ps(vp4, vt4, vc4); vp5 = _mm256_fmadd_ps(vp5, vt5, vc4); vp6 = _mm256_fmadd_ps(vp6, vt6, vc4); vp7 = _mm256_fmadd_ps(vp7, vt7, vc4); vp8 = _mm256_fmadd_ps(vp8, vt8, vc4); vp9 = _mm256_fmadd_ps(vp9, vt9, vc4); vp0 = _mm256_fmadd_ps(vp0, vt0, vc3); vp1 = _mm256_fmadd_ps(vp1, vt1, vc3); vp2 = _mm256_fmadd_ps(vp2, vt2, vc3); vp3 = _mm256_fmadd_ps(vp3, vt3, vc3); vp4 = _mm256_fmadd_ps(vp4, vt4, vc3); vp5 = _mm256_fmadd_ps(vp5, vt5, vc3); vp6 = _mm256_fmadd_ps(vp6, vt6, vc3); vp7 = _mm256_fmadd_ps(vp7, vt7, vc3); vp8 = _mm256_fmadd_ps(vp8, vt8, vc3); vp9 = _mm256_fmadd_ps(vp9, vt9, vc3); vp0 = _mm256_fmadd_ps(vp0, vt0, vc2); vp1 = _mm256_fmadd_ps(vp1, vt1, vc2); vp2 = _mm256_fmadd_ps(vp2, vt2, vc2); vp3 = _mm256_fmadd_ps(vp3, vt3, vc2); vp4 = _mm256_fmadd_ps(vp4, vt4, vc2); vp5 = _mm256_fmadd_ps(vp5, vt5, vc2); vp6 = _mm256_fmadd_ps(vp6, vt6, vc2); vp7 = _mm256_fmadd_ps(vp7, vt7, vc2); vp8 = _mm256_fmadd_ps(vp8, vt8, vc2); vp9 = _mm256_fmadd_ps(vp9, vt9, vc2); vp0 = _mm256_fmadd_ps(vp0, vt0, vtwo); vp1 = _mm256_fmadd_ps(vp1, vt1, vtwo); vp2 = _mm256_fmadd_ps(vp2, vt2, vtwo); vp3 = _mm256_fmadd_ps(vp3, vt3, vtwo); vp4 = _mm256_fmadd_ps(vp4, vt4, vtwo); vp5 = _mm256_fmadd_ps(vp5, vt5, vtwo); vp6 = _mm256_fmadd_ps(vp6, vt6, vtwo); vp7 = _mm256_fmadd_ps(vp7, vt7, vtwo); vp8 = _mm256_fmadd_ps(vp8, vt8, vtwo); vp9 = _mm256_fmadd_ps(vp9, vt9, vtwo); const __m256 vts0 = _mm256_mul_ps(vt0, vs0); const __m256 vsmo0 = _mm256_add_ps(vs0, vminus_one); const __m256 vts1 = _mm256_mul_ps(vt1, vs1); const __m256 vsmo1 = _mm256_add_ps(vs1, vminus_one); const __m256 vts2 = _mm256_mul_ps(vt2, vs2); const __m256 vsmo2 = _mm256_add_ps(vs2, vminus_one); const __m256 vts3 = _mm256_mul_ps(vt3, vs3); const __m256 vsmo3 = _mm256_add_ps(vs3, vminus_one); const __m256 vts4 = _mm256_mul_ps(vt4, vs4); const __m256 vsmo4 = _mm256_add_ps(vs4, vminus_one); const __m256 vts5 = _mm256_mul_ps(vt5, vs5); const __m256 vsmo5 = _mm256_add_ps(vs5, vminus_one); const __m256 vts6 = _mm256_mul_ps(vt6, vs6); const __m256 vsmo6 = _mm256_add_ps(vs6, vminus_one); const __m256 vts7 = _mm256_mul_ps(vt7, vs7); const __m256 vsmo7 = _mm256_add_ps(vs7, vminus_one); const __m256 vts8 = _mm256_mul_ps(vt8, vs8); const __m256 vsmo8 = _mm256_add_ps(vs8, vminus_one); const __m256 vts9 = _mm256_mul_ps(vt9, vs9); const __m256 vsmo9 = _mm256_add_ps(vs9, vminus_one); const __m256 vemo0 = _mm256_fmadd_ps(vp0, vts0, vsmo0); const __m256 vemo1 = _mm256_fmadd_ps(vp1, vts1, vsmo1); const __m256 vemo2 = _mm256_fmadd_ps(vp2, vts2, vsmo2); const __m256 vemo3 = _mm256_fmadd_ps(vp3, vts3, vsmo3); const __m256 vemo4 = _mm256_fmadd_ps(vp4, vts4, vsmo4); const __m256 vemo5 = _mm256_fmadd_ps(vp5, vts5, vsmo5); const __m256 vemo6 = _mm256_fmadd_ps(vp6, vts6, vsmo6); const __m256 vemo7 = _mm256_fmadd_ps(vp7, vts7, vsmo7); const __m256 vemo8 = _mm256_fmadd_ps(vp8, vts8, vsmo8); const __m256 vemo9 = _mm256_fmadd_ps(vp9, vts9, vsmo9); const __m256 vepo0 = _mm256_add_ps(vemo0, vtwo); const __m256 vepo1 = _mm256_add_ps(vemo1, vtwo); const __m256 vepo2 = _mm256_add_ps(vemo2, vtwo); const __m256 vepo3 = _mm256_add_ps(vemo3, vtwo); const __m256 vepo4 = _mm256_add_ps(vemo4, vtwo); const __m256 vepo5 = _mm256_add_ps(vemo5, vtwo); const __m256 vepo6 = _mm256_add_ps(vemo6, vtwo); const __m256 vepo7 = _mm256_add_ps(vemo7, vtwo); const __m256 vepo8 = _mm256_add_ps(vemo8, vtwo); const __m256 vepo9 = _mm256_add_ps(vemo9, vtwo); __m256 vrepo0 = _mm256_rcp_ps(vepo0); __m256 vrepo1 = _mm256_rcp_ps(vepo1); __m256 vrepo2 = _mm256_rcp_ps(vepo2); __m256 vrepo3 = _mm256_rcp_ps(vepo3); __m256 vrepo4 = _mm256_rcp_ps(vepo4); __m256 vrepo5 = _mm256_rcp_ps(vepo5); __m256 vrepo6 = _mm256_rcp_ps(vepo6); __m256 vrepo7 = _mm256_rcp_ps(vepo7); __m256 vrepo8 = _mm256_rcp_ps(vepo8); __m256 vrepo9 = _mm256_rcp_ps(vepo9); const __m256 verepo0 = _mm256_fnmsub_ps(vrepo0, vepo0, vminus_one); const __m256 verepo1 = _mm256_fnmsub_ps(vrepo1, vepo1, vminus_one); const __m256 verepo2 = _mm256_fnmsub_ps(vrepo2, vepo2, vminus_one); const __m256 verepo3 = _mm256_fnmsub_ps(vrepo3, vepo3, vminus_one); const __m256 verepo4 = _mm256_fnmsub_ps(vrepo4, vepo4, vminus_one); const __m256 verepo5 = _mm256_fnmsub_ps(vrepo5, vepo5, vminus_one); const __m256 verepo6 = _mm256_fnmsub_ps(vrepo6, vepo6, vminus_one); const __m256 verepo7 = _mm256_fnmsub_ps(vrepo7, vepo7, vminus_one); const __m256 verepo8 = _mm256_fnmsub_ps(vrepo8, vepo8, vminus_one); const __m256 verepo9 = _mm256_fnmsub_ps(vrepo9, vepo9, vminus_one); vrepo0 = _mm256_fmadd_ps(verepo0, vrepo0, vrepo0); vrepo1 = _mm256_fmadd_ps(verepo1, vrepo1, vrepo1); vrepo2 = _mm256_fmadd_ps(verepo2, vrepo2, vrepo2); vrepo3 = _mm256_fmadd_ps(verepo3, vrepo3, vrepo3); vrepo4 = _mm256_fmadd_ps(verepo4, vrepo4, vrepo4); vrepo5 = _mm256_fmadd_ps(verepo5, vrepo5, vrepo5); vrepo6 = _mm256_fmadd_ps(verepo6, vrepo6, vrepo6); vrepo7 = _mm256_fmadd_ps(verepo7, vrepo7, vrepo7); vrepo8 = _mm256_fmadd_ps(verepo8, vrepo8, vrepo8); vrepo9 = _mm256_fmadd_ps(verepo9, vrepo9, vrepo9); __m256 vy0 = _mm256_mul_ps(vemo0, vrepo0); __m256 vy1 = _mm256_mul_ps(vemo1, vrepo1); __m256 vy2 = _mm256_mul_ps(vemo2, vrepo2); __m256 vy3 = _mm256_mul_ps(vemo3, vrepo3); __m256 vy4 = _mm256_mul_ps(vemo4, vrepo4); __m256 vy5 = _mm256_mul_ps(vemo5, vrepo5); __m256 vy6 = _mm256_mul_ps(vemo6, vrepo6); __m256 vy7 = _mm256_mul_ps(vemo7, vrepo7); __m256 vy8 = _mm256_mul_ps(vemo8, vrepo8); __m256 vy9 = _mm256_mul_ps(vemo9, vrepo9); const __m256 vey0 = _mm256_fnmadd_ps(vy0, vepo0, vemo0); const __m256 vey1 = _mm256_fnmadd_ps(vy1, vepo1, vemo1); const __m256 vey2 = _mm256_fnmadd_ps(vy2, vepo2, vemo2); const __m256 vey3 = _mm256_fnmadd_ps(vy3, vepo3, vemo3); const __m256 vey4 = _mm256_fnmadd_ps(vy4, vepo4, vemo4); const __m256 vey5 = _mm256_fnmadd_ps(vy5, vepo5, vemo5); const __m256 vey6 = _mm256_fnmadd_ps(vy6, vepo6, vemo6); const __m256 vey7 = _mm256_fnmadd_ps(vy7, vepo7, vemo7); const __m256 vey8 = _mm256_fnmadd_ps(vy8, vepo8, vemo8); const __m256 vey9 = _mm256_fnmadd_ps(vy9, vepo9, vemo9); vy0 = _mm256_fmadd_ps(vey0, vrepo0, vy0); vy1 = _mm256_fmadd_ps(vey1, vrepo1, vy1); vy2 = _mm256_fmadd_ps(vey2, vrepo2, vy2); vy3 = _mm256_fmadd_ps(vey3, vrepo3, vy3); vy4 = _mm256_fmadd_ps(vey4, vrepo4, vy4); vy5 = _mm256_fmadd_ps(vey5, vrepo5, vy5); vy6 = _mm256_fmadd_ps(vey6, vrepo6, vy6); vy7 = _mm256_fmadd_ps(vey7, vrepo7, vy7); vy8 = _mm256_fmadd_ps(vey8, vrepo8, vy8); vy9 = _mm256_fmadd_ps(vey9, vrepo9, vy9); vy0 = _mm256_xor_ps(vy0, vinvsignx0); vy1 = _mm256_xor_ps(vy1, vinvsignx1); vy2 = _mm256_xor_ps(vy2, vinvsignx2); vy3 = _mm256_xor_ps(vy3, vinvsignx3); vy4 = _mm256_xor_ps(vy4, vinvsignx4); vy5 = _mm256_xor_ps(vy5, vinvsignx5); vy6 = _mm256_xor_ps(vy6, vinvsignx6); vy7 = _mm256_xor_ps(vy7, vinvsignx7); vy8 = _mm256_xor_ps(vy8, vinvsignx8); vy9 = _mm256_xor_ps(vy9, vinvsignx9); _mm256_storeu_ps(output, vy0); _mm256_storeu_ps(output + 8, vy1); _mm256_storeu_ps(output + 16, vy2); _mm256_storeu_ps(output + 24, vy3); _mm256_storeu_ps(output + 32, vy4); _mm256_storeu_ps(output + 40, vy5); _mm256_storeu_ps(output + 48, vy6); _mm256_storeu_ps(output + 56, vy7); _mm256_storeu_ps(output + 64, vy8); _mm256_storeu_ps(output + 72, vy9); output += 80; } for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); input += 8; __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); _mm256_storeu_ps(output, vy); output += 8; } if XNN_UNLIKELY(batch != 0) { assert(batch >= 1 * sizeof(float)); assert(batch <= 7 * sizeof(float)); const __m256i vmask = _mm256_loadu_si256((const __m256i*) ((uintptr_t) &params->avx_expm1minus_rr1_p6h5.mask_table[7] - batch)); const __m256 vx = _mm256_maskload_ps(input, vmask); __m256 vz = _mm256_or_ps(vx, vsign_mask); const __m256 vinvsignx = _mm256_xor_ps(vx, vz); vz = _mm256_max_ps(vsat_cutoff, vz); __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias); const __m128 vn_hi = _mm256_extractf128_ps(vn, 1); __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23))); const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23)); vs = _mm256_insertf128_ps(vs, vs_hi, 1); vn = _mm256_sub_ps(vn, vmagic_bias); const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz); __m256 vp = vc6; vp = _mm256_fmadd_ps(vp, vt, vc5); vp = _mm256_fmadd_ps(vp, vt, vc4); vp = _mm256_fmadd_ps(vp, vt, vc3); vp = _mm256_fmadd_ps(vp, vt, vc2); vp = _mm256_fmadd_ps(vp, vt, vtwo); const __m256 vts = _mm256_mul_ps(vt, vs); const __m256 vsmo = _mm256_add_ps(vs, vminus_one); const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo); const __m256 vepo = _mm256_add_ps(vemo, vtwo); __m256 vrepo = _mm256_rcp_ps(vepo); const __m256 verepo = _mm256_fnmsub_ps(vrepo, vepo, vminus_one); vrepo = _mm256_fmadd_ps(verepo, vrepo, vrepo); __m256 vy = _mm256_mul_ps(vemo, vrepo); const __m256 vey = _mm256_fnmadd_ps(vy, vepo, vemo); vy = _mm256_fmadd_ps(vey, vrepo, vy); vy = _mm256_xor_ps(vy, vinvsignx); __m128 vy_lo = _mm256_castps256_ps128(vy); if (batch & (4 * sizeof(float))) { _mm_storeu_ps(output, vy_lo); vy_lo = _mm256_extractf128_ps(vy, 1); output += 4; } if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy_lo); vy_lo = _mm_movehl_ps(vy_lo, vy_lo); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy_lo); } } }
20,961
44.372294
132
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neon-expm1minus-rr1-p6h5ts-nr2recps-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neon_expm1minus_rr1_p6h5ts_nr2recps_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vmlaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vmlaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vmlaq_f32(vmagic_bias, vz89AB, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vmlaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vmlaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vmlaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vmlaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vmlaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vmlaq_f32(vc5, vc6, vt89AB); vp0123 = vmlaq_f32(vc4, vp0123, vt0123); vp0123 = vmlaq_f32(vc3, vp0123, vt0123); vp0123 = vmlaq_f32(vc2, vp0123, vt0123); vp4567 = vmlaq_f32(vc4, vp4567, vt4567); vp4567 = vmlaq_f32(vc3, vp4567, vt4567); vp4567 = vmlaq_f32(vc2, vp4567, vt4567); vp89AB = vmlaq_f32(vc4, vp89AB, vt89AB); vp89AB = vmlaq_f32(vc3, vp89AB, vt89AB); vp89AB = vmlaq_f32(vc2, vp89AB, vt89AB); vp0123 = vmlsq_f32(vtwo, vp0123, vt0123); vp4567 = vmlsq_f32(vtwo, vp4567, vt4567); vp89AB = vmlsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vmlsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vmlsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vmlsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
8,293
36.529412
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neon-expm1minus-rr1-p6h5ts-nr2recps-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neon_expm1minus_rr1_p6h5ts_nr2recps_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vmlaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vmlaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vmlaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vmlaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); const float32x4_t vsCDEF = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vnCDEF), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vmlaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vmlaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vmlaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vmlaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vmlaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vmlaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vmlaq_f32(vc5, vc6, vt89AB); float32x4_t vpCDEF = vmlaq_f32(vc5, vc6, vtCDEF); vp0123 = vmlaq_f32(vc4, vp0123, vt0123); vp0123 = vmlaq_f32(vc3, vp0123, vt0123); vp0123 = vmlaq_f32(vc2, vp0123, vt0123); vp4567 = vmlaq_f32(vc4, vp4567, vt4567); vp4567 = vmlaq_f32(vc3, vp4567, vt4567); vp4567 = vmlaq_f32(vc2, vp4567, vt4567); vp89AB = vmlaq_f32(vc4, vp89AB, vt89AB); vp89AB = vmlaq_f32(vc3, vp89AB, vt89AB); vp89AB = vmlaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vmlaq_f32(vc4, vpCDEF, vtCDEF); vpCDEF = vmlaq_f32(vc3, vpCDEF, vtCDEF); vpCDEF = vmlaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vmlsq_f32(vtwo, vp0123, vt0123); vp4567 = vmlsq_f32(vtwo, vp4567, vt4567); vp89AB = vmlsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vmlsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vmlsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vmlsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vmlsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vmlsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,626
