diff --git a/build/VS2008/fullbench/fullbench.vcproj b/build/VS2008/fullbench/fullbench.vcproj index cd9391ec..66ac8223 100644 --- a/build/VS2008/fullbench/fullbench.vcproj +++ b/build/VS2008/fullbench/fullbench.vcproj @@ -510,6 +510,10 @@ RelativePath="..\..\..\lib\compress\zstd_compress_sequences.h" > + + diff --git a/build/VS2008/fuzzer/fuzzer.vcproj b/build/VS2008/fuzzer/fuzzer.vcproj index 5f10bbe7..a9008a67 100644 --- a/build/VS2008/fuzzer/fuzzer.vcproj +++ b/build/VS2008/fuzzer/fuzzer.vcproj @@ -546,6 +546,10 @@ RelativePath="..\..\..\lib\compress\zstd_compress_sequences.h" > + + diff --git a/build/VS2008/zstd/zstd.vcproj b/build/VS2008/zstd/zstd.vcproj index 34ccb6f4..4d75a9d3 100644 --- a/build/VS2008/zstd/zstd.vcproj +++ b/build/VS2008/zstd/zstd.vcproj @@ -626,6 +626,10 @@ RelativePath="..\..\..\lib\compress\zstd_compress_sequences.h" > + + diff --git a/build/VS2008/zstdlib/zstdlib.vcproj b/build/VS2008/zstdlib/zstdlib.vcproj index be46d351..100a501a 100644 --- a/build/VS2008/zstdlib/zstdlib.vcproj +++ b/build/VS2008/zstdlib/zstdlib.vcproj @@ -558,6 +558,10 @@ RelativePath="..\..\..\lib\compress\zstd_compress_sequences.h" > + + diff --git a/build/VS2010/fullbench/fullbench.vcxproj b/build/VS2010/fullbench/fullbench.vcxproj index 105c692b..4597239b 100644 --- a/build/VS2010/fullbench/fullbench.vcxproj +++ b/build/VS2010/fullbench/fullbench.vcxproj @@ -197,6 +197,7 @@ + diff --git a/build/VS2010/fuzzer/fuzzer.vcxproj b/build/VS2010/fuzzer/fuzzer.vcxproj index 7f7e404d..a6f136ee 100644 --- a/build/VS2010/fuzzer/fuzzer.vcxproj +++ b/build/VS2010/fuzzer/fuzzer.vcxproj @@ -200,6 +200,7 @@ + diff --git a/build/VS2010/libzstd-dll/libzstd-dll.vcxproj b/build/VS2010/libzstd-dll/libzstd-dll.vcxproj index 840d1ec3..50c4e817 100644 --- a/build/VS2010/libzstd-dll/libzstd-dll.vcxproj +++ b/build/VS2010/libzstd-dll/libzstd-dll.vcxproj @@ -82,6 +82,7 @@ + diff --git a/build/VS2010/libzstd/libzstd.vcxproj b/build/VS2010/libzstd/libzstd.vcxproj index b84d45b9..c4e82877 100644 --- a/build/VS2010/libzstd/libzstd.vcxproj +++ b/build/VS2010/libzstd/libzstd.vcxproj @@ -82,6 +82,7 @@ + diff --git a/build/VS2010/zstd/zstd.vcxproj b/build/VS2010/zstd/zstd.vcxproj index d65f55b3..1058d229 100644 --- a/build/VS2010/zstd/zstd.vcxproj +++ b/build/VS2010/zstd/zstd.vcxproj @@ -79,6 +79,7 @@ + diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c index 62a407bb..df47ac78 100644 --- a/lib/compress/zstd_compress.c +++ b/lib/compress/zstd_compress.c @@ -42,11 +42,10 @@ size_t ZSTD_compressBound(size_t srcSize) { * Context memory management ***************************************/ struct ZSTD_CDict_s { - void* dictBuffer; const void* dictContent; size_t dictContentSize; - void* workspace; - size_t workspaceSize; + U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */ + ZSTD_cwksp workspace; ZSTD_matchState_t matchState; ZSTD_compressedBlockState_t cBlockState; ZSTD_customMem customMem; @@ -84,23 +83,26 @@ ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem) ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize) { - ZSTD_CCtx* const cctx = (ZSTD_CCtx*) workspace; + ZSTD_cwksp ws; + ZSTD_CCtx* cctx; if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL; /* minimum size */ if ((size_t)workspace & 7) return NULL; /* must be 8-aligned */ - memset(workspace, 0, workspaceSize); /* may be a bit generous, could memset be smaller ? */ + ZSTD_cwksp_init(&ws, workspace, workspaceSize); + + cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx)); + if (cctx == NULL) { + return NULL; + } + memset(cctx, 0, sizeof(ZSTD_CCtx)); + ZSTD_cwksp_move(&cctx->workspace, &ws); cctx->staticSize = workspaceSize; - cctx->workSpace = (void*)(cctx+1); - cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx); /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */ - if (cctx->workSpaceSize < HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t)) return NULL; - assert(((size_t)cctx->workSpace & (sizeof(void*)-1)) == 0); /* ensure correct alignment */ - cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)cctx->workSpace; - cctx->blockState.nextCBlock = cctx->blockState.prevCBlock + 1; - { - void* const ptr = cctx->blockState.nextCBlock + 1; - cctx->entropyWorkspace = (U32*)ptr; - } + if (!ZSTD_cwksp_check_available(&cctx->workspace, HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL; + cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t)); + cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t)); + cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object( + &cctx->workspace, HUF_WORKSPACE_SIZE); cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid()); return cctx; } @@ -128,7 +130,11 @@ static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx) { assert(cctx != NULL); assert(cctx->staticSize == 0); - ZSTD_free(cctx->workSpace, cctx->customMem); cctx->workSpace = NULL; + /* Only free workspace if cctx not in workspace, otherwise the workspace + * will be freed when the cctx itself is freed. */ + if ((void*)cctx->workspace.workspace != (void*)cctx) { + ZSTD_cwksp_free(&cctx->workspace, cctx->customMem); + } ZSTD_clearAllDicts(cctx); #ifdef ZSTD_MULTITHREAD ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL; @@ -160,7 +166,9 @@ static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx) size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx) { if (cctx==NULL) return 0; /* support sizeof on NULL */ - return sizeof(*cctx) + cctx->workSpaceSize + /* cctx may be in the workspace */ + return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx)) + + ZSTD_cwksp_sizeof(&cctx->workspace) + ZSTD_sizeof_localDict(cctx->localDict) + ZSTD_sizeof_mtctx(cctx); } @@ -1101,7 +1109,7 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params) matchStateSize + ldmSpace + ldmSeqSpace; DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)sizeof(ZSTD_CCtx)); - DEBUGLOG(5, "estimate workSpace : %u", (U32)neededSpace); + DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace); return sizeof(ZSTD_CCtx) + neededSpace; } } @@ -1355,9 +1363,9 @@ typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e; typedef enum { ZSTD_resetTarget_CDict, ZSTD_resetTarget_CCtx } ZSTD_resetTarget_e; -static void* +static size_t