diff --git a/.travis.yml b/.travis.yml index 36df67052..a61b82876 100644 --- a/.travis.yml +++ b/.travis.yml @@ -20,23 +20,25 @@ matrix: - env: Cmd='make gcc8install && CC=gcc-8 CFLAGS="-Werror -O3" make -j all' - env: Cmd='make clang38install && CC=clang-3.8 make clean msan-test-zstd' + - env: Cmd='make staticAnalyze' + - env: Cmd='make gcc6install && CC=gcc-6 make clean uasan-fuzztest' - env: Cmd='make gcc6install libc6install && make clean && CC=gcc-6 CFLAGS=-m32 make uasan-fuzztest' - env: Cmd='make clang38install && CC=clang-3.8 make clean msan-fuzztest' - env: Cmd='make clang38install && CC=clang-3.8 make clean tsan-test-zstream' - - env: Cmd='make arminstall && make armfuzz' - - env: Cmd='make arminstall && make aarch64fuzz' - - env: Cmd='make ppcinstall && make ppcfuzz' - - env: Cmd='make ppcinstall && make ppc64fuzz' - - env: Cmd='make -j uasanregressiontest && make clean && make -j msanregressiontest' - env: Cmd='make valgrindinstall && make -C tests clean valgrindTest && make clean && make -C tests test-fuzzer-stackmode' + - env: Cmd='make arminstall && make armfuzz' + - env: Cmd='make arminstall && make aarch64fuzz' + - env: Cmd='make ppcinstall && make ppcfuzz' + - env: Cmd='make ppcinstall && make ppc64fuzz' + - env: Cmd='make lz4install && make -C tests test-lz4 && make clean && make -C tests test-pool && make clean && bash tests/libzstd_partial_builds.sh' diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md new file mode 100644 index 000000000..0f7ad8bfc --- /dev/null +++ b/CODE_OF_CONDUCT.md @@ -0,0 +1,5 @@ +# Code of Conduct + +Facebook has adopted a Code of Conduct that we expect project participants to adhere to. +Please read the [full text](https://code.fb.com/codeofconduct/) +so that you can understand what actions will and will not be tolerated. diff --git a/Makefile b/Makefile index 6c4ab9c9b..a17900450 100644 --- a/Makefile +++ b/Makefile @@ -23,6 +23,7 @@ else EXT = endif +## default: Build lib-release and zstd-release .PHONY: default default: lib-release zstd-release @@ -30,10 +31,9 @@ default: lib-release zstd-release all: allmost examples manual contrib .PHONY: allmost -allmost: allzstd - $(MAKE) -C $(ZWRAPDIR) all +allmost: allzstd zlibwrapper -#skip zwrapper, can't build that on alternate architectures without the proper zlib installed +# skip zwrapper, can't build that on alternate architectures without the proper zlib installed .PHONY: allzstd allzstd: lib $(MAKE) -C $(PRGDIR) all @@ -44,8 +44,8 @@ all32: $(MAKE) -C $(PRGDIR) zstd32 $(MAKE) -C $(TESTDIR) all32 -.PHONY: lib lib-release -lib lib-release: +.PHONY: lib lib-release libzstd.a +lib lib-release : @$(MAKE) -C $(ZSTDDIR) $@ .PHONY: zstd zstd-release @@ -59,8 +59,8 @@ zstdmt: cp $(PRGDIR)/zstd$(EXT) ./zstdmt$(EXT) .PHONY: zlibwrapper -zlibwrapper: - $(MAKE) -C $(ZWRAPDIR) test +zlibwrapper: lib + $(MAKE) -C $(ZWRAPDIR) all .PHONY: test test: MOREFLAGS += -g -DDEBUGLEVEL=1 -Werror @@ -88,6 +88,7 @@ contrib: lib $(MAKE) -C contrib/pzstd all $(MAKE) -C contrib/seekable_format/examples all $(MAKE) -C contrib/adaptive-compression all + $(MAKE) -C contrib/largeNbDicts all .PHONY: cleanTabs cleanTabs: @@ -104,6 +105,7 @@ clean: @$(MAKE) -C contrib/pzstd $@ > $(VOID) @$(MAKE) -C contrib/seekable_format/examples $@ > $(VOID) @$(MAKE) -C contrib/adaptive-compression $@ > $(VOID) + @$(MAKE) -C contrib/largeNbDicts $@ > $(VOID) @$(RM) zstd$(EXT) zstdmt$(EXT) tmp* @$(RM) -r lz4 @echo Cleaning completed @@ -114,11 +116,26 @@ clean: ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD DragonFly NetBSD MSYS_NT)) HOST_OS = POSIX -CMAKE_PARAMS = -DZSTD_BUILD_CONTRIB:BOOL=ON -DZSTD_BUILD_STATIC:BOOL=ON -DZSTD_BUILD_TESTS:BOOL=ON -DZSTD_ZLIB_SUPPORT:BOOL=ON -DZSTD_LZMA_SUPPORT:BOOL=ON -DCMAKE_BUILD_TYPE=Release +CMAKE_PARAMS = -DZSTD_BUILD_CONTRIB:BOOL=ON -DZSTD_BUILD_STATIC:BOOL=ON -DZSTD_BUILD_TESTS:BOOL=ON -DZSTD_ZLIB_SUPPORT:BOOL=ON -DZSTD_LZMA_SUPPORT:BOOL=ON -DCMAKE_BUILD_TYPE=Release +# Print a two column output of targets and their description. To add a target description, put a +# comment in the Makefile with the format "## : ". For example: +# +## list: Print all targets and their descriptions (if provided) .PHONY: list list: - @$(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 '^$@$$' | xargs + @TARGETS=$$($(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null \ + | awk -v RS= -F: '/^# File/,/^# Finished Make data base/ {if ($$1 !~ "^[#.]") {print $$1}}' \ + | egrep -v -e '^[^[:alnum:]]' | sort); \ + { \ + printf "Target Name\tDescription\n"; \ + printf "%0.s-" {1..16}; printf "\t"; printf "%0.s-" {1..40}; printf "\n"; \ + for target in $$TARGETS; do \ + line=$$(egrep "^##[[:space:]]+$$target:" $(lastword $(MAKEFILE_LIST))); \ + description=$$(echo $$line | awk '{i=index($$0,":"); print substr($$0,i+1)}' | xargs); \ + printf "$$target\t$$description\n"; \ + done \ + } | column -t -s $$'\t' .PHONY: install clangtest armtest usan asan uasan install: @@ -198,7 +215,7 @@ gcc6test: clean clangtest: clean clang -v - $(MAKE) all CXX=clang-++ CC=clang MOREFLAGS="-Werror -Wconversion -Wno-sign-conversion -Wdocumentation" + $(MAKE) all CXX=clang++ CC=clang MOREFLAGS="-Werror -Wconversion -Wno-sign-conversion -Wdocumentation" armtest: clean $(MAKE) -C $(TESTDIR) datagen # use native, faster @@ -351,7 +368,10 @@ bmi32build: clean $(CC) -v CFLAGS="-O3 -mbmi -m32 -Werror" $(MAKE) -C $(TESTDIR) test -staticAnalyze: clean +# static analyzer test uses clang's scan-build +# does not analyze zlibWrapper, due to detected issues in zlib source code +staticAnalyze: SCANBUILD ?= scan-build +staticAnalyze: $(CC) -v - CPPFLAGS=-g scan-build --status-bugs -v $(MAKE) all + CC=$(CC) CPPFLAGS=-g $(SCANBUILD) --status-bugs -v $(MAKE) allzstd examples contrib endif diff --git a/NEWS b/NEWS index e3bfb242d..1553f5ad8 100644 --- a/NEWS +++ b/NEWS @@ -1,3 +1,8 @@ +v1.3.6 +perf: much faster dictionary builder, by @jenniferliu +api : reduced DDict size by 2 KB +misc: tests/paramgrill, a parameter optimizer, by @GeorgeLu97 + v1.3.5 perf: much faster dictionary compression, by @felixhandte perf: small quality improvement for dictionary generation, by @terrelln diff --git a/build/VS2008/fuzzer/fuzzer.vcproj b/build/VS2008/fuzzer/fuzzer.vcproj index f6ea1f855..4d444caef 100644 --- a/build/VS2008/fuzzer/fuzzer.vcproj +++ b/build/VS2008/fuzzer/fuzzer.vcproj @@ -336,6 +336,10 @@ RelativePath="..