| /dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack1to4_fp16s.h | 52 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s1_pack1to4_fp16sa_neon() local
434 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s2_pack1to4_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s1_pack1to8_fp16sa_neon() local
434 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s2_pack1to8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s1_pack8_fp16sa_neon() local
662 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s2_pack8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4.h | 52 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s1_pack4_neon() local
841 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s2_pack4_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 52 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s1_pack4_bf16s_neon() local
1090 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s2_pack4_bf16s_neon() local
|
| /dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack1to8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s1_pack1to8_fp16sa_neon() local
434 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s2_pack1to8_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 52 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s1_pack1to4_fp16sa_neon() local
434 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s2_pack1to4_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s1_pack8_fp16sa_neon() local
662 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s2_pack8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4.h | 52 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s1_pack4_neon() local
841 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s2_pack4_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 52 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s1_pack4_bf16s_neon() local
1090 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s2_pack4_bf16s_neon() local
|
| /dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
| H A D | convolution_3x3_pack1to4_fp16s.h | 52 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s1_pack1to4_fp16sa_neon() local
434 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s2_pack1to4_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s1_pack8_fp16sa_neon() local
662 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s2_pack8_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s1_pack1to8_fp16sa_neon() local
434 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s2_pack1to8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4.h | 52 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s1_pack4_neon() local
841 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s2_pack4_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 52 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s1_pack4_bf16s_neon() local
1090 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s2_pack4_bf16s_neon() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s1_pack8_fp16sa_neon() local
662 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s2_pack8_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to4_fp16s.h | 52 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s1_pack1to4_fp16sa_neon() local
434 float16x4_t _k21 = vld1_f16(k0 + 28); in conv3x3s2_pack1to4_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s1_pack1to8_fp16sa_neon() local
434 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s2_pack1to8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4.h | 52 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s1_pack4_neon() local
841 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s2_pack4_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 52 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s1_pack4_bf16s_neon() local
1090 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s2_pack4_bf16s_neon() local
|
| /dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
| H A D | convolutiondepthwise_3x3_pack8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s1_pack8_fp16sa_neon() local
662 float16x8_t _k21 = vld1q_f16(k0 + 56); in convdw3x3s2_pack8_fp16sa_neon() local
|
| H A D | convolution_3x3_pack1to8_fp16s.h | 52 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s1_pack1to8_fp16sa_neon() local
434 float16x8_t _k21 = vld1q_f16(k0 + 56); in conv3x3s2_pack1to8_fp16sa_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4_bf16s.h | 52 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s1_pack4_bf16s_neon() local
1090 float32x4_t _k21 = vcvt_f32_bf16(vld1_u16(k0 + 28)); in convdw3x3s2_pack4_bf16s_neon() local
|
| H A D | convolutiondepthwise_3x3_pack4.h | 52 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s1_pack4_neon() local
841 float32x4_t _k21 = vld1q_f32(k0 + 28); in convdw3x3s2_pack4_neon() local
|
| /dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
| H A D | convolution_3x3_pack1ton.h | 55 vfloat32m1_t _k21 = vle32_v_f32m1(k0 + packn * 7, vl); in conv3x3s1_pack1ton_rvv() local
333 vfloat32m1_t _k21 = vle32_v_f32m1(k0 + packn * 7, vl); in conv3x3s2_pack1ton_rvv() local
|