| /third_party/mindspore/mindspore/ops/_op_impl/_custom_op/ |
| D | dsd_back_impl.py | 142 with tik_inst.for_range(0, block_num, thread_num=2) as w_idx:
167 … tik_inst.data_move(w_local_l1[0, 0, 0, 0], w1_gm[bs_idx, head_idx, w_idx, 0, 0, 0, 0], 0,
171 tik_inst.load2dv1(d_a_l0b[0, 0, 0, 0], d_a_l1[w_idx * block_size//16, 0, 0, 0], 0,
197 … tik_inst.data_move(w_global_l1[0, 0, 0, 0], w2_gm[bs_idx, head_idx, w_idx_1, w_idx, 0, 0, 0], 0,
229 tik_inst.data_move(d_v_gm[bs_idx*seq_len//16+w_idx * (block_size // 16) + h_idx,
236 with tik_inst.for_range(0, block_num, thread_num=2) as w_idx:
255 tik_inst.load2dv1(d_local_l0a[0, 0, 0, 0], d_a_l1[w_idx*(block_size//16), 0, 0, 0],
261 v_gm[bs_idx*seq_len//16+w_idx *
280 … tik_inst.data_move(d_w1_gm[bs_idx, head_idx, w_idx, 0, 0, 0, 0], d_w_local_ub[0, 0, 0, 0], 0,
289 with tik_inst.for_range(0, block_num) as w_idx:
[all …]
|
| D | dsd_impl.py | 104 with tik_inst.for_range(0, block_num, thread_num=2) as w_idx:
114 w2_gm[bs_idx, head_idx, w_idx, cpt_idx *
137 v_gm[bs_idx * seq_len//16 + w_idx * 4, head_idx *
145 … tik_inst.data_move(w1_l1[brick_i, 0, 0, 0], w1_gm[bs_idx, head_idx, w_idx, 0, brick_i, 0, 0], 0,
155 …tik_inst.data_move(output_gm[bs_idx, head_idx, 0, w_idx*(block_size//16), 0, 0], output_ub[0, 0, 0…
|
| /third_party/mesa3d/src/mesa/drivers/dri/i965/ |
| D | gfx6_queryobj.c | 124 int w_idx = 4 * i + idx; in write_xfb_overflow_streams() local
133 w_idx * sizeof(uint64_t)); in write_xfb_overflow_streams()
140 w_idx * sizeof(uint64_t)); in write_xfb_overflow_streams()
|
| /third_party/mindspore/mindspore/ccsrc/common/ |
| D | trans.cc | 1322 for (size_t w_idx = 0; w_idx < w; w_idx++) { in FracZToNchw() local
1323 size_t dst_idx = h_head_addr + w_idx; in FracZToNchw()
1327 size_t src_idx = c1_idx * hwncc0 + h_idx * wncc0 + w_idx * ncc0 + nc_idx * c0 + c0_idx; in FracZToNchw()
1585 for (size_t w_idx = 0; w_idx < w; w_idx++) { in NchwToNc1hwc0() local
1586 size_t w_head_addr = h_head_addr + w_idx * c0; in NchwToNc1hwc0()
1590 size_t src_idx = n_idx * chw + c_idx * hw + h_idx * w + w_idx; in NchwToNc1hwc0()
1638 for (size_t w_idx = 0; w_idx < w; w_idx++) { in Nc1hwc0ToNchw() local
1639 size_t dst_idx = h_head_addr + w_idx; in Nc1hwc0ToNchw()
1642 size_t src_idx = n_idx * c1hwc0 + c1_idx * hwc0 + h_idx * wc0 + w_idx * c0 + c0_idx; in Nc1hwc0ToNchw()
|
| /third_party/mesa3d/src/gallium/drivers/iris/ |
| D | iris_query.c | 261 int w_idx = offset + offsetof(struct iris_query_so_overflow, in write_overflow_values() local
266 bo, w_idx, false); in write_overflow_values()
|
| /third_party/mesa3d/src/gallium/drivers/crocus/ |
| D | crocus_query.c | 289 int w_idx = offset + offsetof(struct crocus_query_so_overflow, in write_overflow_values() local
294 bo, w_idx, false); in write_overflow_values()
|