\..\..\lib\dictBuilder\cover.c" > + + @@ -482,6 +486,10 @@ RelativePath="..\..\..\lib\dictBuilder\zdict.h" > + + diff --git a/build/VS2008/zstd/zstd.vcproj b/build/VS2008/zstd/zstd.vcproj index 5d9f68327..46d7c85dd 100644 --- a/build/VS2008/zstd/zstd.vcproj +++ b/build/VS2008/zstd/zstd.vcproj @@ -348,6 +348,10 @@ RelativePath="..\..\..\lib\dictBuilder\cover.c" > + + @@ -522,6 +526,10 @@ RelativePath="..\..\..\lib\dictBuilder\zdict.h" > + + diff --git a/build/VS2008/zstdlib/zstdlib.vcproj b/build/VS2008/zstdlib/zstdlib.vcproj index 7234b02ac..950a6f39c 100644 --- a/build/VS2008/zstdlib/zstdlib.vcproj +++ b/build/VS2008/zstdlib/zstdlib.vcproj @@ -332,6 +332,10 @@ RelativePath="..\..\..\lib\dictBuilder\cover.c" > + + @@ -502,6 +506,10 @@ RelativePath="..\..\..\lib\dictBuilder\zdict.h" > + + diff --git a/build/VS2010/fuzzer/fuzzer.vcxproj b/build/VS2010/fuzzer/fuzzer.vcxproj index f0d1ab066..6077cd2c1 100644 --- a/build/VS2010/fuzzer/fuzzer.vcxproj +++ b/build/VS2010/fuzzer/fuzzer.vcxproj @@ -176,6 +176,7 @@ + @@ -199,6 +200,7 @@ + diff --git a/build/VS2010/libzstd-dll/libzstd-dll.vcxproj b/build/VS2010/libzstd-dll/libzstd-dll.vcxproj index 92d518d3d..6e14e020e 100644 --- a/build/VS2010/libzstd-dll/libzstd-dll.vcxproj +++ b/build/VS2010/libzstd-dll/libzstd-dll.vcxproj @@ -43,6 +43,7 @@ + diff --git a/build/VS2010/libzstd/libzstd.vcxproj b/build/VS2010/libzstd/libzstd.vcxproj index c306fcec9..18f5cb533 100644 --- a/build/VS2010/libzstd/libzstd.vcxproj +++ b/build/VS2010/libzstd/libzstd.vcxproj @@ -43,6 +43,7 @@ + diff --git a/build/VS2010/zstd/zstd.vcxproj b/build/VS2010/zstd/zstd.vcxproj index 4af281326..936960b58 100644 --- a/build/VS2010/zstd/zstd.vcxproj +++ b/build/VS2010/zstd/zstd.vcxproj @@ -40,6 +40,7 @@ + @@ -61,6 +62,7 @@ + diff --git a/build/cmake/contrib/gen_html/CMakeLists.txt b/build/cmake/contrib/gen_html/CMakeLists.txt index 958e9d173..c4062d475 100644 --- a/build/cmake/contrib/gen_html/CMakeLists.txt +++ b/build/cmake/contrib/gen_html/CMakeLists.txt @@ -27,4 +27,4 @@ ADD_CUSTOM_TARGET(zstd_manual.html ALL ${GENHTML_BINARY} "${LIBVERSION}" "${LIBRARY_DIR}/zstd.h" "${PROJECT_BINARY_DIR}/zstd_manual.html" DEPENDS gen_html COMMENT "Update zstd manual") -INSTALL(FILES "${PROJECT_BINARY_DIR}/zstd_manual.html" DESTINATION "${CMAKE_INSTALL_PREFIX}/${DOC_INSTALL_DIR}") +INSTALL(FILES "${PROJECT_BINARY_DIR}/zstd_manual.html" DESTINATION "${CMAKE_INSTALL_DOCDIR}") diff --git a/build/cmake/lib/CMakeLists.txt b/build/cmake/lib/CMakeLists.txt index e84e06301..ffc196ddc 100644 --- a/build/cmake/lib/CMakeLists.txt +++ b/build/cmake/lib/CMakeLists.txt @@ -47,6 +47,7 @@ SET(Sources ${LIBRARY_DIR}/decompress/huf_decompress.c ${LIBRARY_DIR}/decompress/zstd_decompress.c ${LIBRARY_DIR}/dictBuilder/cover.c + ${LIBRARY_DIR}/dictBuilder/fastcover.c ${LIBRARY_DIR}/dictBuilder/divsufsort.c ${LIBRARY_DIR}/dictBuilder/zdict.c ${LIBRARY_DIR}/deprecated/zbuff_common.c @@ -74,6 +75,7 @@ SET(Headers ${LIBRARY_DIR}/compress/zstd_ldm.h ${LIBRARY_DIR}/compress/zstdmt_compress.h ${LIBRARY_DIR}/dictBuilder/zdict.h + ${LIBRARY_DIR}/dictBuilder/cover.h ${LIBRARY_DIR}/deprecated/zbuff.h) IF (ZSTD_LEGACY_SUPPORT) @@ -178,6 +180,7 @@ INSTALL(FILES ${LIBRARY_DIR}/zstd.h ${LIBRARY_DIR}/deprecated/zbuff.h ${LIBRARY_DIR}/dictBuilder/zdict.h + ${LIBRARY_DIR}/dictBuilder/cover.h ${LIBRARY_DIR}/common/zstd_errors.h DESTINATION "include") diff --git a/contrib/experimental_dict_builders/benchmarkDictBuilder/Makefile b/contrib/experimental_dict_builders/benchmarkDictBuilder/Makefile index 681494888..72ce04f2a 100644 --- a/contrib/experimental_dict_builders/benchmarkDictBuilder/Makefile +++ b/contrib/experimental_dict_builders/benchmarkDictBuilder/Makefile @@ -2,10 +2,9 @@ ARG := CC ?= gcc CFLAGS ?= -O3 -INCLUDES := -I ../randomDictBuilder -I ../fastCover -I ../../../programs -I ../../../lib/common -I ../../../lib -I ../../../lib/dictBuilder +INCLUDES := -I ../randomDictBuilder -I ../../../programs -I ../../../lib/common -I ../../../lib -I ../../../lib/dictBuilder RANDOM_FILE := ../randomDictBuilder/random.c -FAST_FILE := ../fastCover/fastCover.c IO_FILE := ../randomDictBuilder/io.c all: run clean @@ -22,8 +21,8 @@ test: benchmarkTest clean benchmarkTest: benchmark test.sh sh test.sh -benchmark: benchmark.o io.o random.o fastCover.o libzstd.a - $(CC) $(CFLAGS) benchmark.o io.o random.o fastCover.o libzstd.a -o benchmark +benchmark: benchmark.o io.o random.o libzstd.a + $(CC) $(CFLAGS) benchmark.o io.o random.o libzstd.a -o benchmark benchmark.o: benchmark.c $(CC) $(CFLAGS) $(INCLUDES) -c benchmark.c @@ -31,9 +30,6 @@ benchmark.o: benchmark.c random.o: $(RANDOM_FILE) $(CC) $(CFLAGS) $(INCLUDES) -c $(RANDOM_FILE) -fastCover.o: $(FAST_FILE) - $(CC) $(CFLAGS) $(INCLUDES) -c $(FAST_FILE) - io.o: $(IO_FILE) $(CC) $(CFLAGS) $(INCLUDES) -c $(IO_FILE) diff --git a/contrib/experimental_dict_builders/benchmarkDictBuilder/README.md b/contrib/experimental_dict_builders/benchmarkDictBuilder/README.md index 559776e2b..6a6c7f1d2 100644 --- a/contrib/experimental_dict_builders/benchmarkDictBuilder/README.md +++ b/contrib/experimental_dict_builders/benchmarkDictBuilder/README.md @@ -14,113 +14,836 @@ make ARG="in=../../../lib/dictBuilder in=../../../lib/compress" ###Benchmarking Result: - First Cover is optimize cover, second Cover uses optimized d and k from first one. -- For every f value of fastCover, the first one is optimize fastCover and the second one uses optimized d and k from first one. +- For every f value of fastCover, the first one is optimize fastCover and the second one uses optimized d and k from first one. This is run for accel values from 1 to 10. - Fourth column is chosen d and fifth column is chosen k github: -NODICT 0.000025 2.999642 -RANDOM 0.030101 8.791189 -LEGACY 0.913108 8.173529 -COVER 59.234160 10.652243 8 1298 -COVER 6.258459 10.652243 8 1298 -FAST15 9.959246 10.555630 8 1874 -FAST15 0.077719 10.555630 8 1874 -FAST16 10.028343 10.701698 8 1106 -FAST16 