/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 230 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 235 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 236 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 301 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 306 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 307 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 114 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 119 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 120 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 185 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 190 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 191 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 236 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 241 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 242 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 307 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 312 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 313 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 336 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 341 float16x4_t _tmp1m = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 342 float16x4_t _tmp2m = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 407 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 412 float16x4_t _r0tm1 = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 413 float16x4_t _r0tm2 = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 371 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 376 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 377 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 442 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 447 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 448 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 230 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 235 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 236 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 301 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 306 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 307 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 114 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 119 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 120 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 185 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 190 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 191 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 236 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 241 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 242 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 307 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 312 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 313 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 336 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 341 float16x4_t _tmp1m = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 342 float16x4_t _tmp2m = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 407 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 412 float16x4_t _r0tm1 = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 413 float16x4_t _r0tm2 = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 371 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 376 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 377 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 442 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 447 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 448 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 230 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 235 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 236 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 301 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 306 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 307 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 114 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 119 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 120 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 185 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 190 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 191 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 236 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 241 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 242 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 307 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 312 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 313 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 336 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 341 float16x4_t _tmp1m = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 342 float16x4_t _tmp2m = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 407 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 412 float16x4_t _r0tm1 = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 413 float16x4_t _r0tm2 = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 371 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 376 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 377 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 442 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 447 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 448 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 230 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 235 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 236 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 301 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 306 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 307 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 114 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 119 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 120 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 185 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 190 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 191 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 236 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 241 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 242 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 307 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 312 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 313 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 336 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 341 float16x4_t _tmp1m = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 342 float16x4_t _tmp2m = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 407 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 412 float16x4_t _r0tm1 = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 413 float16x4_t _r0tm2 = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 371 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 376 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 377 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 442 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 447 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 448 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 230 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 235 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 236 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 301 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 306 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 307 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 114 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 119 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 120 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 185 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 190 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 191 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 236 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 241 float16x8_t _tmp1m = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 242 float16x8_t _tmp2m = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 307 float16x8_t _tmp12b = vfmsq_n_f16(vaddq_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 312 float16x8_t _r0tm1 = vaddq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 313 float16x8_t _r0tm2 = vsubq_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 336 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 341 float16x4_t _tmp1m = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 342 float16x4_t _tmp2m = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 407 float16x4_t _tmp12b = vfms_n_f16(vadd_f16(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 412 float16x4_t _r0tm1 = vadd_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 413 float16x4_t _r0tm2 = vsub_f16(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 371 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_r01, _r05), _r03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 376 float32x4_t _tmp1m = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 377 float32x4_t _tmp2m = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 442 float32x4_t _tmp12b = vmlsq_n_f32(vaddq_f32(_tmp01, _tmp05), _tmp03, 4.25f); in conv3x3s1_winograd64_pack4to1_neon() local 447 float32x4_t _r0tm1 = vaddq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon() 448 float32x4_t _r0tm2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_pack4to1_neon()
|