| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_1x1_int8.h | 433 int sum1_1 = 0; in conv1x1s1_sgemm_int8_neon() local
461 sum1_1 += tmpptr[2] * kptr[2]; in conv1x1s1_sgemm_int8_neon()
462 sum1_1 += tmpptr[3] * kptr[3]; in conv1x1s1_sgemm_int8_neon()
522 outptr1[1] = sum1_1; in conv1x1s1_sgemm_int8_neon()
1092 int sum1_1 = 0; in conv1x1s1_sgemm_int8_requant_neon() local
1765 int sum1_1 = 0; in conv1x1s1_sgemm_int8_neon() local
1843 outptr1[1] = sum1_1; in conv1x1s1_sgemm_int8_neon()
1975 int sum1_1 = 0; in conv1x1s1_sgemm_int8_neon() local
2021 outptr1[1] = sum1_1; in conv1x1s1_sgemm_int8_neon()
2800 int sum1_1 = 0; in conv1x1s1_sgemm_int8_requant_neon() local
[all …]
|
| H A D | convolution_1x1_bf16s.h | 1159 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1197 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1237 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
1508 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1530 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1554 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
|
| H A D | convolution_1x1.h | 1013 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_neon() local
1051 sum1_1 += tmpptr[1] * kptr[1]; in conv1x1s1_sgemm_neon()
1091 outptr1[1] = sum1_1; in conv1x1s1_sgemm_neon()
1328 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_neon() local
1350 sum1_1 += tmpptr[1] * kptr[1]; in conv1x1s1_sgemm_neon()
1374 outptr1[1] = sum1_1; in conv1x1s1_sgemm_neon()
|
| H A D | convolution_3x3.h | 6309 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6347 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6387 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
6613 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6635 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6659 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_1x1_bf16s.h | 1159 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1197 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1237 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
1508 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1530 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1554 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
|
| H A D | convolution_3x3.h | 6309 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6347 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6387 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
6613 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6635 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6659 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_1x1_bf16s.h | 1159 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1197 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1237 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
1508 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1530 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1554 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
|
| H A D | convolution_3x3.h | 6309 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6347 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6387 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
6613 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6635 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6659 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_1x1_bf16s.h | 1159 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1197 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1237 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
1508 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1530 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1554 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
|
| H A D | convolution_3x3.h | 6309 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6347 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6387 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
6613 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6635 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6659 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_1x1_bf16s.h | 1159 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1197 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1237 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
1508 float sum1_1 = biasptr[1]; in conv1x1s1_sgemm_bf16s_neon() local
1530 sum1_1 += bfloat16_to_float32(tmpptr[1]) * bfloat16_to_float32(kptr[1]); in conv1x1s1_sgemm_bf16s_neon()
1554 outptr1[1] = float32_to_bfloat16(sum1_1); in conv1x1s1_sgemm_bf16s_neon()
|
| H A D | convolution_3x3.h | 6309 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6347 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6387 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
6613 float sum1_1 = 0.f; in conv3x3s1_winograd64_neon5() local
6635 sum1_1 += bb2p0[1] * ktm0[1]; in conv3x3s1_winograd64_neon5()
6659 output1_tm[1] = sum1_1; in conv3x3s1_winograd64_neon5()
|
| /dports/math/ntl/ntl-11.5.1/src/ |
| H A D | mat_lzz_p.cpp | 2277 unsigned long sum1_1 = ll_get_hi(sum_1); in muladd1_by_32_full() local
2287 x[j+1] = sp_ll_red_31_normalized(0, sum1_1, sum0_1, p, ll_red_struct); in muladd1_by_32_full()
2293 x[j+1] = sp_ll_red_31(0, sum1_1, sum0_1, p, ll_red_struct); in muladd1_by_32_full()
2356 unsigned long sum1_1 = ll_get_hi(sum_1); in muladd1_by_32_full_width() local
2366 x[j+1] = sp_ll_red_31_normalized(0, sum1_1, sum0_1, p, ll_red_struct); in muladd1_by_32_full_width()
2372 x[j+1] = sp_ll_red_31(0, sum1_1, sum0_1, p, ll_red_struct); in muladd1_by_32_full_width()
|