38.293878
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neon-expm1minus-rr1-p6h5ts-nr2recps-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neon_expm1minus_rr1_p6h5ts_nr2recps_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
4,208
30.177778
96
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neon-expm1minus-rr1-p6h5ts-nr2recps-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neon_expm1minus_rr1_p6h5ts_nr2recps_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vmlaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vmlaq_f32(vmagic_bias, vz4567, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vmlaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vmlaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vmlaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vmlaq_f32(vc5, vc6, vt4567); vp0123 = vmlaq_f32(vc4, vp0123, vt0123); vp0123 = vmlaq_f32(vc3, vp0123, vt0123); vp0123 = vmlaq_f32(vc2, vp0123, vt0123); vp4567 = vmlaq_f32(vc4, vp4567, vt4567); vp4567 = vmlaq_f32(vc3, vp4567, vt4567); vp4567 = vmlaq_f32(vc2, vp4567, vt4567); vp0123 = vmlsq_f32(vtwo, vp0123, vt0123); vp4567 = vmlsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vmlsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vmlsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vmlaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vmlaq_f32(vz, vn, vln2); float32x4_t vp = vmlaq_f32(vc5, vc6, vt); vp = vmlaq_f32(vc4, vp, vt); vp = vmlaq_f32(vc3, vp, vt); vp = vmlaq_f32(vc2, vp, vt); vp = vmlsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vmlsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
6,957
34.319797
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr1recps1fma-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr1recps1fma_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint32x4_t ve89AB = vshlq_n_u32(vreinterpretq_u32_f32(vn89AB), 20); const uint64x2_t vidx89AB = vandq_u64(vreinterpretq_u64_f32(vn89AB), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); const uint64_t vidx89 = vgetq_lane_u64(vidx89AB, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl89 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const uint64_t vidxAB = vgetq_lane_u64(vidx89AB, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); uint32x2_t vlAB = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); vl89 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], vl89, 1); vlAB = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], vlAB, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const uint32x4_t vl89AB = vcombine_u32(vl89, vlAB); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); const float32x4_t vs89AB = vreinterpretq_f32_u32(vaddq_u32(vl89AB, ve89AB)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); float32x4_t vp89AB = vfmaq_f32(vc3, vc4, vt89AB); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
11,323
43.234375
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr1recps1fma-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr1recps1fma_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vfmaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint32x4_t ve89AB = vshlq_n_u32(vreinterpretq_u32_f32(vn89AB), 20); const uint64x2_t vidx89AB = vandq_u64(vreinterpretq_u64_f32(vn89AB), vindex_mask); const uint32x4_t veCDEF = vshlq_n_u32(vreinterpretq_u32_f32(vnCDEF), 20); const uint64x2_t vidxCDEF = vandq_u64(vreinterpretq_u64_f32(vnCDEF), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); const uint64_t vidx89 = vgetq_lane_u64(vidx89AB, 0); const uint64_t vidxCD = vgetq_lane_u64(vidxCDEF, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl89 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const uint64_t vidxAB = vgetq_lane_u64(vidx89AB, 1); uint32x2_t vlCD = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); const uint64_t vidxEF = vgetq_lane_u64(vidxCDEF, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); uint32x2_t vlAB = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); uint32x2_t vlEF = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); vl89 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], vl89, 1); vlAB = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], vlAB, 1); vlCD = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)], vlCD, 1); vlEF = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)], vlEF, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const uint32x4_t vl89AB = vcombine_u32(vl89, vlAB); const uint32x4_t vlCDEF = vcombine_u32(vlCD, vlEF); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); const float32x4_t vs89AB = vreinterpretq_f32_u32(vaddq_u32(vl89AB, ve89AB)); const float32x4_t vsCDEF = vreinterpretq_f32_u32(vaddq_u32(vlCDEF, veCDEF)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vfmaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); float32x4_t vp89AB = vfmaq_f32(vc3, vc4, vt89AB); float32x4_t vpCDEF = vfmaq_f32(vc3, vc4, vtCDEF); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vfmaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vfmsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vfmsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
13,253
45.181185
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr1recps1fma-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr1recps1fma_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
5,447
35.563758
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr1recps1fma-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr1recps1fma_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,390
40.737778
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr2fma-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr2fma_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint32x4_t ve89AB = vshlq_n_u32(vreinterpretq_u32_f32(vn89AB), 20); const uint64x2_t vidx89AB = vandq_u64(vreinterpretq_u64_f32(vn89AB), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); const uint64_t vidx89 = vgetq_lane_u64(vidx89AB, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl89 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const uint64_t vidxAB = vgetq_lane_u64(vidx89AB, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); uint32x2_t vlAB = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); vl89 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], vl89, 1); vlAB = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], vlAB, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const uint32x4_t vl89AB = vcombine_u32(vl89, vlAB); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); const float32x4_t vs89AB = vreinterpretq_f32_u32(vaddq_u32(vl89AB, ve89AB)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); float32x4_t vp89AB = vfmaq_f32(vc3, vc4, vt89AB); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); float32x4_t verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
11,384
43.472656
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr2fma-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr2fma_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vfmaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint32x4_t ve89AB = vshlq_n_u32(vreinterpretq_u32_f32(vn89AB), 20); const uint64x2_t vidx89AB = vandq_u64(vreinterpretq_u64_f32(vn89AB), vindex_mask); const uint32x4_t veCDEF = vshlq_n_u32(vreinterpretq_u32_f32(vnCDEF), 20); const uint64x2_t vidxCDEF = vandq_u64(vreinterpretq_u64_f32(vnCDEF), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); const uint64_t vidx89 = vgetq_lane_u64(vidx89AB, 0); const uint64_t vidxCD = vgetq_lane_u64(vidxCDEF, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl89 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const uint64_t vidxAB = vgetq_lane_u64(vidx89AB, 1); uint32x2_t vlCD = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); const uint64_t vidxEF = vgetq_lane_u64(vidxCDEF, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); uint32x2_t vlAB = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); uint32x2_t vlEF = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); vl89 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], vl89, 1); vlAB = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], vlAB, 1); vlCD = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)], vlCD, 1); vlEF = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)], vlEF, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const uint32x4_t vl89AB = vcombine_u32(vl89, vlAB); const uint32x4_t vlCDEF = vcombine_u32(vlCD, vlEF); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); const float32x4_t vs89AB = vreinterpretq_f32_u32(vaddq_u32(vl89AB, ve89AB)); const float32x4_t vsCDEF = vreinterpretq_f32_u32(vaddq_u32(vlCDEF, veCDEF)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vfmaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); float32x4_t vp89AB = vfmaq_f32(vc3, vc4, vt89AB); float32x4_t vpCDEF = vfmaq_f32(vc3, vc4, vtCDEF); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vfmaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vfmsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vfmsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); float32x4_t verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
13,329
45.445993
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr2fma-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr2fma_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
5,463
35.671141
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-lut8-p4h3ts-nr2fma-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_lut8_p4h3ts_nr2fma_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.magic_bias); const uint64x2_t vindex_mask = vreinterpretq_u64_u32(vmovq_n_u32(UINT32_C(0x7))); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.ln2); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_lut8_p4h3.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); const uint32x4_t ve0123 = vshlq_n_u32(vreinterpretq_u32_f32(vn0123), 20); const uint64x2_t vidx0123 = vandq_u64(vreinterpretq_u64_f32(vn0123), vindex_mask); const uint32x4_t ve4567 = vshlq_n_u32(vreinterpretq_u32_f32(vn4567), 20); const uint64x2_t vidx4567 = vandq_u64(vreinterpretq_u64_f32(vn4567), vindex_mask); const uint64_t vidx01 = vgetq_lane_u64(vidx0123, 0); const uint64_t vidx45 = vgetq_lane_u64(vidx4567, 0); uint32x2_t vl01 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const uint64_t vidx23 = vgetq_lane_u64(vidx0123, 1); uint32x2_t vl45 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const uint64_t vidx67 = vgetq_lane_u64(vidx4567, 1); uint32x2_t vl23 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); uint32x2_t vl67 = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); vl01 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], vl01, 1); vl23 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], vl23, 1); vl45 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], vl45, 1); vl67 = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], vl67, 1); const uint32x4_t vl0123 = vcombine_u32(vl01, vl23); const uint32x4_t vl4567 = vcombine_u32(vl45, vl67); const float32x4_t vs0123 = vreinterpretq_f32_u32(vaddq_u32(vl0123, ve0123)); const float32x4_t vs4567 = vreinterpretq_f32_u32(vaddq_u32(vl4567, ve4567)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vfmaq_f32(vc3, vc4, vt0123); float32x4_t vp4567 = vfmaq_f32(vc3, vc4, vt4567); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const uint32x4_t ve = vshlq_n_u32(vreinterpretq_u32_f32(vn), 20); const uint64x2_t vidx = vandq_u64(vreinterpretq_u64_f32(vn), vindex_mask); const uint64_t vidx_lo = vgetq_lane_u64(vidx, 0); const uint64_t vidx_hi = vgetq_lane_u64(vidx, 1); uint32x2_t vl_lo = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); uint32x2_t vl_hi = vld1_dup_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); vl_lo = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], vl_lo, 1); vl_hi = vld1_lane_u32(&xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], vl_hi, 1); const uint32x4_t vl = vcombine_u32(vl_lo, vl_hi); const float32x4_t vs = vreinterpretq_f32_u32(vaddq_u32(vl, ve)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc3, vc4, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,436
