1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
|
// Copyright 2023 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifdef UNSAFE_BUFFERS_BUILD
// TODO(crbug.com/390223051): Remove C-library calls to fix the errors.
#pragma allow_unsafe_libc_calls
#endif
#include "media/gpu/v4l2/mt21/mt21_decompressor.h"
#include <sched.h>
#include <stdlib.h>
#include "base/bits.h"
#include "base/memory/scoped_refptr.h"
#include "media/gpu/v4l2/mt21/mt21_util.h"
#include "third_party/libyuv/include/libyuv/planar_functions.h"
namespace media {
namespace {
template <class T>
void MT21ToMM21(const uint8_t* src,
const uint8_t* footer,
uint8_t* dest,
size_t start_offset,
size_t width,
size_t height,
const GolombRiceTableEntry* symbol_cache) {
std::vector<T> subblock_bins[2];
uint8_t scratch[kMT21ScratchMemorySize] __attribute__((aligned(16)));
for (size_t block_offset = 0; block_offset < width * height;
block_offset += kMT21BlockSize) {
BinSubblocks<T>(src, footer, dest, block_offset + start_offset,
subblock_bins);
}
// Handle high-entropy passthrough subblocks.
for (T& subblock : subblock_bins[1]) {
memcpy(subblock.dest, subblock.src, subblock.len);
}
// Vector decompress as many blocks as possible.
size_t i = 0;
for (; i + kNumOutputLanes - 1 < subblock_bins[0].size();
i += kNumOutputLanes) {
VectorDecompressSubblockHelper<T>(subblock_bins[0], i, scratch);
}
// Scalar decompress the remainder.
for (; i < subblock_bins[0].size(); i++) {
DecompressSubblockHelper<T>(subblock_bins[0][i], symbol_cache);
}
}
void DecompressAndDetile(const MT21DecompressionJob& job,
uint8_t* pivot,
const GolombRiceTableEntry* symbol_cache) {
if (job.is_chroma) {
MT21ToMM21<MT21UVSubblock>(job.src, job.footer, pivot, job.offset,
job.width, job.height, symbol_cache);
} else {
MT21ToMM21<MT21YSubblock>(job.src, job.footer, pivot, job.offset, job.width,
job.height, symbol_cache);
}
libyuv::DetilePlane(pivot + job.offset, job.width, job.dest + job.offset,
job.width, job.width, job.height,
job.is_chroma ? kMT21TileHeight / 2 : kMT21TileHeight);
}
// MT8173 has 2 Cortex A72s and 2 Cortex A53s
constexpr size_t kNumLittleThreads = 2;
constexpr size_t kNumBigThreads = 2;
void MT21WorkerEntry(cpu_set_t mask,
std::atomic_bool& should_shutdown,
const GolombRiceTableEntry* symbol_cache,
uint8_t* pivot,
scoped_refptr<MT21DecompressionJob> job) {
sched_setaffinity(0, sizeof(cpu_set_t), &mask);
while (true) {
job->wakeup_event.Wait();
if (should_shutdown) {
break;
}
DecompressAndDetile(*job, pivot, symbol_cache);
job->done_event.Signal();
}
}
} // namespace
MT21DecompressionJob::MT21DecompressionJob(const uint8_t* src,
const uint8_t* footer,
size_t offset,
uint8_t* dest,
size_t width,
size_t height,
bool is_chroma)
: wakeup_event(base::WaitableEvent::ResetPolicy::AUTOMATIC,
base::WaitableEvent::InitialState::NOT_SIGNALED),
done_event(base::WaitableEvent::ResetPolicy::AUTOMATIC,
base::WaitableEvent::InitialState::NOT_SIGNALED) {
this->src = src;
this->footer = footer;
this->offset = offset;
this->dest = dest;
this->width = width;
this->height = height;
this->is_chroma = is_chroma;
}
MT21Decompressor::MT21Decompressor(gfx::Size resolution) {
symbol_cache_ = new GolombRiceTableEntry[kGolombRiceCacheSize];
PopulateGolombRiceCache(symbol_cache_);
aligned_resolution_ =
gfx::Size(base::bits::AlignUp(static_cast<size_t>(resolution.width()),
kMT21TileWidth),
base::bits::AlignUp(static_cast<size_t>(resolution.height()),
kMT21TileHeight));
// Big cores are CPUs 2 and 3, while the little cores are 0 and 1.
