| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | // SPDX-FileCopyrightText: 2024 - 2026 Arm Limited and/or its affiliates <open-source-office@arm.com> | ||
| 2 | // | ||
| 3 | // SPDX-License-Identifier: Apache-2.0 | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <cstdint> | ||
| 7 | #include <cstdlib> | ||
| 8 | #include <memory> | ||
| 9 | |||
| 10 | #include "kleidicv/config.h" | ||
| 11 | #include "kleidicv/ctypes.h" | ||
| 12 | #include "kleidicv/sve2.h" | ||
| 13 | #include "kleidicv/types.h" | ||
| 14 | #include "kleidicv/utils.h" | ||
| 15 | |||
| 16 | namespace KLEIDICV_TARGET_NAMESPACE { | ||
| 17 | |||
| 18 | // Scharr filtering in both horizontal and vertical directions, horizontal and | ||
| 19 | // vertical derivative approximations are stored interleaved. | ||
| 20 | // | ||
| 21 | // The applied weights for the horizontal approximation, as the kernel is | ||
| 22 | // mirrored both vertically and horizontally during the convolution: | ||
| 23 | // [ -3 0 3 ] [ 3 ] | ||
| 24 | // F = [ -10 0 10 ] = [ 10 ] * [ -1, 0, 1 ] | ||
| 25 | // [ -3 0 3 ] [ 3 ] | ||
| 26 | // | ||
| 27 | // The applied weights for the vertical approximation, as the kernel is mirrored | ||
| 28 | // both vertically and horizontally during the convolution: | ||
| 29 | // [ -3 -10 -3 ] [ -1 ] | ||
| 30 | // F = [ 0, 0, 0 ] = [ 0 ] * [ 3, 10, 3 ] | ||
| 31 | // [ 3 10 3 ] [ 1 ] | ||
| 32 | // | ||
| 33 | class ScharrInterleaved { | ||
| 34 | using SourceType = uint8_t; | ||
| 35 | using SourceVecTraits = VecTraits<SourceType>; | ||
| 36 | using SourceVectorType = typename SourceVecTraits::VectorType; | ||
| 37 | using SourceVector2Type = typename SourceVecTraits::Vector2Type; | ||
| 38 | using SourceVector3Type = typename SourceVecTraits::Vector3Type; | ||
| 39 | using BufferType = int16_t; | ||
| 40 | using BufferVecTraits = VecTraits<BufferType>; | ||
| 41 | using BufferVectorType = typename BufferVecTraits::VectorType; | ||
| 42 | using BufferVector2Type = typename BufferVecTraits::Vector2Type; | ||
| 43 | using BufferVector3Type = typename BufferVecTraits::Vector3Type; | ||
| 44 | using DestinationType = int16_t; | ||
| 45 | using DestinationVecTraits = VecTraits<DestinationType>; | ||
| 46 | using DestinationVectorType = typename DestinationVecTraits::VectorType; | ||
| 47 | |||
| 48 | public: | ||
| 49 | 990 | ScharrInterleaved(Rows<BufferType> hori_deriv_buffer, | |
| 50 | Rows<BufferType> vert_deriv_buffer, | ||
| 51 | size_t width) KLEIDICV_STREAMING | ||
| 52 | 990 | : hori_deriv_buffer_(hori_deriv_buffer), | |
| 53 | 990 | vert_deriv_buffer_(vert_deriv_buffer), | |
| 54 | 990 | width_(width) {} | |
| 55 | |||
| 56 | 990 | void process(Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows, | |
| 57 | size_t y_begin, size_t y_end) KLEIDICV_STREAMING { | ||
| 58 |
2/2✓ Branch 0 taken 990 times.
✓ Branch 1 taken 17472 times.