40.942222
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr1recps1fma-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr1recps1fma_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
8,367
36.864253
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr1recps1fma-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr1recps1fma_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vfmaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); const float32x4_t vsCDEF = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vnCDEF), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vfmaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); float32x4_t vpCDEF = vfmaq_f32(vc5, vc6, vtCDEF); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vfmaq_f32(vc4, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc3, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vfmsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vfmsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,715
38.657143
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr1recps1fma-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr1recps1fma_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
4,237
30.392593
96
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr1recps1fma-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr1recps1fma_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
7,016
34.619289
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2fma-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2fma_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); float32x4_t verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
8,428
37.140271
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2fma-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2fma_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vfmaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); const float32x4_t vsCDEF = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vnCDEF), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vfmaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); float32x4_t vpCDEF = vfmaq_f32(vc5, vc6, vtCDEF); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vfmaq_f32(vc4, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc3, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vfmsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vfmsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); float32x4_t verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); verepo89AB = vfmsq_f32(vone, vrepo89AB, vepo89AB); verepoCDEF = vfmsq_f32(vone, vrepoCDEF, vepoCDEF); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); vrepo89AB = vfmaq_f32(vrepo89AB, vrepo89AB, verepo89AB); vrepoCDEF = vfmaq_f32(vrepoCDEF, vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,791
38.967347
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2fma-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2fma_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
4,253
30.511111
96
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2fma-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2fma_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); float32x4_t verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); verepo0123 = vfmsq_f32(vone, vrepo0123, vepo0123); verepo4567 = vfmsq_f32(vone, vrepo4567, vepo4567); vrepo0123 = vfmaq_f32(vrepo0123, vrepo0123, verepo0123); vrepo4567 = vfmaq_f32(vrepo4567, vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); verepo = vfmsq_f32(vone, vrepo, vepo); vrepo = vfmaq_f32(vrepo, vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
7,062
34.852792
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2recps-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2recps_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
8,296
36.542986
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2recps-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2recps_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; const float32x4_t vx89AB = vld1q_f32(input); input += 4; const float32x4_t vxCDEF = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); float32x4_t vz89AB = vabsq_f32(vx89AB); float32x4_t vzCDEF = vabsq_f32(vxCDEF); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); vz89AB = vminq_f32(vz89AB, vsat_cutoff); vzCDEF = vminq_f32(vzCDEF, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); float32x4_t vn89AB = vfmaq_f32(vmagic_bias, vz89AB, vminus_log2e); float32x4_t vnCDEF = vfmaq_f32(vmagic_bias, vzCDEF, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); const float32x4_t vs89AB = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn89AB), 23)); const float32x4_t vsCDEF = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vnCDEF), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); vn89AB = vsubq_f32(vn89AB, vmagic_bias); vnCDEF = vsubq_f32(vnCDEF, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); const float32x4_t vt89AB = vfmaq_f32(vz89AB, vn89AB, vln2); const float32x4_t vtCDEF = vfmaq_f32(vzCDEF, vnCDEF, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); float32x4_t vp89AB = vfmaq_f32(vc5, vc6, vt89AB); float32x4_t vpCDEF = vfmaq_f32(vc5, vc6, vtCDEF); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp89AB = vfmaq_f32(vc4, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc3, vp89AB, vt89AB); vp89AB = vfmaq_f32(vc2, vp89AB, vt89AB); vpCDEF = vfmaq_f32(vc4, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc3, vpCDEF, vtCDEF); vpCDEF = vfmaq_f32(vc2, vpCDEF, vtCDEF); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); vp89AB = vfmsq_f32(vtwo, vp89AB, vt89AB); vpCDEF = vfmsq_f32(vtwo, vpCDEF, vtCDEF); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vts89AB = vmulq_f32(vt89AB, vs89AB); const float32x4_t vsmo89AB = vsubq_f32(vs89AB, vone); const float32x4_t vtsCDEF = vmulq_f32(vtCDEF, vsCDEF); const float32x4_t vsmoCDEF = vsubq_f32(vsCDEF, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vemo89AB = vfmsq_f32(vsmo89AB, vp89AB, vts89AB); const float32x4_t vemoCDEF = vfmsq_f32(vsmoCDEF, vpCDEF, vtsCDEF); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); const float32x4_t vepo89AB = vaddq_f32(vemo89AB, vtwo); const float32x4_t vepoCDEF = vaddq_f32(vemoCDEF, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t vrepo89AB = vrecpeq_f32(vepo89AB); float32x4_t vrepoCDEF = vrecpeq_f32(vepoCDEF); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); float32x4_t verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); float32x4_t verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); verepo89AB = vrecpsq_f32(vrepo89AB, vepo89AB); verepoCDEF = vrecpsq_f32(vrepoCDEF, vepoCDEF); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); vrepo89AB = vmulq_f32(vrepo89AB, verepo89AB); vrepoCDEF = vmulq_f32(vrepoCDEF, verepoCDEF); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); float32x4_t vy89AB = vmulq_f32(vemo89AB, vrepo89AB); float32x4_t vyCDEF = vmulq_f32(vemoCDEF, vrepoCDEF); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vy89AB = vbslq_f32(vsign_mask, vx89AB, vy89AB); vyCDEF = vbslq_f32(vsign_mask, vxCDEF, vyCDEF); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; vst1q_f32(output, vy89AB); output += 4; vst1q_f32(output, vyCDEF); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
9,629
38.306122
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2recps-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2recps_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
4,211
30.2
96
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-neonfma-expm1minus-rr1-p6h5ts-nr2recps-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/tanh-neon-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <arm_neon.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/math-stubs.h> #include <xnnpack/vunary.h> #include <xnnpack/microparams.h> void xnn_f32_vtanh_ukernel__neonfma_expm1minus_rr1_p6h5ts_nr2recps_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { const float32x4_t vsat_cutoff = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.sat_cutoff); const float32x4_t vminus_log2e = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.minus_log2e); const float32x4_t vmagic_bias = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.magic_bias); const float32x4_t vln2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.ln2); const float32x4_t vc6 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c6); const float32x4_t vc5 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c5); const float32x4_t vc4 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c4); const float32x4_t vc3 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c3); const float32x4_t vc2 = vld1q_dup_f32(&params->neon_expm1minus_rr1_p6h5.c2); const float32x4_t vone = vmovq_n_f32(1.0f); const float32x4_t vtwo = vmovq_n_f32(2.0f); const uint32x4_t vsign_mask = vmovq_n_u32(UINT32_C(0x80000000)); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const float32x4_t vx0123 = vld1q_f32(input); input += 4; const float32x4_t vx4567 = vld1q_f32(input); input += 4; float32x4_t vz0123 = vabsq_f32(vx0123); float32x4_t vz4567 = vabsq_f32(vx4567); vz0123 = vminq_f32(vz0123, vsat_cutoff); vz4567 = vminq_f32(vz4567, vsat_cutoff); float32x4_t vn0123 = vfmaq_f32(vmagic_bias, vz0123, vminus_log2e); float32x4_t vn4567 = vfmaq_f32(vmagic_bias, vz4567, vminus_log2e); const float32x4_t vs0123 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn0123), 23)); const float32x4_t vs4567 = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn4567), 23)); vn0123 = vsubq_f32(vn0123, vmagic_bias); vn4567 = vsubq_f32(vn4567, vmagic_bias); const float32x4_t vt0123 = vfmaq_f32(vz0123, vn0123, vln2); const float32x4_t vt4567 = vfmaq_f32(vz4567, vn4567, vln2); float32x4_t vp0123 = vfmaq_f32(vc5, vc6, vt0123); float32x4_t vp4567 = vfmaq_f32(vc5, vc6, vt4567); vp0123 = vfmaq_f32(vc4, vp0123, vt0123); vp0123 = vfmaq_f32(vc3, vp0123, vt0123); vp0123 = vfmaq_f32(vc2, vp0123, vt0123); vp4567 = vfmaq_f32(vc4, vp4567, vt4567); vp4567 = vfmaq_f32(vc3, vp4567, vt4567); vp4567 = vfmaq_f32(vc2, vp4567, vt4567); vp0123 = vfmsq_f32(vtwo, vp0123, vt0123); vp4567 = vfmsq_f32(vtwo, vp4567, vt4567); const float32x4_t vts0123 = vmulq_f32(vt0123, vs0123); const float32x4_t vsmo0123 = vsubq_f32(vs0123, vone); const float32x4_t vts4567 = vmulq_f32(vt4567, vs4567); const float32x4_t vsmo4567 = vsubq_f32(vs4567, vone); const float32x4_t vemo0123 = vfmsq_f32(vsmo0123, vp0123, vts0123); const float32x4_t vemo4567 = vfmsq_f32(vsmo4567, vp4567, vts4567); const float32x4_t vepo0123 = vaddq_f32(vemo0123, vtwo); const float32x4_t vepo4567 = vaddq_f32(vemo4567, vtwo); float32x4_t vrepo0123 = vrecpeq_f32(vepo0123); float32x4_t vrepo4567 = vrecpeq_f32(vepo4567); float32x4_t verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); float32x4_t verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); verepo0123 = vrecpsq_f32(vrepo0123, vepo0123); verepo4567 = vrecpsq_f32(vrepo4567, vepo4567); vrepo0123 = vmulq_f32(vrepo0123, verepo0123); vrepo4567 = vmulq_f32(vrepo4567, verepo4567); float32x4_t vy0123 = vmulq_f32(vemo0123, vrepo0123); float32x4_t vy4567 = vmulq_f32(vemo4567, vrepo4567); vy0123 = vbslq_f32(vsign_mask, vx0123, vy0123); vy4567 = vbslq_f32(vsign_mask, vx4567, vy4567); vst1q_f32(output, vy0123); output += 4; vst1q_f32(output, vy4567); output += 4; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float32x4_t vx = vld1q_f32(input); input += 4; float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); vst1q_f32(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const float32x4_t vx = vld1q_f32(input); float32x4_t vz = vabsq_f32(vx); vz = vminq_f32(vz, vsat_cutoff); float32x4_t vn = vfmaq_f32(vmagic_bias, vz, vminus_log2e); const float32x4_t vs = vreinterpretq_f32_s32(vshlq_n_s32(vreinterpretq_s32_f32(vn), 23)); vn = vsubq_f32(vn, vmagic_bias); const float32x4_t vt = vfmaq_f32(vz, vn, vln2); float32x4_t vp = vfmaq_f32(vc5, vc6, vt); vp = vfmaq_f32(vc4, vp, vt); vp = vfmaq_f32(vc3, vp, vt); vp = vfmaq_f32(vc2, vp, vt); vp = vfmsq_f32(vtwo, vp, vt); const float32x4_t vts = vmulq_f32(vt, vs); const float32x4_t vsmo = vsubq_f32(vs, vone); const float32x4_t vemo = vfmsq_f32(vsmo, vp, vts); const float32x4_t vepo = vaddq_f32(vemo, vtwo); float32x4_t vrepo = vrecpeq_f32(vepo); float32x4_t verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); verepo = vrecpsq_f32(vrepo, vepo); vrepo = vmulq_f32(vrepo, verepo); float32x4_t vy = vmulq_f32(vemo, vrepo); vy = vbslq_f32(vsign_mask, vx, vy); float32x2_t vy_low = vget_low_f32(vy); if (batch & (2 * sizeof(float))) { vst1_f32(output, vy_low); output += 2; vy_low = vget_high_f32(vy); } if (batch & (1 * sizeof(float))) { vst1_lane_f32(output, vy_low, 0); } } }
