| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1069 __fp16 tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1070 __fp16 tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1085 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1984 float tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1985 float tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
2000 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to1_int8.h | 1092 const int* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd42_pack8to1_int8_neon() local
1113 tmp[3][m] = output0_tm_5[0] * 4 + tmp13a + tmp13b * 8; in conv3x3s1_winograd42_pack8to1_int8_neon()
1120 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
1133 tmp[3][m] = (output0_tm_5[0] * 4 + tmp13a + tmp13b * 8) * 4; in conv3x3s1_winograd42_pack8to1_int8_neon()
1140 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
1093 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1142 output0_tm_5 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2172 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
2184 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd42_pack8_fp16sa_neon()
2207 output0_tm_5 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon() local
1703 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd64_pack4_bf16s_neon()
1752 output0_tm_5 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3395 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd42_pack4_bf16s_neon() local
3407 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd42_pack4_bf16s_neon()
3430 output0_tm_5 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1069 __fp16 tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1070 __fp16 tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1085 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1984 float tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1985 float tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
2000 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to1_int8.h | 1092 const int* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd42_pack8to1_int8_neon() local
1113 tmp[3][m] = output0_tm_5[0] * 4 + tmp13a + tmp13b * 8; in conv3x3s1_winograd42_pack8to1_int8_neon()
1120 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
1133 tmp[3][m] = (output0_tm_5[0] * 4 + tmp13a + tmp13b * 8) * 4; in conv3x3s1_winograd42_pack8to1_int8_neon()
1140 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
1093 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1142 output0_tm_5 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2172 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
2184 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd42_pack8_fp16sa_neon()
2207 output0_tm_5 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon() local
1703 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd64_pack4_bf16s_neon()
1752 output0_tm_5 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3395 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd42_pack4_bf16s_neon() local
3407 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd42_pack4_bf16s_neon()
3430 output0_tm_5 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1069 __fp16 tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1070 __fp16 tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1085 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1984 float tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1985 float tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
2000 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to1_int8.h | 1092 const int* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd42_pack8to1_int8_neon() local
1113 tmp[3][m] = output0_tm_5[0] * 4 + tmp13a + tmp13b * 8; in conv3x3s1_winograd42_pack8to1_int8_neon()
1120 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
1133 tmp[3][m] = (output0_tm_5[0] * 4 + tmp13a + tmp13b * 8) * 4; in conv3x3s1_winograd42_pack8to1_int8_neon()
1140 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
1093 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1142 output0_tm_5 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2172 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
2184 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd42_pack8_fp16sa_neon()
2207 output0_tm_5 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon() local
1703 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd64_pack4_bf16s_neon()
1752 output0_tm_5 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3395 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd42_pack4_bf16s_neon() local
3407 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd42_pack4_bf16s_neon()
3430 output0_tm_5 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1069 __fp16 tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1070 __fp16 tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1085 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1984 float tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1985 float tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
2000 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to1_int8.h | 1092 const int* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd42_pack8to1_int8_neon() local
1113 tmp[3][m] = output0_tm_5[0] * 4 + tmp13a + tmp13b * 8; in conv3x3s1_winograd42_pack8to1_int8_neon()
1120 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
1133 tmp[3][m] = (output0_tm_5[0] * 4 + tmp13a + tmp13b * 8) * 4; in conv3x3s1_winograd42_pack8to1_int8_neon()
1140 output0_tm_5 += tiles * 6; in conv3x3s1_winograd42_pack8to1_int8_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
1093 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1142 output0_tm_5 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2172 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
2184 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd42_pack8_fp16sa_neon()
2207 output0_tm_5 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon() local
1703 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd64_pack4_bf16s_neon()
1752 output0_tm_5 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3395 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd42_pack4_bf16s_neon() local
3407 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd42_pack4_bf16s_neon()
3430 output0_tm_5 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1069 __fp16 tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1070 __fp16 tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1085 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1984 float tmp024c = output0_tm_5[0] + output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1985 float tmp135c = output0_tm_5[0] - output0_tm_6[0]; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
2000 output0_tm_5 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
1093 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1142 output0_tm_5 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2172 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
2184 float16x8_t _out0tm5 = vld1q_f16(output0_tm_5); in conv3x3s1_winograd42_pack8_fp16sa_neon()
2207 output0_tm_5 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon() local
1703 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd64_pack4_bf16s_neon()
1752 output0_tm_5 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3395 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd42_pack4_bf16s_neon() local
3407 float32x4_t _out0tm5 = vld1q_f32(output0_tm_5); in conv3x3s1_winograd42_pack4_bf16s_neon()
3430 output0_tm_5 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_packn.h | 649 const float* output0_tm_5 = output0_tm_0 + tiles * packn * 5; in conv3x3s1_winograd64_packn_rvv() local
663 vfloat32m1_t _out0tm5 = vle32_v_f32m1(output0_tm_5, vl); in conv3x3s1_winograd64_packn_rvv()
695 output0_tm_5 += tiles * packn * 8; in conv3x3s1_winograd64_packn_rvv()
1320 const float* output0_tm_5 = output0_tm_0 + tiles * packn * 5; in conv3x3s1_winograd42_packn_rvv() local
1332 vfloat32m1_t _out0tm5 = vle32_v_f32m1(output0_tm_5, vl); in conv3x3s1_winograd42_packn_rvv()
1355 output0_tm_5 += tiles * packn * 6; in conv3x3s1_winograd42_packn_rvv()
|