0.078117 10.701698 8 1106 -FAST17 10.567355 10.650652 8 1106 -FAST17 0.124833 10.650652 8 1106 -FAST18 11.795287 10.499142 8 1826 -FAST18 0.086992 10.499142 8 1826 -FAST19 13.132451 10.527140 8 1826 -FAST19 0.134716 10.527140 8 1826 -FAST20 14.366314 10.494710 8 1826 -FAST20 0.128844 10.494710 8 1826 -FAST21 14.941238 10.503488 8 1778 -FAST21 0.134975 10.503488 8 1778 -FAST22 15.146226 10.509284 8 1826 -FAST22 0.146918 10.509284 8 1826 -FAST23 16.260552 10.509284 8 1826 -FAST23 0.158494 10.509284 8 1826 -FAST24 16.806037 10.512369 8 1826 -FAST24 0.190464 10.512369 8 1826 +NODICT 0.000004 2.999642 +RANDOM 0.024560 8.791189 +LEGACY 0.727109 8.173529 +COVER 40.565676 10.652243 8 1298 +COVER 3.608284 10.652243 8 1298 +FAST f=15 a=1 4.181024 10.570882 8 1154 +FAST f=15 a=1 0.040788 10.570882 8 1154 +FAST f=15 a=2 3.548352 10.574287 6 1970 +FAST f=15 a=2 0.035535 10.574287 6 1970 +FAST f=15 a=3 3.287364 10.613950 6 1010 +FAST f=15 a=3 0.032182 10.613950 6 1010 +FAST f=15 a=4 3.184976 10.573883 6 1058 +FAST f=15 a=4 0.029878 10.573883 6 1058 +FAST f=15 a=5 3.045513 10.580640 8 1154 +FAST f=15 a=5 0.022162 10.580640 8 1154 +FAST f=15 a=6 3.003296 10.583677 6 1010 +FAST f=15 a=6 0.028091 10.583677 6 1010 +FAST f=15 a=7 2.952655 10.622551 6 1106 +FAST f=15 a=7 0.02724 10.622551 6 1106 +FAST f=15 a=8 2.945674 10.614657 6 1010 +FAST f=15 a=8 0.027264 10.614657 6 1010 +FAST f=15 a=9 3.153439 10.564018 8 1154 +FAST f=15 a=9 0.020635 10.564018 8 1154 +FAST f=15 a=10 2.950416 10.511454 6 1010 +FAST f=15 a=10 0.026606 10.511454 6 1010 +FAST f=16 a=1 3.970029 10.681035 8 1154 +FAST f=16 a=1 0.038188 10.681035 8 1154 +FAST f=16 a=2 3.422892 10.484978 6 1874 +FAST f=16 a=2 0.034702 10.484978 6 1874 +FAST f=16 a=3 3.215836 10.632631 8 1154 +FAST f=16 a=3 0.026084 10.632631 8 1154 +FAST f=16 a=4 3.081353 10.626533 6 1106 +FAST f=16 a=4 0.030032 10.626533 6 1106 +FAST f=16 a=5 3.041241 10.545027 8 1922 +FAST f=16 a=5 0.022882 10.545027 8 1922 +FAST f=16 a=6 2.989390 10.638284 6 1874 +FAST f=16 a=6 0.028308 10.638284 6 1874 +FAST f=16 a=7 3.001581 10.797136 6 1106 +FAST f=16 a=7 0.027479 10.797136 6 1106 +FAST f=16 a=8 2.984107 10.658356 8 1058 +FAST f=16 a=8 0.021099 10.658356 8 1058 +FAST f=16 a=9 2.925788 10.523869 6 1010 +FAST f=16 a=9 0.026905 10.523869 6 1010 +FAST f=16 a=10 2.889605 10.745841 6 1874 +FAST f=16 a=10 0.026846 10.745841 6 1874 +FAST f=17 a=1 4.031953 10.672080 8 1202 +FAST f=17 a=1 0.040658 10.672080 8 1202 +FAST f=17 a=2 3.458107 10.589352 8 1106 +FAST f=17 a=2 0.02926 10.589352 8 1106 +FAST f=17 a=3 3.291189 10.662714 8 1154 +FAST f=17 a=3 0.026531 10.662714 8 1154 +FAST f=17 a=4 3.154950 10.549456 8 1346 +FAST f=17 a=4 0.024991 10.549456 8 1346 +FAST f=17 a=5 3.092271 10.541670 6 1202 +FAST f=17 a=5 0.038285 10.541670 6 1202 +FAST f=17 a=6 3.166146 10.729112 6 1874 +FAST f=17 a=6 0.038217 10.729112 6 1874 +FAST f=17 a=7 3.035467 10.810485 6 1106 +FAST f=17 a=7 0.036655 10.810485 6 1106 +FAST f=17 a=8 3.035668 10.530532 6 1058 +FAST f=17 a=8 0.037715 10.530532 6 1058 +FAST f=17 a=9 2.987917 10.589802 8 1922 +FAST f=17 a=9 0.02217 10.589802 8 1922 +FAST f=17 a=10 2.981647 10.722579 8 1106 +FAST f=17 a=10 0.021948 10.722579 8 1106 +FAST f=18 a=1 4.067144 10.634943 8 1154 +FAST f=18 a=1 0.041386 10.634943 8 1154 +FAST f=18 a=2 3.507377 10.546230 6 1970 +FAST f=18 a=2 0.037572 10.546230 6 1970 +FAST f=18 a=3 3.323015 10.648061 8 1154 +FAST f=18 a=3 0.028306 10.648061 8 1154 +FAST f=18 a=4 3.216735 10.705402 6 1010 +FAST f=18 a=4 0.030755 10.705402 6 1010 +FAST f=18 a=5 3.175794 10.588154 8 1874 +FAST f=18 a=5 0.025315 10.588154 8 1874 +FAST f=18 a=6 3.127459 10.751104 8 1106 +FAST f=18 a=6 0.023897 10.751104 8 1106 +FAST f=18 a=7 3.083017 10.780402 6 1106 +FAST f=18 a=7 0.029158 10.780402 6 1106 +FAST f=18 a=8 3.069700 10.547226 8 1346 +FAST f=18 a=8 0.024046 10.547226 8 1346 +FAST f=18 a=9 3.056591 10.674759 6 1010 +FAST f=18 a=9 0.028496 10.674759 6 1010 +FAST f=18 a=10 3.063588 10.737578 8 1106 +FAST f=18 a=10 0.023033 10.737578 8 1106 +FAST f=19 a=1 4.164041 10.650333 8 1154 +FAST f=19 a=1 0.042906 10.650333 8 1154 +FAST f=19 a=2 3.585409 10.577066 6 1058 +FAST f=19 a=2 0.038994 10.577066 6 1058 +FAST f=19 a=3 3.439643 10.639403 8 1154 +FAST f=19 a=3 0.028427 10.639403 8 1154 +FAST f=19 a=4 3.268869 10.554410 8 1298 +FAST f=19 a=4 0.026866 10.554410 8 1298 +FAST f=19 a=5 3.238225 10.615109 6 1010 +FAST f=19 a=5 0.03078 10.615109 6 1010 +FAST f=19 a=6 3.199558 10.609782 6 1874 +FAST f=19 a=6 0.030099 10.609782 6 1874 +FAST f=19 a=7 3.132395 10.794753 6 1106 +FAST f=19 a=7 0.028964 10.794753 6 1106 +FAST f=19 a=8 3.148446 10.554842 8 1298 +FAST f=19 a=8 0.024277 10.554842 8 1298 +FAST f=19 a=9 3.108324 10.668763 6 1010 +FAST f=19 a=9 0.02896 10.668763 6 1010 +FAST f=19 a=10 3.159863 10.757347 8 1106 +FAST f=19 a=10 0.023351 10.757347 8 1106 +FAST f=20 a=1 4.462698 10.661788 8 1154 +FAST f=20 a=1 0.047174 10.661788 8 1154 +FAST f=20 a=2 3.820269 10.678612 6 1106 +FAST f=20 a=2 0.040807 10.678612 6 1106 +FAST f=20 a=3 3.644955 10.648424 8 1154 +FAST f=20 a=3 0.031398 10.648424 8 1154 +FAST f=20 a=4 3.546257 10.559756 8 1298 +FAST f=20 a=4 0.029856 10.559756 8 1298 +FAST f=20 a=5 3.485248 10.646637 6 1010 +FAST f=20 a=5 0.033756 10.646637 6 1010 +FAST f=20 a=6 3.490438 10.775824 8 1106 +FAST f=20 a=6 0.028338 10.775824 8 1106 +FAST f=20 a=7 3.631289 10.801795 6 1106 +FAST f=20 a=7 0.035228 10.801795 6 1106 +FAST f=20 a=8 3.758936 10.545116 8 1346 +FAST f=20 a=8 0.027495 10.545116 8 1346 +FAST f=20 a=9 3.707024 10.677454 6 1010 +FAST f=20 a=9 0.031326 10.677454 6 1010 +FAST f=20 a=10 3.586593 10.756017 8 1106 +FAST f=20 a=10 0.027122 10.756017 8 1106 +FAST f=21 a=1 5.701396 10.655398 8 1154 +FAST f=21 a=1 0.067744 10.655398 8 1154 +FAST f=21 a=2 5.270542 10.650743 6 1106 +FAST f=21 a=2 0.052999 10.650743 6 1106 +FAST f=21 a=3 4.945294 10.652380 8 1154 +FAST f=21 a=3 0.052678 10.652380 8 1154 +FAST f=21 a=4 4.894079 10.543185 8 1298 +FAST f=21 a=4 0.04997 10.543185 8 1298 +FAST f=21 a=5 4.785417 10.630321 6 1010 +FAST f=21 a=5 0.045294 10.630321 6 1010 +FAST f=21 a=6 4.789381 10.664477 6 1874 +FAST f=21 a=6 0.046578 10.664477 6 1874 +FAST