| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_pack1ton.h | 69 vfloat32m1_t _sum5 = vle32_v_f32m1(outptr0 + packn * 5, vl); in conv3x3s1_pack1ton_rvv() local
347 vfloat32m1_t _sum5 = vle32_v_f32m1(outptr0 + packn * 5, vl); in conv3x3s2_pack1ton_rvv() local
|
| H A D | convolution_3x3_pack1ton_fp16s.h | 69 vfloat16m1_t _sum5 = vle16_v_f16m1(outptr0 + packn * 5, vl); in conv3x3s1_pack1ton_fp16sa_rvv() local
347 vfloat16m1_t _sum5 = vle16_v_f16m1(outptr0 + packn * 5, vl); in conv3x3s2_pack1ton_fp16sa_rvv() local
|
| H A D | convolution_sgemm_packnto1.h | 216 vfloat32m1_t _sum5 = vle32_v_f32m1(biasptr, vl); in im2col_sgemm_packnto1_rvv() local
376 vfloat32m1_t _sum5 = vfmv_v_f_f32m1(0.f, vl); in im2col_sgemm_packnto1_rvv() local
|
| H A D | convolution_sgemm_packnto1_fp16s.h | 216 vfloat16m1_t _sum5 = vle16_v_f16m1(biasptr, vl); in im2col_sgemm_packnto1_fp16sa_rvv() local
376 vfloat16m1_t _sum5 = vfmv_v_f_f16m1(0.f, vl); in im2col_sgemm_packnto1_fp16sa_rvv() local
|
| H A D | convolution_sgemm_packn.h | 209 vfloat32m1_t _sum5 = vfmv_v_f_f32m1(0.f, vl); in im2col_sgemm_packn_rvv() local
|
| H A D | convolution_sgemm_packn_fp16s.h | 241 vfloat16m1_t _sum5 = vfmv_v_f_f16m1(0.f, vl); in im2col_sgemm_packn_fp16sa_rvv() local
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
| H A D | innerproduct_x86.cpp | 173 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
332 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
915 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1017 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
1155 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1478 __m256 _sum5 = _mm256_set1_ps(0.f); in forward_fp16() local
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/x86/ |
| H A D | innerproduct_x86.cpp | 180 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
339 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
899 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1001 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
1139 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1437 __m256 _sum5 = _mm256_set1_ps(0.f); in forward_fp16() local
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/x86/ |
| H A D | innerproduct_x86.cpp | 180 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
339 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
899 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1001 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
1139 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1437 __m256 _sum5 = _mm256_set1_ps(0.f); in forward_fp16() local
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/x86/ |
| H A D | innerproduct_x86.cpp | 180 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
339 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
899 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1001 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
1139 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1437 __m256 _sum5 = _mm256_set1_ps(0.f); in forward_fp16() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/x86/ |
| H A D | innerproduct_x86.cpp | 180 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
339 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
899 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1001 __m128 _sum5 = _mm_set1_ps(0.f); in forward() local
1139 __m256 _sum5 = _mm256_set1_ps(0.f); in forward() local
1437 __m256 _sum5 = _mm256_set1_ps(0.f); in forward_fp16() local
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to4_int8.h | 905 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1121 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1739 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1866 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
|
| H A D | innerproduct_arm.cpp | 834 float16x8_t _sum5 = vdupq_n_f16((__fp16)0.f); in forward_fp16sa() local
935 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
1251 float16x8_t _sum5 = vdupq_n_f16(0.f); in forward_fp16sa() local
1374 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
|
| H A D | convolution_1x1_fp16s.h | 165 float16x8_t _sum5 = vdupq_laneq_f16(_bias0, 5); in conv1x1s1_sgemm_fp16sa_neon() local
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to4_int8.h | 905 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1121 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1739 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1866 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
|
| H A D | innerproduct_arm.cpp | 834 float16x8_t _sum5 = vdupq_n_f16((__fp16)0.f); in forward_fp16sa() local
935 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
1251 float16x8_t _sum5 = vdupq_n_f16(0.f); in forward_fp16sa() local
1374 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
|
| H A D | convolution_1x1_fp16s.h | 165 float16x8_t _sum5 = vdupq_laneq_f16(_bias0, 5); in conv1x1s1_sgemm_fp16sa_neon() local
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to4_int8.h | 905 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1121 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1739 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1866 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
|
| H A D | innerproduct_arm.cpp | 834 float16x8_t _sum5 = vdupq_n_f16((__fp16)0.f); in forward_fp16sa() local
935 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
1251 float16x8_t _sum5 = vdupq_n_f16(0.f); in forward_fp16sa() local
1374 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
|
| H A D | convolution_1x1_fp16s.h | 165 float16x8_t _sum5 = vdupq_laneq_f16(_bias0, 5); in conv1x1s1_sgemm_fp16sa_neon() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_pack8to4_int8.h | 905 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1121 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1739 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
1866 int32x4_t _sum5 = vdupq_n_s32(0); in conv3x3s1_winograd42_pack8to4_int8_neon() local
|
| H A D | innerproduct_arm.cpp | 837 float16x8_t _sum5 = vdupq_n_f16((__fp16)0.f); in forward_fp16sa() local
938 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
1254 float16x8_t _sum5 = vdupq_n_f16(0.f); in forward_fp16sa() local
1377 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/mips/ |
| H A D | convolution_sgemm_pack4.h | 287 v4f32 _sum5 = _sum0; in im2col_sgemm_pack4_msa() local
347 v4f32 _sum5 = _sum0; in im2col_sgemm_pack4_msa() local
|
| H A D | convolution_sgemm_pack4to1.h | 264 v4f32 _sum5 = (v4f32)__msa_splati_w(_bias, 1); in im2col_sgemm_pack4to1_msa() local
328 v4f32 _sum5 = (v4f32)__msa_splati_w(_bias, 2); in im2col_sgemm_pack4to1_msa() local
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | innerproduct_arm.cpp | 1016 float16x8_t _sum5 = vdupq_n_f16((__fp16)0.f); in forward_fp16sa() local
1117 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
1456 float16x8_t _sum5 = vdupq_n_f16(0.f); in forward_fp16sa() local
1579 float16x4_t _sum5 = vdup_n_f16(0.f); in forward_fp16sa() local
|