name
string
code
string
asm
string
file
string
ncnn::im2col_sgemm_pack8to1_int8_sse_xop(ncnn::Mat const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Option const&)
void im2col_sgemm_pack8to1_int8_sse_xop(const Mat& bottom_im2col, Mat& top_blob, const Mat& kernel, const Option& opt) { im2col_sgemm_pack8to1_int8_sse(bottom_im2col, top_blob, kernel, opt); }
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x88, %rsp movq %rcx, %rax movq %rdx, 0x10(%rsp) movq %rdi, %r12 movslq 0x2c(%rdi), %r13 movl 0x30(%rdi), %ebx movl 0x38(%rdi), %r15d movq %rsi, 0x28(%rsp) movslq 0x38(%rsi), %rcx movq %rcx, 0x18(%rsp) leaq 0x30(%rsp), %rdi andq $0x0, 0x40(%rdi) mo...
/csukuangfj[P]ncnn/src/layer/x86/convolution_x86_xop.cpp
ncnn::crop_pack4_sse(ncnn::Mat const&, ncnn::Mat&, int, int)
static void crop_pack4_sse(const Mat& src, Mat& dst, int top, int left) { int w = dst.w; int h = dst.h; int right = src.w - dst.w - left; const float* ptr = src.row(top) + left * 4; float* outptr = dst; for (int y = 0; y < h; y++) { for (int x = 0; x < w; x++) { ...
movl 0x2c(%rsi), %eax movl 0x30(%rsi), %r8d movslq 0x2c(%rdi), %r9 movslq %edx, %rdx imulq %r9, %rdx subl %eax, %r9d imulq 0x10(%rdi), %rdx addq (%rdi), %rdx shll $0x2, %ecx movslq %ecx, %rcx leaq (%rdx,%rcx,4), %rcx movq (%rsi), %rdx shll $0x2, %r9d movslq %r9d, %rsi xorl %edi, %edi testl %eax, %eax cmovlel %edi, %eax...
/csukuangfj[P]ncnn/src/layer/x86/crop_x86.cpp
ncnn::Deconvolution_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Deconvolution_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { // deconvolv with NxN kernel // value = value + bias int w = bottom_blob.w; int h = bottom_blob.h; size_t elemsize = bottom_blob.elemsize; int elempack = bottom_blob.elempack; // NCNN_LOGE(...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x258, %rsp # imm = 0x258 movq %rdx, 0x160(%rsp) movq %rdi, %r15 movq (%rdi), %rdi movq -0x18(%rdi), %rax movl 0xd4(%r15,%rax), %edx decl %edx imull 0xdc(%r15,%rax), %edx movl %edx, 0x144(%rsp) movl 0x2c(%rsi), %r13d leal -0x1(%r13), %r1...
/csukuangfj[P]ncnn/src/layer/x86/deconvolution_x86.cpp
virtual thunk to ncnn::Deconvolution_x86_avx::create_pipeline(ncnn::Option const&)
int Deconvolution_x86_avx::create_pipeline(const Option& opt) { activation = create_activation_layer(activation_type, activation_params, opt); const int maxk = kernel_w * kernel_h; int num_input = weight_data_size / maxk / num_output; int elempack = 1; int out_elempack = 1; #if __SSE2__ if (op...
pushq %rax movq (%rdi), %rax addq -0x30(%rax), %rdi callq 0x164cde xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/deconvolution_x86_avx.cpp
ncnn::Deconvolution_x86_avx::destroy_pipeline(ncnn::Option const&)
int Deconvolution_x86_avx::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } if (gemm) { gemm->destroy_pipeline(opt); delete gemm; gemm = 0; } return 0; }
pushq %r14 pushq %rbx pushq %rax movq %rsi, %r14 movq %rdi, %rbx movq 0x8(%rdi), %rdi testq %rdi, %rdi je 0x165840 movq (%rdi), %rax movq %r14, %rsi callq *0x28(%rax) movq 0x8(%rbx), %rdi testq %rdi, %rdi je 0x16583b movq (%rdi), %rax callq *0x8(%rax) andq $0x0, 0x8(%rbx) movq 0x10(%rbx), %rdi testq %rdi, %rdi je 0x165...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/deconvolution_x86_avx.cpp
ncnn::InnerProduct_x86::create_pipeline_int8_x86(ncnn::Option const&)
int InnerProduct_x86::create_pipeline_int8_x86(const Option& opt) { const int num_input = weight_data_size / num_output; int out_elempack = 1; #if __SSE2__ if (opt.use_packing_layout) { out_elempack = num_output % 8 == 0 ? 8 : 1; } #endif // __SSE2__ // src = inch-outch // dst = pb...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x48, %rsp movq %rsi, %rbx movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rax leaq 0x130(%rdi,%rax), %rsi movl -0x60(%rsi), %ecx movl -0x58(%rsi), %eax cltd idivl %ecx movl %eax, %r15d testb $0x7, %cl sete %bpl andb 0x27(%rbx), %bpl pushq $0x...
/csukuangfj[P]ncnn/src/layer/x86/innerproduct_x86.cpp
ncnn::Input::Input()
Input::Input() { one_blob_only = true; support_inplace = true; support_vulkan = true; support_packing = true; support_bf16_storage = true; support_image_storage = true; }
pushq %rbx movq %rdi, %rbx callq 0x77e84 leaq 0x2a7e68(%rip), %rax # 0x4836d0 movq %rax, (%rbx) movb $0x1, %al movb %al, 0xf(%rbx) movb %al, 0xc(%rbx) movl $0x1010101, 0x8(%rbx) # imm = 0x1010101 popq %rbx retq
/csukuangfj[P]ncnn/src/layer/input.cpp
ncnn::Log::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int Log::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; int size = w * h; if (base == -1.f) { #pragma omp parallel for num_threads(opt.num_threads) for (int q = 0; q < chann...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x18, %rsp movq %rsi, %r13 movq %rdi, %rbx movl 0x30(%rsi), %r15d movl 0x38(%rsi), %r12d imull 0x2c(%rsi), %r15d movss 0xd0(%rdi), %xmm0 movss 0x21584b(%rip), %xmm1 # 0x3f11f0 ucomiss %xmm0, %xmm1 jne 0x1dba13 movq (%r13), %r14 movq 0x40(%r13), %...
/csukuangfj[P]ncnn/src/layer/log.cpp
ncnn::LRN::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blob; squa...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x158, %rsp # imm = 0x158 movq %rsi, %rbp movl 0x2c(%rsi), %esi movl 0x30(%rbp), %ecx movl 0x38(%rbp), %r13d movq 0x10(%rbp), %r15 leaq 0x90(%rsp), %rax andq $0x0, 0x40(%rax) movq %rdx, %r14 movq %rdi, %rbx xorps %xmm0, %xmm0 movaps %xmm...
/csukuangfj[P]ncnn/src/layer/lrn.cpp
virtual thunk to ncnn::LRN_x86::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN_x86::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blob; ...
movq (%rdi), %rax addq -0x58(%rax), %rdi jmp 0x1dc3b0
/csukuangfj[P]ncnn/src/layer/x86/lrn_x86.cpp
ncnn::LRN_x86_avx512::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN_x86_avx512::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blo...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x178, %rsp # imm = 0x178 movq %rsi, %r15 movl 0x2c(%rsi), %esi movl 0x30(%r15), %ecx movl 0x38(%r15), %r12d movq 0x10(%r15), %r13 leaq 0x90(%rsp), %rax andq $0x0, 0x40(%rax) movq %rdx, %r14 movq %rdi, %rbx vxorps %xmm0, %xmm0, %xmm0 vmo...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/lrn_x86_avx512.cpp
virtual thunk to ncnn::LRN_x86_avx512::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN_x86_avx512::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blo...