6,960
34.335025
101
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-lut8-p4h3ts-div-x1.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_lut8_p4h3ts_div_x1( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc4 * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); }
2,396
27.535714
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-lut8-p4h3ts-div-x2.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_lut8_p4h3ts_div_x2( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; for (; batch >= 2 * sizeof(float); batch -= 2 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; input += 2; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); float vn0 = vz0 * vminus_log2e + vmagic_bias; float vn1 = vz1 * vminus_log2e + vmagic_bias; const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vidx0 = vb0 & vindex_mask; const uint32_t vidx1 = vb1 & vindex_mask; const uint32_t vl0 = xnn_table_exp2minus_k_over_8[vidx0]; uint32_t ve0 = vb0 << 20; const uint32_t vl1 = xnn_table_exp2minus_k_over_8[vidx1]; uint32_t ve1 = vb1 << 20; ve0 += vl0; ve1 += vl1; const float vt0 = vn0 * vln2 + vz0; const float vs0 = uint32_as_float(ve0); const float vt1 = vn1 * vln2 + vz1; const float vs1 = uint32_as_float(ve1); float vp0 = vc4 * vt0 + vc3; float vp1 = vc4 * vt1 + vc3; vp0 = vp0 * vt0 + vc2; vp1 = vp1 * vt1 + vc2; vp0 = vp0 * vt0 + vminus_two; vp1 = vp1 * vt1 + vminus_two; const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vemo0 = vp0 * vts0 + vsmo0; const float vemo1 = vp1 * vts1 + vsmo1; const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); output[0] = vy0; output[1] = vy1; output += 2; } if XNN_UNLIKELY(batch != 0) { const float vx = *input; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc4 * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output = vy; } }
4,081
27.151724
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-lut8-p4h3ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_lut8_p4h3ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_lut8_p4h3.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_lut8_p4h3.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_lut8_p4h3.magic_bias; const uint32_t vindex_mask = UINT32_C(0x7); const float vln2 = params->scalar_expm1minus_rr1_lut8_p4h3.ln2; const float vc4 = params->scalar_expm1minus_rr1_lut8_p4h3.c4; const float vc3 = params->scalar_expm1minus_rr1_lut8_p4h3.c3; const float vc2 = params->scalar_expm1minus_rr1_lut8_p4h3.c2; const float vminus_two = params->scalar_expm1minus_rr1_lut8_p4h3.minus_two; const float vone = params->scalar_expm1minus_rr1_lut8_p4h3.one; for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; const float vx2 = input[2]; const float vx3 = input[3]; input += 4; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); float vz2 = fabsf(vx2); float vz3 = fabsf(vx3); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); vz2 = math_pmin_f32(vz2, vsat_cutoff); vz3 = math_pmin_f32(vz3, vsat_cutoff); float vn0 = vz0 * vminus_log2e + vmagic_bias; float vn1 = vz1 * vminus_log2e + vmagic_bias; float vn2 = vz2 * vminus_log2e + vmagic_bias; float vn3 = vz3 * vminus_log2e + vmagic_bias; const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vb2 = float_as_uint32(vn2); vn2 -= vmagic_bias; const uint32_t vb3 = float_as_uint32(vn3); vn3 -= vmagic_bias; const uint32_t vidx0 = vb0 & vindex_mask; const uint32_t vidx1 = vb1 & vindex_mask; const uint32_t vidx2 = vb2 & vindex_mask; const uint32_t vidx3 = vb3 & vindex_mask; const uint32_t vl0 = xnn_table_exp2minus_k_over_8[vidx0]; uint32_t ve0 = vb0 << 20; const uint32_t vl1 = xnn_table_exp2minus_k_over_8[vidx1]; uint32_t ve1 = vb1 << 20; const uint32_t vl2 = xnn_table_exp2minus_k_over_8[vidx2]; uint32_t ve2 = vb2 << 20; const uint32_t vl3 = xnn_table_exp2minus_k_over_8[vidx3]; uint32_t ve3 = vb3 << 20; ve0 += vl0; ve1 += vl1; ve2 += vl2; ve3 += vl3; const float vt0 = vn0 * vln2 + vz0; const float vs0 = uint32_as_float(ve0); const float vt1 = vn1 * vln2 + vz1; const float vs1 = uint32_as_float(ve1); const float vt2 = vn2 * vln2 + vz2; const float vs2 = uint32_as_float(ve2); const float vt3 = vn3 * vln2 + vz3; const float vs3 = uint32_as_float(ve3); float vp0 = vc4 * vt0 + vc3; float vp1 = vc4 * vt1 + vc3; float vp2 = vc4 * vt2 + vc3; float vp3 = vc4 * vt3 + vc3; vp0 = vp0 * vt0 + vc2; vp1 = vp1 * vt1 + vc2; vp2 = vp2 * vt2 + vc2; vp3 = vp3 * vt3 + vc2; vp0 = vp0 * vt0 + vminus_two; vp1 = vp1 * vt1 + vminus_two; vp2 = vp2 * vt2 + vminus_two; vp3 = vp3 * vt3 + vminus_two; const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vts2 = vt2 * vs2; const float vsmo2 = vs2 - vone; const float vts3 = vt3 * vs3; const float vsmo3 = vs3 - vone; const float vemo0 = vp0 * vts0 + vsmo0; const float vemo1 = vp1 * vts1 + vsmo1; const float vemo2 = vp2 * vts2 + vsmo2; const float vemo3 = vp3 * vts3 + vsmo3; const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; const float vepo2 = vemo2 - vminus_two; const float vepo3 = vemo3 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; float vy2 = vemo2 / vepo2; float vy3 = vemo3 / vepo3; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); vy2 = copysignf(vy2, vx2); vy3 = copysignf(vy3, vx3); output[0] = vy0; output[1] = vy1; output[2] = vy2; output[3] = vy3; output += 4; } if XNN_UNLIKELY(batch != 0) { do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t vidx = vb & vindex_mask; const uint32_t vl = xnn_table_exp2minus_k_over_8[vidx]; uint32_t ve = vb << 20; ve += vl; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc4 * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); } }
5,789
29
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-p6h5ts-div-x1.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_p6h5ts_div_x1( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc6 * vt + vc5; vp = vp * vt + vc4; vp = vp * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); }
2,285
26.878049
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-p6h5ts-div-x2.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_p6h5ts_div_x2( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; for (; batch >= 2 * sizeof(float); batch -= 2 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; input += 2; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); float vn0 = vz0 * vminus_log2e + vmagic_bias; float vn1 = vz1 * vminus_log2e + vmagic_bias; const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t ve0 = vb0 << 23; const uint32_t ve1 = vb1 << 23; const float vt0 = vn0 * vln2 + vz0; const float vs0 = uint32_as_float(ve0); const float vt1 = vn1 * vln2 + vz1; const float vs1 = uint32_as_float(ve1); float vp0 = vc6 * vt0 + vc5; float vp1 = vc6 * vt1 + vc5; vp0 = vp0 * vt0 + vc4; vp1 = vp1 * vt1 + vc4; vp0 = vp0 * vt0 + vc3; vp1 = vp1 * vt1 + vc3; vp0 = vp0 * vt0 + vc2; vp1 = vp1 * vt1 + vc2; vp0 = vp0 * vt0 + vminus_two; vp1 = vp1 * vt1 + vminus_two; const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vemo0 = vp0 * vts0 + vsmo0; const float vemo1 = vp1 * vts1 + vsmo1; const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); output[0] = vy0; output[1] = vy1; output += 2; } if XNN_UNLIKELY(batch != 0) { const float vx = *input; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc6 * vt + vc5; vp = vp * vt + vc4; vp = vp * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output = vy; } }
3,840
26.633094
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-scalar-expm1minus-rr1-p6h5ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/scalar-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <math.h> #include <xnnpack/common.h> #include <xnnpack/math.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__scalar_expm1minus_rr1_p6h5ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const float vsat_cutoff = params->scalar_expm1minus_rr1_p6h5.sat_cutoff; const float vminus_log2e = params->scalar_expm1minus_rr1_p6h5.minus_log2e; const float vmagic_bias = params->scalar_expm1minus_rr1_p6h5.magic_bias; const float vln2 = params->scalar_expm1minus_rr1_p6h5.ln2; const float vc6 = params->scalar_expm1minus_rr1_p6h5.c6; const float vc5 = params->scalar_expm1minus_rr1_p6h5.c5; const float vc4 = params->scalar_expm1minus_rr1_p6h5.c4; const float vc3 = params->scalar_expm1minus_rr1_p6h5.c3; const float vc2 = params->scalar_expm1minus_rr1_p6h5.c2; const float vminus_two = params->scalar_expm1minus_rr1_p6h5.minus_two; const float vone = params->scalar_expm1minus_rr1_p6h5.one; for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const float vx0 = input[0]; const float vx1 = input[1]; const float vx2 = input[2]; const float vx3 = input[3]; input += 4; float vz0 = fabsf(vx0); float vz1 = fabsf(vx1); float vz2 = fabsf(vx2); float vz3 = fabsf(vx3); vz0 = math_pmin_f32(vz0, vsat_cutoff); vz1 = math_pmin_f32(vz1, vsat_cutoff); vz2 = math_pmin_f32(vz2, vsat_cutoff); vz3 = math_pmin_f32(vz3, vsat_cutoff); float vn0 = vz0 * vminus_log2e + vmagic_bias; float vn1 = vz1 * vminus_log2e + vmagic_bias; float vn2 = vz2 * vminus_log2e + vmagic_bias; float vn3 = vz3 * vminus_log2e + vmagic_bias; const uint32_t vb0 = float_as_uint32(vn0); vn0 -= vmagic_bias; const uint32_t vb1 = float_as_uint32(vn1); vn1 -= vmagic_bias; const uint32_t vb2 = float_as_uint32(vn2); vn2 -= vmagic_bias; const uint32_t vb3 = float_as_uint32(vn3); vn3 -= vmagic_bias; const uint32_t ve0 = vb0 << 23; const uint32_t ve1 = vb1 << 23; const uint32_t ve2 = vb2 << 23; const uint32_t ve3 = vb3 << 23; const float vt0 = vn0 * vln2 + vz0; const float vs0 = uint32_as_float(ve0); const float vt1 = vn1 * vln2 + vz1; const float vs1 = uint32_as_float(ve1); const float vt2 = vn2 * vln2 + vz2; const float vs2 = uint32_as_float(ve2); const float vt3 = vn3 * vln2 + vz3; const float vs3 = uint32_as_float(ve3); float vp0 = vc6 * vt0 + vc5; float vp1 = vc6 * vt1 + vc5; float vp2 = vc6 * vt2 + vc5; float vp3 = vc6 * vt3 + vc5; vp0 = vp0 * vt0 + vc4; vp1 = vp1 * vt1 + vc4; vp2 = vp2 * vt2 + vc4; vp3 = vp3 * vt3 + vc4; vp0 = vp0 * vt0 + vc3; vp1 = vp1 * vt1 + vc3; vp2 = vp2 * vt2 + vc3; vp3 = vp3 * vt3 + vc3; vp0 = vp0 * vt0 + vc2; vp1 = vp1 * vt1 + vc2; vp2 = vp2 * vt2 + vc2; vp3 = vp3 * vt3 + vc2; vp0 = vp0 * vt0 + vminus_two; vp1 = vp1 * vt1 + vminus_two; vp2 = vp2 * vt2 + vminus_two; vp3 = vp3 * vt3 + vminus_two; const float vts0 = vt0 * vs0; const float vsmo0 = vs0 - vone; const float vts1 = vt1 * vs1; const float vsmo1 = vs1 - vone; const float vts2 = vt2 * vs2; const float vsmo2 = vs2 - vone; const float vts3 = vt3 * vs3; const float vsmo3 = vs3 - vone; const float vemo0 = vp0 * vts0 + vsmo0; const float vemo1 = vp1 * vts1 + vsmo1; const float vemo2 = vp2 * vts2 + vsmo2; const float vemo3 = vp3 * vts3 + vsmo3; const float vepo0 = vemo0 - vminus_two; const float vepo1 = vemo1 - vminus_two; const float vepo2 = vemo2 - vminus_two; const float vepo3 = vemo3 - vminus_two; float vy0 = vemo0 / vepo0; float vy1 = vemo1 / vepo1; float vy2 = vemo2 / vepo2; float vy3 = vemo3 / vepo3; vy0 = copysignf(vy0, vx0); vy1 = copysignf(vy1, vx1); vy2 = copysignf(vy2, vx2); vy3 = copysignf(vy3, vx3); output[0] = vy0; output[1] = vy1; output[2] = vy2; output[3] = vy3; output += 4; } if XNN_UNLIKELY(batch != 0) { do { const float vx = *input++; float vz = fabsf(vx); vz = math_pmin_f32(vz, vsat_cutoff); float vn = vz * vminus_log2e + vmagic_bias; const uint32_t vb = float_as_uint32(vn); vn -= vmagic_bias; const uint32_t ve = vb << 23; const float vs = uint32_as_float(ve); const float vt = vn * vln2 + vz; float vp = vc6 * vt + vc5; vp = vp * vt + vc4; vp = vp * vt + vc3; vp = vp * vt + vc2; vp = vp * vt + vminus_two; const float vts = vt * vs; const float vsmo = vs - vone; const float vemo = vp * vts + vsmo; const float vepo = vemo - vminus_two; float vy = vemo / vepo; vy = copysignf(vy, vx); *output++ = vy; batch -= sizeof(float); } while (batch != 0); } }
5,418
28.291892
