1 // Copyright 2020 The libgav1 Authors
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 // http://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 #include "src/post_filter.h"
15 #include "src/utils/blocking_counter.h"
16
17 namespace libgav1 {
18
19 template <typename Pixel>
ApplyLoopRestorationForOneRow(const Pixel * src_buffer,const ptrdiff_t stride,const Plane plane,const int plane_height,const int plane_width,const int unit_y,const int unit_row,const int current_process_unit_height,const int plane_unit_size,Pixel * dst_buffer)20 void PostFilter::ApplyLoopRestorationForOneRow(
21 const Pixel* src_buffer, const ptrdiff_t stride, const Plane plane,
22 const int plane_height, const int plane_width, const int unit_y,
23 const int unit_row, const int current_process_unit_height,
24 const int plane_unit_size, Pixel* dst_buffer) {
25 const int num_horizontal_units =
26 restoration_info_->num_horizontal_units(static_cast<Plane>(plane));
27 const RestorationUnitInfo* const restoration_info =
28 restoration_info_->loop_restoration_info(static_cast<Plane>(plane),
29 unit_row * num_horizontal_units);
30 const bool in_place = DoCdef() || thread_pool_ != nullptr;
31 const Pixel* border = nullptr;
32 ptrdiff_t border_stride = 0;
33 src_buffer += unit_y * stride;
34 if (in_place) {
35 const int border_unit_y = std::max(
36 RightShiftWithCeiling(unit_y, 4 - subsampling_y_[plane]) - 4, 0);
37 border_stride = loop_restoration_border_.stride(plane) / sizeof(Pixel);
38 border =
39 reinterpret_cast<const Pixel*>(loop_restoration_border_.data(plane)) +
40 border_unit_y * border_stride;
41 }
42 int unit_column = 0;
43 int column = 0;
44 do {
45 const int current_process_unit_width =
46 std::min(plane_unit_size, plane_width - column);
47 const Pixel* src = src_buffer + column;
48 unit_column = std::min(unit_column, num_horizontal_units - 1);
49 if (restoration_info[unit_column].type == kLoopRestorationTypeNone) {
50 Pixel* dst = dst_buffer + column;
51 if (in_place) {
52 int k = current_process_unit_height;
53 do {
54 memmove(dst, src, current_process_unit_width * sizeof(Pixel));
55 src += stride;
56 dst += stride;
57 } while (--k != 0);
58 } else {
59 CopyPlane(src, stride, current_process_unit_width,
60 current_process_unit_height, dst, stride);
61 }
62 } else {
63 const Pixel* top_border = src - kRestorationVerticalBorder * stride;
64 ptrdiff_t top_border_stride = stride;
65 const Pixel* bottom_border = src + current_process_unit_height * stride;
66 ptrdiff_t bottom_border_stride = stride;
67 const bool frame_bottom_border =
68 (unit_y + current_process_unit_height >= plane_height);
69 if (in_place && (unit_y != 0 || !frame_bottom_border)) {
70 const Pixel* loop_restoration_border = border + column;
71 if (unit_y != 0) {
72 top_border = loop_restoration_border;
73 top_border_stride = border_stride;
74 loop_restoration_border += 4 * border_stride;
75 }
76 if (!frame_bottom_border) {
77 bottom_border = loop_restoration_border +
78 kRestorationVerticalBorder * border_stride;
79 bottom_border_stride = border_stride;
80 }
81 }
82 RestorationBuffer restoration_buffer;
83 const LoopRestorationType type = restoration_info[unit_column].type;
84 assert(type == kLoopRestorationTypeSgrProj ||
85 type == kLoopRestorationTypeWiener);
86 const dsp::LoopRestorationFunc restoration_func =
87 dsp_.loop_restorations[type - 2];
88 restoration_func(restoration_info[unit_column], src, stride, top_border,
89 top_border_stride, bottom_border, bottom_border_stride,
90 current_process_unit_width, current_process_unit_height,
91 &restoration_buffer, dst_buffer + column);
92 }