movq (%rdi), %rax addq -0x58(%rax), %rdi jmp 0x1dcc74
/csukuangfj[P]ncnn/build_O2/src/layer/x86/lrn_x86_avx512.cpp
ncnn::LRN_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blob; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x178, %rsp # imm = 0x178 movq %rsi, %r15 movl 0x2c(%rsi), %esi movl 0x30(%r15), %ecx movl 0x38(%r15), %r12d movq 0x10(%r15), %r13 leaq 0x90(%rsp), %rax andq $0x0, 0x40(%rax) movq %rdx, %r14 movq %rdi, %rbx vxorps %xmm0, %xmm0, %xmm0 vmo...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/lrn_x86_fma.cpp
virtual thunk to ncnn::LRN_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int LRN_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { int w = bottom_top_blob.w; int h = bottom_top_blob.h; int channels = bottom_top_blob.c; size_t elemsize = bottom_top_blob.elemsize; int size = w * h; // squared values with local_size padding Mat square_blob; ...
movq (%rdi), %rax addq -0x58(%rax), %rdi jmp 0x1dd838
/csukuangfj[P]ncnn/build_O2/src/layer/x86/lrn_x86_fma.cpp
ncnn::MemoryData::load_model(ncnn::ModelBin const&)
int MemoryData::load_model(const ModelBin& mb) { if (d != 0) { data = mb.load(w, h, d, c, 1); } else if (c != 0) { data = mb.load(w, h, c, 1); } else if (h != 0) { data = mb.load(w, h, 1); } else if (w != 0) { data = mb.load(w, 1); } el...
pushq %r14 pushq %rbx subq $0x58, %rsp movq %rdi, %rbx movl 0xd8(%rdi), %r8d testl %r8d, %r8d je 0x1df233 movl 0xd0(%rbx), %edx movl 0xd4(%rbx), %ecx movl 0xdc(%rbx), %r9d movq (%rsi), %rax movl $0x1, (%rsp) leaq 0x10(%rsp), %r14 movq %r14, %rdi callq *0x28(%rax) leaq 0xe0(%rbx), %rcx movq 0x8(%r14), %rax cmpq %r14, %r...
/csukuangfj[P]ncnn/src/layer/memorydata.cpp
ncnn::MVN::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int MVN::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; size_t elemsize = bottom_blob.elemsize; int size = w * h; top_blob.create(w, h, channels, elemsize, opt.blob_allocator); if (top_blo...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0xb8, %rsp movq %rdx, %r14 movq %rsi, %r12 movq %rdi, 0x8(%rsp) movl 0x2c(%rsi), %ebp movl 0x30(%rsi), %r15d movl 0x38(%rsi), %r13d movq 0x10(%rsi), %rbx movq %rcx, 0x10(%rsp) movq 0x8(%rcx), %r9 movq %rdx, %rdi movl %ebp, %esi movl %r15d, %edx mov...
/csukuangfj[P]ncnn/src/layer/mvn.cpp
ncnn::Pooling::load_param(ncnn::ParamDict const&)
int Pooling::load_param(const ParamDict& pd) { pooling_type = pd.get(0, 0); kernel_w = pd.get(1, 0); kernel_h = pd.get(11, kernel_w); stride_w = pd.get(2, 1); stride_h = pd.get(12, stride_w); pad_left = pd.get(3, 0); pad_right = pd.get(14, pad_left); pad_top = pd.get(13, pad_left); p...
pushq %rbp pushq %r14 pushq %rbx movq %rsi, %r14 movq %rdi, %rbx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%rbx) pushq $0x1 popq %rbp movq %r14, %rdi movl %ebp, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd4(%rbx) pushq $0xb popq %rsi movq %r14, %rdi movl %eax, %edx callq 0x718a6...
/csukuangfj[P]ncnn/src/layer/pooling.cpp
ncnn::Pooling_x86_fma::create_pipeline(ncnn::Option const&)
int Pooling_x86_fma::create_pipeline(const Option& /*opt*/) { if (adaptive_pooling) { support_packing = false; support_bf16_storage = false; support_fp16_storage = false; support_int8_storage = false; support_tensor_storage = false; } return 0; }
movq (%rdi), %rax movq -0x18(%rax), %rcx cmpl $0x0, 0x100(%rdi,%rcx) je 0x1e431f xorl %edx, %edx movb %dl, 0xb(%rdi,%rcx) movq -0x18(%rax), %rcx movb %dl, 0xc(%rdi,%rcx) movq -0x18(%rax), %rcx movb %dl, 0xd(%rdi,%rcx) movq -0x18(%rax), %rcx movb %dl, 0xe(%rdi,%rcx) movq -0x18(%rax), %rax movb %dl, 0x10(%rdi,%rax) xorl ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/pooling_x86_fma.cpp
ncnn::binary_op_broadcast_inner(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, int, ncnn::Option const&)
static int binary_op_broadcast_inner(const Mat& a, const Mat& b, Mat& c, int op_type, const Option& opt) { // squeeze inner axes Mat b2 = b; if (b.dims == 2 && b.w == 1) b2 = b.reshape(b.h); else if (b.dims == 3 && b.h == 1) b2 = b.reshape(b.c); else if (b.dims == 3 && b.w == 1) ...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x1a0, %rsp # imm = 0x1A0 movl %ecx, %r13d movq %rdx, 0x48(%rsp) movq %rdi, 0x20(%rsp) movq (%rsi), %rbx movq 0x8(%rsi), %rcx movq 0x10(%rsi), %rax movq %rax, 0x78(%rsp) movq 0x20(%rsi), %rax movq %rax, ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
virtual thunk to ncnn::BinaryOp_x86_fma::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int BinaryOp_x86_fma::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const bool b_is_scalar = bottom_blobs[1].w * bottom_blobs[1].h * bottom_blobs[1].d * bottom_blobs[1].c * bottom_blobs[1].elempack == 1; const bool a_rank_is_lower = bottom_blobs[0].dims < ...
movq (%rdi), %rax addq -0x40(%rax), %rdi jmp 0x25fe94
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
ncnn::BinaryOp_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int BinaryOp_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { using namespace BinaryOp_x86_fma_functor; if (op_type == Operation_ADD) return binary_op_scalar_inplace<binary_op_add>(bottom_top_blob, b, opt); if (op_type == Operation_SUB) return binary_op_scalar_inplace<binary_op_sub...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0xe0, %rsp movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %eax cmpq $0xb, %rax ja 0x26c8c4 movq %rsi, %r14 leaq 0x18b6f8(%rip), %rcx # 0x3f749c movslq (%rcx,%rax,4), %rax addq %rcx, %rax movq %rs...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
virtual thunk to ncnn::BinaryOp_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int BinaryOp_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { using namespace BinaryOp_x86_fma_functor; if (op_type == Operation_ADD) return binary_op_scalar_inplace<binary_op_add>(bottom_top_blob, b, opt); if (op_type == Operation_SUB) return binary_op_scalar_inplace<binary_op_sub...
pushq %rax movq (%rdi), %rax addq -0x58(%rax), %rdi callq 0x26bd6a xorl %eax, %eax popq %rcx retq
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
int ncnn::binary_op_no_broadcast<ncnn::BinaryOp_x86_fma_functor::binary_op_sub>(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static int binary_op_no_broadcast(const Mat& a, const Mat& b, Mat& c, const Option& opt) { Op op; const int channels = a.c; const int size = a.w * a.h * a.d * a.elempack; #pragma omp parallel for num_threads(opt.num_threads) for (int q = 0; q < channels; q++) { const float* ptr = a.cha...
pushq %rbp pushq %r15 pushq %r14 pushq %rbx movl 0x30(%rdi), %eax movl 0x38(%rdi), %ecx imull 0x2c(%rdi), %eax imull 0x34(%rdi), %eax imull 0x18(%rdi), %eax xorl %r8d, %r8d testl %ecx, %ecx cmovlel %r8d, %ecx cmpq %rcx, %r8 je 0x26c9bc movq 0x10(%rsi), %r9 imulq 0x40(%rsi), %r9 movq 0x10(%rdi), %r10 imulq %r8, %r9 addq...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
int ncnn::binary_op_no_broadcast<ncnn::BinaryOp_x86_fma_functor::binary_op_div>(ncnn::Mat const&, ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&)
static int binary_op_no_broadcast(const Mat& a, const Mat& b, Mat& c, const Option& opt) { Op op; const int channels = a.c; const int size = a.w * a.h * a.d * a.elempack; #pragma omp parallel for num_threads(opt.num_threads) for (int q = 0; q < channels; q++) { const float* ptr = a.cha...