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-lut8-p4h3ts-div-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_lut8_p4h3ts_div_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)]); const __m128i vl8 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const __m128i vl9 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const __m128i vl89 = _mm_unpacklo_epi32(vl8, vl9); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)]); const __m128i vlA = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); const __m128i vlB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); const __m128i vlAB = _mm_unpacklo_epi32(vlA, vlB); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); const __m128i vl8 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx5]); const __m128i vl9 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx9]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const __m128i vl89 = _mm_unpacklo_epi32(vl8, vl9); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx6]); const __m128i vlA = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxA]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx7]); const __m128i vlB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxB]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); const __m128i vlAB = _mm_unpacklo_epi32(vlA, vlB); #endif const __m128i vl0123 = _mm_unpacklo_epi64(vl01, vl23); const __m128i vl4567 = _mm_unpacklo_epi64(vl45, vl67); const __m128i vl89AB = _mm_unpacklo_epi64(vl89, vlAB); const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
16,225
46.306122
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-lut8-p4h3ts-div-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_lut8_p4h3ts_div_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); const __m128i veCDEF = _mm_slli_epi32(_mm_castps_si128(vnCDEF), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidxCD = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); vidxCDEF = _mm_unpackhi_epi64(vidxCDEF, vidxCDEF); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); const uint64_t vidxEF = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)]); const __m128i vl8 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); const __m128i vl9 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)]); const __m128i vlC = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); const __m128i vlD = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const __m128i vl89 = _mm_unpacklo_epi32(vl8, vl9); const __m128i vlCD = _mm_unpacklo_epi32(vlC, vlD); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)]); const __m128i vlA = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB]); const __m128i vlB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)]); const __m128i vlE = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF]); const __m128i vlF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); const __m128i vlAB = _mm_unpacklo_epi32(vlA, vlB); const __m128i vlEF = _mm_unpacklo_epi32(vlE, vlF); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); const uint32_t vidxC = (uint32_t) _mm_cvtsi128_si32(vidxCDEF); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); const __m128i vl8 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); const __m128i vlC = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxC]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); const uint32_t vidxD = (uint32_t) _mm_extract_epi16(vidxCDEF, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx5]); const __m128i vl9 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx9]); const __m128i vlD = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxD]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const __m128i vl89 = _mm_unpacklo_epi32(vl8, vl9); const __m128i vlCD = _mm_unpacklo_epi32(vlC, vlD); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); const uint32_t vidxE = (uint32_t) _mm_extract_epi16(vidxCDEF, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx6]); const __m128i vlA = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxA]); const __m128i vlE = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxE]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); const uint32_t vidxF = (uint32_t) _mm_extract_epi16(vidxCDEF, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx7]); const __m128i vlB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxB]); const __m128i vlF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxF]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); const __m128i vlAB = _mm_unpacklo_epi32(vlA, vlB); const __m128i vlEF = _mm_unpacklo_epi32(vlE, vlF); #endif const __m128i vl0123 = _mm_unpacklo_epi64(vl01, vl23); const __m128i vl4567 = _mm_unpacklo_epi64(vl45, vl67); const __m128i vl89AB = _mm_unpacklo_epi64(vl89, vlAB); const __m128i vlCDEF = _mm_unpacklo_epi64(vlCD, vlEF); const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_add_epi32(vlCDEF, veCDEF)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc4, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
19,062
48.643229
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-lut8-p4h3ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_lut8_p4h3ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
7,542
40.674033
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-lut8-p4h3ts-div-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_lut8_p4h3ts_div_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); input += 8; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const __m128i vl4 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl5 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx5]); const __m128i vl01 = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl45 = _mm_unpacklo_epi32(vl4, vl5); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const __m128i vl6 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx6]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl7 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx7]); const __m128i vl23 = _mm_unpacklo_epi32(vl2, vl3); const __m128i vl67 = _mm_unpacklo_epi32(vl6, vl7); #endif const __m128i vl0123 = _mm_unpacklo_epi64(vl01, vl23); const __m128i vl4567 = _mm_unpacklo_epi64(vl45, vl67); const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); output += 8; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi]); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); const __m128i vl0 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); const __m128i vl1 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx1]); const __m128i vl_lo = _mm_unpacklo_epi32(vl0, vl1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); const __m128i vl2 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx2]); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); const __m128i vl3 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx3]); const __m128i vl_hi = _mm_unpacklo_epi32(vl2, vl3); #endif const __m128i vl = _mm_unpacklo_epi64(vl_lo, vl_hi); const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
13,385
43.324503
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-div-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_div_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
7,742
36.770732
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-div-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_div_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
8,936
38.72
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
4,051
30.905512
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-div-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_div_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); input += 8; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); output += 8; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
6,545
34.383784
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr1-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr1_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); const __m128 vm89AB = _mm_cmple_ps(vz89AB, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); const __m128 vepo89AB = _mm_sub_ps(vminus_two, vemo89AB); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); __m128 vrepo89AB = _mm_rcp_ps(vepo89AB); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_add_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); __m128 vy89AB = _mm_mul_ps(vemo89AB, vrepo89AB); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy89AB = _mm_or_ps(_mm_andnot_ps(vm89AB, vy89AB), _mm_and_ps(vminus_one, vm89AB)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
8,915
38.982063
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr1-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr1_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); const __m128 vm89AB = _mm_cmple_ps(vz89AB, vsat_cutoff); const __m128 vmCDEF = _mm_cmple_ps(vzCDEF, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); const __m128 vepo89AB = _mm_sub_ps(vminus_two, vemo89AB); const __m128 vepoCDEF = _mm_sub_ps(vminus_two, vemoCDEF); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); __m128 vrepo89AB = _mm_rcp_ps(vepo89AB); __m128 vrepoCDEF = _mm_rcp_ps(vepoCDEF); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_add_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); vrepoCDEF = _mm_mul_ps(vrepoCDEF, _mm_add_ps(_mm_mul_ps(vrepoCDEF, vepoCDEF), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); __m128 vy89AB = _mm_mul_ps(vemo89AB, vrepo89AB); __m128 vyCDEF = _mm_mul_ps(vemoCDEF, vrepoCDEF); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy89AB = _mm_or_ps(_mm_andnot_ps(vm89AB, vy89AB), _mm_and_ps(vminus_one, vm89AB)); vyCDEF = _mm_or_ps(_mm_andnot_ps(vmCDEF, vyCDEF), _mm_and_ps(vminus_one, vmCDEF)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
10,359
41.113821
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr1-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr1_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
4,473
32.140741
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr1-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr1_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); input += 8; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); output += 8; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
7,468
36.345
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr2-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr2_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); const __m128 vm89AB = _mm_cmple_ps(vz89AB, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); const __m128 vepo89AB = _mm_sub_ps(vminus_two, vemo89AB); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); __m128 vrepo89AB = _mm_rcp_ps(vepo89AB); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_add_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_sub_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_sub_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_sub_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); __m128 vy89AB = _mm_mul_ps(vemo89AB, vrepo89AB); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy89AB = _mm_or_ps(_mm_andnot_ps(vm89AB, vy89AB), _mm_and_ps(vminus_one, vm89AB)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
9,363
40.070175
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr2-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr2_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); const __m128 vm89AB = _mm_cmple_ps(vz89AB, vsat_cutoff); const __m128 vmCDEF = _mm_cmple_ps(vzCDEF, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); const __m128 vepo89AB = _mm_sub_ps(vminus_two, vemo89AB); const __m128 vepoCDEF = _mm_sub_ps(vminus_two, vemoCDEF); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); __m128 vrepo89AB = _mm_rcp_ps(vepo89AB); __m128 vrepoCDEF = _mm_rcp_ps(vepoCDEF); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_add_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); vrepoCDEF = _mm_mul_ps(vrepoCDEF, _mm_add_ps(_mm_mul_ps(vrepoCDEF, vepoCDEF), vminus_two)); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_sub_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_sub_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo89AB = _mm_mul_ps(vrepo89AB, _mm_sub_ps(_mm_mul_ps(vrepo89AB, vepo89AB), vminus_two)); vrepoCDEF = _mm_mul_ps(vrepoCDEF, _mm_sub_ps(_mm_mul_ps(vrepoCDEF, vepoCDEF), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); __m128 vy89AB = _mm_mul_ps(vemo89AB, vrepo89AB); __m128 vyCDEF = _mm_mul_ps(vemoCDEF, vrepoCDEF); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy89AB = _mm_or_ps(_mm_andnot_ps(vm89AB, vy89AB), _mm_and_ps(vminus_one, vm89AB)); vyCDEF = _mm_or_ps(_mm_andnot_ps(vmCDEF, vyCDEF), _mm_and_ps(vminus_one, vmCDEF)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
