/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/ |
H A D | mat_pixel_resize.cpp | 194 short* rows1p = rows1; in resize_bilinear_c1() local 238 rows1p += 8; in resize_bilinear_c1() 280 "1"(rows1p), in resize_bilinear_c1() 436 rows1p += 2; in resize_bilinear_c2() 492 rows1p += 2; in resize_bilinear_c2() 547 rows1p += 8; in resize_bilinear_c2() 589 "1"(rows1p), in resize_bilinear_c2() 747 rows1p += 3; in resize_bilinear_c3() 812 rows1p += 3; in resize_bilinear_c3() 867 rows1p += 8; in resize_bilinear_c3() [all …]
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/ |
H A D | mat_pixel_resize.cpp | 194 short* rows1p = rows1; in resize_bilinear_c1() local 238 rows1p += 8; in resize_bilinear_c1() 280 "1"(rows1p), in resize_bilinear_c1() 436 rows1p += 2; in resize_bilinear_c2() 492 rows1p += 2; in resize_bilinear_c2() 547 rows1p += 8; in resize_bilinear_c2() 589 "1"(rows1p), in resize_bilinear_c2() 747 rows1p += 3; in resize_bilinear_c3() 812 rows1p += 3; in resize_bilinear_c3() 867 rows1p += 8; in resize_bilinear_c3() [all …]
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/ |
H A D | mat_pixel_resize.cpp | 194 short* rows1p = rows1; in resize_bilinear_c1() local 238 rows1p += 8; in resize_bilinear_c1() 280 "1"(rows1p), in resize_bilinear_c1() 436 rows1p += 2; in resize_bilinear_c2() 492 rows1p += 2; in resize_bilinear_c2() 547 rows1p += 8; in resize_bilinear_c2() 589 "1"(rows1p), in resize_bilinear_c2() 747 rows1p += 3; in resize_bilinear_c3() 812 rows1p += 3; in resize_bilinear_c3() 867 rows1p += 8; in resize_bilinear_c3() [all …]
|
/dports/misc/ncnn/ncnn-20211208/src/ |
H A D | mat_pixel_resize.cpp | 194 short* rows1p = rows1; in resize_bilinear_c1() local 238 rows1p += 8; in resize_bilinear_c1() 280 "1"(rows1p), in resize_bilinear_c1() 436 rows1p += 2; in resize_bilinear_c2() 492 rows1p += 2; in resize_bilinear_c2() 547 rows1p += 8; in resize_bilinear_c2() 589 "1"(rows1p), in resize_bilinear_c2() 747 rows1p += 3; in resize_bilinear_c3() 812 rows1p += 3; in resize_bilinear_c3() 867 rows1p += 8; in resize_bilinear_c3() [all …]
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/ |
H A D | mat_pixel_resize.cpp | 194 short* rows1p = rows1; in resize_bilinear_c1() local 238 rows1p += 8; in resize_bilinear_c1() 280 "1"(rows1p), in resize_bilinear_c1() 436 rows1p += 2; in resize_bilinear_c2() 492 rows1p += 2; in resize_bilinear_c2() 547 rows1p += 8; in resize_bilinear_c2() 589 "1"(rows1p), in resize_bilinear_c2() 747 rows1p += 3; in resize_bilinear_c3() 812 rows1p += 3; in resize_bilinear_c3() 867 rows1p += 8; in resize_bilinear_c3() [all …]
|
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 139 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_fp16s() 156 rows1p += 8; in resize_bilinear_image_fp16s() 198 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 207 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 220 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 244 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 255 float16x8_t _rows1 = vld1q_f16(rows1p); in resize_bilinear_image_fp16sa() [all …]
|
H A D | interp_bilinear_pack4_fp16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 58 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 71 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 90 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 102 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 115 rows1p += 4; in resize_bilinear_image_pack4_fp16s() 152 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 165 vst1_f16(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16sa() 178 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 209 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local [all …]
|
H A D | interp_bilinear_bf16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 54 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 67 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 78 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 91 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 107 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_bf16s() 115 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image_bf16s() 124 rows1p += 8; in resize_bilinear_image_bf16s() 130 *Dp++ = float32_to_bfloat16(*rows0p++ * b0 + *rows1p++ * b1); in resize_bilinear_image_bf16s()
|
H A D | interp_bilinear.h | 82 float* rows1p = rows1; in resize_bilinear_image() local 100 vst1_f32(rows1p + dx, _rows1); in resize_bilinear_image() 112 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 125 float* rows1p = rows1; in resize_bilinear_image() local 151 vst1_f32(rows1p + dx, _rows1); in resize_bilinear_image() 165 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 178 float* rows1p = rows1; in resize_bilinear_image() local 194 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image() 202 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image() 211 rows1p += 8; in resize_bilinear_image() [all …]