pushq %rbp pushq %r15 pushq %r14 pushq %rbx movl 0x30(%rdi), %eax movl 0x38(%rdi), %ecx imull 0x2c(%rdi), %eax imull 0x34(%rdi), %eax imull 0x18(%rdi), %eax xorl %r8d, %r8d testl %ecx, %ecx cmovlel %r8d, %ecx cmpq %rcx, %r8 je 0x26cabd movq 0x10(%rsi), %r9 imulq 0x40(%rsi), %r9 movq 0x10(%rdi), %r10 imulq %r8, %r9 addq...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/binaryop_x86_fma.cpp
ncnn::UnaryOp_x86_fma::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int UnaryOp_x86_fma::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { using namespace UnaryOp_x86_fma_functor; if (op_type == Operation_ABS) return unary_op_inplace<unary_op_abs>(bottom_top_blob, opt); if (op_type == Operation_NEG) return unary_op_inplace<unary_op_neg>(botto...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x80, %rsp movq %rsi, 0x18(%rsp) movq (%rdi), %rax movq -0x18(%rax), %rax movl 0xd0(%rdi,%rax), %eax cmpq $0x11, %rax ja 0x281716 leaq 0x177d68(%rip), %rcx # 0x3f8348 movslq (%rcx,%rax,4), %rax addq %rcx, %rax j...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/unaryop_x86_fma.cpp
ncnn::UnaryOp_x86_fma_functor::unary_op_exp::func_pack8(float vector[8] const&) const
static NCNN_FORCEINLINE __m256 exp256_ps(__m256 x) { __m256 tmp = _mm256_setzero_ps(), fx; __m256i imm0; __m256 one = *(__m256*)_ps256_1; x = _mm256_min_ps(x, *(__m256*)_ps256_exp_hi); x = _mm256_max_ps(x, *(__m256*)_ps256_exp_lo); /* express exp(x) as exp(g + n*log(2)) */ fx = _mm256_comp...
vbroadcastss 0x16fa73(%rip), %ymm0 # 0x3f11b8 vminps (%rsi), %ymm0, %ymm0 vbroadcastss 0x16fa6a(%rip), %ymm1 # 0x3f11bc vmaxps %ymm1, %ymm0, %ymm1 vbroadcastss 0x16c8b5(%rip), %ymm0 # 0x3ee014 vbroadcastss 0x16fa58(%rip), %ymm2 # 0x3f11c0 vfmadd213ps %ymm0, %ymm1, %ymm2 # ymm2 = (ymm1 * ymm2) + ymm0 vroundps $0x1, %ymm...
/csukuangfj[P]ncnn/src/layer/x86/avx_mathfun.h
ncnn::UnaryOp_x86_avx::forward_inplace(ncnn::Mat&, ncnn::Option const&) const
int UnaryOp_x86_avx::forward_inplace(Mat& bottom_top_blob, const Option& opt) const { using namespace UnaryOp_x86_avx_functor; if (op_type == Operation_ABS) return unary_op_inplace<unary_op_abs>(bottom_top_blob, opt); if (op_type == Operation_NEG) return unary_op_inplace<unary_op_neg>(botto...
pushq %rbp movq %rsp, %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx andq $-0x20, %rsp subq $0x80, %rsp movq %rsi, 0x18(%rsp) movq (%rdi), %rax movq -0x18(%rax), %rax movl 0xd0(%rdi,%rax), %eax cmpq $0x11, %rax ja 0x283f35 leaq 0x1755e4(%rip), %rcx # 0x3f83c8 movslq (%rcx,%rax,4), %rax addq %rcx, %rax j...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/unaryop_x86_avx.cpp
ncnn::UnaryOp_x86_avx_functor::unary_op_exp::func_pack8(float vector[8] const&) const
static NCNN_FORCEINLINE __m256 exp256_ps(__m256 x) { __m256 tmp = _mm256_setzero_ps(), fx; __m256i imm0; __m256 one = *(__m256*)_ps256_1; x = _mm256_min_ps(x, *(__m256*)_ps256_exp_hi); x = _mm256_max_ps(x, *(__m256*)_ps256_exp_lo); /* express exp(x) as exp(g + n*log(2)) */ fx = _mm256_comp...
vbroadcastss 0x16d253(%rip), %ymm0 # 0x3f11b8 vminps (%rsi), %ymm0, %ymm0 vbroadcastss 0x16d24a(%rip), %ymm1 # 0x3f11bc vmaxps %ymm1, %ymm0, %ymm0 vbroadcastss 0x16d241(%rip), %ymm1 # 0x3f11c0 vmulps %ymm1, %ymm0, %ymm1 vbroadcastss 0x16a088(%rip), %ymm2 # 0x3ee014 vaddps %ymm2, %ymm1, %ymm1 vroundps $0x1, %ymm1, %ymm3...
/csukuangfj[P]ncnn/src/layer/x86/avx_mathfun.h
ncnn::UnaryOp_x86_avx_functor::unary_op_cos::func_pack8(float vector[8] const&) const
static NCNN_FORCEINLINE __m256 cos256_ps(__m256 x) { // any x __m256 xmm1, xmm2 = _mm256_setzero_ps(), xmm3, y; __m256i imm0, imm2; #ifndef __AVX2__ __m128i imm0_1, imm0_2; __m128i imm2_1, imm2_2; #endif /* take the absolute value */ x = _mm256_and_ps(x, *(__m256*)_ps256_inv_sign_mask); ...
vbroadcastss 0x16d2fd(%rip), %ymm0 # 0x3f1900 vandps (%rsi), %ymm0, %ymm2 vbroadcastss 0x173240(%rip), %ymm0 # 0x3f7850 vmulps %ymm0, %ymm2, %ymm0 vcvttps2dq %ymm0, %ymm0 vpcmpeqd %xmm1, %xmm1, %xmm1 vpsubd %xmm1, %xmm0, %xmm3 vextractf128 $0x1, %ymm0, %xmm0 vpsubd %xmm1, %xmm0, %xmm0 vmovddup 0x173d7e(%rip), %xmm1 #...
/csukuangfj[P]ncnn/src/layer/x86/avx_mathfun.h
ncnn::UnaryOp_x86_avx_functor::unary_op_log10::func_pack8(float vector[8] const&) const
static NCNN_FORCEINLINE __m256 log256_ps(__m256 x) { __m256i imm0; __m256 one = *(__m256*)_ps256_1; //__m256 invalid_mask = _mm256_cmple_ps(x, _mm256_setzero_ps()); __m256 invalid_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_LE_OS); x = _mm256_max_ps(x, *(__m256*)_ps256_min_norm_pos); /* cut ...
vbroadcastss 0x16bdc3(%rip), %ymm0 # 0x3f11e0 vmaxps (%rsi), %ymm0, %ymm0 vpsrld $0x17, %xmm0, %xmm1 vextractf128 $0x1, %ymm0, %xmm2 vpsrld $0x17, %xmm2, %xmm2 vbroadcastss 0x16bdaa(%rip), %ymm3 # 0x3f11e4 vandps %ymm3, %ymm0, %ymm0 vbroadcastss 0x168bcd(%rip), %ymm3 # 0x3ee014 vorps %ymm3, %ymm0, %ymm0 vbroadcastss 0x...
/csukuangfj[P]ncnn/src/layer/x86/avx_mathfun.h
ncnn::ConvolutionDepthWise::load_model(ncnn::ModelBin const&)
int ConvolutionDepthWise::load_model(const ModelBin& mb) { if (dynamic_weight) return 0; weight_data = mb.load(weight_data_size, 0); if (weight_data.empty()) return -100; if (bias_term) { bias_data = mb.load(num_output, 1); if (bias_data.empty()) return ...
pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x60, %rsp cmpl $0x0, 0x160(%rdi) je 0x285b4c xorl %r12d, %r12d movl %r12d, %eax addq $0x60, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 retq movq %rsi, %r15 movq %rdi, %rbx movl 0x104(%rdi), %edx movq (%rsi), %rax leaq 0x10(%rsp), %r14 movq %r14, ...
