Project
Loading...
Searching...
No Matches
Codecs.cxx
Go to the documentation of this file.
1// Copyright 2019-2026 CERN and copyright holders of ALICE O2.
2// See https://alice-o2.web.cern.ch/copyright for details of the copyright holders.
3// All rights not expressly granted are reserved.
4//
5// This software is distributed under the terms of the GNU General Public
6// License v3 (GPL Version 3), copied verbatim in the file "COPYING".
7//
8// In applying this license CERN does not waive the privileges and immunities
9// granted to it by virtue of its status as an Intergovernmental Organization
10// or submit itself to any jurisdiction.
11
12#include "AODNative/Codecs.h"
13#include "AODNative/Errors.h"
14#include "Endian.h"
15#include <bit>
16#include <cstring>
17#include <lz4.h>
18#include <lzma.h>
19#include <stdexcept>
20#include <zlib.h>
21// For the buffer-less decompression API, exported by the shared library too.
22#define ZSTD_STATIC_LINKING_ONLY
23#include <zstd.h>
24
25namespace o2::aod::native
26{
27
28namespace
29{
30constexpr size_t kChunkHeaderSize = 9;
31constexpr size_t kLZ4ChecksumSize = 8;
32
33uint32_t le24(uint8_t const* p) { return p[0] | (p[1] << 8) | (p[2] << 16); }
34} // namespace
35
36// One decoder state per codec, created on first use and reset for each chunk.
38 ZSTD_DCtx* zstd = nullptr;
39 z_stream zlib{};
40 bool zlibInit = false;
41 lzma_stream lzma = LZMA_STREAM_INIT;
42 bool lzmaInit = false;
43 bool lz4Partial = false; // the current LZ4 chunk was decoded partially, straight into the destination
44
46 {
47 if (zstd) {
48 ZSTD_freeDCtx(zstd);
49 }
50 if (zlibInit) {
51 inflateEnd(&zlib);
52 }
53 if (lzmaInit) {
54 lzma_end(&lzma);
55 }
56 }
57};
58
59PayloadDecompressor::PayloadDecompressor(std::span<const uint8_t> src, uint32_t objLen, std::string const& file, uint64_t fileOffset)
60 : mSrc(src), mObjLen(objLen), mFile(file), mFileOffset(fileOffset), mCodec(std::make_unique<Codec>())
61{
62}
63
65
66void PayloadDecompressor::nextChunk()
67{
68 auto fail = [this](std::string const& what) { throw CorruptData(mFile, mFileOffset + mSrcPos, what); };
69 if (mSrc.size() - mSrcPos < kChunkHeaderSize) {
70 fail("compressed payload ends after " + std::to_string(mProduced) + " of " + std::to_string(mObjLen) + " bytes");
71 }
72 auto* h = mSrc.data() + mSrcPos;
73 mTag[0] = (char)h[0];
74 mTag[1] = (char)h[1];
75 size_t cSize = le24(h + 3);
76 mChunkOut = le24(h + 6);
77 mChunkDone = 0;
78 if (cSize > mSrc.size() - mSrcPos - kChunkHeaderSize) {
79 fail("compressed chunk of " + std::to_string(cSize) + " bytes past the end of the payload");
80 }
81 if (mChunkOut == 0 || mChunkOut > mObjLen - mProduced) {
82 fail("chunk of " + std::to_string(mChunkOut) + " uncompressed bytes does not fit the object length");
83 }
84 mChunkIn = mSrc.subspan(mSrcPos + kChunkHeaderSize, cSize);
85 auto& codec = *mCodec;
86 mScratch.clear();
87 if (mTag[0] == 'Z' && mTag[1] == 'S') {
88 if (!codec.zstd) {
89 codec.zstd = ZSTD_createDCtx();
90 }
91 ZSTD_DCtx_reset(codec.zstd, ZSTD_reset_session_only);
92 } else if (mTag[0] == 'Z' && mTag[1] == 'L') {
93 if (codec.zlibInit) {
94 inflateReset(&codec.zlib);
95 } else {
96 if (inflateInit(&codec.zlib) != Z_OK) {
97 fail("cannot initialise zlib");
98 }
99 codec.zlibInit = true;
100 }
101 codec.zlib.next_in = (Bytef*)mChunkIn.data();
102 codec.zlib.avail_in = mChunkIn.size();