|
18462 | for (size_t i = y_begin; i < y_end; ++i) { |
| 59 | 17472 | process_vertical(src_rows.at(static_cast<ptrdiff_t>(i))); | |
| 60 | 17472 | process_horizontal(dst_rows.at(static_cast<ptrdiff_t>(i))); | |
| 61 | 17472 | } | |
| 62 | 990 | } | |
| 63 | |||
| 64 | private: | ||
| 65 | 1334 | void vertical_vector_path_x2(svbool_t pg, Rows<const SourceType> src_rows, | |
| 66 | ptrdiff_t index) KLEIDICV_STREAMING { | ||
| 67 | #if KLEIDICV_TARGET_SME2 | ||
| 68 | 82 | svcount_t pc8 = SourceVecTraits::svptrue_c(); | |
| 69 | 82 | SourceVector2Type src0 = svld1_x2(pc8, &src_rows.at(0)[index]); | |
| 70 | 82 | SourceVector2Type src1 = svld1_x2(pc8, &src_rows.at(1)[index]); | |
| 71 | 82 | SourceVector2Type src2 = svld1_x2(pc8, &src_rows.at(2)[index]); | |
| 72 | 164 | SourceVector3Type src_a = | |
| 73 | 82 | svcreate3(svget2(src0, 0), svget2(src1, 0), svget2(src2, 0)); | |
| 74 | 164 | SourceVector3Type src_b = | |
| 75 | 82 | svcreate3(svget2(src0, 1), svget2(src1, 1), svget2(src2, 1)); | |
| 76 | #else | ||
| 77 | 2504 | SourceVector3Type src_a = svcreate3(svld1(pg, &src_rows.at(0)[index]), | |
| 78 | 1252 | svld1(pg, &src_rows.at(1)[index]), | |
| 79 | 1252 | svld1(pg, &src_rows.at(2)[index])); | |
| 80 | 2504 | SourceVector3Type src_b = | |
| 81 | 2504 | svcreate3(svld1_vnum(pg, &src_rows.at(0)[index], 1), | |
| 82 | 1252 | svld1_vnum(pg, &src_rows.at(1)[index], 1), | |
| 83 | 1252 | svld1_vnum(pg, &src_rows.at(2)[index], 1)); | |
| 84 | #endif | ||
| 85 | |||
| 86 | 1334 | svint16x2_t hori_interleaved_a = vertical_compute_hori_approx(pg, src_a); | |
| 87 | 1334 | svint16x2_t vert_interleaved_a = vertical_compute_vert_approx(src_a); | |
| 88 | |||
| 89 | 1334 | svint16x2_t hori_interleaved_b = vertical_compute_hori_approx(pg, src_b); | |
| 90 | 1334 | svint16x2_t vert_interleaved_b = vertical_compute_vert_approx(src_b); | |
| 91 | |||
| 92 | 1334 | svst2(pg, &hori_deriv_buffer_[index], hori_interleaved_a); | |
| 93 | 1334 | svst2_vnum(pg, &hori_deriv_buffer_[index], 2, hori_interleaved_b); | |
| 94 | |||
| 95 | 1334 | svst2(pg, &vert_deriv_buffer_[index], vert_interleaved_a); | |
| 96 | 1334 | svst2_vnum(pg, &vert_deriv_buffer_[index], 2, vert_interleaved_b); | |
| 97 | 1334 | } | |
| 98 | |||
| 99 | 18504 | void vertical_vector_path(svbool_t pg, Rows<const SourceType> src_rows, | |
| 100 | ptrdiff_t index) KLEIDICV_STREAMING { | ||
| 101 | 18504 | SourceVectorType src_0 = svld1(pg, &src_rows.at(0)[index]); | |
| 102 | 18504 | SourceVectorType src_1 = svld1(pg, &src_rows.at(1)[index]); | |
| 103 | 18504 | SourceVectorType src_2 = svld1(pg, &src_rows.at(2)[index]); | |
| 104 | |||
| 105 | 18504 | SourceVector3Type src = svcreate3(src_0, src_1, src_2); | |
| 106 | 18504 | svint16x2_t hori_interleaved = vertical_compute_hori_approx(pg, src); | |
| 107 | 18504 | svint16x2_t vert_interleaved = vertical_compute_vert_approx(src); | |
| 108 | |||
| 109 | 18504 | svst2(pg, &hori_deriv_buffer_[index], hori_interleaved); | |
| 110 | 18504 | svst2(pg, &vert_deriv_buffer_[index], vert_interleaved); | |
| 111 | 18504 | } | |
| 112 | |||
| 113 | 21172 | svint16x2_t vertical_compute_hori_approx(svbool_t pg, SourceVector3Type src) | |
| 114 | KLEIDICV_STREAMING { | ||