/csukuangfj[P]ncnn/src/layer/convolutiondepthwise.cpp
ncnn::ConvolutionDepthWise::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { // convolv with NxN kernel // value = value + bias #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return forward_int8(bottom_blob, top_blob, opt); } #endif ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x48, %rsp movq %rcx, %r12 movq %rdx, %r15 movq %rdi, %r14 cmpb $0x1, 0x1e(%rcx) jne 0x286fbf cmpq $0x1, 0x178(%r14) jne 0x286fbf movq %r14, %rdi movq %r15, %rdx movq %r12, %rcx addq $0x48, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 pop...
/csukuangfj[P]ncnn/src/layer/convolutiondepthwise.cpp
ncnn::ConvolutionDepthWise::forward_int8(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise::forward_int8(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { // convolv with NxN kernel // value = value + bias int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; size_t elemsize = bottom_blob.elemsize; if (channels % gro...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x1a8, %rsp # imm = 0x1A8 movq %rcx, %r14 movq %rdx, %r8 movl 0x38(%rsi), %ebp movl 0x108(%rdi), %ecx movl %ebp, %eax cltd idivl %ecx pushq $-0x64 popq %rbx testl %edx, %edx jne 0x2871a9 movl 0xd0(%rdi), %eax cltd idivl %ecx testl %edx, ...
/csukuangfj[P]ncnn/src/layer/convolutiondepthwise.cpp
ncnn::convolutiondepthwise(ncnn::Mat const&, ncnn::Mat&, ncnn::Mat const&, ncnn::Mat const&, int, int, int, int, int, int, int, int, ncnn::Mat const&, ncnn::Option const&)
static int convolutiondepthwise(const Mat& bottom_blob, Mat& top_blob, const Mat& weight_data, const Mat& bias_data, int kernel_w, int kernel_h, int stride_w, int stride_h, int dilation_w, int dilation_h, int group, int activation_type, const Mat& activation_params, const Option& opt) { const int w = bottom_blob.w;...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x128, %rsp # imm = 0x128 movl %r9d, %ebp movl %r8d, %r15d movq %rdx, 0xa0(%rsp) movq %rsi, 0x58(%rsp) movl 0x170(%rsp), %ebx movq %rdi, 0x28(%rsp) movl 0x2c(%rdi), %r12d cmpq $0x0, (%rcx) movq %rcx, 0xa8(%rsp) je 0x287e14 movslq 0x38(%r...
/csukuangfj[P]ncnn/src/layer/convolutiondepthwise.cpp
ncnn::ConvolutionDepthWise_x86::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_int8_...
pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x50, %rsp movq (%rdi), %rax movq -0x18(%rax), %r13 cmpl $0x0, 0x160(%rdi,%r13) je 0x288dd5 xorl %eax, %eax addq $0x50, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 retq movq %rsi, %r14 movq %rdi, %rbx movl 0x110(%rdi,%r13), %ecx decl %ecx cmpl $0x5...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86::create_pipeline_int8_x86(ncnn::Option const&)
int ConvolutionDepthWise_x86::create_pipeline_int8_x86(const Option& opt) { const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; // depth-wise if (channels == group && group == num_output) { int elempack = 1; #if __SSE2__ ...
pushq %r15 pushq %r14 pushq %r12 pushq %rbx subq $0x48, %rsp movq %rsi, %r15 movq %rdi, %rbx movq (%rdi), %rax movq -0x18(%rax), %r14 movl 0xd8(%rdi,%r14), %r8d imull 0xd4(%rdi,%r14), %r8d movl 0xd0(%rdi,%r14), %esi movl 0x104(%rdi,%r14), %eax movl 0x108(%rdi,%r14), %ecx cltd idivl %ecx cltd idivl %r8d movl %eax, %edi ...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86::create_group_ops(ncnn::Option const&)
int ConvolutionDepthWise_x86::create_group_ops(const Option& opt) { // create Convolution op for each group const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; for (int i = 0; i < (int)group_ops.size(); i++) delete group_ops[i];...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x278, %rsp # imm = 0x278 movq %rsi, 0x250(%rsp) movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %ecx movl 0xd8(%rdi,%rdx), %ebp imull 0xd4(%rdi,%rdx), %ebp movl 0x104(%rdi,%rdx), %eax movl 0x108(%rdi,%rdx)...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_int8_...
pushq %rax movq (%rdi), %rax addq -0x30(%rax), %rdi callq 0x288da6 xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete group_o...
pushq %r15 pushq %r14 pushq %rbx movq %rsi, %r14 movq %rdi, %rbx movq 0x8(%rdi), %rdi testq %rdi, %rdi je 0x28a54b movq (%rdi), %rax movq %r14, %rsi callq *0x28(%rax) movq 0x8(%rbx), %rdi testq %rdi, %rdi je 0x28a546 movq (%rdi), %rax callq *0x8(%rax) andq $0x0, 0x8(%rbx) xorl %r15d, %r15d movq 0x10(%rbx), %rax movq 0x...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete group_o...
pushq %rax movq (%rdi), %rax addq -0x38(%rax), %rdi callq 0x28a51a xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int channels...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x2c8, %rsp # imm = 0x2C8 movq %rcx, %r14 movq (%rdi), %rax movq -0x18(%rax), %rcx cmpb $0x1, 0x1e(%r14) jne 0x28a5f5 cmpl $0x0, 0x10c(%rdi,%rcx) je 0x28a5f5 movq %r14, %rcx addq $0x2c8, %rsp # imm = 0x2C8 popq %rbx popq %r12 ...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int channels...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x28a5b0
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int _ker...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x168, %rsp # imm = 0x168 movq %rsi, %rbp movq %rdi, %r13 movq (%rsi), %r14 leaq 0x48(%r14), %rdi movq (%rdx), %rax movq %rax, 0xb8(%rsp) movl 0x60(%r14), %ebx movl 0x74(%r14), %eax movl %eax, 0x1c(%rsp) movl 0x78(%r14), %eax movl %eax, ...
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int _ker...
movq (%rdi), %rax addq -0x40(%rax), %rdi jmp 0x28febe
/csukuangfj[P]ncnn/src/layer/x86/convolutiondepthwise_x86.cpp
ncnn::ConvolutionDepthWise_x86_avx512::create_group_ops(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx512::create_group_ops(const Option& opt) { // create Convolution op for each group const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; for (int i = 0; i < (int)group_ops.size(); i++) delete group_...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x278, %rsp # imm = 0x278 movq %rsi, 0x250(%rsp) movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %ecx movl 0xd8(%rdi,%rdx), %ebp imull 0xd4(%rdi,%rdx), %ebp movl 0x104(%rdi,%rdx), %eax movl 0x108(%rdi,%rdx)...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
ncnn::ConvolutionDepthWise_x86_avx512::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx512::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete ...
pushq %r15 pushq %r14 pushq %rbx movq %rsi, %r14 movq %rdi, %rbx movq 0x8(%rdi), %rdi testq %rdi, %rdi je 0x292071 movq (%rdi), %rax movq %r14, %rsi callq *0x28(%rax) movq 0x8(%rbx), %rdi testq %rdi, %rdi je 0x29206c movq (%rdi), %rax callq *0x8(%rax) andq $0x0, 0x8(%rbx) xorl %r15d, %r15d movq 0x10(%rbx), %rax movq 0x...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx512::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx512::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete ...
