/dports/misc/mxnet/incubator-mxnet-1.9.0/3rdparty/mkldnn/src/cpu/x64/ |
H A D | jit_avx512_core_f32_wino_conv_4x3.cpp | 293 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_S_G_D() 294 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_S_G_D() 397 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_SGD() 398 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_SGD() 658 alpha, alpha, jcp.oc_block, jcp.ic_block, jcp.ic_simd_block, in _execute_backward_weights_SDGtWo() 673 jcp.ic_simd_block); in _execute_backward_weights_SDGtWo() 816 jcp.ic_simd_block, jcp.oc_reg_block, jcp.oc_simd_block); in _execute_backward_weights_S_D_Giot_W() 821 alpha, alpha, jcp.oc_block, jcp.ic_block, jcp.ic_simd_block, in _execute_backward_weights_S_D_Giot_W() 831 jcp.nb_tile_block_ur, jcp.tile_block_ur, jcp.ic_simd_block); in _execute_backward_weights_S_D_Giot_W() 913 * jcp.ic_simd_block * jcp.oc_reg_block in _execute_backward_weights_S_D_Giot_W() [all …]
|
H A D | jit_avx512_common_convolution_winograd.cpp | 513 jcp.ic / jcp.ic_simd_block, jcp.kh, jcp.kw, simd_w, simd_w); in weight_transform_data() 650 conv.tile_block_ur, conv.ic_simd_block * conv.tile_4fma); in diff_src_transform_bwd_weights() 857 alpha, conv.oc_block, conv.ic_block, conv.ic_simd_block, in diff_weights_transform_bwd_weights() 860 conv.ic / simd_w, conv.kh, conv.kw, conv.ic_simd_block, in diff_weights_transform_bwd_weights() 871 for (int v = 0; v < conv.ic_simd_block; v++) { in diff_weights_transform_bwd_weights() 884 for (int v = 0; v < conv.ic_simd_block; v++) { in diff_weights_transform_bwd_weights() 934 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_S_G_D() 935 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_S_G_D() 1065 jcp.ic_simd_block, jcp.oc_simd_block); in _execute_backward_weights_S_D_G_W() 1074 jcp.ic_simd_block * jcp.tile_4fma); in _execute_backward_weights_S_D_G_W() [all …]
|
H A D | jit_avx512_common_conv_winograd_kernel_f32.cpp | 572 jcp.ic_simd_block = jcp.dimK_reg_block; in init_conf() 603 jcp.ic_simd_block = jcp.dimM_simd_block; in init_conf() 918 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_S_D_G_W_avx512_common() 1096 jcp.ic_simd_block = simd_w; in init_conf()
|
H A D | jit_avx512_common_convolution_winograd.hpp | 55 * jcp.tile_4fma * jcp.ic_simd_block; in init_scratchpad()
|
H A D | jit_avx512_core_f32_wino_conv_4x3_kernel.cpp | 1432 jcp.ic_simd_block = jcp.dimK_reg_block; in init_conf() 1492 jcp.ic_simd_block = jcp.dimM_simd_block; in init_conf() 2346 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_SDGtWo() 2390 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_S_D_Giot_W() 2543 jcp.ic_simd_block = simd_w; in init_conf() 2550 jcp.dimN_reg_block = jcp.ic_simd_block; in init_conf()
|
H A D | jit_primitive_conf.hpp | 378 int ic_simd_block = 16; member
|
/dports/math/onednn/oneDNN-2.5.1/src/cpu/x64/ |
H A D | jit_avx512_core_f32_wino_conv_4x3.cpp | 293 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_S_G_D() 294 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_S_G_D() 397 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_SGD() 398 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_SGD() 658 alpha, alpha, jcp.oc_block, jcp.ic_block, jcp.ic_simd_block, in _execute_backward_weights_SDGtWo() 673 jcp.ic_simd_block); in _execute_backward_weights_SDGtWo() 816 jcp.ic_simd_block, jcp.oc_reg_block, jcp.oc_simd_block); in _execute_backward_weights_S_D_Giot_W() 821 alpha, alpha, jcp.oc_block, jcp.ic_block, jcp.ic_simd_block, in _execute_backward_weights_S_D_Giot_W() 831 jcp.nb_tile_block_ur, jcp.tile_block_ur, jcp.ic_simd_block); in _execute_backward_weights_S_D_Giot_W() 914 * jcp.ic_simd_block * jcp.oc_reg_block in _execute_backward_weights_S_D_Giot_W() [all …]
|
H A D | jit_avx512_common_convolution_winograd.cpp | 513 jcp.ic / jcp.ic_simd_block, jcp.kh, jcp.kw, simd_w, simd_w); in weight_transform_data() 650 conv.tile_block_ur, conv.ic_simd_block * conv.tile_4fma); in diff_src_transform_bwd_weights() 857 alpha, conv.oc_block, conv.ic_block, conv.ic_simd_block, in diff_weights_transform_bwd_weights() 860 conv.ic / simd_w, conv.kh, conv.kw, conv.ic_simd_block, in diff_weights_transform_bwd_weights() 871 for (int v = 0; v < conv.ic_simd_block; v++) { in diff_weights_transform_bwd_weights() 884 for (int v = 0; v < conv.ic_simd_block; v++) { in diff_weights_transform_bwd_weights() 934 jcp.oc / jcp.oc_simd_block, jcp.ic / jcp.ic_simd_block, jcp.kh, in _execute_data_W_S_G_D() 935 jcp.kw, jcp.ic_simd_block, jcp.oc_simd_block); in _execute_data_W_S_G_D() 1065 jcp.ic_simd_block, jcp.oc_simd_block); in _execute_backward_weights_S_D_G_W() 1074 jcp.ic_simd_block * jcp.tile_4fma); in _execute_backward_weights_S_D_G_W() [all …]
|
H A D | jit_avx512_common_conv_winograd_kernel_f32.cpp | 572 jcp.ic_simd_block = jcp.dimK_reg_block; in init_conf() 603 jcp.ic_simd_block = jcp.dimM_simd_block; in init_conf() 918 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_S_D_G_W_avx512_common() 1096 jcp.ic_simd_block = simd_w; in init_conf()
|
H A D | jit_avx512_common_convolution_winograd.hpp | 55 * jcp.tile_4fma * jcp.ic_simd_block; in init_scratchpad()
|
H A D | jit_avx512_core_f32_wino_conv_4x3_kernel.cpp | 1432 jcp.ic_simd_block = jcp.dimK_reg_block; in init_conf() 1492 jcp.ic_simd_block = jcp.dimM_simd_block; in init_conf() 2344 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_SDGtWo() 2387 jcp.dimN_reg_block = jcp.ic_simd_block; in set_wsched_WEI_S_D_Giot_W() 2540 jcp.ic_simd_block = simd_w; in init_conf() 2547 jcp.dimN_reg_block = jcp.ic_simd_block; in init_conf()
|
H A D | jit_primitive_conf.hpp | 380 int ic_simd_block = 16; member
|