f=21 a=7 4.302955 10.805179 6 1106 +FAST f=21 a=7 0.041205 10.805179 6 1106 +FAST f=21 a=8 4.034630 10.551211 8 1298 +FAST f=21 a=8 0.040121 10.551211 8 1298 +FAST f=21 a=9 4.523868 10.799114 6 1010 +FAST f=21 a=9 0.043592 10.799114 6 1010 +FAST f=21 a=10 4.760736 10.750255 8 1106 +FAST f=21 a=10 0.043483 10.750255 8 1106 +FAST f=22 a=1 6.743064 10.640537 8 1154 +FAST f=22 a=1 0.086967 10.640537 8 1154 +FAST f=22 a=2 6.121739 10.626638 6 1970 +FAST f=22 a=2 0.066337 10.626638 6 1970 +FAST f=22 a=3 5.248851 10.640688 8 1154 +FAST f=22 a=3 0.054935 10.640688 8 1154 +FAST f=22 a=4 5.436579 10.588333 8 1298 +FAST f=22 a=4 0.064113 10.588333 8 1298 +FAST f=22 a=5 5.812815 10.652653 6 1010 +FAST f=22 a=5 0.058189 10.652653 6 1010 +FAST f=22 a=6 5.745472 10.666437 6 1874 +FAST f=22 a=6 0.057188 10.666437 6 1874 +FAST f=22 a=7 5.716393 10.806911 6 1106 +FAST f=22 a=7 0.056 10.806911 6 1106 +FAST f=22 a=8 5.698799 10.530784 8 1298 +FAST f=22 a=8 0.0583 10.530784 8 1298 +FAST f=22 a=9 5.710533 10.777391 6 1010 +FAST f=22 a=9 0.054945 10.777391 6 1010 +FAST f=22 a=10 5.685395 10.745023 8 1106 +FAST f=22 a=10 0.056526 10.745023 8 1106 +FAST f=23 a=1 7.836923 10.638828 8 1154 +FAST f=23 a=1 0.099522 10.638828 8 1154 +FAST f=23 a=2 6.627834 10.631061 6 1970 +FAST f=23 a=2 0.066769 10.631061 6 1970 +FAST f=23 a=3 5.602533 10.647288 8 1154 +FAST f=23 a=3 0.064513 10.647288 8 1154 +FAST f=23 a=4 6.005580 10.568747 8 1298 +FAST f=23 a=4 0.062022 10.568747 8 1298 +FAST f=23 a=5 5.481816 10.676921 6 1010 +FAST f=23 a=5 0.058959 10.676921 6 1010 +FAST f=23 a=6 5.460444 10.666194 6 1874 +FAST f=23 a=6 0.057687 10.666194 6 1874 +FAST f=23 a=7 5.659822 10.800377 6 1106 +FAST f=23 a=7 0.06783 10.800377 6 1106 +FAST f=23 a=8 6.826940 10.522167 8 1298 +FAST f=23 a=8 0.070533 10.522167 8 1298 +FAST f=23 a=9 6.804757 10.577799 8 1682 +FAST f=23 a=9 0.069949 10.577799 8 1682 +FAST f=23 a=10 6.774933 10.742093 8 1106 +FAST f=23 a=10 0.068395 10.742093 8 1106 +FAST f=24 a=1 8.444110 10.632783 8 1154 +FAST f=24 a=1 0.094357 10.632783 8 1154 +FAST f=24 a=2 7.289578 10.631061 6 1970 +FAST f=24 a=2 0.098515 10.631061 6 1970 +FAST f=24 a=3 8.619780 10.646289 8 1154 +FAST f=24 a=3 0.098041 10.646289 8 1154 +FAST f=24 a=4 8.508455 10.555199 8 1298 +FAST f=24 a=4 0.093885 10.555199 8 1298 +FAST f=24 a=5 8.471145 10.674363 6 1010 +FAST f=24 a=5 0.088676 10.674363 6 1010 +FAST f=24 a=6 8.426727 10.667228 6 1874 +FAST f=24 a=6 0.087247 10.667228 6 1874 +FAST f=24 a=7 8.356826 10.803027 6 1106 +FAST f=24 a=7 0.085835 10.803027 6 1106 +FAST f=24 a=8 6.756811 10.522049 8 1298 +FAST f=24 a=8 0.07107 10.522049 8 1298 +FAST f=24 a=9 6.548169 10.571882 8 1682 +FAST f=24 a=9 0.0713 10.571882 8 1682 +FAST f=24 a=10 8.238079 10.736453 8 1106 +FAST f=24 a=10 0.07004 10.736453 8 1106 + hg-commands: -NODICT 0.000026 2.425291 -RANDOM 0.046270 3.490331 -LEGACY 0.847904 3.911682 -COVER 71.691804 4.132653 8 386 -COVER 3.187085 4.132653 8 386 -FAST15 11.593687 3.920720 6 1106 -FAST15 0.082431 3.920720 6 1106 -FAST16 11.775958 4.033306 8 674 -FAST16 0.092587 4.033306 8 674 -FAST17 11.965064 4.064132 8 1490 -FAST17 0.106382 4.064132 8 1490 -FAST18 11.438197 4.086714 8 290 -FAST18 0.097293 4.086714 8 290 -FAST19 12.292512 4.097947 8 578 -FAST19 0.104406 4.097947 8 578 -FAST20 13.857857 4.102851 8 434 -FAST20 0.139467 4.102851 8 434 -FAST21 14.599613 4.105350 8 530 -FAST21 0.189416 4.105350 8 530 -FAST22 15.966109 4.104100 8 530 -FAST22 0.183817 4.104100 8 530 -FAST23 18.033645 4.098110 8 914 -FAST23 0.246641 4.098110 8 914 -FAST24 22.992891 4.117367 8 722 -FAST24 0.285994 4.117367 8 722 +NODICT 0.000005 2.425276 +RANDOM 0.046332 3.490331 +LEGACY 0.720351 3.911682 +COVER 45.507731 4.132653 8 386 +COVER 1.868810 4.132653 8 386 +FAST f=15 a=1 4.561427 3.866894 8 1202 +FAST f=15 a=1 0.048946 3.866894 8 1202 +FAST f=15 a=2 3.574462 3.892119 8 1538 +FAST f=15 a=2 0.033677 3.892119 8 1538 +FAST f=15 a=3 3.230227 3.888791 6 1346 +FAST f=15 a=3 0.034312 3.888791 6 1346 +FAST f=15 a=4 3.042388 3.899739 8 1010 +FAST f=15 a=4 0.024307 3.899739 8 1010 +FAST f=15 a=5 2.800148 3.896220 8 818 +FAST f=15 a=5 0.022331 3.896220 8 818 +FAST f=15 a=6 2.706518 3.882039 8 578 +FAST f=15 a=6 0.020955 3.882039 8 578 +FAST f=15 a=7 2.701820 3.885430 6 866 +FAST f=15 a=7 0.026074 3.885430 6 866 +FAST f=15 a=8 2.604445 3.906932 8 1826 +FAST f=15 a=8 0.021789 3.906932 8 1826 +FAST f=15 a=9 2.598568 3.870324 6 1682 +FAST f=15 a=9 0.026004 3.870324 6 1682 +FAST f=15 a=10 2.575920 3.920783 8 1442 +FAST f=15 a=10 0.020228 3.920783 8 1442 +FAST f=16 a=1 4.630623 4.001430 8 770 +FAST f=16 a=1 0.047497 4.001430 8 770 +FAST f=16 a=2 3.674721 3.974431 8 1874 +FAST f=16 a=2 0.035761 3.974431 8 1874 +FAST f=16 a=3 3.338384 3.978703 8 1010 +FAST f=16 a=3 0.029436 3.978703 8 1010 +FAST f=16 a=4 3.004412 3.983035 8 1010 +FAST f=16 a=4 0.025744 3.983035 8 1010 +FAST f=16 a=5 2.881892 3.987710 8 770 +FAST f=16 a=5 0.023211 3.987710 8 770 +FAST f=16 a=6 2.807410 3.952717 8 1298 +FAST f=16 a=6 0.023199 3.952717 8 1298 +FAST f=16 a=7 2.819623 3.994627 8 770 +FAST f=16 a=7 0.021806 3.994627 8 770 +FAST f=16 a=8 2.740092 3.954032 8 1826 +FAST f=16 a=8 0.0226 3.954032 8 1826 +FAST f=16 a=9 2.682564 3.969879 6 1442 +FAST f=16 a=9 0.026324 3.969879 6 1442 +FAST f=16 a=10 2.657959 3.969755 8 674 +FAST f=16 a=10 0.020413 3.969755 8 674 +FAST f=17 a=1 4.729228 4.046000 8 530 +FAST f=17 a=1 0.049703 4.046000 8 530 +FAST f=17 a=2 3.764510 3.991519 8 1970 +FAST f=17 a=2 0.038195 3.991519 8 1970 +FAST f=17 a=3 3.416992 4.006296 6 914 +FAST f=17 a=3 0.036244 4.006296 6 914 +FAST f=17 a=4 3.145626 3.979182 8 1970 +FAST f=17 a=4 0.028676 3.979182 8 1970 +FAST f=17 a=5 2.995070 4.050070 8 770 +FAST f=17 a=5 0.025707 4.050070 8 770 +FAST f=17 a=6 2.911833 4.040024 8 770 +FAST f=17 a=6 0.02453 4.040024 8 770 +FAST f=17 a=7 2.894796 4.015884 8 818 +FAST f=17 a=7 0.023956 4.015884 8 818 +FAST f=17 a=8 2.789962 4.039303 8 530 +FAST f=17 a=8 0.023219 4.039303 8 530 +FAST f=17 a=9 2.787625 3.996762 8 1634 +FAST f=17 a=9 0.023651 3.996762 8 1634 +FAST f=17 a=10 2.754796 4.005059 8 1058 +FAST f=17 a=10 0.022537 4.005059 8 1058 +FAST f=18 a=1 4.779117 4.038214 8 242 +FAST