KleidiCV Coverage Report

Directory:	./
File:	kleidicv/include/kleidicv/filters/filter_2d_neon.h
Date:	2025-09-25 14:13:34

	Exec	Total	Coverage
Lines:	82	82	100.0%
Functions:	261	273	95.6%
Branches:	56	56	100.0%

  
      Line
      Branch
      Exec
      Source
    
      // SPDX-FileCopyrightText: 2025 Arm Limited and/or its affiliates <open-source-office@arm.com>
    
      //
    
      // SPDX-License-Identifier: Apache-2.0
    
      #ifndef KLEIDICV_FILTER_2D_NEON_H
    
      #define KLEIDICV_FILTER_2D_NEON_H
    
      #include "filter_2d_window_loader_3x3.h"
    
      #include "filter_2d_window_loader_5x5.h"
    
      #include "filter_2d_window_loader_7x7.h"
    
      #include "kleidicv/neon.h"
    
      #include "process_filter_2d.h"
    
      namespace KLEIDICV_TARGET_NAMESPACE {
    
      template <typename InnerFilterType, size_t KSize, typename WindowLoaderType>
    
      class Filter2d {
    
       public:
    
        using SourceType = typename InnerFilterType::SourceType;
    
        using DestinationType = typename InnerFilterType::DestinationType;
    
        using SourceVecTraits = typename neon::VecTraits<SourceType>;
    
        using DestinationVecTraits = typename neon::VecTraits<DestinationType>;
    
        using SourceVectorType = typename SourceVecTraits::VectorType;
    
        using DestinationVectorType = typename DestinationVecTraits::VectorType;
    
        using BorderType = FixedBorderType;
    
        static constexpr size_t kMargin = KSize / 2;
    
        using BorderInfoType =
    
            typename ::KLEIDICV_TARGET_NAMESPACE::FixedBorderInfo<SourceType, KSize>;
    
        using BorderOffsets = typename BorderInfoType::Offsets;
    
      1610
        explicit Filter2d(InnerFilterType filter) : filter_{filter} {}
    
      8676
        void process_pixels_without_horizontal_borders(
    
            size_t width, Rows<const SourceType> src_rows,
    
            Rows<DestinationType> dst_rows, BorderOffsets window_row_offsets,
    
            BorderOffsets window_col_offsets) const {
    
      17352
          LoopUnroll2<TryToAvoidTailLoop> loop{width * src_rows.channels(),
    
      8676
                                               SourceVecTraits::num_lanes()};
    
      34632
          loop.unroll_once([&](size_t index) {
    
      25956
            SourceVectorType src[KSize][KSize];
    
      25956
            DestinationVectorType dst_vec;
    
      11224924
            auto KernelWindow = [&](size_t row, size_t col) -> SourceVectorType& {
    
      11198968
              return src[row][col];
    
            };
    
      983368
            auto load_array_element = [](const SourceType& x) { return vld1q(&x); };
    
      51912
            WindowLoaderType::load_window(KernelWindow, load_array_element, src_rows,
    
      25956
                                          window_row_offsets, window_col_offsets,
    
      25956
                                          index);
    
      25956
            filter_.vector_path(KernelWindow, dst_vec);
    
      25956
            vst1q(&dst_rows[index], dst_vec);
    
      25956
          });
    
      13556
          loop.tail([&](size_t index) {
    
      4880
            process_one_element_with_horizontal_borders(
    
      4880
                src_rows, dst_rows, window_row_offsets, window_col_offsets, index);
    
      4880
          });
    
