| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16_grad/ |
| D | convolution_grad_input.c | 42 const float16_t *mat_b[C16NUM]; in ConvDwInputGrad16() local
44 mat_b[j_i] = w + (j + j_i) * k_spatial; in ConvDwInputGrad16()
61 float16x8_t mat_b0 = {mat_b[0][k], mat_b[1][k], mat_b[2][k], mat_b[3][k], in ConvDwInputGrad16()
62 mat_b[4][k], mat_b[5][k], mat_b[6][k], mat_b[7][k]}; in ConvDwInputGrad16()
63 float16x8_t mat_b1 = {mat_b[8][k], mat_b[9][k], mat_b[10][k], mat_b[11][k], in ConvDwInputGrad16()
64 mat_b[12][k], mat_b[13][k], mat_b[14][k], mat_b[15][k]}; in ConvDwInputGrad16()
80 … : "r"(mat_b[0] + k), "r"(mat_b[1] + k), "r"(mat_b[2] + k), "r"(mat_b[3] + k), "r"(mat_b[4] + k), in ConvDwInputGrad16()
81 "r"(mat_b[5] + k), "r"(mat_b[6] + k), "r"(mat_b[7] + k) in ConvDwInputGrad16()
93 …: "r"(mat_b[8] + k), "r"(mat_b[9] + k), "r"(mat_b[10] + k), "r"(mat_b[11] + k), "r"(mat_b[12] + k), in ConvDwInputGrad16()
94 "r"(mat_b[13] + k), "r"(mat_b[14] + k), "r"(mat_b[15] + k) in ConvDwInputGrad16()
[all …]
|
| D | gemm_fp16.h | 31 float16_t *mat_b; member
35 … const float16_t *mat_b, int ldb, float16_t beta, float16_t *mat_c, int ldc, float16_t *workspace);
37 const float16_t *mat_b, int ldb, float16_t beta, float16_t *mat_c, int ldc,
|
| D | gemm_fp16.c | 339 …const float16_t *mat_b, int ldb, float16_t beta, float16_t *mat_c, int ldc, float16_t *workspace) { in GemmMatmulFp16() argument
345 …GemmMatmulPlusFp16(ta, tb, M, N, K, alpha, mat_a, lda, mat_b, ldb, beta, mat_c, ldc, workspace, &g… in GemmMatmulFp16()
349 const float16_t *mat_b, int ldb, float16_t beta, float16_t *mat_c, int ldc, in GemmMatmulPlusFp16() argument
360 float16_t *mat_b_input = (float16_t *)mat_b; in GemmMatmulPlusFp16()
375 RowMajor2Col8MajorStrideFp16(mat_b, mat_b_input, N, K, ldb); in GemmMatmulPlusFp16()
377 RowMajor2Row8MajorStrideFp16(mat_b, mat_b_input, K, N, ldb); in GemmMatmulPlusFp16()
384 gcb->mat_b = mat_b_input; in GemmMatmulPlusFp16()
|
| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/ |
| D | gemm.h | 31 float *mat_b; member
34 …(int ta, int tb, int M, int N, int K, float alpha, const float *mat_a, int lda, const float *mat_b,
36 …(int ta, int tb, int M, int N, int K, float alpha, const float *mat_a, int lda, const float *mat_b,
|
| D | gemm.c | 539 …(int ta, int tb, int M, int N, int K, float alpha, const float *mat_a, int lda, const float *mat_b, argument
547 gcb.mat_b = NULL;
548 GemmMatmulPlus(ta, tb, M, N, K, alpha, mat_a, lda, mat_b, ldb, beta, mat_c, ldc, workspace, &gcb);
551 …(int ta, int tb, int M, int N, int K, float alpha, const float *mat_a, int lda, const float *mat_b, argument
562 float *mat_b_input = (float *)mat_b;
589 RowMajor2Col8MajorStride(mat_b, mat_b_input, N, K, ldb);
591 RowMajor2Row8MajorStride(mat_b, mat_b_input, K, N, ldb);
602 gcb->mat_b = mat_b_input;
|
| D | convolution_grad_input.c | 62 float32x4_t mat_b = {mat_b_0[k], mat_b_1[k], mat_b_2[k], mat_b_3[k]}; in ConvDwInputGrad() local
64 mat_c = vmlaq_f32(mat_c, mat_b, mat_a); in ConvDwInputGrad()
|
| /third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp32_grad/ |
| D | convolution.cc | 108 const float *mat_b = nullptr; in Execute() local
