Commit Graph

300 Commits

Author SHA1 Message Date
cyan4973
5fba09fa41 updated util's time for Windows compatibility
Correctly measures time on Posix systems when running with
Multi-threading

Todo : check Windows measurement under multi-threading
2017-01-20 12:57:31 -08:00
Yann Collet
b459aad5b4 renamed savedRep into repToConfirm 2017-01-19 17:33:37 -08:00
Yann Collet
500014af49 zstd cli can now compress using multi-threading
added : command -T#
added : ZSTD_resetCStream() (zstdmt_compress)
added : FIO_setNbThreads()  (fileio)
2017-01-19 17:04:28 -08:00
Yann Collet
19d670ba9d Added ZSTDMT_initCStream_advanced() variant
Correctly compress with custom params and dictionary
Added relevant fuzzer test in zstreamtest

Also :
new macro ZSTDMT_SECTION_LOGSIZE_MIN, which sets a minimum size for a full job
(note : a flush() command can still generate a partial job anytime)
2017-01-19 15:32:07 -08:00
Yann Collet
736788f8e8 added streaming fuzzer tests for MT API
Also : fixed corner case, where nb of jobs completed becomes > jobQueueSize
which is possible when many flushes are issued
while there is not enough dst buffer to flush completed ones.
2017-01-19 12:15:29 -08:00
Yann Collet
32dfae6f98 fixed Multi-threaded compression
MT compression generates a single frame.
Multi-threading operates by breaking the frames into independent sections.
But from a decoder perspective, there is no difference :
it's just a suite of blocks.

Problem is, decoder preserves repCodes from previous block to start decoding next block.
This is also valid between sections, since they are no different than changing block.

Previous version would incorrectly initialize repcodes to their default value at the beginning of each section.
When using them, there was a mismatch between encoder (default values) and decoder (values from previous block).

This change ensures that repcodes won't be used at the beginning of a new section.
It works by setting them to 0.
This only works with regular (single segment) variants : extDict variants will fail !
Fortunately, sections beyond the 1st one belong to this category.

To be checked : btopt strategy.
This change was only validated from fast to btlazy2 strategies.
2017-01-19 10:32:55 -08:00
Yann Collet
37226c1e9f Simplified compressChunk job
minor refactoring : compression done in a single call on first chunk
Avoid a mutable hSize variable and eventual recombination to cSize at the end
2017-01-19 10:18:17 -08:00
Yann Collet
6073b3e6b8 ZSTDMT_endStream : nullify input buffer after flush
There will be no more input after ZSTDMT_endStream invocation :
only flush/end is allowed (to fully collect compressed result).
2017-01-18 15:32:38 -08:00
Yann Collet
3a01c46b26 ZSTDMT_initCStream() supports restart from invalid state
ZSTDMT_initCStream() will correcly scrub for resources
when it detects that previous compression was not properly finished.
2017-01-18 15:18:17 -08:00
Yann Collet
4885f591b3 trap compression errors, collect back resources from workers 2017-01-18 14:11:37 -08:00
Yann Collet
563ef8acf4 CCtxPool starts empty, as suggested by @terrelln
Also : make zstdmt now a target from root
2017-01-18 12:12:10 -08:00
Yann Collet
a6db7a7b9b fixed cmaketest
(buffer_t){NULL,0} is not considered a constant.
{NULL,0} is.
2017-01-18 11:57:34 -08:00
Yann Collet
0d6b8f65a9 ZSTDMT_free() scrubs potentially unfinished jobs to release their resources
In some complex scenarios (free() without finishing compression),
it is possible that some resources are still into jobs
and not collected back into pools.
In which case, previous version of free() would miss them.
This would be equivalent to a leak.

New version ensures that it even foes after such resource.
It requires job consumers to properly mark resources as released,
by replacing entries by NULL after releasing back to the pool.

Obviously, it's not recommended to free() zstdmt context mid-term,
still that's now a supported scenario.

The same methodology is also used to ensure proper resource collection
after an error is detected.

Still to do :
- detect compression errors (not just allocation ones)
- properly manage resource when init() is called without finishing previous compression.
2017-01-17 17:46:33 -08:00
Yann Collet
d0a1d45582 ZSTDMT_{flush,end}Stream() now block on next job completion when nothing to flush
The main issue was to avoid a caller to continually loop on {flush,end}Stream()
when there was nothing ready to be flushed but still some compression work ongoing in a worker thread.
The continuous loop would have resulted in wasted energy.
The new version makes call to {flush,end}Stream blocking when there is nothing ready to be flushed.
Of course, if all worker threads have exhausted job, it will return zero (all flush completed).