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 139 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_fp16s() 156 rows1p += 8; in resize_bilinear_image_fp16s() 198 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 207 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 220 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 244 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 255 float16x8_t _rows1 = vld1q_f16(rows1p); in resize_bilinear_image_fp16sa() [all …]
|
H A D | interp_bilinear_pack4_fp16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 58 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 71 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 90 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 102 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 115 rows1p += 4; in resize_bilinear_image_pack4_fp16s() 152 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 165 vst1_f16(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16sa() 178 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 209 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local [all …]
|
H A D | interp_bilinear_bf16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 54 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 67 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 78 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 91 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 107 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_bf16s() 115 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image_bf16s() 124 rows1p += 8; in resize_bilinear_image_bf16s() 130 *Dp++ = float32_to_bfloat16(*rows0p++ * b0 + *rows1p++ * b1); in resize_bilinear_image_bf16s()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 139 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_fp16s() 156 rows1p += 8; in resize_bilinear_image_fp16s() 198 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 207 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 220 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 244 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 255 float16x8_t _rows1 = vld1q_f16(rows1p); in resize_bilinear_image_fp16sa() [all …]
|
H A D | interp_bilinear_pack4_fp16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 58 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 71 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 90 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 102 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 115 rows1p += 4; in resize_bilinear_image_pack4_fp16s() 152 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 165 vst1_f16(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16sa() 178 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 209 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local [all …]
|
H A D | interp_bilinear_bf16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 54 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 67 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 78 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 91 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 107 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_bf16s() 115 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image_bf16s() 124 rows1p += 8; in resize_bilinear_image_bf16s() 130 *Dp++ = float32_to_bfloat16(*rows0p++ * b0 + *rows1p++ * b1); in resize_bilinear_image_bf16s()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 139 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_fp16s() 156 rows1p += 8; in resize_bilinear_image_fp16s() 198 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 207 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 220 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 244 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 255 float16x8_t _rows1 = vld1q_f16(rows1p); in resize_bilinear_image_fp16sa() [all …]
|
H A D | interp_bilinear_pack4_fp16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 58 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 71 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 90 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 102 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 115 rows1p += 4; in resize_bilinear_image_pack4_fp16s() 152 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 165 vst1_f16(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16sa() 178 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 209 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local [all …]
|
H A D | interp_bilinear_bf16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 54 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 67 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 78 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 91 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 107 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_bf16s() 115 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image_bf16s() 124 rows1p += 8; in resize_bilinear_image_bf16s() 130 *Dp++ = float32_to_bfloat16(*rows0p++ * b0 + *rows1p++ * b1); in resize_bilinear_image_bf16s()