103 } else if (mTag[0] == 'X' && mTag[1] == 'Z') {
104 if (codec.lzmaInit) {
105 lzma_end(&codec.lzma);
106 }
107 codec.lzma = LZMA_STREAM_INIT;
108 if (lzma_stream_decoder(&codec.lzma, UINT64_MAX, 0) != LZMA_OK) {
109 fail("cannot initialise liblzma");
110 }
111 codec.lzmaInit = true;
112 codec.lzma.next_in = mChunkIn.data();
113 codec.lzma.avail_in = mChunkIn.size();
114 } else if (mTag[0] == 'L' && mTag[1] == '4') {
115 if (cSize < kLZ4ChecksumSize) {
116 fail("LZ4 chunk too small for its checksum");
117 }
118 uint64_t expected = 0;
119 for (size_t i = 0; i < kLZ4ChecksumSize; ++i) {
120 expected = (expected << 8) | mChunkIn[i];
121 }
122 mChunkIn = mChunkIn.subspan(kLZ4ChecksumSize);
123 if (xxh64(mChunkIn) != expected) {
124 fail("LZ4 checksum mismatch");
125 }
126 codec.lz4Partial = false;
127 } else {
128 throw UnsupportedLayout(std::string("compression algorithm \"") + mTag[0] + mTag[1] + "\" in " + mFile);
129 }
130 mSrcPos += kChunkHeaderSize + cSize;
131}
132
133size_t PayloadDecompressor::produceFromChunk(uint8_t* dst, size_t n)
134{
135 auto fail = [this](std::string const& what) { throw CorruptData(mFile, mFileOffset + mSrcPos, what); };
136 auto& codec = *mCodec;
137 bool wholeChunk = mChunkDone == 0 && n == mChunkOut;
138 if (mTag[0] == 'Z' && mTag[1] == 'S') {
139 if (wholeChunk) {
140 auto r = ZSTD_decompressDCtx(codec.zstd, dst, n, mChunkIn.data(), mChunkIn.size());
141 if (ZSTD_isError(r) || r != n) {
142 fail(std::string("zstd: ") + (ZSTD_isError(r) ? ZSTD_getErrorName(r) : "size mismatch"));
143 }
144 return n;
145 }
146 if (mPrefixOnly && mChunkDone == 0) {
147 return zstdPrefix(dst, n);
148 }
149 // Resume where the previous read stopped: the input position is kept in mChunkIn.
150 // The streaming API keeps its own window buffer, sized by the writer's window.
151 ZSTD_outBuffer out{dst, n, 0};
152 while (out.pos < out.size) {
153 ZSTD_inBuffer in{mChunkIn.data(), mChunkIn.size(), 0};
154 auto r = ZSTD_decompressStream(codec.zstd, &out, &in);
155 mChunkIn = mChunkIn.subspan(in.pos);
156 if (ZSTD_isError(r)) {
157 fail(std::string("zstd: ") + ZSTD_getErrorName(r));
158 }
159 if (out.pos < out.size && (r == 0 || (in.pos == 0 && mChunkIn.empty()))) {
160 fail("zstd chunk ended early");
161 }
162 }
163 return n;
164 }
165 if (mTag[0] == 'Z' && mTag[1] == 'L') {
166 auto& z = codec.zlib;
167 z.next_out = dst;
168 z.avail_out = n;
169 while (z.avail_out > 0) {
170 auto r = inflate(&z, Z_NO_FLUSH);
171 if (r == Z_STREAM_END && z.avail_out > 0) {
172 fail("zlib chunk ended early");
173 }
174 if (r != Z_OK && r != Z_STREAM_END) {
175 fail(std::string("zlib: ") + (z.msg ? z.msg : std::to_string(r)));
176 }
177 }
178 return n;
179 }
180 if (mTag[0] == 'X' && mTag[1] == 'Z') {
181 auto& s = codec.lzma;
182 s.next_out = dst;
183 s.avail_out = n;
184 while (s.avail_out > 0) {
185 auto r = lzma_code(&s, LZMA_RUN);
186 if (r == LZMA_STREAM_END && s.avail_out > 0) {
187 fail("xz chunk ended early");
188 }
189 if (r != LZMA_OK && r != LZMA_STREAM_END) {
190 fail("liblzma error " + std::to_string(r));
191 }
192 }
193 return n;
194 }
195 // LZ4 block format cannot be resumed. Decode partially straight into the
196 // destination when this read starts the chunk, which is the common case of
197 // stopping at the end of the value region. Should a later read need the
198 // rest, decode the whole chunk into scratch memory.
