| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8_fp16s.h | 266 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
328 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8_fp16sa_neon()
337 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1406 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
1429 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack8_fp16sa_neon()
1436 r0_tm_4 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 278 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
340 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
349 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
1915 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd42_pack4_bf16s_neon() local
1938 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack4_bf16s_neon()
1945 r0_tm_4 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8_fp16s.h | 266 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
328 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8_fp16sa_neon()
337 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1406 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
1429 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack8_fp16sa_neon()
1436 r0_tm_4 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 278 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
340 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
349 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
1915 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd42_pack4_bf16s_neon() local
1938 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack4_bf16s_neon()
1945 r0_tm_4 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 284 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
346 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
355 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8_fp16s.h | 266 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
328 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8_fp16sa_neon()
337 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1406 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
1429 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack8_fp16sa_neon()
1436 r0_tm_4 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 278 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
340 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
349 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
1915 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd42_pack4_bf16s_neon() local
1938 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack4_bf16s_neon()
1945 r0_tm_4 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 284 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
346 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
355 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack8_fp16s.h | 266 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
328 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8_fp16sa_neon()
337 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1406 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
1429 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack8_fp16sa_neon()
1436 r0_tm_4 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 278 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
340 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
349 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
1915 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd42_pack4_bf16s_neon() local
1938 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack4_bf16s_neon()
1945 r0_tm_4 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 284 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
346 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
355 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolution_3x3_pack8_fp16s.h | 266 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8_fp16sa_neon() local
328 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8_fp16sa_neon()
337 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8_fp16sa_neon()
1406 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd42_pack8_fp16sa_neon() local
1429 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack8_fp16sa_neon()
1436 r0_tm_4 += tiles * 48; in conv3x3s1_winograd42_pack8_fp16sa_neon()
|
| H A D | convolution_3x3_pack8to1_fp16s.h | 278 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local
340 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
349 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
| H A D | convolution_3x3_pack4_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4_bf16s_neon()
1915 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd42_pack4_bf16s_neon() local
1938 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd42_pack4_bf16s_neon()
1945 r0_tm_4 += tiles * 24; in conv3x3s1_winograd42_pack4_bf16s_neon()
|
| H A D | convolution_3x3_pack4to1_bf16s.h | 162 float* r0_tm_4 = r0_tm_0 + tiles * 16; in conv3x3s1_winograd64_pack4to1_bf16s_neon() local
224 vst1q_f32(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
233 r0_tm_4 += tiles * 32; in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
| H A D | convolution_3x3_pack8to4_fp16s.h | 284 __fp16* r0_tm_4 = r0_tm_0 + tiles * 32; in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local
346 vst1q_f16(r0_tm_4, _r0tm4); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
355 r0_tm_4 += tiles * 64; in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_packn.h | 228 float* r0_tm_4 = r0_tm_0 + tiles * packn * 4; in conv3x3s1_winograd64_packn_rvv() local
269 vse32_v_f32m1(r0_tm_4, _r0tm4, vl); in conv3x3s1_winograd64_packn_rvv()
278 r0_tm_4 += tiles * packn * 8; in conv3x3s1_winograd64_packn_rvv()
925 float* r0_tm_4 = r0_tm_0 + tiles * packn * 4; in conv3x3s1_winograd42_packn_rvv() local
948 vse32_v_f32m1(r0_tm_4, _r0tm4, vl); in conv3x3s1_winograd42_packn_rvv()
955 r0_tm_4 += tiles * packn * 6; in conv3x3s1_winograd42_packn_rvv()
|