      8676
        }
    
      2272
        void process_pixels_of_dual_rows_without_horizontal_borders(
    
            size_t width, Rows<const SourceType> src_rows,
    
            Rows<DestinationType> dst_rows, BorderOffsets window_row_offsets_0,
    
            BorderOffsets window_row_offsets_1,
    
            BorderOffsets window_col_offsets) const {
    
      4544
          LoopUnroll2<TryToAvoidTailLoop> loop{width * src_rows.channels(),
    
      2272
                                               SourceVecTraits::num_lanes()};
    
      5750
          loop.unroll_once([&](size_t index) {
    
      3478
            SourceVectorType src[KSize + 1][KSize];
    
      3478
            DestinationVectorType dst_vec_0;
    
      3478
            DestinationVectorType dst_vec_1;
    
      198246
            auto KernelWindow = [&](size_t row, size_t col) -> SourceVectorType& {
    
      194768
              return src[row][col];
    
            };
    
      45214
            auto load_array_element = [](const SourceType& x) { return vld1q(&x); };
    
      3478
            WindowLoaderType::load_window_to_handle_dual_rows(
    
      3478
                KernelWindow, load_array_element, src_rows, window_row_offsets_0,
    
      3478
                window_row_offsets_1, window_col_offsets, index);
    
      3478
            filter_.vector_path_for_dual_row_handling(KernelWindow, dst_vec_0,
    
                                                      dst_vec_1);
    
      3478
            vst1q(&dst_rows.at(0, 0)[index], dst_vec_0);
    
      3478
            vst1q(&dst_rows.at(1, 0)[index], dst_vec_1);
    
      3478
          });
    
      4440
          loop.tail([&](size_t index) {
    
      2168
            process_two_element_vertically_with_or_without_horizontal_borders(
    
      2168
                src_rows, dst_rows, window_row_offsets_0, window_row_offsets_1,
    
      2168
                window_col_offsets, index);
    
      2168
          });
    
      2272
        }
    
      42704
        void process_one_pixel_with_horizontal_borders(
    
            Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
    
            BorderOffsets window_row_offsets,
    
            BorderOffsets window_col_offsets) const KLEIDICV_STREAMING {
    
        42/42✓ Branch 0 taken 64 times.
✓ Branch 1 taken 160 times.
✓ Branch 2 taken 1232 times.
✓ Branch 3 taken 3080 times.
✓ Branch 4 taken 2496 times.
✓ Branch 5 taken 6240 times.
✓ Branch 6 taken 72 times.
✓ Branch 7 taken 168 times.
✓ Branch 8 taken 3272 times.
✓ Branch 9 taken 5120 times.
✓ Branch 10 taken 4488 times.
✓ Branch 11 taken 8232 times.
✓ Branch 12 taken 72 times.
✓ Branch 13 taken 168 times.
✓ Branch 14 taken 3272 times.
✓ Branch 15 taken 5120 times.
✓ Branch 16 taken 4488 times.
✓ Branch 17 taken 8232 times.
✓ Branch 18 taken 72 times.
✓ Branch 19 taken 168 times.
✓ Branch 20 taken 3272 times.
✓ Branch 21 taken 5120 times.
✓ Branch 22 taken 4488 times.
✓ Branch 23 taken 8232 times.
✓ Branch 24 taken 64 times.
✓ Branch 25 taken 160 times.
✓ Branch 26 taken 1232 times.
✓ Branch 27 taken 3080 times.
✓ Branch 28 taken 2496 times.
✓ Branch 29 taken 6240 times.
✓ Branch 30 taken 64 times.
✓ Branch 31 taken 160 times.
✓ Branch 32 taken 1232 times.
✓ Branch 33 taken 3080 times.
✓ Branch 34 taken 2496 times.
✓ Branch 35 taken 6240 times.
✓ Branch 36 taken 72 times.
✓ Branch 37 taken 168 times.
✓ Branch 38 taken 3272 times.
✓ Branch 39 taken 5120 times.
✓ Branch 40 taken 4488 times.
✓ Branch 41 taken 8232 times.