| 115 | 21172 | SourceVectorType src_0 = svget3(src, 0); | |
| 116 | 21172 | SourceVectorType src_1 = svget3(src, 1); | |
| 117 | 21172 | SourceVectorType src_2 = svget3(src, 2); | |
| 118 | 21172 | svuint16_t hori_acc_b = svaddlb(src_0, src_2); | |
| 119 | 21172 | svuint16_t hori_acc_t = svaddlt(src_0, src_2); | |
| 120 | |||
| 121 | 21172 | hori_acc_b = svmul_n_u16_x(pg, hori_acc_b, 3); | |
| 122 | 21172 | hori_acc_t = svmul_n_u16_x(pg, hori_acc_t, 3); | |
| 123 | |||
| 124 | 21172 | hori_acc_b = svmlalb_n_u16(hori_acc_b, src_1, 10); | |
| 125 | 21172 | hori_acc_t = svmlalt_n_u16(hori_acc_t, src_1, 10); | |
| 126 | |||
| 127 | 63516 | return svcreate2(svreinterpret_s16(hori_acc_b), | |
| 128 | 21172 | svreinterpret_s16(hori_acc_t)); | |
| 129 | 21172 | } | |
| 130 | |||
| 131 | 21172 | svint16x2_t vertical_compute_vert_approx(SourceVector3Type src) | |
| 132 | KLEIDICV_STREAMING { | ||
| 133 | 21172 | SourceVectorType src_0 = svget3(src, 0); | |
| 134 | 21172 | SourceVectorType src_2 = svget3(src, 2); | |
| 135 | 21172 | svuint16_t vert_acc_b = svsublb(src_2, src_0); | |
| 136 | 21172 | svuint16_t vert_acc_t = svsublt(src_2, src_0); | |
| 137 | |||
| 138 | 63516 | return svcreate2(svreinterpret_s16(vert_acc_b), | |
| 139 | 21172 | svreinterpret_s16(vert_acc_t)); | |
| 140 | 21172 | } | |
| 141 | |||
| 142 | 17472 | void process_vertical(Rows<const SourceType> src_rows) KLEIDICV_STREAMING { | |
| 143 | 34944 | LoopUnroll2 loop{width_ * src_rows.channels(), | |
| 144 | 17472 | SourceVecTraits::num_lanes()}; | |
| 145 | 17472 | svbool_t pg_all = SourceVecTraits::svptrue(); | |
| 146 | |||
| 147 | 18806 | loop.unroll_twice([&](ptrdiff_t index) KLEIDICV_STREAMING { | |
| 148 | 1334 | vertical_vector_path_x2(pg_all, src_rows, index); | |
| 149 | 1334 | }); | |
| 150 | |||
| 151 | 19184 | loop.unroll_once([&](ptrdiff_t index) KLEIDICV_STREAMING { | |
| 152 | 1712 | vertical_vector_path(pg_all, src_rows, index); | |
| 153 | 1712 | }); | |
| 154 | |||
| 155 | 34264 | loop.remaining([&](ptrdiff_t index, ptrdiff_t length) KLEIDICV_STREAMING { | |
| 156 | 16792 | svbool_t pg = SourceVecTraits::svwhilelt(index, length); | |
| 157 | 16792 | vertical_vector_path(pg, src_rows, index); | |
| 158 | 16792 | }); | |
| 159 | 17472 | } | |
| 160 | |||
| 161 | 3518 | void horizontal_vector_path_x2(svbool_t pg, Rows<DestinationType> dst_rows, | |
| 162 | ptrdiff_t index, | ||
| 163 | ptrdiff_t channel) KLEIDICV_STREAMING { | ||
| 164 | #if KLEIDICV_TARGET_SME2 | ||
| 165 | 300 | svcount_t pc16 = BufferVecTraits::svptrue_c(); | |
| 166 | 300 | BufferVector2Type hori_0 = svld1_x2(pc16, &hori_deriv_buffer_[index]); | |
| 167 | 600 | BufferVector2Type hori_1 = | |
| 168 | 300 | svld1_x2(pc16, &hori_deriv_buffer_[index + channel * 2]); | |
| 169 | 300 | BufferVector2Type vert_0 = svld1_x2(pc16, &vert_deriv_buffer_[index]); | |
| 170 | 600 | BufferVector2Type vert_1 = | |
| 171 | 300 | svld1_x2(pc16, &vert_deriv_buffer_[index + channel]); | |
| 172 | 600 | BufferVector2Type vert_2 = | |
| 173 | 300 | svld1_x2(pc16, &vert_deriv_buffer_[index + channel * 2]); | |
| 174 | 300 | BufferVector2Type hori_a = svcreate2(svget2(hori_0, 0), svget2(hori_1, 0)); | |