199 if (wholeChunk) {
200 // A whole chunk must use all of its input and produce exactly its size.
201 auto r = LZ4_decompress_safe((char const*)mChunkIn.data(), (char*)dst, mChunkIn.size(), n);
202 if (r < 0 || (size_t)r != n) {
203 fail("LZ4 chunk size mismatch");
204 }
205 return n;
206 }
207 if (mScratch.empty() && mChunkDone == 0) {
208 auto r = LZ4_decompress_safe_partial((char const*)mChunkIn.data(), (char*)dst, mChunkIn.size(), n, n);
209 if (r < 0 || (size_t)r != n) {
210 fail("LZ4 decompression failed");
211 }
212 codec.lz4Partial = true;
213 return n;
214 }
215 if (mScratch.empty()) {
216 mScratch.resize(mChunkOut);
217 auto r = LZ4_decompress_safe((char const*)mChunkIn.data(), (char*)mScratch.data(), mChunkIn.size(), mChunkOut);
218 if (r < 0 || (size_t)r != mChunkOut) {
219 fail("LZ4 decompression failed");
220 }
221 }
222 std::memcpy(dst, mScratch.data() + mChunkDone, n);
223 return n;
224}
225
226// Decodes the first @a n bytes of a zstd chunk straight into @a dst, with the
227// buffer-less API: the destination itself is zstd's history, so no window
228// buffer is allocated. Blocks are written in place while a whole block (at most
229// ZSTD_BLOCKSIZE_MAX) still fits; the last ones go to a fixed scratch, contiguous
230// with each other, and only the needed bytes are copied out.
231size_t PayloadDecompressor::zstdPrefix(uint8_t* dst, size_t n)
232{
233 auto fail = [this](std::string const& what) { throw CorruptData(mFile, mFileOffset + mSrcPos, what); };
234 auto* dctx = mCodec->zstd;
235 if (ZSTD_isError(ZSTD_decompressBegin(dctx))) {
236 fail("cannot initialise zstd");
237 }
238 auto in = mChunkIn;
239 size_t written = 0;
240 bool inScratch = false;
241 size_t scratchUsed = 0;
242 while (written < n) {
243 auto need = ZSTD_nextSrcSizeToDecompress(dctx);
244 if (need == 0) {
245 fail("zstd chunk ended early");
246 }
247 if (need > in.size()) {
248 fail("truncated zstd chunk");
249 }
250 size_t r;
251 if (!inScratch && n - written >= ZSTD_BLOCKSIZE_MAX) {
252 r = ZSTD_decompressContinue(dctx, dst + written, n - written, in.data(), need);
253 if (!ZSTD_isError(r)) {
254 written += r;
255 }
256 } else {
257 if (!inScratch) {
258 mScratch.resize(2 * ZSTD_BLOCKSIZE_MAX);
259 inScratch = true;
260 }
261 r = ZSTD_decompressContinue(dctx, mScratch.data() + scratchUsed, mScratch.size() - scratchUsed, in.data(), need);
262 if (!ZSTD_isError(r)) {
263 scratchUsed += r;
264 if (written + scratchUsed >= n) {
265 std::memcpy(dst + written, mScratch.data(), n - written);
266 written = n;
267 }
268 }
269 }
270 if (ZSTD_isError(r)) {
271 fail(std::string("zstd: ") + ZSTD_getErrorName(r));
272 }
273 in = in.subspan(need);
274 }
275 mScratch.clear();
276 mScratch.shrink_to_fit();
277 return n;
278}
279
281{
282 if (mProduced != 0) {
283 throw std::logic_error("readPrefix must be the first read of a payload");
284 }
285 mPrefixOnly = true;
286 read(dst, n);
287}
288
290{
291 if (mPrefixOnly && mProduced != 0) {
292 throw std::logic_error("no read can follow readPrefix");
293 }
294 if (n > mObjLen - mProduced) {
295 throw CorruptData(mFile, mFileOffset, "read of " + std::to_string(n) + " bytes past the object length " + std::to_string(mObjLen));
296 }
297 while (n > 0) {
298 if (mChunkDone == mChunkOut) {
299 nextChunk();
300 }
301 auto k = std::min(n, mChunkOut - mChunkDone);
302 produceFromChunk(dst, k);
303 dst += k;
304 n -= k;
305 mChunkDone += k;
306 mProduced += k;
307 }
308}
309
310size_t decompressBounded(std::span<const uint8_t> src, uint32_t objLen, uint8_t* dst, size_t maxOut,
311 std::string const& file, uint64_t fileOffset)
312{
313 auto n = std::min<size_t>(objLen, maxOut);
314 PayloadDecompressor d{src, objLen, file, fileOffset};
315 d.readPrefix(dst, n);