      125224
          for (size_t index = 0; index < src_rows.channels(); ++index) {
    
      82520
            disable_loop_vectorization();
    
      82520
            process_one_element_with_horizontal_borders(
    
      82520
                src_rows, dst_rows, window_row_offsets, window_col_offsets, index);
    
      82520
          }
    
      42704
        }
    
        // Processes two vertically adjacent pixels in a single column
    
      4544
        void process_two_pixels_with_horizontal_borders(
    
            Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
    
            BorderOffsets window_row_offsets_0, BorderOffsets window_row_offsets_1,
    
            BorderOffsets window_col_offsets) const {
    
        14/14✓ Branch 0 taken 128 times.
✓ Branch 1 taken 320 times.
✓ Branch 2 taken 1040 times.
✓ Branch 3 taken 1232 times.
✓ Branch 4 taken 1040 times.
✓ Branch 5 taken 1232 times.
✓ Branch 6 taken 1040 times.
✓ Branch 7 taken 1232 times.
✓ Branch 8 taken 128 times.
✓ Branch 9 taken 320 times.
✓ Branch 10 taken 128 times.
✓ Branch 11 taken 320 times.
✓ Branch 12 taken 1040 times.
✓ Branch 13 taken 1232 times.

      10432
          for (size_t index = 0; index < src_rows.channels(); ++index) {
    
      5888
            disable_loop_vectorization();
    
      5888
            process_two_element_vertically_with_or_without_horizontal_borders(
    
      5888
                src_rows, dst_rows, window_row_offsets_0, window_row_offsets_1,
    
      5888
                window_col_offsets, index);
    
      5888
          }
    
      4544
        }
    
       private:
    
      87400
        void process_one_element_with_horizontal_borders(
    
            Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
    
            BorderOffsets window_row_offsets, BorderOffsets window_col_offsets,
    
            size_t index) const KLEIDICV_STREAMING {
    
      87400
          SourceType src[KSize][KSize];
    
      87400
          auto KernelWindow =
    
      41533720
              [&](size_t row, size_t col)
    
      41446320
                  KLEIDICV_STREAMING -> SourceType& { return src[row][col]; };
    
      3552016
          auto load_array_element = [&](const SourceType& x)
    
      3464616
                                        KLEIDICV_STREAMING { return x; };
    
      174800
          WindowLoaderType::load_window(KernelWindow, load_array_element, src_rows,
    
      87400
                                        window_row_offsets, window_col_offsets,
    
      87400
                                        index);
    
      87400
          filter_.scalar_path(KernelWindow, dst_rows[index]);
    
      87400
        }
    
      8056
        void process_two_element_vertically_with_or_without_horizontal_borders(
    
            Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
    
            BorderOffsets window_row_offsets_0, BorderOffsets window_row_offsets_1,
    
            BorderOffsets window_col_offsets, size_t index) const {
    
      8056
          SourceType src[KSize + 1][KSize];
    
      459192
          auto KernelWindow = [&](size_t row, size_t col) -> SourceType& {
    
      451136
            return src[row][col];
    
          };
    
      104728
          auto load_array_element = [](const SourceType& x) { return x; };
    
      8056
          WindowLoaderType::load_window_to_handle_dual_rows(
    
      8056
              KernelWindow, load_array_element, src_rows, window_row_offsets_0,
    
      8056
              window_row_offsets_1, window_col_offsets, index);
    
      8056
          filter_.scalar_path_for_dual_row_handling(
    
      8056
              KernelWindow, dst_rows.at(0, 0)[index], dst_rows.at(1, 0)[index]);
    
      8056
        }
    
        InnerFilterType filter_;
    
      };
    
      template <typename InnerFilterType>
    
      using Filter2D3x3 =
    
          Filter2d<InnerFilterType, 3UL,
    
                   Filter2dWindowLoader3x3<typename InnerFilterType::SourceType>>;
    
      template <typename InnerFilterType>
    
      using Filter2D5x5 =
    
          Filter2d<InnerFilterType, 5UL,
    
                   Filter2dWindowLoader5x5<typename InnerFilterType::SourceType>>;
    
      template <typename InnerFilterType>
    
      using Filter2D7x7 =
    
          Filter2d<InnerFilterType, 7UL,
    
                   Filter2dWindowLoader7x7<typename InnerFilterType::SourceType>>;
    
