| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1051 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1052 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1053 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1054 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1055 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1057 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1058 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1072 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1080 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1966 const float* output0_tm_0 = (const float*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1967 const float* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1968 const float* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1969 const float* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1970 const float* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1972 const float* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1973 const float* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1987 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1995 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1075 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1076 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1077 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1078 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1080 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 48; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1081 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 56; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1088 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1137 output0_tm_0 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2179 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd42_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 1076 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
1077 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1078 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1079 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1080 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1081 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1082 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1083 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 28; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1090 float16x4_t _out0tm0 = vld1_f16(output0_tm_0); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1139 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1685 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4_bf16s_neon()
1686 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack4_bf16s_neon()
1687 const float* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack4_bf16s_neon()
1688 const float* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon()
1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon()
1698 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd64_pack4_bf16s_neon()
1747 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3391 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd42_pack4_bf16s_neon()
3392 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd42_pack4_bf16s_neon()
3402 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd42_pack4_bf16s_neon()
[all …]
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1051 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1052 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1053 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1054 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1055 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1057 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1058 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1072 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1080 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1966 const float* output0_tm_0 = (const float*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1967 const float* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1968 const float* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1969 const float* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1970 const float* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1972 const float* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1973 const float* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1987 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1995 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1075 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1076 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1077 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1078 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1080 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 48; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1081 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 56; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1088 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1137 output0_tm_0 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2179 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd42_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 1076 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
1077 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1078 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1079 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1080 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1081 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1082 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1083 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 28; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1090 float16x4_t _out0tm0 = vld1_f16(output0_tm_0); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1139 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1685 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4_bf16s_neon()
1686 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack4_bf16s_neon()
1687 const float* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack4_bf16s_neon()
1688 const float* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon()
1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon()
1698 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd64_pack4_bf16s_neon()
1747 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3391 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd42_pack4_bf16s_neon()
3392 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd42_pack4_bf16s_neon()
3402 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd42_pack4_bf16s_neon()
[all …]
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1051 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1052 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1053 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1054 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1055 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1057 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1058 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1072 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1080 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1966 const float* output0_tm_0 = (const float*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1967 const float* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1968 const float* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1969 const float* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1970 const float* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1972 const float* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1973 const float* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1987 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1995 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1075 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1076 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1077 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1078 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1080 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 48; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1081 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 56; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1088 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1137 output0_tm_0 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2179 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd42_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 1076 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
1077 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1078 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1079 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1080 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1081 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1082 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1083 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 28; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1090 float16x4_t _out0tm0 = vld1_f16(output0_tm_0); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1139 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1685 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4_bf16s_neon()
1686 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack4_bf16s_neon()
1687 const float* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack4_bf16s_neon()
1688 const float* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon()
1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon()
1698 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd64_pack4_bf16s_neon()
1747 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3391 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd42_pack4_bf16s_neon()
3392 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd42_pack4_bf16s_neon()
3402 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd42_pack4_bf16s_neon()
[all …]
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1051 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1052 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1053 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1054 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1055 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1057 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1058 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1072 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1080 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1966 const float* output0_tm_0 = (const float*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1967 const float* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1968 const float* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1969 const float* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1970 const float* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1972 const float* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1973 const float* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1987 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1995 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1075 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1076 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1077 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1078 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1080 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 48; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1081 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 56; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1088 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1137 output0_tm_0 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2179 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd42_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 1076 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
1077 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1078 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1079 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1080 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1081 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1082 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1083 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 28; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1090 float16x4_t _out0tm0 = vld1_f16(output0_tm_0); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1139 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1685 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4_bf16s_neon()
1686 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack4_bf16s_neon()
1687 const float* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack4_bf16s_neon()
1688 const float* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon()
1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon()
1698 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd64_pack4_bf16s_neon()
1747 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3391 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd42_pack4_bf16s_neon()
3392 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd42_pack4_bf16s_neon()
3402 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd42_pack4_bf16s_neon()
[all …]
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8to1_fp16s.h | 1051 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
1052 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1053 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1054 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1055 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1056 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1057 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1058 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1072 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
1080 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack8_fp16s.h | 1075 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1076 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1077 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1078 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1079 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 40; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1080 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 48; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1081 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 56; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1088 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd64_pack8_fp16sa_neon()
1137 output0_tm_0 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
2179 float16x8_t _out0tm0 = vld1q_f16(output0_tm_0); in conv3x3s1_winograd42_pack8_fp16sa_neon()
[all …]
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 1966 const float* output0_tm_0 = (const float*)out0_tm + (i * w_tm / 8 + j) * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
1967 const float* output0_tm_1 = output0_tm_0 + tiles * 1; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1968 const float* output0_tm_2 = output0_tm_0 + tiles * 2; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1969 const float* output0_tm_3 = output0_tm_0 + tiles * 3; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1970 const float* output0_tm_4 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1971 const float* output0_tm_5 = output0_tm_0 + tiles * 5; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1972 const float* output0_tm_6 = output0_tm_0 + tiles * 6; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1973 const float* output0_tm_7 = output0_tm_0 + tiles * 7; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1987 tmp[0][m] = output0_tm_0[0] + tmp024a + tmp024b + tmp024c * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
1995 output0_tm_0 += tiles * 8; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 1076 const __fp16* output0_tm_0 = (const __fp16*)out0_tm + (i * w_tm / 8 + j) * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
1077 const __fp16* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1078 const __fp16* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1079 const __fp16* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1080 const __fp16* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1081 const __fp16* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1082 const __fp16* output0_tm_6 = output0_tm_0 + tiles * 24; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1083 const __fp16* output0_tm_7 = output0_tm_0 + tiles * 28; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1090 float16x4_t _out0tm0 = vld1_f16(output0_tm_0); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
1139 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 1685 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd64_pack4_bf16s_neon()
1686 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd64_pack4_bf16s_neon()
1687 const float* output0_tm_3 = output0_tm_0 + tiles * 12; in conv3x3s1_winograd64_pack4_bf16s_neon()
1688 const float* output0_tm_4 = output0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon()
1689 const float* output0_tm_5 = output0_tm_0 + tiles * 20; in conv3x3s1_winograd64_pack4_bf16s_neon()
1698 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd64_pack4_bf16s_neon()
1747 output0_tm_0 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
3391 const float* output0_tm_1 = output0_tm_0 + tiles * 4; in conv3x3s1_winograd42_pack4_bf16s_neon()
3392 const float* output0_tm_2 = output0_tm_0 + tiles * 8; in conv3x3s1_winograd42_pack4_bf16s_neon()
3402 float32x4_t _out0tm0 = vld1q_f32(output0_tm_0); in conv3x3s1_winograd42_pack4_bf16s_neon()
[all …]
|