Note : There are still some remaining issues to report error codes
and properly collect back resources into pools when an error is triggered.
2017-01-17 16:15:18 -08:00
Yann Collet
a73c412932 completed ZSTDMT streaming compression
Provides the baseline compression API :
size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel);
size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output);
size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output);

Not tested yet
2017-01-17 15:31:16 -08:00
Sean Purcell
57d423c5df Don't create dict in streaming apis if dictSize == 0 2017-01-17 14:31:35 -08:00
Gregory Szorc
7d6f478d15 Set dictionary ID in ZSTD_initCStream_usingCDict()
When porting python-zstandard to use ZSTD_initCStream_usingCDict()
so compression dictionaries could be reused, an automated test
failed due to compressed content changing.

I tracked this down to ZSTD_initCStream_usingCDict() not
setting the dictID field of the ZSTD_CCtx attached to the
ZSTD_CStream instance.

I'm not 100% convinced this is the correct or full solution,
as I'm still seeing one automated test failing with this change.
2017-01-14 17:44:54 -08:00
Yann Collet
5b726dbe4d fix gcc-arm warning "suggest braces around empty body" 2017-01-12 17:46:46 +01:00
Yann Collet
ad9f6bd123 zstdmt : fix : resources properly collected even when early fail
In previous version, main function would return early when detecting a job error.
Late threads resources were therefore not collected back into pools.
New version just register the error, but continue the collecting process.
All buffers and context should be released back to pool before leaving main function.
2017-01-12 03:06:35 +01:00
Yann Collet
b05c4828ea zstdmt : correctly check for cctx and buffer allocation
Result from getBuffer and getCCtx could be NULL when allocation fails.
Now correctly checks : job creation stop and last job reports an allocation error.
releaseBuffer and releaseCCtx are now also compatible with NULL input.

Identified a new potential issue :
when early job fails, later jobs are not collected for resource retrieval.
2017-01-12 02:01:28 +01:00
Yann Collet
107bcbbbc2 zstdmt : changed internal naming from frame to chunk
Since the result of mt compression is a single frame,
changed naming, which implied the concatenation of multiple frames.

minor : ensures that content size is written in header
2017-01-12 01:25:46 +01:00
Yann Collet
5eb749e734 ZSTDMT_compress() creates a single frame
The new strategy involves cutting frame at block level.
The result is a single frame, preserving ZSTD_getDecompressedSize()

As a consequence, bench can now make a full round-trip,
since the result is compatible with ZSTD_decompress().

This strategy will not make it possible to decode the frame with multiple threads
since the exact cut between independent blocks is not known.
MT decoding needs further discussions.
2017-01-11 18:21:25 +01:00
Yann Collet
04cbc36499 minor refactor (release CCtx 1st) and comment clarification 2017-01-11 16:08:08 +01:00
Yann Collet
085179bb78 fixed ZSTDMT_createCCtx() : checked inner objects are properly created 2017-01-11 15:58:05 +01:00
Yann Collet
8ce1cc2bec improved ZSTD_createCCtxPool() cancellation
use ZSTD_freeCCtxPool() to release the partially created pool.
avoids to duplicate logic.