      }  // namespace KLEIDICV_TARGET_NAMESPACE
    
      #endif  // KLEIDICV_FILTER_2D_NEON_H

Line	Branch	Exec	Source
1			// SPDX-FileCopyrightText: 2025 Arm Limited and/or its affiliates <open-source-office@arm.com>
2			//
3			// SPDX-License-Identifier: Apache-2.0
4
5			#ifndef KLEIDICV_FILTER_2D_NEON_H
6			#define KLEIDICV_FILTER_2D_NEON_H
7
8			#include "filter_2d_window_loader_3x3.h"
9			#include "filter_2d_window_loader_5x5.h"
10			#include "filter_2d_window_loader_7x7.h"
11			#include "kleidicv/neon.h"
12			#include "process_filter_2d.h"
13
14			namespace KLEIDICV_TARGET_NAMESPACE {
15
16			template <typename InnerFilterType, size_t KSize, typename WindowLoaderType>
17			class Filter2d {
18			public:
19			using SourceType = typename InnerFilterType::SourceType;
20			using DestinationType = typename InnerFilterType::DestinationType;
21			using SourceVecTraits = typename neon::VecTraits<SourceType>;
22			using DestinationVecTraits = typename neon::VecTraits<DestinationType>;
23			using SourceVectorType = typename SourceVecTraits::VectorType;
24			using DestinationVectorType = typename DestinationVecTraits::VectorType;
25			using BorderType = FixedBorderType;
26			static constexpr size_t kMargin = KSize / 2;
27			using BorderInfoType =
28			typename ::KLEIDICV_TARGET_NAMESPACE::FixedBorderInfo<SourceType, KSize>;
29			using BorderOffsets = typename BorderInfoType::Offsets;
30
31		1610	explicit Filter2d(InnerFilterType filter) : filter_{filter} {}
32
33		8676	void process_pixels_without_horizontal_borders(
34			size_t width, Rows<const SourceType> src_rows,
35			Rows<DestinationType> dst_rows, BorderOffsets window_row_offsets,
36			BorderOffsets window_col_offsets) const {
37		17352	LoopUnroll2<TryToAvoidTailLoop> loop{width * src_rows.channels(),
38		8676	SourceVecTraits::num_lanes()};
39
40		34632	loop.unroll_once([&](size_t index) {
41		25956	SourceVectorType src[KSize][KSize];
42		25956	DestinationVectorType dst_vec;
43
44		11224924	auto KernelWindow = [&](size_t row, size_t col) -> SourceVectorType& {
45		11198968	return src[row][col];
46			};
47
48		983368	auto load_array_element = [](const SourceType& x) { return vld1q(&x); };
49		51912	WindowLoaderType::load_window(KernelWindow, load_array_element, src_rows,
50		25956	window_row_offsets, window_col_offsets,
51		25956	index);
52
53		25956	filter_.vector_path(KernelWindow, dst_vec);
54		25956	vst1q(&dst_rows[index], dst_vec);
55		25956	});
56
57		13556	loop.tail([&](size_t index) {
58		4880	process_one_element_with_horizontal_borders(
59		4880	src_rows, dst_rows, window_row_offsets, window_col_offsets, index);
60		4880	});
61		8676	}
62
63		2272	void process_pixels_of_dual_rows_without_horizontal_borders(
64			size_t width, Rows<const SourceType> src_rows,
65			Rows<DestinationType> dst_rows, BorderOffsets window_row_offsets_0,
66			BorderOffsets window_row_offsets_1,
67			BorderOffsets window_col_offsets) const {
68		4544	LoopUnroll2<TryToAvoidTailLoop> loop{width * src_rows.channels(),
69		2272	SourceVecTraits::num_lanes()};
70
71		5750	loop.unroll_once([&](size_t index) {