10,903
42.269841
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr2-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr2_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
4,633
32.824818
87
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse2-expm1minus-rr1-p6h5ts-nr2-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <emmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse2_expm1minus_rr1_p6h5ts_nr2_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); input += 8; const __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); const __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vm0123 = _mm_cmple_ps(vz0123, vsat_cutoff); const __m128 vm4567 = _mm_cmple_ps(vz4567, vsat_cutoff); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vepo0123 = _mm_sub_ps(vminus_two, vemo0123); const __m128 vepo4567 = _mm_sub_ps(vminus_two, vemo4567); __m128 vrepo0123 = _mm_rcp_ps(vepo0123); __m128 vrepo4567 = _mm_rcp_ps(vepo4567); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_add_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_add_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); vrepo0123 = _mm_mul_ps(vrepo0123, _mm_sub_ps(_mm_mul_ps(vrepo0123, vepo0123), vminus_two)); vrepo4567 = _mm_mul_ps(vrepo4567, _mm_sub_ps(_mm_mul_ps(vrepo4567, vepo4567), vminus_two)); __m128 vy0123 = _mm_mul_ps(vemo0123, vrepo0123); __m128 vy4567 = _mm_mul_ps(vemo4567, vrepo4567); vy0123 = _mm_or_ps(_mm_andnot_ps(vm0123, vy0123), _mm_and_ps(vminus_one, vm0123)); vy4567 = _mm_or_ps(_mm_andnot_ps(vm4567, vy4567), _mm_and_ps(vminus_one, vm4567)); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); output += 8; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); const __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); const __m128 vm = _mm_cmple_ps(vz, vsat_cutoff); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vminus_two, vemo); __m128 vrepo = _mm_rcp_ps(vepo); vrepo = _mm_mul_ps(vrepo, _mm_add_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); vrepo = _mm_mul_ps(vrepo, _mm_sub_ps(_mm_mul_ps(vrepo, vepo), vminus_two)); __m128 vy = _mm_mul_ps(vemo, vrepo); vy = _mm_or_ps(_mm_andnot_ps(vm, vy), _mm_and_ps(vminus_one, vm)); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
7,820
37.338235
95
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], 1); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23], 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], 3); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], 3); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], 3); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx5], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidx9], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx6], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxA], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx7], 3); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxB], 3); #endif const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
14,610
45.680511
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); const __m128i veCDEF = _mm_slli_epi32(_mm_castps_si128(vnCDEF), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidxCD = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); vidxCDEF = _mm_unpackhi_epi64(vidxCDEF, vidxCDEF); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); const uint64_t vidxEF = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], 1); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23], 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], 3); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], 3); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], 3); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)], 3); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); const uint32_t vidxC = (uint32_t) _mm_cvtsi128_si32(vidxCDEF); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxC]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); const uint32_t vidxD = (uint32_t) _mm_extract_epi16(vidxCDEF, 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx5], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidx9], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxD], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); const uint32_t vidxE = (uint32_t) _mm_extract_epi16(vidxCDEF, 4); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx6], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxA], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxE], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); const uint32_t vidxF = (uint32_t) _mm_extract_epi16(vidxCDEF, 6); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx7], 3); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxB], 3); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxF], 3); #endif const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_add_epi32(vlCDEF, veCDEF)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc4, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
17,148
48.137536
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x20.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x20( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 20 * sizeof(float); batch -= 20 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); const __m128 vxGHIJ = _mm_loadu_ps(input + 16); input += 20; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); __m128 vzGHIJ = _mm_or_ps(vxGHIJ, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vinvsignxGHIJ = _mm_xor_ps(vxGHIJ, vzGHIJ); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); vzGHIJ = _mm_max_ps(vsat_cutoff, vzGHIJ); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); __m128 vnGHIJ = _mm_add_ps(_mm_mul_ps(vzGHIJ, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); const __m128i veCDEF = _mm_slli_epi32(_mm_castps_si128(vnCDEF), 20); const __m128i veGHIJ = _mm_slli_epi32(_mm_castps_si128(vnGHIJ), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); __m128i vidxGHIJ = _mm_and_si128(_mm_castps_si128(vnGHIJ), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidxCD = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); vidxCDEF = _mm_unpackhi_epi64(vidxCDEF, vidxCDEF); const uint64_t vidxGH = (uint64_t) _mm_cvtsi128_si64(vidxGHIJ); vidxGHIJ = _mm_unpackhi_epi64(vidxGHIJ, vidxGHIJ); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); const uint64_t vidxEF = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); const uint64_t vidxIJ = (uint64_t) _mm_cvtsi128_si64(vidxGHIJ); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], 1); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23], 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], 3); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], 3); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], 3); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)], 3); __m128i vlGHIJ = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxGH]); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxGH >> 32)], 1); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxIJ], 2); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxIJ >> 32)], 3); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const __m128i vidxGHIJ = _mm_and_si128(_mm_castps_si128(vnGHIJ), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); const uint32_t vidxC = (uint32_t) _mm_cvtsi128_si32(vidxCDEF); const uint32_t vidxG = (uint32_t) _mm_cvtsi128_si32(vidxGHIJ); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxC]); __m128i vlGHIJ = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxG]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); const uint32_t vidxD = (uint32_t) _mm_extract_epi16(vidxCDEF, 2); const uint32_t vidxH = (uint32_t) _mm_extract_epi16(vidxGHIJ, 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx5], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidx9], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxD], 1); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxH], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); const uint32_t vidxE = (uint32_t) _mm_extract_epi16(vidxCDEF, 4); const uint32_t vidxI = (uint32_t) _mm_extract_epi16(vidxGHIJ, 4); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx6], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxA], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxE], 2); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxI], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); const uint32_t vidxF = (uint32_t) _mm_extract_epi16(vidxCDEF, 6); const uint32_t vidxJ = (uint32_t) _mm_extract_epi16(vidxGHIJ, 6); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx7], 3); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxB], 3); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxF], 3); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxJ], 3); #endif const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_add_epi32(vlCDEF, veCDEF)); const __m128 vsGHIJ = _mm_castsi128_ps(_mm_add_epi32(vlGHIJ, veGHIJ)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); vnGHIJ = _mm_sub_ps(vnGHIJ, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); const __m128 vtGHIJ = _mm_add_ps(_mm_mul_ps(vnGHIJ, vminus_ln2), vzGHIJ); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc4, vtCDEF), vc3); __m128 vpGHIJ = _mm_add_ps(_mm_mul_ps(vc4, vtGHIJ), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); vpGHIJ = _mm_sub_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vtsGHIJ = _mm_mul_ps(vtGHIJ, vsGHIJ); const __m128 vsmoGHIJ = _mm_add_ps(vsGHIJ, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vemoGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtsGHIJ), vsmoGHIJ); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); const __m128 vepoGHIJ = _mm_sub_ps(vemoGHIJ, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); __m128 vyGHIJ = _mm_div_ps(vemoGHIJ, vepoGHIJ); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); vyGHIJ = _mm_xor_ps(vyGHIJ, vinvsignxGHIJ); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); _mm_storeu_ps(output + 16, vyGHIJ); output += 20; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
19,686
50.135065
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x24.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x24( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); const __m128 vxGHIJ = _mm_loadu_ps(input + 16); const __m128 vxKLMN = _mm_loadu_ps(input + 20); input += 24; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); __m128 vzGHIJ = _mm_or_ps(vxGHIJ, vsign_mask); __m128 vzKLMN = _mm_or_ps(vxKLMN, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vinvsignxGHIJ = _mm_xor_ps(vxGHIJ, vzGHIJ); const __m128 vinvsignxKLMN = _mm_xor_ps(vxKLMN, vzKLMN); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); vzGHIJ = _mm_max_ps(vsat_cutoff, vzGHIJ); vzKLMN = _mm_max_ps(vsat_cutoff, vzKLMN); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); __m128 vnGHIJ = _mm_add_ps(_mm_mul_ps(vzGHIJ, vlog2e), vmagic_bias); __m128 vnKLMN = _mm_add_ps(_mm_mul_ps(vzKLMN, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); const __m128i ve89AB = _mm_slli_epi32(_mm_castps_si128(vn89AB), 20); const __m128i veCDEF = _mm_slli_epi32(_mm_castps_si128(vnCDEF), 20); const __m128i veGHIJ = _mm_slli_epi32(_mm_castps_si128(vnGHIJ), 20); const __m128i veKLMN = _mm_slli_epi32(_mm_castps_si128(vnKLMN), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); __m128i vidxGHIJ = _mm_and_si128(_mm_castps_si128(vnGHIJ), vindex_mask); __m128i vidxKLMN = _mm_and_si128(_mm_castps_si128(vnKLMN), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx89 = (uint64_t) _mm_cvtsi128_si64(vidx89AB); vidx89AB = _mm_unpackhi_epi64(vidx89AB, vidx89AB); const uint64_t vidxCD = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); vidxCDEF = _mm_unpackhi_epi64(vidxCDEF, vidxCDEF); const uint64_t vidxGH = (uint64_t) _mm_cvtsi128_si64(vidxGHIJ); vidxGHIJ = _mm_unpackhi_epi64(vidxGHIJ, vidxGHIJ); const uint64_t vidxKL = (uint64_t) _mm_cvtsi128_si64(vidxKLMN); vidxKLMN = _mm_unpackhi_epi64(vidxKLMN, vidxKLMN); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); const