ZSTD_reset_matchState(ZSTD_matchState_t* ms, - void* ptr, + ZSTD_cwksp* ws, const ZSTD_compressionParameters* cParams, ZSTD_compResetPolicy_e const crp, ZSTD_resetTarget_e const forWho) { @@ -1365,9 +1373,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms, size_t const hSize = ((size_t)1) << cParams->hashLog; U32 const hashLog3 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0; size_t const h3Size = ((size_t)1) << hashLog3; - size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32); - assert(((size_t)ptr & 3) == 0); ms->hashLog3 = hashLog3; memset(&ms->window, 0, sizeof(ms->window)); @@ -1376,33 +1382,43 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms, ms->window.nextSrc = ms->window.base + 1; /* see issue #1241 */ ZSTD_invalidateMatchState(ms); + assert(!ZSTD_cwksp_reserve_failed(ws)); /* check that allocation hasn't already failed */ + + ZSTD_cwksp_clear_tables(ws); + + DEBUGLOG(5, "reserving table space"); + /* table Space */ + ms->hashTable = (U32*)ZSTD_cwksp_reserve_table(ws, hSize * sizeof(U32)); + ms->chainTable = (U32*)ZSTD_cwksp_reserve_table(ws, chainSize * sizeof(U32)); + ms->hashTable3 = (U32*)ZSTD_cwksp_reserve_table(ws, h3Size * sizeof(U32)); + RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation, + "failed a workspace allocation in ZSTD_reset_matchState"); + + DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_noMemset); + if (crp!=ZSTDcrp_noMemset) { + /* reset tables only */ + memset(ms->hashTable, 0, hSize * sizeof(U32)); + memset(ms->chainTable, 0, chainSize * sizeof(U32)); + memset(ms->hashTable3, 0, h3Size * sizeof(U32)); + } + /* opt parser space */ if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) { DEBUGLOG(4, "reserving optimal parser space"); - ms->opt.litFreq = (unsigned*)ptr; - ms->opt.litLengthFreq = ms->opt.litFreq + (1<opt.matchLengthFreq = ms->opt.litLengthFreq + (MaxLL+1); - ms->opt.offCodeFreq = ms->opt.matchLengthFreq + (MaxML+1); - ptr = ms->opt.offCodeFreq + (MaxOff+1); - ms->opt.matchTable = (ZSTD_match_t*)ptr; - ptr = ms->opt.matchTable + ZSTD_OPT_NUM+1; - ms->opt.priceTable = (ZSTD_optimal_t*)ptr; - ptr = ms->opt.priceTable + ZSTD_OPT_NUM+1; + ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned)); + ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned)); + ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned)); + ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t)); + ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t)); } - /* table Space */ - DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_noMemset); - assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */ - if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace); /* reset tables only */ - ms->hashTable = (U32*)(ptr); - ms->chainTable = ms->hashTable + hSize; - ms->hashTable3 = ms->chainTable + chainSize; - ptr = ms->hashTable3 + h3Size; - ms->cParams = *cParams; - assert(((size_t)ptr & 3) == 0); - return ptr; + RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation, + "failed a workspace allocation in ZSTD_reset_matchState"); + + return 0; } /* ZSTD_indexTooCloseToMax() : @@ -1418,13 +1434,6 @@ static int ZSTD_indexTooCloseToMax(ZSTD_window_t w) return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN); } -#define ZSTD_WORKSPACETOOLARGE_FACTOR 3 /* define "workspace is too large" as this number of times larger than needed */ -#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128 /* when workspace is continuously too large - * during at least this number of times, - * context's memory usage is considered wasteful, - * because it's sized to handle a worst case scenario which rarely happens. - * In which case, resize it down to free some memory */ - /*! ZSTD_resetCCtx_internal() : note : `params` are assumed fully validated at this stage */ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc, @@ -1433,6 +1442,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc, ZSTD_compResetPolicy_e const crp, ZSTD_buffered_policy_e const zbuff) { + ZSTD_cwksp* const ws = &zc->workspace; DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u", (U32)pledgedSrcSize, params.cParams.windowLog); assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams))); @@ -1444,16 +1454,18 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc, zc->seqStore.maxNbSeq, zc->seqStore.maxNbLit, zbuff, pledgedSrcSize) ) { DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> consider continue mode"); - zc->workSpaceOversizedDuration += (zc->workSpaceOversizedDuration > 0); /* if it was too large, it still is */ - if (zc->workSpaceOversizedDuration <= ZSTD_WORKSPACETOOLARGE_MAXDURATION) { + ZSTD_cwksp_bump_oversized_duration(ws, 0); + if (!ZSTD_cwksp_check_wasteful(ws, 0)) { DEBUGLOG(4, "continue mode confirmed (wLog1=%u, blockSize1=%zu)", zc->appliedParams.cParams.windowLog, zc->blockSize); if (ZSTD_indexTooCloseToMax(zc->blockState.matchState.window)) { /* prefer a reset, faster than a rescale */ - ZSTD_reset_matchState(&zc->blockState.matchState, - zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32, - ¶ms.cParams, - crp, ZSTD_resetTarget_CCtx); + FORWARD_IF_ERROR(ZSTD_reset_matchState( + &zc->blockState.matchState, + ws, + ¶ms.cParams, + crp, + ZSTD_resetTarget_CCtx)); } return ZSTD_continueCCtx(zc, ¶ms, pledgedSrcSize); } } } @@ -1476,53 +1488,57 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc, size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0; size_t const matchStateSize = ZSTD_sizeof_matchState(¶ms.cParams, /* forCCtx */ 1); size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize); - void* ptr; /* used to partition workSpace */ - /* Check if workSpace is large enough, alloc a new one if needed */ - { size_t const entropySpace = HUF_WORKSPACE_SIZE; + /* Check if workspace is large enough, alloc a new one if needed */ + { size_t const cctxSpace = zc->staticSize ? sizeof(ZSTD_CCtx) : 0; + size_t const entropySpace = HUF_WORKSPACE_SIZE; size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t); size_t const bufferSpace = buffInSize + buffOutSize; size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams); size_t const ldmSeqSpace = maxNbLdmSeq * sizeof(rawSeq); - size_t const neededSpace = entropySpace + blockStateSpace + ldmSpace + - ldmSeqSpace + matchStateSize + tokenSpace + - bufferSpace; + size_t const neededSpace = + cctxSpace + + entropySpace + + blockStateSpace + + ldmSpace + + ldmSeqSpace + + matchStateSize + + tokenSpace + + bufferSpace; - int const workSpaceTooSmall = zc->workSpaceSize < neededSpace; - int const workSpaceTooLarge = zc->workSpaceSize > ZSTD_WORKSPACETOOLARGE_FACTOR * neededSpace; - int const workSpaceWasteful = workSpaceTooLarge && (zc->workSpaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION); - zc->workSpaceOversizedDuration = workSpaceTooLarge ? zc->workSpaceOversizedDuration+1 : 0; + int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace; + int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace); DEBUGLOG(4, "Need %zuKB workspace, including %zuKB for match state, and %zuKB for buffers", neededSpace>>10, matchStateSize>>10, bufferSpace>>10); DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize); - if (workSpaceTooSmall || workSpaceWasteful) { - DEBUGLOG(4, "Resize workSpaceSize from %zuKB to %zuKB", - zc->workSpaceSize >> 10, + if (workspaceTooSmall || workspaceWasteful) { + DEBUGLOG(4, "Resize workspaceSize from %zuKB to %zuKB", + ZSTD_cwksp_sizeof(ws) >> 10, neededSpace >> 10); RETURN_ERROR_IF(zc->staticSize, memory_allocation, "static cctx : no resize"); - zc->workSpaceSize = 0; - ZSTD_free(zc->workSpace, zc->customMem); - zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem); - RETURN_ERROR_IF(zc->workSpace == NULL, memory_allocation); - zc->workSpaceSize = neededSpace; - zc->workSpaceOversizedDuration = 0; + ZSTD_cwksp_free(ws, zc->customMem); + FORWARD_IF_ERROR(ZSTD_cwksp_create(ws, neededSpace, zc->customMem)); + DEBUGLOG(5, "reserving object space"); /* Statically sized space. * entropyWorkspace never moves, * though prev/next block swap places */ - assert(((size_t)zc->workSpace & 3) == 0); /* ensure correct alignment */ - assert(zc->workSpaceSize >= 2 * sizeof(ZSTD_compressedBlockState_t)); - zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)zc->workSpace; - zc->blockState.nextCBlock = zc->blockState.prevCBlock + 1; - ptr = zc->blockState.nextCBlock + 1; - zc->entropyWorkspace = (U32*)ptr; + assert(ZSTD_cwksp_check_available(ws, 2 * sizeof(ZSTD_compressedBlockState_t))); + zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t)); + RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock"); + zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t)); + RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock"); + zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, HUF_WORKSPACE_SIZE); + RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate entropyWorkspace"); } } + ZSTD_cwksp_clear(ws); + /* init params */ zc->appliedParams = params; zc->blockState.matchState.cParams = params.cParams; @@ -1541,58 +1557,55 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc, ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock); - ptr = ZSTD_reset_matchState(&zc->blockState.matchState, - zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32, - ¶ms.cParams, - crp, ZSTD_resetTarget_CCtx); - - /* ldm hash table */ - /* initialize bucketOffsets table later for pointer alignment */ - if (params.ldmParams.enableLdm) { - size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog; - memset(ptr, 0, ldmHSize * sizeof(ldmEntry_t)); - assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */ - zc->ldmState.hashTable = (ldmEntry_t*)ptr; - ptr = zc->ldmState.hashTable + ldmHSize; - zc->ldmSequences = (rawSeq*)ptr; - ptr = zc->ldmSequences + maxNbLdmSeq; - zc->maxNbLdmSequences = maxNbLdmSeq; - - memset(&zc->ldmState.window, 0, sizeof(zc->ldmState.window)); - } - assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */ - - /* sequences storage */ - zc->seqStore.maxNbSeq = maxNbSeq; - zc->seqStore.sequencesStart = (seqDef*)ptr; - ptr = zc->seqStore.sequencesStart + maxNbSeq; - zc->seqStore.llCode = (BYTE*) ptr; - zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq; - zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq; - zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq; /* ZSTD_wildcopy() is used to copy into the literals buffer, * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes. */ + zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH); zc->seqStore.maxNbLit = blockSize; - ptr = zc->seqStore.litStart + blockSize + WILDCOPY_OVERLENGTH; + + /* buffers */ + zc->inBuffSize = buffInSize; + zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize); + zc->outBuffSize = buffOutSize; + zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize); /* ldm bucketOffsets table */ if (params.ldmParams.enableLdm) { size_t const ldmBucketSize = ((size_t)1) << (params.ldmParams.hashLog - params.ldmParams.bucketSizeLog); - memset(ptr, 0, ldmBucketSize); - zc->ldmState.bucketOffsets = (BYTE*)ptr; - ptr = zc->ldmState.bucketOffsets + ldmBucketSize; + zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, ldmBucketSize); + memset(zc->ldmState.bucketOffsets, 0, ldmBucketSize); + } + + /* sequences storage */ + ZSTD_referenceExternalSequences(zc, NULL, 0); + zc->seqStore.maxNbSeq = maxNbSeq; + zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE)); + zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE)); + zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE)); + zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef)); + + FORWARD_IF_ERROR(ZSTD_reset_matchState( + &zc->blockState.matchState, + ws, + ¶ms.cParams, + crp, ZSTD_resetTarget_CCtx)); + + /* ldm hash table */ + /* initialize bucketOffsets table separately for pointer alignment */ + if (params.ldmParams.enableLdm) { + size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog; + zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t)); + memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t)); + zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq)); + zc->maxNbLdmSequences = maxNbLdmSeq; + + memset(&zc->ldmState.window, 0, sizeof(zc->ldmState.window)); ZSTD_window_clear(&zc->ldmState.window); } - ZSTD_referenceExternalSequences(zc, NULL, 0); - /* buffers */ - zc->inBuffSize = buffInSize; - zc->inBuff = (char*)ptr; - zc->outBuffSize = buffOutSize; - zc->outBuff = zc->inBuff + buffInSize; + DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws)); return 