f=18 a=1 0.048814 4.038214 8 242 +FAST f=18 a=2 3.829753 4.045768 8 722 +FAST f=18 a=2 0.036541 4.045768 8 722 +FAST f=18 a=3 3.495053 4.021497 8 770 +FAST f=18 a=3 0.032648 4.021497 8 770 +FAST f=18 a=4 3.221395 4.039623 8 770 +FAST f=18 a=4 0.027818 4.039623 8 770 +FAST f=18 a=5 3.059369 4.050414 8 530 +FAST f=18 a=5 0.026296 4.050414 8 530 +FAST f=18 a=6 3.019292 4.010714 6 962 +FAST f=18 a=6 0.031104 4.010714 6 962 +FAST f=18 a=7 2.949322 4.031439 6 770 +FAST f=18 a=7 0.030745 4.031439 6 770 +FAST f=18 a=8 2.876425 4.032088 6 386 +FAST f=18 a=8 0.027407 4.032088 6 386 +FAST f=18 a=9 2.850958 4.053372 8 674 +FAST f=18 a=9 0.023799 4.053372 8 674 +FAST f=18 a=10 2.884352 4.020148 8 1730 +FAST f=18 a=10 0.024401 4.020148 8 1730 +FAST f=19 a=1 4.815669 4.061203 8 674 +FAST f=19 a=1 0.051425 4.061203 8 674 +FAST f=19 a=2 3.951356 4.013822 8 1442 +FAST f=19 a=2 0.039968 4.013822 8 1442 +FAST f=19 a=3 3.554682 4.050425 8 722 +FAST f=19 a=3 0.032725 4.050425 8 722 +FAST f=19 a=4 3.242585 4.054677 8 722 +FAST f=19 a=4 0.028194 4.054677 8 722 +FAST f=19 a=5 3.105909 4.064524 8 818 +FAST f=19 a=5 0.02675 4.064524 8 818 +FAST f=19 a=6 3.059901 4.036857 8 1250 +FAST f=19 a=6 0.026396 4.036857 8 1250 +FAST f=19 a=7 3.016151 4.068234 6 770 +FAST f=19 a=7 0.031501 4.068234 6 770 +FAST f=19 a=8 2.962902 4.077509 8 530 +FAST f=19 a=8 0.023333 4.077509 8 530 +FAST f=19 a=9 2.899607 4.067328 8 530 +FAST f=19 a=9 0.024553 4.067328 8 530 +FAST f=19 a=10 2.950978 4.059901 8 434 +FAST f=19 a=10 0.023852 4.059901 8 434 +FAST f=20 a=1 5.259834 4.027579 8 1634 +FAST f=20 a=1 0.061123 4.027579 8 1634 +FAST f=20 a=2 4.382150 4.025093 8 1634 +FAST f=20 a=2 0.048009 4.025093 8 1634 +FAST f=20 a=3 4.104323 4.060842 8 530 +FAST f=20 a=3 0.040965 4.060842 8 530 +FAST f=20 a=4 3.853340 4.023504 6 914 +FAST f=20 a=4 0.041072 4.023504 6 914 +FAST f=20 a=5 3.728841 4.018089 6 1634 +FAST f=20 a=5 0.037469 4.018089 6 1634 +FAST f=20 a=6 3.683045 4.069138 8 578 +FAST f=20 a=6 0.028011 4.069138 8 578 +FAST f=20 a=7 3.726973 4.063160 8 722 +FAST f=20 a=7 0.028437 4.063160 8 722 +FAST f=20 a=8 3.555073 4.057690 8 386 +FAST f=20 a=8 0.027588 4.057690 8 386 +FAST f=20 a=9 3.551095 4.067253 8 482 +FAST f=20 a=9 0.025976 4.067253 8 482 +FAST f=20 a=10 3.490127 4.068518 8 530 +FAST f=20 a=10 0.025971 4.068518 8 530 +FAST f=21 a=1 7.343816 4.064945 8 770 +FAST f=21 a=1 0.085035 4.064945 8 770 +FAST f=21 a=2 5.930894 4.048206 8 386 +FAST f=21 a=2 0.067349 4.048206 8 386 +FAST f=21 a=3 6.770775 4.063417 8 578 +FAST f=21 a=3 0.077104 4.063417 8 578 +FAST f=21 a=4 6.889409 4.066761 8 626 +FAST f=21 a=4 0.0717 4.066761 8 626 +FAST f=21 a=5 6.714896 4.051813 8 914 +FAST f=21 a=5 0.071026 4.051813 8 914 +FAST f=21 a=6 6.539890 4.047263 8 1922 +FAST f=21 a=6 0.07127 4.047263 8 1922 +FAST f=21 a=7 6.511052 4.068373 8 482 +FAST f=21 a=7 0.065467 4.068373 8 482 +FAST f=21 a=8 6.458788 4.071597 8 482 +FAST f=21 a=8 0.063817 4.071597 8 482 +FAST f=21 a=9 6.377591 4.052905 8 434 +FAST f=21 a=9 0.063112 4.052905 8 434 +FAST f=21 a=10 6.360752 4.047773 8 530 +FAST f=21 a=10 0.063606 4.047773 8 530 +FAST f=22 a=1 10.523471 4.040812 8 962 +FAST f=22 a=1 0.14214 4.040812 8 962 +FAST f=22 a=2 9.454758 4.059396 8 914 +FAST f=22 a=2 0.118343 4.059396 8 914 +FAST f=22 a=3 9.043197 4.043019 8 1922 +FAST f=22 a=3 0.109798 4.043019 8 1922 +FAST f=22 a=4 8.716261 4.044819 8 770 +FAST f=22 a=4 0.099687 4.044819 8 770 +FAST f=22 a=5 8.529472 4.070576 8 530 +FAST f=22 a=5 0.093127 4.070576 8 530 +FAST f=22 a=6 8.424241 4.070565 8 722 +FAST f=22 a=6 0.093703 4.070565 8 722 +FAST f=22 a=7 8.403391 4.070591 8 578 +FAST f=22 a=7 0.089763 4.070591 8 578 +FAST f=22 a=8 8.285221 4.089171 8 530 +FAST f=22 a=8 0.087716 4.089171 8 530 +FAST f=22 a=9 8.282506 4.047470 8 722 +FAST f=22 a=9 0.089773 4.047470 8 722 +FAST f=22 a=10 8.241809 4.064151 8 818 +FAST f=22 a=10 0.090413 4.064151 8 818 +FAST f=23 a=1 12.389208 4.051635 6 530 +FAST f=23 a=1 0.147796 4.051635 6 530 +FAST f=23 a=2 11.300910 4.042835 6 914 +FAST f=23 a=2 0.133178 4.042835 6 914 +FAST f=23 a=3 10.879455 4.047415 8 626 +FAST f=23 a=3 0.129571 4.047415 8 626 +FAST f=23 a=4 10.522718 4.038269 6 914 +FAST f=23 a=4 0.118121 4.038269 6 914 +FAST f=23 a=5 10.348043 4.066884 8 434 +FAST f=23 a=5 0.112098 4.066884 8 434 +FAST f=23 a=6 10.238630 4.048635 8 1010 +FAST f=23 a=6 0.120281 4.048635 8 1010 +FAST f=23 a=7 10.213255 4.061809 8 530 +FAST f=23 a=7 0.1121 4.061809 8 530 +FAST f=23 a=8 10.107879 4.074104 8 818 +FAST f=23 a=8 0.116544 4.074104 8 818 +FAST f=23 a=9 10.063424 4.064811 8 674 +FAST f=23 a=9 0.109045 4.064811 8 674 +FAST f=23 a=10 10.035801 4.054918 8 530 +FAST f=23 a=10 0.108735 4.054918 8 530 +FAST f=24 a=1 14.963878 4.073490 8 722 +FAST f=24 a=1 0.206344 4.073490 8 722 +FAST f=24 a=2 13.833472 4.036100 8 962 +FAST f=24 a=2 0.17486 4.036100 8 962 +FAST f=24 a=3 13.404631 4.026281 6 1106 +FAST f=24 a=3 0.153961 4.026281 6 1106 +FAST f=24 a=4 13.041164 4.065448 8 674 +FAST f=24 a=4 0.155509 4.065448 8 674 +FAST f=24 a=5 12.879412 4.054636 8 674 +FAST f=24 a=5 0.148282 4.054636 8 674 +FAST f=24 a=6 12.773736 4.081376 8 530 +FAST f=24 a=6 0.142563 4.081376 8 530 +FAST f=24 a=7 12.711310 4.059834 8 770 +FAST f=24 a=7 0.149321 4.059834 8 770 +FAST f=24 a=8 12.635459 4.052050 8 1298 +FAST f=24 a=8 0.15095 4.052050 8 1298 +FAST f=24 a=9 12.558104 4.076516 8 722 +FAST f=24 a=9 0.144361 4.076516 8 722 +FAST f=24 a=10 10.661348 4.062137 8 818 +FAST f=24 a=10 0.108232 4.062137 8 818 + hg-changelog: -NODICT 0.000007 1.377613 -RANDOM 0.297345 2.097487 -LEGACY 2.633992 2.058907 -COVER 219.179786 2.189685 8 98 -COVER 6.620852 2.189685 8 98 -FAST15 47.635082 2.130794 6 386 -FAST15 0.321297 2.130794 6 386 -FAST16 43.837676 2.144845 8 194 -FAST16 0.312640 2.144845 8 194 -FAST17 49.349017 2.156099 8 242 -FAST17 0.348459 2.156099 8 242 -FAST18 51.153784 2.172439 6 98 -FAST18 0.353106 2.172439 6 98 -FAST19 52.627045 2.180321 6 98 -FAST19 0.390612 2.180321 6 98 -FAST20 63.748782 2.187431 6 98 -FAST20 0.489544 2.187431 6 98 -FAST21 68.709198 2.184185 6 146 -FAST21 0.530852 2.184185 6 146 -FAST22 68.491639 2.182830 6 98 -FAST22 0.645699 2.182830 6 98 -FAST23 72.558688 