| 175 | 300 | BufferVector2Type hori_b = svcreate2(svget2(hori_0, 1), svget2(hori_1, 1)); | |
| 176 | |||
| 177 | 600 | BufferVector3Type vert_a = | |
| 178 | 300 | svcreate3(svget2(vert_0, 0), svget2(vert_1, 0), svget2(vert_2, 0)); | |
| 179 | 600 | BufferVector3Type vert_b = | |
| 180 | 300 | svcreate3(svget2(vert_0, 1), svget2(vert_1, 1), svget2(vert_2, 1)); | |
| 181 | #else | ||
| 182 | 3218 | BufferVectorType hori_buff_0 = svld1(pg, &hori_deriv_buffer_[index]); | |
| 183 | 6436 | BufferVectorType hori_buff_3 = | |
| 184 | 3218 | svld1_vnum(pg, &hori_deriv_buffer_[index], 1); | |
| 185 | 6436 | BufferVectorType hori_buff_2 = | |
| 186 | 3218 | svld1(pg, &hori_deriv_buffer_[index + channel * 2]); | |
| 187 | 6436 | BufferVectorType hori_buff_4 = | |
| 188 | 3218 | svld1_vnum(pg, &hori_deriv_buffer_[index + channel * 2], 1); | |
| 189 | |||
| 190 | 3218 | BufferVectorType vert_buff_0 = svld1(pg, &vert_deriv_buffer_[index]); | |
| 191 | 6436 | BufferVectorType vert_buff_3 = | |
| 192 | 3218 | svld1_vnum(pg, &vert_deriv_buffer_[index], 1); | |
| 193 | 6436 | BufferVectorType vert_buff_1 = | |
| 194 | 3218 | svld1(pg, &vert_deriv_buffer_[index + channel]); | |
| 195 | 6436 | BufferVectorType vert_buff_2 = | |
| 196 | 3218 | svld1(pg, &vert_deriv_buffer_[index + channel * 2]); | |
| 197 | 6436 | BufferVectorType vert_buff_4 = | |
| 198 | 3218 | svld1_vnum(pg, &vert_deriv_buffer_[index + channel], 1); | |
| 199 | 6436 | BufferVectorType vert_buff_5 = | |
| 200 | 3218 | svld1_vnum(pg, &vert_deriv_buffer_[index + channel * 2], 1); | |
| 201 | |||
| 202 | 3218 | BufferVector2Type hori_a = svcreate2(hori_buff_0, hori_buff_2); | |
| 203 | 3218 | BufferVector2Type hori_b = svcreate2(hori_buff_3, hori_buff_4); | |
| 204 | |||
| 205 | 3218 | BufferVector3Type vert_a = svcreate3(vert_buff_0, vert_buff_1, vert_buff_2); | |
| 206 | 3218 | BufferVector3Type vert_b = svcreate3(vert_buff_3, vert_buff_4, vert_buff_5); | |
| 207 | #endif | ||
| 208 | |||
| 209 | 7036 | svint16x2_t interleaved_result_a = | |
| 210 | 3518 | horizontal_compute_interleaved(pg, hori_a, vert_a); | |
| 211 | |||
| 212 | 7036 | svint16x2_t interleaved_result_b = | |
| 213 | 3518 | horizontal_compute_interleaved(pg, hori_b, vert_b); | |
| 214 | |||
| 215 | 3518 | svst2(pg, &dst_rows[index * 2], interleaved_result_a); | |
| 216 | 3518 | svst2_vnum(pg, &dst_rows[index * 2], 2, interleaved_result_b); | |
| 217 | 3518 | } | |
| 218 | |||
| 219 | 18324 | void horizontal_vector_path(svbool_t pg, Rows<DestinationType> dst_rows, | |
| 220 | ptrdiff_t index, | ||
| 221 | ptrdiff_t channel) KLEIDICV_STREAMING { | ||
| 222 | // Horizontal derivative approximation | ||
| 223 | 36648 | BufferVector2Type hori = | |
| 224 | 36648 | svcreate2(svld1(pg, &hori_deriv_buffer_[index]), | |
| 225 | 18324 | svld1(pg, &hori_deriv_buffer_[index + channel * 2])); | |
| 226 | |||
| 227 | // Vertical derivative approximation | ||
| 228 | 36648 | BufferVector3Type vert = | |
| 229 | 36648 | svcreate3(svld1(pg, &vert_deriv_buffer_[index]), | |
| 230 | 18324 | svld1(pg, &vert_deriv_buffer_[index + channel]), | |
| 231 | 18324 | svld1(pg, &vert_deriv_buffer_[index + channel * 2])); | |
| 232 | |||
| 233 | 36648 | svint16x2_t interleaved_result = | |
