| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 211 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
311 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
391 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
505 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
606 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
684 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
757 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
946 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1081 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1162 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 283 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
429 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
534 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
686 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
819 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
915 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1006 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1033 _sum0 = vmlaq_f32(_sum0, _k12, _r12); in convdw3x3s1_pack4_bf16s_neon()
1248 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
1356 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s1_pack8_fp16sa_neon() local
207 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
302 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
381 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
486 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
553 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
610 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
660 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s2_pack8_fp16sa_neon() local
764 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
839 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 50 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s1_pack1to4_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
432 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s2_pack1to4_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s1_pack1to8_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
432 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s2_pack1to8_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
631 "w"(_k12), // %13 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 | 211 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
311 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
391 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
505 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
606 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
684 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
757 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
946 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1081 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1162 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 283 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
429 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
534 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
686 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
819 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
915 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1006 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1033 _sum0 = vmlaq_f32(_sum0, _k12, _r12); in convdw3x3s1_pack4_bf16s_neon()
1248 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
1356 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s1_pack8_fp16sa_neon() local
207 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
302 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
381 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
486 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
553 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
610 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
660 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s2_pack8_fp16sa_neon() local
764 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
839 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s1_pack1to8_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
432 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s2_pack1to8_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 50 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s1_pack1to4_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
432 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s2_pack1to4_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 211 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
311 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
391 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
505 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
606 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
684 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
757 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
946 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1081 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1162 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 283 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
429 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
534 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
686 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
819 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
915 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1006 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1033 _sum0 = vmlaq_f32(_sum0, _k12, _r12); in convdw3x3s1_pack4_bf16s_neon()
1248 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
1356 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s1_pack8_fp16sa_neon() local
207 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
302 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
381 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
486 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
553 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
610 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
660 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s2_pack8_fp16sa_neon() local
764 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
839 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 50 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s1_pack1to4_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
432 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s2_pack1to4_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s1_pack1to8_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
432 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s2_pack1to8_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack4.h | 211 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
311 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
391 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
505 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
606 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
684 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
757 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
946 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1081 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1162 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 283 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
429 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
534 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
686 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
819 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
915 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1006 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1033 _sum0 = vmlaq_f32(_sum0, _k12, _r12); in convdw3x3s1_pack4_bf16s_neon()
1248 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
1356 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s1_pack8_fp16sa_neon() local
207 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
302 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
381 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
486 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
553 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
610 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
660 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s2_pack8_fp16sa_neon() local
764 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
839 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 50 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s1_pack1to4_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to4_fp16sa_neon()
432 float16x4_t _k12 = vld1_f16(k0 + 20); in conv3x3s2_pack1to4_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s1_pack1to8_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
432 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s2_pack1to8_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
631 "w"(_k12), // %13 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 | 211 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
311 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
391 "w"(_k12), // %17 in convdw3x3s1_pack4_neon()
505 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
606 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
684 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
757 "w"(_k12), // %13 in convdw3x3s1_pack4_neon()
946 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1081 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
1162 "w"(_k12), // %13 in convdw3x3s2_pack4_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 283 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
429 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
534 "w"(_k12), // %17 in convdw3x3s1_pack4_bf16s_neon()
686 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
819 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
915 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1006 "w"(_k12), // %13 in convdw3x3s1_pack4_bf16s_neon()
1033 _sum0 = vmlaq_f32(_sum0, _k12, _r12); in convdw3x3s1_pack4_bf16s_neon()
1248 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
1356 "w"(_k12), // %13 in convdw3x3s2_pack4_bf16s_neon()
[all …]
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s1_pack8_fp16sa_neon() local
207 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
302 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
381 "w"(_k12), // %17 in convdw3x3s1_pack8_fp16sa_neon()
486 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
553 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
610 "w"(_k12), // %13 in convdw3x3s1_pack8_fp16sa_neon()
660 float16x8_t _k12 = vld1q_f16(k0 + 40); in convdw3x3s2_pack8_fp16sa_neon() local
764 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
839 "w"(_k12), // %13 in convdw3x3s2_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 50 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s1_pack1to8_fp16sa_neon() local
179 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
264 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
325 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
377 "w"(_k12), // %13 in conv3x3s1_pack1to8_fp16sa_neon()
432 float16x8_t _k12 = vld1q_f16(k0 + 40); in conv3x3s2_pack1to8_fp16sa_neon() local
519 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
579 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
631 "w"(_k12), // %13 in conv3x3s2_pack1to8_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_pack1ton.h | 53 vfloat32m1_t _k12 = vle32_v_f32m1(k0 + packn * 5, vl); in conv3x3s1_pack1ton_rvv() local
114 _sum0 = vfmacc_vf_f32m1(_sum0, r1[2], _k12, vl); in conv3x3s1_pack1ton_rvv()
115 _sum1 = vfmacc_vf_f32m1(_sum1, r1[3], _k12, vl); in conv3x3s1_pack1ton_rvv()
116 _sum2 = vfmacc_vf_f32m1(_sum2, r1[4], _k12, vl); in conv3x3s1_pack1ton_rvv()
117 _sum3 = vfmacc_vf_f32m1(_sum3, r1[5], _k12, vl); in conv3x3s1_pack1ton_rvv()
118 _sum4 = vfmacc_vf_f32m1(_sum4, r1[6], _k12, vl); in conv3x3s1_pack1ton_rvv()
119 _sum5 = vfmacc_vf_f32m1(_sum5, r1[7], _k12, vl); in conv3x3s1_pack1ton_rvv()
120 _sum6 = vfmacc_vf_f32m1(_sum6, r1[8], _k12, vl); in conv3x3s1_pack1ton_rvv()
121 _sum7 = vfmacc_vf_f32m1(_sum7, r1[9], _k12, vl); in conv3x3s1_pack1ton_rvv()
191 _sum0 = vfmacc_vf_f32m1(_sum0, r1[2], _k12, vl); in conv3x3s1_pack1ton_rvv()
[all …]
|