pushq %rax movq (%rdi), %rax addq -0x38(%rax), %rdi callq 0x292040 xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
ncnn::ConvolutionDepthWise_x86_avx512::forward_int8_x86(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx512::forward_int8_x86(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; int elempack = bottom_blob.elempack; int elembits = bottom_blob.elembits(); const int kernel_exten...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x318, %rsp # imm = 0x318 movq %rcx, 0x30(%rsp) movq %rdx, 0x20(%rsp) movq %rsi, %r15 movl 0x38(%rsi), %ecx movl 0x18(%rsi), %r12d movq 0x10(%rsi), %rsi testl %r12d, %r12d je 0x2975b9 leal (,%rsi,8), %eax cltd idivl %r12d cmpl $0x8, %eax...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx512::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx512::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int c...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x2920d6
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
ncnn::ConvolutionDepthWise_x86_avx512::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx512::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const i...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x168, %rsp # imm = 0x168 movq %rsi, %rbp movq %rdi, %r13 movq (%rsi), %r14 leaq 0x48(%r14), %rdi movq (%rdx), %rax movq %rax, 0xb8(%rsp) movl 0x60(%r14), %ebx movl 0x74(%r14), %eax movl %eax, 0x1c(%rsp) movl 0x78(%r14), %eax movl %eax, ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx512::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx512::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const i...
movq (%rdi), %rax addq -0x40(%rax), %rdi jmp 0x299cb8
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx512.cpp
ncnn::ConvolutionDepthWise_x86_fma::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_fma::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_i...
pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x50, %rsp movq (%rdi), %rax movq -0x18(%rax), %r13 cmpl $0x0, 0x160(%rdi,%r13) je 0x29a51d xorl %eax, %eax addq $0x50, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 retq movq %rsi, %r14 movq %rdi, %rbx movl 0x110(%rdi,%r13), %ecx decl %ecx cmpl $0x5...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_fma::create_group_ops(ncnn::Option const&)
int ConvolutionDepthWise_x86_fma::create_group_ops(const Option& opt) { // create Convolution op for each group const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; for (int i = 0; i < (int)group_ops.size(); i++) delete group_ops...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x278, %rsp # imm = 0x278 movq %rsi, 0x250(%rsp) movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %ecx movl 0xd8(%rdi,%rdx), %ebp imull 0xd4(%rdi,%rdx), %ebp movl 0x104(%rdi,%rdx), %eax movl 0x108(%rdi,%rdx)...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_fma::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_fma::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_i...
pushq %rax movq (%rdi), %rax addq -0x30(%rax), %rdi callq 0x29a4ee xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_fma::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_fma::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete gro...
pushq %r15 pushq %r14 pushq %rbx movq %rsi, %r14 movq %rdi, %rbx movq 0x8(%rdi), %rdi testq %rdi, %rdi je 0x29bd9d movq (%rdi), %rax movq %r14, %rsi callq *0x28(%rax) movq 0x8(%rbx), %rdi testq %rdi, %rdi je 0x29bd98 movq (%rdi), %rax callq *0x8(%rax) andq $0x0, 0x8(%rbx) xorl %r15d, %r15d movq 0x10(%rbx), %rax movq 0x...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_fma::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_fma::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete gro...
pushq %rax movq (%rdi), %rax addq -0x38(%rax), %rdi callq 0x29bd6c xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_fma::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_fma::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int chan...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x2e8, %rsp # imm = 0x2E8 movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdi movq %rcx, 0x58(%rsp) cmpb $0x1, 0x1e(%rcx) jne 0x29be51 cmpl $0x0, 0x10c(%r14,%rdi) je 0x29be51 movq %r14, %rdi movq 0x58(%rsp), %rcx addq $0x2e8, %rsp ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_fma::forward_int8_x86(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_fma::forward_int8_x86(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; int elempack = bottom_blob.elempack; int elembits = bottom_blob.elembits(); const int kernel_extent_w...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x318, %rsp # imm = 0x318 movq %rcx, 0x30(%rsp) movq %rdx, 0x20(%rsp) movq %rsi, %r15 movl 0x38(%rsi), %ecx movl 0x18(%rsi), %r12d movq 0x10(%rsi), %rsi testl %r12d, %r12d je 0x29ffd3 leal (,%rsi,8), %eax cltd idivl %r12d cmpl $0x8, %eax...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_fma::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_fma::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int chan...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x29be02
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_fma::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_fma::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x168, %rsp # imm = 0x168 movq %rsi, %rbp movq %rdi, %r13 movq (%rsi), %r14 leaq 0x48(%r14), %rdi movq (%rdx), %rax movq %rax, 0xb8(%rsp) movl 0x60(%r14), %ebx movl 0x74(%r14), %eax movl %eax, 0x1c(%rsp) movl 0x78(%r14), %eax movl %eax, ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_fma::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_fma::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int ...
movq (%rdi), %rax addq -0x40(%rax), %rdi jmp 0x2a2888
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_fma.cpp
ncnn::ConvolutionDepthWise_x86_avx::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_i...
pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x50, %rsp movq (%rdi), %rax movq -0x18(%rax), %r13 cmpl $0x0, 0x160(%rdi,%r13) je 0x2a30ed xorl %eax, %eax addq $0x50, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 retq movq %rsi, %r14 movq %rdi, %rbx movl 0x110(%rdi,%r13), %ecx decl %ecx cmpl $0x5...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::create_pipeline_int8_x86(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::create_pipeline_int8_x86(const Option& opt) { const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; // depth-wise if (channels == group && group == num_output) { int elempack = 1; #if __SSE2__...
pushq %r15 pushq %r14 pushq %r12 pushq %rbx subq $0x48, %rsp movq %rsi, %r15 movq %rdi, %rbx movq (%rdi), %rax movq -0x18(%rax), %r14 movl 0xd8(%rdi,%r14), %r8d imull 0xd4(%rdi,%r14), %r8d movl 0xd0(%rdi,%r14), %esi movl 0x104(%rdi,%r14), %eax movl 0x108(%rdi,%r14), %ecx cltd idivl %ecx cltd idivl %r8d movl %eax, %edi ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::create_group_ops(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::create_group_ops(const Option& opt) { // create Convolution op for each group const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; for (int i = 0; i < (int)group_ops.size(); i++) delete group_ops...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x278, %rsp # imm = 0x278 movq %rsi, 0x250(%rsp) movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %ecx movl 0xd8(%rdi,%rdx), %ebp imull 0xd4(%rdi,%rdx), %ebp movl 0x104(%rdi,%rdx), %eax movl 0x108(%rdi,%rdx)...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx::create_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::create_pipeline(const Option& opt) { if (dynamic_weight) return 0; activation = create_activation_layer(activation_type, activation_params, opt); #if NCNN_INT8 if (opt.use_int8_inference && weight_data.elemsize == (size_t)1u) { return create_pipeline_i...
pushq %rax movq (%rdi), %rax addq -0x30(%rax), %rdi callq 0x2a30be xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete gro...
pushq %r15 pushq %r14 pushq %rbx movq %rsi, %r14 movq %rdi, %rbx movq 0x8(%rdi), %rdi testq %rdi, %rdi je 0x2a496d movq (%rdi), %rax movq %r14, %rsi callq *0x28(%rax) movq 0x8(%rbx), %rdi testq %rdi, %rdi je 0x2a4968 movq (%rdi), %rax callq *0x8(%rax) andq $0x0, 0x8(%rbx) xorl %r15d, %r15d movq 0x10(%rbx), %rax movq 0x...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx::destroy_pipeline(ncnn::Option const&)
int ConvolutionDepthWise_x86_avx::destroy_pipeline(const Option& opt) { if (activation) { activation->destroy_pipeline(opt); delete activation; activation = 0; } for (int i = 0; i < (int)group_ops.size(); i++) { group_ops[i]->destroy_pipeline(opt); delete gro...
