diff options
author | Conrad Meyer <cem@FreeBSD.org> | 2018-10-22 20:00:30 +0000 |
---|---|---|
committer | Conrad Meyer <cem@FreeBSD.org> | 2018-10-22 20:00:30 +0000 |
commit | 706cfae467a217cc786fd96a72cc2e33c61987e4 (patch) | |
tree | e7673904660df47b5abd9a1c33cf982a514dac66 /tests | |
parent | 42239e68a5cfba3b37b054425eace8d14e0844e3 (diff) |
Notes
Diffstat (limited to 'tests')
-rw-r--r-- | tests/.gitignore | 2 | ||||
-rw-r--r-- | tests/Makefile | 57 | ||||
-rw-r--r-- | tests/README.md | 53 | ||||
-rw-r--r-- | tests/decodecorpus.c | 14 | ||||
-rw-r--r-- | tests/fullbench.c | 477 | ||||
-rw-r--r-- | tests/fuzz/fuzz.h | 6 | ||||
-rwxr-xr-x | tests/fuzz/fuzz.py | 43 | ||||
-rw-r--r-- | tests/fuzz/regression_driver.c | 2 | ||||
-rw-r--r-- | tests/fuzz/zstd_helpers.c | 4 | ||||
-rw-r--r-- | tests/fuzzer.c | 294 | ||||
-rw-r--r-- | tests/gzip/Makefile | 2 | ||||
-rw-r--r-- | tests/legacy.c | 5 | ||||
-rwxr-xr-x | tests/libzstd_partial_builds.sh | 36 | ||||
-rw-r--r-- | tests/longmatch.c | 16 | ||||
-rw-r--r-- | tests/paramgrill.c | 2957 | ||||
-rwxr-xr-x | tests/playTests.sh | 229 | ||||
-rw-r--r-- | tests/poolTests.c | 193 | ||||
-rwxr-xr-x | tests/rateLimiter.py | 40 | ||||
-rw-r--r-- | tests/roundTripCrash.c | 2 | ||||
-rw-r--r-- | tests/symbols.c | 2 | ||||
-rwxr-xr-x | tests/test-zstd-versions.py | 2 | ||||
-rw-r--r-- | tests/zstreamtest.c | 368 |
22 files changed, 3808 insertions, 996 deletions
diff --git a/tests/.gitignore b/tests/.gitignore index 4911b2d62a568..1f08c3995e858 100644 --- a/tests/.gitignore +++ b/tests/.gitignore @@ -1,6 +1,7 @@ # local binary (Makefile) fullbench fullbench32 +fullbench-lib fuzzer fuzzer32 fuzzer-dll @@ -26,6 +27,7 @@ invalidDictionaries checkTag zcat zstdcat +tm # Tmp test directory zstdtest diff --git a/tests/Makefile b/tests/Makefile index 5b35ad406318d..2a96829f61c5d 100644 --- a/tests/Makefile +++ b/tests/Makefile @@ -24,15 +24,18 @@ PYTHON ?= python3 TESTARTEFACT := versionsTest DEBUGLEVEL ?= 1 -DEBUGFLAGS = -g -DZSTD_DEBUG=$(DEBUGLEVEL) +DEBUGFLAGS = -g -DDEBUGLEVEL=$(DEBUGLEVEL) CPPFLAGS += -I$(ZSTDDIR) -I$(ZSTDDIR)/common -I$(ZSTDDIR)/compress \ -I$(ZSTDDIR)/dictBuilder -I$(ZSTDDIR)/deprecated -I$(PRGDIR) +ifeq ($(OS),Windows_NT) # MinGW assumed +CPPFLAGS += -D__USE_MINGW_ANSI_STDIO # compatibility with %zu formatting +endif CFLAGS ?= -O3 CFLAGS += -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \ -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \ -Wstrict-prototypes -Wundef -Wformat-security \ -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \ - -Wredundant-decls + -Wredundant-decls -Wmissing-prototypes CFLAGS += $(DEBUGFLAGS) $(MOREFLAGS) FLAGS = $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) @@ -78,7 +81,8 @@ DECODECORPUS_TESTTIME ?= -T30 default: fullbench @echo $(ZSTDMT_OBJECTS) -all: fullbench fuzzer zstreamtest paramgrill datagen decodecorpus roundTripCrash +all: fullbench fuzzer zstreamtest paramgrill datagen decodecorpus roundTripCrash \ + fullbench-lib all32: fullbench32 fuzzer32 zstreamtest32 @@ -88,13 +92,8 @@ allnothread: fullbench fuzzer paramgrill datagen decodecorpus dll: fuzzer-dll zstreamtest-dll -zstd: - $(MAKE) -C $(PRGDIR) $@ MOREFLAGS+="$(DEBUGFLAGS)" - -zstd32: - $(MAKE) -C $(PRGDIR) $@ MOREFLAGS+="$(DEBUGFLAGS)" - -zstd-nolegacy: +PHONY: zstd zstd32 zstd-nolegacy # must be phony, only external makefile knows how to build them, or if they need an update +zstd zstd32 zstd-nolegacy: $(MAKE) -C $(PRGDIR) $@ MOREFLAGS+="$(DEBUGFLAGS)" gzstd: @@ -131,13 +130,14 @@ zstdmt_d_%.o : $(ZSTDDIR)/decompress/%.c fullbench32: CPPFLAGS += -m32 fullbench fullbench32 : CPPFLAGS += $(MULTITHREAD_CPP) fullbench fullbench32 : LDFLAGS += $(MULTITHREAD_LD) -fullbench fullbench32 : DEBUGFLAGS = # turn off assert() for speed measurements +fullbench fullbench32 : DEBUGFLAGS = -DNDEBUG # turn off assert() for speed measurements fullbench fullbench32 : $(ZSTD_FILES) -fullbench fullbench32 : $(PRGDIR)/datagen.c fullbench.c +fullbench fullbench32 : $(PRGDIR)/datagen.c $(PRGDIR)/bench.c fullbench.c $(CC) $(FLAGS) $^ -o $@$(EXT) +fullbench-lib : CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ fullbench-lib : zstd-staticLib -fullbench-lib : $(PRGDIR)/datagen.c fullbench.c +fullbench-lib : $(PRGDIR)/datagen.c $(PRGDIR)/bench.c fullbench.c $(CC) $(FLAGS) $(filter %.c,$^) -o $@$(EXT) $(ZSTDDIR)/libzstd.a # note : broken : requires unavailable symbols @@ -202,8 +202,8 @@ zstreamtest-dll : $(ZSTDDIR)/common/xxhash.c # xxh symbols not exposed from dll zstreamtest-dll : $(ZSTREAM_LOCAL_FILES) $(CC) $(CPPFLAGS) $(CFLAGS) $(filter %.c,$^) $(LDFLAGS) -o $@$(EXT) -paramgrill : DEBUGFLAGS = # turn off assert() for speed measurements -paramgrill : $(ZSTD_FILES) $(PRGDIR)/datagen.c paramgrill.c +paramgrill : DEBUGFLAGS = # turn off assert() by default for speed measurements +paramgrill : $(ZSTD_FILES) $(PRGDIR)/bench.c $(PRGDIR)/datagen.c paramgrill.c $(CC) $(FLAGS) $^ -lm -o $@$(EXT) datagen : $(PRGDIR)/datagen.c datagencli.c @@ -245,13 +245,14 @@ checkTag: checkTag.c $(ZSTDDIR)/zstd.h clean: $(MAKE) -C $(ZSTDDIR) clean + $(MAKE) -C $(PRGDIR) clean @$(RM) -fR $(TESTARTEFACT) @$(RM) -f core *.o tmp* result* *.gcda dictionary *.zst \ $(PRGDIR)/zstd$(EXT) $(PRGDIR)/zstd32$(EXT) \ fullbench$(EXT) fullbench32$(EXT) \ fullbench-lib$(EXT) fullbench-dll$(EXT) \ fuzzer$(EXT) fuzzer32$(EXT) zbufftest$(EXT) zbufftest32$(EXT) \ - fuzzer-dll$(EXT) zstreamtest-dll$(EXT) zbufftest-dll$(EXT)\ + fuzzer-dll$(EXT) zstreamtest-dll$(EXT) zbufftest-dll$(EXT) \ zstreamtest$(EXT) zstreamtest32$(EXT) \ datagen$(EXT) paramgrill$(EXT) roundTripCrash$(EXT) longmatch$(EXT) \ symbols$(EXT) invalidDictionaries$(EXT) legacy$(EXT) poolTests$(EXT) \ @@ -260,7 +261,7 @@ clean: #---------------------------------------------------------------------------------- -#make valgrindTest is validated only for Linux, OSX, BSD, Hurd and Solaris targets +#make valgrindTest is validated only for Linux, macOS, BSD, Hurd and Solaris targets #---------------------------------------------------------------------------------- ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS)) HOST_OS = POSIX @@ -301,11 +302,6 @@ endif list: @$(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null | awk -v RS= -F: '/^# File/,/^# Finished Make data base/ {if ($$1 !~ "^[#.]") {print $$1}}' | sort | egrep -v -e '^[^[:alnum:]]' -e '^$@$$' | xargs -.PHONY: zstd-playTests -zstd-playTests: datagen - file $(ZSTD) - ZSTD="$(QEMU_SYS) $(ZSTD)" ./playTests.sh $(ZSTDRTTEST) - .PHONY: shortest shortest: ZSTDRTTEST= shortest: test-zstd @@ -323,14 +319,21 @@ test32: test-zstd32 test-fullbench32 test-fuzzer32 test-zstream32 test-all: test test32 valgrindTest test-decodecorpus-cli + +.PHONY: test-zstd test-zstd32 test-zstd-nolegacy test-zstd: ZSTD = $(PRGDIR)/zstd -test-zstd: zstd zstd-playTests +test-zstd: zstd test-zstd32: ZSTD = $(PRGDIR)/zstd32 -test-zstd32: zstd32 zstd-playTests +test-zstd32: zstd32 test-zstd-nolegacy: ZSTD = $(PRGDIR)/zstd-nolegacy -test-zstd-nolegacy: zstd-nolegacy zstd-playTests +test-zstd-nolegacy: zstd-nolegacy + +test-zstd test-zstd32 test-zstd-nolegacy: datagen + file $(ZSTD) + ZSTD="$(QEMU_SYS) $(ZSTD)" ./playTests.sh $(ZSTDRTTEST) + test-gzstd: gzstd $(PRGDIR)/zstd -f README.md test-zstd-speed.py @@ -360,6 +363,9 @@ test-fullbench32: fullbench32 datagen test-fuzzer: fuzzer $(QEMU_SYS) ./fuzzer -v $(FUZZERTEST) $(FUZZER_FLAGS) +test-fuzzer-stackmode: MOREFLAGS += -DZSTD_HEAPMODE=0 +test-fuzzer-stackmode: test-fuzzer + test-fuzzer32: fuzzer32 $(QEMU_SYS) ./fuzzer32 -v $(FUZZERTEST) $(FUZZER_FLAGS) @@ -373,7 +379,6 @@ test-zstream: zstreamtest $(QEMU_SYS) ./zstreamtest -v $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS) $(QEMU_SYS) ./zstreamtest --mt -t1 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS) $(QEMU_SYS) ./zstreamtest --newapi -t1 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS) - $(QEMU_SYS) ./zstreamtest --opaqueapi -t1 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS) test-zstream32: zstreamtest32 $(QEMU_SYS) ./zstreamtest32 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS) diff --git a/tests/README.md b/tests/README.md index 24a28ab7b9d03..f28766bd19423 100644 --- a/tests/README.md +++ b/tests/README.md @@ -88,3 +88,56 @@ as well as the 10,000 original files for more detailed comparison of decompressi will choose a random seed, and for 1 minute, generate random test frames and ensure that the zstd library correctly decompresses them in both simple and streaming modes. + +#### `paramgrill` - tool for generating compression table parameters and optimizing parameters on file given constraints + +Full list of arguments +``` + -T# : set level 1 speed objective + -B# : cut input into blocks of size # (default : single block) + -S : benchmarks a single run (example command: -Sl3w10h12) + w# - windowLog + h# - hashLog + c# - chainLog + s# - searchLog + l# - searchLength + t# - targetLength + S# - strategy + L# - level + --zstd= : Single run, parameter selection syntax same as zstdcli with more parameters + (Added forceAttachDictionary / fadt) + When invoked with --optimize, this represents the sample to exceed. + --optimize= : find parameters to maximize compression ratio given parameters + Can use all --zstd= commands to constrain the type of solution found in addition to the following constraints + cSpeed= : Minimum compression speed + dSpeed= : Minimum decompression speed + cMem= : Maximum compression memory + lvl= : Searches for solutions which are strictly better than that compression lvl in ratio and cSpeed, + stc= : When invoked with lvl=, represents percentage slack in ratio/cSpeed allowed for a solution to be considered (Default 100%) + : In normal operation, represents percentage slack in choosing viable starting strategy selection in choosing the default parameters + (Lower value will begin with stronger strategies) (Default 90%) + speedRatio= (accepts decimals) + : determines value of gains in speed vs gains in ratio + when determining overall winner (default 5 (1% ratio = 5% speed)). + tries= : Maximum number of random restarts on a single strategy before switching (Default 5) + Higher values will make optimizer run longer, more chances to find better solution. + memLog : Limits the log of the size of each memotable (1 per strategy). Will use hash tables when state space is larger than max size. + Setting memLog = 0 turns off memoization + --display= : specifiy which parameters are included in the output + can use all --zstd parameter names and 'cParams' as a shorthand for all parameters used in ZSTD_compressionParameters + (Default: display all params available) + -P# : generated sample compressibility (when no file is provided) + -t# : Caps runtime of operation in seconds (default : 99999 seconds (about 27 hours )) + -v : Prints Benchmarking output + -D : Next argument dictionary file + -s : Benchmark all files separately + -q : Quiet, repeat for more quiet + -q Prints parameters + results whenever a new best is found + -qq Only prints parameters whenever a new best is found, prints final parameters + results + -qqq Only print final parameters + results + -qqqq Only prints final parameter set in the form --zstd= + -v : Verbose, cancels quiet, repeat for more volume + -v Prints all candidate parameters and results + +``` + Any inputs afterwards are treated as files to benchmark. diff --git a/tests/decodecorpus.c b/tests/decodecorpus.c index 407653119dd25..2c2276004a955 100644 --- a/tests/decodecorpus.c +++ b/tests/decodecorpus.c @@ -437,7 +437,8 @@ static size_t writeHufHeader(U32* seed, HUF_CElt* hufTable, void* dst, size_t ds U32 count[HUF_SYMBOLVALUE_MAX+1]; /* Scan input and build symbol stats */ - { size_t const largest = FSE_count_wksp (count, &maxSymbolValue, (const BYTE*)src, srcSize, WKSP); + { size_t const largest = HIST_count_wksp (count, &maxSymbolValue, (const BYTE*)src, srcSize, WKSP); + assert(!HIST_isError(largest)); if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 0; } /* single symbol, rle */ if (largest <= (srcSize >> 7)+1) return 0; /* Fast heuristic : not compressible enough */ } @@ -619,6 +620,8 @@ static size_t writeLiteralsBlock(U32* seed, frame_t* frame, size_t contentSize) } static inline void initSeqStore(seqStore_t *seqStore) { + seqStore->maxNbSeq = MAX_NB_SEQ; + seqStore->maxNbLit = ZSTD_BLOCKSIZE_MAX; seqStore->sequencesStart = SEQUENCE_BUFFER; seqStore->litStart = SEQUENCE_LITERAL_BUFFER; seqStore->llCode = SEQUENCE_LLCODE; @@ -834,7 +837,8 @@ static size_t writeSequences(U32* seed, frame_t* frame, seqStore_t* seqStorePtr, /* CTable for Literal Lengths */ { U32 max = MaxLL; - size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, WKSP); + size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, WKSP); /* cannot fail */ + assert(!HIST_isError(mostFrequent)); if (mostFrequent == nbSeq) { /* do RLE if we have the chance */ *op++ = llCodeTable[0]; @@ -865,7 +869,8 @@ static size_t writeSequences(U32* seed, frame_t* frame, seqStore_t* seqStorePtr, /* CTable for Offsets */ /* see Literal Lengths for descriptions of mode choices */ { U32 max = MaxOff; - size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, WKSP); + size_t const mostFrequent = HIST_countFast_wksp(count, &max, ofCodeTable, nbSeq, WKSP); /* cannot fail */ + assert(!HIST_isError(mostFrequent)); if (mostFrequent == nbSeq) { *op++ = ofCodeTable[0]; FSE_buildCTable_rle(CTable_OffsetBits, (BYTE)max); @@ -892,7 +897,8 @@ static size_t writeSequences(U32* seed, frame_t* frame, seqStore_t* seqStorePtr, /* CTable for MatchLengths */ /* see Literal Lengths for descriptions of mode choices */ { U32 max = MaxML; - size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, WKSP); + size_t const mostFrequent = HIST_countFast_wksp(count, &max, mlCodeTable, nbSeq, WKSP); /* cannot fail */ + assert(!HIST_isError(mostFrequent)); if (mostFrequent == nbSeq) { *op++ = *mlCodeTable; FSE_buildCTable_rle(CTable_MatchLength, (BYTE)max); diff --git a/tests/fullbench.c b/tests/fullbench.c index 6abdd4da00fa2..b05f1537cd706 100644 --- a/tests/fullbench.c +++ b/tests/fullbench.c @@ -30,6 +30,7 @@ #include "zstd.h" /* ZSTD_versionString */ #include "util.h" /* time functions */ #include "datagen.h" +#include "bench.h" /* CustomBench*/ /*_************************************ @@ -45,9 +46,13 @@ #define KNUTH 2654435761U #define MAX_MEM (1984 MB) +#define DEFAULT_CLEVEL 1 + #define COMPRESSIBILITY_DEFAULT 0.50 static const size_t g_sampleSize = 10000000; +#define TIMELOOP_NANOSEC (1*1000000000ULL) /* 1 second */ + /*_************************************ * Macros @@ -93,14 +98,26 @@ static size_t BMK_findMaxMem(U64 requiredMem) /*_******************************************************* * Benchmark wrappers *********************************************************/ -size_t local_ZSTD_compress(void* dst, size_t dstSize, void* buff2, const void* src, size_t srcSize) + +static ZSTD_CCtx* g_zcc = NULL; + +static size_t +local_ZSTD_compress(const void* src, size_t srcSize, + void* dst, size_t dstSize, + void* buff2) { - (void)buff2; - return ZSTD_compress(dst, dstSize, src, srcSize, 1); + ZSTD_parameters p; + ZSTD_frameParameters f = { 1 /* contentSizeHeader*/, 0, 0 }; + p.fParams = f; + p.cParams = *(ZSTD_compressionParameters*)buff2; + return ZSTD_compress_advanced (g_zcc, dst, dstSize, src, srcSize, NULL ,0, p); + //return ZSTD_compress(dst, dstSize, src, srcSize, cLevel); } static size_t g_cSize = 0; -size_t local_ZSTD_decompress(void* dst, size_t dstSize, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_decompress(const void* src, size_t srcSize, + void* dst, size_t dstSize, + void* buff2) { (void)src; (void)srcSize; return ZSTD_decompress(dst, dstSize, buff2, g_cSize); @@ -110,14 +127,14 @@ static ZSTD_DCtx* g_zdc = NULL; #ifndef ZSTD_DLL_IMPORT extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize); -size_t local_ZSTD_decodeLiteralsBlock(void* dst, size_t dstSize, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_decodeLiteralsBlock(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2) { (void)src; (void)srcSize; (void)dst; (void)dstSize; return ZSTD_decodeLiteralsBlock((ZSTD_DCtx*)g_zdc, buff2, g_cSize); } extern size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeq, const void* src, size_t srcSize); -size_t local_ZSTD_decodeSeqHeaders(void* dst, size_t dstSize, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_decodeSeqHeaders(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2) { int nbSeq; (void)src; (void)srcSize; (void)dst; (void)dstSize; @@ -126,12 +143,18 @@ size_t local_ZSTD_decodeSeqHeaders(void* dst, size_t dstSize, void* buff2, const #endif static ZSTD_CStream* g_cstream= NULL; -size_t local_ZSTD_compressStream(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_compressStream(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; - (void)buff2; - ZSTD_initCStream(g_cstream, 1); + ZSTD_parameters p; + ZSTD_frameParameters f = {1 /* contentSizeHeader*/, 0, 0}; + p.fParams = f; + p.cParams = *(ZSTD_compressionParameters*)buff2; + ZSTD_initCStream_advanced(g_cstream, NULL, 0, p, ZSTD_CONTENTSIZE_UNKNOWN); buffOut.dst = dst; buffOut.size = dstCapacity; buffOut.pos = 0; @@ -143,12 +166,14 @@ size_t local_ZSTD_compressStream(void* dst, size_t dstCapacity, void* buff2, con return buffOut.pos; } -static size_t local_ZSTD_compress_generic_end(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_compress_generic_end(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; (void)buff2; - ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionLevel, 1); buffOut.dst = dst; buffOut.size = dstCapacity; buffOut.pos = 0; @@ -159,12 +184,14 @@ static size_t local_ZSTD_compress_generic_end(void* dst, size_t dstCapacity, voi return buffOut.pos; } -static size_t local_ZSTD_compress_generic_continue(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_compress_generic_continue(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; (void)buff2; - ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionLevel, 1); buffOut.dst = dst; buffOut.size = dstCapacity; buffOut.pos = 0; @@ -176,12 +203,14 @@ static size_t local_ZSTD_compress_generic_continue(void* dst, size_t dstCapacity return buffOut.pos; } -static size_t local_ZSTD_compress_generic_T2_end(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_compress_generic_T2_end(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; (void)buff2; - ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionLevel, 1); ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_nbWorkers, 2); buffOut.dst = dst; buffOut.size = dstCapacity; @@ -193,12 +222,14 @@ static size_t local_ZSTD_compress_generic_T2_end(void* dst, size_t dstCapacity, return buffOut.pos; } -static size_t local_ZSTD_compress_generic_T2_continue(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_compress_generic_T2_continue(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; (void)buff2; - ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionLevel, 1); ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_nbWorkers, 2); buffOut.dst = dst; buffOut.size = dstCapacity; @@ -212,7 +243,10 @@ static size_t local_ZSTD_compress_generic_T2_continue(void* dst, size_t dstCapac } static ZSTD_DStream* g_dstream= NULL; -static size_t local_ZSTD_decompressStream(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t +local_ZSTD_decompressStream(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { ZSTD_outBuffer buffOut; ZSTD_inBuffer buffIn; @@ -228,34 +262,52 @@ static size_t local_ZSTD_decompressStream(void* dst, size_t dstCapacity, void* b return buffOut.pos; } -static ZSTD_CCtx* g_zcc = NULL; - #ifndef ZSTD_DLL_IMPORT -size_t local_ZSTD_compressContinue(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_compressContinue(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { - (void)buff2; - ZSTD_compressBegin(g_zcc, 1 /* compressionLevel */); + ZSTD_parameters p; + ZSTD_frameParameters f = { 1 /* contentSizeHeader*/, 0, 0 }; + p.fParams = f; + p.cParams = *(ZSTD_compressionParameters*)buff2; + ZSTD_compressBegin_advanced(g_zcc, NULL, 0, p, srcSize); return ZSTD_compressEnd(g_zcc, dst, dstCapacity, src, srcSize); } #define FIRST_BLOCK_SIZE 8 -size_t local_ZSTD_compressContinue_extDict(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_compressContinue_extDict(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { BYTE firstBlockBuf[FIRST_BLOCK_SIZE]; - (void)buff2; + ZSTD_parameters p; + ZSTD_frameParameters f = { 1, 0, 0 }; + p.fParams = f; + p.cParams = *(ZSTD_compressionParameters*)buff2; + ZSTD_compressBegin_advanced(g_zcc, NULL, 0, p, srcSize); memcpy(firstBlockBuf, src, FIRST_BLOCK_SIZE); - ZSTD_compressBegin(g_zcc, 1); - { size_t const compressResult = ZSTD_compressContinue(g_zcc, dst, dstCapacity, firstBlockBuf, FIRST_BLOCK_SIZE); - if (ZSTD_isError(compressResult)) { DISPLAY("local_ZSTD_compressContinue_extDict error : %s\n", ZSTD_getErrorName(compressResult)); return compressResult; } + { size_t const compressResult = ZSTD_compressContinue(g_zcc, + dst, dstCapacity, + firstBlockBuf, FIRST_BLOCK_SIZE); + if (ZSTD_isError(compressResult)) { + DISPLAY("local_ZSTD_compressContinue_extDict error : %s\n", + ZSTD_getErrorName(compressResult)); + return compressResult; + } dst = (BYTE*)dst + compressResult; dstCapacity -= compressResult; } - return ZSTD_compressEnd(g_zcc, dst, dstCapacity, (const BYTE*)src + FIRST_BLOCK_SIZE, srcSize - FIRST_BLOCK_SIZE); + return ZSTD_compressEnd(g_zcc, dst, dstCapacity, + (const BYTE*)src + FIRST_BLOCK_SIZE, + srcSize - FIRST_BLOCK_SIZE); } -size_t local_ZSTD_decompressContinue(void* dst, size_t dstCapacity, void* buff2, const void* src, size_t srcSize) +static size_t local_ZSTD_decompressContinue(const void* src, size_t srcSize, + void* dst, size_t dstCapacity, + void* buff2) { size_t regeneratedSize = 0; const BYTE* ip = (const BYTE*)buff2; @@ -263,7 +315,7 @@ size_t local_ZSTD_decompressContinue(void* dst, size_t dstCapacity, void* buff2, BYTE* op = (BYTE*)dst; size_t remainingCapacity = dstCapacity; - (void)src; (void)srcSize; + (void)src; (void)srcSize; /* unused */ ZSTD_decompressBegin(g_zdc); while (ip < iend) { size_t const iSize = ZSTD_nextSrcSizeToDecompress(g_zdc); @@ -282,27 +334,30 @@ size_t local_ZSTD_decompressContinue(void* dst, size_t dstCapacity, void* buff2, /*_******************************************************* * Bench functions *********************************************************/ -static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) +static size_t benchMem(U32 benchNb, + const void* src, size_t srcSize, + int cLevel, ZSTD_compressionParameters cparams) { + size_t dstBuffSize = ZSTD_compressBound(srcSize); BYTE* dstBuff; - size_t const dstBuffSize = ZSTD_compressBound(srcSize); + void* dstBuff2; void* buff2; const char* benchName; - size_t (*benchFunction)(void* dst, size_t dstSize, void* verifBuff, const void* src, size_t srcSize); - double bestTime = 100000000.; + BMK_benchFn_t benchFunction; + int errorcode = 0; /* Selection */ switch(benchNb) { case 1: - benchFunction = local_ZSTD_compress; benchName = "compress(1)"; + benchFunction = local_ZSTD_compress; benchName = "compress"; break; case 2: benchFunction = local_ZSTD_decompress; benchName = "decompress"; break; #ifndef ZSTD_DLL_IMPORT case 11: - benchFunction = local_ZSTD_compressContinue; benchName = "compressContinue(1)"; + benchFunction = local_ZSTD_compressContinue; benchName = "compressContinue"; break; case 12: benchFunction = local_ZSTD_compressContinue_extDict; benchName = "compressContinue_extDict"; @@ -318,7 +373,7 @@ static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) break; #endif case 41: - benchFunction = local_ZSTD_compressStream; benchName = "compressStream(1)"; + benchFunction = local_ZSTD_compressStream; benchName = "compressStream"; break; case 42: benchFunction = local_ZSTD_decompressStream; benchName = "decompressStream"; @@ -341,32 +396,65 @@ static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) /* Allocation */ dstBuff = (BYTE*)malloc(dstBuffSize); - buff2 = malloc(dstBuffSize); - if ((!dstBuff) || (!buff2)) { + dstBuff2 = malloc(dstBuffSize); + if ((!dstBuff) || (!dstBuff2)) { DISPLAY("\nError: not enough memory!\n"); - free(dstBuff); free(buff2); + free(dstBuff); free(dstBuff2); return 12; } + buff2 = dstBuff2; if (g_zcc==NULL) g_zcc = ZSTD_createCCtx(); if (g_zdc==NULL) g_zdc = ZSTD_createDCtx(); if (g_cstream==NULL) g_cstream = ZSTD_createCStream(); if (g_dstream==NULL) g_dstream = ZSTD_createDStream(); + /* DISPLAY("params: cLevel %d, wlog %d hlog %d clog %d slog %d slen %d tlen %d strat %d \n", + cLevel, cparams->windowLog, cparams->hashLog, cparams->chainLog, cparams->searchLog, + cparams->searchLength, cparams->targetLength, cparams->strategy); */ + + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_compressionLevel, cLevel); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_windowLog, cparams.windowLog); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_hashLog, cparams.hashLog); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_chainLog, cparams.chainLog); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_searchLog, cparams.searchLog); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_minMatch, cparams.searchLength); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_targetLength, cparams.targetLength); + ZSTD_CCtx_setParameter(g_zcc, ZSTD_p_compressionStrategy, cparams.strategy); + + + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionLevel, cLevel); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_windowLog, cparams.windowLog); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_hashLog, cparams.hashLog); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_chainLog, cparams.chainLog); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_searchLog, cparams.searchLog); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_minMatch, cparams.searchLength); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_targetLength, cparams.targetLength); + ZSTD_CCtx_setParameter(g_cstream, ZSTD_p_compressionStrategy, cparams.strategy); + /* Preparation */ switch(benchNb) { + case 1: + buff2 = &cparams; + break; case 2: - g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, 1); + g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, cLevel); break; #ifndef ZSTD_DLL_IMPORT + case 11: + buff2 = &cparams; + break; + case 12: + buff2 = &cparams; + break; case 13 : - g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, 1); + g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, cLevel); break; case 31: /* ZSTD_decodeLiteralsBlock */ { blockProperties_t bp; ZSTD_frameHeader zfp; size_t frameHeaderSize, skippedSize; - g_cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, 1); + g_cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, cLevel); frameHeaderSize = ZSTD_getFrameHeader(&zfp, dstBuff, ZSTD_frameHeaderSize_min); if (frameHeaderSize==0) frameHeaderSize = ZSTD_frameHeaderSize_min; ZSTD_getcBlockSize(dstBuff+frameHeaderSize, dstBuffSize, &bp); /* Get 1st block type */ @@ -386,8 +474,8 @@ static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) const BYTE* ip = dstBuff; const BYTE* iend; size_t frameHeaderSize, cBlockSize; - ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, 1); /* it would be better to use direct block compression here */ - g_cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, 1); + ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, cLevel); /* it would be better to use direct block compression here */ + g_cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, cLevel); frameHeaderSize = ZSTD_getFrameHeader(&zfp, dstBuff, ZSTD_frameHeaderSize_min); if (frameHeaderSize==0) frameHeaderSize = ZSTD_frameHeaderSize_min; ip += frameHeaderSize; /* Skip frame Header */ @@ -409,8 +497,11 @@ static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) case 31: goto _cleanOut; #endif + case 41 : + buff2 = &cparams; + break; case 42 : - g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, 1); + g_cSize = ZSTD_compress(buff2, dstBuffSize, src, srcSize, cLevel); break; /* test functions */ @@ -419,138 +510,190 @@ static size_t benchMem(const void* src, size_t srcSize, U32 benchNb) default : ; } - /* warming up memory */ + /* warming up dstBuff */ { size_t i; for (i=0; i<dstBuffSize; i++) dstBuff[i]=(BYTE)i; } /* benchmark loop */ - { U32 loopNb; - U32 nbRounds = (U32)((50 MB) / (srcSize+1)) + 1; /* initial conservative speed estimate */ -# define TIME_SEC_MICROSEC (1*1000000ULL) /* 1 second */ -# define TIME_SEC_NANOSEC (1*1000000000ULL) /* 1 second */ - DISPLAY("%2i- %-30.30s : \r", benchNb, benchName); - for (loopNb = 1; loopNb <= g_nbIterations; loopNb++) { - UTIL_time_t clockStart; - size_t