uint64_t vidxAB = (uint64_t) _mm_cvtsi128_si64(vidx89AB); const uint64_t vidxEF = (uint64_t) _mm_cvtsi128_si64(vidxCDEF); const uint64_t vidxIJ = (uint64_t) _mm_cvtsi128_si64(vidxGHIJ); const uint64_t vidxMN = (uint64_t) _mm_cvtsi128_si64(vidxKLMN); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], 1); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23], 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], 3); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], 3); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx89]); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx89 >> 32)], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxAB], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxAB >> 32)], 3); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxCD]); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxCD >> 32)], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxEF], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxEF >> 32)], 3); __m128i vlGHIJ = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxGH]); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxGH >> 32)], 1); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxIJ], 2); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxIJ >> 32)], 3); __m128i vlKLMN = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxKL]); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxKL >> 32)], 1); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidxMN], 2); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidxMN >> 32)], 3); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const __m128i vidx89AB = _mm_and_si128(_mm_castps_si128(vn89AB), vindex_mask); const __m128i vidxCDEF = _mm_and_si128(_mm_castps_si128(vnCDEF), vindex_mask); const __m128i vidxGHIJ = _mm_and_si128(_mm_castps_si128(vnGHIJ), vindex_mask); const __m128i vidxKLMN = _mm_and_si128(_mm_castps_si128(vnKLMN), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); const uint32_t vidx8 = (uint32_t) _mm_cvtsi128_si32(vidx89AB); const uint32_t vidxC = (uint32_t) _mm_cvtsi128_si32(vidxCDEF); const uint32_t vidxG = (uint32_t) _mm_cvtsi128_si32(vidxGHIJ); const uint32_t vidxK = (uint32_t) _mm_cvtsi128_si32(vidxKLMN); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); __m128i vl89AB = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx8]); __m128i vlCDEF = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxC]); __m128i vlGHIJ = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxG]); __m128i vlKLMN = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidxK]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); const uint32_t vidx9 = (uint32_t) _mm_extract_epi16(vidx89AB, 2); const uint32_t vidxD = (uint32_t) _mm_extract_epi16(vidxCDEF, 2); const uint32_t vidxH = (uint32_t) _mm_extract_epi16(vidxGHIJ, 2); const uint32_t vidxL = (uint32_t) _mm_extract_epi16(vidxKLMN, 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx5], 1); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidx9], 1); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxD], 1); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxH], 1); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[vidxL], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); const uint32_t vidxA = (uint32_t) _mm_extract_epi16(vidx89AB, 4); const uint32_t vidxE = (uint32_t) _mm_extract_epi16(vidxCDEF, 4); const uint32_t vidxI = (uint32_t) _mm_extract_epi16(vidxGHIJ, 4); const uint32_t vidxM = (uint32_t) _mm_extract_epi16(vidxKLMN, 4); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx6], 2); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxA], 2); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxE], 2); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxI], 2); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[vidxM], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); const uint32_t vidxB = (uint32_t) _mm_extract_epi16(vidx89AB, 6); const uint32_t vidxF = (uint32_t) _mm_extract_epi16(vidxCDEF, 6); const uint32_t vidxJ = (uint32_t) _mm_extract_epi16(vidxGHIJ, 6); const uint32_t vidxN = (uint32_t) _mm_extract_epi16(vidxKLMN, 6); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx7], 3); vl89AB = _mm_insert_epi32(vl89AB, (int) xnn_table_exp2minus_k_over_8[vidxB], 3); vlCDEF = _mm_insert_epi32(vlCDEF, (int) xnn_table_exp2minus_k_over_8[vidxF], 3); vlGHIJ = _mm_insert_epi32(vlGHIJ, (int) xnn_table_exp2minus_k_over_8[vidxJ], 3); vlKLMN = _mm_insert_epi32(vlKLMN, (int) xnn_table_exp2minus_k_over_8[vidxN], 3); #endif const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); const __m128 vs89AB = _mm_castsi128_ps(_mm_add_epi32(vl89AB, ve89AB)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_add_epi32(vlCDEF, veCDEF)); const __m128 vsGHIJ = _mm_castsi128_ps(_mm_add_epi32(vlGHIJ, veGHIJ)); const __m128 vsKLMN = _mm_castsi128_ps(_mm_add_epi32(vlKLMN, veKLMN)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); vnGHIJ = _mm_sub_ps(vnGHIJ, vmagic_bias); vnKLMN = _mm_sub_ps(vnKLMN, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); const __m128 vtGHIJ = _mm_add_ps(_mm_mul_ps(vnGHIJ, vminus_ln2), vzGHIJ); const __m128 vtKLMN = _mm_add_ps(_mm_mul_ps(vnKLMN, vminus_ln2), vzKLMN); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc4, vt89AB), vc3); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc4, vtCDEF), vc3); __m128 vpGHIJ = _mm_add_ps(_mm_mul_ps(vc4, vtGHIJ), vc3); __m128 vpKLMN = _mm_add_ps(_mm_mul_ps(vc4, vtKLMN), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc2); vpKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtKLMN), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); vpGHIJ = _mm_sub_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vminus_two); vpKLMN = _mm_sub_ps(_mm_mul_ps(vpKLMN, vtKLMN), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vtsGHIJ = _mm_mul_ps(vtGHIJ, vsGHIJ); const __m128 vsmoGHIJ = _mm_add_ps(vsGHIJ, vminus_one); const __m128 vtsKLMN = _mm_mul_ps(vtKLMN, vsKLMN); const __m128 vsmoKLMN = _mm_add_ps(vsKLMN, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vemoGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtsGHIJ), vsmoGHIJ); const __m128 vemoKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtsKLMN), vsmoKLMN); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); const __m128 vepoGHIJ = _mm_sub_ps(vemoGHIJ, vminus_two); const __m128 vepoKLMN = _mm_sub_ps(vemoKLMN, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); __m128 vyGHIJ = _mm_div_ps(vemoGHIJ, vepoGHIJ); __m128 vyKLMN = _mm_div_ps(vemoKLMN, vepoKLMN); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); vyGHIJ = _mm_xor_ps(vyGHIJ, vinvsignxGHIJ); vyKLMN = _mm_xor_ps(vyKLMN, vinvsignxKLMN); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); _mm_storeu_ps(output + 16, vyGHIJ); _mm_storeu_ps(output + 20, vyKLMN); output += 24; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
22,224
51.790974
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
6,829
38.94152
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-lut8-p4h3ts-div-x8.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> // Table of exp2(k / 8) values decremented (as integer) by (k << 20), k = 0..7 extern XNN_INTERNAL const uint32_t xnn_table_exp2minus_k_over_8[8]; void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_lut8_p4h3ts_div_x8( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.magic_bias); const __m128i vindex_mask = _mm_load_si128((const __m128i*) params->sse_expm1minus_rr1_lut8_p4h3.index_mask); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_ln2); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_lut8_p4h3.minus_one); for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); input += 8; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); const __m128i ve0123 = _mm_slli_epi32(_mm_castps_si128(vn0123), 20); const __m128i ve4567 = _mm_slli_epi32(_mm_castps_si128(vn4567), 20); #if XNN_ARCH_X86_64 __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const uint64_t vidx01 = (uint64_t) _mm_cvtsi128_si64(vidx0123); vidx0123 = _mm_unpackhi_epi64(vidx0123, vidx0123); const uint64_t vidx45 = (uint64_t) _mm_cvtsi128_si64(vidx4567); vidx4567 = _mm_unpackhi_epi64(vidx4567, vidx4567); const uint64_t vidx23 = (uint64_t) _mm_cvtsi128_si64(vidx0123); const uint64_t vidx67 = (uint64_t) _mm_cvtsi128_si64(vidx4567); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx01]); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx01 >> 32)], 1); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx23], 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx23 >> 32)], 3); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx45]); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx45 >> 32)], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx67], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx67 >> 32)], 3); #else const __m128i vidx0123 = _mm_and_si128(_mm_castps_si128(vn0123), vindex_mask); const __m128i vidx4567 = _mm_and_si128(_mm_castps_si128(vn4567), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx0123); const uint32_t vidx4 = (uint32_t) _mm_cvtsi128_si32(vidx4567); __m128i vl0123 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); __m128i vl4567 = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx4]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx0123, 2); const uint32_t vidx5 = (uint32_t) _mm_extract_epi16(vidx4567, 2); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx5], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx0123, 4); const uint32_t vidx6 = (uint32_t) _mm_extract_epi16(vidx4567, 4); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx6], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx0123, 6); const uint32_t vidx7 = (uint32_t) _mm_extract_epi16(vidx4567, 6); vl0123 = _mm_insert_epi32(vl0123, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); vl4567 = _mm_insert_epi32(vl4567, (int) xnn_table_exp2minus_k_over_8[vidx7], 3); #endif const __m128 vs0123 = _mm_castsi128_ps(_mm_add_epi32(vl0123, ve0123)); const __m128 vs4567 = _mm_castsi128_ps(_mm_add_epi32(vl4567, ve4567)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc4, vt0123), vc3); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc4, vt4567), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); output += 8; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128i ve = _mm_slli_epi32(_mm_castps_si128(vn), 20); #if XNN_ARCH_X86_64 __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint64_t vidx_lo = (uint64_t) _mm_cvtsi128_si64(vidx); vidx = _mm_unpackhi_epi64(vidx, vidx); const uint64_t vidx_hi = (uint64_t) _mm_cvtsi128_si64(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_lo]); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_lo >> 32)], 1); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) vidx_hi], 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[(uint32_t) (vidx_hi >> 32)], 3); #else const __m128i vidx = _mm_and_si128(_mm_castps_si128(vn), vindex_mask); const uint32_t vidx0 = (uint32_t) _mm_cvtsi128_si32(vidx); __m128i vl = _mm_cvtsi32_si128((int) xnn_table_exp2minus_k_over_8[vidx0]); const uint32_t vidx1 = (uint32_t) _mm_extract_epi16(vidx, 2); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx1], 1); const uint32_t vidx2 = (uint32_t) _mm_extract_epi16(vidx, 4); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx2], 2); const uint32_t vidx3 = (uint32_t) _mm_extract_epi16(vidx, 6); vl = _mm_insert_epi32(vl, (int) xnn_table_exp2minus_k_over_8[vidx3], 3); #endif const __m128 vs = _mm_castsi128_ps(_mm_add_epi32(vl, ve)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc4, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
12,069
42.574007
111
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-p6h5ts-div-x12.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_p6h5ts_div_x12( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 12 * sizeof(float); batch -= 12 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); input += 12; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); output += 12; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
7,743
36.77561
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-p6h5ts-div-x16.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_p6h5ts_div_x16( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 16 * sizeof(float); batch -= 16 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); input += 16; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); output += 16; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