93 ++unit_column;
94 column += plane_unit_size;
95 } while (column < plane_width);
96 }
97
98 template <typename Pixel>
ApplyLoopRestorationForOneSuperBlockRow(const int row4x4_start,const int sb4x4)99 void PostFilter::ApplyLoopRestorationForOneSuperBlockRow(const int row4x4_start,
100 const int sb4x4) {
101 assert(row4x4_start >= 0);
102 assert(DoRestoration());
103 int plane = kPlaneY;
104 const int upscaled_width = frame_header_.upscaled_width;
105 const int height = frame_header_.height;
106 do {
107 if (loop_restoration_.type[plane] == kLoopRestorationTypeNone) {
108 continue;
109 }
110 const ptrdiff_t stride = frame_buffer_.stride(plane) / sizeof(Pixel);
111 const int unit_height_offset =
112 kRestorationUnitOffset >> subsampling_y_[plane];
113 const int plane_height = SubsampledValue(height, subsampling_y_[plane]);
114 const int plane_width =
115 SubsampledValue(upscaled_width, subsampling_x_[plane]);
116 const int plane_unit_size = 1 << loop_restoration_.unit_size_log2[plane];
117 const int plane_process_unit_height =
118 kRestorationUnitHeight >> subsampling_y_[plane];
119 int y = (row4x4_start == 0)
120 ? 0
121 : (MultiplyBy4(row4x4_start) >> subsampling_y_[plane]) -
122 unit_height_offset;
123 int expected_height = plane_process_unit_height -
124 ((row4x4_start == 0) ? unit_height_offset : 0);
125 int current_process_unit_height;
126 for (int sb_y = 0; sb_y < sb4x4;
127 sb_y += 16, y += current_process_unit_height) {
128 if (y >= plane_height) break;
129 const int unit_row = std::min(
130 (y + unit_height_offset) >> loop_restoration_.unit_size_log2[plane],
131 restoration_info_->num_vertical_units(static_cast<Plane>(plane)) - 1);
132 current_process_unit_height = std::min(expected_height, plane_height - y);
133 expected_height = plane_process_unit_height;
134 ApplyLoopRestorationForOneRow<Pixel>(
135 reinterpret_cast<Pixel*>(superres_buffer_[plane]), stride,
136 static_cast<Plane>(plane), plane_height, plane_width, y, unit_row,
137 current_process_unit_height, plane_unit_size,
138 reinterpret_cast<Pixel*>(loop_restoration_buffer_[plane]) +
139 y * stride);
140 }
141 } while (++plane < planes_);
142 }
143
ApplyLoopRestoration(const int row4x4_start,const int sb4x4)144 void PostFilter::ApplyLoopRestoration(const int row4x4_start, const int sb4x4) {
145 #if LIBGAV1_MAX_BITDEPTH >= 10
146 if (bitdepth_ >= 10) {
147 ApplyLoopRestorationForOneSuperBlockRow<uint16_t>(row4x4_start, sb4x4);
148 return;
149 }
150 #endif
151 ApplyLoopRestorationForOneSuperBlockRow<uint8_t>(row4x4_start, sb4x4);
152 }
153
ApplyLoopRestorationWorker(std::atomic<int> * row4x4_atomic)154 void PostFilter::ApplyLoopRestorationWorker(std::atomic<int>* row4x4_atomic) {
155 int row4x4;
156 // Loop Restoration operates with a lag of 8 rows (4 for chroma with
157 // subsampling) and hence we need to make sure to cover the last 8 rows of the
158 // last superblock row. So we run this loop for an extra iteration to
159 // accomplish that.
160 const int row4x4_end = frame_header_.rows4x4 + kNum4x4InLoopRestorationUnit;
161 while ((row4x4 = row4x4_atomic->fetch_add(kNum4x4InLoopRestorationUnit,
162 std::memory_order_relaxed)) <
163 row4x4_end) {
164 CopyBordersForOneSuperBlockRow(row4x4, kNum4x4InLoopRestorationUnit,
165 /*for_loop_restoration=*/true);
166 #if LIBGAV1_MAX_BITDEPTH >= 10
167 if (bitdepth_ >= 10) {
168 ApplyLoopRestorationForOneSuperBlockRow<uint16_t>(
169 row4x4, kNum4x4InLoopRestorationUnit);
170 continue;
171 }
172 #endif
173 ApplyLoopRestorationForOneSuperBlockRow<uint8_t>(
174 row4x4, kNum4x4InLoopRestorationUnit);
175 }
176 }
177
178 } // namespace libgav1
179