| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 209 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
309 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
389 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
503 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
604 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
682 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
755 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
944 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1079 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1160 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 281 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
427 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
532 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
684 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
817 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
913 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1004 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1031 _sum0 = vmlaq_f32(_sum0, _k10, _r10); in convdw3x3s1_pack4_bf16s_neon()
1246 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
1354 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s1_pack8_fp16sa_neon() local
205 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
300 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
379 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
484 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
551 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
608 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
658 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s2_pack8_fp16sa_neon() local
762 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
837 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 48 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s1_pack1to4_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
430 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s2_pack1to4_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s1_pack1to8_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
430 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s2_pack1to8_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 209 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
309 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
389 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
503 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
604 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
682 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
755 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
944 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1079 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1160 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 281 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
427 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
532 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
684 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
817 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
913 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1004 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1031 _sum0 = vmlaq_f32(_sum0, _k10, _r10); in convdw3x3s1_pack4_bf16s_neon()
1246 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
1354 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s1_pack8_fp16sa_neon() local
205 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
300 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
379 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
484 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
551 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
608 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
658 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s2_pack8_fp16sa_neon() local
762 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
837 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s1_pack1to8_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
430 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s2_pack1to8_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 48 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s1_pack1to4_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
430 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s2_pack1to4_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 209 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
309 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
389 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
503 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
604 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
682 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
755 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
944 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1079 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1160 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 281 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
427 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
532 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
684 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
817 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
913 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1004 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1031 _sum0 = vmlaq_f32(_sum0, _k10, _r10); in convdw3x3s1_pack4_bf16s_neon()
1246 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
1354 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s1_pack8_fp16sa_neon() local
205 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
300 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
379 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
484 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
551 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
608 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
658 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s2_pack8_fp16sa_neon() local
762 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
837 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 48 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s1_pack1to4_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
430 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s2_pack1to4_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s1_pack1to8_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
430 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s2_pack1to8_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 209 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
309 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
389 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
503 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
604 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
682 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
755 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
944 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1079 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1160 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 281 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
427 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
532 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
684 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
817 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
913 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1004 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1031 _sum0 = vmlaq_f32(_sum0, _k10, _r10); in convdw3x3s1_pack4_bf16s_neon()
1246 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
1354 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s1_pack8_fp16sa_neon() local
205 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
300 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
379 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
484 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
551 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
608 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
658 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s2_pack8_fp16sa_neon() local
762 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
837 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 48 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s1_pack1to4_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to4_fp16sa_neon()
430 float16x4_t _k10 = vld1_f16(k0 + 12); in conv3x3s2_pack1to4_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s1_pack1to8_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
430 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s2_pack1to8_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 209 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
309 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
389 "w"(_k10), // %15 in convdw3x3s1_pack4_neon()
503 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
604 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
682 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
755 "w"(_k10), // %11 in convdw3x3s1_pack4_neon()
944 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1079 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
1160 "w"(_k10), // %11 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 281 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
427 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
532 "w"(_k10), // %15 in convdw3x3s1_pack4_bf16s_neon()
684 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
817 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
913 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1004 "w"(_k10), // %11 in convdw3x3s1_pack4_bf16s_neon()
1031 _sum0 = vmlaq_f32(_sum0, _k10, _r10); in convdw3x3s1_pack4_bf16s_neon()
1246 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
1354 "w"(_k10), // %11 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s1_pack8_fp16sa_neon() local
205 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
300 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
379 "w"(_k10), // %15 in convdw3x3s1_pack8_fp16sa_neon()
484 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
551 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
608 "w"(_k10), // %11 in convdw3x3s1_pack8_fp16sa_neon()
658 float16x8_t _k10 = vld1q_f16(k0 + 24); in convdw3x3s2_pack8_fp16sa_neon() local
762 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
837 "w"(_k10), // %11 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 48 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s1_pack1to8_fp16sa_neon() local
177 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
262 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
323 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
375 "w"(_k10), // %11 in conv3x3s1_pack1to8_fp16sa_neon()
430 float16x8_t _k10 = vld1q_f16(k0 + 24); in conv3x3s2_pack1to8_fp16sa_neon() local
517 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
577 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
629 "w"(_k10), // %11 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_pack1ton.h | 51 vfloat32m1_t _k10 = vle32_v_f32m1(k0 + packn * 3, vl); in conv3x3s1_pack1ton_rvv() local
98 _sum0 = vfmacc_vf_f32m1(_sum0, r1[0], _k10, vl); in conv3x3s1_pack1ton_rvv()
99 _sum1 = vfmacc_vf_f32m1(_sum1, r1[1], _k10, vl); in conv3x3s1_pack1ton_rvv()
100 _sum2 = vfmacc_vf_f32m1(_sum2, r1[2], _k10, vl); in conv3x3s1_pack1ton_rvv()
101 _sum3 = vfmacc_vf_f32m1(_sum3, r1[3], _k10, vl); in conv3x3s1_pack1ton_rvv()
102 _sum4 = vfmacc_vf_f32m1(_sum4, r1[4], _k10, vl); in conv3x3s1_pack1ton_rvv()
103 _sum5 = vfmacc_vf_f32m1(_sum5, r1[5], _k10, vl); in conv3x3s1_pack1ton_rvv()
104 _sum6 = vfmacc_vf_f32m1(_sum6, r1[6], _k10, vl); in conv3x3s1_pack1ton_rvv()
105 _sum7 = vfmacc_vf_f32m1(_sum7, r1[7], _k10, vl); in conv3x3s1_pack1ton_rvv()
183 _sum0 = vfmacc_vf_f32m1(_sum0, r1[0], _k10, vl); in conv3x3s1_pack1ton_rvv()
[all …]
|