| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_1x1_int8.h | 580 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_neon()
581 _sum = vpadalq_s16(_sum, _tp0); in conv1x1s1_sgemm_int8_neon()
592 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_neon()
594 _sum = vaddw_s16(_sum, vget_low_s16(_tp0)); in conv1x1s1_sgemm_int8_neon()
681 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_neon() local
682 _sum = vpadalq_s16(_sum, _tp0); in conv1x1s1_sgemm_int8_neon()
1239 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_requant_neon() local
1240 _sum = vpadalq_s16(_sum, _tp0); in conv1x1s1_sgemm_int8_requant_neon()
1251 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_requant_neon() local
1365 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv1x1s1_sgemm_int8_requant_neon() local
[all …]
|
| H A D | convolution_3x3_int8.h | 2179 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_int8_neon() local
2199 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_int8_neon()
2202 _t1p2 = vmlaq_lane_s32(_t1p2, _t3p4, _tp0, 1); in conv3x3s1_winograd43_int8_neon()
2204 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_int8_neon()
2210 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_int8_neon()
2215 _t1s2n = vmul_s32(_t1s2n, _tp0); in conv3x3s1_winograd43_int8_neon()
3327 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_dequant_int8_neon() local
3347 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3352 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3358 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
[all …]
|
| H A D | convolution_sgemm_int8.h | 569 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv_im2col_sgemm_int8_neon()
570 _sum = vpadalq_s16(_sum, _tp0); in conv_im2col_sgemm_int8_neon()
581 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv_im2col_sgemm_int8_neon()
583 _sum = vaddw_s16(_sum, vget_low_s16(_tp0)); in conv_im2col_sgemm_int8_neon()
669 int16x8_t _tp0 = vmull_s8(_k, _r0); in conv_im2col_sgemm_int8_neon() local
670 _sum = vpadalq_s16(_sum, _tp0); in conv_im2col_sgemm_int8_neon()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_int8.h | 2179 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_int8_neon() local
2199 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_int8_neon()
2202 _t1p2 = vmlaq_lane_s32(_t1p2, _t3p4, _tp0, 1); in conv3x3s1_winograd43_int8_neon()
2204 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_int8_neon()
2210 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_int8_neon()
2215 _t1s2n = vmul_s32(_t1s2n, _tp0); in conv3x3s1_winograd43_int8_neon()
3327 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_dequant_int8_neon() local
3347 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3352 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3358 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
[all …]
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_int8.h | 2179 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_int8_neon() local
2199 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_int8_neon()
2202 _t1p2 = vmlaq_lane_s32(_t1p2, _t3p4, _tp0, 1); in conv3x3s1_winograd43_int8_neon()
2204 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_int8_neon()
2210 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_int8_neon()
2215 _t1s2n = vmul_s32(_t1s2n, _tp0); in conv3x3s1_winograd43_int8_neon()
3327 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_dequant_int8_neon() local
3347 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3352 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3358 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
[all …]
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_int8.h | 2179 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_int8_neon() local
2199 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_int8_neon()
2202 _t1p2 = vmlaq_lane_s32(_t1p2, _t3p4, _tp0, 1); in conv3x3s1_winograd43_int8_neon()
2204 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_int8_neon()
2210 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_int8_neon()
2215 _t1s2n = vmul_s32(_t1s2n, _tp0); in conv3x3s1_winograd43_int8_neon()
3327 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_dequant_int8_neon() local
3347 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3352 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3358 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
[all …]
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_int8.h | 2179 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_int8_neon() local
2199 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_int8_neon()
2202 _t1p2 = vmlaq_lane_s32(_t1p2, _t3p4, _tp0, 1); in conv3x3s1_winograd43_int8_neon()
2204 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_int8_neon()
2210 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_int8_neon()
2215 _t1s2n = vmul_s32(_t1s2n, _tp0); in conv3x3s1_winograd43_int8_neon()
3327 int32x2_t _tp0 = {1, 4}; in conv3x3s1_winograd43_dequant_int8_neon() local
3347 _w0n = vmul_s32(_w0n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3352 _t1p2n = vmul_s32(_t1p2n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
3358 _w3n = vmul_s32(_w3n, _tp0); in conv3x3s1_winograd43_dequant_int8_neon()
[all …]
|