File size: 4,369 Bytes
be903e2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 | // Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
#version 450
#if NCNN_fp16_storage
#extension GL_EXT_shader_16bit_storage: require
#endif
#if NCNN_fp16_arithmetic
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
#endif
layout (constant_id = 0) const int op_type = 0;
layout (constant_id = 1) const int with_scalar = 0;
layout (constant_id = 2) const float const_b = 0;
#define shape_constant_id_offset 3
layout (constant_id = shape_constant_id_offset + 0) const int adims = 0;
layout (constant_id = shape_constant_id_offset + 1) const int aw = 0;
layout (constant_id = shape_constant_id_offset + 2) const int ah = 0;
layout (constant_id = shape_constant_id_offset + 3) const int ac = 0;
layout (constant_id = shape_constant_id_offset + 4) const int acstep = 0;
layout (constant_id = shape_constant_id_offset + 5) const int bdims = 0;
layout (constant_id = shape_constant_id_offset + 6) const int bw = 0;
layout (constant_id = shape_constant_id_offset + 7) const int bh = 0;
layout (constant_id = shape_constant_id_offset + 8) const int bc = 0;
layout (constant_id = shape_constant_id_offset + 9) const int bcstep = 0;
layout (constant_id = shape_constant_id_offset + 10) const int outdims = 0;
layout (constant_id = shape_constant_id_offset + 11) const int outw = 0;
layout (constant_id = shape_constant_id_offset + 12) const int outh = 0;
layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
#if NCNN_image_shader
layout (binding = 0) uniform unfp sampler3D a_blob_3d;
layout (binding = 1) uniform unfp sampler3D b_blob_3d;
layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
#else
layout (binding = 0) buffer a_blob { sfpvec4 a_blob_data[]; };
layout (binding = 1) readonly buffer b_blob { sfpvec4 b_blob_data[]; };
layout (binding = 2) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
#endif
layout (push_constant) uniform parameter
{
int adims;
int aw;
int ah;
int ac;
int acstep;
int bdims;
int bw;
int bh;
int bc;
int bcstep;
int outdims;
int outw;
int outh;
int outc;
int outcstep;
} p;
void main()
{
int gx = int(gl_GlobalInvocationID.x);
int gy = int(gl_GlobalInvocationID.y);
int gz = int(gl_GlobalInvocationID.z);
if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
return;
#if NCNN_image_shader
afpvec4 v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
#else
const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
afpvec4 v1 = buffer_ld4(a_blob_data, gi);
#endif
afpvec4 v2;
if (with_scalar == 1)
{
// scalar
v2 = afpvec4(afp(const_b));
}
else
{
// no broadcast
#if NCNN_image_shader
v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
#else
v2 = buffer_ld4(b_blob_data, gi);
#endif
}
afpvec4 res;
if (op_type == 0) res = v1 + v2;
if (op_type == 1) res = v1 - v2;
if (op_type == 2) res = v1 * v2;
if (op_type == 3) res = v1 / v2;
if (op_type == 4) res = max(v1, v2);
if (op_type == 5) res = min(v1, v2);
if (op_type == 6) res = pow(v1, v2);
if (op_type == 7) res = v2 - v1;
if (op_type == 8) res = v2 / v1;
if (op_type == 9) res = pow(v2, v1);
#if NCNN_moltenvk
if (op_type == 10) res = afpvec4(atan(vec4(v1), vec4(v2)));
if (op_type == 11) res = afpvec4(atan(vec4(v2), vec4(v1)));
#else
if (op_type == 10) res = atan(v1, v2);
if (op_type == 11) res = atan(v2, v1);
#endif
#if NCNN_image_shader
image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
#else
buffer_st4(top_blob_data, gi, res);
#endif
}
|