pushq %rax movq (%rdi), %rax addq -0x38(%rax), %rdi callq 0x2a493c xorl %eax, %eax popq %rcx retq nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int chan...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x3b8, %rsp # imm = 0x3B8 movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdi movq %rcx, 0x48(%rsp) cmpb $0x1, 0x1e(%rcx) jne 0x2a4a21 cmpl $0x0, 0x10c(%r14,%rdi) je 0x2a4a21 movq %r14, %rdi movq 0x48(%rsp), %rcx addq $0x3b8, %rsp ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::forward_int8_x86(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx::forward_int8_x86(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int channels = bottom_blob.c; int elempack = bottom_blob.elempack; int elembits = bottom_blob.elembits(); const int kernel_extent_w...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x318, %rsp # imm = 0x318 movq %rcx, 0x30(%rsp) movq %rdx, 0x20(%rsp) movq %rsi, %r15 movl 0x38(%rsi), %ecx movl 0x18(%rsi), %r12d movq 0x10(%rsi), %rsi testl %r12d, %r12d je 0x2a9903 leal (,%rsi,8), %eax cltd idivl %r12d cmpl $0x8, %eax...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { #if NCNN_INT8 if (opt.use_int8_inference && int8_scale_term) { return forward_int8_x86(bottom_blob, top_blob, opt); } #endif int w = bottom_blob.w; int h = bottom_blob.h; int chan...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x2a49d2
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::ConvolutionDepthWise_x86_avx::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x168, %rsp # imm = 0x168 movq %rsi, %rbp movq %rdi, %r13 movq (%rsi), %r14 leaq 0x48(%r14), %rdi movq (%rdx), %rax movq %rax, 0xb8(%rsp) movl 0x60(%r14), %ebx movl 0x74(%r14), %eax movl %eax, 0x1c(%rsp) movl 0x78(%r14), %eax movl %eax, ...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
virtual thunk to ncnn::ConvolutionDepthWise_x86_avx::forward(std::vector<ncnn::Mat, std::allocator<ncnn::Mat>> const&, std::vector<ncnn::Mat, std::allocator<ncnn::Mat>>&, ncnn::Option const&) const
int ConvolutionDepthWise_x86_avx::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const { const Mat& bottom_blob = bottom_blobs[0]; const Mat& _weight_data = bottom_blobs[1]; Mat& top_blob = top_blobs[0]; const int _kernel_w = _weight_data.w; const int ...
movq (%rdi), %rax addq -0x40(%rax), %rdi jmp 0x2ac246 nop
/csukuangfj[P]ncnn/build_O2/src/layer/x86/convolutiondepthwise_x86_avx.cpp
ncnn::Padding::load_param(ncnn::ParamDict const&)
int Padding::load_param(const ParamDict& pd) { top = pd.get(0, 0); bottom = pd.get(1, 0); left = pd.get(2, 0); right = pd.get(3, 0); type = pd.get(4, 0); value = pd.get(5, 0.f); per_channel_pad_data_size = pd.get(6, 0); front = pd.get(7, 0); behind = pd.get(8, 0); return 0; }
pushq %r14 pushq %rbx pushq %rax movq %rsi, %r14 movq %rdi, %rbx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%rbx) pushq $0x1 popq %rsi movq %r14, %rdi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd4(%rbx) pushq $0x2 popq %rsi movq %r14, %rdi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd8...
/csukuangfj[P]ncnn/src/layer/padding.cpp
ncnn::Padding::load_model(ncnn::ModelBin const&)
int Padding::load_model(const ModelBin& mb) { if (per_channel_pad_data_size) { per_channel_pad_data = mb.load(per_channel_pad_data_size, 1); } return 0; }
pushq %r14 pushq %rbx subq $0x48, %rsp movl 0xf0(%rdi), %edx testl %edx, %edx je 0x2acbe2 movq %rdi, %rbx movq (%rsi), %rax movq %rsp, %r14 pushq $0x1 popq %rcx movq %r14, %rdi callq *0x10(%rax) leaq 0xf8(%rbx), %rcx movq 0x8(%r14), %rax cmpq %r14, %rcx je 0x2acbba testq %rax, %rax je 0x2acb2b lock incl (%rax) movq 0x1...
/csukuangfj[P]ncnn/src/layer/padding.cpp
ncnn::Padding::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x118, %rsp # imm = 0x118 movq %rcx, 0x20(%rsp) movq %rdx, %rbx movq %rsi, %r14 movl 0xd0(%rdi), %eax movl 0xd4(%rdi), %edx testl %eax, %eax sete %cl testl %edx, %edx sete %sil movq 0xd8(%rdi), %xmm1 movq %rdi, 0x8(%rsp) movq 0xe8(%rdi),...
/csukuangfj[P]ncnn/src/layer/padding.cpp
void ncnn::copy_make_border_image<signed char>(ncnn::Mat const&, ncnn::Mat&, int, int, int, signed char)
static void copy_make_border_image(const Mat& src, Mat& dst, int top, int left, int type, T v) { int w = dst.w; int h = dst.h; const T* ptr = src; T* outptr = dst; if (type == 0) { int y = 0; // fill top for (; y < top; y++) { int x = 0; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x38, %rsp movl %r9d, %r13d movl %ecx, %r9d movq %rdi, %r14 movslq 0x2c(%rsi), %r12 movl 0x30(%rsi), %eax movl %eax, 0x2c(%rsp) movq (%rdi), %r15 movq (%rsi), %rbp movslq %ecx, %rax movq %rax, (%rsp) testl %r8d, %r8d movl %ecx, 0xc(%rsp) movl %edx,...
/csukuangfj[P]ncnn/src/layer/padding.cpp
void ncnn::copy_make_border_image<unsigned short>(ncnn::Mat const&, ncnn::Mat&, int, int, int, unsigned short)
static void copy_make_border_image(const Mat& src, Mat& dst, int top, int left, int type, T v) { int w = dst.w; int h = dst.h; const T* ptr = src; T* outptr = dst; if (type == 0) { int y = 0; // fill top for (; y < top; y++) { int x = 0; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x58, %rsp movslq 0x2c(%rsi), %rbx movl 0x30(%rsi), %eax movl %eax, 0x4c(%rsp) movq (%rdi), %r12 movq (%rsi), %rbp movq %rcx, 0x28(%rsp) movslq %ecx, %rax movq %rax, 0x18(%rsp) testl %r8d, %r8d movq %rdi, 0x40(%rsp) movl %edx, 0xc(%rsp) jne 0x2ae0d...
/csukuangfj[P]ncnn/src/layer/padding.cpp
void ncnn::copy_make_border_image<float>(ncnn::Mat const&, ncnn::Mat&, int, int, int, float)
static void copy_make_border_image(const Mat& src, Mat& dst, int top, int left, int type, T v) { int w = dst.w; int h = dst.h; const T* ptr = src; T* outptr = dst; if (type == 0) { int y = 0; // fill top for (; y < top; y++) { int x = 0; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x58, %rsp movslq 0x2c(%rsi), %rbx movl 0x30(%rsi), %eax movl %eax, 0x4c(%rsp) movq (%rdi), %r12 movq (%rsi), %rbp movq %rcx, 0x30(%rsp) movslq %ecx, %rax movq %rax, 0x18(%rsp) testl %r8d, %r8d movq %rdi, 0x40(%rsp) movl %edx, 0xc(%rsp) jne 0x2ae81...
/csukuangfj[P]ncnn/src/layer/padding.cpp
ncnn::Padding_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) re...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0xe8, %rsp movq %rcx, %rbx movq %rdx, %r15 movq %rsi, %rcx movq (%rdi), %rax movq -0x18(%rax), %rbp movq %rdi, (%rsp) movl 0xd0(%rdi,%rbp), %r10d testl %r10d, %r10d movq %rsi, 0x8(%rsp) jne 0x2aeef1 movq (%rsp), %rax cmpl $0x0, 0xd4(%rax,%rbp) jne ...
/csukuangfj[P]ncnn/src/layer/x86/padding_x86.cpp
ncnn::Padding_x86::forward_int8(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86::forward_int8(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; int channels = bottom_blob.c; int dims = bottom_blob.dims; size_t elemsize = bottom_blob.elemsize; int elempack = bottom_blob.e...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x118, %rsp # imm = 0x118 movq %rcx, %r12 movq %rdx, %rbx movq %rsi, %rax movq %rdi, %r13 movl 0x2c(%rsi), %esi movl 0x30(%rax), %edx movl 0x34(%rax), %ecx movl %ecx, 0x18(%rsp) movl 0x38(%rax), %ebp movq 0x10(%rax), %r9 movl 0x18(%rax),...
