1
0
mirror of https://github.com/facebook/zstd.git synced 2025-07-05 15:09:03 +02:00
Commit Graph

103 Commits

Author SHA1 Message Date
052a95f77c fix : ZSTDMT_compress_advanced() correctly generates checksum
when params.fParams.checksumFlag==1.
This use case used to be impossible when only ZSTD_compress() was available
2017-07-11 17:18:26 -07:00
2a62f48bf4 release input buffers from inside worker thread
buffers are released sooner, which makes them available faster for next job.
=> decreases total nb of buffers necessary
2017-07-11 15:56:40 -07:00
57236184af buffer pool : all buffers have same size
to reduce memory fragmentation.
They can be used for in or out, interchangeably.
2017-07-11 15:17:25 -07:00
34b2b95631 zstdmt : intermediate outBuffer allocated from within worker
reduces total amount of memory needed,
since jobs in queue do not have an outBuffer pre-reserved now
2017-07-11 14:59:10 -07:00
16261e6951 buffer pool can be invoked from multiple threads 2017-07-11 14:14:07 -07:00
ef0ff7fe7f zstdmt: removed margin for improved memory usage 2017-07-11 08:54:29 -07:00
4616fad18b improved ZSTDMT_compress() memory usage
does not need the input buffer for streaming operations

also : reduced a few tests time length
2017-07-10 17:16:41 -07:00
670b1fc547 optimized memory usage for ZSTDMT_compress()
Previously, each job would reserve a CCtx right before being posted.
The CCtx would be "part of the job description",
and only released when the job is completed (aka flushed).
For ZSTDMT_compress(), which creates all jobs first and only join at the end,
that meant one CCtx per job.
The nb of jobs used to be == nb of threads,
but since latest modification,
which reduces the size of jobs in order to spread the load of difficult areas,
it also increases the nb of jobs for large sources / small compression level.
This resulted in many more CCtx being created.

In this new version, CCtx are reserved within the worker thread.
It guaranteea there cannot be more CCtx reserved than workers (<= nb threads).

To do that, it required to make the CCtx Pool multi-threading-safe :
it can now be called from multiple threads in parallel.
2017-07-10 16:30:55 -07:00
3510efb02d fix : custom allocator correctly propagated to child contexts 2017-07-10 14:21:40 -07:00
e32fb0c1fe added ZSTD_sizeof_CCtx() test 2017-07-10 12:29:57 -07:00
27e883371d fixed wrong assert() condition
A single job created by ZSTDMT_compress() can be < 256KB
if data to compress is < 256 KB
(in which case it is delegated to single thread mode)
2017-07-04 19:33:16 -07:00
6383372dec fixed : 0-copy in NULL is UB 2017-07-04 10:36:41 -07:00
2de2396a36 refactor ZSTDMT_compress() 2017-07-03 16:23:36 -07:00
2084b041f4 fixed comments 2017-07-03 15:52:19 -07:00
1bafe393e4 fix : ZSTDMT_compressStream_generic() can accept NULL input
also : converge implementations towards new version of ZSTDMT_compressStream_generic()
2017-07-01 06:59:24 -07:00
58bd0e70fc fixed : dictionary compression with new advanced API in Multi-threading mode 2017-06-30 16:01:02 -07:00
d8b33a598d Optimized ZSTDMT single-pass mode speed on large sources
by ensuring job sizes remain "not too large"
2017-06-30 15:44:57 -07:00
d5c046c609 implemented shortcut for zstd_compress_generic() in MT mode
added ZSTDMT_compress_advanced() API
2017-06-30 14:51:01 -07:00
33a6639039 fixed ZSTD_refPrefix with Multithread-enabled CCtx 2017-06-28 11:09:43 -07:00
204b6b7ef6 fixed streaming buffered allocation with CDict compression 2017-06-21 15:13:00 -07:00
83095970e6 free cdictLocal faster, suggested by @terrelln 2017-06-21 12:26:40 -07:00
7bd1a2900e added ZSTD_dictMode_e to control dictionary loading mode 2017-06-21 11:50:33 -07:00
c08e649e95 first implementation of bench.c with new API ZSTD_compress_generic()
Doesn't speed optimize this buffer-to-buffer scenario yet.
Still internally defers to streaming implementation.

