| /dports/misc/mxnet/incubator-mxnet-1.9.0/3rdparty/mkldnn/src/cpu/ |
| H A D | gemm_x8s8s32x_convolution.cpp | 208 const int nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr() local
210 const size_t work_amount = (size_t)jcp.ngroups * jcp.mb * nb_oh * nb_ow; in execute_forward_thr()
212 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr()
289 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr()
|
| H A D | gemm_convolution.cpp | 100 const int nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr_nspc() local
104 = static_cast<size_t>(jcp.mb) * jcp.ngroups * nb_oh * nb_ow; in execute_forward_thr_nspc()
106 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
218 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
|
| /dports/math/onednn/oneDNN-2.5.1/src/cpu/ |
| H A D | gemm_x8s8s32x_convolution.cpp | 204 const dim_t nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr() local
206 const dim_t work_amount = jcp.ngroups * jcp.mb * nb_oh * nb_ow; in execute_forward_thr()
208 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr()
300 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr()
|
| H A D | gemm_convolution.cpp | 98 const dim_t nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr_nspc() local
101 const dim_t work_amount = jcp.mb * jcp.ngroups * nb_oh * nb_ow; in execute_forward_thr_nspc()
103 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
215 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
|
| /dports/misc/mxnet/incubator-mxnet-1.9.0/3rdparty/mkldnn/src/cpu/x64/ |
| H A D | jit_uni_dw_convolution.cpp | 359 const int nb_oh = div_up(jcp.oh, jcp.oh_blk_size); in execute_backward_weights_nxc() local
361 balance211(nb_oh, jcp.nthr_oh, ithr_oh, nb_oh_start, nb_oh_end); in execute_backward_weights_nxc()
394 for (int nb_oh = nb_oh_start; nb_oh < nb_oh_end; ++nb_oh) { in execute_backward_weights_nxc() local
395 const int oh_s = nb_oh * h_block_size; in execute_backward_weights_nxc()
|
| H A D | jit_brgemm_conv_utils.cpp | 558 nb_oh = div_up(oh, oh_blk_size); in update_blocks()
607 const auto sp_amount = nb_od * nb_oh * nb_sp; in est_eff()
724 const auto nb_oh_thr = nstl::min(nb_oh, div_up(job, dim_oh)); in est_eff()
727 const auto dim_od = nb_oh * dim_oh; in est_eff()
972 : nb_od * nb_oh * nb_sp; in est_eff_1x1()
991 const auto nb_oh_thr = nstl::min(nb_oh, div_up(job, dim_oh)); in est_eff_1x1()
995 const auto dim_od = nb_oh * dim_oh; in est_eff_1x1()
1574 jcp.nb_oh = div_up(jcp.oh, jcp.oh_blk_size); in init_conf()
1581 jcp.inp_buffer_mask_size = rnd_up((dim_t)jcp.nb_od * jcp.nb_oh in init_conf()
|
| H A D | jit_brgemm_conv.cpp | 524 * jcp.nb_oh * jcp.nb_ow; in execute()
583 BRGC_WO(n, jcp.mb, odb, jcp.nb_od, ohb, jcp.nb_oh, owb, jcp.nb_ow, g, in execute()
587 jcp.nb_oh, owb, jcp.nb_ow) in execute()
692 inp_buffer_mask[(((icb)*jcp.nb_od + (odb)) * jcp.nb_oh + (ohb)) \ in maybe_conv_inp()
|
| H A D | gemm_bf16_convolution.cpp | 446 const int nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr_nspc() local
448 const size_t work_amount = (size_t)jcp.ngroups * jcp.mb * nb_oh * nb_ow; in execute_forward_thr_nspc()
450 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
527 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
|
| H A D | jit_primitive_conf.hpp | 800 nb_oh; // blocking - included in parallelization member
|
| /dports/math/onednn/oneDNN-2.5.1/src/cpu/x64/ |
| H A D | jit_uni_dw_convolution.cpp | 362 const int nb_oh = div_up(jcp.oh, jcp.oh_blk_size); in execute_backward_weights_nxc() local
364 balance211(nb_oh, jcp.nthr_oh, ithr_oh, nb_oh_start, nb_oh_end); in execute_backward_weights_nxc()
396 for (int nb_oh = nb_oh_start; nb_oh < nb_oh_end; ++nb_oh) { in execute_backward_weights_nxc() local
397 const int oh_s = nb_oh * h_block_size; in execute_backward_weights_nxc()
|
| H A D | jit_brgemm_conv_utils.cpp | 711 nb_oh = div_up(oh, oh_block); in update_blocks()
761 const auto sp_amount = nb_od * nb_oh * nb_sp; in est_eff()
880 const auto nb_oh_thr = nstl::min(nb_oh, div_up(job, dim_oh)); in est_eff()
883 const auto dim_od = nb_oh * dim_oh; in est_eff()
1216 : nb_od * nb_oh * nb_sp; in est_eff_1x1()
1236 const auto nb_oh_thr = nstl::min(nb_oh, div_up(job, dim_oh)); in est_eff_1x1()
1240 const auto dim_od = nb_oh * dim_oh; in est_eff_1x1()
1908 jcp.nb_oh = div_up(jcp.oh, jcp.oh_block); in init_conf()
1929 * jcp.nb_oh * jcp.nb_ow * jcp.ngroups * jcp.nb_ic, in init_conf()
|
| H A D | jit_brgemm_conv.cpp | 618 * jcp.nb_oc * jcp.nb_od * jcp.nb_oh * jcp.nb_ow; in execute()
646 nd_iterator_init(start, n, jcp.mb, odb, jcp.nb_od, ohb, jcp.nb_oh, in execute()
650 odb, jcp.nb_od, ohb, jcp.nb_oh, owb, jcp.nb_ow); in execute()
712 nd_iterator_step(n, jcp.mb, odb, jcp.nb_od, ohb, jcp.nb_oh, owb, in execute()
716 jcp.nb_od, ohb, jcp.nb_oh, owb, jcp.nb_ow); in execute()
846 inp_buffer_mask[(((icb)*jcp.nb_od + (odb)) * jcp.nb_oh + (ohb)) \ in maybe_conv_inp()
|
| H A D | gemm_bf16_convolution.cpp | 436 const dim_t nb_oh = div_up(jcp.oh, jcp.oh_block); in execute_forward_thr_nspc() local
438 const dim_t work_amount = jcp.ngroups * jcp.mb * nb_oh * nb_ow; in execute_forward_thr_nspc()
440 nd_iterator_init(start, n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
517 nd_iterator_step(n, jcp.mb, g, jcp.ngroups, ohb, nb_oh, owb, nb_ow); in execute_forward_thr_nspc()
|
| H A D | jit_primitive_conf.hpp | 842 nb_oh; // blocking - included in parallelization member
|