| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 110 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm_fp16() local
123 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_xc_O_0), xi, _sumO_0); in lstm_fp16()
147 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_hc_O_0), h_cont, _sumO_0); in lstm_fp16()
193 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), xi, _sumO_0); in lstm_fp16()
230 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), h_cont, _sumO_0); in lstm_fp16()
238 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm_fp16()
511 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm() local
524 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_O_0), xi, _sumO_0); in lstm()
548 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_O_0), h_cont, _sumO_0); in lstm()
565 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 110 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm_fp16() local
123 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_xc_O_0), xi, _sumO_0); in lstm_fp16()
147 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_hc_O_0), h_cont, _sumO_0); in lstm_fp16()
193 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), xi, _sumO_0); in lstm_fp16()
230 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), h_cont, _sumO_0); in lstm_fp16()
238 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm_fp16()
511 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm() local
524 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_O_0), xi, _sumO_0); in lstm()
548 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_O_0), h_cont, _sumO_0); in lstm()
565 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 110 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm_fp16() local
123 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_xc_O_0), xi, _sumO_0); in lstm_fp16()
147 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_hc_O_0), h_cont, _sumO_0); in lstm_fp16()
193 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), xi, _sumO_0); in lstm_fp16()
230 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), h_cont, _sumO_0); in lstm_fp16()
238 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm_fp16()
511 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm() local
524 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_O_0), xi, _sumO_0); in lstm()
548 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_O_0), h_cont, _sumO_0); in lstm()
565 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/x86/ |
| H A D | lstm_x86.cpp | 111 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm_fp16() local
124 _sumO_0 = _mm256_comp_fmadd_ps(loadfp16(weight_xc_O_0), xi, _sumO_0); in lstm_fp16()
148 _sumO_0 = _mm256_comp_fmadd_ps(loadfp16(weight_hc_O_0), h_cont, _sumO_0); in lstm_fp16()
194 _sumO_0 = _mm256_comp_fmadd_ps(loadfp16(fp16_weights[2]), xi, _sumO_0); in lstm_fp16()
231 _sumO_0 = _mm256_comp_fmadd_ps(loadfp16(fp16_weights[2]), h_cont, _sumO_0); in lstm_fp16()
239 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm_fp16()
512 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm() local
525 _sumO_0 = _mm256_comp_fmadd_ps(_mm256_loadu_ps(weight_xc_O_0), xi, _sumO_0); in lstm()
549 _sumO_0 = _mm256_comp_fmadd_ps(_mm256_loadu_ps(weight_hc_O_0), h_cont, _sumO_0); in lstm()
566 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm()
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/x86/ |
| H A D | lstm_x86.cpp | 112 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm_fp16() local
125 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_xc_O_0), xi, _sumO_0); in lstm_fp16()
149 _sumO_0 = _mm256_fmadd_ps(loadfp16(weight_hc_O_0), h_cont, _sumO_0); in lstm_fp16()
195 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), xi, _sumO_0); in lstm_fp16()
232 _sumO_0 = _mm256_fmadd_ps(loadfp16(fp16_weights[2]), h_cont, _sumO_0); in lstm_fp16()
240 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm_fp16()
513 __m256 _sumO_0 = _mm256_setzero_ps(); in lstm() local
526 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_xc_O_0), xi, _sumO_0); in lstm()
550 _sumO_0 = _mm256_fmadd_ps(_mm256_loadu_ps(weight_hc_O_0), h_cont, _sumO_0); in lstm()
567 …_mm256_storeu_ps(sums, HorizontalSums(_sumI_0, _sumF_0, _sumO_0, _sumG_0, _sumI_1, _sumF_1, _sumO_… in lstm()
|