benchResult=0; - U32 roundNb; - - UTIL_sleepMilli(5); /* give processor time to other processes */ - UTIL_waitForNextTick(); - clockStart = UTIL_getTime(); - for (roundNb=0; roundNb < nbRounds; roundNb++) { - benchResult = benchFunction(dstBuff, dstBuffSize, buff2, src, srcSize); - if (ZSTD_isError(benchResult)) { - DISPLAY("ERROR ! %s() => %s !! \n", benchName, ZSTD_getErrorName(benchResult)); - exit(1); - } } - { U64 const clockSpanNano = UTIL_clockSpanNano(clockStart); - double const averageTime = (double)clockSpanNano / TIME_SEC_NANOSEC / nbRounds; - if (clockSpanNano > 0) { - if (averageTime < bestTime) bestTime = averageTime; - assert(bestTime > (1./2000000000)); - nbRounds = (U32)(1. / bestTime); /* aim for 1 sec */ - DISPLAY("%2i- %-30.30s : %7.1f MB/s (%9u)\r", - loopNb, benchName, - (double)srcSize / (1 MB) / bestTime, - (U32)benchResult); - } else { - assert(nbRounds < 40000000); /* avoid overflow */ - nbRounds *= 100; - } - } } } - DISPLAY("%2u\n", benchNb); + { BMK_timedFnState_t* const tfs = BMK_createTimedFnState(g_nbIterations * 1000, 1000); + BMK_runTime_t bestResult; + bestResult.sumOfReturn = 0; + bestResult.nanoSecPerRun = (unsigned long long)(-1LL); + assert(tfs != NULL); + for (;;) { + void* const dstBuffv = dstBuff; + BMK_runOutcome_t const bOutcome = + BMK_benchTimedFn( tfs, + benchFunction, buff2, + NULL, NULL, /* initFn */ + 1, /* blockCount */ + &src, &srcSize, + &dstBuffv, &dstBuffSize, + NULL); + + if (!BMK_isSuccessful_runOutcome(bOutcome)) { + DISPLAY("ERROR benchmarking function ! ! \n"); + errorcode = 1; + goto _cleanOut; + } + + { BMK_runTime_t const newResult = BMK_extract_runTime(bOutcome); + if (newResult.nanoSecPerRun < bestResult.nanoSecPerRun ) + bestResult.nanoSecPerRun = newResult.nanoSecPerRun; + DISPLAY("\r%2u#%-29.29s:%8.1f MB/s (%8u) ", + benchNb, benchName, + (double)srcSize * TIMELOOP_NANOSEC / bestResult.nanoSecPerRun / MB_UNIT, + (unsigned)newResult.sumOfReturn ); + } + + if ( BMK_isCompleted_TimedFn(tfs) ) break; + } + BMK_freeTimedFnState(tfs); + } + DISPLAY("\n"); _cleanOut: free(dstBuff); - free(buff2); + free(dstBuff2); ZSTD_freeCCtx(g_zcc); g_zcc=NULL; ZSTD_freeDCtx(g_zdc); g_zdc=NULL; ZSTD_freeCStream(g_cstream); g_cstream=NULL; ZSTD_freeDStream(g_dstream); g_dstream=NULL; - return 0; + return errorcode; } -static int benchSample(U32 benchNb) +static int benchSample(U32 benchNb, + int cLevel, ZSTD_compressionParameters cparams) { size_t const benchedSize = g_sampleSize; - const char* name = "Sample 10MiB"; + const char* const name = "Sample 10MiB"; /* Allocation */ - void* origBuff = malloc(benchedSize); + void* const origBuff = malloc(benchedSize); if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); return 12; } /* Fill buffer */ RDG_genBuffer(origBuff, benchedSize, g_compressibility, 0.0, 0); /* bench */ - DISPLAY("\r%79s\r", ""); + DISPLAY("\r%70s\r", ""); DISPLAY(" %s : \n", name); - if (benchNb) - benchMem(origBuff, benchedSize, benchNb); - else - for (benchNb=0; benchNb<100; benchNb++) benchMem(origBuff, benchedSize, benchNb); + if (benchNb) { + benchMem(benchNb, origBuff, benchedSize, cLevel, cparams); + } else { /* 0 == run all tests */ + for (benchNb=0; benchNb<100; benchNb++) { + benchMem(benchNb, origBuff, benchedSize, cLevel, cparams); + } } free(origBuff); return 0; } -static int benchFiles(const char** fileNamesTable, const int nbFiles, U32 benchNb) +static int benchFiles(U32 benchNb, + const char** fileNamesTable, const int nbFiles, + int cLevel, ZSTD_compressionParameters cparams) { /* Loop for each file */ int fileIdx; for (fileIdx=0; fileIdx<nbFiles; fileIdx++) { const char* const inFileName = fileNamesTable[fileIdx]; FILE* const inFile = fopen( inFileName, "rb" ); - U64 inFileSize; size_t benchedSize; - void* origBuff; /* Check file existence */ if (inFile==NULL) { DISPLAY( "Pb opening %s\n", inFileName); return 11; } /* Memory allocation & restrictions */ - inFileSize = UTIL_getFileSize(inFileName); - if (inFileSize == UTIL_FILESIZE_UNKNOWN) { - DISPLAY( "Cannot measure size of %s\n", inFileName); - fclose(inFile); - return 11; - } - benchedSize = BMK_findMaxMem(inFileSize*3) / 3; - if ((U64)benchedSize > inFileSize) benchedSize = (size_t)inFileSize; - if (benchedSize < inFileSize) - DISPLAY("Not enough memory for '%s' full size; testing %u MB only...\n", inFileName, (U32)(benchedSize>>20)); + { U64 const inFileSize = UTIL_getFileSize(inFileName); + if (inFileSize == UTIL_FILESIZE_UNKNOWN) { + DISPLAY( "Cannot measure size of %s\n", inFileName); + fclose(inFile); + return 11; + } + benchedSize = BMK_findMaxMem(inFileSize*3) / 3; + if ((U64)benchedSize > inFileSize) + benchedSize = (size_t)inFileSize; + if ((U64)benchedSize < inFileSize) { + DISPLAY("Not enough memory for '%s' full size; testing %u MB only... \n", + inFileName, (U32)(benchedSize>>20)); + } } /* Alloc */ - origBuff = malloc(benchedSize); - if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); fclose(inFile); return 12; } - - /* Fill input buffer */ - DISPLAY("Loading %s... \r", inFileName); - { - size_t readSize = fread(origBuff, 1, benchedSize, inFile); - fclose(inFile); - if (readSize != benchedSize) { - DISPLAY("\nError: problem reading file '%s' !! \n", inFileName); - free(origBuff); - return 13; - } } + { void* const origBuff = malloc(benchedSize); + if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); fclose(inFile); return 12; } + + /* Fill input buffer */ + DISPLAY("Loading %s... \r", inFileName); + { size_t const readSize = fread(origBuff, 1, benchedSize, inFile); + fclose(inFile); + if (readSize != benchedSize) { + DISPLAY("\nError: problem reading file '%s' !! \n", inFileName); + free(origBuff); + return 13; + } } - /* bench */ - DISPLAY("\r%79s\r", ""); - DISPLAY(" %s : \n", inFileName); - if (benchNb) - benchMem(origBuff, benchedSize, benchNb); - else - for (benchNb=0; benchNb<100; benchNb++) benchMem(origBuff, benchedSize, benchNb); + /* bench */ + DISPLAY("\r%70s\r", ""); /* blank line */ + DISPLAY(" %s : \n", inFileName); + if (benchNb) { + benchMem(benchNb, origBuff, benchedSize, cLevel, cparams); + } else { + for (benchNb=0; benchNb<100; benchNb++) { + benchMem(benchNb, origBuff, benchedSize, cLevel, cparams); + } } - free(origBuff); - } + free(origBuff); + } } return 0; } + +/*_******************************************************* +* Argument Parsing +*********************************************************/ + +#define ERROR_OUT(msg) { DISPLAY("%s \n", msg); exit(1); } + +static unsigned readU32FromChar(const char** stringPtr) +{ + const char errorMsg[] = "error: numeric value too large"; + unsigned result = 0; + while ((**stringPtr >='0') && (**stringPtr <='9')) { + unsigned const max = (((unsigned)(-1)) / 10) - 1; + if (result > max) ERROR_OUT(errorMsg); + result *= 10, result += **stringPtr - '0', (*stringPtr)++ ; + } + if ((**stringPtr=='K') || (**stringPtr=='M')) { + unsigned const maxK = ((unsigned)(-1)) >> 10; + if (result > maxK) ERROR_OUT(errorMsg); + result <<= 10; + if (**stringPtr=='M') { + if (result > maxK) ERROR_OUT(errorMsg); + result <<= 10; + } + (*stringPtr)++; /* skip `K` or `M` */ + if (**stringPtr=='i') (*stringPtr)++; + if (**stringPtr=='B') (*stringPtr)++; + } + return result; +} + +static unsigned longCommandWArg(const char** stringPtr, const char* longCommand) +{ + size_t const comSize = strlen(longCommand); + int const result = !strncmp(*stringPtr, longCommand, comSize); + if (result) *stringPtr += comSize; + return result; +} + + +/*_******************************************************* +* Command line +*********************************************************/ + static int usage(const char* exename) { DISPLAY( "Usage :\n"); @@ -567,6 +710,8 @@ static int usage_advanced(const char* exename) DISPLAY( " -b# : test only function # \n"); DISPLAY( " -i# : iteration loops [1-9](default : %i)\n", NBLOOPS); DISPLAY( " -P# : sample compressibility (default : %.1f%%)\n", COMPRESSIBILITY_DEFAULT * 100); + DISPLAY( " -l# : benchmark functions at that compression level (default : %i)\n", DEFAULT_CLEVEL); + DISPLAY( " --zstd : custom parameter selection. Format same as zstdcli \n"); return 0; } @@ -579,23 +724,45 @@ static int badusage(const char* exename) int main(int argc, const char** argv) { - int i, filenamesStart=0, result; - const char* exename = argv[0]; + int argNb, filenamesStart=0, result; + const char* const exename = argv[0]; const char* input_filename = NULL; U32 benchNb = 0, main_pause = 0; + int cLevel = DEFAULT_CLEVEL; + ZSTD_compressionParameters cparams = ZSTD_getCParams(cLevel, 0, 0); DISPLAY(WELCOME_MESSAGE); if (argc<1) return badusage(exename); - for(i=1; i<argc; i++) { - const char* argument = argv[i]; + for (argNb=1; argNb<argc; argNb++) { + const char* argument = argv[argNb]; assert(argument != NULL); - /* Commands (note : aggregated commands are allowed) */ - if (argument[0]=='-') { + if (longCommandWArg(&argument, "--zstd=")) { + for ( ; ;) { + if (longCommandWArg(&argument, "windowLog=") || longCommandWArg(&argument, "wlog=")) { cparams.windowLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "chainLog=") || longCommandWArg(&argument, "clog=")) { cparams.chainLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "hashLog=") || longCommandWArg(&argument, "hlog=")) { cparams.hashLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "searchLog=") || longCommandWArg(&argument, "slog=")) { cparams.searchLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "searchLength=") || longCommandWArg(&argument, "slen=")) { cparams.searchLength = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "targetLength=") || longCommandWArg(&argument, "tlen=")) { cparams.targetLength = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "strategy=") || longCommandWArg(&argument, "strat=")) { cparams.strategy = (ZSTD_strategy)(readU32FromChar(&argument)); if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "level=") || longCommandWArg(&argument, "lvl=")) { cLevel = (int)readU32FromChar(&argument); cparams = ZSTD_getCParams(cLevel, 0, 0); if (argument[0]==',') { argument++; continue; } else break; } + DISPLAY("invalid compression parameter \n"); + return 1; + } + + /* check end of string */ + if (argument[0] != 0) { + DISPLAY("invalid --zstd= format \n"); + return 1; + } else { + continue; + } - while (argument[1]!=0) { - argument++; + } else if (argument[0]=='-') { /* Commands (note : aggregated commands are allowed) */ + argument++; + while (argument[0]!=0) { switch(argument[0]) { @@ -608,33 +775,25 @@ int main(int argc, const char** argv) /* Select specific algorithm to bench */ case 'b': - benchNb = 0; - while ((argument[1]>= '0') && (argument[1]<= '9')) { - benchNb *= 10; - benchNb += argument[1] - '0'; - argument++; - } + argument++; + benchNb = readU32FromChar(&argument); break; /* Modify Nb Iterations */ case 'i': - if ((argument[1] >='0') && (argument[1] <='9')) { - int iters = argument[1] - '0'; - BMK_SetNbIterations(iters); - argument++; - } + argument++; + BMK_SetNbIterations((int)readU32FromChar(&argument)); break; /* Select compressibility of synthetic sample */ case 'P': - { U32 proba32 = 0; - while ((argument[1]>= '0') && (argument[1]<= '9')) { - proba32 *= 10; - proba32 += argument[1] - '0'; - argument++; - } - g_compressibility = (double)proba32 / 100.; - } + argument++; + g_compressibility = (double)readU32FromChar(&argument) / 100.; + break; + case 'l': + argument++; + cLevel = readU32FromChar(&argument); + cparams = ZSTD_getCParams(cLevel, 0, 0); break; /* Unknown command */ @@ -645,13 +804,15 @@ int main(int argc, const char** argv) } /* first provided filename is input */ - if (!input_filename) { input_filename=argument; filenamesStart=i; continue; } + if (!input_filename) { input_filename=argument; filenamesStart=argNb; continue; } } + + if (filenamesStart==0) /* no input file */ - result = benchSample(benchNb); + result = benchSample(benchNb, cLevel, cparams); else - result = benchFiles(argv+filenamesStart, argc-filenamesStart, benchNb); + result = benchFiles(benchNb, argv+filenamesStart, argc-filenamesStart, cLevel, cparams); if (main_pause) { int unused; printf("press enter...\n"); unused = getchar(); (void)unused; } diff --git a/tests/fuzz/fuzz.h b/tests/fuzz/fuzz.h index a64845473c2b0..8850025b0fd25 100644 --- a/tests/fuzz/fuzz.h +++ b/tests/fuzz/fuzz.h @@ -23,10 +23,10 @@ * the data to zstd functions. Every fuzzer initializes the RNG exactly * once before doing anything else, even if it is unused. * Default: 4. - * @param ZSTD_DEBUG: - * This is a parameter for the zstd library. Defining `ZSTD_DEBUG=1` + * @param DEBUGLEVEL: + * This is a parameter for the zstd library. Defining `DEBUGLEVEL=1` * enables assert() statements in the zstd library. Higher levels enable - * logging, so aren't recommended. Defining `ZSTD_DEBUG=1` is + * logging, so aren't recommended. Defining `DEBUGLEVEL=1` is * recommended. * @param MEM_FORCE_MEMORY_ACCESS: * This flag controls how the zstd library accesses unaligned memory. diff --git a/tests/fuzz/fuzz.py b/tests/fuzz/fuzz.py index b591e4f6734e3..8ce293a3a695f 100755 --- a/tests/fuzz/fuzz.py +++ b/tests/fuzz/fuzz.py @@ -13,6 +13,7 @@ import argparse import contextlib import os import re +import shlex import shutil import subprocess import sys @@ -147,15 +148,18 @@ def compiler_version(cc, cxx): """ cc_version_bytes = subprocess.check_output([cc, "--version"]) cxx_version_bytes = subprocess.check_output([cxx, "--version"]) - if cc_version_bytes.startswith(b'clang'): - assert(cxx_version_bytes.startswith(b'clang')) + compiler = None + version = None + if b'clang' in cc_version_bytes: + assert(b'clang' in cxx_version_bytes) compiler = 'clang' - if cc_version_bytes.startswith(b'gcc'): - assert(cxx_version_bytes.startswith(b'g++')) + elif b'gcc' in cc_version_bytes: + assert(b'gcc' in cxx_version_bytes) compiler = 'gcc' - version_regex = b'([0-9])+\.([0-9])+\.([0-9])+' - version_match = re.search(version_regex, cc_version_bytes) - version = tuple(int(version_match.group(i)) for i in range(1, 4)) + if compiler is not None: + version_regex = b'([0-9])+\.([0-9])+\.([0-9])+' + version_match = re.search(version_regex, cc_version_bytes) + version = tuple(int(version_match.group(i)) for i in range(1, 4)) return compiler, version @@ -248,7 +252,7 @@ def build_parser(args): dest='debug', type=int, default=1, - help='Set ZSTD_DEBUG (default: 1)') + help='Set DEBUGLEVEL (default: 1)') parser.add_argument( '--force-memory-access', dest='memory_access', @@ -265,7 +269,7 @@ def build_parser(args): '--disable-fuzzing-mode', dest='fuzzing_mode', action='store_false', - help='Do not define FUZZING_BUILD_MORE_UNSAFE_FOR_PRODUCTION') + help='Do not define FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION') parser.add_argument( '--enable-stateful-fuzzing', dest='stateful_fuzzing', @@ -346,16 +350,16 @@ def build(args): targets = args.TARGET cc = args.cc cxx = args.cxx - cppflags = [args.cppflags] - cflags = [args.cflags] - ldflags = [args.ldflags] - cxxflags = [args.cxxflags] - mflags = [args.mflags] if args.mflags else [] + cppflags = shlex.split(args.cppflags) + cflags = shlex.split(args.cflags) + ldflags = shlex.split(args.ldflags) + cxxflags = shlex.split(args.cxxflags) + mflags = shlex.split(args.mflags) # Flags to be added to both cflags and cxxflags common_flags = [] cppflags += [ - '-DZSTD_DEBUG={}'.format(args.debug), + '-DDEBUGLEVEL={}'.format(args.debug), '-DMEM_FORCE_MEMORY_ACCESS={}'.format(args.memory_access), '-DFUZZ_RNG_SEED_SIZE={}'.format(args.fuzz_rng_seed_size), ] @@ -399,7 +403,7 @@ def build(args): cppflags += ['-DSTATEFUL_FUZZING'] if args.fuzzing_mode: - cppflags += ['-DFUZZING_BUILD_MORE_UNSAFE_FOR_PRODUCTION'] + cppflags += ['-DFUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION'] if args.lib_fuzzing_engine == 'libregression.a': targets = ['libregression.a'] + targets @@ -750,11 +754,10 @@ def zip_cmd(args): for target in args.TARGET: # Zip the seed_corpus seed_corpus = abs_join(CORPORA_DIR, "{}_seed_corpus".format(target)) - seeds = [abs_join(seed_corpus, f) for f in os.listdir(seed_corpus)] zip_file = "{}.zip".format(seed_corpus) - cmd = ["zip", "-q", "-j", "-9", zip_file] - print(' '.join(cmd + [abs_join(seed_corpus, '*')])) - subprocess.check_call(cmd + seeds) + cmd = ["zip", "-r", "-q", "-j", "-9", zip_file, "."] + print(' '.join(cmd)) + subprocess.check_call(cmd, cwd=seed_corpus) def list_cmd(args): diff --git a/tests/fuzz/regression_driver.c b/tests/fuzz/regression_driver.c index 2b714d29e9dd6..1553d436ce038 100644 --- a/tests/fuzz/regression_driver.c +++ b/tests/fuzz/regression_driver.c @@ -16,7 +16,7 @@ #include <stdlib.h> int main(int argc, char const **argv) { - size_t const kMaxFileSize = (size_t)1 << 20; + size_t const kMaxFileSize = (size_t)1 << 27; int const kFollowLinks = 1; char *fileNamesBuf = NULL; char const **files = argv + 1; diff --git a/tests/fuzz/zstd_helpers.c b/tests/fuzz/zstd_helpers.c index 6fc38361b7adc..bf5eccff83c70 100644 --- a/tests/fuzz/zstd_helpers.c +++ b/tests/fuzz/zstd_helpers.c @@ -34,8 +34,7 @@ ZSTD_compressionParameters FUZZ_randomCParams(size_t srcSize, uint32_t *state) cParams.searchLog = FUZZ_rand32(state, ZSTD_SEARCHLOG_MIN, 9); cParams.searchLength = FUZZ_rand32(state, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX); - cParams.targetLength = FUZZ_rand32(state, ZSTD_TARGETLENGTH_MIN, - 512); + cParams.targetLength = FUZZ_rand32(state, 0, 512); cParams.strategy = FUZZ_rand32(state, ZSTD_fast, ZSTD_btultra); return ZSTD_adjustCParams(cParams, srcSize, 0); } @@ -72,6 +71,7 @@ void FUZZ_setRandomParameters(ZSTD_CCtx *cctx, size_t srcSize, uint32_t *state) setRand(cctx, ZSTD_p_contentSizeFlag, 0, 1, state); setRand(cctx, ZSTD_p_checksumFlag, 0, 1, state); setRand(cctx, ZSTD_p_dictIDFlag, 0, 1, state); + setRand(cctx, ZSTD_p_forceAttachDict, -2, 2, state); /* Select long distance matchig parameters */ setRand(cctx, ZSTD_p_enableLongDistanceMatching, 0, 1, state); setRand(cctx, ZSTD_p_ldmHashLog, ZSTD_HASHLOG_MIN, 16, state); diff --git a/tests/fuzzer.c b/tests/fuzzer.c index e97b841e8535c..5616285b9ed70 100644 --- a/tests/fuzzer.c +++ b/tests/fuzzer.c @@ -27,6 +27,7 @@ #include <string.h> /* strcmp */ #include <assert.h> #define ZSTD_STATIC_LINKING_ONLY /* ZSTD_compressContinue, ZSTD_compressBlock */ +#include "fse.h" #include "zstd.h" /* ZSTD_VERSION_STRING */ #include "zstd_errors.h" /* ZSTD_getErrorCode */ #include "zstdmt_compress.h" @@ -66,14 +67,20 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER; if (g_displayLevel>=4) fflush(stderr); } } +/*-******************************************************* +* Compile time test +*********************************************************/ #undef MIN #undef MAX +/* Declaring the function is it isn't unused */ +void FUZ_bug976(void); void FUZ_bug976(void) { /* these constants shall not depend on MIN() macro */ assert(ZSTD_HASHLOG_MAX < 31); assert(ZSTD_CHAINLOG_MAX < 31); } + /*-******************************************************* * Internal functions *********************************************************/ @@ -117,6 +124,13 @@ static unsigned FUZ_highbit32(U32 v32) #define CHECK(fn) { CHECK_V(err, fn); } #define CHECKPLUS(var, fn, more) { CHECK_V(var, fn); more; } +#define CHECK_EQ(lhs, rhs) { \ + if ((lhs) != (rhs)) { \ + DISPLAY("Error L%u => %s != %s ", __LINE__, #lhs, #rhs); \ + goto _output_error; \ + } \ +} + /*============================================= * Memory Tests @@ -167,13 +181,9 @@ static void FUZ_displayMallocStats(mallocCounter_t count) (U32)(count.totalMalloc >> 10)); } -static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part) +static int FUZ_mallocTests_internal(unsigned seed, double compressibility, unsigned part, + void* inBuffer, size_t inSize, void* outBuffer, size_t outSize) { - size_t const inSize = 64 MB + 16 MB + 4 MB + 1 MB + 256 KB + 64 KB; /* 85.3 MB */ - size_t const outSize = ZSTD_compressBound(inSize); - void* const inBuffer = malloc(inSize); - void* const outBuffer = malloc(outSize); - /* test only played in verbose mode, as they are long */ if (g_displayLevel<3) return 0; @@ -258,6 +268,28 @@ static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part) return 0; } +static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part) +{ + size_t const inSize = 64 MB + 16 MB + 4 MB + 1 MB + 256 KB + 64 KB; /* 85.3 MB */ + size_t const outSize = ZSTD_compressBound(inSize); + void* const inBuffer = malloc(inSize); + void* const outBuffer = malloc(outSize); + int result; + + /* Create compressible noise */ + if (!inBuffer || !outBuffer) { + DISPLAY("Not enough memory, aborting \n"); + exit(1); + } + + result = FUZ_mallocTests_internal(seed, compressibility, part, + inBuffer, inSize, outBuffer, outSize); + + free(inBuffer); + free(outBuffer); + return result; +} + #else static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part) @@ -303,9 +335,13 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(3, "OK : %s \n", errorString); } + DISPLAYLEVEL(3, "test%3i : min compression level : ", testNb++); + { int const mcl = ZSTD_minCLevel(); + DISPLAYLEVEL(3, "%i (OK) \n", mcl); + } DISPLAYLEVEL(3, "test%3i : compress %u bytes : ", testNb++, (U32)CNBuffSize); - { ZSTD_CCtx* cctx = ZSTD_createCCtx(); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); if (cctx==NULL) goto _output_error; CHECKPLUS(r, ZSTD_compressCCtx(cctx, compressedBuffer, compressedBufferSize, @@ -368,6 +404,12 @@ static int basicUnitTests(U32 seed, double compressibility) if (ZSTD_getErrorCode(r) != ZSTD_error_srcSize_wrong) goto _output_error; } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : decompress too large input : ", testNb++); + { size_t const r = ZSTD_decompress(decodedBuffer, CNBuffSize, compressedBuffer, compressedBufferSize); + if (!ZSTD_isError(r)) goto _output_error; + if (ZSTD_getErrorCode(r) != ZSTD_error_srcSize_wrong) goto _output_error; } + DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3d : check CCtx size after compressing empty input : ", testNb++); { ZSTD_CCtx* cctx = ZSTD_createCCtx(); size_t const r = ZSTD_compressCCtx(cctx, compressedBuffer, compressedBufferSize, NULL, 0, 19); @@ -394,14 +436,80 @@ static int basicUnitTests(U32 seed, double compressibility) } DISPLAYLEVEL(3, "OK \n"); - DISPLAYLEVEL(3, "test%3d : large window log smaller data : ", testNb++); + DISPLAYLEVEL(3, "test%3d : re-using a CCtx should compress the same : ", testNb++); + { int i; + for (i=0; i<20; i++) + ((char*)CNBuffer)[i] = (char)i; /* ensure no match during initial section */ + memcpy((char*)CNBuffer + 20, CNBuffer, 10); /* create one match, starting from beginning of sample, which is the difficult case (see #1241) */ + for (i=1; i<=19; i++) { + ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + size_t size1, size2; + DISPLAYLEVEL(5, "l%i ", i); + size1 = ZSTD_compressCCtx(cctx, compressedBuffer, compressedBufferSize, CNBuffer, 30, i); + CHECK_Z(size1); + size2 = ZSTD_compressCCtx(cctx, compressedBuffer, compressedBufferSize, CNBuffer, 30, i); + CHECK_Z(size2); + CHECK_EQ(size1, size2); + + ZSTD_freeCCtx(cctx); + } + } + DISPLAYLEVEL(3, "OK \n"); + + DISPLAYLEVEL(3, "test%3d : ZSTD_CCtx_getParameter() : ", testNb++); { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); - ZSTD_parameters params = ZSTD_getParams(1, ZSTD_CONTENTSIZE_UNKNOWN, 0); - size_t const nbCompressions = (1U << 31) / CNBuffSize + 1; - size_t i; + ZSTD_outBuffer out = {NULL, 0, 0}; + ZSTD_inBuffer in = {NULL, 0, 0}; + unsigned value; + + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 3); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, 0); + CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_p_hashLog, ZSTD_HASHLOG_MIN)); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 3); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, ZSTD_HASHLOG_MIN); + CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, 7)); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 7); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, ZSTD_HASHLOG_MIN); + /* Start a compression job */ + ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_continue); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 7); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, ZSTD_HASHLOG_MIN); + /* Reset the CCtx */ + ZSTD_CCtx_reset(cctx); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 7); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, ZSTD_HASHLOG_MIN); + /* Reset the parameters */ + ZSTD_CCtx_resetParameters(cctx); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_compressionLevel, &value)); + CHECK_EQ(value, 3); + CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_p_hashLog, &value)); + CHECK_EQ(value, 0); + + ZSTD_freeCCtx(cctx); + } + DISPLAYLEVEL(3, "OK \n"); + + /* this test is really too long, and should be made faster */ + DISPLAYLEVEL(3, "test%3d : overflow protection with large windowLog : ", testNb++); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + ZSTD_parameters params = ZSTD_getParams(-9, ZSTD_CONTENTSIZE_UNKNOWN, 0); + size_t const nbCompressions = ((1U << 31) / CNBuffSize) + 1; /* ensure U32 overflow protection is triggered */ + size_t cnb; + assert(cctx != NULL); params.fParams.contentSizeFlag = 0; params.cParams.windowLog = ZSTD_WINDOWLOG_MAX; - for (i = 0; i < nbCompressions; ++i) { + for (cnb = 0; cnb < nbCompressions; ++cnb) { + DISPLAYLEVEL(6, "run %zu / %zu \n", cnb, nbCompressions); CHECK_Z( ZSTD_compressBegin_advanced(cctx, NULL, 0, params, ZSTD_CONTENTSIZE_UNKNOWN) ); /* re-use same parameters */ CHECK_Z( ZSTD_compressEnd(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize) ); } @@ -409,6 +517,39 @@ static int basicUnitTests(U32 seed, double compressibility) } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3d : size down context : ", testNb++); + { ZSTD_CCtx* const largeCCtx = ZSTD_createCCtx(); + assert(largeCCtx != NULL); + CHECK_Z( ZSTD_compressBegin(largeCCtx, 19) ); /* streaming implies ZSTD_CONTENTSIZE_UNKNOWN, which maximizes memory usage */ + CHECK_Z( ZSTD_compressEnd(largeCCtx, compressedBuffer, compressedBufferSize, CNBuffer, 1) ); + { size_t const largeCCtxSize = ZSTD_sizeof_CCtx(largeCCtx); /* size of context must be measured after compression */ + { ZSTD_CCtx* const smallCCtx = ZSTD_createCCtx(); + assert(smallCCtx != NULL); + CHECK_Z(ZSTD_compressCCtx(smallCCtx, compressedBuffer, compressedBufferSize, CNBuffer, 1, 1)); + { size_t const smallCCtxSize = ZSTD_sizeof_CCtx(smallCCtx); + DISPLAYLEVEL(5, "(large) %zuKB > 32*%zuKB (small) : ", + largeCCtxSize>>10, smallCCtxSize>>10); + assert(largeCCtxSize > 32* smallCCtxSize); /* note : "too large" definition is handled within zstd_compress.c . + * make this test case extreme, so that it doesn't depend on a possibly fluctuating definition */ + } + ZSTD_freeCCtx(smallCCtx); + } + { U32 const maxNbAttempts = 1100; /* nb of usages before triggering size down is handled within zstd_compress.c. + * currently defined as 128x, but could be adjusted in the future. + * make this test long enough so that it's not too much tied to the current definition within zstd_compress.c */ + U32 u; + for (u=0; u<maxNbAttempts; u++) { + CHECK_Z(ZSTD_compressCCtx(largeCCtx, compressedBuffer, compressedBufferSize, CNBuffer, 1, 1)); + if (ZSTD_sizeof_CCtx(largeCCtx) < largeCCtxSize) break; /* sized down */ + } + DISPLAYLEVEL(5, "size down after %u attempts : ", u); + if (u==maxNbAttempts) goto _output_error; /* no sizedown happened */ + } + } + ZSTD_freeCCtx(largeCCtx); + } + DISPLAYLEVEL(3, "OK \n"); + /* Static CCtx tests */ #define STATIC_CCTX_LEVEL 3 DISPLAYLEVEL(3, "test%3i : create static CCtx for level %u :", testNb++, STATIC_CCTX_LEVEL); @@ -1024,12 +1165,40 @@ static int basicUnitTests(U32 seed, double compressibility) ZSTD_freeCCtx(cctx); } + /* negative compression level test : ensure simple API and advanced API produce same result */ + DISPLAYLEVEL(3, "test%3i : negative compression level : ", testNb++); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + size_t const srcSize = CNBuffSize / 5; + int const compressionLevel = -1; + + assert(cctx != NULL); + { ZSTD_parameters const params = ZSTD_getParams(compressionLevel, srcSize, 0); + size_t const cSize_1pass = ZSTD_compress_advanced(cctx, + compressedBuffer, compressedBufferSize, + CNBuffer, srcSize, + NULL, 0, + params); + if (ZSTD_isError(cSize_1pass)) goto _output_error; + + CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, (unsigned)compressionLevel) ); + { ZSTD_inBuffer in = { CNBuffer, srcSize, 0 }; + ZSTD_outBuffer out = { compressedBuffer, compressedBufferSize, 0 }; + size_t const compressionResult = ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end); + DISPLAYLEVEL(5, "simple=%zu vs %zu=advanced : ", cSize_1pass, out.pos); + if (ZSTD_isError(compressionResult)) goto _output_error; + if (out.pos != cSize_1pass) goto _output_error; + } } + ZSTD_freeCCtx(cctx); + } + DISPLAYLEVEL(3, "OK \n"); + /* parameters order test */ { size_t const inputSize = CNBuffSize / 2; U64 xxh64; - { ZSTD_CCtx* cctx = ZSTD_createCCtx(); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); DISPLAYLEVEL(3, "test%3i : parameters in order : ", testNb++); + assert(cctx != NULL); CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, 2) ); CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_p_enableLongDistanceMatching, 1) ); CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_p_windowLog, 18) ); @@ -1085,9 +1254,13 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(3, "OK : %s \n", ZSTD_getErrorName(decodeResult)); } - DISPLAYLEVEL(3, "test%3i : decompress with magic-less instruction : ", testNb++); + DISPLAYLEVEL(3, "test%3i : decompress of magic-less frame : ", testNb++); ZSTD_DCtx_reset(dctx); CHECK( ZSTD_DCtx_setFormat(dctx, ZSTD_f_zstd1_magicless) ); + { ZSTD_frameHeader zfh; + size_t const zfhrt = ZSTD_getFrameHeader_advanced(&zfh, compressedBuffer, cSize, ZSTD_f_zstd1_magicless); + if (zfhrt != 0) goto _output_error; + } { ZSTD_inBuffer in = { compressedBuffer, cSize, 0 }; ZSTD_outBuffer out = { decodedBuffer, CNBuffSize, 0 }; size_t const result = ZSTD_decompress_generic(dctx, &out, &in); @@ -1120,6 +1293,20 @@ static int basicUnitTests(U32 seed, double compressibility) if (r != blockSize) goto _output_error; } DISPLAYLEVEL(3, "OK \n"); + /* very long stream of block compression */ + DISPLAYLEVEL(3, "test%3i : Huge block streaming compression test : ", testNb++); + CHECK( ZSTD_compressBegin(cctx, -99) ); /* we just want to quickly overflow internal U32 index */ + CHECK( ZSTD_getBlockSize(cctx) >= blockSize); + { U64 const toCompress = 5000000000ULL; /* > 4 GB */ + U64 compressed = 0; + while (compressed < toCompress) { + size_t const blockCSize = ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), CNBuffer, blockSize); + if (ZSTD_isError(cSize)) goto _output_error; + compressed += blockCSize; + } + } + DISPLAYLEVEL(3, "OK \n"); + /* dictionary block compression */ DISPLAYLEVEL(3, "test%3i : Dictionary Block compression test : ", testNb++); CHECK( ZSTD_compressBegin_usingDict(cctx, CNBuffer, dictSize, 5) ); @@ -1142,6 +1329,15 @@ static int basicUnitTests(U32 seed, double compressibility) if (r != blockSize) goto _output_error; } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : Block compression with CDict : ", testNb++); + { ZSTD_CDict* const cdict = ZSTD_createCDict(CNBuffer, dictSize, 3); + if (cdict==NULL) goto _output_error; + CHECK( ZSTD_compressBegin_usingCDict(cctx, cdict) ); + CHECK( ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), (char*)CNBuffer+dictSize, blockSize) ); + ZSTD_freeCDict(cdict); + } + DISPLAYLEVEL(3, "OK \n"); + ZSTD_freeCCtx(cctx); } ZSTD_freeDCtx(dctx); @@ -1199,6 +1395,24 @@ static int basicUnitTests(U32 seed, double compressibility) ((BYTE*)CNBuffer)[i+1] = _3BytesSeqs[id][1]; ((BYTE*)CNBuffer)[i+2] = _3BytesSeqs[id][2]; } } } + DISPLAYLEVEL(3, "test%3i : growing nbSeq : ", testNb++); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + size_t const maxNbSeq = _3BYTESTESTLENGTH / 3; + size_t const bound = ZSTD_compressBound(_3BYTESTESTLENGTH); + size_t nbSeq = 1; + while (nbSeq <= maxNbSeq) { + CHECK(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, nbSeq * 3, 19)); + /* Check every sequence for the first 100, then skip more rapidly. */ + if (nbSeq < 100) { + ++nbSeq; + } else { + nbSeq += (nbSeq >> 2); + } + } + ZSTD_freeCCtx(cctx); + } + DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : compress lots 3-bytes sequences : ", testNb++); { CHECK_V(r, ZSTD_compress(compressedBuffer, ZSTD_compressBound(_3BYTESTESTLENGTH), CNBuffer, _3BYTESTESTLENGTH, 19) ); @@ -1210,8 +1424,26 @@ static int basicUnitTests(U32 seed, double compressibility) if (r != _3BYTESTESTLENGTH) goto _output_error; } DISPLAYLEVEL(3, "OK \n"); - DISPLAYLEVEL(3, "test%3i : incompressible data and ill suited dictionary : ", testNb++); + + DISPLAYLEVEL(3, "test%3i : growing literals buffer : ", testNb++); RDG_genBuffer(CNBuffer, CNBuffSize, 0.0, 0.1, seed); + { ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + size_t const bound = ZSTD_compressBound(CNBuffSize); + size_t size = 1; + while (size <= CNBuffSize) { + CHECK(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, size, 3)); + /* Check every size for the first 100, then skip more rapidly. */ + if (size < 100) { + ++size; + } else { + size += (size >> 2); + } + } + ZSTD_freeCCtx(cctx); + } + DISPLAYLEVEL(3, "OK \n"); + + DISPLAYLEVEL(3, "test%3i : incompressible data and ill suited dictionary : ", testNb++); { /* Train a dictionary on low characters */ size_t dictSize = 16 KB; void* const dictBuffer = malloc(dictSize); @@ -1286,6 +1518,24 @@ static int basicUnitTests(U32 seed, double compressibility) } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : testing FSE_normalizeCount() PR#1255: ", testNb++); + { + short norm[32]; + unsigned count[32]; + unsigned const tableLog = 5; + size_t const nbSeq = 32; + unsigned const maxSymbolValue = 31; + size_t i; + + for (i = 0; i < 32; ++i) + count[i] = 1; + /* Calling FSE_normalizeCount() on a uniform distribution should not + * cause a division by zero. + */ + FSE_normalizeCount(norm, tableLog, count, nbSeq, maxSymbolValue); + } + DISPLAYLEVEL(3, "OK \n"); + _end: free(CNBuffer); free(compressedBuffer); @@ -1359,7 +1609,6 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD size_t const dstBufferSize = (size_t)1<<maxSampleLog; size_t const cBufferSize = ZSTD_compressBound(dstBufferSize); BYTE* cNoiseBuffer[5]; - BYTE* srcBuffer; /* jumping pointer */ BYTE* const cBuffer = (BYTE*) malloc (cBufferSize); BYTE* const dstBuffer = (BYTE*) malloc (dstBufferSize); BYTE* const mirrorBuffer = (BYTE*) malloc (dstBufferSize); @@ -1368,7 +1617,7 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD ZSTD_DCtx* const dctx = ZSTD_createDCtx(); U32 result = 0; U32 testNb = 0; - U32 coreSeed = seed, lseed = 0; + U32 coreSeed = seed; UTIL_time_t const startClock = UTIL_getTime(); U64 const maxClockSpan = maxDurationS * SEC_TO_MICRO; int const cLevelLimiter = bigTests ? 3 : 2; @@ -1389,13 +1638,14 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD RDG_genBuffer(cNoiseBuffer[2], srcBufferSize, compressibility, 0., coreSeed); RDG_genBuffer(cNoiseBuffer[3], srcBufferSize, 0.95, 0., coreSeed); /* highly compressible */ RDG_genBuffer(cNoiseBuffer[4], srcBufferSize, 1.00, 0., coreSeed); /* sparse content */ - srcBuffer = cNoiseBuffer[2]; /* catch up testNb */ for (testNb=1; testNb < startTest; testNb++) FUZ_rand(&coreSeed); /* main test loop */ for ( ; (testNb <= nbTests) || (UTIL_clockSpanMicro(startClock) < maxClockSpan); testNb++ ) { + BYTE* srcBuffer; /* jumping pointer */ + U32 lseed; size_t sampleSize, maxTestSize, totalTestSize; size_t cSize, totalCSize, totalGenSize; U64 crcOrig; @@ -1626,11 +1876,9 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD CHECK (totalGenSize != totalTestSize, "streaming decompressed data : wrong size") CHECK (totalCSize != cSize, "compressed data should be fully read") { U64 const crcDest = XXH64(dstBuffer, totalTestSize, 0); - if (crcDest!=crcOrig) { - size_t const errorPos = findDiff(mirrorBuffer, dstBuffer, totalTestSize); - CHECK (1, "streaming decompressed data corrupted : byte %u / %u (%02X!=%02X)", - (U32)errorPos, (U32)totalTestSize, dstBuffer[errorPos], mirrorBuffer[errorPos]); - } } + CHECK(crcOrig != crcDest, "streaming decompressed data corrupted (pos %u / %u)", + (U32)findDiff(mirrorBuffer, dstBuffer, totalTestSize), (U32)totalTestSize); + } } /* for ( ; (testNb <= nbTests) */ DISPLAY("\r%u fuzzer tests completed \n", testNb-1); diff --git a/tests/gzip/Makefile b/tests/gzip/Makefile index 40a0ba97d2b37..c5d67206b99d9 100644 --- a/tests/gzip/Makefile +++ b/tests/gzip/Makefile @@ -33,7 +33,7 @@ clean: #------------------------------------------------------------------------------ -# validated only for Linux, OSX, Hurd and some BSD targets +# validated only for Linux, macOS, Hurd and some BSD targets #------------------------------------------------------------------------------ ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU FreeBSD DragonFly NetBSD)) diff --git a/tests/legacy.c b/tests/legacy.c index 847e1d25e96b7..e1cf82f2f9d83 100644 --- a/tests/legacy.c +++ b/tests/legacy.c @@ -36,7 +36,7 @@ size_t const COMPRESSED_SIZE = 917; const char* const EXPECTED; /* content is at end of file */ -int testSimpleAPI(void) +static int testSimpleAPI(void) { size_t const size = strlen(EXPECTED); char* const output = malloc(size); @@ -71,7 +71,8 @@ int testSimpleAPI(void) return 0; } -int testStreamingAPI(void) + +static int testStreamingAPI(void) { size_t const outBuffSize = ZSTD_DStreamOutSize(); char* const outBuff = malloc(outBuffSize); diff --git a/tests/libzstd_partial_builds.sh b/tests/libzstd_partial_builds.sh new file mode 100755 index 0000000000000..34d8ea55231eb --- /dev/null +++ b/tests/libzstd_partial_builds.sh @@ -0,0 +1,36 @@ +#!/bin/sh -e + +die() { + $ECHO "$@" 1>&2 + exit 1 +} + +DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )" + +INTOVOID="/dev/null" +case "$OS" in + Windows*) + INTOVOID="NUL" + ;; +esac + +ZSTD_LIB_COMPRESSION=0 CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID +nm $DIR/../lib/libzstd.a | grep ".*\.o:" > tmplog +! grep -q "zstd_compress" tmplog && grep -q "zstd_decompress" tmplog && ! grep -q "dict" tmplog && grep -q "zstd_v" tmplog && ! grep -q "zbuff" tmplog && make clean && rm -f tmplog || die "Compression macro failed" + + +ZSTD_LIB_DECOMPRESSION=0 CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID +nm $DIR/../lib/libzstd.a | grep ".*\.o:" > tmplog +grep -q "zstd_compress" tmplog && ! grep -q "zstd_decompress" tmplog && grep -q "dict" tmplog && ! grep -q "zstd_v" tmplog && ! grep -q "zbuff" tmplog && make clean && rm -f tmplog || die "Decompression macro failed" + +ZSTD_LIB_DEPRECATED=0 CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID +nm $DIR/../lib/libzstd.a | grep ".*\.o:" > tmplog +grep -q "zstd_compress" tmplog && grep -q "zstd_decompress" tmplog && grep -q "dict" tmplog && grep -q "zstd_v" tmplog && ! grep -q "zbuff" tmplog && make clean && rm -f tmplog || die "Deprecated macro failed" + +ZSTD_LIB_DICTBUILDER=0 CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID +nm $DIR/../lib/libzstd.a | grep ".*\.o:" > tmplog +grep -q "zstd_compress" tmplog && grep -q "zstd_decompress" tmplog && ! grep -q "dict" tmplog && grep -q "zstd_v" tmplog && grep -q "zbuff" tmplog && make clean && rm -f tmplog || die "Dictbuilder macro failed" + +ZSTD_LIB_DECOMPRESSION=0 ZSTD_LIB_DICTBUILDER=0 CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID +nm $DIR/../lib/libzstd.a | grep ".*\.o:" > tmplog +grep -q "zstd_compress" tmplog && ! grep -q "zstd_decompress" tmplog && ! grep -q "dict" tmplog && ! grep -q "zstd_v" tmplog && ! grep -q "zbuff" tmplog && make clean && rm -f tmplog || die "Multi-macro failed"
\ No newline at end of file diff --git a/tests/longmatch.c b/tests/longmatch.c index ed3861571d9b7..1271e9ab1039c 100644 --- a/tests/longmatch.c +++ b/tests/longmatch.c @@ -17,25 +17,25 @@ #define ZSTD_STATIC_LINKING_ONLY #include "zstd.h" -int compress(ZSTD_CStream *ctx, ZSTD_outBuffer out, const void *data, size_t size) { +static int +compress(ZSTD_CStream *ctx, ZSTD_outBuffer out, const void *data, size_t size) +{ ZSTD_inBuffer in = { data, size, 0 }; while (in.pos < in.size) { ZSTD_outBuffer tmp = out; const size_t rc = ZSTD_compressStream(ctx, &tmp, &in); - if (ZSTD_isError(rc)) { - return 1; - } + if (ZSTD_isError(rc)) return 1; } - { - ZSTD_outBuffer tmp = out; + { ZSTD_outBuffer tmp = out; const size_t rc = ZSTD_flushStream(ctx, &tmp); if (rc != 0) { return 1; } } return 0; } -int main(int argc, const char** argv) { - ZSTD_CStream *ctx; +int main(int argc, const char** argv) +{ + ZSTD_CStream* ctx; ZSTD_parameters params; size_t rc; unsigned windowLog; diff --git a/tests/paramgrill.c b/tests/paramgrill.c index 13b102b2d042d..7a4be854a46bf 100644 --- a/tests/paramgrill.c +++ b/tests/paramgrill.c @@ -17,7 +17,7 @@ #include <stdio.h> /* fprintf, fopen, ftello64 */ #include <string.h> /* strcmp */ #include <math.h> /* log */ -#include <time.h> +#include <assert.h> #include "mem.h" #define ZSTD_STATIC_LINKING_ONLY /* ZSTD_parameters, ZSTD_estimateCCtxSize */ @@ -25,6 +25,9 @@ #include "datagen.h" #include "xxhash.h" #include "util.h" +#include "bench.h" +#include "zstd_errors.h" +#include "zstd_internal.h" /* should not be needed */ /*-************************************ @@ -32,25 +35,15 @@ **************************************/ #define PROGRAM_DESCRIPTION "ZSTD parameters tester" #define AUTHOR "Yann Collet" -#define WELCOME_MESSAGE "*** %s %s %i-bits, by %s (%s) ***\n", PROGRAM_DESCRIPTION, ZSTD_VERSION_STRING, (int)(sizeof(void*)*8), AUTHOR, __DATE__ +#define WELCOME_MESSAGE "*** %s %s %i-bits, by %s ***\n", PROGRAM_DESCRIPTION, ZSTD_VERSION_STRING, (int)(sizeof(void*)*8), AUTHOR - -#define KB *(1<<10) -#define MB *(1<<20) -#define GB *(1ULL<<30) - -#define NBLOOPS 2 -#define TIMELOOP (2 * SEC_TO_MICRO) - -#define NB_LEVELS_TRACKED 30 +#define TIMELOOP_NANOSEC (1*1000000000ULL) /* 1 second */ +#define NB_LEVELS_TRACKED 22 /* ensured being >= ZSTD_maxCLevel() in BMK_init_level_constraints() */ static const size_t maxMemory = (sizeof(size_t)==4) ? (2 GB - 64 MB) : (size_t)(1ULL << ((sizeof(size_t)*8)-31)); #define COMPRESSIBILITY_DEFAULT 0.50 -static const size_t sampleSize = 10000000; -static const double g_grillDuration_s = 90000; /* about 24 hours */ -static const U64 g_maxParamTime = 15 * SEC_TO_MICRO; static const U64 g_maxVariationTime = 60 * SEC_TO_MICRO; static const int g_maxNbVariations = 64; @@ -59,38 +52,301 @@ static const int g_maxNbVariations = 64; * Macros **************************************/ #define DISPLAY(...) fprintf(stderr, __VA_ARGS__) +#define DISPLAYLEVEL(n, ...) if(g_displayLevel >= n) { fprintf(stderr, __VA_ARGS__); } +#define DEBUGOUTPUT(...) { if (DEBUG) DISPLAY(__VA_ARGS__); } + +#define TIMED 0 +#ifndef DEBUG +# define DEBUG 0 +#endif #undef MIN #undef MAX #define MIN(a,b) ( (a) < (b) ? (a) : (b) ) #define MAX(a,b) ( (a) > (b) ? (a) : (b) ) +#define CUSTOM_LEVEL 99 +#define BASE_CLEVEL 1 + +#define FADT_MIN 0 +#define FADT_MAX ((U32)-1) + +#define WLOG_RANGE (ZSTD_WINDOWLOG_MAX - ZSTD_WINDOWLOG_MIN + 1) +#define CLOG_RANGE (ZSTD_CHAINLOG_MAX - ZSTD_CHAINLOG_MIN + 1) +#define HLOG_RANGE (ZSTD_HASHLOG_MAX - ZSTD_HASHLOG_MIN + 1) +#define SLOG_RANGE (ZSTD_SEARCHLOG_MAX - ZSTD_SEARCHLOG_MIN + 1) +#define SLEN_RANGE (ZSTD_SEARCHLENGTH_MAX - ZSTD_SEARCHLENGTH_MIN + 1) +#define TLEN_RANGE 17 +#define STRT_RANGE (ZSTD_btultra - ZSTD_fast + 1) +#define FADT_RANGE 3 + +#define CHECKTIME(r) { if(BMK_timeSpan(g_time) > g_timeLimit_s) { DEBUGOUTPUT("Time Limit Reached\n"); return r; } } +#define CHECKTIMEGT(ret, val, _gototag) {if(BMK_timeSpan(g_time) > g_timeLimit_s) { DEBUGOUTPUT("Time Limit Reached\n"); ret = val; goto _gototag; } } + +#define PARAM_UNSET ((U32)-2) /* can't be -1 b/c fadt uses -1 */ + +static const char* g_stratName[ZSTD_btultra+1] = { + "(none) ", "ZSTD_fast ", "ZSTD_dfast ", + "ZSTD_greedy ", "ZSTD_lazy ", "ZSTD_lazy2 ", + "ZSTD_btlazy2 ", "ZSTD_btopt ", "ZSTD_btultra "}; + +static const U32 tlen_table[TLEN_RANGE] = { 0, 1, 2, 4, 6, 8, 12, 16, 24, 32, 48, 64, 96, 128, 256, 512, 999 }; /*-************************************ -* Benchmark Parameters +* Setup for Adding new params **************************************/ -static U32 g_nbIterations = NBLOOPS; -static double g_compressibility = COMPRESSIBILITY_DEFAULT; + +/* indices for each of the variables */ +typedef enum { + wlog_ind = 0, + clog_ind = 1, + hlog_ind = 2, + slog_ind = 3, + slen_ind = 4, + tlen_ind = 5, + strt_ind = 6, + fadt_ind = 7, /* forceAttachDict */ + NUM_PARAMS = 8 +} varInds_t; + +typedef struct { + U32 vals[NUM_PARAMS]; +} paramValues_t; + +/* maximum value of parameters */ +static const U32 mintable[NUM_PARAMS] = + { ZSTD_WINDOWLOG_MIN, ZSTD_CHAINLOG_MIN, ZSTD_HASHLOG_MIN, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLENGTH_MIN, ZSTD_TARGETLENGTH_MIN, ZSTD_fast, FADT_MIN }; + +/* minimum value of parameters */ +static const U32 maxtable[NUM_PARAMS] = + { ZSTD_WINDOWLOG_MAX, ZSTD_CHAINLOG_MAX, ZSTD_HASHLOG_MAX, ZSTD_SEARCHLOG_MAX, ZSTD_SEARCHLENGTH_MAX, ZSTD_TARGETLENGTH_MAX, ZSTD_btultra, FADT_MAX }; + +/* # of values parameters can take on */ +static const U32 rangetable[NUM_PARAMS] = + { WLOG_RANGE, CLOG_RANGE, HLOG_RANGE, SLOG_RANGE, SLEN_RANGE, TLEN_RANGE, STRT_RANGE, FADT_RANGE }; + +/* ZSTD_cctxSetParameter() index to set */ +static const ZSTD_cParameter cctxSetParamTable[NUM_PARAMS] = + { ZSTD_p_windowLog, ZSTD_p_chainLog, ZSTD_p_hashLog, ZSTD_p_searchLog, ZSTD_p_minMatch, ZSTD_p_targetLength, ZSTD_p_compressionStrategy, ZSTD_p_forceAttachDict }; + +/* names of parameters */ +static const char* g_paramNames[NUM_PARAMS] = + { "windowLog", "chainLog", "hashLog","searchLog", "searchLength", "targetLength", "strategy", "forceAttachDict" }; + +/* shortened names of parameters */ +static const char* g_shortParamNames[NUM_PARAMS] = + { "wlog", "clog", "hlog","slog", "slen", "tlen", "strt", "fadt" }; + +/* maps value from { 0 to rangetable[param] - 1 } to valid paramvalues */ +static U32 rangeMap(varInds_t param, int ind) { + ind = MAX(MIN(ind, (int)rangetable[param] - 1), 0); + switch(param) { + case tlen_ind: + return tlen_table[ind]; + case fadt_ind: /* 0, 1, 2 -> -1, 0, 1 */ + return ind - 1; + case wlog_ind: /* using default: triggers -Wswitch-enum */ + case clog_ind: + case hlog_ind: + case slog_ind: + case slen_ind: + case strt_ind: + return mintable[param] + ind; + case NUM_PARAMS: + DISPLAY("Error, not a valid param\n "); + return (U32)-1; + } + return 0; /* should never happen, stop compiler warnings */ +} + +/* inverse of rangeMap */ +static int invRangeMap(varInds_t param, U32 value) { + value = MIN(MAX(mintable[param], value), maxtable[param]); + switch(param) { + case tlen_ind: /* bin search */ + { + int lo = 0; + int hi = TLEN_RANGE; + while(lo < hi) { + int mid = (lo + hi) / 2; + if(tlen_table[mid] < value) { + lo = mid + 1; + } if(tlen_table[mid] == value) { + return mid; + } else { + hi = mid; + } + } + return lo; + } + case fadt_ind: + return (int)value + 1; + case wlog_ind: + case clog_ind: + case hlog_ind: + case slog_ind: + case slen_ind: + case strt_ind: + return value - mintable[param]; + case NUM_PARAMS: + DISPLAY("Error, not a valid param\n "); + return -2; + } + return 0; /* should never happen, stop compiler warnings */ +} + +/* display of params */ +static void displayParamVal(FILE* f, varInds_t param, U32 value, int width) { + switch(param) { + case fadt_ind: if(width) { fprintf(f, "%*d", width, (int)value); } else { fprintf(f, "%d", (int)value); } break; + case strt_ind: if(width) { fprintf(f, "%*s", width, g_stratName[value]); } else { fprintf(f, "%s", g_stratName[value]); } break; + case wlog_ind: + case clog_ind: + case hlog_ind: + case slog_ind: + case slen_ind: + case tlen_ind: if(width) { fprintf(f, "%*u", width, value); } else { fprintf(f, "%u", value); } break; + case NUM_PARAMS: + DISPLAY("Error, not a valid param\n "); break; + } +} + + +/*-************************************ +* Benchmark Parameters/Global Variables +**************************************/ + +typedef BYTE U8; + +/* General Utility */ +static U32 g_timeLimit_s = 99999; /* about 27 hours */ +static UTIL_time_t g_time; /* to be used to compare solution finding speeds to compare to original */ static U32 g_blockSize = 0; static U32 g_rand = 1; + +/* Display */ +static int g_displayLevel = 3; +static BYTE g_silenceParams[NUM_PARAMS]; + +/* Mode Selection */ static U32 g_singleRun = 0; +static U32 g_optimizer = 0; +static int g_optmode = 0; + +/* For cLevel Table generation */ static U32 g_target = 0; static U32 g_noSeed = 0; -static ZSTD_compressionParameters g_params = { 0, 0, 0, 0, 0, 0, ZSTD_greedy }; -void BMK_SetNbIterations(int nbLoops) -{ - g_nbIterations = nbLoops; - DISPLAY("- %u iterations -\n", g_nbIterations); -} +/* For optimizer */ +static paramValues_t g_params; /* Initialized at the beginning of main w/ emptyParams() function */ +static double g_ratioMultiplier = 5.; +static U32 g_strictness = PARAM_UNSET; /* range 1 - 100, measure of how strict */ +static BMK_benchResult_t g_lvltarget; + +typedef enum { + directMap, + xxhashMap, + noMemo +} memoTableType_t; + +typedef struct { + memoTableType_t tableType; + BYTE* table; + size_t tableLen; + varInds_t varArray[NUM_PARAMS]; + size_t varLen; +} memoTable_t; + +typedef struct { + BMK_benchResult_t result; + paramValues_t params; +} winnerInfo_t; + +typedef struct { + U32 cSpeed; /* bytes / sec */ + U32 dSpeed; + U32 cMem; /* bytes */ +} constraint_t; + +typedef struct winner_ll_node winner_ll_node; +struct winner_ll_node { + winnerInfo_t res; + winner_ll_node* next; +}; + +static winner_ll_node* g_winners; /* linked list sorted ascending by cSize & cSpeed */ + +/* + * Additional Global Variables (Defined Above Use) + * g_level_constraint + * g_alreadyTested + * g_maxTries + * g_clockGranularity + */ /*-******************************************************* -* Private functions +* General Util Functions *********************************************************/ -/* accuracy in seconds only, span can be multiple years */ -static double BMK_timeSpan(time_t tStart) { return difftime(time(NULL), tStart); } +/* nullified useless params, to ensure count stats */ +/* cleans up params for memoizing / display */ +static paramValues_t sanitizeParams(paramValues_t params) +{ + if (params.vals[strt_ind] == ZSTD_fast) + params.vals[clog_ind] = 0, params.vals[slog_ind] = 0; + if (params.vals[strt_ind] == ZSTD_dfast) + params.vals[slog_ind] = 0; + if (params.vals[strt_ind] != ZSTD_btopt && params.vals[strt_ind] != ZSTD_btultra && params.vals[strt_ind] != ZSTD_fast) + params.vals[tlen_ind] = 0; + + return params; +} + +static ZSTD_compressionParameters pvalsToCParams(paramValues_t p) { + ZSTD_compressionParameters c; + memset(&c, 0, sizeof(ZSTD_compressionParameters)); + c.windowLog = p.vals[wlog_ind]; + c.chainLog = p.vals[clog_ind]; + c.hashLog = p.vals[hlog_ind]; + c.searchLog = p.vals[slog_ind]; + c.searchLength = p.vals[slen_ind]; + c.targetLength = p.vals[tlen_ind]; + c.strategy = p.vals[strt_ind]; + /* no forceAttachDict */ + return c; +} + +static paramValues_t cParamsToPVals(ZSTD_compressionParameters c) { + paramValues_t p; + varInds_t i; + p.vals[wlog_ind] = c.windowLog; + p.vals[clog_ind] = c.chainLog; + p.vals[hlog_ind] = c.hashLog; + p.vals[slog_ind] = c.searchLog; + p.vals[slen_ind] = c.searchLength; + p.vals[tlen_ind] = c.targetLength; + p.vals[strt_ind] = c.strategy; + + /* set all other params to their minimum value */ + for(i = strt_ind + 1; i < NUM_PARAMS; i++) { + p.vals[i] = mintable[i]; + } + return p; +} + +/* equivalent of ZSTD_adjustCParams for paramValues_t */ +static paramValues_t adjustParams(paramValues_t p, const size_t maxBlockSize, const size_t dictSize) { + paramValues_t ot = p; + varInds_t i; + p = cParamsToPVals(ZSTD_adjustCParams(pvalsToCParams(p), maxBlockSize, dictSize)); + if(!dictSize) { p.vals[fadt_ind] = 0; } + /* retain value of all other parameters */ + for(i = strt_ind + 1; i < NUM_PARAMS; i++) { + p.vals[i] = ot.vals[i]; + } + return p; +} static size_t BMK_findMaxMem(U64 requiredMem) { @@ -100,23 +356,25 @@ static size_t BMK_findMaxMem(U64 requiredMem) requiredMem = (((requiredMem >> 26) + 1) << 26); if (requiredMem > maxMemory) requiredMem = maxMemory; - requiredMem += 2*step; - while (!testmem) { - requiredMem -= step; + requiredMem += 2 * step; + while (!testmem && requiredMem > 0) { testmem = malloc ((size_t)requiredMem); + requiredMem -= step; } free (testmem); - return (size_t) (requiredMem - step); + return (size_t) requiredMem; } +/* accuracy in seconds only, span can be multiple years */ +static U32 BMK_timeSpan(const UTIL_time_t tStart) { return (U32)(UTIL_clockSpanMicro(tStart) / 1000000ULL); } static U32 FUZ_rotl32(U32 x, U32 r) { return ((x << r) | (x >> (32 - r))); } -U32 FUZ_rand(U32* src) +static U32 FUZ_rand(U32* src) { const U32 prime1 = 2654435761U; const U32 prime2 = 2246822519U; @@ -128,221 +386,489 @@ U32 FUZ_rand(U32* src) return rand32 >> 5; } +/* allows zeros */ +#define CLAMPCHECK(val,min,max) { \ + if (((val)<(min)) | ((val)>(max))) { \ + DISPLAY("INVALID PARAMETER CONSTRAINTS\n"); \ + return 0; \ +} } + +static int paramValid(const paramValues_t paramTarget) { + U32 i; + for(i = 0; i < NUM_PARAMS; i++) { + CLAMPCHECK(paramTarget.vals[i], mintable[i], maxtable[i]); + } + return 1; +} -/*-******************************************************* -* Bench functions -*********************************************************/ -typedef struct { - size_t cSize; - double cSpeed; /* bytes / sec */ - double dSpeed; -} BMK_result_t; +static paramValues_t cParamUnsetMin(paramValues_t paramTarget) { + varInds_t i; + for(i = 0; i < NUM_PARAMS; i++) { + if(paramTarget.vals[i] == PARAM_UNSET) { + paramTarget.vals[i] = mintable[i]; + } + } + return paramTarget; +} -typedef struct -{ - const char* srcPtr; - size_t srcSize; - char* cPtr; - size_t cRoom; - size_t cSize; - char* resPtr; - size_t resSize; -} blockParam_t; - - -static size_t BMK_benchParam(BMK_result_t* resultPtr, - const void* srcBuffer, size_t srcSize, - ZSTD_CCtx* ctx, - const ZSTD_compressionParameters cParams) +static paramValues_t emptyParams(void) { + U32 i; + paramValues_t p; + for(i = 0; i < NUM_PARAMS; i++) { + p.vals[i] = PARAM_UNSET; + } + return p; +} + +static winnerInfo_t initWinnerInfo(const paramValues_t p) { + winnerInfo_t w1; + w1.result.cSpeed = 0.; + w1.result.dSpeed = 0.; + w1.result.cMem = (size_t)-1; + w1.result.cSize = (size_t)-1; + w1.params = p; + return w1; +} + +static paramValues_t overwriteParams(paramValues_t base, const paramValues_t mask) { + U32 i; + for(i = 0; i < NUM_PARAMS; i++) { + if(mask.vals[i] != PARAM_UNSET) { + base.vals[i] = mask.vals[i]; + } + } + return base; +} + +static void paramVaryOnce(const varInds_t paramIndex, const int amt, paramValues_t* ptr) { + ptr->vals[paramIndex] = rangeMap(paramIndex, invRangeMap(paramIndex, ptr->vals[paramIndex]) + amt); +} + +/* varies ptr by nbChanges respecting varyParams*/ +static void paramVariation(paramValues_t* ptr, memoTable_t* mtAll, const U32 nbChanges) { - const size_t blockSize = g_blockSize ? g_blockSize : srcSize; - const U32 nbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize); - blockParam_t* const blockTable = (blockParam_t*) malloc(nbBlocks * sizeof(blockParam_t)); - const size_t maxCompressedSize = (size_t)nbBlocks * ZSTD_compressBound(blockSize); - void* const compressedBuffer = malloc(maxCompressedSize); - void* const resultBuffer = malloc(srcSize); - ZSTD_parameters params; - U32 Wlog = cParams.windowLog; - U32 Clog = cParams.chainLog; - U32 Hlog = cParams.hashLog; - U32 Slog = cParams.searchLog; - U32 Slength = cParams.searchLength; - U32 Tlength = cParams.targetLength; - ZSTD_strategy strat = cParams.strategy; - char name[30] = { 0 }; - U64 crcOrig; - - /* init result for early exit */ - resultPtr->cSize = srcSize; - resultPtr->cSpeed = 0.; - resultPtr->dSpeed = 0.; - - /* Memory allocation & restrictions */ - snprintf(name, 30, "Sw%02uc%02uh%02us%02ul%1ut%03uS%1u", Wlog, Clog, Hlog, Slog, Slength, Tlength, strat); - if (!compressedBuffer || !resultBuffer || !blockTable) { - DISPLAY("\nError: not enough memory!