Also : identified a new difficult corner case :
when freeing the Pool, all CCtx should be previously released back to the pool.
Otherwise, it means some CCtx are still in use.
There is currently no clear policy on what to do in such a case.
Note : it's supposed to never happen.
Since pool creation/usage is static, it has no external user,
which limits risks.
2017-01-11 15:44:26 +01:00
Yann Collet
47557ba2b2 fixed ZSTDMT_createCCtxPool() when inner CCtx creation fails 2017-01-11 15:35:56 +01:00
Yann Collet
f1cb55192c fixed linux warnings 2017-01-02 01:11:55 +01:00
Yann Collet
0ec6a95ba1 minor fixes 2017-01-02 00:49:42 +01:00
Yann Collet
2ec635a162 use pthread_cond to send signals between threads 2017-01-01 17:31:33 +01:00
Yann Collet
3b9d434356 extended ZSTDMT code support for non-MT systems and WIN32 (preliminary) 2016-12-31 16:32:19 +01:00
Yann Collet
c8efc1c874 simplified Buffer Pool 2016-12-31 14:45:33 +01:00
Yann Collet
3b29dbd9e8 new zstdmt version using generic treadpool 2016-12-31 06:04:25 +01:00
Yann Collet
c6a6417458 bench correctly measures time for multi-threaded compression (posix only) 2016-12-31 03:31:26 +01:00
Yann Collet
e70912c72b Changed : input divided into roughly equal parts.
Debug : can measure time waiting for mutexes to unlock.
2016-12-29 01:24:01 +01:00
Yann Collet
6c0ed9483a compression threads use ZSTD_compressCCtx() 2016-12-28 17:08:28 +01:00
Yann Collet
ce9e1452fd protect buffer pool with a mutex 2016-12-28 15:31:19 +01:00
Yann Collet
3d93f2fce7 first zstdmt sketch 2016-12-27 07:19:36 +01:00
Yann Collet
aca113f4f5 fixed ZSTD_sizeof_?Dict() 2016-12-23 22:25:03 +01:00
Yann Collet
4e5eea61a8 added ZSTD_createDDict_byReference() 2016-12-21 16:44:35 +01:00
Yann Collet
1f57c2ed32 added : ZSTD_createCDict_byReference() 2016-12-21 16:20:11 +01:00
Nick Terrell
8157a4c3cc Fix dictionary loading bug causing an MSAN failure
Offset rep codes must be in the range `[1, dictSize)`.
Fix dictionary loading to reject `0` as a offset rep code.
2016-12-20 10:47:52 -08:00
Yann Collet
d564faa3c6 fix : ZSTD_initCStream_srcSize() correctly set srcSize in frame header 2016-12-18 21:39:15 +01:00
Yann Collet
e795c8a5f6 Added ZSTD_initCStream_srcSize().
Added relevant test cases in zstreamtest
2016-12-13 17:00:14 +01:00
Yann Collet
c3a5c4bef8 introduced cycleLog 2016-12-12 00:47:30 +01:00
Yann Collet
c261f71f6a minor variation of rescale fix 2016-12-12 00:25:07 +01:00
Nick Terrell
3826207a70 Simplify segfault fix
Take advantage of the fact that `chainLog <= windowLog`.
2016-12-10 18:46:55 -08:00
Nick Terrell
0012332ce0 Fix compression segfault
When the overflow protection kicks in, it makes sure that ip - ctx->base
isn't too large.  However, it didn't ensure that saved offsets are
still valid.  This change ensures that any valid offsets (<= windowLog)
are still representable after the update.

The bug would shop up on line 1056, when `offset_1 > current + 1`, which
causes an underflow.  This in turn, would cause a segfault on line 1063.

The input must necessarily be longer than 1 GB for this issue to occur.
Even then, it only occurs if one of the last 3 matches is larger than
the chain size and block size.
2016-12-09 17:15:33 -08:00
Yann Collet
825dffbc43 moved zbuff source files into lib/deprecated 2016-12-05 19:28:19 -08:00
Yann Collet
a0d742b1e4 introduced HUF_buildCTable_wksp(), to reduce stack memory usage 2016-12-01 17:47:30 -08:00
Yann Collet
643d9a234b replaced usage of FSE_buildCTable by FSE_buildCTable_wksp, using less stack space in the process 2016-12-01 16:24:04 -08:00
Yann Collet
e928f7e16d introduced ext_wksp variants of count to reduce stack memory usage 2016-12-01 16:13:35 -08:00
Yann Collet
979cab412b fixed some minor visual silent cast warnings.
introduced FSE_count_parallel_wksp().
2016-11-30 18:10:38 -08:00
Yann Collet
5e00b848a8 FSE_compress_wksp() uses less stack space 2016-11-30 16:46:13 -08:00
Yann Collet
d79a9a00d9 Introduced FSE_compress_wksp() and FSE_buildCTable_wksp() to reduce stack memory usage 2016-11-30 15:52:20 -08:00
Yann Collet
25f46dcc0f minor const 2016-11-29 16:59:27 -08:00
Przemyslaw Skibinski
fc4193bda5 fixed g++ warnings 2016-11-23 18:17:18 +01:00
Przemyslaw Skibinski
9ca65af810 zstd_opt.h: improved price function 2016-11-23 17:22:54 +01:00
Przemyslaw Skibinski
ad3e94512c fixed warnings from static analyzer in zstd_opt.h 2016-11-21 20:22:12 +01:00
Przemyslaw Skibinski
3d18088b38 updated windres 2016-11-17 18:04:41 +01:00
Yann Collet
407a11f63e fixed Visual compatibility 2016-11-03 15:52:01 -07:00
Nick Terrell
d82efd8a70 ZSTD_compress_usingDict() when dict gets loaded
Specify that when `dict == NULL || dictSize < 8` no dictionary
gets loaded.
Also add some periods.
2016-11-02 18:07:16 -07:00
Yann Collet
ee5b725823 ZSTD_initCStream() optimization : do not allocate a CDict when no dictionary used 2016-10-27 14:20:55 -07:00
Yann Collet
335ad5d4d4 added ZSTD_initDStream_usingDDict() .
slightly optimized ZSTD_initDStream() when no dictionary .
fixed ZSTD_sizeof_CStream() .
2016-10-25 17:47:02 -07:00
Yann Collet
9516234e67 first sketch for ZSTD_initCStream_usingCDict() 2016-10-25 16:19:52 -07:00
Yann Collet
62d9a7ddfd Merge pull request #429 from inikep/btopt2
Btopt2
2016-10-25 14:48:43 -07:00
Przemyslaw Skibinski
5c5f01f3da added ZSTD_btopt2 strategy 2016-10-25 12:25:07 +02:00
Nick Terrell
b2c39a22b0 Fix compiler narrowing warning 2016-10-24 14:50:13 -07:00
Nick Terrell
f698ad6deb Merge remote-tracking branch 'upstream/dev' into fixes
* upstream/dev:
  added doc\zstd_manual.html
  added contrib\gen_html
  zstd_compression_format.md moved to doc/
  Fix small bug in ZSTD_execSequence()
  improved ZSTD_compressBlock_opt_extDict_generic
  protect ZSTD_decodeFrameHeader() from invalid usage, as suggested by @spaskob
  zstd_opt.h: small improvement in compression ratio
  improved dicitonary segment merge
  use implicit rules to compile zstd_decompress.c
  detect early impossible decompression scenario in legacy decoder v0.5
  no repeat mode in legacy v0.5
  fixed invalid invocation of dictionary in legacy decoder v0.5
  fix edge case
  fix command line interpretation
  fixed minor corner case
  zstd.h: added the Introduction section
  fixed clang 3.5 warnings
  zstd.h: updated comments
2016-10-24 13:10:13 -07:00
Nick Terrell
f9c9af3c2e Reject dictionaries with incomplete entropy tables
If a dictionary specifies that a symbol has probability zero in its
`matchLength`, `literalLength`, or `offset` FSE table, but the symbol
appears when compressing input, the compressor fails.

