KleidiCV Coverage Report

Directory: ./
File: kleidicv/src/filters/sobel_sc.h
Date: 2025-09-25 14:13:34
Exec Total Coverage
Lines: 68 68 100.0%
Functions: 16 16 100.0%
Branches: 36 36 100.0%
Line Branch Exec Source
1 // SPDX-FileCopyrightText: 2023 - 2025 Arm Limited and/or its affiliates <open-source-office@arm.com>
2 //
3 // SPDX-License-Identifier: Apache-2.0
4
5 #ifndef KLEIDICV_SOBEL_SC_H
6 #define KLEIDICV_SOBEL_SC_H
7
8 #include "kleidicv/filters/separable_filter_3x3_sc.h"
9 #include "kleidicv/filters/sobel.h"
10 #include "kleidicv/kleidicv.h"
11 #include "kleidicv/sve2.h"
12 #include "kleidicv/workspace/separable.h"
13
14 namespace KLEIDICV_TARGET_NAMESPACE {
15
16 // Template for 3x3 Sobel filters which calculate horizontal derivative
17 // approximations, often denoted as Gx.
18 //
19 // [ -1, 0, 1 ] [ 1 ]
20 // F = [ -2, 0, 2 ] = [ 2 ] * [ -1, 0, 1 ]
21 // [ -1, 0, 1 ] [ 1 ]
22 template <typename T>
23 class HorizontalSobel3x3;
24
25 // 3x3 Sobel filter for uint8_t types which calculates horizontal derivative
26 // approximations, often denoted as Gx.
27 template <>
28 class HorizontalSobel3x3<uint8_t> {
29 public:
30 using SourceType = uint8_t;
31 using BufferType = int16_t;
32 using DestinationType = int16_t;
33
34 // Applies vertical filtering vector using SIMD operations.
35 //
36 // DST = [ SRC0, SRC1, SRC2 ] * [ 1, 2, 1 ]T
37 5072 void vertical_vector_path(svbool_t pg,
38 std::reference_wrapper<svuint8_t> src[3],
39 BufferType *dst) const KLEIDICV_STREAMING {
40 5072 svuint16_t acc_u16_b = svaddlb(src[0], src[2]);
41 5072 svuint16_t acc_u16_t = svaddlt(src[0], src[2]);
42 5072 acc_u16_b = svmlalb(acc_u16_b, src[1], svdup_n_u8(2));
43 5072 acc_u16_t = svmlalt(acc_u16_t, src[1], svdup_n_u8(2));
44
45 10144 svint16x2_t interleaved =
46 5072 svcreate2(svreinterpret_s16(acc_u16_b), svreinterpret_s16(acc_u16_t));
47 5072 svst2(pg, &dst[0], interleaved);
48 5072 }
49
50 // Applies horizontal filtering vector using SIMD operations.
51 //
52 // DST = [ SRC0, SRC1, SRC2 ] * [ -1, 0, 1 ]T
53 4448 void horizontal_vector_path(svbool_t pg,
54 std::reference_wrapper<svint16_t> src[3],
55 DestinationType *dst) const KLEIDICV_STREAMING {
56 4448 svst1(pg, &dst[0], svsub_x(pg, src[2], src[0]));
57 4448 }
58
59 // Applies horizontal filtering vector using scalar operations.
60 //
61 // DST = [ SRC0, SRC1, SRC2 ] * [ -1, 0, 1 ]T
62 15296 void horizontal_scalar_path(const BufferType src[3],
63 DestinationType *dst) const KLEIDICV_STREAMING {
64 // Explicitly narrow. Overflow is permitted.
65 15296 dst[0] = static_cast<DestinationType>(src[2] - src[0]);
66 15296 }
67 }; // end of class HorizontalSobel3x3<uint8_t>
68
69 // Template for 3x3 Sobel filters which calculate vertical derivative
70 // approximations, often denoted as Gy.
71 //
72 // [ -1, -2, 1 ] [ -1 ]
73 // F = [ 0, 0, 0 ] = [ 0 ] * [ 1, 2, 1 ]
74 // [ 1, 2, 1 ] [ 1 ]
75 template <typename T>
76 class VerticalSobel3x3;
77
78 // 3x3 Sobel filter for uint8_t types which calculates vertical derivative
79 // approximations, often denoted as Gy.
80 template <>
81 class VerticalSobel3x3<uint8_t> {
82 public:
83 using SourceType = uint8_t;
84 using BufferType = int16_t;
85 using DestinationType = int16_t;
86
87 // Applies vertical filtering vector using SIMD operations.
88 //
89 // DST = [ SRC0, SRC1, SRC2 ] * [ -1, 0, 1 ]T
90 5072 void vertical_vector_path(svbool_t pg,
91 std::reference_wrapper<svuint8_t> src[3],
92 BufferType *dst) const KLEIDICV_STREAMING {
93 5072 svuint16_t acc_u16_b = svsublb(src[2], src[0]);
94 5072 svuint16_t acc_u16_t = svsublt(src[2], src[0]);
95
96 10144 svint16x2_t interleaved =
97 5072 svcreate2(svreinterpret_s16(acc_u16_b), svreinterpret_s16(acc_u16_t));
98 5072 svst2(pg, &dst[0], interleaved);
99 5072 }
100
101 // Applies horizontal filtering vector using SIMD operations.