| 234 | 18324 | horizontal_compute_interleaved(pg, hori, vert); | |
| 235 | 18324 | svst2(pg, &dst_rows[index * 2], interleaved_result); | |
| 236 | 18324 | } | |
| 237 | |||
| 238 | 25360 | svint16x2_t horizontal_compute_interleaved( | |
| 239 | svbool_t pg, BufferVector2Type hori, | ||
| 240 | BufferVector3Type vert) KLEIDICV_STREAMING { | ||
| 241 | 25360 | BufferVectorType hori_buff_0 = svget2(hori, 0); | |
| 242 | 25360 | BufferVectorType hori_buff_2 = svget2(hori, 1); | |
| 243 | 25360 | BufferVectorType vert_buff_0 = svget3(vert, 0); | |
| 244 | 25360 | BufferVectorType vert_buff_1 = svget3(vert, 1); | |
| 245 | 25360 | BufferVectorType vert_buff_2 = svget3(vert, 2); | |
| 246 | 25360 | DestinationVectorType hori_result = svsub_x(pg, hori_buff_2, hori_buff_0); | |
| 247 | |||
| 248 | 25360 | DestinationVectorType vert_result = svadd_x(pg, vert_buff_0, vert_buff_2); | |
| 249 | 25360 | vert_result = svmul_n_s16_x(pg, vert_result, 3); | |
| 250 | 25360 | vert_result = svmad_s16_x(pg, vert_buff_1, svdup_n_s16(10), vert_result); | |
| 251 | |||
| 252 | 50720 | return svcreate2(hori_result, vert_result); | |
| 253 | 25360 | } | |
| 254 | |||
| 255 | 17472 | void process_horizontal(Rows<DestinationType> dst_rows) KLEIDICV_STREAMING { | |
| 256 | // width is decremented by 2 as the result has less columns. | ||
| 257 | 34944 | LoopUnroll2 loop{(width_ - 2) * hori_deriv_buffer_.channels(), | |
| 258 | 17472 | BufferVecTraits::num_lanes()}; | |
| 259 | 17472 | svbool_t pg_all = BufferVecTraits::svptrue(); | |
| 260 | 34944 | const ptrdiff_t channel = | |
| 261 | 17472 | static_cast<ptrdiff_t>(hori_deriv_buffer_.channels()); | |
| 262 | |||
| 263 | 20990 | loop.unroll_twice([&](ptrdiff_t index) KLEIDICV_STREAMING { | |
| 264 | 3518 | horizontal_vector_path_x2(pg_all, dst_rows, index, channel); | |
| 265 | 3518 | }); | |
| 266 | |||
| 267 | 19888 | loop.unroll_once([&](ptrdiff_t index) KLEIDICV_STREAMING { | |
| 268 | 2416 | horizontal_vector_path(pg_all, dst_rows, index, channel); | |
| 269 | 2416 | }); | |
| 270 | |||
| 271 | 33380 | loop.remaining([&](ptrdiff_t index, ptrdiff_t length) KLEIDICV_STREAMING { | |
| 272 | 15908 | svbool_t pg = BufferVecTraits::svwhilelt(index, length); | |
| 273 | 15908 | horizontal_vector_path(pg, dst_rows, index, channel); | |
| 274 | 15908 | }); | |
| 275 | 17472 | } | |
| 276 | |||
| 277 | Rows<BufferType> hori_deriv_buffer_; | ||
| 278 | Rows<BufferType> vert_deriv_buffer_; | ||
| 279 | size_t width_; | ||
| 280 | }; // end of class ScharrInterleaved | ||
| 281 | |||
| 282 | class ScharrBufferDeleter { | ||
| 283 | public: | ||
| 284 | 990 | void operator()(void *ptr) const { std::free(ptr); } | |
| 285 | }; | ||
| 286 | |||
| 287 | 1008 | static kleidicv_error_t kleidicv_scharr_interleaved_stripe_s16_u8_sc( | |
| 288 | const uint8_t *src, size_t src_stride, size_t src_width, size_t src_height, | ||
| 289 | size_t src_channels, int16_t *dst, size_t dst_stride, size_t y_begin, | ||
| 290 | size_t y_end) KLEIDICV_STREAMING { | ||
| 291 | // Does not include checks for whether the operation is implemented. | ||
| 292 | // This must be done earlier, by scharr_interleaved_is_implemented. | ||
| 293 |
4/4✓ Branch 0 taken 3 times.