316 return n;
317}
318
319uint64_t xxh64(std::span<const uint8_t> data)
320{
321 constexpr uint64_t P1 = 11400714785074694791ULL;
322 constexpr uint64_t P2 = 14029467366897019727ULL;
323 constexpr uint64_t P3 = 1609587929392839161ULL;
324 constexpr uint64_t P4 = 9650029242287828579ULL;
325 constexpr uint64_t P5 = 2870177450012600261ULL;
326 auto read64 = [](uint8_t const* p) {
327 uint64_t v;
328 std::memcpy(&v, p, 8);
329 return std::endian::native == std::endian::little ? v : bswap(v);
330 };
331 auto read32 = [](uint8_t const* p) {
332 uint32_t v;
333 std::memcpy(&v, p, 4);
334 return std::endian::native == std::endian::little ? v : bswap(v);
335 };
336 auto round = [](uint64_t acc, uint64_t input) {
337 acc += input * P2;
338 return std::rotl(acc, 31) * P1;
339 };
340 auto merge = [&round](uint64_t acc, uint64_t v) {
341 acc ^= round(0, v);
342 return acc * P1 + P4;
343 };
344
345 auto const* p = data.data();
346 auto const* end = p + data.size();
347 uint64_t h;
348 if (data.size() >= 32) {
349 uint64_t v1 = P1 + P2, v2 = P2, v3 = 0, v4 = -P1;
350 for (; end - p >= 32; p += 32) {
351 v1 = round(v1, read64(p));
352 v2 = round(v2, read64(p + 8));
353 v3 = round(v3, read64(p + 16));
354 v4 = round(v4, read64(p + 24));
355 }
356 h = std::rotl(v1, 1) + std::rotl(v2, 7) + std::rotl(v3, 12) + std::rotl(v4, 18);
357 h = merge(merge(merge(merge(h, v1), v2), v3), v4);
358 } else {
359 h = P5;
360 }
361 h += data.size();
362 for (; end - p >= 8; p += 8) {
363 h ^= round(0, read64(p));
364 h = std::rotl(h, 27) * P1 + P4;
365 }
366 if (end - p >= 4) {
367 h ^= (uint64_t)read32(p) * P1;
368 h = std::rotl(h, 23) * P2 + P3;
369 p += 4;
370 }
371 for (; p < end; ++p) {
372 h ^= *p * P5;
373 h = std::rotl(h, 11) * P1;
374 }
375 h ^= h >> 33;
376 h *= P2;
377 h ^= h >> 29;
378 h *= P3;
379 h ^= h >> 32;
380 return h;
381}
382
383} // namespace o2::aod::native
unsigned int uint32_t
int32_t i
configStandalone Light OpenGL display display_light Apply smoothing to points Enable Z buffer Highlight links Highlight seeds Highlight sector tracks Highlight final tracks Show individual Show related Highlight grid Show only individual collision Color clusters belonging to track objects Set line Enable drawing TPC data Enable drawing TOF data Enable drawing magnetic field Set field line step count Invert colors configStandalone Heavy OpenGL display display_heavy Show tracks with ITS contribution Show tracks with TOF contribution Enable propagation of loopers Show clusters at nominal x of pad row for early transformed data Split CE tracks when they cross the central Mark clusters Mark adjacent clusters Hide rejected tracks Apply filter on tracks to be displayed Separate increase X h configStandalone Camera window renderer OpenGL display display_camera Orient the camera such that the y axis is always upwards Full Screen MultiSample Anti Downsampling Anti Enable Vertical Sync Target frame or if the backend cannot limit to the refresh Use OpenGL indirect draws to reduce number of draw calls Display h configStandalone Vulkan display display_vulkan Have uniform buffers in host accessible device memory h configStandalone Internal display display_internal Number of updates the Qt GUI may send ahead of the last one the display has before it waits for the display Opacity of the console background Number of console lines the history is shortened to this when it reaches twice the size Time in seconds an info message fades out after setInfoTime h OpenGL display display Font size Do not use Freetype for font Renderer for event Frontend to use for event ROOT macros used as track filter Console script file
