/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 790 __m256 _sum15 = _mm256_loadu_ps(outptr1 + 40); in conv3x3s2_pack1to8_avx() local 801 _sum15 = _mm256_fmadd_ps(_r011, _k00_1, _sum15); in conv3x3s2_pack1to8_avx() 802 _sum15 = _mm256_fmadd_ps(_r012, _k01_1, _sum15); in conv3x3s2_pack1to8_avx() 803 _sum15 = _mm256_fmadd_ps(_r013, _k02_1, _sum15); in conv3x3s2_pack1to8_avx() 804 _sum15 = _mm256_fmadd_ps(_r111, _k10_1, _sum15); in conv3x3s2_pack1to8_avx() 805 _sum15 = _mm256_fmadd_ps(_r112, _k11_1, _sum15); in conv3x3s2_pack1to8_avx() 806 _sum15 = _mm256_fmadd_ps(_r113, _k12_1, _sum15); in conv3x3s2_pack1to8_avx() 807 _sum15 = _mm256_fmadd_ps(_r211, _k20_1, _sum15); in conv3x3s2_pack1to8_avx() 808 _sum15 = _mm256_fmadd_ps(_r212, _k21_1, _sum15); in conv3x3s2_pack1to8_avx() 809 _sum15 = _mm256_fmadd_ps(_r213, _k22_1, _sum15); 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 | 790 __m256 _sum15 = _mm256_loadu_ps(outptr1 + 40); in conv3x3s2_pack1to8_avx() local 801 _sum15 = _mm256_fmadd_ps(_r011, _k00_1, _sum15); in conv3x3s2_pack1to8_avx() 802 _sum15 = _mm256_fmadd_ps(_r012, _k01_1, _sum15); in conv3x3s2_pack1to8_avx() 803 _sum15 = _mm256_fmadd_ps(_r013, _k02_1, _sum15); in conv3x3s2_pack1to8_avx() 804 _sum15 = _mm256_fmadd_ps(_r111, _k10_1, _sum15); in conv3x3s2_pack1to8_avx() 805 _sum15 = _mm256_fmadd_ps(_r112, _k11_1, _sum15); in conv3x3s2_pack1to8_avx() 806 _sum15 = _mm256_fmadd_ps(_r113, _k12_1, _sum15); in conv3x3s2_pack1to8_avx() 807 _sum15 = _mm256_fmadd_ps(_r211, _k20_1, _sum15); in conv3x3s2_pack1to8_avx() 808 _sum15 = _mm256_fmadd_ps(_r212, _k21_1, _sum15); in conv3x3s2_pack1to8_avx() 809 _sum15 = _mm256_fmadd_ps(_r213, _k22_1, _sum15); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 790 __m256 _sum15 = _mm256_loadu_ps(outptr1 + 40); in conv3x3s2_pack1to8_avx() local 801 _sum15 = _mm256_fmadd_ps(_r011, _k00_1, _sum15); in conv3x3s2_pack1to8_avx() 802 _sum15 = _mm256_fmadd_ps(_r012, _k01_1, _sum15); in conv3x3s2_pack1to8_avx() 803 _sum15 = _mm256_fmadd_ps(_r013, _k02_1, _sum15); in conv3x3s2_pack1to8_avx() 804 _sum15 = _mm256_fmadd_ps(_r111, _k10_1, _sum15); in conv3x3s2_pack1to8_avx() 805 _sum15 = _mm256_fmadd_ps(_r112, _k11_1, _sum15); in conv3x3s2_pack1to8_avx() 806 _sum15 = _mm256_fmadd_ps(_r113, _k12_1, _sum15); in conv3x3s2_pack1to8_avx() 807 _sum15 = _mm256_fmadd_ps(_r211, _k20_1, _sum15); in conv3x3s2_pack1to8_avx() 808 _sum15 = _mm256_fmadd_ps(_r212, _k21_1, _sum15); in conv3x3s2_pack1to8_avx() 809 _sum15 = _mm256_fmadd_ps(_r213, _k22_1, _sum15); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/misc/ncnn/ncnn-20211208/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 790 __m256 _sum15 = _mm256_loadu_ps(outptr1 + 40); in conv3x3s2_pack1to8_avx() local 801 _sum15 = _mm256_comp_fmadd_ps(_r011, _k00_1, _sum15); in conv3x3s2_pack1to8_avx() 802 _sum15 = _mm256_comp_fmadd_ps(_r012, _k01_1, _sum15); in conv3x3s2_pack1to8_avx() 803 _sum15 = _mm256_comp_fmadd_ps(_r013, _k02_1, _sum15); in conv3x3s2_pack1to8_avx() 804 _sum15 = _mm256_comp_fmadd_ps(_r111, _k10_1, _sum15); in conv3x3s2_pack1to8_avx() 805 _sum15 = _mm256_comp_fmadd_ps(_r112, _k11_1, _sum15); in conv3x3s2_pack1to8_avx() 806 _sum15 = _mm256_comp_fmadd_ps(_r113, _k12_1, _sum15); in conv3x3s2_pack1to8_avx() 807 _sum15 = _mm256_comp_fmadd_ps(_r211, _k20_1, _sum15); in conv3x3s2_pack1to8_avx() 808 _sum15 = _mm256_comp_fmadd_ps(_r212, _k21_1, _sum15); in conv3x3s2_pack1to8_avx() 809 _sum15 = _mm256_comp_fmadd_ps(_r213, _k22_1, _sum15); in conv3x3s2_pack1to8_avx() [all …]
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
H A D | convolution_3x3_pack1to8.h | 790 __m256 _sum15 = _mm256_loadu_ps(outptr1 + 40); in conv3x3s2_pack1to8_avx() local 801 _sum15 = _mm256_fmadd_ps(_r011, _k00_1, _sum15); in conv3x3s2_pack1to8_avx() 802 _sum15 = _mm256_fmadd_ps(_r012, _k01_1, _sum15); in conv3x3s2_pack1to8_avx() 803 _sum15 = _mm256_fmadd_ps(_r013, _k02_1, _sum15); in conv3x3s2_pack1to8_avx() 804 _sum15 = _mm256_fmadd_ps(_r111, _k10_1, _sum15); in conv3x3s2_pack1to8_avx() 805 _sum15 = _mm256_fmadd_ps(_r112, _k11_1, _sum15); in conv3x3s2_pack1to8_avx() 806 _sum15 = _mm256_fmadd_ps(_r113, _k12_1, _sum15); in conv3x3s2_pack1to8_avx() 807 _sum15 = _mm256_fmadd_ps(_r211, _k20_1, _sum15); in conv3x3s2_pack1to8_avx() 808 _sum15 = _mm256_fmadd_ps(_r212, _k21_1, _sum15); in conv3x3s2_pack1to8_avx() 809 _sum15 = _mm256_fmadd_ps(_r213, _k22_1, _sum15); in conv3x3s2_pack1to8_avx() [all …]
|