\n"); - free(compressedBuffer); - free(resultBuffer); - free(blockTable); - return 12; - } - - /* Calculating input Checksum */ - crcOrig = XXH64(srcBuffer, srcSize, 0); - - /* Init blockTable data */ - { + paramValues_t p; + U32 validated = 0; + while (!validated) { U32 i; - size_t remaining = srcSize; - const char* srcPtr = (const char*)srcBuffer; - char* cPtr = (char*)compressedBuffer; - char* resPtr = (char*)resultBuffer; - for (i=0; i<nbBlocks; i++) { - size_t thisBlockSize = MIN(remaining, blockSize); - blockTable[i].srcPtr = srcPtr; - blockTable[i].cPtr = cPtr; - blockTable[i].resPtr = resPtr; - blockTable[i].srcSize = thisBlockSize; - blockTable[i].cRoom = ZSTD_compressBound(thisBlockSize); - srcPtr += thisBlockSize; - cPtr += blockTable[i].cRoom; - resPtr += thisBlockSize; - remaining -= thisBlockSize; - } } + p = *ptr; + for (i = 0 ; i < nbChanges ; i++) { + const U32 changeID = (U32)FUZ_rand(&g_rand) % (mtAll[p.vals[strt_ind]].varLen << 1); + paramVaryOnce(mtAll[p.vals[strt_ind]].varArray[changeID >> 1], ((changeID & 1) << 1) - 1, &p); + } + validated = paramValid(p); + } + *ptr = p; +} - /* warmimg up memory */ - RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.10, 1); +/* Completely random parameter selection */ +static paramValues_t randomParams(void) +{ + varInds_t v; paramValues_t p; + for(v = 0; v < NUM_PARAMS; v++) { + p.vals[v] = rangeMap(v, FUZ_rand(&g_rand) % rangetable[v]); + } + return p; +} - /* Bench */ - { U32 loopNb; - size_t cSize = 0; - double fastestC = 100000000., fastestD = 100000000.; - double ratio = 0.; - UTIL_time_t const benchStart = UTIL_getTime(); - - DISPLAY("\r%79s\r", ""); - memset(¶ms, 0, sizeof(params)); - params.cParams = cParams; - for (loopNb = 1; loopNb <= g_nbIterations; loopNb++) { - int nbLoops; - U32 blockNb; - UTIL_time_t roundStart; - U64 roundClock; - - { U64 const benchTime = UTIL_clockSpanMicro(benchStart); - if (benchTime > g_maxParamTime) break; } - - /* Compression */ - DISPLAY("\r%1u-%s : %9u ->", loopNb, name, (U32)srcSize); - memset(compressedBuffer, 0xE5, maxCompressedSize); - - nbLoops = 0; - UTIL_waitForNextTick(); - roundStart = UTIL_getTime(); - while (UTIL_clockSpanMicro(roundStart) < TIMELOOP) { - for (blockNb=0; blockNb<nbBlocks; blockNb++) - blockTable[blockNb].cSize = ZSTD_compress_advanced(ctx, - blockTable[blockNb].cPtr, blockTable[blockNb].cRoom, - blockTable[blockNb].srcPtr, blockTable[blockNb].srcSize, - NULL, 0, - params); - nbLoops++; +static U64 g_clockGranularity = 100000000ULL; + +static void findClockGranularity(void) { + UTIL_time_t clockStart = UTIL_getTime(); + U64 el1 = 0, el2 = 0; + int i = 0; + do { + el1 = el2; + el2 = UTIL_clockSpanNano(clockStart); + if(el1 < el2) { + U64 iv = el2 - el1; + if(g_clockGranularity > iv) { + g_clockGranularity = iv; + i = 0; + } else { + i++; } - roundClock = UTIL_clockSpanMicro(roundStart); - - cSize = 0; - for (blockNb=0; blockNb<nbBlocks; blockNb++) - cSize += blockTable[blockNb].cSize; - ratio = (double)srcSize / (double)cSize; - if ((double)roundClock < fastestC * SEC_TO_MICRO * nbLoops) fastestC = ((double)roundClock / SEC_TO_MICRO) / nbLoops; - DISPLAY("\r"); - DISPLAY("%1u-%s : %9u ->", loopNb, name, (U32)srcSize); - DISPLAY(" %9u (%4.3f),%7.1f MB/s", (U32)cSize, ratio, (double)srcSize / fastestC / 1000000.); - resultPtr->cSize = cSize; - resultPtr->cSpeed = (double)srcSize / fastestC; - -#if 1 - /* Decompression */ - memset(resultBuffer, 0xD6, srcSize); - - nbLoops = 0; - UTIL_waitForNextTick(); - roundStart = UTIL_getTime(); - for ( ; UTIL_clockSpanMicro(roundStart) < TIMELOOP; nbLoops++) { - for (blockNb=0; blockNb<nbBlocks; blockNb++) - blockTable[blockNb].resSize = ZSTD_decompress(blockTable[blockNb].resPtr, blockTable[blockNb].srcSize, - blockTable[blockNb].cPtr, blockTable[blockNb].cSize); + } + } while(i < 10); + DEBUGOUTPUT("Granularity: %llu\n", (unsigned long long)g_clockGranularity); +} + +/*-************************************ +* Optimizer Util Functions +**************************************/ + +/* checks results are feasible */ +static int feasible(const BMK_benchResult_t results, const constraint_t target) { + return (results.cSpeed >= target.cSpeed) + && (results.dSpeed >= target.dSpeed) + && (results.cMem <= target.cMem) + && (!g_optmode || results.cSize <= g_lvltarget.cSize); +} + +/* hill climbing value for part 1 */ +/* Scoring here is a linear reward for all set constraints normalized between 0 to 1 + * (with 0 at 0 and 1 being fully fulfilling the constraint), summed with a logarithmic + * bonus to exceeding the constraint value. We also give linear ratio for compression ratio. + * The constant factors are experimental. + */ +static double resultScore(const BMK_benchResult_t res, const size_t srcSize, const constraint_t target) { + double cs = 0., ds = 0., rt, cm = 0.; + const double r1 = 1, r2 = 0.1, rtr = 0.5; + double ret; + if(target.cSpeed) { cs = res.cSpeed / (double)target.cSpeed; } + if(target.dSpeed) { ds = res.dSpeed / (double)target.dSpeed; } + if(target.cMem != (U32)-1) { cm = (double)target.cMem / res.cMem; } + rt = ((double)srcSize / res.cSize); + + ret = (MIN(1, cs) + MIN(1, ds) + MIN(1, cm))*r1 + rt * rtr + + (MAX(0, log(cs))+ MAX(0, log(ds))+ MAX(0, log(cm))) * r2; + + return ret; +} + +/* calculates normalized squared euclidean distance of result1 if it is in the first quadrant relative to lvlRes */ +static double resultDistLvl(const BMK_benchResult_t result1, const BMK_benchResult_t lvlRes) { + double normalizedCSpeedGain1 = (result1.cSpeed / lvlRes.cSpeed) - 1; + double normalizedRatioGain1 = ((double)lvlRes.cSize / result1.cSize) - 1; + if(normalizedRatioGain1 < 0 || normalizedCSpeedGain1 < 0) { + return 0.0; + } + return normalizedRatioGain1 * g_ratioMultiplier + normalizedCSpeedGain1; +} + +/* return true if r2 strictly better than r1 */ +static int compareResultLT(const BMK_benchResult_t result1, const BMK_benchResult_t result2, const constraint_t target, size_t srcSize) { + if(feasible(result1, target) && feasible(result2, target)) { + if(g_optmode) { + return resultDistLvl(result1, g_lvltarget) < resultDistLvl(result2, g_lvltarget); + } else { + return (result1.cSize > result2.cSize) || (result1.cSize == result2.cSize && result2.cSpeed > result1.cSpeed) + || (result1.cSize == result2.cSize && result2.cSpeed == result1.cSpeed && result2.dSpeed > result1.dSpeed); + } + } + return feasible(result2, target) || (!feasible(result1, target) && (resultScore(result1, srcSize, target) < resultScore(result2, srcSize, target))); +} + +static constraint_t relaxTarget(constraint_t target) { + target.cMem = (U32)-1; + target.cSpeed *= ((double)g_strictness) / 100; + target.dSpeed *= ((double)g_strictness) / 100; + return target; +} + +static void optimizerAdjustInput(paramValues_t* pc, const size_t maxBlockSize) { + varInds_t v; + for(v = 0; v < NUM_PARAMS; v++) { + if(pc->vals[v] != PARAM_UNSET) { + U32 newval = MIN(MAX(pc->vals[v], mintable[v]), maxtable[v]); + if(newval != pc->vals[v]) { + pc->vals[v] = newval; + DISPLAY("Warning: parameter %s not in valid range, adjusting to ", g_paramNames[v]); displayParamVal(stderr, v, newval, 0); DISPLAY("\n"); } - roundClock = UTIL_clockSpanMicro(roundStart); - - if ((double)roundClock < fastestD * SEC_TO_MICRO * nbLoops) fastestD = ((double)roundClock / SEC_TO_MICRO) / nbLoops; - DISPLAY("\r"); - DISPLAY("%1u-%s : %9u -> ", loopNb, name, (U32)srcSize); - DISPLAY("%9u (%4.3f),%7.1f MB/s, ", (U32)cSize, ratio, (double)srcSize / fastestC / 1000000.); - DISPLAY("%7.1f MB/s", (double)srcSize / fastestD / 1000000.); - resultPtr->dSpeed = (double)srcSize / fastestD; - - /* CRC Checking */ - { U64 const crcCheck = XXH64(resultBuffer, srcSize, 0); - if (crcOrig!=crcCheck) { - unsigned u; - unsigned eBlockSize = (unsigned)(MIN(65536*2, blockSize)); - DISPLAY("\n!!! WARNING !!! Invalid Checksum : %x != %x\n", (unsigned)crcOrig, (unsigned)crcCheck); - for (u=0; u<srcSize; u++) { - if (((const BYTE*)srcBuffer)[u] != ((BYTE*)resultBuffer)[u]) { - printf("Decoding error at pos %u (block %u, pos %u) \n", u, u / eBlockSize, u % eBlockSize); - break; - } } - break; - } } -#endif - } } + } + } - /* End cleaning */ - DISPLAY("\r"); - free(compressedBuffer); - free(resultBuffer); - return 0; + if(pc->vals[wlog_ind] != PARAM_UNSET) { + + U32 sshb = maxBlockSize > 1 ? ZSTD_highbit32((U32)(maxBlockSize-1)) + 1 : 1; + /* edge case of highBit not working for 0 */ + + if(maxBlockSize < (1ULL << 31) && sshb + 1 < pc->vals[wlog_ind]) { + U32 adjust = MAX(mintable[wlog_ind], sshb); + if(adjust != pc->vals[wlog_ind]) { + pc->vals[wlog_ind] = adjust; + DISPLAY("Warning: windowLog larger than src/block size, adjusted to %u\n", pc->vals[wlog_ind]); + } + } + } + + if(pc->vals[wlog_ind] != PARAM_UNSET && pc->vals[clog_ind] != PARAM_UNSET) { + U32 maxclog; + if(pc->vals[strt_ind] == PARAM_UNSET || pc->vals[strt_ind] >= (U32)ZSTD_btlazy2) { + maxclog = pc->vals[wlog_ind] + 1; + } else { + maxclog = pc->vals[wlog_ind]; + } + + if(pc->vals[clog_ind] > maxclog) { + pc->vals[clog_ind] = maxclog; + DISPLAY("Warning: chainlog too much larger than windowLog size, adjusted to %u\n", pc->vals[clog_ind]); + } + } + + if(pc->vals[wlog_ind] != PARAM_UNSET && pc->vals[hlog_ind] != PARAM_UNSET) { + if(pc->vals[wlog_ind] + 1 < pc->vals[hlog_ind]) { + pc->vals[hlog_ind] = pc->vals[wlog_ind] + 1; + DISPLAY("Warning: hashlog too much larger than windowLog size, adjusted to %u\n", pc->vals[hlog_ind]); + } + } + + if(pc->vals[slog_ind] != PARAM_UNSET && pc->vals[clog_ind] != PARAM_UNSET) { + if(pc->vals[slog_ind] > pc->vals[clog_ind]) { + pc->vals[clog_ind] = pc->vals[slog_ind]; + DISPLAY("Warning: searchLog larger than chainLog, adjusted to %u\n", pc->vals[slog_ind]); + } + } } +static int redundantParams(const paramValues_t paramValues, const constraint_t target, const size_t maxBlockSize) { + return + (ZSTD_estimateCStreamSize_usingCParams(pvalsToCParams(paramValues)) > (size_t)target.cMem) /* Uses too much memory */ + || ((1ULL << (paramValues.vals[wlog_ind] - 1)) >= maxBlockSize && paramValues.vals[wlog_ind] != mintable[wlog_ind]) /* wlog too much bigger than src size */ + || (paramValues.vals[clog_ind] > (paramValues.vals[wlog_ind] + (paramValues.vals[strt_ind] > ZSTD_btlazy2))) /* chainLog larger than windowLog*/ + || (paramValues.vals[slog_ind] > paramValues.vals[clog_ind]) /* searchLog larger than chainLog */ + || (paramValues.vals[hlog_ind] > paramValues.vals[wlog_ind] + 1); /* hashLog larger than windowLog + 1 */ -const char* g_stratName[ZSTD_btultra+1] = { - "(none) ", "ZSTD_fast ", "ZSTD_dfast ", - "ZSTD_greedy ", "ZSTD_lazy ", "ZSTD_lazy2 ", - "ZSTD_btlazy2 ", "ZSTD_btopt ", "ZSTD_btultra "}; +} + +/*-************************************ +* Display Functions +**************************************/ -static void BMK_printWinner(FILE* f, U32 cLevel, BMK_result_t result, ZSTD_compressionParameters params, size_t srcSize) +static void BMK_translateAdvancedParams(FILE* f, const paramValues_t params) { + varInds_t v; + int first = 1; + fprintf(f,"--zstd="); + for (v = 0; v < NUM_PARAMS; v++) { + if (g_silenceParams[v]) { continue; } + if (!first) { fprintf(f, ","); } + fprintf(f,"%s=", g_paramNames[v]); + + if (v == strt_ind) { fprintf(f,"%u", params.vals[v]); } + else { displayParamVal(f, v, params.vals[v], 0); } + first = 0; + } + fprintf(f, "\n"); +} + +static void BMK_displayOneResult(FILE* f, winnerInfo_t res, const size_t srcSize) { - DISPLAY("\r%79s\r", ""); - fprintf(f," {%3u,%3u,%3u,%3u,%3u,%3u, %s }, ", - params.windowLog, params.chainLog, params.hashLog, params.searchLog, params.searchLength, - params.targetLength, g_stratName[(U32)(params.strategy)]); - fprintf(f, - "/* level %2u */ /* R:%5.3f at %5.1f MB/s - %5.1f MB/s */\n", - cLevel, (double)srcSize / result.cSize, result.cSpeed / 1000000., result.dSpeed / 1000000.); + varInds_t v; + int first = 1; + res.params = cParamUnsetMin(res.params); + fprintf(f, " {"); + for (v = 0; v < NUM_PARAMS; v++) { + if (g_silenceParams[v]) { continue; } + if (!first) { fprintf(f, ","); } + displayParamVal(f, v, res.params.vals[v], 3); + first = 0; + } + + { double const ratio = res.result.cSize ? + (double)srcSize / res.result.cSize : 0; + double const cSpeedMBps = (double)res.result.cSpeed / MB_UNIT; + double const dSpeedMBps = (double)res.result.dSpeed / MB_UNIT; + + fprintf(f, " }, /* R:%5.3f at %5.1f MB/s - %5.1f MB/s */\n", + ratio, cSpeedMBps, dSpeedMBps); + } } +/* Writes to f the results of a parameter benchmark */ +/* when used with --optimize, will only print results better than previously discovered */ +static void BMK_printWinner(FILE* f, const int cLevel, const BMK_benchResult_t result, const paramValues_t params, const size_t srcSize) +{ + char lvlstr[15] = "Custom Level"; + winnerInfo_t w; + w.params = params; + w.result = result; -static double g_cSpeedTarget[NB_LEVELS_TRACKED] = { 0. }; /* NB_LEVELS_TRACKED : checked at main() */ + fprintf(f, "\r%79s\r", ""); -typedef struct { - BMK_result_t result; - ZSTD_compressionParameters params; -} winnerInfo_t; + if(cLevel != CUSTOM_LEVEL) { + snprintf(lvlstr, 15, " Level %2d ", cLevel); + } + + if(TIMED) { + const U64 time = UTIL_clockSpanNano(g_time); + const U64 minutes = time / (60ULL * TIMELOOP_NANOSEC); + fprintf(f, "%1lu:%2lu:%05.2f - ", (unsigned long) minutes / 60,(unsigned long) minutes % 60, (double)(time - minutes * TIMELOOP_NANOSEC * 60ULL)/TIMELOOP_NANOSEC); + } + + fprintf(f, "/* %s */ ", lvlstr); + BMK_displayOneResult(f, w, srcSize); +} -static void BMK_printWinners2(FILE* f, const winnerInfo_t* winners, size_t srcSize) +/* comparison function: */ +/* strictly better, strictly worse, equal, speed-side adv, size-side adv */ +#define WORSE_RESULT 0 +#define BETTER_RESULT 1 +#define ERROR_RESULT 2 + +#define SPEED_RESULT 4 +#define SIZE_RESULT 5 +/* maybe have epsilon-eq to limit table size? */ +static int speedSizeCompare(const BMK_benchResult_t r1, const BMK_benchResult_t r2) { + if(r1.cSpeed < r2.cSpeed) { + if(r1.cSize >= r2.cSize) { + return BETTER_RESULT; + } + return SPEED_RESULT; /* r2 is smaller but not faster. */ + } else { + if(r1.cSize <= r2.cSize) { + return WORSE_RESULT; + } + return SIZE_RESULT; /* r2 is faster but not smaller */ + } +} + +/* 0 for insertion, 1 for no insert */ +/* maintain invariant speedSizeCompare(n, n->next) = SPEED_RESULT */ +static int insertWinner(const winnerInfo_t w, const constraint_t targetConstraints) { + BMK_benchResult_t r = w.result; + winner_ll_node* cur_node = g_winners; + /* first node to insert */ + if(!feasible(r, targetConstraints)) { + return 1; + } + + if(g_winners == NULL) { + winner_ll_node* first_node = malloc(sizeof(winner_ll_node)); + if(first_node == NULL) { + return 1; + } + first_node->next = NULL; + first_node->res = w; + g_winners = first_node; + return 0; + } + + while(cur_node->next != NULL) { + switch(speedSizeCompare(cur_node->res.result, r)) { + case WORSE_RESULT: + { + return 1; /* never insert if better */ + } + case BETTER_RESULT: + { + winner_ll_node* tmp; + cur_node->res = cur_node->next->res; + tmp = cur_node->next; + cur_node->next = cur_node->next->next; + free(tmp); + break; + } + case SIZE_RESULT: + { + cur_node = cur_node->next; + break; + } + case SPEED_RESULT: /* insert after first size result, then return */ + { + winner_ll_node* newnode = malloc(sizeof(winner_ll_node)); + if(newnode == NULL) { + return 1; + } + newnode->res = cur_node->res; + cur_node->res = w; + newnode->next = cur_node->next; + cur_node->next = newnode; + return 0; + } + } + + } + + assert(cur_node->next == NULL); + switch(speedSizeCompare(cur_node->res.result, r)) { + case WORSE_RESULT: + { + return 1; /* never insert if better */ + } + case BETTER_RESULT: + { + cur_node->res = w; + return 0; + } + case SIZE_RESULT: + { + winner_ll_node* newnode = malloc(sizeof(winner_ll_node)); + if(newnode == NULL) { + return 1; + } + newnode->res = w; + newnode->next = NULL; + cur_node->next = newnode; + return 0; + } + case SPEED_RESULT: /* insert before first size result, then return */ + { + winner_ll_node* newnode = malloc(sizeof(winner_ll_node)); + if(newnode == NULL) { + return 1; + } + newnode->res = cur_node->res; + cur_node->res = w; + newnode->next = cur_node->next; + cur_node->next = newnode; + return 0; + } + default: + return 1; + } +} + +static void BMK_printWinnerOpt(FILE* f, const U32 cLevel, const BMK_benchResult_t result, const paramValues_t params, const constraint_t targetConstraints, const size_t srcSize) +{ + /* global winner used for constraints */ + /* cSize, cSpeed, dSpeed, cMem */ + static winnerInfo_t g_winner = { { (size_t)-1LL, 0, 0, (size_t)-1LL }, { { PARAM_UNSET, PARAM_UNSET, PARAM_UNSET, PARAM_UNSET, PARAM_UNSET, PARAM_UNSET, PARAM_UNSET, PARAM_UNSET } } }; + if(DEBUG || compareResultLT(g_winner.result, result, targetConstraints, srcSize) || g_displayLevel >= 4) { + if(DEBUG && compareResultLT(g_winner.result, result, targetConstraints, srcSize)) { + DISPLAY("New Winner: \n"); + } + + if(g_displayLevel >= 2) { BMK_printWinner(f, cLevel, result, params, srcSize); } + + if(compareResultLT(g_winner.result, result, targetConstraints, srcSize)) { + if(g_displayLevel >= 1) { BMK_translateAdvancedParams(f, params); } + g_winner.result = result; + g_winner.params = params; + } + } + + if(g_optmode && g_optimizer && (DEBUG || g_displayLevel == 3)) { + winnerInfo_t w; + winner_ll_node* n; + w.result = result; + w.params = params; + insertWinner(w, targetConstraints); + + if(!DEBUG) { fprintf(f, "\033c"); } + fprintf(f, "\n"); + + /* the table */ + fprintf(f, "================================\n"); + for(n = g_winners; n != NULL; n = n->next) { + BMK_displayOneResult(f, n->res, srcSize); + } + fprintf(f, "================================\n"); + fprintf(f, "Level Bounds: R: > %.3f AND C: < %.1f MB/s \n\n", + (double)srcSize / g_lvltarget.cSize, (double)g_lvltarget.cSpeed / MB_UNIT); + + + fprintf(f, "Overall Winner: \n"); + BMK_displayOneResult(f, g_winner, srcSize); + BMK_translateAdvancedParams(f, g_winner.params); + + fprintf(f, "Latest BMK: \n");\ + BMK_displayOneResult(f, w, srcSize); + } +} + +static void BMK_printWinners2(FILE* f, const winnerInfo_t* winners, const size_t srcSize) { int cLevel; fprintf(f, "\n /* Proposed configurations : */ \n"); fprintf(f, " /* W, C, H, S, L, T, strat */ \n"); - for (cLevel=0; cLevel <= ZSTD_maxCLevel(); cLevel++) + for (cLevel=0; cLevel <= NB_LEVELS_TRACKED; cLevel++) BMK_printWinner(f, cLevel, winners[cLevel].result, winners[cLevel].params, srcSize); } -static void BMK_printWinners(FILE* f, const winnerInfo_t* winners, size_t srcSize) +static void BMK_printWinners(FILE* f, const winnerInfo_t* winners, const size_t srcSize) { fseek(f, 0, SEEK_SET); BMK_printWinners2(f, winners, srcSize); @@ -350,41 +876,837 @@ static void BMK_printWinners(FILE* f, const winnerInfo_t* winners, size_t srcSiz BMK_printWinners2(stdout, winners, srcSize); } -static int BMK_seed(winnerInfo_t* winners, const ZSTD_compressionParameters params, - const void* srcBuffer, size_t srcSize, - ZSTD_CCtx* ctx) + +/*-******************************************************* +* Functions to Benchmark +*********************************************************/ + +typedef struct { + ZSTD_CCtx* cctx; + const void* dictBuffer; + size_t dictBufferSize; + int cLevel; + const paramValues_t* comprParams; +} BMK_initCCtxArgs; + +static size_t local_initCCtx(void* payload) { + const BMK_initCCtxArgs* ag = (const BMK_initCCtxArgs*)payload; + varInds_t i; + ZSTD_CCtx_reset(ag->cctx); + ZSTD_CCtx_resetParameters(ag->cctx); + ZSTD_CCtx_setParameter(ag->cctx, ZSTD_p_compressionLevel, ag->cLevel); + + for(i = 0; i < NUM_PARAMS; i++) { + if(ag->comprParams->vals[i] != PARAM_UNSET) + ZSTD_CCtx_setParameter(ag->cctx, cctxSetParamTable[i], ag->comprParams->vals[i]); + } + ZSTD_CCtx_loadDictionary(ag->cctx, ag->dictBuffer, ag->dictBufferSize); + + return 0; +} + +typedef struct { + ZSTD_DCtx* dctx; + const void* dictBuffer; + size_t dictBufferSize; +} BMK_initDCtxArgs; + +static size_t local_initDCtx(void* payload) { + const BMK_initDCtxArgs* ag = (const BMK_initDCtxArgs*)payload; + ZSTD_DCtx_reset(ag->dctx); + ZSTD_DCtx_loadDictionary(ag->dctx, ag->dictBuffer, ag->dictBufferSize); + return 0; +} + +/* additional argument is just the context */ +static size_t local_defaultCompress( + const void* srcBuffer, size_t srcSize, + void* dstBuffer, size_t dstSize, + void* addArgs) { + size_t moreToFlush = 1; + ZSTD_CCtx* ctx = (ZSTD_CCtx*)addArgs; + ZSTD_inBuffer in; + ZSTD_outBuffer out; + in.src = srcBuffer; + in.size = srcSize; + in.pos = 0; + out.dst = dstBuffer; + out.size = dstSize; + out.pos = 0; + assert(dstSize == ZSTD_compressBound(srcSize)); /* specific to this version, which is only used in paramgrill */ + while (moreToFlush) { + if(out.pos == out.size) { + return (size_t)-ZSTD_error_dstSize_tooSmall; + } + moreToFlush = ZSTD_compress_generic(ctx, &out, &in, ZSTD_e_end); + if (ZSTD_isError(moreToFlush)) { + return moreToFlush; + } + } + return out.pos; +} + +/* additional argument is just the context */ +static size_t local_defaultDecompress( + const void* srcBuffer, size_t srcSize, + void* dstBuffer, size_t dstSize, + void* addArgs) { + size_t moreToFlush = 1; + ZSTD_DCtx* dctx = (ZSTD_DCtx*)addArgs; + ZSTD_inBuffer in; + ZSTD_outBuffer out; + in.src = srcBuffer; + in.size = srcSize; + in.pos = 0; + out.dst = dstBuffer; + out.size = dstSize; + out.pos = 0; + while (moreToFlush) { + if(out.pos == out.size) { + return (size_t)-ZSTD_error_dstSize_tooSmall; + } + moreToFlush = ZSTD_decompress_generic(dctx, + &out, &in); + if (ZSTD_isError(moreToFlush)) { + return moreToFlush; + } + } + return out.pos; + +} + +/*-************************************ +* Data Initialization Functions +**************************************/ + +typedef struct { + void* srcBuffer; + size_t srcSize; + const void** srcPtrs; + size_t* srcSizes; + void** dstPtrs; + size_t* dstCapacities; + size_t* dstSizes; + void** resPtrs; + size_t* resSizes; + size_t nbBlocks; + size_t maxBlockSize; +} buffers_t; + +typedef struct { + size_t dictSize; + void* dictBuffer; + ZSTD_CCtx* cctx; + ZSTD_DCtx* dctx; +} contexts_t; + +static void freeNonSrcBuffers(const buffers_t b) { + free(b.srcPtrs); + free(b.srcSizes); + + if(b.dstPtrs != NULL) { + free(b.dstPtrs[0]); + } + free(b.dstPtrs); + free(b.dstCapacities); + free(b.dstSizes); + + if(b.resPtrs != NULL) { + free(b.resPtrs[0]); + } + free(b.resPtrs); + free(b.resSizes); +} + +static void freeBuffers(const buffers_t b) { + if(b.srcPtrs != NULL) { + free(b.srcBuffer); + } + freeNonSrcBuffers(b); +} + +/* srcBuffer will be freed by freeBuffers now */ +static int createBuffersFromMemory(buffers_t* buff, void * srcBuffer, const size_t nbFiles, + const size_t* fileSizes) { - BMK_result_t testResult; + size_t pos = 0, n, blockSize; + U32 maxNbBlocks, blockNb = 0; + buff->srcSize = 0; + for(n = 0; n < nbFiles; n++) { + buff->srcSize += fileSizes[n]; + } + + if(buff->srcSize == 0) { + DISPLAY("No data to bench\n"); + return 1; + } + + blockSize = g_blockSize ? g_blockSize : buff->srcSize; + maxNbBlocks = (U32) ((buff->srcSize + (blockSize-1)) / blockSize) + (U32)nbFiles; + + buff->srcPtrs = (const void**)calloc(maxNbBlocks, sizeof(void*)); + buff->srcSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t)); + + buff->dstPtrs = (void**)calloc(maxNbBlocks, sizeof(void*)); + buff->dstCapacities = (size_t*)malloc(maxNbBlocks * sizeof(size_t)); + buff->dstSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t)); + + buff->resPtrs = (void**)calloc(maxNbBlocks, sizeof(void*)); + buff->resSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t)); + + if(!buff->srcPtrs || !buff->srcSizes || !buff->dstPtrs || !buff->dstCapacities || !buff->dstSizes || !buff->resPtrs || !buff->resSizes) { + DISPLAY("alloc error\n"); + freeNonSrcBuffers(*buff); + return 1; + } + + buff->srcBuffer = srcBuffer; + buff->srcPtrs[0] = (const void*)buff->srcBuffer; + buff->dstPtrs[0] = malloc(ZSTD_compressBound(buff->srcSize) + (maxNbBlocks * 1024)); + buff->resPtrs[0] = malloc(buff->srcSize); + + if(!buff->dstPtrs[0] || !buff->resPtrs[0]) { + DISPLAY("alloc error\n"); + freeNonSrcBuffers(*buff); + return 1; + } + + for(n = 0; n < nbFiles; n++) { + size_t pos_end = pos + fileSizes[n]; + for(; pos < pos_end; blockNb++) { + buff->srcPtrs[blockNb] = (const void*)((char*)srcBuffer + pos); + buff->srcSizes[blockNb] = blockSize; + pos += blockSize; + } + + if(fileSizes[n] > 0) { buff->srcSizes[blockNb - 1] = ((fileSizes[n] - 1) % blockSize) + 1; } + pos = pos_end; + } + + buff->dstCapacities[0] = ZSTD_compressBound(buff->srcSizes[0]); + buff->dstSizes[0] = buff->dstCapacities[0]; + buff->resSizes[0] = buff->srcSizes[0]; + buff->maxBlockSize = buff->srcSizes[0]; + + for(n = 1; n < blockNb; n++) { + buff->dstPtrs[n] = ((char*)buff->dstPtrs[n-1]) + buff->dstCapacities[n-1]; + buff->resPtrs[n] = ((char*)buff->resPtrs[n-1]) + buff->resSizes[n-1]; + buff->dstCapacities[n] = ZSTD_compressBound(buff->srcSizes[n]); + buff->dstSizes[n] = buff->dstCapacities[n]; + buff->resSizes[n] = buff->srcSizes[n]; + + buff->maxBlockSize = MAX(buff->maxBlockSize, buff->srcSizes[n]); + } + + buff->nbBlocks = blockNb; + + return 0; +} + +/* allocates buffer's arguments. returns success / failuere */ +static int createBuffers(buffers_t* buff, const char* const * const fileNamesTable, + size_t nbFiles) { + size_t pos = 0; + size_t n; + size_t totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, (U32)nbFiles); + size_t benchedSize = MIN(BMK_findMaxMem(totalSizeToLoad * 3) / 3, totalSizeToLoad); + size_t* fileSizes = calloc(sizeof(size_t), nbFiles); + void* srcBuffer = NULL; + int ret = 0; + + if(!totalSizeToLoad || !benchedSize) { + ret = 1; + DISPLAY("Nothing to Bench\n"); + goto _cleanUp; + } + + srcBuffer = malloc(benchedSize); + + if(!fileSizes || !srcBuffer) { + ret = 1; + goto _cleanUp; + } + + for(n = 0; n < nbFiles; n++) { + FILE* f; + U64 fileSize = UTIL_getFileSize(fileNamesTable[n]); + if (UTIL_isDirectory(fileNamesTable[n])) { + DISPLAY("Ignoring %s directory... \n", fileNamesTable[n]); + continue; + } + if (fileSize == UTIL_FILESIZE_UNKNOWN) { + DISPLAY("Cannot evaluate size of %s, ignoring ... \n", fileNamesTable[n]); + continue; + } + f = fopen(fileNamesTable[n], "rb"); + if (f==NULL) { + DISPLAY("impossible to open file %s\n", fileNamesTable[n]); + fclose(f); + ret = 10; + goto _cleanUp; + } + + DISPLAYLEVEL(2, "Loading %s... \r", fileNamesTable[n]); + + if (fileSize + pos > benchedSize) fileSize = benchedSize - pos, nbFiles=n; /* buffer too small - stop after this file */ + { + char* buffer = (char*)(srcBuffer); + size_t const readSize = fread((buffer)+pos, 1, (size_t)fileSize, f); + fclose(f); + if (readSize != (size_t)fileSize) { + DISPLAY("could not read %s", fileNamesTable[n]); + ret = 1; + goto _cleanUp; + } + + fileSizes[n] = readSize; + pos += readSize; + } + } + + ret = createBuffersFromMemory(buff, srcBuffer, nbFiles, fileSizes); + +_cleanUp: + if(ret) { free(srcBuffer); } + free(fileSizes); + return ret; +} + +static void freeContexts(const contexts_t ctx) { + free(ctx.dictBuffer); + ZSTD_freeCCtx(ctx.cctx); + ZSTD_freeDCtx(ctx.dctx); +} + +static int createContexts(contexts_t* ctx, const char* dictFileName) { + FILE* f; + size_t readSize; + ctx->cctx = ZSTD_createCCtx(); + ctx->dctx = ZSTD_createDCtx(); + assert(ctx->cctx != NULL); + assert(ctx->dctx != NULL); + + if(dictFileName == NULL) { + ctx->dictSize = 0; + ctx->dictBuffer = NULL; + return 0; + } + { U64 const dictFileSize = UTIL_getFileSize(dictFileName); + assert(dictFileSize != UTIL_FILESIZE_UNKNOWN); + ctx->dictSize = dictFileSize; + assert((U64)ctx->dictSize == dictFileSize); /* check overflow */ + } + ctx->dictBuffer = malloc(ctx->dictSize); + + f = fopen(dictFileName, "rb"); + + if (f==NULL) { + DISPLAY("unable to open file\n"); + freeContexts(*ctx); + return 1; + } + + if (ctx->dictSize > 64 MB || !(ctx->dictBuffer)) { + DISPLAY("dictionary too large\n"); + fclose(f); + freeContexts(*ctx); + return 1; + } + readSize = fread(ctx->dictBuffer, 1, ctx->dictSize, f); + fclose(f); + if (readSize != ctx->dictSize) { + DISPLAY("unable to read file\n"); + freeContexts(*ctx); + return 1; + } + return 0; +} + +/*-************************************ +* Optimizer Memoization Functions +**************************************/ + +/* return: new length */ +/* keep old array, will need if iter over strategy. */ +/* prunes useless params */ +static size_t sanitizeVarArray(varInds_t* varNew, const size_t varLength, const varInds_t* varArray, const ZSTD_strategy strat) { + size_t i, j = 0; + for(i = 0; i < varLength; i++) { + if( !