Timeframe settings Average distance in cm of events merged into Randomize distance around average distance of merged events Simulate correct bunch interactions instead of placing only the average number of events A value[n] sets TFlen for[n] collisions in Number of bunches per trainsort Number of bunch trains Instantaneous interaction rate Do not simulate border Total number of collisions to be placed in the interior of all time Do not select random but walk over array of events in stride steps h Event generator settings h configStandalone Use CPU for GPU Force usage of the specified GPU device no CPU fallback Continue processing after an error s
Timeframe settings Average distance in cm of events merged into Randomize distance around average distance of merged events Simulate correct bunch interactions instead of placing only the average number of events A value[n] sets TFlen for[n] collisions in Number of bunches per trainsort Number of bunch trains Instantaneous interaction rate Do not simulate border Total number of collisions to be placed in the interior of all time Do not select random but walk over array of events in stride steps h Event generator settings h configStandalone Use CPU for GPU Force usage of the specified GPU device no CPU fallback Continue processing after an error First event to Number of events to Number of iterations to Number of iterations to Number of initial iterations excluded from const char Directory with events to Events directory is and not inside events Show standalone event Run event generator maximum time bin of continuous for triggered for automatic continuous for automatic continuous triggered uint64_t
Timeframe settings Average distance in cm of events merged into Randomize distance around average distance of merged events Simulate correct bunch interactions instead of placing only the average number of events A value[n] sets TFlen for[n] collisions in Number of bunches per trainsort Number of bunch trains Instantaneous interaction rate Do not simulate border Total number of collisions to be placed in the interior of all time Do not select random but walk over array of events in stride steps h Event generator settings h configStandalone Use CPU for GPU Force usage of the specified GPU device no CPU fallback Continue processing after an error First event to Number of events to Number of iterations to Number of iterations to Number of initial iterations excluded from const char Directory with events to Events directory is and not inside events d
void merge(Options const &options)
Class for time synchronization of RawReader instances.
PayloadDecompressor(std::span< const uint8_t > src, uint32_t objLen, std::string const &file, uint64_t fileOffset)
Definition Codecs.cxx:59
void read(uint8_t *dst, size_t n)
Definition Codecs.cxx:289
void readPrefix(uint8_t *dst, size_t n)
Definition Codecs.cxx:280
GLdouble n
Definition glcorearb.h:1982
GLenum src
Definition glcorearb.h:1767
GLuint GLuint end
Definition glcorearb.h:469
const GLdouble * v
Definition glcorearb.h:832
GLenum GLenum dst
Definition glcorearb.h:1767
GLfloat GLfloat v1
Definition glcorearb.h:812
GLboolean r
Definition glcorearb.h:1233
GLfloat GLfloat GLfloat GLfloat v3
Definition glcorearb.h:814
GLfloat GLfloat GLfloat v2
Definition glcorearb.h:813
GLdouble GLdouble GLdouble z
Definition glcorearb.h:843
size_t decompressBounded(std::span< const uint8_t > src, uint32_t objLen, uint8_t *dst, size_t maxOut, std::string const &file="<memory>", uint64_t fileOffset=0)
Definition Codecs.cxx:310
constexpr T bswap(T v)
Definition Endian.h:22
uint64_t xxh64(std::span< const uint8_t > data)
XXH64 with seed 0, as used for the LZ4 chunk checksum.
Definition Codecs.cxx:319
std::string to_string(gsl::span< T, Size > span)
Definition common.h:52
std::map< std::string, ID > expected