/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 759 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() 760 _sum04 = _mm256_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to8_avx() 761 _sum04 = _mm256_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to8_avx() 762 _sum04 = _mm256_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to8_avx() 763 _sum04 = _mm256_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to8_avx() 764 _sum04 = _mm256_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to8_avx() 765 _sum04 = _mm256_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to8_avx() 766 _sum04 = _mm256_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to8_avx() 767 _sum04 = _mm256_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to8_avx() 1292 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 759 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() 760 _sum04 = _mm256_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to8_avx() 761 _sum04 = _mm256_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to8_avx() 762 _sum04 = _mm256_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to8_avx() 763 _sum04 = _mm256_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to8_avx() 764 _sum04 = _mm256_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to8_avx() 765 _sum04 = _mm256_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to8_avx() 766 _sum04 = _mm256_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to8_avx() 767 _sum04 = _mm256_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to8_avx() 1292 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 759 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() 760 _sum04 = _mm256_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to8_avx() 761 _sum04 = _mm256_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to8_avx() 762 _sum04 = _mm256_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to8_avx() 763 _sum04 = _mm256_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to8_avx() 764 _sum04 = _mm256_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to8_avx() 765 _sum04 = _mm256_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to8_avx() 766 _sum04 = _mm256_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to8_avx() 767 _sum04 = _mm256_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to8_avx() 1292 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/misc/ncnn/ncnn-20211208/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 759 _sum04 = _mm256_comp_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() 760 _sum04 = _mm256_comp_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to8_avx() 761 _sum04 = _mm256_comp_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to8_avx() 762 _sum04 = _mm256_comp_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to8_avx() 763 _sum04 = _mm256_comp_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to8_avx() 764 _sum04 = _mm256_comp_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to8_avx() 765 _sum04 = _mm256_comp_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to8_avx() 766 _sum04 = _mm256_comp_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to8_avx() 767 _sum04 = _mm256_comp_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to8_avx() 1292 _sum04 = _mm256_comp_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() [all …]
|
H A D | convolution_3x3_pack1to4.h | 1017 __m128 _sum04 = _mm_loadu_ps(outptr0 + 16); in conv3x3s2_pack1to4_sse() local 1019 _sum04 = _mm_comp_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to4_sse() 1020 _sum04 = _mm_comp_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to4_sse() 1021 _sum04 = _mm_comp_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to4_sse() 1022 _sum04 = _mm_comp_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to4_sse() 1023 _sum04 = _mm_comp_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to4_sse() 1024 _sum04 = _mm_comp_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to4_sse() 1025 _sum04 = _mm_comp_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to4_sse() 1026 _sum04 = _mm_comp_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to4_sse() 1027 _sum04 = _mm_comp_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to4_sse() [all …]
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 759 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() 760 _sum04 = _mm256_fmadd_ps(_r010, _k01_0, _sum04); in conv3x3s2_pack1to8_avx() 761 _sum04 = _mm256_fmadd_ps(_r011, _k02_0, _sum04); in conv3x3s2_pack1to8_avx() 762 _sum04 = _mm256_fmadd_ps(_r19, _k10_0, _sum04); in conv3x3s2_pack1to8_avx() 763 _sum04 = _mm256_fmadd_ps(_r110, _k11_0, _sum04); in conv3x3s2_pack1to8_avx() 764 _sum04 = _mm256_fmadd_ps(_r111, _k12_0, _sum04); in conv3x3s2_pack1to8_avx() 765 _sum04 = _mm256_fmadd_ps(_r29, _k20_0, _sum04); in conv3x3s2_pack1to8_avx() 766 _sum04 = _mm256_fmadd_ps(_r210, _k21_0, _sum04); in conv3x3s2_pack1to8_avx() 767 _sum04 = _mm256_fmadd_ps(_r211, _k22_0, _sum04); in conv3x3s2_pack1to8_avx() 1292 _sum04 = _mm256_fmadd_ps(_r09, _k00_0, _sum04); in conv3x3s2_pack1to8_avx() [all …]
|