((varArray[i] == clog_ind && strat == ZSTD_fast) + || (varArray[i] == slog_ind && strat == ZSTD_fast) + || (varArray[i] == slog_ind && strat == ZSTD_dfast) + || (varArray[i] == tlen_ind && strat != ZSTD_btopt && strat != ZSTD_btultra && strat != ZSTD_fast))) { + varNew[j] = varArray[i]; + j++; + } + } + return j; +} + +/* res should be NUM_PARAMS size */ +/* constructs varArray from paramValues_t style parameter */ +/* pass in using dict. */ +static size_t variableParams(const paramValues_t paramConstraints, varInds_t* res, const int usingDictionary) { + varInds_t i; + size_t j = 0; + for(i = 0; i < NUM_PARAMS; i++) { + if(paramConstraints.vals[i] == PARAM_UNSET) { + if(i == fadt_ind && !usingDictionary) continue; /* don't use fadt if no dictionary */ + res[j] = i; j++; + } + } + return j; +} + +/* length of memo table given free variables */ +static size_t memoTableLen(const varInds_t* varyParams, const size_t varyLen) { + size_t arrayLen = 1; + size_t i; + for(i = 0; i < varyLen; i++) { + if(varyParams[i] == strt_ind) continue; /* strategy separated by table */ + arrayLen *= rangetable[varyParams[i]]; + } + return arrayLen; +} + +/* returns unique index in memotable of compression parameters */ +static unsigned memoTableIndDirect(const paramValues_t* ptr, const varInds_t* varyParams, const size_t varyLen) { + size_t i; + unsigned ind = 0; + for(i = 0; i < varyLen; i++) { + varInds_t v = varyParams[i]; + if(v == strt_ind) continue; /* exclude strategy from memotable */ + ind *= rangetable[v]; ind += (unsigned)invRangeMap(v, ptr->vals[v]); + } + return ind; +} + +static size_t memoTableGet(const memoTable_t* memoTableArray, const paramValues_t p) { + const memoTable_t mt = memoTableArray[p.vals[strt_ind]]; + switch(mt.tableType) { + case directMap: + return mt.table[memoTableIndDirect(&p, mt.varArray, mt.varLen)]; + case xxhashMap: + return mt.table[(XXH64(&p.vals, sizeof(U32) * NUM_PARAMS, 0) >> 3) % mt.tableLen]; + case noMemo: + return 0; + } + return 0; /* should never happen, stop compiler warnings */ +} + +static void memoTableSet(const memoTable_t* memoTableArray, const paramValues_t p, const BYTE value) { + const memoTable_t mt = memoTableArray[p.vals[strt_ind]]; + switch(mt.tableType) { + case directMap: + mt.table[memoTableIndDirect(&p, mt.varArray, mt.varLen)] = value; break; + case xxhashMap: + mt.table[(XXH64(&p.vals, sizeof(U32) * NUM_PARAMS, 0) >> 3) % mt.tableLen] = value; break; + case noMemo: + break; + } +} + +/* frees all allocated memotables */ +static void freeMemoTableArray(memoTable_t* const mtAll) { + int i; + if(mtAll == NULL) { return; } + for(i = 1; i <= (int)ZSTD_btultra; i++) { + free(mtAll[i].table); + } + free(mtAll); +} + +/* inits memotables for all (including mallocs), all strategies */ +/* takes unsanitized varyParams */ +static memoTable_t* createMemoTableArray(const paramValues_t p, const varInds_t* const varyParams, const size_t varyLen, const U32 memoTableLog) { + memoTable_t* mtAll = (memoTable_t*)calloc(sizeof(memoTable_t),(ZSTD_btultra + 1)); + ZSTD_strategy i, stratMin = ZSTD_fast, stratMax = ZSTD_btultra; + + if(mtAll == NULL) { + return NULL; + } + + for(i = 1; i <= (int)ZSTD_btultra; i++) { + mtAll[i].varLen = sanitizeVarArray(mtAll[i].varArray, varyLen, varyParams, i); + } + + /* no memoization */ + if(memoTableLog == 0) { + for(i = 1; i <= (int)ZSTD_btultra; i++) { + mtAll[i].tableType = noMemo; + mtAll[i].table = NULL; + mtAll[i].tableLen = 0; + } + return mtAll; + } + + + if(p.vals[strt_ind] != PARAM_UNSET) { + stratMin = p.vals[strt_ind]; + stratMax = p.vals[strt_ind]; + } + + + for(i = stratMin; i <= stratMax; i++) { + size_t mtl = memoTableLen(mtAll[i].varArray, mtAll[i].varLen); + mtAll[i].tableType = directMap; + + if(memoTableLog != PARAM_UNSET && mtl > (1ULL << memoTableLog)) { /* use hash table */ /* provide some option to only use hash tables? */ + mtAll[i].tableType = xxhashMap; + mtl = (1ULL << memoTableLog); + } + + mtAll[i].table = (BYTE*)calloc(sizeof(BYTE), mtl); + mtAll[i].tableLen = mtl; + + if(mtAll[i].table == NULL) { + freeMemoTableArray(mtAll); + return NULL; + } + } + + return mtAll; +} + +/* Sets pc to random unmeasured set of parameters */ +/* specifiy strategy */ +static void randomConstrainedParams(paramValues_t* pc, const memoTable_t* memoTableArray, const ZSTD_strategy st) +{ + size_t j; + const memoTable_t mt = memoTableArray[st]; + pc->vals[strt_ind] = st; + for(j = 0; j < mt.tableLen; j++) { + int i; + for(i = 0; i < NUM_PARAMS; i++) { + varInds_t v = mt.varArray[i]; + if(v == strt_ind) continue; + pc->vals[v] = rangeMap(v, FUZ_rand(&g_rand) % rangetable[v]); + } + + if(!(memoTableGet(memoTableArray, *pc))) break; /* only pick unpicked params. */ + } +} + +/*-************************************ +* Benchmarking Functions +**************************************/ + +/* Replicate functionality of benchMemAdvanced, but with pre-split src / dst buffers */ +/* The purpose is so that sufficient information is returned so that a decompression call to benchMemInvertible is possible */ +/* BMK_benchMemAdvanced(srcBuffer,srcSize, dstBuffer, dstSize, fileSizes, nbFiles, 0, &cParams, dictBuffer, dictSize, ctx, dctx, 0, "File", &adv); */ +/* nbSeconds used in same way as in BMK_advancedParams_t */ +/* if in decodeOnly, then srcPtr's will be compressed blocks, and uncompressedBlocks will be written to dstPtrs */ +/* dictionary nullable, nothing else though. */ +/* note : it would be better if this function was in bench.c, sharing code with benchMemAdvanced(), since it's technically a part of it */ +static BMK_benchOutcome_t +BMK_benchMemInvertible( buffers_t buf, contexts_t ctx, + int cLevel, const paramValues_t* comprParams, + BMK_mode_t mode, unsigned nbSeconds) +{ + U32 i; + BMK_benchResult_t bResult; + const void *const *const srcPtrs = (const void *const *const)buf.srcPtrs; + size_t const *const srcSizes = buf.srcSizes; + void** const dstPtrs = buf.dstPtrs; + size_t const *const dstCapacities = buf.dstCapacities; + size_t* const dstSizes = buf.dstSizes; + void** const resPtrs = buf.resPtrs; + size_t const *const resSizes = buf.resSizes; + const void* dictBuffer = ctx.dictBuffer; + const size_t dictBufferSize = ctx.dictSize; + const size_t nbBlocks = buf.nbBlocks; + const size_t srcSize = buf.srcSize; + ZSTD_CCtx* cctx = ctx.cctx; + ZSTD_DCtx* dctx = ctx.dctx; + + /* init */ + memset(&bResult, 0, sizeof(bResult)); + + /* warmimg up memory */ + for (i = 0; i < buf.nbBlocks; i++) { + if (mode != BMK_decodeOnly) { + RDG_genBuffer(dstPtrs[i], dstCapacities[i], 0.10, 0.50, 1); + } else { + RDG_genBuffer(resPtrs[i], resSizes[i], 0.10, 0.50, 1); + } + } + + /* Bench */ + { + /* init args */ + int compressionCompleted = (mode == BMK_decodeOnly); + int decompressionCompleted = (mode == BMK_compressOnly); + BMK_timedFnState_t* timeStateCompress = BMK_createTimedFnState(nbSeconds * 1000, 1000); + BMK_timedFnState_t* timeStateDecompress = BMK_createTimedFnState(nbSeconds * 1000, 1000); + BMK_initCCtxArgs cctxprep; + BMK_initDCtxArgs dctxprep; + cctxprep.cctx = cctx; + cctxprep.dictBuffer = dictBuffer; + cctxprep.dictBufferSize = dictBufferSize; + cctxprep.cLevel = cLevel; + cctxprep.comprParams = comprParams; + dctxprep.dctx = dctx; + dctxprep.dictBuffer = dictBuffer; + dctxprep.dictBufferSize = dictBufferSize; + + assert(timeStateCompress != NULL); + assert(timeStateDecompress != NULL); + while(!compressionCompleted) { + BMK_runOutcome_t const cOutcome = BMK_benchTimedFn(timeStateCompress, + &local_defaultCompress, cctx, + &local_initCCtx, &cctxprep, + nbBlocks, + srcPtrs, srcSizes, + dstPtrs, dstCapacities, + dstSizes); + + if (!BMK_isSuccessful_runOutcome(cOutcome)) { + BMK_benchOutcome_t bOut; + memset(&bOut, 0, sizeof(bOut)); + bOut.tag = 1; /* should rather be a function or a constant */ + BMK_freeTimedFnState(timeStateCompress); + BMK_freeTimedFnState(timeStateDecompress); + return bOut; + } + { BMK_runTime_t const rResult = BMK_extract_runTime(cOutcome); + bResult.cSpeed = (srcSize * TIMELOOP_NANOSEC) / rResult.nanoSecPerRun; + bResult.cSize = rResult.sumOfReturn; + } + compressionCompleted = BMK_isCompleted_TimedFn(timeStateCompress); + } + + while (!decompressionCompleted) { + BMK_runOutcome_t const dOutcome = BMK_benchTimedFn(timeStateDecompress, + &local_defaultDecompress, dctx, + &local_initDCtx, &dctxprep, + nbBlocks, + (const void* const*)dstPtrs, dstSizes, + resPtrs, resSizes, + NULL); + + if (!BMK_isSuccessful_runOutcome(dOutcome)) { + BMK_benchOutcome_t bOut; + memset(&bOut, 0, sizeof(bOut)); + bOut.tag = 1; /* should rather be a function or a constant */ + BMK_freeTimedFnState(timeStateCompress); + BMK_freeTimedFnState(timeStateDecompress); + return bOut; + } + { BMK_runTime_t const rResult = BMK_extract_runTime(dOutcome); + bResult.dSpeed = (srcSize * TIMELOOP_NANOSEC) / rResult.nanoSecPerRun; + } + decompressionCompleted = BMK_isCompleted_TimedFn(timeStateDecompress); + } + + BMK_freeTimedFnState(timeStateCompress); + BMK_freeTimedFnState(timeStateDecompress); + } + + /* Bench */ + bResult.cMem = (1 << (comprParams->vals[wlog_ind])) + ZSTD_sizeof_CCtx(cctx); + + { BMK_benchOutcome_t bOut; + bOut.tag = 0; + bOut.internal_never_use_directly = bResult; /* should be a function */ + return bOut; + } +} + +static int BMK_benchParam ( BMK_benchResult_t* resultPtr, + buffers_t buf, contexts_t ctx, + paramValues_t cParams) +{ + BMK_benchOutcome_t const outcome = BMK_benchMemInvertible(buf, ctx, + BASE_CLEVEL, &cParams, + BMK_both, 3); + int const success = BMK_isSuccessful_benchOutcome(outcome); + if (!success) return 1; + *resultPtr = BMK_extract_benchResult(outcome); + return 0; +} + + +#define CBENCHMARK(conditional, resultvar, tmpret, mode, sec) { \ + if(conditional) { \ + BMK_benchOutcome_t const outcome = BMK_benchMemInvertible(buf, ctx, BASE_CLEVEL, &cParams, mode, sec); \ + if (!BMK_isSuccessful_benchOutcome(outcome)) { \ + DEBUGOUTPUT("Benchmarking failed\n"); \ + return ERROR_RESULT; \ + } \ + { BMK_benchResult_t const tmpResult = BMK_extract_benchResult(outcome); \ + if (mode != BMK_decodeOnly) { \ + resultvar.cSpeed = tmpResult.cSpeed; \ + resultvar.cSize = tmpResult.cSize; \ + resultvar.cMem = tmpResult.cMem; \ + } \ + if (mode != BMK_compressOnly) { resultvar.dSpeed = tmpResult.dSpeed; } \ + } } \ +} + +/* Benchmarking which stops when we are sufficiently sure the solution is infeasible / worse than the winner */ +#define VARIANCE 1.2 +static int allBench(BMK_benchResult_t* resultPtr, + const buffers_t buf, const contexts_t ctx, + const paramValues_t cParams, + const constraint_t target, + BMK_benchResult_t* winnerResult, int feas) +{ + BMK_benchResult_t benchres; + double uncertaintyConstantC = 3., uncertaintyConstantD = 3.; + double winnerRS; + + BMK_benchOutcome_t const outcome = BMK_benchMemInvertible(buf, ctx, BASE_CLEVEL, &cParams, BMK_both, 2); + if (!BMK_isSuccessful_benchOutcome(outcome)) { + DEBUGOUTPUT("Benchmarking failed \n"); + return ERROR_RESULT; + } + benchres = BMK_extract_benchResult(outcome); + + winnerRS = resultScore(*winnerResult, buf.srcSize, target); + DEBUGOUTPUT("WinnerScore: %f \n ", winnerRS); + + *resultPtr = benchres; + + /* anything with worse ratio in feas is definitely worse, discard */ + if(feas && benchres.cSize < winnerResult->cSize && !g_optmode) { + return WORSE_RESULT; + } + + /* calculate uncertainty in compression / decompression runs */ + if (benchres.cSpeed) { + U64 const loopDurationC = (((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.cSpeed); + uncertaintyConstantC = ((loopDurationC + (double)(2 * g_clockGranularity))/loopDurationC); + } + + if (benchres.dSpeed) { + U64 const loopDurationD = (((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.dSpeed); + uncertaintyConstantD = ((loopDurationD + (double)(2 * g_clockGranularity))/loopDurationD); + } + + /* optimistic assumption of benchres */ + { BMK_benchResult_t resultMax = benchres; + resultMax.cSpeed *= uncertaintyConstantC * VARIANCE; + resultMax.dSpeed *= uncertaintyConstantD * VARIANCE; + + /* disregard infeasible results in feas mode */ + /* disregard if resultMax < winner in infeas mode */ + if((feas && !feasible(resultMax, target)) || + (!feas && (winnerRS > resultScore(resultMax, buf.srcSize, target)))) { + return WORSE_RESULT; + } + } + + /* compare by resultScore when in infeas */ + /* compare by compareResultLT when in feas */ + if((!feas && (resultScore(benchres, buf.srcSize, target) > resultScore(*winnerResult, buf.srcSize, target))) || + (feas && (compareResultLT(*winnerResult, benchres, target, buf.srcSize))) ) { + return BETTER_RESULT; + } else { + return WORSE_RESULT; + } +} + + +#define INFEASIBLE_THRESHOLD 200 +/* Memoized benchmarking, won't benchmark anything which has already been benchmarked before. */ +static int benchMemo(BMK_benchResult_t* resultPtr, + const buffers_t buf, const contexts_t ctx, + const paramValues_t cParams, + const constraint_t target, + BMK_benchResult_t* winnerResult, memoTable_t* const memoTableArray, + const int feas) { + static int bmcount = 0; + int res; + + if ( memoTableGet(memoTableArray, cParams) >= INFEASIBLE_THRESHOLD + || redundantParams(cParams, target, buf.maxBlockSize) ) { + return WORSE_RESULT; + } + + res = allBench(resultPtr, buf, ctx, cParams, target, winnerResult, feas); + + if(DEBUG && !(bmcount % 250)) { + DISPLAY("Count: %d\n", bmcount); + bmcount++; + } + BMK_printWinnerOpt(stdout, CUSTOM_LEVEL, *resultPtr, cParams, target, buf.srcSize); + + if(res == BETTER_RESULT || feas) { + memoTableSet(memoTableArray, cParams, 255); /* what happens if collisions are frequent */ + } + return res; +} + + +typedef struct { + U64 cSpeed_min; + U64 dSpeed_min; + U32 windowLog_max; + ZSTD_strategy strategy_max; +} level_constraints_t; + +static level_constraints_t g_level_constraint[NB_LEVELS_TRACKED+1]; + +static void BMK_init_level_constraints(int bytePerSec_level1) +{ + assert(NB_LEVELS_TRACKED >= ZSTD_maxCLevel()); + memset(g_level_constraint, 0, sizeof(g_level_constraint)); + g_level_constraint[1].cSpeed_min = bytePerSec_level1; + g_level_constraint[1].dSpeed_min = 0.; + g_level_constraint[1].windowLog_max = 19; + g_level_constraint[1].strategy_max = ZSTD_fast; + + /* establish speed objectives (relative to level 1) */ + { int l; + for (l=2; l<=NB_LEVELS_TRACKED; l++) { + g_level_constraint[l].cSpeed_min = (g_level_constraint[l-1].cSpeed_min * 49) / 64; + g_level_constraint[l].dSpeed_min = 0.; + g_level_constraint[l].windowLog_max = (l<20) ? 23 : l+5; /* only --ultra levels >= 20 can use windowlog > 23 */ + g_level_constraint[l].strategy_max = (l<19) ? ZSTD_btopt : ZSTD_btultra; /* level 19 is allowed to use btultra */ + } } +} + +static int BMK_seed(winnerInfo_t* winners, const paramValues_t params, + const buffers_t buf, const contexts_t ctx) +{ + BMK_benchResult_t testResult; int better = 0; int cLevel; - BMK_benchParam(&testResult, srcBuffer, srcSize, ctx, params); + BMK_benchParam(&testResult, buf, ctx, params); + - for (cLevel = 1; cLevel <= ZSTD_maxCLevel(); cLevel++) { - if (testResult.cSpeed < g_cSpeedTarget[cLevel]) + for (cLevel = 1; cLevel <= NB_LEVELS_TRACKED; cLevel++) { + if (testResult.cSpeed < g_level_constraint[cLevel].cSpeed_min) + continue; /* not fast enough for this level */ + if (testResult.dSpeed < g_level_constraint[cLevel].dSpeed_min) continue; /* not fast enough for this level */ + if (params.vals[wlog_ind] > g_level_constraint[cLevel].windowLog_max) + continue; /* too much memory for this level */ + if (params.vals[strt_ind] > g_level_constraint[cLevel].strategy_max) + continue; /* forbidden strategy for this level */ if (winners[cLevel].result.cSize==0) { /* first solution for this cLevel */ winners[cLevel].result = testResult; winners[cLevel].params = params; - BMK_printWinner(stdout, cLevel, testResult, params, srcSize); + BMK_printWinner(stdout, cLevel, testResult, params, buf.srcSize); better = 1; continue; } if ((double)testResult.cSize <= ((double)winners[cLevel].result.cSize * (1. + (0.02 / cLevel))) ) { /* Validate solution is "good enough" */ - double W_ratio = (double)srcSize / testResult.cSize; - double O_ratio = (double)srcSize / winners[cLevel].result.cSize; + double W_ratio = (double)buf.srcSize / testResult.cSize; + double O_ratio = (double)buf.srcSize / winners[cLevel].result.cSize; double W_ratioNote = log (W_ratio); double O_ratioNote = log (O_ratio); - size_t W_DMemUsed = (1 << params.windowLog) + (16 KB); - size_t O_DMemUsed = (1 << winners[cLevel].params.windowLog) + (16 KB); + size_t W_DMemUsed = (1 << params.vals[wlog_ind]) + (16 KB); + size_t O_DMemUsed = (1 << winners[cLevel].params.vals[wlog_ind]) + (16 KB); double W_DMemUsed_note = W_ratioNote * ( 40 + 9*cLevel) - log((double)W_DMemUsed); double O_DMemUsed_note = O_ratioNote * ( 40 + 9*cLevel) - log((double)O_DMemUsed); - size_t W_CMemUsed = (1 << params.windowLog) + ZSTD_estimateCCtxSize_usingCParams(params); - size_t O_CMemUsed = (1 << winners[cLevel].params.windowLog) + ZSTD_estimateCCtxSize_usingCParams(winners[cLevel].params); + size_t W_CMemUsed = (1 << params.vals[wlog_ind]) + ZSTD_estimateCCtxSize_usingCParams(pvalsToCParams(params)); + size_t O_CMemUsed = (1 << winners[cLevel].params.vals[wlog_ind]) + ZSTD_estimateCCtxSize_usingCParams(pvalsToCParams(winners[cLevel].params)); double W_CMemUsed_note = W_ratioNote * ( 50 + 13*cLevel) - log((double)W_CMemUsed); double O_CMemUsed_note = O_ratioNote * ( 50 + 13*cLevel) - log((double)O_CMemUsed); @@ -414,16 +1736,16 @@ static int BMK_seed(winnerInfo_t* winners, const ZSTD_compressionParameters para /* too large compression speed difference for the compression benefit */ if (W_ratio > O_ratio) DISPLAY ("Compression Speed : %5.3f @ %4.1f MB/s vs %5.3f @ %4.1f MB/s : not enough for level %i\n", - W_ratio, testResult.cSpeed / 1000000, - O_ratio, winners[cLevel].result.cSpeed / 1000000., cLevel); + W_ratio, (double)testResult.cSpeed / MB_UNIT, + O_ratio, (double)winners[cLevel].result.cSpeed / MB_UNIT, cLevel); continue; } if (W_DSpeed_note < O_DSpeed_note ) { /* too large decompression speed difference for the compression benefit */ if (W_ratio > O_ratio) DISPLAY ("Decompression Speed : %5.3f @ %4.1f MB/s vs %5.3f @ %4.1f MB/s : not enough for level %i\n", - W_ratio, testResult.dSpeed / 1000000., - O_ratio, winners[cLevel].result.dSpeed / 1000000., cLevel); + W_ratio, (double)testResult.dSpeed / MB_UNIT, + O_ratio, (double)winners[cLevel].result.dSpeed / MB_UNIT, cLevel); continue; } @@ -432,7 +1754,7 @@ static int BMK_seed(winnerInfo_t* winners, const ZSTD_compressionParameters para winners[cLevel].result = testResult; winners[cLevel].params = params; - BMK_printWinner(stdout, cLevel, testResult, params, srcSize); + BMK_printWinner(stdout, cLevel, testResult, params, buf.srcSize); better = 1; } } @@ -440,409 +1762,683 @@ static int BMK_seed(winnerInfo_t* winners, const ZSTD_compressionParameters para return better; } - -/* nullified useless params, to ensure count stats */ -static ZSTD_compressionParameters* sanitizeParams(ZSTD_compressionParameters params) -{ - g_params = params; - if (params.strategy == ZSTD_fast) - g_params.chainLog = 0, g_params.searchLog = 0; - if (params.strategy == ZSTD_dfast) - g_params.searchLog = 0; - if (params.strategy != ZSTD_btopt && params.strategy != ZSTD_btultra) - g_params.targetLength = 0; - return &g_params; -} - - -static void paramVariation(ZSTD_compressionParameters* ptr) -{ - ZSTD_compressionParameters p; - U32 validated = 0; - while (!validated) { - U32 nbChanges = (FUZ_rand(&g_rand) & 3) + 1; - p = *ptr; - for ( ; nbChanges ; nbChanges--) { - const U32 changeID = FUZ_rand(&g_rand) % 14; - switch(changeID) - { - case 0: - p.chainLog++; break; - case 1: - p.chainLog--; break; - case 2: - p.hashLog++; break; - case 3: - p.hashLog--; break; - case 4: - p.searchLog++; break; - case 5: - p.searchLog--; break; - case 6: - p.windowLog++; break; - case 7: - p.windowLog--; break; - case 8: - p.searchLength++; break; - case 9: - p.searchLength--; break; - case 10: - p.strategy = (ZSTD_strategy)(((U32)p.strategy)+1); break; - case 11: - p.strategy = (ZSTD_strategy)(((U32)p.strategy)-1); break; - case 12: - p.targetLength *= 1 + ((double)(FUZ_rand(&g_rand)&255)) / 256.; break; - case 13: - p.targetLength /= 1 + ((double)(FUZ_rand(&g_rand)&255)) / 256.; break; - } - } - validated = !ZSTD_isError(ZSTD_checkCParams(p)); - } - *ptr = p; -} - +/*-************************************ +* Compression Level Table Generation Functions +**************************************/ #define PARAMTABLELOG 25 #define PARAMTABLESIZE (1<<PARAMTABLELOG) #define PARAMTABLEMASK (PARAMTABLESIZE-1) static BYTE g_alreadyTested[PARAMTABLESIZE] = {0}; /* init to zero */ -#define NB_TESTS_PLAYED(p) \ - g_alreadyTested[(XXH64(sanitizeParams(p), sizeof(p), 0) >> 3) & PARAMTABLEMASK] - +static BYTE* NB_TESTS_PLAYED(paramValues_t p) { + ZSTD_compressionParameters p2 = pvalsToCParams(sanitizeParams(p)); + return &g_alreadyTested[(XXH64((void*)&p2, sizeof(p2), 0) >> 3) & PARAMTABLEMASK]; +} static void playAround(FILE* f, winnerInfo_t* winners, - ZSTD_compressionParameters params, - const void* srcBuffer, size_t srcSize, - ZSTD_CCtx* ctx) + paramValues_t p, + const buffers_t buf, const contexts_t ctx) { - int nbVariations = 0; + int nbVariations = 0, i; UTIL_time_t const clockStart = UTIL_getTime(); while (UTIL_clockSpanMicro(clockStart) < g_maxVariationTime) { - ZSTD_compressionParameters p = params; + BYTE* b; if (nbVariations++ > g_maxNbVariations) break; - paramVariation(&p); + + do { for(i = 0; i < 4; i++) { paramVaryOnce(FUZ_rand(&g_rand) % (strt_ind + 1), ((FUZ_rand(&g_rand) & 1) << 1) - 1, &p); } } + while(!paramValid(p)); /* exclude faster if already played params */ - if (FUZ_rand(&g_rand) & ((1 << NB_TESTS_PLAYED(p))-1)) + if (FUZ_rand(&g_rand) & ((1 << *NB_TESTS_PLAYED(p))-1)) continue; /* test */ - NB_TESTS_PLAYED(p)++; - if (!BMK_seed(winners, p, srcBuffer, srcSize, ctx)) continue; + b = NB_TESTS_PLAYED(p); + (*b)++; + if (!BMK_seed(winners, p, buf, ctx)) continue; /* improvement found => search more */ - BMK_printWinners(f, winners, srcSize); - playAround(f, winners, p, srcBuffer, srcSize, ctx); + BMK_printWinners(f, winners, buf.srcSize); + playAround(f, winners, p, buf, ctx); } } - -static ZSTD_compressionParameters randomParams(void) -{ - ZSTD_compressionParameters p; - U32 validated = 0; - while (!validated) { - /* totally random entry */ - p.chainLog = (FUZ_rand(&g_rand) % (ZSTD_CHAINLOG_MAX+1 - ZSTD_CHAINLOG_MIN)) + ZSTD_CHAINLOG_MIN; - p.hashLog = (FUZ_rand(&g_rand) % (ZSTD_HASHLOG_MAX+1 - ZSTD_HASHLOG_MIN)) + ZSTD_HASHLOG_MIN; - p.searchLog = (FUZ_rand(&g_rand) % (ZSTD_SEARCHLOG_MAX+1 - ZSTD_SEARCHLOG_MIN)) + ZSTD_SEARCHLOG_MIN; - p.windowLog = (FUZ_rand(&g_rand) % (ZSTD_WINDOWLOG_MAX+1 - ZSTD_WINDOWLOG_MIN)) + ZSTD_WINDOWLOG_MIN; - p.searchLength=(FUZ_rand(&g_rand) % (ZSTD_SEARCHLENGTH_MAX+1 - ZSTD_SEARCHLENGTH_MIN)) + ZSTD_SEARCHLENGTH_MIN; - p.targetLength=(FUZ_rand(&g_rand) % (512)) + ZSTD_TARGETLENGTH_MIN; - p.strategy = (ZSTD_strategy) (FUZ_rand(&g_rand) % (ZSTD_btultra +1)); - validated = !ZSTD_isError(ZSTD_checkCParams(p)); - } - return p; -} - static void BMK_selectRandomStart( FILE* f, winnerInfo_t* winners, - const void* srcBuffer, size_t srcSize, - ZSTD_CCtx* ctx) + const buffers_t buf, const contexts_t ctx) { - U32 const id = (FUZ_rand(&g_rand) % (ZSTD_maxCLevel()+1)); - if ((id==0) || (winners[id].params.windowLog==0)) { - /* totally random entry */ - ZSTD_compressionParameters const p = ZSTD_adjustCParams(randomParams(), srcSize, 0); - playAround(f, winners, p, srcBuffer, srcSize, ctx); + U32 const id = FUZ_rand(&g_rand) % (NB_LEVELS_TRACKED+1); + if ((id==0) || (winners[id].params.vals[wlog_ind]==0)) { + /* use some random entry */ + paramValues_t const p = adjustParams(cParamsToPVals(pvalsToCParams(randomParams())), /* defaults nonCompression parameters */ + buf.srcSize, 0); + playAround(f, winners, p, buf, ctx); + } else { + playAround(f, winners, winners[id].params, buf, ctx); } - else - playAround(f, winners, winners[id].params, srcBuffer, srcSize, ctx); } - -static void BMK_benchMem(void* srcBuffer, size_t srcSize) +static void BMK_benchFullTable(const buffers_t buf, const contexts_t ctx) { - ZSTD_CCtx* const ctx = ZSTD_createCCtx(); - ZSTD_compressionParameters params; - winnerInfo_t winners[NB_LEVELS_TRACKED]; + paramValues_t params; + winnerInfo_t winners[NB_LEVELS_TRACKED+1]; const char* const rfName = "grillResults.txt"; FILE* const f = fopen(rfName, "w"); - const size_t blockSize = g_blockSize ? g_blockSize : srcSize; /* init */ - if (ctx==NULL) { DISPLAY("ZSTD_createCCtx() failed \n"); exit(1); } + assert(g_singleRun==0); memset(winners, 0, sizeof(winners)); if (f==NULL) { DISPLAY("error opening %s \n", rfName); exit(1); } - if (g_singleRun) { - BMK_result_t testResult; - g_params = ZSTD_adjustCParams(g_params, srcSize, 0); - BMK_benchParam(&testResult, srcBuffer, srcSize, ctx, g_params); - DISPLAY("\n"); - return; - } - - if (g_target) - g_cSpeedTarget[1] = g_target * 1000000; - else { + if (g_target) { + BMK_init_level_constraints(g_target * MB_UNIT); + } else { /* baseline config for level 1 */ - BMK_result_t testResult; - params = ZSTD_getCParams(1, blockSize, 0); - BMK_benchParam(&testResult, srcBuffer, srcSize, ctx, params); - g_cSpeedTarget[1] = (testResult.cSpeed * 31) / 32; - } - - /* establish speed objectives (relative to level 1) */ - { int i; - for (i=2; i<=ZSTD_maxCLevel(); i++) - g_cSpeedTarget[i] = (g_cSpeedTarget[i-1] * 25) / 32; + paramValues_t const l1params = cParamsToPVals(ZSTD_getCParams(1, buf.maxBlockSize, ctx.dictSize)); + BMK_benchResult_t testResult; + BMK_benchParam(&testResult, buf, ctx, l1params); + BMK_init_level_constraints((int)((testResult.cSpeed * 31) / 32)); } /* populate initial solution */ { const int maxSeeds = g_noSeed ? 1 : ZSTD_maxCLevel(); int i; for (i=0; i<=maxSeeds; i++) { - params = ZSTD_getCParams(i, blockSize, 0); - BMK_seed(winners, params, srcBuffer, srcSize, ctx); + params = cParamsToPVals(ZSTD_getCParams(i, buf.maxBlockSize, 0)); + BMK_seed(winners, params, buf, ctx); } } - BMK_printWinners(f, winners, srcSize); + BMK_printWinners(f, winners, buf.srcSize); /* start tests */ - { const time_t grillStart = time(NULL); + { const UTIL_time_t grillStart = UTIL_getTime(); do { - BMK_selectRandomStart(f, winners, srcBuffer, srcSize, ctx); - } while (BMK_timeSpan(grillStart) < g_grillDuration_s); + BMK_selectRandomStart(f, winners, buf, ctx); + } while (BMK_timeSpan(grillStart) < g_timeLimit_s); } /* end summary */ - BMK_printWinners(f, winners, srcSize); + BMK_printWinners(f, winners, buf.srcSize); DISPLAY("grillParams operations completed \n"); /* clean up*/ fclose(f); - ZSTD_freeCCtx(ctx); } -static int benchSample(void) +/*-************************************ +* Single Benchmark Functions +**************************************/ + +static int benchOnce(const buffers_t buf, const contexts_t ctx, const int cLevel) { + BMK_benchResult_t testResult; + g_params = adjustParams(overwriteParams(cParamsToPVals(ZSTD_getCParams(cLevel, buf.maxBlockSize, ctx.dictSize)), g_params), buf.maxBlockSize, ctx.dictSize); + + if (BMK_benchParam(&testResult, buf, ctx, g_params)) { + DISPLAY("Error during benchmarking\n"); + return 1; + } + + BMK_printWinner(stdout, CUSTOM_LEVEL, testResult, g_params, buf.srcSize); + + return 0; +} + +static int benchSample(double compressibility, int cLevel) { - void* origBuff; - size_t const benchedSize = sampleSize; - const char* const name = "Sample 10MiB"; + const char* const name = "Sample 10MB"; + size_t const benchedSize = 10 MB; + void* const srcBuffer = malloc(benchedSize); + int ret = 0; - /* Allocation */ - origBuff = malloc(benchedSize); - if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); return 12; } + buffers_t buf; + contexts_t ctx; - /* Fill buffer */ - RDG_genBuffer(origBuff, benchedSize, g_compressibility, 0.0, 0); + if(srcBuffer == NULL) { + DISPLAY("Out of Memory\n"); + return 2; + } + + RDG_genBuffer(srcBuffer, benchedSize, compressibility, 0.0, 0); + + if(createBuffersFromMemory(&buf, srcBuffer, 1, &benchedSize)) { + DISPLAY("Buffer Creation Error\n"); + free(srcBuffer); + return 3; + } + + if(createContexts(&ctx, NULL)) { + DISPLAY("Context Creation Error\n"); + freeBuffers(buf); + return 1; + } /* bench */ DISPLAY("\r%79s\r", ""); - DISPLAY("using %s %i%%: \n", name, (int)(g_compressibility*100)); - BMK_benchMem(origBuff, benchedSize); + DISPLAY("using %s %i%%: \n", name, (int)(compressibility*100)); - free(origBuff); - return 0; -} + if(g_singleRun) { + ret = benchOnce(buf, ctx, cLevel); + } else { + BMK_benchFullTable(buf, ctx); + } + freeBuffers(buf); + freeContexts(ctx); -int benchFiles(const char** fileNamesTable, int nbFiles) + return ret; +} + +/* benchFiles() : + * note: while this function takes a table of filenames, + * in practice, only the first filename will be used */ +static int benchFiles(const char** fileNamesTable, int nbFiles, + const char* dictFileName, int cLevel) { - int fileIdx=0; - - /* Loop for each file */ - while (fileIdx<nbFiles) { - const char* const inFileName = fileNamesTable[fileIdx++]; - FILE* const inFile = fopen( inFileName, "rb" ); - U64 const inFileSize = UTIL_getFileSize(inFileName); - size_t benchedSize; - void* origBuff; - - /* Check file existence */ - if (inFile==NULL) { - DISPLAY( "Pb opening %s\n", inFileName); - return 11; - } - if (inFileSize == UTIL_FILESIZE_UNKNOWN) { - DISPLAY("Pb evaluatin size of %s \n", inFileName); - fclose(inFile); - return 11; - } + buffers_t buf; + contexts_t ctx; + int ret = 0; - /* Memory allocation */ - benchedSize = BMK_findMaxMem(inFileSize*3) / 3; - if ((U64)benchedSize > inFileSize) benchedSize = (size_t)inFileSize; - if (benchedSize < inFileSize) - DISPLAY("Not enough memory for '%s' full size; testing %i MB only...\n", inFileName, (int)(benchedSize>>20)); - origBuff = malloc(benchedSize); - if (origBuff==NULL) { - DISPLAY("\nError: not enough memory!