0; } @@ -1652,8 +1665,8 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx, * has its own tables. */ params.cParams = ZSTD_adjustCParams_internal(*cdict_cParams, pledgedSrcSize, 0); params.cParams.windowLog = windowLog; - ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize, - ZSTDcrp_continue, zbuff); + FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize, + ZSTDcrp_continue, zbuff)); assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy); } @@ -1701,8 +1714,8 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx, /* Copy only compression parameters related to tables. */ params.cParams = *cdict_cParams; params.cParams.windowLog = windowLog; - ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize, - ZSTDcrp_noMemset, zbuff); + FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize, + ZSTDcrp_noMemset, zbuff)); assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy); assert(cctx->appliedParams.cParams.hashLog == cdict_cParams->hashLog); assert(cctx->appliedParams.cParams.chainLog == cdict_cParams->chainLog); @@ -1960,7 +1973,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, ZSTD_entropyCTables_t* nextEntropy, const ZSTD_CCtx_params* cctxParams, void* dst, size_t dstCapacity, - void* workspace, size_t wkspSize, + void* entropyWorkspace, size_t entropyWkspSize, const int bmi2) { const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN; @@ -1993,7 +2006,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, ZSTD_disableLiteralsCompression(cctxParams), op, dstCapacity, literals, litSize, - workspace, wkspSize, + entropyWorkspace, entropyWkspSize, bmi2); FORWARD_IF_ERROR(cSize); assert(cSize <= dstCapacity); @@ -2029,7 +2042,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, ZSTD_seqToCodes(seqStorePtr); /* build CTable for Literal Lengths */ { unsigned max = MaxLL; - size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace, wkspSize); /* can't fail */ + size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */ DEBUGLOG(5, "Building LL table"); nextEntropy->fse.litlength_repeatMode = prevEntropy->fse.litlength_repeatMode; LLtype = ZSTD_selectEncodingType(&nextEntropy->fse.litlength_repeatMode, @@ -2039,10 +2052,14 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, ZSTD_defaultAllowed, strategy); assert(set_basic < set_compressed && set_rle < set_compressed); assert(!(LLtype < set_compressed && nextEntropy->fse.litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */ - { size_t const countSize = ZSTD_buildCTable(op, (size_t)(oend - op), CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype, - count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL, - prevEntropy->fse.litlengthCTable, sizeof(prevEntropy->fse.litlengthCTable), - workspace, wkspSize); + { size_t const countSize = ZSTD_buildCTable( + op, (size_t)(oend - op), + CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype, + count, max, llCodeTable, nbSeq, + LL_defaultNorm, LL_defaultNormLog, MaxLL, + prevEntropy->fse.litlengthCTable, + sizeof(prevEntropy->fse.litlengthCTable), + entropyWorkspace, entropyWkspSize); FORWARD_IF_ERROR(countSize); if (LLtype == set_compressed) lastNCount = op; @@ -2051,7 +2068,8 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, } } /* build CTable for Offsets */ { unsigned max = MaxOff; - size_t const mostFrequent = HIST_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace, wkspSize); /* can't fail */ + size_t const mostFrequent = HIST_countFast_wksp( + count, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */ /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */ ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed; DEBUGLOG(5, "Building OF table"); @@ -2062,10 +2080,14 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, OF_defaultNorm, OF_defaultNormLog, defaultPolicy, strategy); assert(!(Offtype < set_compressed && nextEntropy->fse.offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */ - { size_t const countSize = ZSTD_buildCTable(op, (size_t)(oend - op), CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype, - count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff, - prevEntropy->fse.offcodeCTable, sizeof(prevEntropy->fse.offcodeCTable), - workspace, wkspSize); + { size_t const countSize = ZSTD_buildCTable( + op, (size_t)(oend - op), + CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype, + count, max, ofCodeTable, nbSeq, + OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff, + prevEntropy->fse.offcodeCTable, + sizeof(prevEntropy->fse.offcodeCTable), + entropyWorkspace, entropyWkspSize); FORWARD_IF_ERROR(countSize); if (Offtype == set_compressed) lastNCount = op; @@ -2074,7 +2096,8 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, } } /* build CTable for MatchLengths */ { unsigned max = MaxML; - size_t const mostFrequent = HIST_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace, wkspSize); /* can't fail */ + size_t const mostFrequent = HIST_countFast_wksp( + count, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */ DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op)); nextEntropy->fse.matchlength_repeatMode = prevEntropy->fse.matchlength_repeatMode; MLtype = ZSTD_selectEncodingType(&nextEntropy->fse.matchlength_repeatMode, @@ -2083,10 +2106,14 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr, ML_defaultNorm, ML_defaultNormLog, ZSTD_defaultAllowed, strategy); assert(!