✓ Branch 1 taken 1005 times.
✓ Branch 2 taken 3 times.
✓ Branch 3 taken 1005 times.
|
1008 | CHECK_POINTER_AND_STRIDE(src, src_stride, src_height); |
| 294 |
4/4✓ Branch 0 taken 6 times.
✓ Branch 1 taken 999 times.
✓ Branch 2 taken 6 times.
✓ Branch 3 taken 999 times.
|
1005 | CHECK_POINTER_AND_STRIDE(dst, dst_stride, src_height); |
| 295 |
6/6✓ Branch 0 taken 3 times.
✓ Branch 1 taken 996 times.
✓ Branch 2 taken 3 times.
✓ Branch 3 taken 993 times.
✓ Branch 4 taken 6 times.
✓ Branch 5 taken 993 times.
|
999 | CHECK_IMAGE_SIZE(src_width, src_height); |
| 296 | |||
| 297 | // Allocating more elements because in case of SVE interleaving stores are | ||
| 298 | // governed by one predicate. For example, if a predicate requires 7 uint8_t | ||
| 299 | // elements and an algorithm performs widening to 16 bits, the resulting | ||
| 300 | // interleaving store will still be governed by the same predicate, thus | ||
| 301 | // storing 8 elements. Choosing '3' to account for svst4(). | ||
| 302 | 993 | size_t buffer_stride = ((src_width * src_channels) + 3) * sizeof(int16_t); | |
| 303 | // Buffer has two rows, one for the horizontal derivative approximation, one | ||
| 304 | // for the vertical one. | ||
| 305 | 993 | size_t buffer_height = 2; | |
| 306 | // Memory is allocated with malloc to avoid its initialization. | ||
| 307 | 993 | void *allocation = std::malloc(buffer_stride * buffer_height); | |
| 308 | |||
| 309 |
2/2✓ Branch 0 taken 990 times.
✓ Branch 1 taken 3 times.
|
993 | if (!allocation) { |
| 310 | 3 | return KLEIDICV_ERROR_ALLOCATION; | |
| 311 | } | ||
| 312 | |||
| 313 | 1980 | std::unique_ptr<int16_t, ScharrBufferDeleter> buffer( | |
| 314 | 990 | reinterpret_cast<int16_t *>(allocation)); | |
| 315 | |||
| 316 | 990 | Rows<const uint8_t> src_rows{src, src_stride, src_channels}; | |
| 317 | |||
| 318 | // Result is treated as it has double the channel number compared to the | ||
| 319 | // input. | ||
| 320 | 990 | Rows<int16_t> dst_rows{dst, dst_stride, src_channels * 2}; | |
| 321 | |||
| 322 | 990 | Rows<int16_t> hori_deriv_buffer{buffer.get(), buffer_stride, src_channels}; | |
| 323 | |||
| 324 | 1980 | int16_t *vert_deriv_ptr = reinterpret_cast<int16_t *>( | |
| 325 | 990 | reinterpret_cast<uint8_t *>(buffer.get()) + buffer_stride); | |
| 326 | 990 | Rows<int16_t> vert_deriv_buffer{vert_deriv_ptr, buffer_stride, src_channels}; | |
| 327 | |||
| 328 | 1980 | ScharrInterleaved(hori_deriv_buffer, vert_deriv_buffer, src_width) | |
| 329 | 990 | .process(src_rows, dst_rows, y_begin, y_end); | |
| 330 | |||
| 331 | 990 | return KLEIDICV_OK; | |
| 332 | 1008 | } | |
| 333 | } // namespace KLEIDICV_TARGET_NAMESPACE | ||
| 334 |