\n"); - fclose(inFile); - return 12; - } + if (createBuffers(&buf, fileNamesTable, nbFiles)) { + DISPLAY("unable to load files\n"); + return 1; + } - /* Fill input buffer */ - DISPLAY("Loading %s... \r", inFileName); - { size_t const readSize = fread(origBuff, 1, benchedSize, inFile); - fclose(inFile); - if(readSize != benchedSize) { - DISPLAY("\nError: problem reading file '%s' !! \n", inFileName); - free(origBuff); - return 13; - } } + if (createContexts(&ctx, dictFileName)) { + DISPLAY("unable to load dictionary\n"); + freeBuffers(buf); + return 2; + } - /* bench */ - DISPLAY("\r%79s\r", ""); - DISPLAY("using %s : \n", inFileName); - BMK_benchMem(origBuff, benchedSize); + DISPLAY("\r%79s\r", ""); + if (nbFiles == 1) { + DISPLAY("using %s : \n", fileNamesTable[0]); + } else { + DISPLAY("using %d Files : \n", nbFiles); + } - /* clean */ - free(origBuff); + if (g_singleRun) { + ret = benchOnce(buf, ctx, cLevel); + } else { + BMK_benchFullTable(buf, ctx); } - return 0; + freeBuffers(buf); + freeContexts(ctx); + return ret; } -static void BMK_translateAdvancedParams(ZSTD_compressionParameters params) +/*-************************************ +* Local Optimization Functions +**************************************/ + +/* One iteration of hill climbing. Specifically, it first tries all + * valid parameter configurations w/ manhattan distance 1 and picks the best one + * failing that, it progressively tries candidates further and further away (up to #dim + 2) + * if it finds a candidate exceeding winnerInfo, it will repeat. Otherwise, it will stop the + * current stage of hill climbing. + * Each iteration of hill climbing proceeds in 2 'phases'. Phase 1 climbs according to + * the resultScore function, which is effectively a linear increase in reward until it reaches + * the constraint-satisfying value, it which point any excess results in only logarithmic reward. + * This aims to find some constraint-satisfying point. + * Phase 2 optimizes in accordance with what the original function sets out to maximize, with + * all feasible solutions valued over all infeasible solutions. + */ + +/* sanitize all params here. + * all generation after random should be sanitized. (maybe sanitize random) + */ +static winnerInfo_t climbOnce(const constraint_t target, + memoTable_t* mtAll, + const buffers_t buf, const contexts_t ctx, + const paramValues_t init) { - DISPLAY("--zstd=windowLog=%u,chainLog=%u,hashLog=%u,searchLog=%u,searchLength=%u,targetLength=%u,strategy=%u \n", - params.windowLog, params.chainLog, params.hashLog, params.searchLog, params.searchLength, params.targetLength, (U32)(params.strategy)); + /* + * cparam - currently considered 'center' + * candidate - params to benchmark/results + * winner - best option found so far. + */ + paramValues_t cparam = init; + winnerInfo_t candidateInfo, winnerInfo; + int better = 1; + int feas = 0; + + winnerInfo = initWinnerInfo(init); + candidateInfo = winnerInfo; + + { winnerInfo_t bestFeasible1 = initWinnerInfo(cparam); + DEBUGOUTPUT("Climb Part 1\n"); + while(better) { + int offset; + size_t i, dist; + const size_t varLen = mtAll[cparam.vals[strt_ind]].varLen; + better = 0; + DEBUGOUTPUT("Start\n"); + cparam = winnerInfo.params; + candidateInfo.params = cparam; + /* all dist-1 candidates */ + for (i = 0; i < varLen; i++) { + for (offset = -1; offset <= 1; offset += 2) { + CHECKTIME(winnerInfo); + candidateInfo.params = cparam; + paramVaryOnce(mtAll[cparam.vals[strt_ind]].varArray[i], offset, &candidateInfo.params); + + if(paramValid(candidateInfo.params)) { + int res; + res = benchMemo(&candidateInfo.result, buf, ctx, + sanitizeParams(candidateInfo.params), target, &winnerInfo.result, mtAll, feas); + DEBUGOUTPUT("Res: %d\n", res); + if(res == BETTER_RESULT) { /* synonymous with better when called w/ infeasibleBM */ + winnerInfo = candidateInfo; + better = 1; + if(compareResultLT(bestFeasible1.result, winnerInfo.result, target, buf.srcSize)) { + bestFeasible1 = winnerInfo; + } + } + } + } + } /* for (i = 0; i < varLen; i++) */ + + if(better) { + continue; + } + + for(dist = 2; dist < varLen + 2; dist++) { /* varLen is # dimensions */ + for(i = 0; i < (1 << varLen) / varLen + 2; i++) { + int res; + CHECKTIME(winnerInfo); + candidateInfo.params = cparam; + /* param error checking already done here */ + paramVariation(&candidateInfo.params, mtAll, (U32)dist); + + res = benchMemo(&candidateInfo.result, + buf, ctx, + sanitizeParams(candidateInfo.params), target, + &winnerInfo.result, mtAll, feas); + DEBUGOUTPUT("Res: %d\n", res); + if (res == BETTER_RESULT) { /* synonymous with better in this case*/ + winnerInfo = candidateInfo; + better = 1; + if (compareResultLT(bestFeasible1.result, winnerInfo.result, target, buf.srcSize)) { + bestFeasible1 = winnerInfo; + } + break; + } + } + + if (better) { + break; + } + } /* for(dist = 2; dist < varLen + 2; dist++) */ + + if (!better) { /* infeas -> feas -> stop */ + if (feas) return winnerInfo; + feas = 1; + better = 1; + winnerInfo = bestFeasible1; /* note with change, bestFeasible may not necessarily be feasible, but if one has been benchmarked, it will be. */ + DEBUGOUTPUT("Climb Part 2\n"); + } + } + winnerInfo = bestFeasible1; + } + + return winnerInfo; } -/* optimizeForSize(): - * targetSpeed : expressed in MB/s */ -int optimizeForSize(const char* inFileName, U32 targetSpeed) +/* Optimizes for a fixed strategy */ + +/* flexible parameters: iterations of failed climbing (or if we do non-random, maybe this is when everything is close to visitied) + weight more on visit for bad results, less on good results/more on later results / ones with more failures. + allocate memoTable here. + */ +static winnerInfo_t optimizeFixedStrategy( + const buffers_t buf, const contexts_t ctx, + const constraint_t target, paramValues_t paramTarget, + const ZSTD_strategy strat, + memoTable_t* memoTableArray, const int tries) { + int i = 0; + + paramValues_t init; + winnerInfo_t winnerInfo, candidateInfo; + winnerInfo = initWinnerInfo(emptyParams()); + /* so climb is given the right fixed strategy */ + paramTarget.vals[strt_ind] = strat; + /* to pass ZSTD_checkCParams */ + paramTarget = cParamUnsetMin(paramTarget); + + init = paramTarget; + + for(i = 0; i < tries; i++) { + DEBUGOUTPUT("Restart\n"); + do { randomConstrainedParams(&init, memoTableArray, strat); } while(redundantParams(init, target, buf.maxBlockSize)); + candidateInfo = climbOnce(target, memoTableArray, buf, ctx, init); + if(compareResultLT(winnerInfo.result, candidateInfo.result, target, buf.srcSize)) { + winnerInfo = candidateInfo; + BMK_printWinnerOpt(stdout, CUSTOM_LEVEL, winnerInfo.result, winnerInfo.params, target, buf.srcSize); + i = 0; + continue; + } + CHECKTIME(winnerInfo); + i++; + } + return winnerInfo; +} + +/* goes best, best-1, best+1, best-2, ... */ +/* return 0 if nothing remaining */ +static int nextStrategy(const int currentStrategy, const int bestStrategy) { + if(bestStrategy <= currentStrategy) { + int candidate = 2 * bestStrategy - currentStrategy - 1; + if(candidate < 1) { + candidate = currentStrategy + 1; + if(candidate > (int)ZSTD_btultra) { + return 0; + } else { + return candidate; + } + } else { + return candidate; + } + } else { /* bestStrategy >= currentStrategy */ + int candidate = 2 * bestStrategy - currentStrategy; + if(candidate > (int)ZSTD_btultra) { + candidate = currentStrategy - 1; + if(candidate < 1) { + return 0; + } else { + return candidate; + } + } else { + return candidate; + } + } +} + +/* experiment with playing with this and decay value */ + +/* main fn called when using --optimize */ +/* Does strategy selection by benchmarking default compression levels + * then optimizes by strategy, starting with the best one and moving + * progressively moving further away by number + * args: + * fileNamesTable - list of files to benchmark + * nbFiles - length of fileNamesTable + * dictFileName - name of dictionary file if one, else NULL + * target - performance constraints (cSpeed, dSpeed, cMem) + * paramTarget - parameter constraints (i.e. restriction search space to where strategy = ZSTD_fast) + * cLevel - compression level to exceed (all solutions must be > lvl in cSpeed + ratio) + */ + +static int g_maxTries = 5; +#define TRY_DECAY 1 + +static int optimizeForSize(const char* const * const fileNamesTable, const size_t nbFiles, const char* dictFileName, constraint_t target, paramValues_t paramTarget, + const int cLevelOpt, const int cLevelRun, const U32 memoTableLog) { - FILE* const inFile = fopen( inFileName, "rb" ); - U64 const inFileSize = UTIL_getFileSize(inFileName); - size_t benchedSize = BMK_findMaxMem(inFileSize*3) / 3; - void* origBuff; - - /* Init */ - if (inFile==NULL) { DISPLAY( "Pb opening %s\n", inFileName); return 11; } - if (inFileSize == UTIL_FILESIZE_UNKNOWN) { - DISPLAY("Pb evaluatin size of %s \n", inFileName); - fclose(inFile); - return 11; - } - - /* Memory allocation & restrictions */ - if ((U64)benchedSize > inFileSize) benchedSize = (size_t)inFileSize; - if (benchedSize < inFileSize) { - DISPLAY("Not enough memory for '%s' \n", inFileName); - fclose(inFile); - return 11; - } - - /* Alloc */ - origBuff = malloc(benchedSize); - if(!origBuff) { - DISPLAY("\nError: not enough memory!\n"); - fclose(inFile); - return 12; - } - - /* Fill input buffer */ - DISPLAY("Loading %s... \r", inFileName); - { size_t const readSize = fread(origBuff, 1, benchedSize, inFile); - fclose(inFile); - if(readSize != benchedSize) { - DISPLAY("\nError: problem reading file '%s' !! \n", inFileName); - free(origBuff); - return 13; - } } + varInds_t varArray [NUM_PARAMS]; + int ret = 0; + const size_t varLen = variableParams(paramTarget, varArray, dictFileName != NULL); + winnerInfo_t winner = initWinnerInfo(emptyParams()); + memoTable_t* allMT = NULL; + paramValues_t paramBase; + contexts_t ctx; + buffers_t buf; + g_time = UTIL_getTime(); + + if(createBuffers(&buf, fileNamesTable, nbFiles)) { + DISPLAY("unable to load files\n"); + return 1; + } + + if(createContexts(&ctx, dictFileName)) { + DISPLAY("unable to load dictionary\n"); + freeBuffers(buf); + return 2; + } + + if(nbFiles == 1) { + DISPLAYLEVEL(2, "Loading %s... \r", fileNamesTable[0]); + } else { + DISPLAYLEVEL(2, "Loading %lu Files... \r", (unsigned long)nbFiles); + } + + /* sanitize paramTarget */ + optimizerAdjustInput(¶mTarget, buf.maxBlockSize); + paramBase = cParamUnsetMin(paramTarget); + + allMT = createMemoTableArray(paramTarget, varArray, varLen, memoTableLog); + + if (!allMT) { + DISPLAY("MemoTable Init Error\n"); + ret = 2; + goto _cleanUp; + } + + /* default strictnesses */ + if (g_strictness == PARAM_UNSET) { + if(g_optmode) { + g_strictness = 100; + } else { + g_strictness = 90; + } + } else { + if(0 >= g_strictness || g_strictness > 100) { + DISPLAY("Strictness Outside of Bounds\n"); + ret = 4; + goto _cleanUp; + } + } + + /* use level'ing mode instead of normal target mode */ + if (g_optmode) { + winner.params = cParamsToPVals(ZSTD_getCParams(cLevelOpt, buf.maxBlockSize, ctx.dictSize)); + if(BMK_benchParam(&winner.result, buf, ctx, winner.params)) { + ret = 3; + goto _cleanUp; + } + + g_lvltarget = winner.result; + g_lvltarget.cSpeed *= ((double)g_strictness) / 100; + g_lvltarget.dSpeed *= ((double)g_strictness) / 100; + g_lvltarget.cSize /= ((double)g_strictness) / 100; + + target.cSpeed = (U32)g_lvltarget.cSpeed; + target.dSpeed = (U32)g_lvltarget.dSpeed; + + BMK_printWinnerOpt(stdout, cLevelOpt, winner.result, winner.params, target, buf.srcSize); + } + + /* Don't want it to return anything worse than the best known result */ + if (g_singleRun) { + BMK_benchResult_t res; + g_params = adjustParams(overwriteParams(cParamsToPVals(ZSTD_getCParams(cLevelRun, buf.maxBlockSize, ctx.dictSize)), g_params), buf.maxBlockSize, ctx.dictSize); + if (BMK_benchParam(&res, buf, ctx, g_params)) { + ret = 45; + goto _cleanUp; + } + if(compareResultLT(winner.result, res, relaxTarget(target), buf.srcSize)) { + winner.result = res; + winner.params = g_params; + } + } /* bench */ - DISPLAY("\r%79s\r", ""); - DISPLAY("optimizing for %s - limit speed %u MB/s \n", inFileName, targetSpeed); - targetSpeed *= 1000000; + DISPLAYLEVEL(2, "\r%79s\r", ""); + if(nbFiles == 1) { + DISPLAYLEVEL(2, "optimizing for %s", fileNamesTable[0]); + } else { + DISPLAYLEVEL(2, "optimizing for %lu Files", (unsigned long)nbFiles); + } + + if(target.cSpeed != 0) { DISPLAYLEVEL(2," - limit compression speed %u MB/s", target.cSpeed >> 20); } + if(target.dSpeed != 0) { DISPLAYLEVEL(2, " - limit decompression speed %u MB/s", target.dSpeed >> 20); } + if(target.cMem != (U32)-1) { DISPLAYLEVEL(2, " - limit memory %u MB", target.cMem >> 20); } - { ZSTD_CCtx* const ctx = ZSTD_createCCtx(); - winnerInfo_t winner; - BMK_result_t candidate; - const size_t blockSize = g_blockSize ? g_blockSize : benchedSize; + DISPLAYLEVEL(2, "\n"); + findClockGranularity(); - /* init */ - if (ctx==NULL) { DISPLAY("\n ZSTD_createCCtx error \n"); free(origBuff); return 14;} - memset(&winner, 0, sizeof(winner)); - winner.result.cSize = (size_t)(-1); + { paramValues_t CParams; /* find best solution from default params */ - { const int maxSeeds = g_noSeed ? 1 : ZSTD_maxCLevel(); - int i; - for (i=1; i<=maxSeeds; i++) { - ZSTD_compressionParameters const CParams = ZSTD_getCParams(i, blockSize, 0); - BMK_benchParam(&candidate, origBuff, benchedSize, ctx, CParams); - if (candidate.cSpeed < targetSpeed) - break; - if ( (candidate.cSize < winner.result.cSize) - | ((candidate.cSize == winner.result.cSize) & (candidate.cSpeed > winner.result.cSpeed)) ) - { - winner.params = CParams; - winner.result = candidate; - BMK_printWinner(stdout, i, winner.result, winner.params, benchedSize); - } } + { + /* strategy selection */ + const int maxSeeds = g_noSeed ? 1 : ZSTD_maxCLevel(); + DEBUGOUTPUT("Strategy Selection\n"); + if(paramTarget.vals[strt_ind] == PARAM_UNSET) { + BMK_benchResult_t candidate; + int i; + for (i=1; i<=maxSeeds; i++) { + int ec; + CParams = overwriteParams(cParamsToPVals(ZSTD_getCParams(i, buf.maxBlockSize, ctx.dictSize)), paramTarget); + ec = BMK_benchParam(&candidate, buf, ctx, CParams); + BMK_printWinnerOpt(stdout, i, candidate, CParams, target, buf.srcSize); + + if(!ec && compareResultLT(winner.result, candidate, relaxTarget(target), buf.srcSize)) { + winner.result = candidate; + winner.params = CParams; + } + + CHECKTIMEGT(ret, 0, _displayCleanUp); /* if pass time limit, stop */ + /* if the current params are too slow, just stop. */ + if(target.cSpeed > candidate.cSpeed * 3 / 2) { break; } + } + + BMK_printWinnerOpt(stdout, CUSTOM_LEVEL, winner.result, winner.params, target, buf.srcSize); + } } - BMK_printWinner(stdout, 99, winner.result, winner.params, benchedSize); - BMK_translateAdvancedParams(winner.params); - - /* start tests */ - { time_t const grillStart = time(NULL); - do { - ZSTD_compressionParameters params = winner.params; - paramVariation(¶ms); - if ((FUZ_rand(&g_rand) & 31) == 3) params = randomParams(); /* totally random config to improve search space */ - params = ZSTD_adjustCParams(params, blockSize, 0); - - /* exclude faster if already played set of params */ - if (FUZ_rand(&g_rand) & ((1 << NB_TESTS_PLAYED(params))-1)) continue; - - /* test */ - NB_TESTS_PLAYED(params)++; - BMK_benchParam(&candidate, origBuff, benchedSize, ctx, params); - - /* improvement found => new winner */ - if ( (candidate.cSpeed > targetSpeed) - & ( (candidate.cSize < winner.result.cSize) - | ((candidate.cSize == winner.result.cSize) & (candidate.cSpeed > winner.result.cSpeed)) ) ) + + DEBUGOUTPUT("Real Opt\n"); + /* start 'real' optimization */ + { + int bestStrategy = (int)winner.params.vals[strt_ind]; + if(paramTarget.vals[strt_ind] == PARAM_UNSET) { + int st = bestStrategy; + int tries = g_maxTries; + { - winner.params = params; - winner.result = candidate; - BMK_printWinner(stdout, 99, winner.result, winner.params, benchedSize); - BMK_translateAdvancedParams(winner.params); + /* one iterations of hill climbing with the level-defined parameters. */ + winnerInfo_t w1 = climbOnce(target, allMT, buf, ctx, winner.params); + if(compareResultLT(winner.result, w1.result, target, buf.srcSize)) { + winner = w1; + } + CHECKTIMEGT(ret, 0, _displayCleanUp); + } + + while(st && tries > 0) { + winnerInfo_t wc; + DEBUGOUTPUT("StrategySwitch: %s\n", g_stratName[st]); + + wc = optimizeFixedStrategy(buf, ctx, target, paramBase, st, allMT, tries); + + if(compareResultLT(winner.result, wc.result, target, buf.srcSize)) { + winner = wc; + tries = g_maxTries; + bestStrategy = st; + } else { + st = nextStrategy(st, bestStrategy); + tries -= TRY_DECAY; + } + CHECKTIMEGT(ret, 0, _displayCleanUp); } - } while (BMK_timeSpan(grillStart) < g_grillDuration_s); + } else { + winner = optimizeFixedStrategy(buf, ctx, target, paramBase, paramTarget.vals[strt_ind], allMT, g_maxTries); + } + } + /* no solution found */ + if(winner.result.cSize == (size_t)-1) { + ret = 1; + DISPLAY("No feasible solution found\n"); + goto _cleanUp; + } /* end summary */ - BMK_printWinner(stdout, 99, winner.result, winner.params, benchedSize); - DISPLAY("grillParams size - optimizer completed \n"); +_displayCleanUp: + if(g_displayLevel >= 0) { BMK_displayOneResult(stdout, winner, buf.srcSize); } + BMK_translateAdvancedParams(stdout, winner.params); + DISPLAYLEVEL(1, "grillParams size - optimizer completed \n"); - /* clean up*/ - ZSTD_freeCCtx(ctx); } +_cleanUp: + freeContexts(ctx); + freeBuffers(buf); + freeMemoTableArray(allMT); + return ret; +} - free(origBuff); - return 0; +/*-************************************ +* CLI parsing functions +**************************************/ + +/** longCommandWArg() : + * check if *stringPtr is the same as longCommand. + * If yes, @return 1 and advances *stringPtr to the position which immediately follows longCommand. + * @return 0 and doesn't modify *stringPtr otherwise. + * from zstdcli.c + */ +static unsigned longCommandWArg(const char** stringPtr, const char* longCommand) +{ + size_t const comSize = strlen(longCommand); + int const result = !strncmp(*stringPtr, longCommand, comSize); + if (result) *stringPtr += comSize; + return result; +} + +static void errorOut(const char* msg) +{ + DISPLAY("%s \n", msg); exit(1); } +/*! readU32FromChar() : + * @return : unsigned integer value read from input in `char` format. + * allows and interprets K, KB, KiB, M, MB and MiB suffix. + * Will also modify `*stringPtr`, advancing it to position where it stopped reading. + * Note : function will exit() program if digit sequence overflows */ +static unsigned readU32FromChar(const char** stringPtr) +{ + const char errorMsg[] = "error: numeric value too large"; + unsigned sign = 1; + unsigned result = 0; + if(**stringPtr == '-') { sign = (unsigned)-1; (*stringPtr)++; } + while ((**stringPtr >='0') && (**stringPtr <='9')) { + unsigned const max = (((unsigned)(-1)) / 10) - 1; + if (result > max) errorOut(errorMsg); + result *= 10, result += **stringPtr - '0', (*stringPtr)++ ; + } + if ((**stringPtr=='K') || (**stringPtr=='M')) { + unsigned const maxK = ((unsigned)(-1)) >> 10; + if (result > maxK) errorOut(errorMsg); + result <<= 10; + if (**stringPtr=='M') { + if (result > maxK) errorOut(errorMsg); + result <<= 10; + } + (*stringPtr)++; /* skip `K` or `M` */ + if (**stringPtr=='i') (*stringPtr)++; + if (**stringPtr=='B') (*stringPtr)++; + } + return result * sign; +} + +static double readDoubleFromChar(const char** stringPtr) +{ + double result = 0, divide = 10; + while ((**stringPtr >='0') && (**stringPtr <='9')) { + result *= 10, result += **stringPtr - '0', (*stringPtr)++ ; + } + if(**stringPtr!='.') { + return result; + } + (*stringPtr)++; + while ((**stringPtr >='0') && (**stringPtr <='9')) { + result += (double)(**stringPtr - '0') / divide, divide *= 10, (*stringPtr)++ ; + } + return result; +} static int usage(const char* exename) { @@ -857,12 +2453,16 @@ static int usage(const char* exename) static int usage_advanced(void) { DISPLAY( "\nAdvanced options :\n"); - DISPLAY( " -T# : set level 1 speed objective \n"); - DISPLAY( " -B# : cut input into blocks of size # (default : single block) \n"); - DISPLAY( " -i# : iteration loops [1-9](default : %i) \n", NBLOOPS); - DISPLAY( " -O# : find Optimized parameters for # MB/s compression speed (default : 0) \n"); - DISPLAY( " -S : Single run \n"); - DISPLAY( " -P# : generated sample compressibility (default : %.1f%%) \n", COMPRESSIBILITY_DEFAULT * 100); + DISPLAY( " -T# : set level 1 speed objective \n"); + DISPLAY( " -B# : cut input into blocks of size # (default : single block) \n"); + DISPLAY( " --optimize= : same as -O with more verbose syntax (see README.md)\n"); + DISPLAY( " -S : Single run \n"); + DISPLAY( " --zstd : Single run, parameter selection same as zstdcli \n"); + DISPLAY( " -P# : generated sample compressibility (default : %.1f%%) \n", COMPRESSIBILITY_DEFAULT * 100); + DISPLAY( " -t# : Caps runtime of operation in seconds (default : %u seconds (%.1f hours)) \n", g_timeLimit_s, (double)g_timeLimit_s / 3600); + DISPLAY( " -v : Prints Benchmarking output\n"); + DISPLAY( " -D : Next argument dictionary file\n"); + DISPLAY( " -s : Seperate Files\n"); return 0; } @@ -873,37 +2473,135 @@ static int badusage(const char* exename) return 1; } +#define PARSE_SUB_ARGS(stringLong, stringShort, variable) { if (longCommandWArg(&argument, stringLong) || longCommandWArg(&argument, stringShort)) { variable = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } } +/* 1 if successful parse, 0 otherwise */ +static int parse_params(const char** argptr, paramValues_t* pv) { + int matched = 0; + const char* argOrig = *argptr; + varInds_t v; + for(v = 0; v < NUM_PARAMS; v++) { + if(longCommandWArg(argptr,g_shortParamNames[v]) || longCommandWArg(argptr, g_paramNames[v])) { + if(**argptr == '=') { + (*argptr)++; + pv->vals[v] = readU32FromChar(argptr); + matched = 1; + break; + } + } + /* reset and try again */ + *argptr = argOrig; + } + return matched; +} + +/*-************************************ +* Main +**************************************/ + int main(int argc, const char** argv) { int i, filenamesStart=0, result; const char* exename=argv[0]; - const char* input_filename=0; - U32 optimizer = 0; + const char* input_filename = NULL; + const char* dictFileName = NULL; U32 main_pause = 0; - U32 targetSpeed = 0; + int cLevelOpt = 0, cLevelRun = 0; + int seperateFiles = 0; + double compressibility = COMPRESSIBILITY_DEFAULT; + U32 memoTableLog = PARAM_UNSET; + constraint_t target = { 0, 0, (U32)-1 }; - /* checks */ - if (NB_LEVELS_TRACKED <= ZSTD_maxCLevel()) { - DISPLAY("Error : NB_LEVELS_TRACKED <= ZSTD_maxCLevel() \n"); - exit(1); - } + paramValues_t paramTarget = emptyParams(); + g_params = emptyParams(); - /* Welcome message */ - DISPLAY(WELCOME_MESSAGE); - - if (argc<1) { badusage(exename); return 1; } + assert(argc>=1); /* for exename */ for(i=1; i<argc; i++) { const char* argument = argv[i]; - - if(!argument) continue; /* Protection if argument empty */ + DEBUGOUTPUT("%d: %s\n", i, argument); + assert(argument != NULL); if(!strcmp(argument,"--no-seed")) { g_noSeed = 1; continue; } + if (longCommandWArg(&argument, "--optimize=")) { + g_optimizer = 1; + for ( ; ;) { + if(parse_params(&argument, ¶mTarget)) { if(argument[0] == ',') { argument++; continue; } else break; } + PARSE_SUB_ARGS("compressionSpeed=" , "cSpeed=", target.cSpeed); + PARSE_SUB_ARGS("decompressionSpeed=", "dSpeed=", target.dSpeed); + PARSE_SUB_ARGS("compressionMemory=" , "cMem=", target.cMem); + PARSE_SUB_ARGS("strict=", "stc=", g_strictness); + PARSE_SUB_ARGS("maxTries=", "tries=", g_maxTries); + PARSE_SUB_ARGS("memoLimitLog=", "memLog=", memoTableLog); + if (longCommandWArg(&argument, "level=") || longCommandWArg(&argument, "lvl=")) { cLevelOpt = readU32FromChar(&argument); g_optmode = 1; if (argument[0]==',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "speedForRatio=") || longCommandWArg(&argument, "speedRatio=")) { g_ratioMultiplier = readDoubleFromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; } + + DISPLAY("invalid optimization parameter \n"); + return 1; + } + + if (argument[0] != 0) { + DISPLAY("invalid --optimize= format\n"); + return 1; /* check the end of string */ + } + continue; + } else if (longCommandWArg(&argument, "--zstd=")) { /* Decode command (note : aggregated commands are allowed) */ - if (argument[0]=='-') { + g_singleRun = 1; + for ( ; ;) { + if(parse_params(&argument, &g_params)) { if(argument[0] == ',') { argument++; continue; } else break; } + if (longCommandWArg(&argument, "level=") || longCommandWArg(&argument, "lvl=")) { cLevelRun = readU32FromChar(&argument); g_params = emptyParams(); if (argument[0]==',') { argument++; continue; } else break; } + + DISPLAY("invalid compression parameter \n"); + return 1; + } + + if (argument[0] != 0) { + DISPLAY("invalid --zstd= format\n"); + return 1; /* check the end of string */ + } + continue; + /* if not return, success */ + + } else if (longCommandWArg(&argument, "--display=")) { + /* Decode command (note : aggregated commands are allowed) */ + memset(g_silenceParams, 1, sizeof(g_silenceParams)); + for ( ; ;) { + int found = 0; + varInds_t v; + for(v = 0; v < NUM_PARAMS; v++) { + if(longCommandWArg(&argument, g_shortParamNames[v]) || longCommandWArg(&argument, g_paramNames[v])) { + g_silenceParams[v] = 0; + found = 1; + } + } + if(longCommandWArg(&argument, "compressionParameters") || longCommandWArg(&argument, "cParams")) { + for(v = 0; v <= strt_ind; v++) { + g_silenceParams[v] = 0; + } + found = 1; + } + + + if(found) { + if(argument[0]==',') { + continue; + } else { + break; + } + } + DISPLAY("invalid parameter name parameter \n"); + return 1; + } + + if (argument[0] != 0) { + DISPLAY("invalid --display format\n"); + return 1; /* check the end of string */ + } + continue; + } else if (argument[0]=='-') { argument++; while (argument[0]!=0) { @@ -917,114 +2615,110 @@ int main(int argc, const char** argv) /* Pause at the end (hidden option) */ case 'p': main_pause = 1; argument++; break; - /* Modify Nb Iterations */ - case 'i': - argument++; - if ((argument[0] >='0') & (argument[0] <='9')) - g_nbIterations = *argument++ - '0'; - break; - /* Sample compressibility (when no file provided) */ case 'P': argument++; - { U32 proba32 = 0; - while ((argument[0]>= '0') & (argument[0]<= '9')) - proba32 = (proba32*10) + (*argument++ - '0'); - g_compressibility = (double)proba32 / 100.; + { U32 const proba32 = readU32FromChar(&argument); + compressibility = (double)proba32 / 100.; } break; - case 'O': - argument++; - optimizer=1; - targetSpeed = 0; - while ((*argument >= '0') & (*argument <= '9')) - targetSpeed = (targetSpeed*10) + (*argument++ - '0'); - break; - /* Run Single conf */ case 'S': g_singleRun = 1; argument++; - g_params = ZSTD_getCParams(2, g_blockSize, 0); for ( ; ; ) { switch(*argument) { case 'w': - g_params.windowLog = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.windowLog *= 10, g_params.windowLog += *argument++ - '0'; + g_params.vals[wlog_ind] = readU32FromChar(&argument); continue; case 'c': - g_params.chainLog = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.chainLog *= 10, g_params.chainLog += *argument++ - '0'; + g_params.vals[clog_ind] = readU32FromChar(&argument); continue; case 'h': - g_params.hashLog = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.hashLog *= 10, g_params.hashLog += *argument++ - '0'; + g_params.vals[hlog_ind] = readU32FromChar(&argument); continue; case 's': - g_params.searchLog = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.searchLog *= 10, g_params.searchLog += *argument++ - '0'; + g_params.vals[slog_ind] = readU32FromChar(&argument); continue; case 'l': /* search length */ - g_params.searchLength = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.searchLength *= 10, g_params.searchLength += *argument++ - '0'; + g_params.vals[slen_ind] = readU32FromChar(&argument); continue; case 't': /* target length */ - g_params.targetLength = 0; argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.targetLength *= 10, g_params.targetLength += *argument++ - '0'; + g_params.vals[tlen_ind] = readU32FromChar(&argument); continue; case 'S': /* strategy */ argument++; - while ((*argument>= '0') && (*argument<='9')) - g_params.strategy = (ZSTD_strategy)(*argument++ - '0'); + g_params.vals[strt_ind] = readU32FromChar(&argument); + continue; + case 'f': /* forceAttachDict */ + argument++; + g_params.vals[fadt_ind] = readU32FromChar(&argument); continue; case 'L': - { int cLevel = 0; - argument++; - while ((*argument>= '0') && (*argument<='9')) - cLevel *= 10, cLevel += *argument++ - '0'; - g_params = ZSTD_getCParams(cLevel, g_blockSize, 0); + { argument++; + cLevelRun = readU32FromChar(&argument); + g_params = emptyParams(); continue; } default : ; } break; } + break; /* target level1 speed objective, in MB/s */ case 'T': argument++; - g_target = 0; - while ((*argument >= '0') && (*argument <= '9')) - g_target = (g_target*10) + (*argument++ - '0'); + g_target = readU32FromChar(&argument); break; /* cut input into blocks */ case 'B': - g_blockSize = 0; argument++; - while ((*argument >='0') & (*argument <='9')) - g_blockSize = (g_blockSize*10) + (*argument++ - '0'); - if (*argument=='K') g_blockSize<<=10, argument++; /* allows using KB notation */ - if (*argument=='M') g_blockSize<<=20, argument++; - if (*argument=='B') argument++; + g_blockSize = readU32FromChar(&argument); DISPLAY("using %u KB block size \n", g_blockSize>>10); break; + /* caps runtime (in seconds) */ + case 't': + argument++; + g_timeLimit_s = readU32FromChar(&argument); + break; + + case 's': + argument++; + seperateFiles = 1; + break; + + case 'q': + while (argument[0] == 'q') { argument++; g_displayLevel--; } + break; + + case 'v': + while (argument[0] == 'v') { argument++; g_displayLevel++; } + break; + + /* load dictionary file (only applicable for optimizer rn) */ + case 'D': + if(i == argc - 1) { /* last argument, return error. */ + DISPLAY("Dictionary file expected but not given : %d\n", i); + return 1; + } else { + i++; + dictFileName = argv[i]; + argument += strlen(argument); + } + break; + /* Unknown command */ default : return badusage(exename); } @@ -1036,13 +2730,34 @@ int main(int argc, const char** argv) if (!input_filename) { input_filename=argument; filenamesStart=i; continue; } } - if (filenamesStart==0) - result = benchSample(); - else { - if (optimizer) - result = optimizeForSize(input_filename, targetSpeed); - else - result = benchFiles(argv+filenamesStart, argc-filenamesStart); + /* Welcome message */ + DISPLAYLEVEL(2, WELCOME_MESSAGE); + + if (filenamesStart==0) { + if (g_optimizer) { + DISPLAY("Optimizer Expects File\n"); + return 1; + } else { + result = benchSample(compressibility, cLevelRun); + } + } else { + if(seperateFiles) { + for(i = 0; i < argc - filenamesStart; i++) { + if (g_optimizer) { + result = optimizeForSize(argv+filenamesStart + i, 1, dictFileName, target, paramTarget, cLevelOpt, cLevelRun, memoTableLog); + if(result) { DISPLAY("Error on File %d", i); return result; } + } else { + result = benchFiles(argv+filenamesStart + i, 1, dictFileName, cLevelRun); + if(result) { DISPLAY("Error on File %d", i); return result; } + } + } + } else { + if (g_optimizer) { + result = optimizeForSize(argv+filenamesStart, argc-filenamesStart, dictFileName, target, paramTarget, cLevelOpt, cLevelRun, memoTableLog); + } else { + result = benchFiles(argv+filenamesStart, argc-filenamesStart, dictFileName, cLevelRun); + } + } } if (main_pause) { int unused; printf("press enter...