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 139 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_fp16s() 156 rows1p += 8; in resize_bilinear_image_fp16s() 198 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 207 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 220 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 244 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 255 float16x8_t _rows1 = vld1q_f16(rows1p); in resize_bilinear_image_fp16sa() [all …]
|
H A D | interp_bilinear_pack4_fp16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 58 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 71 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 90 vst1q_f32(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16s() 102 float* rows1p = rows1; in resize_bilinear_image_pack4_fp16s() local 115 rows1p += 4; in resize_bilinear_image_pack4_fp16s() 152 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 165 vst1_f16(rows1p + dx * 4, _rows1); in resize_bilinear_image_pack4_fp16sa() 178 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local 209 __fp16* rows1p = rows1; in resize_bilinear_image_pack4_fp16sa() local [all …]
|
H A D | interp_bilinear_bf16s.h | 45 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 54 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 67 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 78 rows1p[dx] = bfloat16_to_float32(S1p[0]) * a0 + bfloat16_to_float32(S1p[1]) * a1; in resize_bilinear_image_bf16s() 91 float* rows1p = rows1; in resize_bilinear_image_bf16s() local 107 float32x4_t _rows1 = vld1q_f32(rows1p); in resize_bilinear_image_bf16s() 115 float32x4_t _rows1n = vld1q_f32(rows1p + 4); in resize_bilinear_image_bf16s() 124 rows1p += 8; in resize_bilinear_image_bf16s() 130 *Dp++ = float32_to_bfloat16(*rows0p++ * b0 + *rows1p++ * b1); in resize_bilinear_image_bf16s()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/riscv/ |
H A D | interp_bilinear_fp16s.h | 82 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 104 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 128 float* rows1p = rows1; in resize_bilinear_image_fp16s() local 137 vfloat32m8_t _rows1 = vle32_v_f32m8(rows1p, vl); in resize_bilinear_image_fp16s() 145 rows1p += vl; in resize_bilinear_image_fp16s() 183 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 192 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 205 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local 216 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image_fp16sa() 229 __fp16* rows1p = rows1; in resize_bilinear_image_fp16sa() local [all …]
|
H A D | interp_bilinear.h | 82 float* rows1p = rows1; in resize_bilinear_image() local 102 vse32_v_f32m4(rows1p, _rows1, vl); in resize_bilinear_image() 106 rows1p += vl; in resize_bilinear_image() 117 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 131 float* rows1p = rows1; in resize_bilinear_image() local 156 vse32_v_f32m4(rows1p, _rows1, vl); in resize_bilinear_image() 161 rows1p += vl; in resize_bilinear_image() 174 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 188 float* rows1p = rows1; in resize_bilinear_image() local 206 rows1p += vl; in resize_bilinear_image() [all …]
|
H A D | interp_bilinear_packn_fp16s.h | 48 float* rows1p = rows1; in resize_bilinear_image_packn_fp16s() local 59 vse32_v_f32m2(rows1p + dx * packn, _rows1, vl); in resize_bilinear_image_packn_fp16s() 72 float* rows1p = rows1; in resize_bilinear_image_packn_fp16s() local 101 float* rows1p = rows1; in resize_bilinear_image_packn_fp16s() local 107 vfloat32m2_t _rows1 = vle32_v_f32m2(rows1p, vl); in resize_bilinear_image_packn_fp16s() 115 rows1p += packn; in resize_bilinear_image_packn_fp16s() 155 __fp16* rows1p = rows1; in resize_bilinear_image_packn_fp16sa() local 179 __fp16* rows1p = rows1; in resize_bilinear_image_packn_fp16sa() local 208 __fp16* rows1p = rows1; in resize_bilinear_image_packn_fp16sa() local 214 vfloat16m1_t _rows1 = vle16_v_f16m1(rows1p, vl); in resize_bilinear_image_packn_fp16sa() [all …]
|
/dports/misc/ncnn/ncnn-20211208/src/layer/mips/ |
H A D | interp_bilinear.h | 82 float* rows1p = rows1; in resize_bilinear_image() local 91 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 104 float* rows1p = rows1; in resize_bilinear_image() local 115 rows1p[dx] = S1p[0] * a0 + S1p[1] * a1; in resize_bilinear_image() 128 float* rows1p = rows1; in resize_bilinear_image() local 144 v4f32 _rows1 = (v4f32)__msa_ld_w(rows1p, 0); in resize_bilinear_image() 152 v4f32 _rows1n = (v4f32)__msa_ld_w(rows1p + 4, 0); in resize_bilinear_image() 161 rows1p += 8; in resize_bilinear_image() 167 *Dp++ = *rows0p++ * b0 + *rows1p++ * b1; in resize_bilinear_image()
|