72		3478	SourceVectorType src[KSize + 1][KSize];
73		3478	DestinationVectorType dst_vec_0;
74		3478	DestinationVectorType dst_vec_1;
75		198246	auto KernelWindow = [&](size_t row, size_t col) -> SourceVectorType& {
76		194768	return src[row][col];
77			};
78
79		45214	auto load_array_element = [](const SourceType& x) { return vld1q(&x); };
80		3478	WindowLoaderType::load_window_to_handle_dual_rows(
81		3478	KernelWindow, load_array_element, src_rows, window_row_offsets_0,
82		3478	window_row_offsets_1, window_col_offsets, index);
83
84		3478	filter_.vector_path_for_dual_row_handling(KernelWindow, dst_vec_0,
85			dst_vec_1);
86		3478	vst1q(&dst_rows.at(0, 0)[index], dst_vec_0);
87		3478	vst1q(&dst_rows.at(1, 0)[index], dst_vec_1);
88		3478	});
89
90		4440	loop.tail([&](size_t index) {
91		2168	process_two_element_vertically_with_or_without_horizontal_borders(
92		2168	src_rows, dst_rows, window_row_offsets_0, window_row_offsets_1,
93		2168	window_col_offsets, index);
94		2168	});
95		2272	}
96
97		42704	void process_one_pixel_with_horizontal_borders(
98			Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
99			BorderOffsets window_row_offsets,
100			BorderOffsets window_col_offsets) const KLEIDICV_STREAMING {
101	42/42 ✓ Branch 0 taken 64 times. ✓ Branch 1 taken 160 times. ✓ Branch 2 taken 1232 times. ✓ Branch 3 taken 3080 times. ✓ Branch 4 taken 2496 times. ✓ Branch 5 taken 6240 times. ✓ Branch 6 taken 72 times. ✓ Branch 7 taken 168 times. ✓ Branch 8 taken 3272 times. ✓ Branch 9 taken 5120 times. ✓ Branch 10 taken 4488 times. ✓ Branch 11 taken 8232 times. ✓ Branch 12 taken 72 times. ✓ Branch 13 taken 168 times. ✓ Branch 14 taken 3272 times. ✓ Branch 15 taken 5120 times. ✓ Branch 16 taken 4488 times. ✓ Branch 17 taken 8232 times. ✓ Branch 18 taken 72 times. ✓ Branch 19 taken 168 times. ✓ Branch 20 taken 3272 times. ✓ Branch 21 taken 5120 times. ✓ Branch 22 taken 4488 times. ✓ Branch 23 taken 8232 times. ✓ Branch 24 taken 64 times. ✓ Branch 25 taken 160 times. ✓ Branch 26 taken 1232 times. ✓ Branch 27 taken 3080 times. ✓ Branch 28 taken 2496 times. ✓ Branch 29 taken 6240 times. ✓ Branch 30 taken 64 times. ✓ Branch 31 taken 160 times. ✓ Branch 32 taken 1232 times. ✓ Branch 33 taken 3080 times. ✓ Branch 34 taken 2496 times. ✓ Branch 35 taken 6240 times. ✓ Branch 36 taken 72 times. ✓ Branch 37 taken 168 times. ✓ Branch 38 taken 3272 times. ✓ Branch 39 taken 5120 times. ✓ Branch 40 taken 4488 times. ✓ Branch 41 taken 8232 times.	125224	for (size_t index = 0; index < src_rows.channels(); ++index) {
102		82520	disable_loop_vectorization();
103		82520	process_one_element_with_horizontal_borders(
104		82520	src_rows, dst_rows, window_row_offsets, window_col_offsets, index);
105		82520	}
106		42704	}
107
108			// Processes two vertically adjacent pixels in a single column
109		4544	void process_two_pixels_with_horizontal_borders(
110			Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
111			BorderOffsets window_row_offsets_0, BorderOffsets window_row_offsets_1,
112			BorderOffsets window_col_offsets) const {