\n"); unused = getchar(); (void)unused; } diff --git a/tests/playTests.sh b/tests/playTests.sh index 41d8263b6a674..b86a0dc40cb20 100755 --- a/tests/playTests.sh +++ b/tests/playTests.sh @@ -48,6 +48,12 @@ fileRoundTripTest() { $DIFF -q tmp.md5.1 tmp.md5.2 } +truncateLastByte() { + dd bs=1 count=$(($(wc -c < "$1") - 1)) if="$1" status=none +} + +UNAME=$(uname) + isTerminal=false if [ -t 0 ] && [ -t 1 ] then @@ -56,7 +62,10 @@ fi isWindows=false INTOVOID="/dev/null" -DEVDEVICE="/dev/zero" +case "$UNAME" in + GNU) DEVDEVICE="/dev/random" ;; + *) DEVDEVICE="/dev/zero" ;; +esac case "$OS" in Windows*) isWindows=true @@ -65,10 +74,10 @@ case "$OS" in ;; esac -UNAME=$(uname) case "$UNAME" in Darwin) MD5SUM="md5 -r" ;; FreeBSD) MD5SUM="gmd5sum" ;; + OpenBSD) MD5SUM="md5" ;; *) MD5SUM="md5sum" ;; esac @@ -94,6 +103,7 @@ else fi + $ECHO "\n===> simple tests " ./datagen > tmp @@ -103,10 +113,15 @@ $ECHO "test : basic decompression" $ZSTD -df tmp.zst # trivial decompression case (overwrites tmp) $ECHO "test : too large compression level => auto-fix" $ZSTD -99 -f tmp # too large compression level, automatic sized down +$ZSTD -5000000000 -f tmp && die "too large numeric value : must fail" $ECHO "test : --fast aka negative compression levels" $ZSTD --fast -f tmp # == -1 $ZSTD --fast=3 -f tmp # == -3 -$ZSTD --fast=200000 -f tmp # == no compression +$ZSTD --fast=200000 -f tmp # too low compression level, automatic fixed +$ZSTD --fast=5000000000 -f tmp && die "too large numeric value : must fail" +$ZSTD -c --fast=0 tmp > $INTOVOID && die "--fast must not accept value 0" +$ECHO "test : too large numeric argument" +$ZSTD --fast=9999999999 -f tmp && die "should have refused numeric value" $ECHO "test : compress to stdout" $ZSTD tmp -c > tmpCompressed $ZSTD tmp --stdout > tmpCompressed # long command format @@ -165,6 +180,8 @@ chmod 400 tmpro.zst $ZSTD -q tmpro && die "should have refused to overwrite read-only file" $ZSTD -q -f tmpro rm -f tmpro tmpro.zst + + $ECHO "test : file removal" $ZSTD -f --rm tmp test ! -f tmp # tmp should no longer be present @@ -175,12 +192,23 @@ $ECHO hello > tmp $ZSTD tmp -f -o "$DEVDEVICE" 2>tmplog > "$INTOVOID" grep -v "Refusing to remove non-regular file" tmplog rm -f tmplog -$ZSTD tmp -f -o "$INTONULL" 2>&1 | grep -v "Refusing to remove non-regular file" +$ZSTD tmp -f -o "$INTOVOID" 2>&1 | grep -v "Refusing to remove non-regular file" $ECHO "test : --rm on stdin" $ECHO a | $ZSTD --rm > $INTOVOID # --rm should remain silent rm tmp $ZSTD -f tmp && die "tmp not present : should have failed" test ! -f tmp.zst # tmp.zst should not be created +$ECHO "test : -d -f do not delete destination when source is not present" +touch tmp # create destination file +$ZSTD -d -f tmp.zst && die "attempt to decompress a non existing file" +test -f tmp # destination file should still be present +$ECHO "test : -f do not delete destination when source is not present" +rm tmp # erase source file +touch tmp.zst # create destination file +$ZSTD -f tmp && die "attempt to compress a non existing file" +test -f tmp.zst # destination file should still be present +rm tmp* + $ECHO "test : compress multiple files" $ECHO hello > tmp1 @@ -258,7 +286,7 @@ rm ./*.tmp ./*.zstd $ECHO "frame concatenation tests completed" -if [ "$isWindows" = false ] && [ "$UNAME" != 'SunOS' ] ; then +if [ "$isWindows" = false ] && [ "$UNAME" != 'SunOS' ] && [ "$UNAME" != "OpenBSD" ] ; then $ECHO "\n**** flush write error test **** " $ECHO "$ECHO foo | $ZSTD > /dev/full" @@ -395,28 +423,54 @@ $ECHO "Hello World" > tmp $ZSTD --train-legacy -q tmp && die "Dictionary training should fail : not enough input source" ./datagen -P0 -g10M > tmp $ZSTD --train-legacy -q tmp && die "Dictionary training should fail : source is pure noise" -rm tmp* +$ECHO "- Test -o before --train" +rm -f tmpDict dictionary +$ZSTD -o tmpDict --train *.c ../programs/*.c +test -f tmpDict +$ZSTD --train *.c ../programs/*.c +test -f dictionary +rm tmp* dictionary -$ECHO "\n===> cover dictionary builder : advanced options " +$ECHO "\n===> fastCover dictionary builder : advanced options " TESTFILE=../programs/zstdcli.c ./datagen > tmpDict $ECHO "- Create first dictionary" -$ZSTD --train-cover=k=46,d=8 *.c ../programs/*.c -o tmpDict +$ZSTD --train-fastcover=k=46,d=8,f=15,split=80 *.c ../programs/*.c -o tmpDict cp $TESTFILE tmp $ZSTD -f tmp -D tmpDict $ZSTD -d tmp.zst -D tmpDict -fo result $DIFF $TESTFILE result $ECHO "- Create second (different) dictionary" -$ZSTD --train-cover=k=56,d=8 *.c ../programs/*.c ../programs/*.h -o tmpDictC +$ZSTD --train-fastcover=k=56,d=8 *.c ../programs/*.c ../programs/*.h -o tmpDictC $ZSTD -d tmp.zst -D tmpDictC -fo result && die "wrong dictionary not detected!" $ECHO "- Create dictionary with short dictID" -$ZSTD --train-cover=k=46,d=8 *.c ../programs/*.c --dictID=1 -o tmpDict1 +$ZSTD --train-fastcover=k=46,d=8,f=15,split=80 *.c ../programs/*.c --dictID=1 -o tmpDict1 cmp tmpDict tmpDict1 && die "dictionaries should have different ID !" $ECHO "- Create dictionary with size limit" -$ZSTD --train-cover=steps=8 *.c ../programs/*.c -o tmpDict2 --maxdict=4K -rm tmp* +$ZSTD --train-fastcover=steps=8 *.c ../programs/*.c -o tmpDict2 --maxdict=4K +$ECHO "- Compare size of dictionary from 90% training samples with 80% training samples" +$ZSTD --train-fastcover=split=90 -r *.c ../programs/*.c +$ZSTD --train-fastcover=split=80 -r *.c ../programs/*.c +$ECHO "- Create dictionary using all samples for both training and testing" +$ZSTD --train-fastcover=split=100 -r *.c ../programs/*.c +$ECHO "- Create dictionary using f=16" +$ZSTD --train-fastcover=f=16 -r *.c ../programs/*.c +$ECHO "- Create dictionary using accel=2" +$ZSTD --train-fastcover=accel=2 -r *.c ../programs/*.c +$ECHO "- Create dictionary using accel=10" +$ZSTD --train-fastcover=accel=10 -r *.c ../programs/*.c +$ECHO "- Create dictionary with multithreading" +$ZSTD --train-fastcover -T4 -r *.c ../programs/*.c +$ECHO "- Test -o before --train-fastcover" +rm -f tmpDict dictionary +$ZSTD -o tmpDict --train-fastcover *.c ../programs/*.c +test -f tmpDict +$ZSTD --train-fastcover *.c ../programs/*.c +test -f dictionary +rm tmp* dictionary + $ECHO "\n===> legacy dictionary builder " @@ -436,7 +490,13 @@ $ZSTD --train-legacy -s5 *.c ../programs/*.c --dictID=1 -o tmpDict1 cmp tmpDict tmpDict1 && die "dictionaries should have different ID !" $ECHO "- Create dictionary with size limit" $ZSTD --train-legacy -s9 *.c ../programs/*.c -o tmpDict2 --maxdict=4K -rm tmp* +$ECHO "- Test -o before --train-legacy" +rm -f tmpDict dictionary +$ZSTD -o tmpDict --train-legacy *.c ../programs/*.c +test -f tmpDict +$ZSTD --train-legacy *.c ../programs/*.c +test -f dictionary +rm tmp* dictionary $ECHO "\n===> integrity tests " @@ -482,6 +542,12 @@ $ZSTD -bi0 --fast tmp1 $ECHO "with recursive and quiet modes" $ZSTD -rqi1b1e2 tmp1 +$ECHO "\n===> zstd compatibility tests " + +./datagen > tmp +rm -f tmp.zst +$ZSTD --format=zstd -f tmp +test -f tmp.zst $ECHO "\n===> gzip compatibility tests " @@ -513,12 +579,18 @@ if [ $GZIPMODE -eq 1 ]; then $ZSTD -f --format=gzip tmp $ZSTD -f tmp cat tmp.gz tmp.zst tmp.gz tmp.zst | $ZSTD -d -f -o tmp - head -c -1 tmp.gz | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" + truncateLastByte tmp.gz | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" rm tmp* else $ECHO "gzip mode not supported" fi +if [ $GZIPMODE -eq 1 ]; then + ./datagen > tmp + rm -f tmp.zst + $ZSTD --format=gzip --format=zstd -f tmp + test -f tmp.zst +fi $ECHO "\n===> xz compatibility tests " @@ -527,16 +599,16 @@ $ZSTD --format=xz -V || LZMAMODE=0 if [ $LZMAMODE -eq 1 ]; then $ECHO "xz support detected" XZEXE=1 - xz -V && lzma -V || XZEXE=0 + xz -Q -V && lzma -Q -V || XZEXE=0 if [ $XZEXE -eq 1 ]; then $ECHO "Testing zstd xz and lzma support" ./datagen > tmp $ZSTD --format=lzma -f tmp $ZSTD --format=xz -f tmp - xz -t -v tmp.xz - xz -t -v tmp.lzma - xz -f -k tmp - lzma -f -k --lzma1 tmp + xz -Q -t -v tmp.xz + xz -Q -t -v tmp.lzma + xz -Q -f -k tmp + lzma -Q -f -k --lzma1 tmp $ZSTD -d -f -v tmp.xz $ZSTD -d -f -v tmp.lzma rm tmp* @@ -548,13 +620,13 @@ if [ $LZMAMODE -eq 1 ]; then $ECHO "Testing xz and lzma symlinks" ./datagen > tmp ./xz tmp - xz -d tmp.xz + xz -Q -d tmp.xz ./lzma tmp - lzma -d tmp.lzma + lzma -Q -d tmp.lzma $ECHO "Testing unxz and unlzma symlinks" - xz tmp + xz -Q tmp ./xz -d tmp.xz - lzma tmp + lzma -Q tmp ./lzma -d tmp.lzma rm xz unxz lzma unlzma rm tmp* @@ -574,8 +646,8 @@ if [ $LZMAMODE -eq 1 ]; then $ZSTD -f --format=lzma tmp $ZSTD -f tmp cat tmp.xz tmp.lzma tmp.zst tmp.lzma tmp.xz tmp.zst | $ZSTD -d -f -o tmp - head -c -1 tmp.xz | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" - head -c -1 tmp.lzma | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" + truncateLastByte tmp.xz | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" + truncateLastByte tmp.lzma | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" rm tmp* else $ECHO "xz mode not supported" @@ -611,12 +683,29 @@ if [ $LZ4MODE -eq 1 ]; then $ZSTD -f --format=lz4 tmp $ZSTD -f tmp cat tmp.lz4 tmp.zst tmp.lz4 tmp.zst | $ZSTD -d -f -o tmp - head -c -1 tmp.lz4 | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" + truncateLastByte tmp.lz4 | $ZSTD -t > $INTOVOID && die "incomplete frame not detected !" rm tmp* else $ECHO "lz4 mode not supported" fi +$ECHO "\n===> suffix list test" + +! $ZSTD -d tmp.abc 2> tmplg + +if [ $GZIPMODE -ne 1 ]; then + grep ".gz" tmplg > $INTOVOID && die "Unsupported suffix listed" +fi + +if [ $LZMAMODE -ne 1 ]; then + grep ".lzma" tmplg > $INTOVOID && die "Unsupported suffix listed" + grep ".xz" tmplg > $INTOVOID && die "Unsupported suffix listed" +fi + +if [ $LZ4MODE -ne 1 ]; then + grep ".lz4" tmplg > $INTOVOID && die "Unsupported suffix listed" +fi + $ECHO "\n===> zstd round-trip tests " roundTripTest @@ -650,6 +739,25 @@ then $ECHO "\n===> zstdmt long distance matching round-trip tests " roundTripTest -g8M "3 --long=24 -T2" + + $ECHO "\n===> ovLog tests " + ./datagen -g2MB > tmp + refSize=$($ZSTD tmp -6 -c --zstd=wlog=18 | wc -c) + ov9Size=$($ZSTD tmp -6 -c --zstd=wlog=18,ovlog=9 | wc -c) + ov0Size=$($ZSTD tmp -6 -c --zstd=wlog=18,ovlog=0 | wc -c) + if [ $refSize -eq $ov9Size ]; then + echo ov9Size should be different from refSize + exit 1 + fi + if [ $refSize -eq $ov0Size ]; then + echo ov0Size should be different from refSize + exit 1 + fi + if [ $ov9Size -ge $ov0Size ]; then + echo ov9Size=$ov9Size should be smaller than ov0Size=$ov0Size + exit 1 + fi + else $ECHO "\n===> no multithreading, skipping zstdmt tests " fi @@ -673,23 +781,32 @@ $ZSTD -l *.zst $ZSTD -lv *.zst $ECHO "\n===> zstd --list/-l error detection tests " -! $ZSTD -l tmp1 tmp1.zst -! $ZSTD --list tmp* -! $ZSTD -lv tmp1* -! $ZSTD --list -v tmp2 tmp12.zst +$ZSTD -l tmp1 tmp1.zst && die "-l must fail on non-zstd file" +$ZSTD --list tmp* && die "-l must fail on non-zstd file" +$ZSTD -lv tmp1* && die "-l must fail on non-zstd file" +$ZSTD --list -v tmp2 tmp12.zst && die "-l must fail on non-zstd file" + +$ECHO "\n===> zstd --list/-l errors when presented with stdin / no files" +$ZSTD -l && die "-l must fail on empty list of files" +$ZSTD -l - && die "-l does not work on stdin" +$ZSTD -l < tmp1.zst && die "-l does not work on stdin" +$ZSTD -l - < tmp1.zst && die "-l does not work on stdin" +$ZSTD -l - tmp1.zst && die "-l does not work on stdin" +$ZSTD -l - tmp1.zst < tmp1.zst && die "-l does not work on stdin" +$ZSTD -l tmp1.zst < tmp2.zst # this will check tmp1.zst, but not tmp2.zst, which is not an error : zstd simply doesn't read stdin in this case. It must not error just because stdin is not a tty $ECHO "\n===> zstd --list/-l test with null files " ./datagen -g0 > tmp5 $ZSTD tmp5 $ZSTD -l tmp5.zst -! $ZSTD -l tmp5* +$ZSTD -l tmp5* && die "-l must fail on non-zstd file" $ZSTD -lv tmp5.zst | grep "Decompressed Size: 0.00 KB (0 B)" # check that 0 size is present in header -! $ZSTD -lv tmp5* +$ZSTD -lv tmp5* && die "-l must fail on non-zstd file" $ECHO "\n===> zstd --list/-l test with no content size field " ./datagen -g513K | $ZSTD > tmp6.zst $ZSTD -l tmp6.zst -! $ZSTD -lv tmp6.zst | grep "Decompressed Size:" # must NOT be present in header +$ZSTD -lv tmp6.zst | grep "Decompressed Size:" && die "Field :Decompressed Size: should not be available in this compressed file" $ECHO "\n===> zstd --list/-l test with no checksum " $ZSTD -f --no-check tmp1 @@ -709,11 +826,22 @@ roundTripTest -g1M -P50 "1 --single-thread --long=29" " --long=28 --memory=512MB roundTripTest -g1M -P50 "1 --single-thread --long=29" " --zstd=wlog=28 --memory=512MB" +$ECHO "\n===> adaptive mode " +roundTripTest -g270000000 " --adapt" +roundTripTest -g27000000 " --adapt=min=1,max=4" +$ECHO "===> test: --adapt must fail on incoherent bounds " +./datagen > tmp +$ZSTD -f -vv --adapt=min=10,max=9 tmp && die "--adapt must fail on incoherent bounds" + + if [ "$1" != "--test-large-data" ]; then $ECHO "Skipping large data tests" exit 0 fi + +############################################################################# + $ECHO "\n===> large files tests " roundTripTest -g270000000 1 @@ -768,4 +896,37 @@ else $ECHO "\n**** no multithreading, skipping zstdmt tests **** " fi -rm tmp* + +$ECHO "\n===> cover dictionary builder : advanced options " + +TESTFILE=../programs/zstdcli.c +./datagen > tmpDict +$ECHO "- Create first dictionary" +$ZSTD --train-cover=k=46,d=8,split=80 *.c ../programs/*.c -o tmpDict +cp $TESTFILE tmp +$ZSTD -f tmp -D tmpDict +$ZSTD -d tmp.zst -D tmpDict -fo result +$DIFF $TESTFILE result +$ECHO "- Create second (different) dictionary" +$ZSTD --train-cover=k=56,d=8 *.c ../programs/*.c ../programs/*.h -o tmpDictC +$ZSTD -d tmp.zst -D tmpDictC -fo result && die "wrong dictionary not detected!" +$ECHO "- Create dictionary with short dictID" +$ZSTD --train-cover=k=46,d=8,split=80 *.c ../programs/*.c --dictID=1 -o tmpDict1 +cmp tmpDict tmpDict1 && die "dictionaries should have different ID !" +$ECHO "- Create dictionary with size limit" +$ZSTD --train-cover=steps=8 *.c ../programs/*.c -o tmpDict2 --maxdict=4K +$ECHO "- Compare size of dictionary from 90% training samples with 80% training samples" +$ZSTD --train-cover=split=90 -r *.c ../programs/*.c +$ZSTD --train-cover=split=80 -r *.c ../programs/*.c +$ECHO "- Create dictionary using all samples for both training and testing" +$ZSTD --train-cover=split=100 -r *.c ../programs/*.c +$ECHO "- Test -o before --train-cover" +rm -f tmpDict dictionary +$ZSTD -o tmpDict --train-cover *.c ../programs/*.c +test -f tmpDict +$ZSTD --train-cover *.c ../programs/*.c +test -f dictionary +rm -f tmp* dictionary + + +rm -f tmp* diff --git a/tests/poolTests.c b/tests/poolTests.c index 00ee830154c99..9661b5299e523 100644 --- a/tests/poolTests.c +++ b/tests/poolTests.c @@ -15,11 +15,11 @@ #include <stddef.h> #include <stdio.h> -#define ASSERT_TRUE(p) \ - do { \ - if (!(p)) { \ - return 1; \ - } \ +#define ASSERT_TRUE(p) \ + do { \ + if (!(p)) { \ + return 1; \ + } \ } while (0) #define ASSERT_FALSE(p) ASSERT_TRUE(!(p)) #define ASSERT_EQ(lhs, rhs) ASSERT_TRUE((lhs) == (rhs)) @@ -32,10 +32,10 @@ struct data { void fn(void *opaque) { struct data *data = (struct data *)opaque; - pthread_mutex_lock(&data->mutex); + ZSTD_pthread_mutex_lock(&data->mutex); data->data[data->i] = data->i; ++data->i; - pthread_mutex_unlock(&data->mutex); + ZSTD_pthread_mutex_unlock(&data->mutex); } int testOrder(size_t numThreads, size_t queueSize) { @@ -43,25 +43,26 @@ int testOrder(size_t numThreads, size_t queueSize) { POOL_ctx *ctx = POOL_create(numThreads, queueSize); ASSERT_TRUE(ctx); data.i = 0; - pthread_mutex_init(&data.mutex, NULL); - { - size_t i; + ZSTD_pthread_mutex_init(&data.mutex, NULL); + { size_t i; for (i = 0; i < 16; ++i) { POOL_add(ctx, &fn, &data); } } POOL_free(ctx); ASSERT_EQ(16, data.i); - { - size_t i; + { size_t i; for (i = 0; i < data.i; ++i) { ASSERT_EQ(i, data.data[i]); } } - pthread_mutex_destroy(&data.mutex); + ZSTD_pthread_mutex_destroy(&data.mutex); return 0; } + +/* --- test deadlocks --- */ + void waitFn(void *opaque) { (void)opaque; UTIL_sleepMilli(1); @@ -72,8 +73,7 @@ int testWait(size_t numThreads, size_t queueSize) { struct data data; POOL_ctx *ctx = POOL_create(numThreads, queueSize); ASSERT_TRUE(ctx); - { - size_t i; + { size_t i; for (i = 0; i < 16; ++i) { POOL_add(ctx, &waitFn, &data); } @@ -82,25 +82,178 @@ int testWait(size_t numThreads, size_t queueSize) { return 0; } + +/* --- test POOL_resize() --- */ + +typedef struct { + ZSTD_pthread_mutex_t mut; + int val; + int max; + ZSTD_pthread_cond_t cond; +} poolTest_t; + +void waitLongFn(void *opaque) { + poolTest_t* test = (poolTest_t*) opaque; + UTIL_sleepMilli(10); + ZSTD_pthread_mutex_lock(&test->mut); + test->val = test->val + 1; + if (test->val == test->max) + ZSTD_pthread_cond_signal(&test->cond); + ZSTD_pthread_mutex_unlock(&test->mut); +} + +static int testThreadReduction_internal(POOL_ctx* ctx, poolTest_t test) +{ + int const nbWaits = 16; + UTIL_time_t startTime; + U64 time4threads, time2threads; + + test.val = 0; + test.max = nbWaits; + + startTime = UTIL_getTime(); + { int i; + for (i=0; i<nbWaits; i++) + POOL_add(ctx, &waitLongFn, &test); + } + ZSTD_pthread_mutex_lock(&test.mut); + ZSTD_pthread_cond_wait(&test.cond, &test.mut); + ASSERT_EQ(test.val, nbWaits); + ZSTD_pthread_mutex_unlock(&test.mut); + time4threads = UTIL_clockSpanNano(startTime); + + ASSERT_EQ( POOL_resize(ctx, 2/*nbThreads*/) , 0 ); + test.val = 0; + startTime = UTIL_getTime(); + { int i; + for (i=0; i<nbWaits; i++) + POOL_add(ctx, &waitLongFn, &test); + } + ZSTD_pthread_mutex_lock(&test.mut); + ZSTD_pthread_cond_wait(&test.cond, &test.mut); + ASSERT_EQ(test.val, nbWaits); + ZSTD_pthread_mutex_unlock(&test.mut); + time2threads = UTIL_clockSpanNano(startTime); + + if (time4threads >= time2threads) return 1; /* check 4 threads were effectively faster than 2 */ + return 0; +} + +static int testThreadReduction(void) { + int result; + poolTest_t test; + POOL_ctx* const ctx = POOL_create(4 /*nbThreads*/, 2 /*queueSize*/); + + ASSERT_TRUE(ctx); + + memset(&test, 0, sizeof(test)); + ASSERT_FALSE( ZSTD_pthread_mutex_init(&test.mut, NULL) ); + ASSERT_FALSE( ZSTD_pthread_cond_init(&test.cond, NULL) ); + + result = testThreadReduction_internal(ctx, test); + + ZSTD_pthread_mutex_destroy(&test.mut); + ZSTD_pthread_cond_destroy(&test.cond); + POOL_free(ctx); + + return result; +} + + +/* --- test abrupt ending --- */ + +typedef struct { + ZSTD_pthread_mutex_t mut; + int val; +} abruptEndCanary_t; + +void waitIncFn(void *opaque) { + abruptEndCanary_t* test = (abruptEndCanary_t*) opaque; + UTIL_sleepMilli(10); + ZSTD_pthread_mutex_lock(&test->mut); + test->val = test->val + 1; + ZSTD_pthread_mutex_unlock(&test->mut); +} + +static int testAbruptEnding_internal(abruptEndCanary_t test) +{ + int const nbWaits = 16; + + POOL_ctx* const ctx = POOL_create(3 /*numThreads*/, nbWaits /*queueSize*/); + ASSERT_TRUE(ctx); + test.val = 0; + + { int i; + for (i=0; i<nbWaits; i++) + POOL_add(ctx, &waitIncFn, &test); /* all jobs pushed into queue */ + } + ASSERT_EQ( POOL_resize(ctx, 1 /*numThreads*/) , 0 ); /* downsize numThreads, to try to break end condition */ + + POOL_free(ctx); /* must finish all jobs in queue before giving back control */ + ASSERT_EQ(test.val, nbWaits); + return 0; +} + +static int testAbruptEnding(void) { + int result; + abruptEndCanary_t test; + + memset(&test, 0, sizeof(test)); + ASSERT_FALSE( ZSTD_pthread_mutex_init(&test.mut, NULL) ); + + result = testAbruptEnding_internal(test); + + ZSTD_pthread_mutex_destroy(&test.mut); + return result; +} + + + +/* --- test launcher --- */ + int main(int argc, const char **argv) { size_t numThreads; + (void)argc; + (void)argv; + + if (POOL_create(0, 1)) { /* should not be possible */ + printf("FAIL: should not create POOL with 0 threads\n"); + return 1; + } + for (numThreads = 1; numThreads <= 4; ++numThreads) { size_t queueSize; for (queueSize = 0; queueSize <= 2; ++queueSize) { + printf("queueSize==%u, numThreads=%u \n", + (unsigned)queueSize, (unsigned)numThreads); if (testOrder(numThreads, queueSize)) { printf("FAIL: testOrder\n"); return 1; } + printf("SUCCESS: testOrder\n"); if (testWait(numThreads, queueSize)) { printf("FAIL: testWait\n"); return 1; } + printf("SUCCESS: testWait\n"); } } - printf("PASS: testOrder\n"); - (void)argc; - (void)argv; - return (POOL_create(0, 1)) ? printf("FAIL: testInvalid\n"), 1 - : printf("PASS: testInvalid\n"), 0; + + if (testThreadReduction()) { + printf("FAIL: thread reduction not effective \n"); + return 1; + } else { + printf("SUCCESS: thread reduction effective (slower execution) \n"); + } + + if (testAbruptEnding()) { + printf("FAIL: jobs in queue not completed on early end \n"); + return 1; + } else { + printf("SUCCESS: all jobs in queue completed on early end \n"); + } + + printf("PASS: all POOL tests\n"); + return 0; } diff --git a/tests/rateLimiter.py b/tests/rateLimiter.py new file mode 100755 index 0000000000000..da0baf01464ff --- /dev/null +++ b/tests/rateLimiter.py @@ -0,0 +1,40 @@ +#!/usr/bin/env python3 + +# ################################################################ +# Copyright (c) 2018-present, 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). +# ########################################################################## + +# Rate limiter, replacement for pv +# this rate limiter does not "catch up" after a blocking period +# Limitations: +# - only accepts limit speed in MB/s + +import sys +import time + +MB = 1024 * 1024 +rate = float(sys.argv[1]) * MB +start = time.time() +total_read = 0 + +# sys.stderr.close() # remove error message, for Ctrl+C + +try: + buf = " " + while len(buf): + now = time.time() + to_read = max(int(rate * (now - start)), 1) + max_buf_size = 1 * MB + to_read = min(to_read, max_buf_size) + start = now + + buf = sys.stdin.buffer.read(to_read) + sys.stdout.buffer.write(buf) + +except (KeyboardInterrupt, BrokenPipeError) as e: + pass diff --git a/tests/roundTripCrash.c b/tests/roundTripCrash.c index 7d937fceebc01..90afcd4b2a8be 100644 --- a/tests/roundTripCrash.c +++ b/tests/roundTripCrash.c @@ -212,7 +212,7 @@ static void loadFile(void* buffer, const char* fileName, size_t fileSize) static void fileCheck(const char* fileName, int testCCtxParams) { size_t const fileSize = getFileSize(fileName); - void* buffer = malloc(fileSize); + void* const buffer = malloc(fileSize + !fileSize /* avoid 0 */); if (!buffer) { fprintf(stderr, "not enough memory \n"); exit(4); diff --git a/tests/symbols.c b/tests/symbols.c index c0bed2e5d96e9..b370821314647 100644 --- a/tests/symbols.c +++ b/tests/symbols.c @@ -144,6 +144,8 @@ static const void *symbols[] = { /* zdict.h: advanced functions */ &ZDICT_trainFromBuffer_cover, &ZDICT_optimizeTrainFromBuffer_cover, + &ZDICT_trainFromBuffer_fastCover, + &ZDICT_optimizeTrainFromBuffer_fastCover, &ZDICT_finalizeDictionary, &ZDICT_trainFromBuffer_legacy, &ZDICT_addEntropyTablesFromBuffer, diff --git a/tests/test-zstd-versions.py b/tests/test-zstd-versions.py index f2deac1f28da0..8e88b869b0d28 100755 --- a/tests/test-zstd-versions.py +++ b/tests/test-zstd-versions.py @@ -213,7 +213,7 @@ if __name__ == '__main__': print('Retrieve all release tags :') os.chdir(clone_dir) alltags = get_git_tags() + [head] - tags = [t for t in alltags if t >= 'v0.4.0'] + tags = [t for t in alltags if t >= 'v0.5.0'] print(tags) # Build all release zstd diff --git a/tests/zstreamtest.c b/tests/zstreamtest.c index b94f282f5802c..f47451a3c3d50 100644 --- a/tests/zstreamtest.c +++ b/tests/zstreamtest.c @@ -10,8 +10,8 @@ /*-************************************ -* Compiler specific -**************************************/ + * Compiler specific + **************************************/ #ifdef _MSC_VER /* Visual Studio */ # define _CRT_SECURE_NO_WARNINGS /* fgets */ # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ @@ -20,8 +20,8 @@ /*-************************************ -* Includes -**************************************/ + * Includes + **************************************/ #include <stdlib.h> /* free */ #include <stdio.h> /* fgets, sscanf */ #include <string.h> /* strcmp */ @@ -40,8 +40,8 @@ /*-************************************ -* Constants -**************************************/ + * Constants + **************************************/ #define KB *(1U<<10) #define MB *(1U<<20) #define GB *(1U<<30) @@ -54,8 +54,8 @@ static const U32 prime32 = 2654435761U; /*-************************************ -* Display Macros -**************************************/ + * Display Macros + **************************************/ #define DISPLAY(...) fprintf(stderr, __VA_ARGS__) #define DISPLAYLEVEL(l, ...) if (g_displayLevel>=l) { \ DISPLAY(__VA_ARGS__); \ @@ -74,8 +74,8 @@ static U64 g_clockTime = 0; /*-******************************************************* -* Fuzzer functions -*********************************************************/ + * Check macros + *********************************************************/ #undef MIN #undef MAX #define MIN(a,b) ((a)<(b)?