(MLtype < set_compressed && nextEntropy->fse.matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */ - { size_t const countSize = ZSTD_buildCTable(op, (size_t)(oend - op), CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype, - count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML, - prevEntropy->fse.matchlengthCTable, sizeof(prevEntropy->fse.matchlengthCTable), - workspace, wkspSize); + { size_t const countSize = ZSTD_buildCTable( + op, (size_t)(oend - op), + CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype, + count, max, mlCodeTable, nbSeq, + ML_defaultNorm, ML_defaultNormLog, MaxML, + prevEntropy->fse.matchlengthCTable, + sizeof(prevEntropy->fse.matchlengthCTable), + entropyWorkspace, entropyWkspSize); FORWARD_IF_ERROR(countSize); if (MLtype == set_compressed) lastNCount = op; @@ -2134,13 +2161,13 @@ ZSTD_compressSequences(seqStore_t* seqStorePtr, const ZSTD_CCtx_params* cctxParams, void* dst, size_t dstCapacity, size_t srcSize, - void* workspace, size_t wkspSize, + void* entropyWorkspace, size_t entropyWkspSize, int bmi2) { size_t const cSize = ZSTD_compressSequences_internal( seqStorePtr, prevEntropy, nextEntropy, cctxParams, dst, dstCapacity, - workspace, wkspSize, bmi2); + entropyWorkspace, entropyWkspSize, bmi2); if (cSize == 0) return 0; /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block. * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block. @@ -2380,6 +2407,7 @@ static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms, ZSTD_CCtx_params } } + /*! ZSTD_compress_frameChunk() : * Compress a chunk of data into one or multiple blocks. * All blocks will be terminated, all input will be consumed. @@ -2848,7 +2876,8 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx, ZSTDcrp_continue, zbuff) ); { size_t const dictID = ZSTD_compress_insertDictionary( cctx->blockState.prevCBlock, &cctx->blockState.matchState, - params, dict, dictSize, dictContentType, dtlm, cctx->entropyWorkspace); + params, dict, dictSize, dictContentType, dtlm, + cctx->entropyWorkspace); FORWARD_IF_ERROR(dictID); assert(dictID <= UINT_MAX); cctx->dictID = (U32)dictID; @@ -3063,7 +3092,7 @@ size_t ZSTD_estimateCDictSize_advanced( { DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (unsigned)sizeof(ZSTD_CDict)); return sizeof(ZSTD_CDict) + HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0) - + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize); + + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : ZSTD_cwksp_align(dictSize, sizeof(void *))); } size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel) @@ -3076,7 +3105,9 @@ size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict) { if (cdict==NULL) return 0; /* support sizeof on NULL */ DEBUGLOG(5, "sizeof(*cdict) : %u", (unsigned)sizeof(*cdict)); - return cdict->workspaceSize + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict); + /* cdict may be in the workspace */ + return (cdict->workspace.workspace == cdict ? 0 : sizeof(*cdict)) + + ZSTD_cwksp_sizeof(&cdict->workspace); } static size_t ZSTD_initCDict_internal( @@ -3090,26 +3121,25 @@ static size_t ZSTD_initCDict_internal( assert(!ZSTD_checkCParams(cParams)); cdict->matchState.cParams = cParams; if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) { - cdict->dictBuffer = NULL; cdict->dictContent = dictBuffer; } else { - void* const internalBuffer = ZSTD_malloc(dictSize, cdict->customMem); - cdict->dictBuffer = internalBuffer; - cdict->dictContent = internalBuffer; + void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*))); RETURN_ERROR_IF(!internalBuffer, memory_allocation); + cdict->dictContent = internalBuffer; memcpy(internalBuffer, dictBuffer, dictSize); } cdict->dictContentSize = dictSize; + cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE); + + /* Reset the state to no dictionary */ ZSTD_reset_compressedBlockState(&cdict->cBlockState); - { void* const end = ZSTD_reset_matchState(&cdict->matchState, - (U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32, - &cParams, - ZSTDcrp_continue, ZSTD_resetTarget_CDict); - assert(end == (char*)cdict->workspace + cdict->workspaceSize); - (void)end; - } + FORWARD_IF_ERROR(ZSTD_reset_matchState( + &cdict->matchState, + &cdict->workspace, + &cParams, + ZSTDcrp_continue, ZSTD_resetTarget_CDict)); /* (Maybe) load the dictionary * Skips loading the dictionary if it is <= 8 bytes. */ @@ -3121,7 +3151,7 @@ static size_t ZSTD_initCDict_internal( { size_t const dictID = ZSTD_compress_insertDictionary( &cdict->cBlockState, &cdict->matchState, ¶ms, cdict->dictContent, cdict->dictContentSize, - dictContentType, ZSTD_dtlm_full, cdict->workspace); + dictContentType, ZSTD_dtlm_full, cdict->entropyWorkspace); FORWARD_IF_ERROR(dictID); assert(dictID <= (size_t)(U32)-1); cdict->dictID = (U32)dictID; @@ -3139,18 +3169,27 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize, DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (unsigned)dictContentType); if (!customMem.customAlloc ^ !customMem.customFree) return NULL; - { ZSTD_CDict* const cdict = (ZSTD_CDict*)ZSTD_malloc(sizeof(ZSTD_CDict), customMem); - size_t const workspaceSize = HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0); + { size_t const workspaceSize = + sizeof(ZSTD_CDict) + + HUF_WORKSPACE_SIZE + + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0) + + (dictLoadMethod == ZSTD_dlm_byRef ? 0 + : ZSTD_cwksp_align(dictSize, sizeof(void*))); void* const workspace = ZSTD_malloc(workspaceSize, customMem); + ZSTD_cwksp ws; + ZSTD_CDict* cdict; - if (!cdict || !workspace) { - ZSTD_free(cdict, customMem); + if (!workspace) { ZSTD_free(workspace, customMem); return NULL; } + + ZSTD_cwksp_init(&ws, workspace, workspaceSize); + + cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict)); + assert(cdict != NULL); + ZSTD_cwksp_move(&cdict->workspace, &ws); cdict->customMem = customMem; - cdict->workspace = workspace; - cdict->workspaceSize = workspaceSize; if (ZSTD_isError( ZSTD_initCDict_internal(cdict, dictBuffer, dictSize, dictLoadMethod, dictContentType, @@ -3183,8 +3222,11 @@ size_t ZSTD_freeCDict(ZSTD_CDict* cdict) { if (cdict==NULL) return 0; /* support free on NULL */ { ZSTD_customMem const cMem = cdict->customMem; - ZSTD_free(cdict->workspace, cMem); - ZSTD_free(cdict->dictBuffer, cMem); + /* Only free workspace if cdict not in workspace, otherwise the + * workspace will be freed when the cdict itself is freed. */ + if ((void*)cdict->workspace.workspace != (void*)cdict) { + ZSTD_cwksp_free(&cdict->workspace, cMem); + } ZSTD_free(cdict, cMem); return 0; } @@ -3211,28 +3253,27 @@ const ZSTD_CDict* ZSTD_initStaticCDict( ZSTD_compressionParameters cParams) { size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0); - size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize) + size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : ZSTD_cwksp_align(dictSize, sizeof(void*))) + HUF_WORKSPACE_SIZE + matchStateSize; - ZSTD_CDict* const cdict = (ZSTD_CDict*) workspace; - void* ptr; + ZSTD_CDict* cdict; + if ((size_t)workspace & 7) return NULL; /* 8-aligned */ + + { + ZSTD_cwksp ws; + ZSTD_cwksp_init(&ws, workspace, workspaceSize); + cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict)); + if (cdict == NULL) return NULL; + ZSTD_cwksp_move(&cdict->workspace, &ws); + } + DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u", (unsigned)workspaceSize, (unsigned)neededSize, (unsigned)(workspaceSize < neededSize)); if (workspaceSize < neededSize) return NULL; - if (dictLoadMethod == ZSTD_dlm_byCopy) { - memcpy(cdict+1, dict, dictSize); - dict = cdict+1; - ptr = (char*)workspace + sizeof(ZSTD_CDict) + dictSize; - } else { - ptr = cdict+1; - } - cdict->workspace = ptr; - cdict->workspaceSize = HUF_WORKSPACE_SIZE + matchStateSize; - if (ZSTD_isError( ZSTD_initCDict_internal(cdict, dict, dictSize, - ZSTD_dlm_byRef, dictContentType, + dictLoadMethod, dictContentType, cParams) )) return NULL; @@ -3393,7 +3434,8 @@ size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pss) * Assumption 2 : either dict, or cdict, is defined, not both */ size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs, const void* dict, size_t dictSize, const ZSTD_CDict* cdict, - const ZSTD_CCtx_params* params, unsigned long long pledgedSrcSize) + const ZSTD_CCtx_params* params, + unsigned long long pledgedSrcSize) { DEBUGLOG(4, "ZSTD_initCStream_internal"); FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) ); diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h index 262a6309..ae106e02 100644 --- a/lib/compress/zstd_compress_internal.h +++ b/lib/compress/zstd_compress_internal.h @@ -19,6 +19,7 @@ * Dependencies ***************************************/ #include "zstd_internal.h" +#include "zstd_cwksp.h" #ifdef ZSTD_MULTITHREAD # include "zstdmt_compress.h" #endif @@ -231,9 +232,7 @@ struct ZSTD_CCtx_s { ZSTD_CCtx_params appliedParams; U32 dictID; - int workSpaceOversizedDuration; - void* workSpace; - size_t workSpaceSize; + ZSTD_cwksp workspace; /* manages buffer for dynamic allocations */ size_t blockSize; unsigned long long pledgedSrcSizePlusOne; /* this way, 0 (default) == unknown */ unsigned long long consumedSrcSize; diff --git a/lib/compress/zstd_compress_literals.c b/lib/compress/zstd_compress_literals.c index eb3e5a44..6c133311 100644 --- a/lib/compress/zstd_compress_literals.c +++ b/lib/compress/zstd_compress_literals.c @@ -70,7 +70,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf, ZSTD_strategy strategy, int disableLiteralCompression, void* dst, size_t dstCapacity, const void* src, size_t srcSize, - void* workspace, size_t wkspSize, + void* entropyWorkspace, size_t entropyWorkspaceSize, const int bmi2) { size_t const minGain = ZSTD_minGain(srcSize, strategy); @@ -99,10 +99,15 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf, { HUF_repeat repeat = prevHuf->repeatMode; int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0; if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1; - cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, - workspace, wkspSize, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) - : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, - workspace, wkspSize, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2); + cLitSize = singleStream ? + HUF_compress1X_repeat( + ostart+lhSize, dstCapacity-lhSize, src, srcSize, + 255, 11, entropyWorkspace, entropyWorkspaceSize, + (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) : + HUF_compress4X_repeat( + ostart+lhSize, dstCapacity-lhSize, src, srcSize, + 255, 11, entropyWorkspace, entropyWorkspaceSize, + (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2); if (repeat != HUF_repeat_none) { /* reused the existing table */ hType = set_repeat; diff --git a/lib/compress/zstd_compress_literals.h b/lib/compress/zstd_compress_literals.h index 7adbecc0..97273d7c 100644 --- a/lib/compress/zstd_compress_literals.h +++ b/lib/compress/zstd_compress_literals.h @@ -23,7 +23,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf, ZSTD_strategy strategy, int disableLiteralCompression, void* dst, size_t dstCapacity, const void* src, size_t srcSize, - void* workspace, size_t wkspSize, + void* entropyWorkspace, size_t entropyWorkspaceSize, const int bmi2); #endif /* ZSTD_COMPRESS_LITERALS_H */ diff --git a/lib/compress/zstd_compress_sequences.c b/lib/compress/zstd_compress_sequences.c index 3c3deae0..0ff7a268 100644 --- a/lib/compress/zstd_compress_sequences.c +++ b/lib/compress/zstd_compress_sequences.c @@ -222,7 +222,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity, const BYTE* codeTable, size_t nbSeq, const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax, const FSE_CTable* prevCTable, size_t prevCTableSize, - void* workspace, size_t workspaceSize) + void* entropyWorkspace, size_t entropyWorkspaceSize) { BYTE* op = (BYTE*)dst; const BYTE* const oend = op + dstCapacity; @@ -238,7 +238,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity, memcpy(nextCTable, prevCTable, prevCTableSize); return 0; case set_basic: - FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize)); /* note : could be pre-calculated */ + FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize)); /* note : could be pre-calculated */ return 0; case set_compressed: { S16 norm[MaxSeq + 1]; @@ -252,7 +252,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity, FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max)); { size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog); /* overflow protected */ FORWARD_IF_ERROR(NCountSize); - FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, workspace, workspaceSize)); + FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize)); return NCountSize; } } diff --git a/lib/compress/zstd_compress_sequences.h b/lib/compress/zstd_compress_sequences.h index f5234d94..57e8e367 100644 --- a/lib/compress/zstd_compress_sequences.h +++ b/lib/compress/zstd_compress_sequences.h @@ -35,7 +35,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity, const BYTE* codeTable, size_t nbSeq, const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax, const FSE_CTable* prevCTable, size_t prevCTableSize, - void* workspace, size_t workspaceSize); + void* entropyWorkspace, size_t entropyWorkspaceSize); size_t ZSTD_encodeSequences( void* dst, size_t dstCapacity, diff --git