2.186399 8 98 -FAST23 0.593539 2.186399 8 98 -FAST24 76.137195 2.185608 6 98 -FAST24 0.680132 2.185608 6 98 +NODICT 0.000017 1.377590 +RANDOM 0.186171 2.097487 +LEGACY 1.670867 2.058907 +COVER 173.561948 2.189685 8 98 +COVER 4.811180 2.189685 8 98 +FAST f=15 a=1 18.685906 2.129682 8 434 +FAST f=15 a=1 0.173376 2.129682 8 434 +FAST f=15 a=2 12.928259 2.131890 8 482 +FAST f=15 a=2 0.102582 2.131890 8 482 +FAST f=15 a=3 11.132343 2.128027 8 386 +FAST f=15 a=3 0.077122 2.128027 8 386 +FAST f=15 a=4 10.120683 2.125797 8 434 +FAST f=15 a=4 0.065175 2.125797 8 434 +FAST f=15 a=5 9.479092 2.127697 8 386 +FAST f=15 a=5 0.057905 2.127697 8 386 +FAST f=15 a=6 9.159523 2.127132 8 1682 +FAST f=15 a=6 0.058604 2.127132 8 1682 +FAST f=15 a=7 8.724003 2.129914 8 434 +FAST f=15 a=7 0.0493 2.129914 8 434 +FAST f=15 a=8 8.595001 2.127137 8 338 +FAST f=15 a=8 0.0474 2.127137 8 338 +FAST f=15 a=9 8.356405 2.125512 8 482 +FAST f=15 a=9 0.046126 2.125512 8 482 +FAST f=15 a=10 8.207111 2.126066 8 338 +FAST f=15 a=10 0.043292 2.126066 8 338 +FAST f=16 a=1 18.464436 2.144040 8 242 +FAST f=16 a=1 0.172156 2.144040 8 242 +FAST f=16 a=2 12.844825 2.148171 8 194 +FAST f=16 a=2 0.099619 2.148171 8 194 +FAST f=16 a=3 11.082568 2.140837 8 290 +FAST f=16 a=3 0.079165 2.140837 8 290 +FAST f=16 a=4 10.066749 2.144405 8 386 +FAST f=16 a=4 0.068411 2.144405 8 386 +FAST f=16 a=5 9.501121 2.140720 8 386 +FAST f=16 a=5 0.061316 2.140720 8 386 +FAST f=16 a=6 9.179332 2.139478 8 386 +FAST f=16 a=6 0.056322 2.139478 8 386 +FAST f=16 a=7 8.849438 2.142412 8 194 +FAST f=16 a=7 0.050493 2.142412 8 194 +FAST f=16 a=8 8.810919 2.143454 8 434 +FAST f=16 a=8 0.051304 2.143454 8 434 +FAST f=16 a=9 8.553900 2.140339 8 194 +FAST f=16 a=9 0.047285 2.140339 8 194 +FAST f=16 a=10 8.398027 2.143130 8 386 +FAST f=16 a=10 0.046386 2.143130 8 386 +FAST f=17 a=1 18.644657 2.157192 8 98 +FAST f=17 a=1 0.173884 2.157192 8 98 +FAST f=17 a=2 13.071242 2.159830 8 146 +FAST f=17 a=2 0.10388 2.159830 8 146 +FAST f=17 a=3 11.332366 2.153654 6 194 +FAST f=17 a=3 0.08983 2.153654 6 194 +FAST f=17 a=4 10.362413 2.156813 8 242 +FAST f=17 a=4 0.070389 2.156813 8 242 +FAST f=17 a=5 9.808159 2.155098 6 338 +FAST f=17 a=5 0.072661 2.155098 6 338 +FAST f=17 a=6 9.451165 2.153845 6 146 +FAST f=17 a=6 0.064959 2.153845 6 146 +FAST f=17 a=7 9.163097 2.155424 6 242 +FAST f=17 a=7 0.064323 2.155424 6 242 +FAST f=17 a=8 9.047276 2.156640 8 242 +FAST f=17 a=8 0.053382 2.156640 8 242 +FAST f=17 a=9 8.807671 2.152396 8 146 +FAST f=17 a=9 0.049617 2.152396 8 146 +FAST f=17 a=10 8.649827 2.152370 8 146 +FAST f=17 a=10 0.047849 2.152370 8 146 +FAST f=18 a=1 18.809502 2.168116 8 98 +FAST f=18 a=1 0.175226 2.168116 8 98 +FAST f=18 a=2 13.756502 2.170870 6 242 +FAST f=18 a=2 0.119507 2.170870 6 242 +FAST f=18 a=3 12.059748 2.163094 6 98 +FAST f=18 a=3 0.093912 2.163094 6 98 +FAST f=18 a=4 11.410294 2.172372 8 98 +FAST f=18 a=4 0.073048 2.172372 8 98 +FAST f=18 a=5 10.560297 2.166388 8 98 +FAST f=18 a=5 0.065136 2.166388 8 98 +FAST f=18 a=6 10.071390 2.162672 8 98 +FAST f=18 a=6 0.059402 2.162672 8 98 +FAST f=18 a=7 10.084214 2.166624 6 194 +FAST f=18 a=7 0.073276 2.166624 6 194 +FAST f=18 a=8 9.953226 2.167454 8 98 +FAST f=18 a=8 0.053659 2.167454 8 98 +FAST f=18 a=9 8.982461 2.161593 6 146 +FAST f=18 a=9 0.05955 2.161593 6 146 +FAST f=18 a=10 8.986092 2.164373 6 242 +FAST f=18 a=10 0.059135 2.164373 6 242 +FAST f=19 a=1 18.908277 2.176021 8 98 +FAST f=19 a=1 0.177316 2.176021 8 98 +FAST f=19 a=2 13.471313 2.176103 8 98 +FAST f=19 a=2 0.106344 2.176103 8 98 +FAST f=19 a=3 11.571406 2.172812 8 98 +FAST f=19 a=3 0.083293 2.172812 8 98 +FAST f=19 a=4 10.632775 2.177770 6 146 +FAST f=19 a=4 0.079864 2.177770 6 146 +FAST f=19 a=5 10.030190 2.175574 6 146 +FAST f=19 a=5 0.07223 2.175574 6 146 +FAST f=19 a=6 9.717818 2.169997 8 98 +FAST f=19 a=6 0.060049 2.169997 8 98 +FAST f=19 a=7 9.397531 2.172770 8 146 +FAST f=19 a=7 0.057188 2.172770 8 146 +FAST f=19 a=8 9.281061 2.175822 8 98 +FAST f=19 a=8 0.053711 2.175822 8 98 +FAST f=19 a=9 9.165242 2.169849 6 146 +FAST f=19 a=9 0.059898 2.169849 6 146 +FAST f=19 a=10 9.048763 2.173394 8 98 +FAST f=19 a=10 0.049757 2.173394 8 98 +FAST f=20 a=1 21.166917 2.183923 6 98 +FAST f=20 a=1 0.205425 2.183923 6 98 +FAST f=20 a=2 15.642753 2.182349 6 98 +FAST f=20 a=2 0.135957 2.182349 6 98 +FAST f=20 a=3 14.053730 2.173544 6 98 +FAST f=20 a=3 0.11266 2.173544 6 98 +FAST f=20 a=4 15.270019 2.183656 8 98 +FAST f=20 a=4 0.107892 2.183656 8 98 +FAST f=20 a=5 15.497927 2.174661 6 98 +FAST f=20 a=5 0.100305 2.174661 6 98 +FAST f=20 a=6 13.973505 2.172391 8 98 +FAST f=20 a=6 0.087565 2.172391 8 98 +FAST f=20 a=7 14.083296 2.172443 8 98 +FAST f=20 a=7 0.078062 2.172443 8 98 +FAST f=20 a=8 12.560048 2.175581 8 98 +FAST f=20 a=8 0.070282 2.175581 8 98 +FAST f=20 a=9 13.078645 2.173975 6 146 +FAST f=20 a=9 0.081041 2.173975 6 146 +FAST f=20 a=10 12.823328 2.177778 8 98 +FAST f=20 a=10 0.074522 2.177778 8 98 +FAST f=21 a=1 29.825370 2.183057 6 98 +FAST f=21 a=1 0.334453 2.183057 6 98 +FAST f=21 a=2 29.476474 2.182752 8 98 +FAST f=21 a=2 0.286602 2.182752 8 98 +FAST f=21 a=3 25.937186 2.175867 8 98 +FAST f=21 a=3 0.17626 2.175867 8 98 +FAST f=21 a=4 20.413865 2.179780 8 98 +FAST f=21 a=4 0.206085 2.179780 8 98 +FAST f=21 a=5 20.541889 2.178328 6 146 +FAST f=21 a=5 0.199157 2.178328 6 146 +FAST f=21 a=6 21.090670 2.174443 6 146 +FAST f=21 a=6 0.190645 2.174443 6 146 +FAST f=21 a=7 20.221569 2.177384 6 146 +FAST f=21 a=7 0.184278 2.177384 6 146 +FAST f=21 a=8 20.322357 2.179456 6 98 +FAST f=21 a=8 0.178458 2.179456 6 98 +FAST f=21 a=9 20.683912 2.174396 6 146 +FAST f=21 a=9 0.190829 2.174396 6 146 +FAST f=21 a=10 20.840865 2.174905 8 98 +FAST f=21 a=10 0.172515 2.174905 8 98 +FAST f=22 a=1 36.822827 2.181612 6 98 +FAST f=22 a=1 0.437389 2.181612 6 98 +FAST f=22 a=2 30.616902 2.183142 8 98 +FAST f=22 a=2 0.324284 2.183142 8 98 +FAST f=22 a=3 28.472482 2.178130 8 98 +FAST f=22 a=3 0.236538 2.178130 8 98 +FAST f=22 a=4 25.847028 2.181878 8 98 +FAST f=22 a=4 0.263744 2.181878 8 98 +FAST f=22 a=5 27.095881 2.180775 8 98 +FAST f=22 a=5 0.24988 2.180775 8 98 +FAST f=22 a=6 25.939172 2.170916 8 98 +FAST f=22 a=6 0.240033 