119 mat_b = w_addr + j * nweights / groups; in Execute()
121 …GemmMatmul(0, 1, real_chunk, n, k, 1, mat_a + (j * kernel_spatial), k * groups, mat_b, k, 0, mat_c… in Execute()
132 mat_b = w_addr + j * nweights / groups; in Execute()
136 … GemmMatmul(0, 1, real_chunk, n, k, 1, mat_a, k, mat_b, k, 0, mat_c, out_ch, mat_workspace); in Execute()
141 mat_b = w_addr; in Execute()
151 … GemmMatmul(0, 1, real_chunk, n, k, 1, im + offset, k, mat_b, k, 0, mat_c, out_ch, mat_workspace); in Execute()
|
| D | convolution_grad_input.cc | 129 float *mat_b = nullptr; in Execute() local
131 mat_b = w_addr + j * nweights / groups; in Execute()
137 mat_b = gcb.mat_b; in Execute()
144 …GemmMatmulPlus(0, 0, real_chunk, n, k, 1, mat_a, out_ch, mat_b, n, 0, mat_c, n, mat_workspace, &gc… in Execute()
150 …GemmMatmulPlus(0, 0, real_chunk, n, k, 1, mat_a, out_ch, mat_b, n, 0, mat_c, n, mat_workspace, &gc… in Execute()
|
| D | convolution_grad_filter.cc | 122 float *mat_b = nullptr; in Execute() local
142 mat_b = workspace_temp + task_id * ws_size_; in Execute()
144 RollingIm2ColPackDwUnitFp32(im, conv_param, mat_b, real_chunk, ci); in Execute()
148 …GemmMatmul(1, 0, k, n, real_chunk, 1, mat_a, out_ch, mat_b + (j * kernel_spatial), n * groups, 1, … in Execute()
160 mat_b = workspace_temp + task_id * ws_size_; in Execute()
163 RollingIm2ColPackUnitFp32(im, conv_param, mat_b, real_chunk, ci); in Execute()
164 … GemmMatmul(1, 0, k, n, real_chunk, 1, mat_a, out_ch, mat_b, n, 0, mat_tmp, n, mat_workspace); in Execute()
|
| D | deconvolution_grad_filter.cc | 112 float *mat_b = workspace_temp; in Execute() local
115 rolling_im2row_hwc(im, mat_b, conv_param, real_chunk, ci); in Execute()
116 …GemmMatmul(0, 0, n, m, real_chunk * in_w, 1, mat_b, real_chunk * in_w, mat_a, in_ch, 1, mat_c, in_… in Execute()
|
| /third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp16_grad/ |
| D | convolution_fp16_grad_input.cc | 125 float16_t *mat_b = nullptr; in DoExecute() local
127 mat_b = w_addr + j * nweights / groups; in DoExecute()
133 mat_b = gcb.mat_b; in DoExecute()
140 …GemmMatmulPlusFp16(0, 0, real_chunk, n, k, 1, mat_a, out_ch, mat_b, n, 0, mat_c, n, mat_workspace,… in DoExecute()
147 …GemmMatmulPlusFp16(0, 0, real_chunk, n, k, 1, mat_a, out_ch, mat_b, n, 0, mat_c, n, mat_workspace,… in DoExecute()
|
| D | convolution_fp16_grad_filter.cc | 135 float16_t *mat_b = workspace_temp + task_id * ws_size_; in DoExecute()
137 RollingIm2ColPackDwUnitFp16(im, conv_param, mat_b, real_chunk, ci); in DoExecute()
141 …GemmMatmulFp16(1, 0, k, n, real_chunk, 1, mat_a, out_ch, mat_b + (j * kernel_spatial), n * groups,… in DoExecute()
153 float16_t *mat_b = workspace_temp + task_id * ws_size_; in DoExecute() local
156 RollingIm2ColPackUnitFp16(im, conv_param, mat_b, real_chunk, ci); in DoExecute()
157 … GemmMatmulFp16(1, 0, k, n, real_chunk, 1, mat_a, out_ch, mat_b, n, 0, mat_tmp, n, mat_workspace); in DoExecute()
|
| /third_party/boost/libs/qvm/test/ |
| D | cmp_mm_test.cpp | 63 typedef boost::qvm::mat<scalar_b, 3, 3> mat_b; in test2() typedef
65 mat_b const b = { { {42, 94, 96}, {72, 95, 81}, {12, 84, 33} } }; in test2()
67 mat_b const d = { { {21, 47, 48}, {36, 47, 65}, {79, 27, 41} } }; in test2()
|