/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3_pack4.h | 45 float32x4_t _k00 = vld1q_f32(k0); in convdw3x3s1_pack4_neon() local 206 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 306 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 386 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 500 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 601 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 679 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 752 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 941 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() 1076 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 45 float32x4_t _k00 = vcvt_f32_bf16(vld1_u16(k0)); in convdw3x3s1_pack4_bf16s_neon() local 278 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 424 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 529 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 681 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 814 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 910 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1001 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1243 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() 1351 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s1_pack8_fp16sa_neon() local 202 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 297 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 376 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 481 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 548 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 605 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 655 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s2_pack8_fp16sa_neon() local 759 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() 834 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() [all …]
|
H A D | convolution_3x3_pack1to4_fp16s.h | 45 float16x4_t _k00 = vld1_f16(k0); in conv3x3s1_pack1to4_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 427 float16x4_t _k00 = vld1_f16(k0); in conv3x3s2_pack1to4_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon()
|
H A D | convolution_3x3_pack1to8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s1_pack1to8_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 427 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s2_pack1to8_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon()
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3_pack4.h | 45 float32x4_t _k00 = vld1q_f32(k0); in convdw3x3s1_pack4_neon() local 206 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 306 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 386 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 500 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 601 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 679 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 752 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 941 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() 1076 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 45 float32x4_t _k00 = vcvt_f32_bf16(vld1_u16(k0)); in convdw3x3s1_pack4_bf16s_neon() local 278 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 424 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 529 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 681 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 814 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 910 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1001 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1243 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() 1351 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s1_pack8_fp16sa_neon() local 202 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 297 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 376 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 481 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 548 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 605 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 655 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s2_pack8_fp16sa_neon() local 759 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() 834 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() [all …]
|
H A D | convolution_3x3_pack1to8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s1_pack1to8_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 427 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s2_pack1to8_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon()
|
H A D | convolution_3x3_pack1to4_fp16s.h | 45 float16x4_t _k00 = vld1_f16(k0); in conv3x3s1_pack1to4_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 427 float16x4_t _k00 = vld1_f16(k0); in conv3x3s2_pack1to4_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3_pack4.h | 45 float32x4_t _k00 = vld1q_f32(k0); in convdw3x3s1_pack4_neon() local 206 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 306 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 386 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 500 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 601 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 679 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 752 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 941 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() 1076 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 45 float32x4_t _k00 = vcvt_f32_bf16(vld1_u16(k0)); in convdw3x3s1_pack4_bf16s_neon() local 278 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 424 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 529 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 681 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 814 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 910 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1001 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1243 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() 1351 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s1_pack8_fp16sa_neon() local 202 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 297 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 376 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 481 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 548 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 605 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 655 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s2_pack8_fp16sa_neon() local 759 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() 834 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() [all …]