(a):(b)) @@ -84,7 +84,7 @@ static U64 g_clockTime = 0; @return : a 27 bits random value, from a 32-bits `seed`. `seed` is also modified */ #define FUZ_rotl32(x,r) ((x << r) | (x >> (32 - r))) -unsigned int FUZ_rand(unsigned int* seedPtr) +static unsigned int FUZ_rand(unsigned int* seedPtr) { static const U32 prime2 = 2246822519U; U32 rand32 = *seedPtr; @@ -110,10 +110,24 @@ unsigned int FUZ_rand(unsigned int* seedPtr) #f, ZSTD_getErrorName(err)); \ } +#define CHECK_RET(ret, cond, ...) { \ + if (cond) { \ + DISPLAY("Error %llu => ", (unsigned long long)ret); \ + DISPLAY(__VA_ARGS__); \ + DISPLAY(" (line %u)\n", __LINE__); \ + return ret; \ +} } + +#define CHECK_RET_Z(f) { \ + size_t const err = f; \ + CHECK_RET(err, ZSTD_isError(err), "%s : %s ", \ + #f, ZSTD_getErrorName(err)); \ +} + /*====================================================== -* Basic Unit tests -======================================================*/ + * Basic Unit tests + *======================================================*/ typedef struct { void* start; @@ -121,34 +135,34 @@ typedef struct { size_t filled; } buffer_t; -static const buffer_t g_nullBuffer = { NULL, 0 , 0 }; +static const buffer_t kBuffNull = { NULL, 0 , 0 }; + +static void FUZ_freeDictionary(buffer_t dict) +{ + free(dict.start); +} static buffer_t FUZ_createDictionary(const void* src, size_t srcSize, size_t blockSize, size_t requestedDictSize) { - buffer_t dict = { NULL, 0, 0 }; + buffer_t dict = kBuffNull; size_t const nbBlocks = (srcSize + (blockSize-1)) / blockSize; - size_t* const blockSizes = (size_t*) malloc(nbBlocks * sizeof(size_t)); - if (!blockSizes) return dict; + size_t* const blockSizes = (size_t*)malloc(nbBlocks * sizeof(size_t)); + if (!blockSizes) return kBuffNull; dict.start = malloc(requestedDictSize); - if (!dict.start) { free(blockSizes); return dict; } + if (!dict.start) { free(blockSizes); return kBuffNull; } { size_t nb; for (nb=0; nb<nbBlocks-1; nb++) blockSizes[nb] = blockSize; blockSizes[nbBlocks-1] = srcSize - (blockSize * (nbBlocks-1)); } { size_t const dictSize = ZDICT_trainFromBuffer(dict.start, requestedDictSize, src, blockSizes, (unsigned)nbBlocks); free(blockSizes); - if (ZDICT_isError(dictSize)) { free(dict.start); return g_nullBuffer; } + if (ZDICT_isError(dictSize)) { FUZ_freeDictionary(dict); return kBuffNull; } dict.size = requestedDictSize; dict.filled = dictSize; - return dict; /* how to return dictSize ? */ + return dict; } } -static void FUZ_freeDictionary(buffer_t dict) -{ - free(dict.start); -} - /* Round trips data and updates xxh with the decompressed data produced */ static size_t SEQ_roundTrip(ZSTD_CCtx* cctx, ZSTD_DCtx* dctx, XXH64_state_t* xxh, void* data, size_t size, @@ -207,6 +221,42 @@ static size_t SEQ_generateRoundTrip(ZSTD_CCtx* cctx, ZSTD_DCtx* dctx, return 0; } +static size_t getCCtxParams(ZSTD_CCtx* zc, ZSTD_parameters* savedParams) +{ + unsigned value; + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_windowLog, &savedParams->cParams.windowLog)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_hashLog, &savedParams->cParams.hashLog)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_chainLog, &savedParams->cParams.chainLog)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_searchLog, &savedParams->cParams.searchLog)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_minMatch, &savedParams->cParams.searchLength)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_targetLength, &savedParams->cParams.targetLength)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_compressionStrategy, &value)); + savedParams->cParams.strategy = value; + + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_checksumFlag, &savedParams->fParams.checksumFlag)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_contentSizeFlag, &savedParams->fParams.contentSizeFlag)); + CHECK_RET_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_dictIDFlag, &value)); + savedParams->fParams.noDictIDFlag = !value; + return 0; +} + +static U32 badParameters(ZSTD_CCtx* zc, ZSTD_parameters const savedParams) +{ + ZSTD_parameters params; + if (ZSTD_isError(getCCtxParams(zc, ¶ms))) return 10; + CHECK_RET(1, params.cParams.windowLog != savedParams.cParams.windowLog, "windowLog"); + CHECK_RET(2, params.cParams.hashLog != savedParams.cParams.hashLog, "hashLog"); + CHECK_RET(3, params.cParams.chainLog != savedParams.cParams.chainLog, "chainLog"); + CHECK_RET(4, params.cParams.searchLog != savedParams.cParams.searchLog, "searchLog"); + CHECK_RET(5, params.cParams.searchLength != savedParams.cParams.searchLength, "searchLength"); + CHECK_RET(6, params.cParams.targetLength != savedParams.cParams.targetLength, "targetLength"); + + CHECK_RET(7, params.fParams.checksumFlag != savedParams.fParams.checksumFlag, "checksumFlag"); + CHECK_RET(8, params.fParams.contentSizeFlag != savedParams.fParams.contentSizeFlag, "contentSizeFlag"); + CHECK_RET(9, params.fParams.noDictIDFlag != savedParams.fParams.noDictIDFlag, "noDictIDFlag"); + return 0; +} + static int basicUnitTests(U32 seed, double compressibility) { size_t const CNBufferSize = COMPRESSIBLE_NOISE_LENGTH; @@ -226,7 +276,7 @@ static int basicUnitTests(U32 seed, double compressibility) ZSTD_inBuffer inBuff, inBuff2; ZSTD_outBuffer outBuff; - buffer_t dictionary = g_nullBuffer; + buffer_t dictionary = kBuffNull; size_t const dictSize = 128 KB; unsigned dictID = 0; @@ -383,11 +433,12 @@ static int basicUnitTests(U32 seed, double compressibility) inBuff.pos = 0; outBuff.pos = 0; while (r) { /* skippable frame */ - size_t const inSize = FUZ_rand(&coreSeed) & 15; - size_t const outSize = FUZ_rand(&coreSeed) & 15; + size_t const inSize = (FUZ_rand(&coreSeed) & 15) + 1; + size_t const outSize = (FUZ_rand(&coreSeed) & 15) + 1; inBuff.size = inBuff.pos + inSize; outBuff.size = outBuff.pos + outSize; r = ZSTD_decompressStream(zd, &outBuff, &inBuff); + if (ZSTD_isError(r)) DISPLAYLEVEL(4, "ZSTD_decompressStream on skippable frame error : %s \n", ZSTD_getErrorName(r)); if (ZSTD_isError(r)) goto _output_error; } /* normal frame */ @@ -395,14 +446,17 @@ static int basicUnitTests(U32 seed, double compressibility) r=1; while (r) { size_t const inSize = FUZ_rand(&coreSeed) & 15; - size_t const outSize = FUZ_rand(&coreSeed) & 15; + size_t const outSize = (FUZ_rand(&coreSeed) & 15) + (!inSize); /* avoid having both sizes at 0 => would trigger a no_forward_progress error */ inBuff.size = inBuff.pos + inSize; outBuff.size = outBuff.pos + outSize; r = ZSTD_decompressStream(zd, &outBuff, &inBuff); + if (ZSTD_isError(r)) DISPLAYLEVEL(4, "ZSTD_decompressStream error : %s \n", ZSTD_getErrorName(r)); if (ZSTD_isError(r)) goto _output_error; } } + if (outBuff.pos != CNBufferSize) DISPLAYLEVEL(4, "outBuff.pos != CNBufferSize : should have regenerated same amount ! \n"); if (outBuff.pos != CNBufferSize) goto _output_error; /* should regenerate the same amount */ + if (inBuff.pos != cSize) DISPLAYLEVEL(4, "inBuff.pos != cSize : should have real all input ! \n"); if (inBuff.pos != cSize) goto _output_error; /* should have read the entire frame */ DISPLAYLEVEL(3, "OK \n"); @@ -414,6 +468,30 @@ static int basicUnitTests(U32 seed, double compressibility) } } DISPLAYLEVEL(3, "OK \n"); + /* Decompression forward progress */ + DISPLAYLEVEL(3, "test%3i : generate error when ZSTD_decompressStream() doesn't progress : ", testNb++); + { /* skippable frame */ + size_t r = 0; + int decNb = 0; + int const maxDec = 100; + inBuff.src = compressedBuffer; + inBuff.size = cSize; + inBuff.pos = 0; + + outBuff.dst = decodedBuffer; + outBuff.pos = 0; + outBuff.size = CNBufferSize-1; /* 1 byte missing */ + + for (decNb=0; decNb<maxDec; decNb++) { + if (r==0) ZSTD_initDStream_usingDict(zd, CNBuffer, dictSize); + r = ZSTD_decompressStream(zd, &outBuff, &inBuff); + if (ZSTD_isError(r)) break; + } + if (!ZSTD_isError(r)) DISPLAYLEVEL(4, "ZSTD_decompressStream should have triggered a no_forward_progress error \n"); + if (!ZSTD_isError(r)) goto _output_error; /* should have triggered no_forward_progress error */ + } + DISPLAYLEVEL(3, "OK \n"); + /* _srcSize compression test */ DISPLAYLEVEL(3, "test%3i : compress_srcSize %u bytes : ", testNb++, COMPRESSIBLE_NOISE_LENGTH); ZSTD_initCStream_srcSize(zc, 1, CNBufferSize); @@ -460,6 +538,21 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(3, "OK (error detected : %s) \n", ZSTD_getErrorName(r)); } + DISPLAYLEVEL(3, "test%3i : wrong srcSize !contentSizeFlag : %u bytes : ", testNb++, COMPRESSIBLE_NOISE_LENGTH-1); + { ZSTD_parameters params = ZSTD_getParams(1, CNBufferSize, 0); + params.fParams.contentSizeFlag = 0; + CHECK_Z(ZSTD_initCStream_advanced(zc, NULL, 0, params, CNBufferSize - MIN(CNBufferSize, 200 KB))); + outBuff.dst = (char*)compressedBuffer; + outBuff.size = compressedBufferSize; + outBuff.pos = 0; + inBuff.src = CNBuffer; + inBuff.size = CNBufferSize; + inBuff.pos = 0; + { size_t const r = ZSTD_compressStream(zc, &outBuff, &inBuff); + if (ZSTD_getErrorCode(r) != ZSTD_error_srcSize_wrong) goto _output_error; /* must fail : wrong srcSize */ + DISPLAYLEVEL(3, "OK (error detected : %s) \n", ZSTD_getErrorName(r)); + } } + /* Complex context re-use scenario */ DISPLAYLEVEL(3, "test%3i : context re-use : ", testNb++); ZSTD_freeCStream(zc); @@ -507,7 +600,6 @@ static int basicUnitTests(U32 seed, double compressibility) size_t const initError = ZSTD_initCStream_usingCDict(zc, cdict); DISPLAYLEVEL(5, "ZSTD_initCStream_usingCDict result : %u ", (U32)initError); if (ZSTD_isError(initError)) goto _output_error; - cSize = 0; outBuff.dst = compressedBuffer; outBuff.size = compressedBufferSize; outBuff.pos = 0; @@ -591,6 +683,8 @@ static int basicUnitTests(U32 seed, double compressibility) for (size = 512; size <= maxSize; size <<= 1) { U64 const crcOrig = XXH64(CNBuffer, size, 0); ZSTD_CCtx* const cctx = ZSTD_createCCtx(); + ZSTD_parameters savedParams; + getCCtxParams(cctx, &savedParams); outBuff.dst = compressedBuffer; outBuff.size = compressedBufferSize; outBuff.pos = 0; @@ -599,6 +693,7 @@ static int basicUnitTests(U32 seed, double compressibility) inBuff.pos = 0; CHECK_Z(ZSTD_CCtx_refCDict(cctx, cdict)); CHECK_Z(ZSTD_compress_generic(cctx, &outBuff, &inBuff, ZSTD_e_end)); + CHECK(badParameters(cctx, savedParams), "Bad CCtx params"); if (inBuff.pos != inBuff.size) goto _output_error; { ZSTD_outBuffer decOut = {decodedBuffer, size, 0}; ZSTD_inBuffer decIn = {outBuff.dst, outBuff.pos, 0}; @@ -622,7 +717,6 @@ static int basicUnitTests(U32 seed, double compressibility) ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dictionary.start, dictionary.filled, ZSTD_dlm_byRef, ZSTD_dct_auto, cParams, ZSTD_defaultCMem); size_t const initError = ZSTD_initCStream_usingCDict_advanced(zc, cdict, fParams, CNBufferSize); if (ZSTD_isError(initError)) goto _output_error; - cSize = 0; outBuff.dst = compressedBuffer; outBuff.size = compressedBufferSize; outBuff.pos = 0; @@ -748,7 +842,12 @@ static int basicUnitTests(U32 seed, double compressibility) /* Basic multithreading compression test */ DISPLAYLEVEL(3, "test%3i : compress %u bytes with multiple threads : ", testNb++, COMPRESSIBLE_NOISE_LENGTH); { ZSTD_parameters const params = ZSTD_getParams(1, 0, 0); + unsigned jobSize; + CHECK_Z( ZSTDMT_getMTCtxParameter(mtctx, ZSTDMT_p_jobSize, &jobSize)); + CHECK(jobSize != 0, "job size non-zero"); CHECK_Z( ZSTDMT_initCStream_advanced(mtctx, CNBuffer, dictSize, params, CNBufferSize) ); + CHECK_Z( ZSTDMT_getMTCtxParameter(mtctx, ZSTDMT_p_jobSize, &jobSize)); + CHECK(jobSize != 0, "job size non-zero"); } outBuff.dst = compressedBuffer; outBuff.size = compressedBufferSize; @@ -868,6 +967,26 @@ static int basicUnitTests(U32 seed, double compressibility) } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : ZSTD_initCStream_srcSize sets requestedParams : ", testNb++); + { unsigned level; + CHECK_Z(ZSTD_initCStream_srcSize(zc, 11, ZSTD_CONTENTSIZE_UNKNOWN)); + CHECK_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_compressionLevel, &level)); + CHECK(level != 11, "Compression level does not match"); + ZSTD_resetCStream(zc, ZSTD_CONTENTSIZE_UNKNOWN); + CHECK_Z(ZSTD_CCtx_getParameter(zc, ZSTD_p_compressionLevel, &level)); + CHECK(level != 11, "Compression level does not match"); + } + DISPLAYLEVEL(3, "OK \n"); + + DISPLAYLEVEL(3, "test%3i : ZSTD_initCStream_advanced sets requestedParams : ", testNb++); + { ZSTD_parameters const params = ZSTD_getParams(9, 0, 0); + CHECK_Z(ZSTD_initCStream_advanced(zc, NULL, 0, params, ZSTD_CONTENTSIZE_UNKNOWN)); + CHECK(badParameters(zc, params), "Compression parameters do not match"); + ZSTD_resetCStream(zc, ZSTD_CONTENTSIZE_UNKNOWN); + CHECK(badParameters(zc, params), "Compression parameters do not match"); + } + DISPLAYLEVEL(3, "OK \n"); + /* Overlen overwriting window data bug */ DISPLAYLEVEL(3, "test%3i : wildcopy doesn't overwrite potential match data : ", testNb++); { /* This test has a window size of 1024 bytes and consists of 3 blocks: @@ -901,6 +1020,97 @@ static int basicUnitTests(U32 seed, double compressibility) } DISPLAYLEVEL(3, "OK \n"); + DISPLAYLEVEL(3, "test%3i : dictionary + uncompressible block + reusing tables checks offset table validity: ", testNb++); + { ZSTD_CDict* const cdict = ZSTD_createCDict_advanced( + dictionary.start, dictionary.filled, + ZSTD_dlm_byRef, ZSTD_dct_fullDict, + ZSTD_getCParams(3, 0, dictionary.filled), + ZSTD_defaultCMem); + const size_t inbufsize = 2 * 128 * 1024; /* 2 blocks */ + const size_t outbufsize = ZSTD_compressBound(inbufsize); + size_t inbufpos = 0; + size_t cursegmentlen; + BYTE *inbuf = (BYTE *)malloc(inbufsize); + BYTE *outbuf = (BYTE *)malloc(outbufsize); + BYTE *checkbuf = (BYTE *)malloc(inbufsize); + size_t ret; + + CHECK(cdict == NULL, "failed to alloc cdict"); + CHECK(inbuf == NULL, "failed to alloc input buffer"); + + /* first block is uncompressible */ + cursegmentlen = 128 * 1024; + RDG_genBuffer(inbuf + inbufpos, cursegmentlen, 0., 0., seed); + inbufpos += cursegmentlen; + + /* second block is compressible */ + cursegmentlen = 128 * 1024 - 256; + RDG_genBuffer(inbuf + inbufpos, cursegmentlen, 0.05, 0., seed); + inbufpos += cursegmentlen; + + /* and includes a very long backref */ + cursegmentlen = 128; + memcpy(inbuf + inbufpos, dictionary.start + 256, cursegmentlen); + inbufpos += cursegmentlen; + + /* and includes a very long backref */ + cursegmentlen = 128; + memcpy(inbuf + inbufpos, dictionary.start + 128, cursegmentlen); + inbufpos += cursegmentlen; + + ret = ZSTD_compress_usingCDict(zc, outbuf, outbufsize, inbuf, inbufpos, cdict); + CHECK_Z(ret); + + ret = ZSTD_decompress_usingDict(zd, checkbuf, inbufsize, outbuf, ret, dictionary.start, dictionary.filled); + CHECK_Z(ret); + + CHECK(memcmp(inbuf, checkbuf, inbufpos), "start and finish buffers don't match"); + + ZSTD_freeCDict(cdict); + free(inbuf); + free(outbuf); + free(checkbuf); + } + DISPLAYLEVEL(3, "OK \n"); + + DISPLAYLEVEL(3, "test%3i : dictionary + small blocks + reusing tables checks offset table validity: ", testNb++); + { ZSTD_CDict* const cdict = ZSTD_createCDict_advanced( + dictionary.start, dictionary.filled, + ZSTD_dlm_byRef, ZSTD_dct_fullDict, + ZSTD_getCParams(3, 0, dictionary.filled), + ZSTD_defaultCMem); + ZSTD_outBuffer out = {compressedBuffer, compressedBufferSize, 0}; + int remainingInput = 256 * 1024; + int offset; + + ZSTD_CCtx_reset(zc); + CHECK_Z(ZSTD_CCtx_resetParameters(zc)); + CHECK_Z(ZSTD_CCtx_refCDict(zc, cdict)); + CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_p_checksumFlag, 1)); + /* Write a bunch of 6 byte blocks */ + while (remainingInput > 0) { + char testBuffer[6] = "\xAA\xAA\xAA\xAA\xAA\xAA"; + const size_t kSmallBlockSize = sizeof(testBuffer); + ZSTD_inBuffer in = {testBuffer, kSmallBlockSize, 0}; + + CHECK_Z(ZSTD_compress_generic(zc, &out, &in, ZSTD_e_flush)); + CHECK(in.pos != in.size, "input not fully consumed"); + remainingInput -= kSmallBlockSize; + } + /* Write several very long offset matches into the dictionary */ + for (offset = 1024; offset >= 0; offset -= 128) { + ZSTD_inBuffer in = {dictionary.start + offset, 128, 0}; + ZSTD_EndDirective flush = offset > 0 ? ZSTD_e_continue : ZSTD_e_end; + CHECK_Z(ZSTD_compress_generic(zc, &out, &in, flush)); + CHECK(in.pos != in.size, "input not fully consumed"); + } + /* Ensure decompression works */ + CHECK_Z(ZSTD_decompress_usingDict(zd, decodedBuffer, CNBufferSize, out.dst, out.pos, dictionary.start, dictionary.filled)); + + ZSTD_freeCDict(cdict); + } + DISPLAYLEVEL(3, "OK \n"); + _end: FUZ_freeDictionary(dictionary); ZSTD_freeCStream(zc); @@ -1216,8 +1426,9 @@ _output_error: } -/* Multi-threading version of fuzzer Tests */ -static int fuzzerTests_MT(U32 seed, U32 nbTests, unsigned startTest, double compressibility, int bigTests) +/* fuzzing ZSTDMT_* interface */ +static int fuzzerTests_MT(U32 seed, U32 nbTests, unsigned startTest, + double compressibility, int bigTests) { const U32 maxSrcLog = bigTests ? 24 : 22; static const U32 maxSampleLog = 19; @@ -1491,7 +1702,7 @@ _output_error: * Otherwise, sets the param in zc. */ static size_t setCCtxParameter(ZSTD_CCtx* zc, ZSTD_CCtx_params* cctxParams, ZSTD_cParameter param, unsigned value, - U32 useOpaqueAPI) + int useOpaqueAPI) { if (useOpaqueAPI) { return ZSTD_CCtxParam_setParameter(cctxParams, param, value); @@ -1501,7 +1712,8 @@ static size_t setCCtxParameter(ZSTD_CCtx* zc, ZSTD_CCtx_params* cctxParams, } /* Tests for ZSTD_compress_generic() API */ -static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double compressibility, int bigTests, U32 const useOpaqueAPI) +static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, + double compressibility, int bigTests) { U32 const maxSrcLog = bigTests ? 24 : 22; static const U32 maxSampleLog = 19; @@ -1554,12 +1766,14 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double /* test loop */ for ( ; (testNb <= nbTests) || (UTIL_clockSpanMicro(startClock) < g_clockTime) ; testNb++ ) { U32 lseed; + int opaqueAPI; const BYTE* srcBuffer; size_t totalTestSize, totalGenSize, cSize; XXH64_state_t xxhState; U64 crcOrig; U32 resetAllowed = 1; size_t maxTestSize; + ZSTD_parameters savedParams; /* init */ if (nbTests >= testNb) { DISPLAYUPDATE(2, "\r%6u/%6u ", testNb, nbTests); } @@ -1567,6 +1781,7 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double FUZ_rand(&coreSeed); lseed = coreSeed ^ prime32; DISPLAYLEVEL(5, " *** Test %u *** \n", testNb); + opaqueAPI = FUZ_rand(&lseed) & 1; /* states full reset (deliberately not synchronized) */ /* some issues can only happen when reusing states */ @@ -1574,13 +1789,13 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double DISPLAYLEVEL(5, "Creating new context \n"); ZSTD_freeCCtx(zc); zc = ZSTD_createCCtx(); - CHECK(zc==NULL, "ZSTD_createCCtx allocation error"); - resetAllowed=0; + CHECK(zc == NULL, "ZSTD_createCCtx allocation error"); + resetAllowed = 0; } if ((FUZ_rand(&lseed) & 0xFF) == 132) { ZSTD_freeDStream(zd); zd = ZSTD_createDStream(); - CHECK(zd==NULL, "ZSTD_createDStream allocation error"); + CHECK(zd == NULL, "ZSTD_createDStream allocation error"); ZSTD_initDStream_usingDict(zd, NULL, 0); /* ensure at least one init */ } @@ -1602,11 +1817,14 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double /* compression init */ CHECK_Z( ZSTD_CCtx_loadDictionary(zc, NULL, 0) ); /* cancel previous dict /*/ if ((FUZ_rand(&lseed)&1) /* at beginning, to keep same nb of rand */ - && oldTestLog /* at least one test happened */ && resetAllowed) { + && oldTestLog /* at least one test happened */ + && resetAllowed) { + /* just set a compression level */ maxTestSize = FUZ_randomLength(&lseed, oldTestLog+2); if (maxTestSize >= srcBufferSize) maxTestSize = srcBufferSize-1; { int const compressionLevel = (FUZ_rand(&lseed) % 5) + 1; - CHECK_Z (setCCtxParameter(zc, cctxParams, ZSTD_p_compressionLevel, compressionLevel, useOpaqueAPI) ); + DISPLAYLEVEL(5, "t%u : compression level : %i \n", testNb, compressionLevel); + CHECK_Z (setCCtxParameter(zc, cctxParams, ZSTD_p_compressionLevel, compressionLevel, opaqueAPI) ); } } else { U32 const testLog = FUZ_rand(&lseed) % maxSrcLog; @@ -1628,7 +1846,10 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double } { U64 const pledgedSrcSize = (FUZ_rand(&lseed) & 3) ? ZSTD_CONTENTSIZE_UNKNOWN : maxTestSize; ZSTD_compressionParameters cParams = ZSTD_getCParams(cLevel, pledgedSrcSize, dictSize); - static const U32 windowLogMax = 24; + const U32 windowLogMax = bigTests ? 24 : 20; + const U32 searchLogMax = bigTests ? 15 : 13; + if (dictSize) + DISPLAYLEVEL(5, "t%u: with dictionary of size : %zu \n", testNb, dictSize); /* mess with compression parameters */ cParams.windowLog += (FUZ_rand(&lseed) & 3) - 1; @@ -1636,68 +1857,70 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double cParams.hashLog += (FUZ_rand(&lseed) & 3) - 1; cParams.chainLog += (FUZ_rand(&lseed) & 3) - 1; cParams.searchLog += (FUZ_rand(&lseed) & 3) - 1; + cParams.searchLog = MIN(searchLogMax, cParams.searchLog); cParams.searchLength += (FUZ_rand(&lseed) & 3) - 1; cParams.targetLength = (U32)((cParams.targetLength + 1 ) * (0.5 + ((double)(FUZ_rand(&lseed) & 127) / 128))); - cParams = ZSTD_adjustCParams(cParams, 0, 0); + cParams = ZSTD_adjustCParams(cParams, pledgedSrcSize, dictSize); if (FUZ_rand(&lseed) & 1) { DISPLAYLEVEL(5, "t%u: windowLog : %u \n", testNb, cParams.windowLog); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_windowLog, cParams.windowLog, useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_windowLog, cParams.windowLog, opaqueAPI) ); assert(cParams.windowLog >= ZSTD_WINDOWLOG_MIN); /* guaranteed by ZSTD_adjustCParams() */ windowLogMalus = (cParams.windowLog - ZSTD_WINDOWLOG_MIN) / 5; } if (FUZ_rand(&lseed) & 1) { DISPLAYLEVEL(5, "t%u: hashLog : %u \n", testNb, cParams.hashLog); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_hashLog, cParams.hashLog, useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_hashLog, cParams.hashLog, opaqueAPI) ); } if (FUZ_rand(&lseed) & 1) { DISPLAYLEVEL(5, "t%u: chainLog : %u \n", testNb, cParams.chainLog); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_chainLog, cParams.chainLog, useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_chainLog, cParams.chainLog, opaqueAPI) ); } - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_searchLog, cParams.searchLog, useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_minMatch, cParams.searchLength, useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_targetLength, cParams.targetLength, useOpaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_searchLog, cParams.searchLog, opaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_minMatch, cParams.searchLength, opaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_targetLength, cParams.targetLength, opaqueAPI) ); /* mess with long distance matching parameters */ if (bigTests) { - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_enableLongDistanceMatching, FUZ_rand(&lseed) & 63, useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmHashLog, FUZ_randomClampedLength(&lseed, ZSTD_HASHLOG_MIN, 23), useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmMinMatch, FUZ_randomClampedLength(&lseed, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX), useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmBucketSizeLog, FUZ_randomClampedLength(&lseed, 0, ZSTD_LDM_BUCKETSIZELOG_MAX), useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmHashEveryLog, FUZ_randomClampedLength(&lseed, 0, ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN), useOpaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_enableLongDistanceMatching, FUZ_rand(&lseed) & 63, opaqueAPI) ); + if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmHashLog, FUZ_randomClampedLength(&lseed, ZSTD_HASHLOG_MIN, 23), opaqueAPI) ); + if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmMinMatch, FUZ_randomClampedLength(&lseed, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX), opaqueAPI) ); + if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmBucketSizeLog, FUZ_randomClampedLength(&lseed, 0, ZSTD_LDM_BUCKETSIZELOG_MAX), opaqueAPI) ); + if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_ldmHashEveryLog, FUZ_randomClampedLength(&lseed, 0, ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN), opaqueAPI) ); } /* mess with frame parameters */ if (FUZ_rand(&lseed) & 1) { U32 const checksumFlag = FUZ_rand(&lseed) & 1; DISPLAYLEVEL(5, "t%u: frame checksum : %u \n", testNb, checksumFlag); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_checksumFlag, checksumFlag, useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_checksumFlag, checksumFlag, opaqueAPI) ); } - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_dictIDFlag, FUZ_rand(&lseed) & 1, useOpaqueAPI) ); - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_contentSizeFlag, FUZ_rand(&lseed) & 1, useOpaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_dictIDFlag, FUZ_rand(&lseed) & 1, opaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_contentSizeFlag, FUZ_rand(&lseed) & 1, opaqueAPI) ); if (FUZ_rand(&lseed) & 1) { DISPLAYLEVEL(5, "t%u: pledgedSrcSize : %u \n", testNb, (U32)pledgedSrcSize); CHECK_Z( ZSTD_CCtx_setPledgedSrcSize(zc, pledgedSrcSize) ); } - /* multi-threading parameters */ - { U32 const nbThreadsCandidate = (FUZ_rand(&lseed) & 4) + 1; + /* multi-threading parameters. Only adjust ocassionally for small tests. */ + if (bigTests || (FUZ_rand(&lseed) & 0xF) == 0xF) { + U32 const nbThreadsCandidate = (FUZ_rand(&lseed) & 4) + 1; U32 const nbThreadsAdjusted = (windowLogMalus < nbThreadsCandidate) ? nbThreadsCandidate - windowLogMalus : 1; U32 const nbThreads = MIN(nbThreadsAdjusted, nbThreadsMax); DISPLAYLEVEL(5, "t%u: nbThreads : %u \n", testNb, nbThreads); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_nbWorkers, nbThreads, useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_nbWorkers, nbThreads, opaqueAPI) ); if (nbThreads > 1) { U32 const jobLog = FUZ_rand(&lseed) % (testLog+1); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_overlapSizeLog, FUZ_rand(&lseed) % 10, useOpaqueAPI) ); - CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_jobSize, (U32)FUZ_rLogLength(&lseed, jobLog), useOpaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_overlapSizeLog, FUZ_rand(&lseed) % 10, opaqueAPI) ); + CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_jobSize, (U32)FUZ_rLogLength(&lseed, jobLog), opaqueAPI) ); } } - if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_forceMaxWindow, FUZ_rand(&lseed) & 1, useOpaqueAPI) ); + if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_forceMaxWindow, FUZ_rand(&lseed) & 1, opaqueAPI) ); /* Apply parameters */ - if (useOpaqueAPI) { - DISPLAYLEVEL(6," t%u: applying CCtxParams \n", testNb); + if (opaqueAPI) { + DISPLAYLEVEL(5, "t%u: applying CCtxParams \n", testNb); CHECK_Z (ZSTD_CCtx_setParametersUsingCCtxParams(zc, cctxParams) ); } @@ -1709,7 +1932,7 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double } if (dict && dictSize) { /* test that compression parameters are rejected (correctly) after loading a non-NULL dictionary */ - if (useOpaqueAPI) { + if (opaqueAPI) { size_t const setError = ZSTD_CCtx_setParametersUsingCCtxParams(zc, cctxParams); CHECK(!ZSTD_isError(setError), "ZSTD_CCtx_setParametersUsingCCtxParams should have failed"); } else { @@ -1722,6 +1945,8 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double } } } + CHECK_Z(getCCtxParams(zc, &savedParams)); + /* multi-segments compression test */ XXH64_reset(&xxhState, 0); { ZSTD_outBuffer outBuff = { cBuffer, cBufferSize, 0 } ; @@ -1761,15 +1986,18 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double } } crcOrig = XXH64_digest(&xxhState); cSize = outBuff.pos; - DISPLAYLEVEL(5, "Frame completed : %u bytes \n", (U32)cSize); + DISPLAYLEVEL(5, "Frame completed : %zu bytes \n", cSize); } + CHECK(badParameters(zc, savedParams), "CCtx params are wrong"); + /* multi - fragments decompression test */ if (!dictSize /* don't reset if dictionary : could be different */ && (FUZ_rand(&lseed) & 1)) { DISPLAYLEVEL(5, "resetting DCtx (dict:%08X) \n", (U32)(size_t)dict); CHECK_Z( ZSTD_resetDStream(zd) ); } else { - DISPLAYLEVEL(5, "using dict of size %u \n", (U32)dictSize); + if (dictSize) + DISPLAYLEVEL(5, "using dictionary of size %zu \n", dictSize); CHECK_Z( ZSTD_initDStream_usingDict(zd, dict, dictSize) ); } { size_t decompressionResult = 1; @@ -1853,7 +2081,7 @@ _output_error: /*-******************************************************* * Command line *********************************************************/ -int FUZ_usage(const char* programName) +static int FUZ_usage(const char* programName) { DISPLAY( "Usage :\n"); DISPLAY( " %s [args]\n", programName); @@ -1883,7 +2111,6 @@ int main(int argc, const char** argv) int bigTests = (sizeof(size_t) == 8); e_api selected_api = simple_api; const char* const programName = argv[0]; - U32 useOpaqueAPI = 0; int argNb; /* Check command line */ @@ -1896,7 +2123,6 @@ int main(int argc, const char** argv) if (!strcmp(argument, "--mt")) { selected_api=mt_api; testNb += !testNb; continue; } if (!strcmp(argument, "--newapi")) { selected_api=advanced_api; testNb += !testNb; continue; } - if (!strcmp(argument, "--opaqueapi")) { selected_api=advanced_api; testNb += !testNb; useOpaqueAPI = 1; continue; } if (!strcmp(argument, "--no-big-tests")) { bigTests=0; continue; } argument++; @@ -2012,7 +2238,7 @@ int main(int argc, const char** argv) result = fuzzerTests_MT(seed, nbTests, testNb, ((double)proba) / 100, bigTests); break; case advanced_api : - result = fuzzerTests_newAPI(seed, nbTests, testNb, ((double)proba) / 100, bigTests, useOpaqueAPI); + result = fuzzerTests_newAPI(seed, nbTests, testNb, ((double)proba) / 100, bigTests); break; default : assert(0); /* impossible */ |