/csukuangfj[P]ncnn/src/layer/x86/padding_x86.cpp
virtual thunk to ncnn::Padding_x86::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) re...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x2aee70
/csukuangfj[P]ncnn/src/layer/x86/padding_x86.cpp
ncnn::padding_constant_pack8_int8_sse(ncnn::Mat const&, ncnn::Mat&, int, int, int, int, long)
static void padding_constant_pack8_int8_sse(const Mat& src, Mat& dst, int top, int bottom, int left, int right, int64_t _v) { const int64_t* ptr = src; int64_t* outptr = dst; // fill top for (int y = 0; y < top; y++) { for (int x = 0; x < dst.w; x++) { *outptr++ = _v; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r12 pushq %rbx movq 0x30(%rsp), %rax movq (%rdi), %r11 movl 0x2c(%rsi), %ebx xorl %ebp, %ebp testl %ebx, %ebx cmovlel %ebp, %ebx movq (%rsi), %r10 testl %edx, %edx cmovlel %ebp, %edx cmpl %edx, %ebp je 0x2b0cfa movl %ebx, %r14d subl $0x1, %r14d jb 0x2b0cf6 movq %rax, (%r10) addq ...
/csukuangfj[P]ncnn/src/layer/x86/padding_pack8_int8.h
ncnn::Padding_x86_avx512::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_avx512::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x118, %rsp # imm = 0x118 movq %rcx, %rbp movq %rdx, 0x8(%rsp) movq %rsi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rax leaq (%rdi,%rax), %r11 movq %rdi, 0x10(%rsp) movl 0xd0(%rdi,%rax), %r10d testl %r10d, %r10d jne 0x2b0e4f cmpl $0x0, 0...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_avx512.cpp
ncnn::Padding_x86_avx512::forward_int8(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_avx512::forward_int8(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; int channels = bottom_blob.c; int dims = bottom_blob.dims; size_t elemsize = bottom_blob.elemsize; int elempack = bottom...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x128, %rsp # imm = 0x128 movq %rcx, %r12 movq %rdx, %rbx movq %rdi, %r14 movl 0x38(%rsi), %r15d movq 0x10(%rsi), %r9 movl 0x18(%rsi), %edi movq %rsi, 0x28(%rsp) vmovdqu 0x28(%rsi), %xmm0 cmpl $0x8, %edi jne 0x2b42dc vmovd %xmm0, %eax de...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_avx512.cpp
virtual thunk to ncnn::Padding_x86_avx512::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_avx512::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) ...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x2b0de0
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_avx512.cpp
ncnn::padding_constant_pack8_int8_sse(ncnn::Mat const&, ncnn::Mat&, int, int, int, int, long)
static void padding_constant_pack8_int8_sse(const Mat& src, Mat& dst, int top, int bottom, int left, int right, int64_t _v) { const int64_t* ptr = src; int64_t* outptr = dst; // fill top for (int y = 0; y < top; y++) { for (int x = 0; x < dst.w; x++) { *outptr++ = _v; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r12 pushq %rbx movq 0x30(%rsp), %rax movq (%rdi), %r11 movl 0x2c(%rsi), %ebx xorl %ebp, %ebp testl %ebx, %ebx cmovlel %ebp, %ebx movq (%rsi), %r10 testl %edx, %edx cmovlel %ebp, %edx cmpl %edx, %ebp je 0x2b471c movl %ebx, %r14d subl $0x1, %r14d jb 0x2b4718 movq %rax, (%r10) addq ...
/csukuangfj[P]ncnn/src/layer/x86/padding_pack8_int8.h
ncnn::Padding_x86_fma::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_fma::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) ...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0xf8, %rsp movq %rcx, %rbp movq %rdx, 0x8(%rsp) movq %rsi, %r14 movq (%rdi), %rax movq -0x18(%rax), %r11 movl 0xd0(%rdi,%r11), %r10d testl %r10d, %r10d jne 0x2b486b cmpl $0x0, 0xd4(%rdi,%r11) jne 0x2b486b cmpl $0x0, 0xd8(%rdi,%r11) jne 0x2b486b cmp...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_fma.cpp
ncnn::Padding_x86_fma::forward_int8(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_fma::forward_int8(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int d = bottom_blob.d; int channels = bottom_blob.c; int dims = bottom_blob.dims; size_t elemsize = bottom_blob.elemsize; int elempack = bottom_bl...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x128, %rsp # imm = 0x128 movq %rcx, %r12 movq %rdx, %rbx movq %rdi, %r14 movl 0x38(%rsi), %r15d movq 0x10(%rsi), %r9 movl 0x18(%rsi), %edi movq %rsi, 0x28(%rsp) vmovdqu 0x28(%rsi), %xmm0 cmpl $0x8, %edi jne 0x2b70ac vmovd %xmm0, %eax de...
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_fma.cpp
virtual thunk to ncnn::Padding_x86_fma::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Padding_x86_fma::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { if (top == 0 && bottom == 0 && left == 0 && right == 0 && front == 0 && behind == 0) { top_blob = bottom_blob; return 0; } int elembits = bottom_blob.elembits(); if (elembits == 8) ...
movq (%rdi), %rax addq -0x48(%rax), %rdi jmp 0x2b4800
/csukuangfj[P]ncnn/build_O2/src/layer/x86/padding_x86_fma.cpp
ncnn::padding_constant_pack8_int8_sse(ncnn::Mat const&, ncnn::Mat&, int, int, int, int, long)
static void padding_constant_pack8_int8_sse(const Mat& src, Mat& dst, int top, int bottom, int left, int right, int64_t _v) { const int64_t* ptr = src; int64_t* outptr = dst; // fill top for (int y = 0; y < top; y++) { for (int x = 0; x < dst.w; x++) { *outptr++ = _v; ...
pushq %rbp pushq %r15 pushq %r14 pushq %r12 pushq %rbx movq 0x30(%rsp), %rax movq (%rdi), %r11 movl 0x2c(%rsi), %ebx xorl %ebp, %ebp testl %ebx, %ebx cmovlel %ebp, %ebx movq (%rsi), %r10 testl %edx, %edx cmovlel %ebp, %edx cmpl %edx, %ebp je 0x2b744a movl %ebx, %r14d subl $0x1, %r14d jb 0x2b7446 movq %rax, (%r10) addq ...
/csukuangfj[P]ncnn/src/layer/x86/padding_pack8_int8.h
ncnn::ExpandDims::load_param(ncnn::ParamDict const&)
int ExpandDims::load_param(const ParamDict& pd) { expand_w = pd.get(0, 0); expand_h = pd.get(1, 0); expand_d = pd.get(11, 0); expand_c = pd.get(2, 0); axes = pd.get(3, Mat()); return 0; }
pushq %r15 pushq %r14 pushq %rbx subq $0xa0, %rsp movq %rsi, %r14 movq %rdi, %rbx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%rbx) pushq $0x1 popq %rsi movq %r14, %rdi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd4(%rbx) pushq $0xb popq %rsi movq %r14, %rdi xorl %edx, %edx callq 0x718a...
/csukuangfj[P]ncnn/src/layer/expanddims.cpp
ncnn::PriorBox::load_param(ncnn::ParamDict const&)
int PriorBox::load_param(const ParamDict& pd) { min_sizes = pd.get(0, Mat()); max_sizes = pd.get(1, Mat()); aspect_ratios = pd.get(2, Mat()); variances[0] = pd.get(3, 0.1f); variances[1] = pd.get(4, 0.1f); variances[2] = pd.get(5, 0.2f); variances[3] = pd.get(6, 0.2f); flip = pd.get(7, 1...
pushq %r15 pushq %r14 pushq %rbx subq $0xa0, %rsp movq %rsi, %r14 movq %rdi, %rbx leaq 0x50(%rsp), %rcx andq $0x0, 0x40(%rcx) xorps %xmm0, %xmm0 movaps %xmm0, (%rcx) movups %xmm0, 0xc(%rcx) movaps %xmm0, 0x20(%rcx) movups %xmm0, 0x2c(%rcx) movq %rsp, %r15 movq %r15, %rdi xorl %edx, %edx callq 0x718da leaq 0xd0(%rbx), %...