|
H A D | convolution_3x3_pack1to4_fp16s.h | 45 float16x4_t _k00 = vld1_f16(k0); in conv3x3s1_pack1to4_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 427 float16x4_t _k00 = vld1_f16(k0); in conv3x3s2_pack1to4_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon()
|
H A D | convolution_3x3_pack1to8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s1_pack1to8_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 427 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s2_pack1to8_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | convolutiondepthwise_3x3_pack4.h | 45 float32x4_t _k00 = vld1q_f32(k0); in convdw3x3s1_pack4_neon() local 206 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 306 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 386 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 500 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 601 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 679 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 752 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 941 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() 1076 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 45 float32x4_t _k00 = vcvt_f32_bf16(vld1_u16(k0)); in convdw3x3s1_pack4_bf16s_neon() local 278 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 424 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 529 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 681 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 814 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 910 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1001 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1243 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() 1351 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s1_pack8_fp16sa_neon() local 202 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 297 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 376 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 481 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 548 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 605 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 655 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s2_pack8_fp16sa_neon() local 759 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() 834 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() [all …]
|
H A D | convolution_3x3_pack1to4_fp16s.h | 45 float16x4_t _k00 = vld1_f16(k0); in conv3x3s1_pack1to4_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to4_fp16sa_neon() 427 float16x4_t _k00 = vld1_f16(k0); in conv3x3s2_pack1to4_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to4_fp16sa_neon()
|
H A D | convolution_3x3_pack1to8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s1_pack1to8_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 427 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s2_pack1to8_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon()
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | convolutiondepthwise_3x3_pack4.h | 45 float32x4_t _k00 = vld1q_f32(k0); in convdw3x3s1_pack4_neon() local 206 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 306 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 386 "w"(_k00), // %12 in convdw3x3s1_pack4_neon() 500 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 601 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 679 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 752 "w"(_k00), // %8 in convdw3x3s1_pack4_neon() 941 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() 1076 "w"(_k00), // %8 in convdw3x3s2_pack4_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 45 float32x4_t _k00 = vcvt_f32_bf16(vld1_u16(k0)); in convdw3x3s1_pack4_bf16s_neon() local 278 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 424 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 529 "w"(_k00), // %12 in convdw3x3s1_pack4_bf16s_neon() 681 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 814 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 910 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1001 "w"(_k00), // %8 in convdw3x3s1_pack4_bf16s_neon() 1243 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() 1351 "w"(_k00), // %8 in convdw3x3s2_pack4_bf16s_neon() [all …]
|
H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s1_pack8_fp16sa_neon() local 202 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 297 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 376 "w"(_k00), // %12 in convdw3x3s1_pack8_fp16sa_neon() 481 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 548 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 605 "w"(_k00), // %8 in convdw3x3s1_pack8_fp16sa_neon() 655 float16x8_t _k00 = vld1q_f16(k0); in convdw3x3s2_pack8_fp16sa_neon() local 759 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() 834 "w"(_k00), // %8 in convdw3x3s2_pack8_fp16sa_neon() [all …]
|
H A D | convolution_3x3_pack1to8_fp16s.h | 45 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s1_pack1to8_fp16sa_neon() local 174 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 259 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 320 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 372 "w"(_k00), // %8 in conv3x3s1_pack1to8_fp16sa_neon() 427 float16x8_t _k00 = vld1q_f16(k0); in conv3x3s2_pack1to8_fp16sa_neon() local 514 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 574 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon() 626 "w"(_k00), // %8 in conv3x3s2_pack1to8_fp16sa_neon()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
H A D | convolution_3x3_pack1ton.h | 48 vfloat32m1_t _k00 = vle32_v_f32m1(k0, vl); in conv3x3s1_pack1ton_rvv() local 73 _sum0 = vfmacc_vf_f32m1(_sum0, r0[0], _k00, vl); in conv3x3s1_pack1ton_rvv() 74 _sum1 = vfmacc_vf_f32m1(_sum1, r0[1], _k00, vl); in conv3x3s1_pack1ton_rvv() 75 _sum2 = vfmacc_vf_f32m1(_sum2, r0[2], _k00, vl); in conv3x3s1_pack1ton_rvv() 76 _sum3 = vfmacc_vf_f32m1(_sum3, r0[3], _k00, vl); in conv3x3s1_pack1ton_rvv() 77 _sum4 = vfmacc_vf_f32m1(_sum4, r0[4], _k00, vl); in conv3x3s1_pack1ton_rvv() 78 _sum5 = vfmacc_vf_f32m1(_sum5, r0[5], _k00, vl); in conv3x3s1_pack1ton_rvv() 79 _sum6 = vfmacc_vf_f32m1(_sum6, r0[6], _k00, vl); in conv3x3s1_pack1ton_rvv() 80 _sum7 = vfmacc_vf_f32m1(_sum7, r0[7], _k00, vl); in conv3x3s1_pack1ton_rvv() 170 _sum0 = vfmacc_vf_f32m1(_sum0, r0[0], _k00, vl); in conv3x3s1_pack1ton_rvv() [all …]
|