Also : fixed a long standing bug in ZSTDMT streaming API.
2017-06-19 18:25:35 -07:00
381e66cfbd added ZSTD_clampCParams()
now ZSTD_adjustCParams() is always successful,
it always produces a valid CParams
2017-06-16 17:34:54 -07:00
559ee82e90 fixed : calling ZSTD_compress_generic() to end-flush a stream in multiple steps 2017-06-16 11:58:21 -07:00
23aace9778 added control stage to MT mode 2017-06-11 18:32:36 -07:00
f35e2de61c linked newAPI to ZSTDMT 2017-06-05 18:32:48 -07:00
8c910d2097 updated ZSTDMT streaming API
ZSTDMT streaming API is now similar
and has same capabilites as single-thread streaming API.
It makes it easier to blend them together.
2017-06-03 01:15:02 -07:00
58e8d793e1 made debug definitions common within zstd_internal.h 2017-06-02 18:20:48 -07:00
c35e535002 added support for multithreading parameters 2017-06-01 18:44:06 -07:00
c4a5a21c5c created ZSTDMT_sizeof_CCtx() and POOL_sizeof()
required by ZSTD_sizeofCCtx() while adding a ZSTDMT_CCtx*
2017-06-01 17:56:14 -07:00
deee6e523f expose ZSTD_compress_generic_simpleArgs()
which is a binding towards ZSTD_compress_generic()
using only integral types for arguments.
2017-05-30 17:42:00 -07:00
5bcef1ada2 removed mtctx->cstream
use the first cctx in pool when ZSTDMT is used in single-thread mode
now that cctx and cstream are the same object.
2017-05-30 16:37:19 -07:00
44e45e8423 added ZSTDMT_createCCtx_advanced()
make it possible to use custom allocators
2017-05-30 16:12:06 -07:00
b6dec4c3ae fixed minor cast warning 2017-05-27 17:09:06 -07:00
e071159101 mtctx->jobs allocate its own memory space
to make ZSTDMT_CCtx_s size predictable
so that it can be included in CCtx
2017-05-27 00:21:33 -07:00
31533bacce Changed ZSTD_createCDict_advanced()
It now only uses compressionParameters as argument.
It produces many changes throughout user code,
though hopefully they tend to be simple :
just provide the cParams part from existing ZSTD_parameters.

Some programs might depend on ZSTD_createCDict_advanced() to pass frame parameters.
This change will force them to revisit this strategy and fix it,
since frame parameters are effectively silently ignored in current version.
2017-04-27 00:29:04 -07:00
2c5514c759 fixed ZSTDMT_initCStream_advanced()
Must use the new ZSTD_compressBegin_usingCDict_advanced()
to enforce correct frame parameters
2017-04-18 22:52:41 -07:00
20d5e03893 content size is controlled at bufferless level
so it's active for all entry points

Also : added relevant test (wrong content size) in fuzzer
2017-04-11 18:34:02 -07:00
30c7698970 optimize ZSTDMT_compress() memory usage
does no longer allocate temporary buffers
when there is enough room in dstBuffer to decompress directly there.
(previous method would skip that for 1st chunk only).

Also : fix ZSTD_compressBound() for small srcSize
2017-03-31 18:27:03 -07:00
eea7858e2b fixed minor warnings in debug code 2017-03-30 16:47:19 -07:00
34cc487d05 overlap at full windowSize for max compression level
as it provides max compression ratio
2017-03-30 16:23:22 -07:00
458e955c23 improved ZSTDMT_compress()
Use a bit more threads by default.
Uses overlap segments to boost compression ratio (like the streaming variant)
2017-03-30 15:51:58 -07:00
ca5a8bbe36 re-added patch ... 2017-03-29 17:15:27 -07:00
4bcc69b761 solves warnings when compiling with global XXH_STATIC_LINKING_ONLY
XXH_STATIC_LINKING_ONLY protection macro is intended to be triggered just before the include.
The main idea is to keep this setting local :
user module shall explicitly understand and accept the static linking restriction
which becomes transparent when triggering the macro at project level.
Global definition also triggers redefinition warnings for user modules which do locally define the macro.

This new version compiles lib and cli without warning when the macro is set globally.
That's not a scenario to be recommended, since it trades a local effect for a global one,
but it was easy enough to provide from zstd side.
2017-03-01 11:33:25 -08:00
14312d833e zstdmt : fix : loading prefix from previous segments
There used to be a (very small) chance that
loading prefix from previous segment
would be confused with a real zstd dictionary.
For that to happen, the prefix needs to start
with the same value as dictionary magic.
That's 1 chance in 4 billions if all values have equal probability.
But in fact, since some values are more common (0x00000000 for example)
others are less common, and dictionary magic was selected to be one of them,
so probabilities are likely even lower.

Anyway, this risk is no down to zero
by adding a new CCtx parameter : ZSTD_p_forceRawDict

Current parameter policy : the parameter "stick" to its CCtx,
so any dictionary loading after ZSTD_p_forceRawDict is set
will be loaded in "raw" ("content only") mode,
even if CCtx is re-used multiple times with multiple different dictionary.
It's up to the user to reset this value differently if it needs so.
2017-02-23 23:42:12 -08:00
831b4890ce minor tests/Makefile refactoring
and update of zstd_manual,html
2017-02-23 23:09:10 -08:00
48bed91606 Merge pull request #527 from facebook/zstdmt
zstdmt refinements
2017-01-31 16:36:46 -08:00
b2e1b3d670 fixed overlapLog==0 => no overlap 2017-01-30 14:54:46 -08:00
3672d06d06 zstdmt : section size is set to be a minimum of overlapSize
the minimum size condition size is applied transparently (no warning, no error)
like previous minimum section size condition (1 KB) which still applies.
2017-01-30 13:35:45 -08:00