/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | interp_bicubic_pack4_fp16s.h | 60 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 64 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 100 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 106 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 318 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 322 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 358 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 364 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 411 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 419 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() [all …]
|
H A D | interp_bicubic_pack4.h | 60 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 100 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 153 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 212 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4()
|
H A D | interp_bicubic_pack4_bf16s.h | 60 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 100 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 153 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 212 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s()
|
H A D | interp_bicubic_pack8_fp16s.h | 60 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 64 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 100 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 106 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 153 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 161 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 211 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 221 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | interp_bicubic_pack4_fp16s.h | 60 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 64 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 100 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 106 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 318 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 322 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 358 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 364 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 411 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 419 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() [all …]
|
H A D | interp_bicubic_pack4.h | 60 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 100 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 153 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 212 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4()
|
H A D | interp_bicubic_pack4_bf16s.h | 60 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 100 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 153 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 212 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s()
|
H A D | interp_bicubic_pack8_fp16s.h | 60 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 64 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 100 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 106 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 153 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 161 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 211 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 221 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | interp_bicubic_pack4_fp16s.h | 60 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 64 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 100 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 106 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 318 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 322 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 358 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 364 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 411 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 419 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() [all …]
|
H A D | interp_bicubic_pack8_fp16s.h | 60 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 64 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 100 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 106 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 153 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 161 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 211 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 221 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa()
|
H A D | interp_bicubic_pack4.h | 60 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 100 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 153 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 212 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4()
|
H A D | interp_bicubic_pack4_bf16s.h | 60 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 100 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 153 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 212 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s()
|
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | interp_bicubic_pack4_fp16s.h | 60 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 64 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 100 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 106 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 318 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 322 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 358 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 364 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 411 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 419 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() [all …]
|
H A D | interp_bicubic_pack4.h | 60 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 100 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 153 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 212 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4()
|
H A D | interp_bicubic_pack4_bf16s.h | 60 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 100 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 153 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 212 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s()
|
H A D | interp_bicubic_pack8_fp16s.h | 60 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 64 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 100 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 106 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 153 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 161 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 211 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 221 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa()
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | interp_bicubic_pack4_fp16s.h | 60 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 64 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 100 float32x4_t _S31 = vcvt_f32_f16(vld1_f16(S3p + 0)); in resize_bicubic_image_pack4_fp16s() local 106 _rows3 = vfmaq_laneq_f32(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16s() 318 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 322 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 358 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 364 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() 411 float16x4_t _S31 = vld1_f16(S3p + 0); in resize_bicubic_image_pack4_fp16sa() local 419 _rows3 = vfma_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack4_fp16sa() [all …]
|
H A D | interp_bicubic_pack4.h | 60 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 100 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 153 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4() 212 float32x4_t _S31 = vld1q_f32(S3p + 0); in resize_bicubic_image_pack4() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4()
|
H A D | interp_bicubic_pack8_fp16s.h | 60 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 64 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 100 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 106 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 153 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 161 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa() 211 float16x8_t _S31 = vld1q_f16(S3p + 0); in resize_bicubic_image_pack8_fp16sa() local 221 _rows3 = vfmaq_lane_f16(_rows3, _S31, _a0123, 1); in resize_bicubic_image_pack8_fp16sa()
|
H A D | interp_bicubic_pack4_bf16s.h | 60 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 64 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 100 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 106 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 153 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 161 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s() 212 float32x4_t _S31 = vcvt_f32_bf16(vld1_u16(S3p + 0)); in resize_bicubic_image_pack4_bf16s() local 222 _rows3 = vmlaq_lane_f32(_rows3, _S31, vget_low_f32(_a0123), 1); in resize_bicubic_image_pack4_bf16s()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
H A D | interp_bicubic_packn_fp16s.h | 61 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16s() local 64 …f_f32m2(vfwmacc_vf_f32m2(vfwmul_vf_f32m2(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn_fp16s() 97 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16s() local 101 …f_f32m2(vfwmacc_vf_f32m2(vfwmul_vf_f32m2(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn_fp16s() 143 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16s() local 191 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16s() local 290 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16sa() local 293 …_vf_f16m1(vfmacc_vf_f16m1(vfmul_vf_f16m1(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn_fp16sa() 326 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16sa() local 372 vfloat16m1_t _S31 = vle16_v_f16m1(S3p, vl); in resize_bicubic_image_packn_fp16sa() local [all …]
|
H A D | interp_bicubic_packn.h | 61 vfloat32m1_t _S31 = vle32_v_f32m1(S3p, vl); in resize_bicubic_image_packn() local 64 …_vf_f32m1(vfmacc_vf_f32m1(vfmul_vf_f32m1(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn() 97 vfloat32m1_t _S31 = vle32_v_f32m1(S3p, vl); in resize_bicubic_image_packn() local 101 …_vf_f32m1(vfmacc_vf_f32m1(vfmul_vf_f32m1(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn() 143 vfloat32m1_t _S31 = vle32_v_f32m1(S3p, vl); in resize_bicubic_image_packn() local 148 …_vf_f32m1(vfmacc_vf_f32m1(vfmul_vf_f32m1(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn() 191 vfloat32m1_t _S31 = vle32_v_f32m1(S3p, vl); in resize_bicubic_image_packn() local 197 …_vf_f32m1(vfmacc_vf_f32m1(vfmul_vf_f32m1(_S30, alphap[0], vl), alphap[1], _S31, vl), alphap[2], _S… in resize_bicubic_image_packn()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/mips/ |
H A D | interp_bicubic_pack4.h | 63 v4f32 _S31 = (v4f32)__msa_ld_w(S3p + 0, 0); in resize_bicubic_image_pack4() local 67 _rows3 = __msa_fmadd_w(_rows3, _S31, _a1); in resize_bicubic_image_pack4() 106 v4f32 _S31 = (v4f32)__msa_ld_w(S3p + 0, 0); in resize_bicubic_image_pack4() local 112 _rows3 = __msa_fmadd_w(_rows3, _S31, _a1); in resize_bicubic_image_pack4() 162 v4f32 _S31 = (v4f32)__msa_ld_w(S3p + 0, 0); in resize_bicubic_image_pack4() local 170 _rows3 = __msa_fmadd_w(_rows3, _S31, _a1); in resize_bicubic_image_pack4() 223 v4f32 _S31 = (v4f32)__msa_ld_w(S3p + 0, 0); in resize_bicubic_image_pack4() local 233 _rows3 = __msa_fmadd_w(_rows3, _S31, _a1); in resize_bicubic_image_pack4()
|
/dports/games/libretro-pcsx_rearmed/pcsx_rearmed-c2d67cd/deps/lightning/include/lightning/ |
H A D | jit_arm.h | 86 _S31, enumerator
|
/dports/games/libretro-beetle_psx/beetle-psx-libretro-3ec155d/deps/lightning/include/lightning/ |
H A D | jit_arm.h | 86 _S31, enumerator
|