8,937
38.724444
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-p6h5ts-div-x20.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_p6h5ts_div_x20( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 20 * sizeof(float); batch -= 20 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); const __m128 vxGHIJ = _mm_loadu_ps(input + 16); input += 20; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); __m128 vzGHIJ = _mm_or_ps(vxGHIJ, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vinvsignxGHIJ = _mm_xor_ps(vxGHIJ, vzGHIJ); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); vzGHIJ = _mm_max_ps(vsat_cutoff, vzGHIJ); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); __m128 vnGHIJ = _mm_add_ps(_mm_mul_ps(vzGHIJ, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); const __m128 vsGHIJ = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnGHIJ), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); vnGHIJ = _mm_sub_ps(vnGHIJ, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); const __m128 vtGHIJ = _mm_add_ps(_mm_mul_ps(vnGHIJ, vminus_ln2), vzGHIJ); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); __m128 vpGHIJ = _mm_add_ps(_mm_mul_ps(vc6, vtGHIJ), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); vpGHIJ = _mm_sub_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vtsGHIJ = _mm_mul_ps(vtGHIJ, vsGHIJ); const __m128 vsmoGHIJ = _mm_add_ps(vsGHIJ, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vemoGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtsGHIJ), vsmoGHIJ); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); const __m128 vepoGHIJ = _mm_sub_ps(vemoGHIJ, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); __m128 vyGHIJ = _mm_div_ps(vemoGHIJ, vepoGHIJ); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); vyGHIJ = _mm_xor_ps(vyGHIJ, vinvsignxGHIJ); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); _mm_storeu_ps(output + 16, vyGHIJ); output += 20; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
10,131
40.355102
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-p6h5ts-div-x24.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_p6h5ts_div_x24( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) { const __m128 vx0123 = _mm_loadu_ps(input); const __m128 vx4567 = _mm_loadu_ps(input + 4); const __m128 vx89AB = _mm_loadu_ps(input + 8); const __m128 vxCDEF = _mm_loadu_ps(input + 12); const __m128 vxGHIJ = _mm_loadu_ps(input + 16); const __m128 vxKLMN = _mm_loadu_ps(input + 20); input += 24; __m128 vz0123 = _mm_or_ps(vx0123, vsign_mask); __m128 vz4567 = _mm_or_ps(vx4567, vsign_mask); __m128 vz89AB = _mm_or_ps(vx89AB, vsign_mask); __m128 vzCDEF = _mm_or_ps(vxCDEF, vsign_mask); __m128 vzGHIJ = _mm_or_ps(vxGHIJ, vsign_mask); __m128 vzKLMN = _mm_or_ps(vxKLMN, vsign_mask); const __m128 vinvsignx0123 = _mm_xor_ps(vx0123, vz0123); const __m128 vinvsignx4567 = _mm_xor_ps(vx4567, vz4567); const __m128 vinvsignx89AB = _mm_xor_ps(vx89AB, vz89AB); const __m128 vinvsignxCDEF = _mm_xor_ps(vxCDEF, vzCDEF); const __m128 vinvsignxGHIJ = _mm_xor_ps(vxGHIJ, vzGHIJ); const __m128 vinvsignxKLMN = _mm_xor_ps(vxKLMN, vzKLMN); vz0123 = _mm_max_ps(vsat_cutoff, vz0123); vz4567 = _mm_max_ps(vsat_cutoff, vz4567); vz89AB = _mm_max_ps(vsat_cutoff, vz89AB); vzCDEF = _mm_max_ps(vsat_cutoff, vzCDEF); vzGHIJ = _mm_max_ps(vsat_cutoff, vzGHIJ); vzKLMN = _mm_max_ps(vsat_cutoff, vzKLMN); __m128 vn0123 = _mm_add_ps(_mm_mul_ps(vz0123, vlog2e), vmagic_bias); __m128 vn4567 = _mm_add_ps(_mm_mul_ps(vz4567, vlog2e), vmagic_bias); __m128 vn89AB = _mm_add_ps(_mm_mul_ps(vz89AB, vlog2e), vmagic_bias); __m128 vnCDEF = _mm_add_ps(_mm_mul_ps(vzCDEF, vlog2e), vmagic_bias); __m128 vnGHIJ = _mm_add_ps(_mm_mul_ps(vzGHIJ, vlog2e), vmagic_bias); __m128 vnKLMN = _mm_add_ps(_mm_mul_ps(vzKLMN, vlog2e), vmagic_bias); const __m128 vs0123 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn0123), 23)); const __m128 vs4567 = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn4567), 23)); const __m128 vs89AB = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn89AB), 23)); const __m128 vsCDEF = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnCDEF), 23)); const __m128 vsGHIJ = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnGHIJ), 23)); const __m128 vsKLMN = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vnKLMN), 23)); vn0123 = _mm_sub_ps(vn0123, vmagic_bias); vn4567 = _mm_sub_ps(vn4567, vmagic_bias); vn89AB = _mm_sub_ps(vn89AB, vmagic_bias); vnCDEF = _mm_sub_ps(vnCDEF, vmagic_bias); vnGHIJ = _mm_sub_ps(vnGHIJ, vmagic_bias); vnKLMN = _mm_sub_ps(vnKLMN, vmagic_bias); const __m128 vt0123 = _mm_add_ps(_mm_mul_ps(vn0123, vminus_ln2), vz0123); const __m128 vt4567 = _mm_add_ps(_mm_mul_ps(vn4567, vminus_ln2), vz4567); const __m128 vt89AB = _mm_add_ps(_mm_mul_ps(vn89AB, vminus_ln2), vz89AB); const __m128 vtCDEF = _mm_add_ps(_mm_mul_ps(vnCDEF, vminus_ln2), vzCDEF); const __m128 vtGHIJ = _mm_add_ps(_mm_mul_ps(vnGHIJ, vminus_ln2), vzGHIJ); const __m128 vtKLMN = _mm_add_ps(_mm_mul_ps(vnKLMN, vminus_ln2), vzKLMN); __m128 vp0123 = _mm_add_ps(_mm_mul_ps(vc6, vt0123), vc5); __m128 vp4567 = _mm_add_ps(_mm_mul_ps(vc6, vt4567), vc5); __m128 vp89AB = _mm_add_ps(_mm_mul_ps(vc6, vt89AB), vc5); __m128 vpCDEF = _mm_add_ps(_mm_mul_ps(vc6, vtCDEF), vc5); __m128 vpGHIJ = _mm_add_ps(_mm_mul_ps(vc6, vtGHIJ), vc5); __m128 vpKLMN = _mm_add_ps(_mm_mul_ps(vc6, vtKLMN), vc5); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc4); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc4); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc4); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc4); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc4); vpKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtKLMN), vc4); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc3); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc3); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc3); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc3); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc3); vpKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtKLMN), vc3); vp0123 = _mm_add_ps(_mm_mul_ps(vp0123, vt0123), vc2); vp4567 = _mm_add_ps(_mm_mul_ps(vp4567, vt4567), vc2); vp89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vt89AB), vc2); vpCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtCDEF), vc2); vpGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vc2); vpKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtKLMN), vc2); vp0123 = _mm_sub_ps(_mm_mul_ps(vp0123, vt0123), vminus_two); vp4567 = _mm_sub_ps(_mm_mul_ps(vp4567, vt4567), vminus_two); vp89AB = _mm_sub_ps(_mm_mul_ps(vp89AB, vt89AB), vminus_two); vpCDEF = _mm_sub_ps(_mm_mul_ps(vpCDEF, vtCDEF), vminus_two); vpGHIJ = _mm_sub_ps(_mm_mul_ps(vpGHIJ, vtGHIJ), vminus_two); vpKLMN = _mm_sub_ps(_mm_mul_ps(vpKLMN, vtKLMN), vminus_two); const __m128 vts0123 = _mm_mul_ps(vt0123, vs0123); const __m128 vsmo0123 = _mm_add_ps(vs0123, vminus_one); const __m128 vts4567 = _mm_mul_ps(vt4567, vs4567); const __m128 vsmo4567 = _mm_add_ps(vs4567, vminus_one); const __m128 vts89AB = _mm_mul_ps(vt89AB, vs89AB); const __m128 vsmo89AB = _mm_add_ps(vs89AB, vminus_one); const __m128 vtsCDEF = _mm_mul_ps(vtCDEF, vsCDEF); const __m128 vsmoCDEF = _mm_add_ps(vsCDEF, vminus_one); const __m128 vtsGHIJ = _mm_mul_ps(vtGHIJ, vsGHIJ); const __m128 vsmoGHIJ = _mm_add_ps(vsGHIJ, vminus_one); const __m128 vtsKLMN = _mm_mul_ps(vtKLMN, vsKLMN); const __m128 vsmoKLMN = _mm_add_ps(vsKLMN, vminus_one); const __m128 vemo0123 = _mm_add_ps(_mm_mul_ps(vp0123, vts0123), vsmo0123); const __m128 vemo4567 = _mm_add_ps(_mm_mul_ps(vp4567, vts4567), vsmo4567); const __m128 vemo89AB = _mm_add_ps(_mm_mul_ps(vp89AB, vts89AB), vsmo89AB); const __m128 vemoCDEF = _mm_add_ps(_mm_mul_ps(vpCDEF, vtsCDEF), vsmoCDEF); const __m128 vemoGHIJ = _mm_add_ps(_mm_mul_ps(vpGHIJ, vtsGHIJ), vsmoGHIJ); const __m128 vemoKLMN = _mm_add_ps(_mm_mul_ps(vpKLMN, vtsKLMN), vsmoKLMN); const __m128 vepo0123 = _mm_sub_ps(vemo0123, vminus_two); const __m128 vepo4567 = _mm_sub_ps(vemo4567, vminus_two); const __m128 vepo89AB = _mm_sub_ps(vemo89AB, vminus_two); const __m128 vepoCDEF = _mm_sub_ps(vemoCDEF, vminus_two); const __m128 vepoGHIJ = _mm_sub_ps(vemoGHIJ, vminus_two); const __m128 vepoKLMN = _mm_sub_ps(vemoKLMN, vminus_two); __m128 vy0123 = _mm_div_ps(vemo0123, vepo0123); __m128 vy4567 = _mm_div_ps(vemo4567, vepo4567); __m128 vy89AB = _mm_div_ps(vemo89AB, vepo89AB); __m128 vyCDEF = _mm_div_ps(vemoCDEF, vepoCDEF); __m128 vyGHIJ = _mm_div_ps(vemoGHIJ, vepoGHIJ); __m128 vyKLMN = _mm_div_ps(vemoKLMN, vepoKLMN); vy0123 = _mm_xor_ps(vy0123, vinvsignx0123); vy4567 = _mm_xor_ps(vy4567, vinvsignx4567); vy89AB = _mm_xor_ps(vy89AB, vinvsignx89AB); vyCDEF = _mm_xor_ps(vyCDEF, vinvsignxCDEF); vyGHIJ = _mm_xor_ps(vyGHIJ, vinvsignxGHIJ); vyKLMN = _mm_xor_ps(vyKLMN, vinvsignxKLMN); _mm_storeu_ps(output, vy0123); _mm_storeu_ps(output + 4, vy4567); _mm_storeu_ps(output + 8, vy89AB); _mm_storeu_ps(output + 12, vyCDEF); _mm_storeu_ps(output + 16, vyGHIJ); _mm_storeu_ps(output + 20, vyKLMN); output += 24; } for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
11,325
41.739623
89
c
XNNPACK
XNNPACK-master/src/f32-vtanh/gen/f32-vtanh-sse41-expm1minus-rr1-p6h5ts-div-x4.c
// Auto-generated file. Do not edit! // Template: src/f32-vtanh/sse-expm1minus.c.in // Generator: tools/xngen // // Copyright 2023 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include <assert.h> #include <stddef.h> #include <stdint.h> #include <smmintrin.h> #include <xnnpack/common.h> #include <xnnpack/microparams.h> #include <xnnpack/vunary.h> void xnn_f32_vtanh_ukernel__sse41_expm1minus_rr1_p6h5ts_div_x4( size_t batch, const float* input, float* output, const union xnn_f32_tanh_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(batch != 0); assert(batch % sizeof(float) == 0); assert(input != NULL); assert(output != NULL); const __m128 vsign_mask = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sign_mask); const __m128 vsat_cutoff = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.sat_cutoff); const __m128 vlog2e = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.log2e); const __m128 vmagic_bias = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.magic_bias); const __m128 vminus_ln2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_ln2); const __m128 vc6 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c6); const __m128 vc5 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c5); const __m128 vc4 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c4); const __m128 vc3 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c3); const __m128 vc2 = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.c2); const __m128 vminus_two = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_two); const __m128 vminus_one = _mm_load_ps(params->sse_expm1minus_rr1_p6h5.minus_one); for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) { const __m128 vx = _mm_loadu_ps(input); input += 4; __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); _mm_storeu_ps(output, vy); output += 4; } if XNN_UNLIKELY(batch != 0) { const __m128 vx = _mm_loadu_ps(input); __m128 vz = _mm_or_ps(vx, vsign_mask); const __m128 vinvsignx = _mm_xor_ps(vx, vz); vz = _mm_max_ps(vsat_cutoff, vz); __m128 vn = _mm_add_ps(_mm_mul_ps(vz, vlog2e), vmagic_bias); const __m128 vs = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn), 23)); vn = _mm_sub_ps(vn, vmagic_bias); const __m128 vt = _mm_add_ps(_mm_mul_ps(vn, vminus_ln2), vz); __m128 vp = _mm_add_ps(_mm_mul_ps(vc6, vt), vc5); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc4); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc3); vp = _mm_add_ps(_mm_mul_ps(vp, vt), vc2); vp = _mm_sub_ps(_mm_mul_ps(vp, vt), vminus_two); const __m128 vts = _mm_mul_ps(vt, vs); const __m128 vsmo = _mm_add_ps(vs, vminus_one); const __m128 vemo = _mm_add_ps(_mm_mul_ps(vp, vts), vsmo); const __m128 vepo = _mm_sub_ps(vemo, vminus_two); __m128 vy = _mm_div_ps(vemo, vepo); vy = _mm_xor_ps(vy, vinvsignx); if (batch & (2 * sizeof(float))) { _mm_storel_pi((__m64*) output, vy); vy = _mm_movehl_ps(vy, vy); output += 2; } if (batch & (1 * sizeof(float))) { _mm_store_ss(output, vy); } } }
4,052
30.913386
87
c