Home
last modified time | relevance | path

Searched defs:_k01 (Results 1 – 25 of 107) sorted by relevance

12345

/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/
H A Dconvolution_3x3_pack1to4_fp16s.h46 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s1_pack1to4_fp16sa_neon() local
428 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s2_pack1to4_fp16sa_neon() local
H A Dconvolution_3x3_pack1to8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s1_pack1to8_fp16sa_neon() local
428 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s2_pack1to8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s1_pack8_fp16sa_neon() local
656 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s2_pack8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack4.h46 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s1_pack4_neon() local
835 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s2_pack4_neon() local
H A Dconvolutiondepthwise_3x3_pack4_bf16s.h46 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s1_pack4_bf16s_neon() local
1084 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s2_pack4_bf16s_neon() local
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/
H A Dconvolution_3x3_pack1to8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s1_pack1to8_fp16sa_neon() local
428 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s2_pack1to8_fp16sa_neon() local
H A Dconvolution_3x3_pack1to4_fp16s.h46 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s1_pack1to4_fp16sa_neon() local
428 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s2_pack1to4_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s1_pack8_fp16sa_neon() local
656 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s2_pack8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack4.h46 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s1_pack4_neon() local
835 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s2_pack4_neon() local
H A Dconvolutiondepthwise_3x3_pack4_bf16s.h46 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s1_pack4_bf16s_neon() local
1084 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s2_pack4_bf16s_neon() local
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/
H A Dconvolution_3x3_pack1to4_fp16s.h46 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s1_pack1to4_fp16sa_neon() local
428 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s2_pack1to4_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s1_pack8_fp16sa_neon() local
656 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s2_pack8_fp16sa_neon() local
H A Dconvolution_3x3_pack1to8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s1_pack1to8_fp16sa_neon() local
428 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s2_pack1to8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack4.h46 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s1_pack4_neon() local
835 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s2_pack4_neon() local
H A Dconvolutiondepthwise_3x3_pack4_bf16s.h46 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s1_pack4_bf16s_neon() local
1084 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s2_pack4_bf16s_neon() local
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/
H A Dconvolutiondepthwise_3x3_pack8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s1_pack8_fp16sa_neon() local
656 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s2_pack8_fp16sa_neon() local
H A Dconvolution_3x3_pack1to4_fp16s.h46 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s1_pack1to4_fp16sa_neon() local
428 float16x4_t _k01 = vld1_f16(k0 + 4); in conv3x3s2_pack1to4_fp16sa_neon() local
H A Dconvolution_3x3_pack1to8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s1_pack1to8_fp16sa_neon() local
428 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s2_pack1to8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack4.h46 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s1_pack4_neon() local
835 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s2_pack4_neon() local
H A Dconvolutiondepthwise_3x3_pack4_bf16s.h46 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s1_pack4_bf16s_neon() local
1084 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s2_pack4_bf16s_neon() local
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/
H A Dconvolutiondepthwise_3x3_pack8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s1_pack8_fp16sa_neon() local
656 float16x8_t _k01 = vld1q_f16(k0 + 8); in convdw3x3s2_pack8_fp16sa_neon() local
H A Dconvolution_3x3_pack1to8_fp16s.h46 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s1_pack1to8_fp16sa_neon() local
428 float16x8_t _k01 = vld1q_f16(k0 + 8); in conv3x3s2_pack1to8_fp16sa_neon() local
H A Dconvolutiondepthwise_3x3_pack4_bf16s.h46 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s1_pack4_bf16s_neon() local
1084 float32x4_t _k01 = vcvt_f32_bf16(vld1_u16(k0 + 4)); in convdw3x3s2_pack4_bf16s_neon() local
H A Dconvolutiondepthwise_3x3_pack4.h46 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s1_pack4_neon() local
835 float32x4_t _k01 = vld1q_f32(k0 + 4); in convdw3x3s2_pack4_neon() local
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/x86/
H A Dconvolution_2x2_pack8.h64 __m256 _k01 = _mm256_loadu_ps(kptr + 8); in conv2x2s1_pack8_avx() local
257 __m256 _k01 = _mm256_loadu_ps(kptr + 8); in conv2x2s1_pack8_avx() local

12345