| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 113 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm_fp16() local
126 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_xc_F_1), xi, _sumF_1); in lstm_fp16()
150 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_hc_F_1), h_cont, _sumF_1); in lstm_fp16()
196 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), xi, _sumF_1); in lstm_fp16()
233 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), h_cont, _sumF_1); in lstm_fp16()
238 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm_fp16()
514 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm() local
527 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_F_1), xi, _sumF_1); in lstm()
551 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_F_1), h_cont, _sumF_1); in lstm()
565 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 113 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm_fp16() local
126 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_xc_F_1), xi, _sumF_1); in lstm_fp16()
150 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_hc_F_1), h_cont, _sumF_1); in lstm_fp16()
196 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), xi, _sumF_1); in lstm_fp16()
233 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), h_cont, _sumF_1); in lstm_fp16()
238 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm_fp16()
514 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm() local
527 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_F_1), xi, _sumF_1); in lstm()
551 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_F_1), h_cont, _sumF_1); in lstm()
565 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 113 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm_fp16() local
126 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_xc_F_1), xi, _sumF_1); in lstm_fp16()
150 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_hc_F_1), h_cont, _sumF_1); in lstm_fp16()
196 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), xi, _sumF_1); in lstm_fp16()
233 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), h_cont, _sumF_1); in lstm_fp16()
238 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm_fp16()
514 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm() local
527 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_F_1), xi, _sumF_1); in lstm()
551 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_F_1), h_cont, _sumF_1); in lstm()
565 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/x86/ |
| H A D | lstm_x86.cpp | 114 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm_fp16() local
127 _sumF_1 = _mm256_comp_fmadd_ps(loadfp16(weight_xc_F_1), xi, _sumF_1); in lstm_fp16()
151 _sumF_1 = _mm256_comp_fmadd_ps(loadfp16(weight_hc_F_1), h_cont, _sumF_1); in lstm_fp16()
197 _sumF_1 = _mm256_comp_fmadd_ps(loadfp16(fp16_weights[5]), xi, _sumF_1); in lstm_fp16()
234 _sumF_1 = _mm256_comp_fmadd_ps(loadfp16(fp16_weights[5]), h_cont, _sumF_1); in lstm_fp16()
239 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm_fp16()
515 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm() local
528 _sumF_1 = _mm256_comp_fmadd_ps(_mm256_loadu_ps(weight_xc_F_1), xi, _sumF_1); in lstm()
552 _sumF_1 = _mm256_comp_fmadd_ps(_mm256_loadu_ps(weight_hc_F_1), h_cont, _sumF_1); in lstm()
566 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm()
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 115 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm_fp16() local
128 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_xc_F_1), xi, _sumF_1); in lstm_fp16()
152 _sumF_1 = _mm256_fmadd_ps(loadfp16(weight_hc_F_1), h_cont, _sumF_1); in lstm_fp16()
198 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), xi, _sumF_1); in lstm_fp16()
235 _sumF_1 = _mm256_fmadd_ps(loadfp16(fp16_weights[5]), h_cont, _sumF_1); in lstm_fp16()
240 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm_fp16()
516 __m256 _sumF_1 = _mm256_setzero_ps(); in lstm() local
529 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_F_1), xi, _sumF_1); in lstm()
553 _sumF_1 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_F_1), h_cont, _sumF_1); in lstm()
567 …reu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_1, _sumG_1… in lstm()
|