| /dports/science/elk/elk-7.2.42/src/ |
| H A D | tm2pol.f90 | 6 subroutine tm2pol(l,k,w2,tm2p) argument
26 real(8),intent(out) :: tm2p local
|
| H A D | writetm2du.f90 | 29 real(8) tm2ptot,tm2p,tm2p0 local
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack4_bf16s.h | 288 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_bf16s_neon() local
328 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_bf16s_neon() local
389 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_bf16s_neon() local
427 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
465 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
1998 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2038 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2099 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2137 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2175 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 284 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
324 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
390 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
436 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + i % 12 % 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
|
| H A D | convolution_3x3_pack4.h | 519 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_neon() local
559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_neon() local
620 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_neon() local
658 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_neon() local
696 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_neon() local
2458 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_neon() local
2498 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_neon() local
2559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_neon() local
2597 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_neon() local
2635 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_neon() local
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 388 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
425 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
449 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
|
| H A D | convolution_3x3_pack8to1_int8.h | 316 short* tm2p = tm2.row<short>(i / 8); in conv3x3s1_winograd42_pack8to1_int8_neon() local
355 short* tm2p = tm2.row<short>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd42_pack8to1_int8_neon() local
398 short* tm2p = tm2.row<short>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd42_pack8to1_int8_neon() local
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 394 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
431 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
455 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack4_bf16s.h | 288 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_bf16s_neon() local
328 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_bf16s_neon() local
389 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_bf16s_neon() local
427 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
465 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
1998 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2038 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2099 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2137 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2175 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 284 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
324 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
390 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
436 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + i % 12 % 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
|
| H A D | convolution_3x3_pack4.h | 519 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_neon() local
559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_neon() local
620 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_neon() local
658 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_neon() local
696 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_neon() local
2458 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_neon() local
2498 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_neon() local
2559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_neon() local
2597 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_neon() local
2635 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_neon() local
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 388 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
425 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
449 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack4_bf16s.h | 288 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_bf16s_neon() local
328 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_bf16s_neon() local
389 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_bf16s_neon() local
427 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
465 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
1998 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2038 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2099 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2137 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2175 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 284 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
324 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
390 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
436 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + i % 12 % 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
|
| H A D | convolution_3x3_pack4.h | 519 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_neon() local
559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_neon() local
620 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_neon() local
658 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_neon() local
696 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_neon() local
2458 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_neon() local
2498 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_neon() local
2559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_neon() local
2597 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_neon() local
2635 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_neon() local
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 388 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
425 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
449 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_pack4_bf16s.h | 288 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_bf16s_neon() local
328 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_bf16s_neon() local
389 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_bf16s_neon() local
427 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
465 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
1998 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2038 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2099 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2137 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2175 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 284 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
324 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
390 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
436 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + i % 12 % 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
|
| H A D | convolution_3x3_pack4.h | 519 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_neon() local
559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_neon() local
620 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_neon() local
658 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_neon() local
696 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_neon() local
2458 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_neon() local
2498 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_neon() local
2559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_neon() local
2597 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_neon() local
2635 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_neon() local
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 388 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
425 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
449 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack4_bf16s.h | 288 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_bf16s_neon() local
328 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_bf16s_neon() local
389 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_bf16s_neon() local
427 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
465 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_bf16s_neon() local
1998 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2038 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2099 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2137 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
2175 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_bf16s_neon() local
|
| H A D | convolution_3x3_pack4.h | 519 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4_neon() local
559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4_neon() local
620 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4_neon() local
658 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack4_neon() local
696 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack4_neon() local
2458 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd42_pack4_neon() local
2498 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd42_pack4_neon() local
2559 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd42_pack4_neon() local
2597 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd42_pack4_neon() local
2635 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd42_pack4_neon() local
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 284 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
324 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
390 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
436 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + i % 12 % 4); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 388 __fp16* tm2p = tm2.row<__fp16>(i / 8); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
425 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
449 __fp16* tm2p = tm2.row<__fp16>(i / 8 + (i % 8) / 4 + i % 4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
| H A D | convolution_3x3_pack8.h | 498 float* tm2p = tm2.row(i / 12); in conv3x3s1_winograd64_pack8_avx() local
536 float* tm2p = tm2.row(i / 12 + (i % 12) / 8); in conv3x3s1_winograd64_pack8_avx() local
566 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4); in conv3x3s1_winograd64_pack8_avx() local
588 float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2); in conv3x3s1_winograd64_pack8_avx() local
607 … float* tm2p = tm2.row(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2); in conv3x3s1_winograd64_pack8_avx() local
|