a/lib/compress/zstd_cwksp.h b/lib/compress/zstd_cwksp.h new file mode 100644 index 00000000..76ace25e --- /dev/null +++ b/lib/compress/zstd_cwksp.h @@ -0,0 +1,369 @@ +/* + * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +#ifndef ZSTD_CWKSP_H +#define ZSTD_CWKSP_H + +/*-************************************* +* Dependencies +***************************************/ +#include "zstd_internal.h" + +#if defined (__cplusplus) +extern "C" { +#endif + +/*-************************************* +* Constants +***************************************/ + +/* define "workspace is too large" as this number of times larger than needed */ +#define ZSTD_WORKSPACETOOLARGE_FACTOR 3 + +/* when workspace is continuously too large + * during at least this number of times, + * context's memory usage is considered wasteful, + * because it's sized to handle a worst case scenario which rarely happens. + * In which case, resize it down to free some memory */ +#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128 + +/*-************************************* +* Structures +***************************************/ +typedef enum { + ZSTD_cwksp_alloc_objects, + ZSTD_cwksp_alloc_buffers, + ZSTD_cwksp_alloc_aligned +} ZSTD_cwksp_alloc_phase_e; + +/** + * Zstd fits all its internal datastructures into a single continuous buffer, + * so that it only needs to perform a single OS allocation (or so that a buffer + * can be provided to it and it can perform no allocations at all). This buffer + * is called the workspace. + * + * Several optimizations complicate that process of allocating memory ranges + * from this workspace for each internal datastructure: + * + * - These different internal datastructures have different setup requirements: + * + * - The static objects need to be cleared once and can then be trivially + * reused for each compression. + * + * - Various buffers don't need to be initialized at all--they are always + * written into before they're read. + * + * - The matchstate tables have a unique requirement that they don't need + * their memory to be totally cleared, but they do need the memory to have + * some bound, i.e., a guarantee that all values in the memory they've been + * allocated is less than some maximum value (which is the starting value + * for the indices that they will then use for compression). When this + * guarantee is provided to them, they can use the memory without any setup + * work. When it can't, they have to clear the area. + * + * - These buffers also have different alignment requirements. + * + * - We would like to reuse the objects in the workspace for multiple + * compressions without having to perform any expensive reallocation or + * reinitialization work. + * + * - We would like to be able to efficiently reuse the workspace across + * multiple compressions **even when the compression parameters change** and + * we need to resize some of the objects (where possible). + * + * To attempt to manage this buffer, given these constraints, the ZSTD_cwksp + * abstraction was created. It works as follows: + * + * Workspace Layout: + * + * [ ... workspace ... ] + * [objects][tables ... ->] free space [<- ... aligned][<- ... buffers] + * + * The various objects that live in the workspace are divided into the + * following categories, and are allocated separately: + * + * - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict, + * so that literally everything fits in a single buffer. Note: if present, + * this must be the first object in the workspace, since ZSTD_free{CCtx, + * CDict}() rely on a pointer comparison to see whether one or two frees are + * required. + * + * - Fixed size objects: these are fixed-size, fixed-count objects that are + * nonetheless "dynamically" allocated in the workspace so that we can + * control how they're initialized separately from the broader ZSTD_CCtx. + * Examples: + * - Entropy Workspace + * - 2 x ZSTD_compressedBlockState_t + * - CDict dictionary contents + * + * - Tables: these are any of several different datastructures (hash tables, + * chain tables, binary trees) that all respect a common format: they are + * uint32_t arrays, all of whose values are between 0 and (nextSrc - base). + * Their sizes depend on the cparams. + * + * - Aligned: these buffers are used for various purposes that require 4 byte + * alignment, but don't require any initialization before they're used. + * + * - Buffers: these buffers are used for various purposes that don't require + * any alignment or initialization before they're used. This means they can + * be moved around at no cost for a new compression. + * + * Allocating Memory: + * + * The various types of objects must be allocated in order, so they can be + * correctly packed into the workspace buffer. That order is: + * + * 1. Objects + * 2. Buffers + * 3. Aligned + * 4. Tables + * + * Attempts to reserve objects of different types out of order will fail. + */ +typedef struct { + void* workspace; + void* workspaceEnd; + + void* objectEnd; + void* tableEnd; + void* allocStart; + + int allocFailed; + int workspaceOversizedDuration; + ZSTD_cwksp_alloc_phase_e phase; +} ZSTD_cwksp; + +/*-************************************* +* Functions +***************************************/ + +MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws); + +/** + * Align must be a power of 2. + */ +MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) { + size_t const mask = align - 1; + assert((align & mask) == 0); + return (size + mask) & ~mask; +} + +MEM_STATIC void ZSTD_cwksp_internal_advance_phase( + ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase) { + assert(phase >= ws->phase); + if (phase > ws->phase) { + if (ws->phase < ZSTD_cwksp_alloc_buffers && + phase >= ZSTD_cwksp_alloc_buffers) { + } + if (ws->phase < ZSTD_cwksp_alloc_aligned && + phase >= ZSTD_cwksp_alloc_aligned) { + /* If unaligned allocations down from a too-large top have left us + * unaligned, we need to realign our alloc ptr. Technically, this + * can consume space that is unaccounted for in the neededSpace + * calculation. However, I believe this can only happen when the + * workspace is too large, and specifically when it is too large + * by a larger margin than the