Ensure that dictionaries support all `matchLength`, and `literalLength`
codes.  They must also support all of the `offset` codes required to
represent every possible offset that can appear in the first block.
2016-10-24 10:42:44 -07:00
Przemyslaw Skibinski
3ee94a7600 zstd_compression_format.md moved to doc/ 2016-10-24 15:58:07 +02:00
Przemyslaw Skibinski
4732074a71 improved ZSTD_compressBlock_opt_extDict_generic 2016-10-21 11:19:00 +02:00
Przemyslaw Skibinski
d365ae3497 zstd_opt.h: small improvement in compression ratio 2016-10-20 11:49:02 +02:00
Yann Collet
197a55ee7b fix edge case 2016-10-18 11:27:52 -07:00
Nick Terrell
fd98087047 Fix stack buffer overflow in HUF_readCTable()
If `w ==0` on line 153, then `CTable[n].nbBits == tableLog + 1`.
Then `nbPerRank[CTable[n].nbBits]` and `valPerRank[CTable[n].nbBits]`
are stack buffer overflows.
2016-10-17 18:16:59 -07:00
Nick Terrell
bfd943ace5 Fix buffer overrun in ZSTD_loadDictEntropyStats()
The table log set by `FSE_readNCount()` was not checked in
`ZSTD_loadDictEntropyStats()`.  This caused `FSE_buildCTable()`
to stack/heap overflow in a few places.

The benchmarks look good, there is no obvious compression performance regression:

  > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar
   1#silesia.tar       : 211988480 ->  73656930 (2.878), 271.6 MB/s , 716.8 MB/s
   2#silesia.tar       : 211988480 ->  70162842 (3.021), 204.8 MB/s , 671.1 MB/s
   3#silesia.tar       : 211988480 ->  66997986 (3.164), 156.8 MB/s , 658.6 MB/s
   4#silesia.tar       : 211988480 ->  66002591 (3.212), 136.4 MB/s , 665.3 MB/s
   5#silesia.tar       : 211988480 ->  65008480 (3.261),  98.9 MB/s , 647.0 MB/s
   6#silesia.tar       : 211988480 ->  62979643 (3.366),  65.2 MB/s , 670.4 MB/s
   7#silesia.tar       : 211988480 ->  61974560 (3.421),  44.9 MB/s , 688.2 MB/s
   8#silesia.tar       : 211988480 ->  61028308 (3.474),  32.4 MB/s , 711.9 MB/s
   9#silesia.tar       : 211988480 ->  60416751 (3.509),  21.1 MB/s , 718.1 MB/s
  10#silesia.tar       : 211988480 ->  60174239 (3.523),  22.2 MB/s , 721.8 MB/s

  > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar
   1#silesia.tar       : 211988480 ->  73656930 (2.878), 273.8 MB/s , 722.0 MB/s
   2#silesia.tar       : 211988480 ->  70162842 (3.021), 203.2 MB/s , 666.6 MB/s
   3#silesia.tar       : 211988480 ->  66997986 (3.164), 157.4 MB/s , 666.5 MB/s
   4#silesia.tar       : 211988480 ->  66002591 (3.212), 132.1 MB/s , 661.9 MB/s
   5#silesia.tar       : 211988480 ->  65008480 (3.261),  96.8 MB/s , 641.6 MB/s
   6#silesia.tar       : 211988480 ->  62979643 (3.366),  63.1 MB/s , 677.0 MB/s
   7#silesia.tar       : 211988480 ->  61974560 (3.421),  44.3 MB/s , 678.2 MB/s
   8#silesia.tar       : 211988480 ->  61028308 (3.474),  33.1 MB/s , 708.9 MB/s
   9#silesia.tar       : 211988480 ->  60416751 (3.509),  21.5 MB/s , 710.1 MB/s
  10#silesia.tar       : 211988480 ->  60174239 (3.523),  21.9 MB/s , 723.9 MB/s
2016-10-17 16:55:52 -07:00
Yann Collet
2b361cf2f1 minor opt 2016-10-14 16:09:07 -07:00
Nick Terrell
3b9cdf9220 Fix ubsan failures (pass NULL to memcpy) 2016-10-12 20:54:42 -07:00
Yann Collet
cf409a7e2a fixed : init*_advanced() followed by reset() with different pledgedSrcSiz 2016-09-26 16:41:05 +02:00
Yann Collet
97b378a6f8 Streaming : dictionary compression on multiple files / segments can correctly provide srcSize into header (when provided) using pledgedSrcSize. 2016-09-21 17:20:19 +02:00
Yann Collet
993060e0f2 cli : better adaptation to small files 2016-09-21 16:46:08 +02:00
Yann Collet
a6bdf55759 fixed memory leak 2016-09-15 17:02:06 +02:00
Yann Collet
644a8da88a fixed minor conversion warning 2016-09-15 16:16:21 +02:00
Yann Collet
4cb212938c introduced ZSTD_resetCStream() 2016-09-15 14:54:07 +02:00
Yann Collet
fa0c09760c variable renaming 2016-09-15 14:11:01 +02:00
Yann Collet
d7c6589df8 support ZSTD_sizeof_*() on NULL
added ZSTD_sizeof_CDict()
2016-09-15 02:57:27 +02:00
Yann Collet
64deef3bee Fixed srcSize=1 2016-09-14 00:16:07 +02:00
Yann Collet
ac8bace6b1 support large skippable frames 2016-09-07 14:54:23 +02:00
Yann Collet
95d07d7447 introduced CHECK_E 2016-09-06 16:38:51 +02:00
Yann Collet
3e21ec5b01 introduced CHECK_F 2016-09-06 15:36:19 +02:00
Yann Collet
5c956d593c FORCE_INLINE common definition 2016-09-06 15:05:19 +02:00
Yann Collet
edbcd9f5b2 fixed zbufftest 2016-09-06 14:30:57 +02:00
Yann Collet
b624922b14 fixed checksum 2016-09-06 11:16:57 +02:00
Yann Collet
a7737f6a60 improved compression on small files when using same parameters 2016-09-06 09:44:59 +02:00
Yann Collet
7ae67bb18a small compression speed gains with using_CDict 2016-09-06 06:28:05 +02:00
Yann Collet
855766d73d clarified dictionary in format description 2016-09-02 17:04:49 -07:00
Yann Collet
1563bfeabc fixing FORCE_INLINE for older compilers (#330) 2016-09-02 11:44:21 -07:00
David Lam
e10f7f3dcb merge 2016-08-30 12:03:36 -07:00
Yann Collet
4ded9e591c added boilerplate 2016-08-30 11:06:28 -07:00
Yann Collet
14200a20f0 Fixed issue #304, reported by @borzunov 2016-08-30 06:51:00 -07:00
David Lam
da9d3b7057 Cleanup some errors in typedef comments and remove duplicated HOWTO from zbuff_decompress.c 2016-08-29 17:31:51 -07:00