102 //
103 // DST = [ SRC0, SRC1, SRC2 ] * [ 1, 2, 1 ]T
104 4448 void horizontal_vector_path(svbool_t pg,
105 std::reference_wrapper<svint16_t> src[3],
106 DestinationType *dst) const KLEIDICV_STREAMING {
107 4448 svint16_t acc = svadd_x(pg, src[0], src[2]);
108 4448 acc = svmad_s16_x(pg, src[1], svdup_n_s16(2), acc);
109 4448 svst1(pg, &dst[0], acc);
110 4448 }
111
112 // Applies horizontal filtering vector using scalar operations.
113 //
114 // DST = [ SRC0, SRC1, SRC2 ] * [ 1, 2, 1 ]T
115 15296 void horizontal_scalar_path(const BufferType src[3],
116 DestinationType *dst) const KLEIDICV_STREAMING {
117 // Explicitly narrow. Overflow is permitted.
118 15296 dst[0] = static_cast<DestinationType>(src[0] + 2 * src[1] + src[2]);
119 15296 }
120 }; // end of class VerticalSobel3x3<uint8_t>
121
122 KLEIDICV_TARGET_FN_ATTRS
123 180 static kleidicv_error_t sobel_3x3_horizontal_stripe_s16_u8_sc(
124 const uint8_t *src, size_t src_stride, int16_t *dst, size_t dst_stride,
125 size_t width, size_t height, size_t y_begin, size_t y_end,
126 size_t channels) KLEIDICV_STREAMING {
127
4/4
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 178 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 178 times.
 180 CHECK_POINTER_AND_STRIDE(src, src_stride, height);
128
4/4
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 174 times.
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 174 times.
 178 CHECK_POINTER_AND_STRIDE(dst, dst_stride, height);
129
6/6
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 172 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 170 times.
✓ Branch 4 taken 4 times.
✓ Branch 5 taken 170 times.
 174 CHECK_IMAGE_SIZE(width, height);
130
131
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 168 times.
 170 if (channels > KLEIDICV_MAXIMUM_CHANNEL_COUNT) {
132 2 return KLEIDICV_ERROR_NOT_IMPLEMENTED;
133 }
134
135 168 Rectangle rect{width, height};
136 168 Rows<const uint8_t> src_rows{src, src_stride, channels};
137 168 Rows<int16_t> dst_rows{dst, dst_stride, channels};
138
139 168 auto workspace =
140 168 SeparableFilterWorkspace::create(rect, channels, sizeof(int16_t));
141
2/2
✓ Branch 0 taken 166 times.
✓ Branch 1 taken 2 times.
 168 if (!workspace) {
142 2 return KLEIDICV_ERROR_ALLOCATION;
143 }
144
145 166 HorizontalSobel3x3<uint8_t> horizontal_sobel;
146 166 SeparableFilter3x3<HorizontalSobel3x3<uint8_t>> filter{horizontal_sobel};
147 332 workspace->process(rect, y_begin, y_end, src_rows, dst_rows, channels,
148 166 FixedBorderType::REPLICATE, filter);
149 166 return KLEIDICV_OK;
150 180 }
151
152 KLEIDICV_TARGET_FN_ATTRS
153 180 static kleidicv_error_t sobel_3x3_vertical_stripe_s16_u8_sc(
154 const uint8_t *src, size_t src_stride, int16_t *dst, size_t dst_stride,
155 size_t width, size_t height, size_t y_begin, size_t y_end,
156 size_t channels) KLEIDICV_STREAMING {
157
4/4
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 178 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 178 times.
 180 CHECK_POINTER_AND_STRIDE(src, src_stride, height);
158
4/4
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 174 times.
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 174 times.
 178 CHECK_POINTER_AND_STRIDE(dst, dst_stride, height);
159
6/6
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 172 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 170 times.
✓ Branch 4 taken 4 times.
✓ Branch 5 taken 170 times.
 174 CHECK_IMAGE_SIZE(width, height);
160
161
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 168 times.
 170 if (channels > KLEIDICV_MAXIMUM_CHANNEL_COUNT) {
162 2 return KLEIDICV_ERROR_NOT_IMPLEMENTED;
163 }
164
165 168 Rectangle rect{width, height};
166 168 Rows<const uint8_t> src_rows{src, src_stride, channels};
167 168 Rows<int16_t> dst_rows{dst, dst_stride, channels};
168
169 168 auto workspace =
170 168 SeparableFilterWorkspace::create(rect, channels, sizeof(int16_t));
171
2/2
✓ Branch 0 taken 166 times.
✓ Branch 1 taken 2 times.
 168 if (!workspace) {
172 2 return KLEIDICV_ERROR_ALLOCATION;
173 }
174
175 166 VerticalSobel3x3<uint8_t> vertical_sobel;
176 166 SeparableFilter3x3<VerticalSobel3x3<uint8_t>> filter{vertical_sobel};
177 332 workspace->process(rect, y_begin, y_end, src_rows, dst_rows, channels,
178 166 FixedBorderType::REPLICATE, filter);
179 166 return KLEIDICV_OK;
180 180 }
181
182 } // namespace KLEIDICV_TARGET_NAMESPACE
183
184 #endif // KLEIDICV_SOBEL_SC_H
185