/csukuangfj[P]ncnn/src/layer/priorbox.cpp
ncnn::DetectionOutput::load_param(ncnn::ParamDict const&)
int DetectionOutput::load_param(const ParamDict& pd) { num_class = pd.get(0, 0); nms_threshold = pd.get(1, 0.05f); nms_top_k = pd.get(2, 300); keep_top_k = pd.get(3, 100); confidence_threshold = pd.get(4, 0.5f); variances[0] = pd.get(5, 0.1f); variances[1] = pd.get(6, 0.1f); variances[2]...
pushq %r14 pushq %rbx pushq %rax movq %rsi, %r14 movq %rdi, %rbx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%rbx) pushq $0x1 popq %rsi movss 0x1371b7(%rip), %xmm0 # 0x3f88f0 movq %r14, %rdi callq 0x718c0 movss %xmm0, 0xd4(%rbx) pushq $0x2 popq %rsi movq %r14, %rdi movl $0x12c, %edx ...
/csukuangfj[P]ncnn/src/layer/detectionoutput.cpp
ncnn::Interp::load_param(ncnn::ParamDict const&)
int Interp::load_param(const ParamDict& pd) { resize_type = pd.get(0, 0); height_scale = pd.get(1, 1.f); width_scale = pd.get(2, 1.f); output_height = pd.get(3, 0); output_width = pd.get(4, 0); dynamic_target_size = pd.get(5, 0); align_corner = pd.get(6, 0); if (resize_type < 0 || resiz...
pushq %r15 pushq %r14 pushq %rbx movq %rsi, %r15 movq %rdi, %r14 xorl %ebx, %ebx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%r14) pushq $0x1 popq %rsi movss 0x12bc91(%rip), %xmm0 # 0x3eec88 movq %r15, %rdi callq 0x718c0 movss %xmm0, 0xd8(%r14) pushq $0x2 popq %rsi movq %r15, %rdi mo...
/csukuangfj[P]ncnn/src/layer/interp.cpp
ncnn::Interp::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int Interp::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; int outw = output_width; int outh = output_height; if (bottom_blob.dims == 1) { w = 1; h = 1; } if (outw == 0 || outh == 0) { o...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x58, %rsp movq %rcx, %r14 movq %rdx, %rbx movq %rsi, %r12 movq %rdi, %r15 movq 0xdc(%rdi), %xmm0 movdqa %xmm0, 0x40(%rsp) pshufd $0x50, %xmm0, %xmm0 # xmm0 = xmm0[0,0,1,1] pxor %xmm1, %xmm1 pcmpeqd %xmm0, %xmm1 movmskpd %xmm1, %eax testl %eax,...
/csukuangfj[P]ncnn/src/layer/interp.cpp
ncnn::linear_coeffs(int, int, int*, float*, int)
static void linear_coeffs(int w, int outw, int* xofs, float* alpha, int align_corner) { double scale = (double)w / outw; if (align_corner) { scale = (double)(w - 1) / (outw - 1); } for (int dx = 0; dx < outw; dx++) { float fx = (float)((dx + 0.5) * scale - 0.5); if (alig...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x18, %rsp movl %r8d, %ebx movq %rcx, %r14 movq %rdx, 0x10(%rsp) movl %esi, %r12d movl %edi, %r13d testl %r8d, %r8d je 0x2c7078 leal -0x1(%r13), %ebp cvtsi2sd %ebp, %xmm1 leal -0x1(%r12), %eax cvtsi2sd %eax, %xmm0 divsd %xmm0, %xmm1 movsd %xmm1, (%...
/csukuangfj[P]ncnn/src/layer/x86/interp_bilinear.h
ncnn::cubic_coeffs(int, int, int*, float*, int)
static void cubic_coeffs(int w, int outw, int* xofs, float* alpha, int align_corner) { double scale = (double)w / outw; if (align_corner) { scale = (double)(w - 1) / (outw - 1); } for (int dx = 0; dx < outw; dx++) { float fx = (float)((dx + 0.5) * scale - 0.5); if (align...
testl %r8d, %r8d je 0x2d3720 leal -0x1(%rdi), %eax vcvtsi2sd %eax, %xmm0, %xmm0 leal -0x1(%rsi), %r9d vcvtsi2sd %r9d, %xmm1, %xmm1 vdivsd %xmm1, %xmm0, %xmm0 jmp 0x2d372f vcvtsi2sd %edi, %xmm0, %xmm0 vcvtsi2sd %esi, %xmm1, %xmm1 vdivsd %xmm1, %xmm0, %xmm0 leal -0x1(%rdi), %eax pushq %rbp pushq %r14 pushq %rbx leal -0x2...
/csukuangfj[P]ncnn/src/layer/x86/interp_bicubic.h
ncnn::DeconvolutionDepthWise::load_param(ncnn::ParamDict const&)
int DeconvolutionDepthWise::load_param(const ParamDict& pd) { num_output = pd.get(0, 0); kernel_w = pd.get(1, 0); kernel_h = pd.get(11, kernel_w); dilation_w = pd.get(2, 1); dilation_h = pd.get(12, dilation_w); stride_w = pd.get(3, 1); stride_h = pd.get(13, stride_w); pad_left = pd.get(4...
pushq %rbp pushq %r15 pushq %r14 pushq %rbx subq $0x98, %rsp movq %rsi, %r14 movq %rdi, %rbx movq %rsi, %rdi xorl %esi, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd0(%rbx) pushq $0x1 popq %rbp movq %r14, %rdi movl %ebp, %esi xorl %edx, %edx callq 0x718a6 movl %eax, 0xd4(%rbx) pushq $0xb popq %rsi movq %r14, %rdi m...
/csukuangfj[P]ncnn/src/layer/deconvolutiondepthwise.cpp
ncnn::DeconvolutionDepthWise::forward(ncnn::Mat const&, ncnn::Mat&, ncnn::Option const&) const
int DeconvolutionDepthWise::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const { int w = bottom_blob.w; int h = bottom_blob.h; size_t elemsize = bottom_blob.elemsize; const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1; const int kernel_extent_h = dilation_h * (kernel_h...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x198, %rsp # imm = 0x198 movl 0x2c(%rsi), %r14d movl 0x30(%rsi), %ebp movl 0xd4(%rdi), %ebx decl %ebx imull 0xdc(%rdi), %ebx movq %rsi, 0x20(%rsp) movq 0x10(%rsi), %r8 decl %r14d imull 0xe4(%rdi), %r14d movl 0xd8(%rdi), %r12d decl %r12d...
/csukuangfj[P]ncnn/src/layer/deconvolutiondepthwise.cpp
ncnn::DeconvolutionDepthWise_x86::create_group_ops(ncnn::Option const&)
int DeconvolutionDepthWise_x86::create_group_ops(const Option& opt) { // create Deconvolution op for each group const int maxk = kernel_w * kernel_h; int channels = (weight_data_size / group) / maxk / (num_output / group) * group; for (int i = 0; i < (int)group_ops.size(); i++) delete group_ops...
pushq %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $0x158, %rsp # imm = 0x158 movq %rsi, 0x120(%rsp) movq %rdi, %r14 movq (%rdi), %rax movq -0x18(%rax), %rdx movl 0xd0(%rdi,%rdx), %ecx movl 0xd8(%rdi,%rdx), %ebp imull 0xd4(%rdi,%rdx), %ebp movl 0x110(%rdi,%rdx), %eax movl 0x114(%rdi,%rdx)...
/csukuangfj[P]ncnn/src/layer/x86/deconvolutiondepthwise_x86.cpp