/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 244 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 249 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 250 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 313 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 318 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 319 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 128 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 133 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 134 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 197 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 202 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 203 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 250 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 255 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 256 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 319 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 324 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 325 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 350 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 355 float16x4_t _tmp3m = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 356 float16x4_t _tmp4m = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 419 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 424 float16x4_t _r0tm3 = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 425 float16x4_t _r0tm4 = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 385 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_neon() local 390 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 391 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 454 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_neon() local 459 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 460 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); 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 | 244 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 249 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 250 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 313 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 318 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 319 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 128 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 133 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 134 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 197 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 202 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 203 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 250 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 255 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 256 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 319 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 324 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 325 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 350 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 355 float16x4_t _tmp3m = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 356 float16x4_t _tmp4m = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 419 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 424 float16x4_t _r0tm3 = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 425 float16x4_t _r0tm4 = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 385 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_neon() local 390 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 391 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 454 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_neon() local 459 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 460 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 244 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 249 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 250 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 313 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 318 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 319 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 128 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 133 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 134 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 197 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 202 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 203 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 250 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 255 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 256 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 319 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 324 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 325 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 350 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 355 float16x4_t _tmp3m = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 356 float16x4_t _tmp4m = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 419 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 424 float16x4_t _r0tm3 = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 425 float16x4_t _r0tm4 = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 385 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_neon() local 390 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 391 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 454 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_neon() local 459 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 460 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | convolution_3x3_pack8to1_fp16s.h | 244 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 249 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 250 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 313 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 318 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 319 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 128 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 133 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 134 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 197 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 202 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 203 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 250 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 255 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 256 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 319 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 324 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 325 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 350 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 355 float16x4_t _tmp3m = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 356 float16x4_t _tmp4m = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 419 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 424 float16x4_t _r0tm3 = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 425 float16x4_t _r0tm4 = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 385 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_neon() local 390 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 391 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 454 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_neon() local 459 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 460 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); 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 | 244 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 249 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 250 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 313 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to1_fp16sa_neon() local 318 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon() 319 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to1_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1_bf16s.h | 128 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 133 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 134 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 197 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_bf16s_neon() local 202 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon() 203 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_bf16s_neon()
|
H A D | convolution_3x3_pack8to4_fp16s.h | 250 … float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 255 float16x8_t _tmp3m = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 256 float16x8_t _tmp4m = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 319 …float16x8_t _tmp34b = vfmaq_n_f16(vfmsq_n_f16(vmulq_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack8to4_fp16sa_neon() local 324 float16x8_t _r0tm3 = vaddq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon() 325 float16x8_t _r0tm4 = vsubq_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack8to4_fp16sa_neon()
|
H A D | convolution_3x3_pack4_fp16s.h | 350 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 355 float16x4_t _tmp3m = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 356 float16x4_t _tmp4m = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 419 … float16x4_t _tmp34b = vfma_n_f16(vfms_n_f16(vmul_n_f16(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.f); in conv3x3s1_winograd64_pack4_fp16sa_neon() local 424 float16x4_t _r0tm3 = vadd_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon() 425 float16x4_t _r0tm4 = vsub_f16(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4_fp16sa_neon()
|
H A D | convolution_3x3_pack4to1.h | 385 … float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_r01, 0.5f), _r03, 2.5f), _r05, 2.f); in conv3x3s1_winograd64_pack4to1_neon() local 390 float32x4_t _tmp3m = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 391 float32x4_t _tmp4m = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 454 …float32x4_t _tmp34b = vmlaq_n_f32(vmlsq_n_f32(vmulq_n_f32(_tmp01, 0.5f), _tmp03, 2.5f), _tmp05, 2.… in conv3x3s1_winograd64_pack4to1_neon() local 459 float32x4_t _r0tm3 = vaddq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon() 460 float32x4_t _r0tm4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_pack4to1_neon()
|