space that will be consumed. */ + /* TODO: cleaner, compiler warning friendly way to do this??? */ + ws->allocStart = (BYTE*)ws->allocStart - ((size_t)ws->allocStart & (sizeof(U32)-1)); + } + ws->phase = phase; + } +} + +/** + * Internal function. Do not use directly. + */ +MEM_STATIC void* ZSTD_cwksp_reserve_internal( + ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase) { + void* alloc; + void* bottom = ws->tableEnd; + ZSTD_cwksp_internal_advance_phase(ws, phase); + alloc = (BYTE *)ws->allocStart - bytes; + DEBUGLOG(4, "cwksp: reserving %zd bytes, %zd bytes remaining", + bytes, ZSTD_cwksp_available_space(ws) - bytes); + assert(alloc >= bottom); + if (alloc < bottom) { + ws->allocFailed = 1; + return NULL; + } + ws->allocStart = alloc; + return alloc; +} + +/** + * Reserves and returns unaligned memory. + */ +MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes) { + return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers); +} + +/** + * Reserves and returns memory sized on and aligned on sizeof(unsigned). + */ +MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes) { + assert((bytes & (sizeof(U32)-1)) == 0); + return ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, sizeof(U32)), ZSTD_cwksp_alloc_aligned); +} + +/** + * Aligned on sizeof(unsigned). These buffers have the special property that + * their values remain constrained, allowing us to re-use them without + * memset()-ing them. + */ +MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) { + const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned; + void* alloc = ws->tableEnd; + void* end = (BYTE *)alloc + bytes; + void* top = ws->allocStart; + DEBUGLOG(4, "cwksp: reserving table %zd bytes, %zd bytes remaining", + bytes, ZSTD_cwksp_available_space(ws) - bytes); + assert((bytes & (sizeof(U32)-1)) == 0); + ZSTD_cwksp_internal_advance_phase(ws, phase); + assert(end <= top); + if (end > top) { + DEBUGLOG(4, "cwksp: object alloc failed!"); + ws->allocFailed = 1; + return NULL; + } + ws->tableEnd = end; + return alloc; +} + +/** + * Aligned on sizeof(void*). + */ +MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) { + size_t roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*)); + void* start = ws->objectEnd; + void* end = (BYTE*)start + roundedBytes; + DEBUGLOG(4, + "cwksp: reserving object %zd bytes (rounded to %zd), %zd bytes remaining", + bytes, roundedBytes, ZSTD_cwksp_available_space(ws) - roundedBytes); + assert(((size_t)start & (sizeof(void*)-1)) == 0); + assert((bytes & (sizeof(void*)-1)) == 0); + /* we must be in the first phase, no advance is possible */ + if (ws->phase != ZSTD_cwksp_alloc_objects || end > ws->workspaceEnd) { + DEBUGLOG(4, "cwksp: object alloc failed!"); + ws->allocFailed = 1; + return NULL; + } + ws->objectEnd = end; + ws->tableEnd = end; + return start; +} + +/** + * Invalidates table allocations. + * All other allocations remain valid. + */ +MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) { + DEBUGLOG(4, "cwksp: clearing tables!"); + ws->tableEnd = ws->objectEnd; +} + +/** + * Invalidates all buffer, aligned, and table allocations. + * Object allocations remain valid. + */ +MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) { + DEBUGLOG(4, "cwksp: clearing!"); + ws->tableEnd = ws->objectEnd; + ws->allocStart = ws->workspaceEnd; + ws->allocFailed = 0; + if (ws->phase > ZSTD_cwksp_alloc_buffers) { + ws->phase = ZSTD_cwksp_alloc_buffers; + } +} + +/** + * The provided workspace takes ownership of the buffer [start, start+size). + * Any existing values in the workspace are ignored (the previously managed + * buffer, if present, must be separately freed). + */ +MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) { + DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size); + assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */ + ws->workspace = start; + ws->workspaceEnd = (BYTE*)start + size; + ws->objectEnd = ws->workspace; + ws->phase = ZSTD_cwksp_alloc_objects; + ZSTD_cwksp_clear(ws); + ws->workspaceOversizedDuration = 0; +} + +MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) { + void* workspace = ZSTD_malloc(size, customMem); + DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size); + RETURN_ERROR_IF(workspace == NULL, memory_allocation); + ZSTD_cwksp_init(ws, workspace, size); + return 0; +} + +MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) { + DEBUGLOG(4, "cwksp: freeing workspace"); + ZSTD_free(ws->workspace, customMem); + ws->workspace = NULL; + ws->workspaceEnd = NULL; + ZSTD_cwksp_clear(ws); +} + +/** + * Moves the management of a workspace from one cwksp to another. The src cwksp + * is left in an invalid state (src must be re-init()'ed before its used again). + */ +MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) { + *dst = *src; + memset(src, 0, sizeof(ZSTD_cwksp)); +} + +MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) { + return (BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace; +} + +MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) { + return ws->allocFailed; +} + +/*-************************************* +* Functions Checking Free Space +***************************************/ + +MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws) { + return (size_t)((BYTE*)ws->allocStart - (BYTE*)ws->tableEnd); +} + +MEM_STATIC int ZSTD_cwksp_check_available(ZSTD_cwksp* ws, size_t additionalNeededSpace) { + return ZSTD_cwksp_available_space(ws) >= additionalNeededSpace; +} + +MEM_STATIC int ZSTD_cwksp_check_too_large(ZSTD_cwksp* ws, size_t additionalNeededSpace) { + return ZSTD_cwksp_check_available( + ws, additionalNeededSpace * ZSTD_WORKSPACETOOLARGE_FACTOR); +} + +MEM_STATIC int ZSTD_cwksp_check_wasteful(ZSTD_cwksp* ws, size_t additionalNeededSpace) { + return ZSTD_cwksp_check_too_large(ws, additionalNeededSpace) + && ws->workspaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION; +} + +MEM_STATIC void ZSTD_cwksp_bump_oversized_duration( + ZSTD_cwksp* ws, size_t additionalNeededSpace) { + if (ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)) { + ws->workspaceOversizedDuration++; + } else { + ws->workspaceOversizedDuration = 0; + } +} + +#if defined (__cplusplus) +} +#endif + +#endif /* ZSTD_CWKSP_H */