cpu_set_t mask;
CPU_ZERO(&mask);
for (size_t i = kNumLittleThreads; i < kNumLittleThreads + kNumBigThreads;
i++) {
CPU_SET(i, &mask);
}
big_core_pivot_ =
static_cast<uint8_t*>(aligned_alloc(16, aligned_resolution_.GetArea()));
for (size_t i = 0; i < kNumBigThreads; i++) {
scoped_refptr<MT21DecompressionJob> job =
base::MakeRefCounted<MT21DecompressionJob>(nullptr, nullptr, 0, nullptr,
0, 0, false);
big_core_jobs_.push_back(job);
big_core_threads_.emplace_back(MT21WorkerEntry, mask,
std::ref(should_shutdown_), symbol_cache_,
big_core_pivot_, job);
}
CPU_ZERO(&mask);
for (size_t i = 0; i < kNumLittleThreads; i++) {
CPU_SET(i, &mask);
}
little_core_pivot_ =
static_cast<uint8_t*>(aligned_alloc(16, aligned_resolution_.GetArea()));
for (size_t i = 0; i < kNumLittleThreads; i++) {
scoped_refptr<MT21DecompressionJob> job =
base::MakeRefCounted<MT21DecompressionJob>(nullptr, nullptr, 0, nullptr,
0, 0, true);
little_core_jobs_.push_back(job);
little_core_threads_.emplace_back(MT21WorkerEntry, mask,
std::ref(should_shutdown_), symbol_cache_,
little_core_pivot_, job);
}
// Experimental evidence shows that A53s decompress MT21 blocks at about half
// the speed of A72s. This conveniently means that if split the chroma plane
// between the A53s and the luma plane between the A72s, we should perfectly
// balance the load.
size_t uv_split_height = base::bits::AlignUp(
static_cast<size_t>(aligned_resolution_.height() / 2 / 2),
kMT21TileHeight / 2);
size_t uv_split_offset = uv_split_height * aligned_resolution_.width();
little_core_jobs_[0]->offset = 0;
little_core_jobs_[0]->width = aligned_resolution_.width();
little_core_jobs_[0]->height = uv_split_height;
little_core_jobs_[1]->offset = uv_split_offset;
little_core_jobs_[1]->width = aligned_resolution_.width();
little_core_jobs_[1]->height =
aligned_resolution_.height() / 2 - uv_split_height;
size_t y_split_height = base::bits::AlignUp(
static_cast<size_t>(aligned_resolution_.height() / 2), kMT21TileHeight);
size_t y_split_offset = y_split_height * aligned_resolution_.width();
big_core_jobs_[0]->offset = 0;
big_core_jobs_[0]->width = aligned_resolution_.width();
big_core_jobs_[0]->height = y_split_height;
big_core_jobs_[1]->offset = y_split_offset;
big_core_jobs_[1]->width = aligned_resolution_.width();
big_core_jobs_[1]->height = aligned_resolution_.height() - y_split_height;
}
MT21Decompressor::~MT21Decompressor() {
should_shutdown_ = true;
for (auto& job : little_core_jobs_) {
job->wakeup_event.Signal();
}
for (auto& job : big_core_jobs_) {
job->wakeup_event.Signal();
}
for (size_t i = 0; i < kNumLittleThreads; i++) {
little_core_threads_[i].join();
}
for (size_t i = 0; i < kNumBigThreads; i++) {
big_core_threads_[i].join();
}
delete little_core_pivot_;
delete big_core_pivot_;
delete symbol_cache_;
}
void MT21Decompressor::MT21ToNV12(const uint8_t* src_y,
const uint8_t* src_uv,
const size_t y_buf_size,
const size_t uv_buf_size,
uint8_t* dest_y,
uint8_t* dest_uv) {
const uint8_t* y_footer =
ComputeFooterOffset(aligned_resolution_.GetArea(), y_buf_size,
kMT21YFooterAlignment) +
src_y;
const uint8_t* uv_footer =
ComputeFooterOffset(aligned_resolution_.GetArea() / 2, uv_buf_size,
kMT21UVFooterAlignment) +
src_uv;
// Start little core jobs.
for (auto& job : little_core_jobs_) {
job->src = src_uv;
job->footer = uv_footer;
job->dest = dest_uv;
job->wakeup_event.Signal();
}
// Start big core jobs.
for (auto& job : big_core_jobs_) {
job->src = src_y;
job->footer = y_footer;
job->dest = dest_y;
job->wakeup_event.Signal();
}
// Wait for everything to finish.
for (auto& job : little_core_jobs_) {
job->done_event.Wait();
}
for (auto& job : big_core_jobs_) {
job->done_event.Wait();
}
}
} // namespace media
|