113	14/14 ✓ Branch 0 taken 128 times. ✓ Branch 1 taken 320 times. ✓ Branch 2 taken 1040 times. ✓ Branch 3 taken 1232 times. ✓ Branch 4 taken 1040 times. ✓ Branch 5 taken 1232 times. ✓ Branch 6 taken 1040 times. ✓ Branch 7 taken 1232 times. ✓ Branch 8 taken 128 times. ✓ Branch 9 taken 320 times. ✓ Branch 10 taken 128 times. ✓ Branch 11 taken 320 times. ✓ Branch 12 taken 1040 times. ✓ Branch 13 taken 1232 times.	10432	for (size_t index = 0; index < src_rows.channels(); ++index) {
114		5888	disable_loop_vectorization();
115		5888	process_two_element_vertically_with_or_without_horizontal_borders(
116		5888	src_rows, dst_rows, window_row_offsets_0, window_row_offsets_1,
117		5888	window_col_offsets, index);
118		5888	}
119		4544	}
120
121			private:
122		87400	void process_one_element_with_horizontal_borders(
123			Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
124			BorderOffsets window_row_offsets, BorderOffsets window_col_offsets,
125			size_t index) const KLEIDICV_STREAMING {
126		87400	SourceType src[KSize][KSize];
127
128		87400	auto KernelWindow =
129		41533720	[&](size_t row, size_t col)
130		41446320	KLEIDICV_STREAMING -> SourceType& { return src[row][col]; };
131
132		3552016	auto load_array_element = [&](const SourceType& x)
133		3464616	KLEIDICV_STREAMING { return x; };
134
135		174800	WindowLoaderType::load_window(KernelWindow, load_array_element, src_rows,
136		87400	window_row_offsets, window_col_offsets,
137		87400	index);
138
139		87400	filter_.scalar_path(KernelWindow, dst_rows[index]);
140		87400	}
141
142		8056	void process_two_element_vertically_with_or_without_horizontal_borders(
143			Rows<const SourceType> src_rows, Rows<DestinationType> dst_rows,
144			BorderOffsets window_row_offsets_0, BorderOffsets window_row_offsets_1,
145			BorderOffsets window_col_offsets, size_t index) const {
146		8056	SourceType src[KSize + 1][KSize];
147		459192	auto KernelWindow = [&](size_t row, size_t col) -> SourceType& {
148		451136	return src[row][col];
149			};
150		104728	auto load_array_element = [](const SourceType& x) { return x; };
151		8056	WindowLoaderType::load_window_to_handle_dual_rows(
152		8056	KernelWindow, load_array_element, src_rows, window_row_offsets_0,
153		8056	window_row_offsets_1, window_col_offsets, index);
154
155		8056	filter_.scalar_path_for_dual_row_handling(
156		8056	KernelWindow, dst_rows.at(0, 0)[index], dst_rows.at(1, 0)[index]);
157		8056	}
158
159			InnerFilterType filter_;
160			};
161
162			template <typename InnerFilterType>
163			using Filter2D3x3 =
164			Filter2d<InnerFilterType, 3UL,
165			Filter2dWindowLoader3x3<typename InnerFilterType::SourceType>>;
166
167			template <typename InnerFilterType>
168			using Filter2D5x5 =
169			Filter2d<InnerFilterType, 5UL,
170			Filter2dWindowLoader5x5<typename InnerFilterType::SourceType>>;
171
172			template <typename InnerFilterType>
173			using Filter2D7x7 =
174			Filter2d<InnerFilterType, 7UL,
175			Filter2dWindowLoader7x7<typename InnerFilterType::SourceType>>;
176
177			} // namespace KLEIDICV_TARGET_NAMESPACE
178
179			#endif // KLEIDICV_FILTER_2D_NEON_H
180