2.170916 8 98 +FAST f=22 a=7 27.064194 2.177849 8 98 +FAST f=22 a=7 0.242383 2.177849 8 98 +FAST f=22 a=8 25.140221 2.178216 8 98 +FAST f=22 a=8 0.237601 2.178216 8 98 +FAST f=22 a=9 25.505283 2.177455 6 146 +FAST f=22 a=9 0.223217 2.177455 6 146 +FAST f=22 a=10 24.529362 2.176705 6 98 +FAST f=22 a=10 0.222876 2.176705 6 98 +FAST f=23 a=1 39.127310 2.183006 6 98 +FAST f=23 a=1 0.417338 2.183006 6 98 +FAST f=23 a=2 32.468161 2.183524 6 98 +FAST f=23 a=2 0.351645 2.183524 6 98 +FAST f=23 a=3 31.577620 2.172604 6 98 +FAST f=23 a=3 0.319659 2.172604 6 98 +FAST f=23 a=4 30.129247 2.183932 6 98 +FAST f=23 a=4 0.307239 2.183932 6 98 +FAST f=23 a=5 29.103376 2.183529 6 146 +FAST f=23 a=5 0.285533 2.183529 6 146 +FAST f=23 a=6 29.776045 2.174367 8 98 +FAST f=23 a=6 0.276846 2.174367 8 98 +FAST f=23 a=7 28.940407 2.178022 6 146 +FAST f=23 a=7 0.274082 2.178022 6 146 +FAST f=23 a=8 29.256009 2.179462 6 98 +FAST f=23 a=8 0.26949 2.179462 6 98 +FAST f=23 a=9 29.347312 2.170407 8 98 +FAST f=23 a=9 0.265034 2.170407 8 98 +FAST f=23 a=10 29.140081 2.171762 8 98 +FAST f=23 a=10 0.259183 2.171762 8 98 +FAST f=24 a=1 44.871179 2.182115 6 98 +FAST f=24 a=1 0.509433 2.182115 6 98 +FAST f=24 a=2 38.694867 2.180549 8 98 +FAST f=24 a=2 0.406695 2.180549 8 98 +FAST f=24 a=3 38.363769 2.172821 8 98 +FAST f=24 a=3 0.359581 2.172821 8 98 +FAST f=24 a=4 36.580797 2.184142 8 98 +FAST f=24 a=4 0.340614 2.184142 8 98 +FAST f=24 a=5 33.125701 2.183301 8 98 +FAST f=24 a=5 0.324874 2.183301 8 98 +FAST f=24 a=6 34.776068 2.173019 6 146 +FAST f=24 a=6 0.340397 2.173019 6 146 +FAST f=24 a=7 34.417625 2.176561 6 146 +FAST f=24 a=7 0.308223 2.176561 6 146 +FAST f=24 a=8 35.470291 2.182161 6 98 +FAST f=24 a=8 0.307724 2.182161 6 98 +FAST f=24 a=9 34.927252 2.172682 6 146 +FAST f=24 a=9 0.300598 2.172682 6 146 +FAST f=24 a=10 33.238355 2.173395 6 98 +FAST f=24 a=10 0.249916 2.173395 6 98 + hg-manifest: -NODICT 0.000026 1.866385 -RANDOM 0.784554 2.309436 -LEGACY 10.193714 2.506977 -COVER 988.206583 2.582528 8 434 -COVER 39.726199 2.582528 8 434 -FAST15 168.388819 2.392920 6 1826 -FAST15 1.272178 2.392920 6 1826 -FAST16 161.822607 2.480762 6 1922 -FAST16 1.164908 2.480762 6 1922 -FAST17 157.688544 2.548285 6 1682 -FAST17 1.222439 2.548285 6 1682 -FAST18 154.529585 2.567634 6 386 -FAST18 1.217596 2.567634 6 386 -FAST19 160.244979 2.581653 8 338 -FAST19 1.282450 2.581653 8 338 -FAST20 191.503297 2.586881 8 194 -FAST20 2.009748 2.586881 8 194 -FAST21 226.389709 2.590051 6 242 -FAST21 2.494543 2.590051 6 242 -FAST22 217.859055 2.591376 6 194 -FAST22 2.295693 2.591376 6 194 -FAST23 236.819791 2.591131 8 434 -FAST23 2.744711 2.591131 8 434 -FAST24 269.187800 2.591548 6 290 -FAST24 2.923671 2.591548 6 290 +NODICT 0.000004 1.866377 +RANDOM 0.696346 2.309436 +LEGACY 7.064527 2.506977 +COVER 876.312865 2.582528 8 434 +COVER 35.684533 2.582528 8 434 +FAST f=15 a=1 76.618201 2.404013 8 1202 +FAST f=15 a=1 0.700722 2.404013 8 1202 +FAST f=15 a=2 49.213058 2.409248 6 1826 +FAST f=15 a=2 0.473393 2.409248 6 1826 +FAST f=15 a=3 41.753197 2.409677 8 1490 +FAST f=15 a=3 0.336848 2.409677 8 1490 +FAST f=15 a=4 38.648295 2.407996 8 1538 +FAST f=15 a=4 0.283952 2.407996 8 1538 +FAST f=15 a=5 36.144936 2.402895 8 1874 +FAST f=15 a=5 0.270128 2.402895 8 1874 +FAST f=15 a=6 35.484675 2.394873 8 1586 +FAST f=15 a=6 0.251637 2.394873 8 1586 +FAST f=15 a=7 34.280599 2.397311 8 1778 +FAST f=15 a=7 0.23984 2.397311 8 1778 +FAST f=15 a=8 32.122572 2.396089 6 1490 +FAST f=15 a=8 0.251508 2.396089 6 1490 +FAST f=15 a=9 29.909842 2.390092 6 1970 +FAST f=15 a=9 0.251233 2.390092 6 1970 +FAST f=15 a=10 30.102938 2.400086 6 1682 +FAST f=15 a=10 0.23688 2.400086 6 1682 +FAST f=16 a=1 67.750401 2.475460 6 1346 +FAST f=16 a=1 0.796035 2.475460 6 1346 +FAST f=16 a=2 52.812027 2.480860 6 1730 +FAST f=16 a=2 0.480384 2.480860 6 1730 +FAST f=16 a=3 44.179259 2.469304 8 1970 +FAST f=16 a=3 0.332657 2.469304 8 1970 +FAST f=16 a=4 37.612728 2.478208 6 1970 +FAST f=16 a=4 0.32498 2.478208 6 1970 +FAST f=16 a=5 35.056222 2.475568 6 1298 +FAST f=16 a=5 0.302824 2.475568 6 1298 +FAST f=16 a=6 34.713012 2.486079 8 1730 +FAST f=16 a=6 0.24755 2.486079 8 1730 +FAST f=16 a=7 33.713687 2.477180 6 1682 +FAST f=16 a=7 0.280358 2.477180 6 1682 +FAST f=16 a=8 31.571412 2.475418 8 1538 +FAST f=16 a=8 0.241241 2.475418 8 1538 +FAST f=16 a=9 31.608069 2.478263 8 1922 +FAST f=16 a=9 0.241764 2.478263 8 1922 +FAST f=16 a=10 31.358002 2.472263 8 1442 +FAST f=16 a=10 0.221661 2.472263 8 1442 +FAST f=17 a=1 66.185775 2.536085 6 1346 +FAST f=17 a=1 0.713549 2.536085 6 1346 +FAST f=17 a=2 50.365000 2.546105 8 1298 +FAST f=17 a=2 0.467846 2.546105 8 1298 +FAST f=17 a=3 42.712843 2.536250 8 1298 +FAST f=17 a=3 0.34047 2.536250 8 1298 +FAST f=17 a=4 39.514227 2.535555 8 1442 +FAST f=17 a=4 0.302989 2.535555 8 1442 +FAST f=17 a=5 35.189292 2.524925 8 1202 +FAST f=17 a=5 0.273451 2.524925 8 1202 +FAST f=17 a=6 35.791683 2.523466 8 1202 +FAST f=17 a=6 0.268261 2.523466 8 1202 +FAST f=17 a=7 37.416136 2.526625 6 1010 +FAST f=17 a=7 0.277558 2.526625 6 1010 +FAST f=17 a=8 37.084707 2.533274 6 1250 +FAST f=17 a=8 0.285104 2.533274 6 1250 +FAST f=17 a=9 34.183814 2.532765 8 1298 +FAST f=17 a=9 0.235133 2.532765 8 1298 +FAST f=17 a=10 31.149235 2.528722 8 1346 +FAST f=17 a=10 0.232679 2.528722 8 1346 +FAST f=18 a=1 72.942176 2.559857 6 386 +FAST f=18 a=1 0.718618 2.559857 6 386 +FAST f=18 a=2 51.690440 2.559572 8 290 +FAST f=18 a=2 0.403978 2.559572 8 290 +FAST f=18 a=3 45.344908 2.561040 8 962 +FAST f=18 a=3 0.357205 2.561040 8 962 +FAST f=18 a=4 39.804522 2.558446 8 1010 +FAST f=18 a=4 0.310526 2.558446 8 1010 +FAST f=18 a=5 38.134888 2.561811 8 626 +FAST f=18 a=5 0.273743 2.561811 8 626 +FAST f=18 a=6 35.091890 2.555518 8 722 +FAST f=18 a=6 0.260135 2.555518 8 722 +FAST f=18 a=7 34.639523 2.562938 8 290 +FAST f=18 a=7 0.234294 2.562938 8 290 +FAST f=18 a=8 36.076431 2.563567 8 1586 +FAST f=18 a=8 0.274075 2.563567 8 1586 +FAST f=18 a=9 36.376433 2.560950 8 722 +FAST f=18 a=9 0.240106 2.560950 8 722 +FAST f=18 a=10 32.624790 2.559340 8 578 +FAST f=18 a=10 0.234704 2.559340 8 578 +FAST f=19 a=1 70.513761 2.572441 8 194 +FAST f=19 a=1 0.726112 2.572441 8 194 +FAST f=19 a=2 59.263032 2.574560 8 482 +FAST f=19 a=2 0.451554 2.574560 8 482 +FAST f=19 a=3 51.509594 2.571546 6 194 +FAST f=19 a=3 0.393014 2.571546 6 194 +FAST f=19 a=4 55.393906 2.573386 8 482 +FAST f=19 a=4 0.38819 2.573386 8 482 +FAST f=19 a=5 43.201736 2.567589 8 674 +FAST f=19 a=5 0.292155 2.567589 8 674 +FAST f=19 a=6 42.911687 2.572666 6 434 +FAST f=19 a=6 0.303988 2.572666 6 434 +FAST f=19 a=7 44.687591 2.573613 6 290 +FAST f=19 a=7 0.308721 2.573613 6 290 +FAST f=19 a=8 37.372868 2.571039 6 194 +FAST f=19 a=8 0.287137 2.571039 6 194 +FAST f=19 a=9 36.074230 2.566473 6 482 +FAST f=19 a=9 0.280721 2.566473 6 482 +FAST f=19 a=10 33.731720 2.570306 8 194 +FAST f=19 a=10 0.224073 2.570306 8 194 +FAST f=20 a=1 79.670634 2.581146 6 290 +FAST f=20 a=1 0.899986 2.581146 6 290 +FAST f=20 a=2 58.827141 2.579782 8 386 +FAST f=20 a=2 0.602288 2.579782 8 386 +FAST f=20 a=3 51.289004 2.579627 8 722 +FAST f=20 a=3 0.446091 2.579627 8 722 +FAST f=20 a=4 47.711068 2.581508 8 722 +FAST f=20 a=4 0.473007 2.581508 8 722 +FAST f=20 a=5 47.402929 2.578062 6 434 +FAST f=20 a=5 0.497131 2.578062 6 434 +FAST f=20 a=6 54.797102 2.577365 8 482 +FAST f=20 a=6 0.515061 2.577365 8 482 +FAST f=20 a=7 51.370877 2.583050 8 386 +FAST f=20 a=7 0.402878 2.583050 8 386 +FAST f=20 a=8 51.437931 2.574875 6 242 +FAST f=20 a=8 0.453094 2.574875 6 242 +FAST f=20 a=9 44.105456 2.576700 6 242 +FAST f=20 a=9 0.456633 2.576700 6 242 +FAST f=20 a=10 44.447580 2.578305 8 338 +FAST f=20 a=10 0.409121 2.578305 8 338 +FAST f=21 a=1 113.031686 2.582449 6 242 +FAST f=21 a=1 1.456971 2.582449 6 242 +FAST f=21 a=2 97.700932 2.582124 8 194 +FAST f=21 a=2 1.072078 2.582124 8 194 +FAST f=21 a=3 96.563648 2.585479 8 434 +FAST f=21 a=3 0.949528 2.585479 8 434 +FAST f=21 a=4 90.597813 2.582366 6 386 +FAST f=21 a=4 0.76944 2.582366 6 386 +FAST f=21 a=5 86.815980 2.579043 8 434 +FAST f=21 a=5 0.858167 2.579043 8 434 +FAST f=21 a=6 91.235820 2.578378 8 530 +FAST f=21 a=6 0.684274 2.578378 8 530 +FAST f=21 a=7 84.392788 2.581243 8 386 +FAST f=21 a=7 0.814386 2.581243 8 386 +FAST f=21 a=8 82.052310 2.582547 8 338 +FAST f=21 a=8 0.822633 2.582547 8 338 +FAST f=21 a=9 74.696074 2.579319 8 194 +FAST f=21 a=9 0.811028 2.579319 8 194 +FAST f=21 a=10 76.211170 2.578766 8 290 +FAST f=21 a=10 0.809715 2.578766 8 290 +FAST f=22 a=1 138.976871 2.580478 8 194 +FAST f=22 a=1 1.748932 2.580478 8 194 +FAST f=22 a=2 120.164097 2.583633 8 386 +FAST f=22 a=2 1.333239 2.583633 8 386 +FAST f=22 a=3 111.986474 2.582566 6 194 +FAST f=22 a=3 1.305734 2.582566 6 194 +FAST f=22 a=4 108.548148 2.583068 6 194 +FAST f=22 a=4 1.314026 2.583068 6 194 +FAST f=22 a=5 103.173017 2.583495 6 290 +FAST f=22 a=5 1.228664 2.583495 6 290 +FAST f=22 a=6 108.421262 2.582349 8 530 +FAST f=22 a=6 1.076773 2.582349 8 530 +FAST f=22 a=7 103.284127 2.581022 8 386 +FAST f=22 a=7 1.112117 2.581022 8 386 +FAST f=22 a=8 96.330279 2.581073 8 290 +FAST f=22 a=8 1.109303 2.581073 8 290 +FAST f=22 a=9 97.651348 2.580075 6 194 +FAST f=22 a=9 0.933032 2.580075 6 194 +FAST f=22 a=10 101.660621 2.584886 8 194 +FAST f=22 a=10 0.796823 2.584886 8 194 +FAST f=23 a=1 159.322978 2.581474 6 242 +FAST f=23 a=1 2.015878 2.581474 6 242 +FAST f=23 a=2 134.331775 2.581619 8 194 +FAST f=23 a=2 1.545845 2.581619 8 194 +FAST f=23 a=3 127.724552 2.579888 6 338 +FAST f=23 a=3 1.444496 2.579888 6 338 +FAST f=23 a=4 126.077675 2.578137 6 242 +FAST f=23 a=4 1.364394 2.578137 6 242 +FAST f=23 a=5 124.914027 2.580843 8 338 +FAST f=23 a=5 1.116059 2.580843 8 338 +FAST f=23 a=6 122.874153 2.577637 6 338 +FAST f=23 a=6 1.164584 2.577637 6 338 +FAST f=23 a=7 123.099257 2.582715 6 386 +FAST f=23 a=7 1.354042 2.582715 6 386 +FAST f=23 a=8 122.026753 2.577681 8 194 +FAST f=23 a=8 1.210966 2.577681 8 194 +FAST f=23 a=9 121.164312 2.584599 6 290 +FAST f=23 a=9 1.174859 2.584599 6 290 +FAST f=23 a=10 117.462222 2.580358 8 194 +FAST f=23 a=10 1.075258 2.580358 8 194 +FAST f=24 a=1 169.539659 2.581642 6 194 +FAST f=24 a=1 1.916804 2.581642 6 194 +FAST f=24 a=2 160.539270 2.580421 6 290 +FAST f=24 a=2 1.71087 2.580421 6 290 +FAST f=24 a=3 155.455874 2.580449 6 242 +FAST f=24 a=3 1.60307 2.580449 6 242 +FAST f=24 a=4 147.630320 2.582953 6 338 +FAST f=24 a=4 1.396364 2.582953 6 338 +FAST f=24 a=5 133.767428 2.580589 6 290 +FAST f=24 a=5 1.19933 2.580589 6 290 +FAST f=24 a=6 146.437535 2.579453 8 194 +FAST f=24 a=6 1.385405 2.579453 8 194 +FAST f=24 a=7 147.227507 2.584155 8 386 +FAST f=24 a=7 1.48942 2.584155 8 386 +FAST f=24 a=8 138.005773 2.584115 8 194 +FAST f=24 a=8 1.352 2.584115 8 194 +FAST f=24 a=9 141.442625 2.582902 8 290 +FAST f=24 a=9 1.39647 2.582902 8 290 +FAST f=24 a=10 142.157446 2.582701 8 434 +FAST f=24 a=10 1.498889 2.582701 8 434 diff --git a/contrib/experimental_dict_builders/benchmarkDictBuilder/benchmark.c b/contrib/experimental_dict_builders/benchmarkDictBuilder/benchmark.c index d92e8d5cb..b19345692 100644 --- a/contrib/experimental_dict_builders/benchmarkDictBuilder/benchmark.c +++ b/contrib/experimental_dict_builders/benchmarkDictBuilder/benchmark.c @@ -5,7 +5,6 @@ #include #include #include "random.h" -#include "fastCover.h" #include "dictBuilder.h" #include "zstd_internal.h" /* includes zstd.h */ #include "io.h" @@ -149,7 +148,7 @@ double compressWithDict(sampleInfo *srcInfo, dictInfo* dInfo, int compressionLev /* Allocate dst with enough space to compress the maximum sized sample */ { size_t maxSampleSize = 0; - for (int i = 0; i < srcInfo->nbSamples; i++) { + for (i = 0; i < srcInfo->nbSamples; i++) { maxSampleSize = MAX(srcInfo->samplesSizes[i], maxSampleSize); } dstCapacity = ZSTD_compressBound(maxSampleSize); @@ -291,6 +290,9 @@ int main(int argCount, const char* argv[]) /* Initialize arguments to default values */ unsigned k = 200; unsigned d = 8; + unsigned f; + unsigned accel; + unsigned i; const unsigned cLevel = DEFAULT_CLEVEL; const unsigned dictID = 0; const unsigned maxDictSize = g_defaultMaxDictSize; @@ -305,7 +307,7 @@ int main(int argCount, const char* argv[]) const char** extendedFileList = NULL; /* Parse arguments */ - for (int i = 1; i < argCount; i++) { + for (i = 1; i < argCount; i++) { const char* argument = argv[i]; if (longCommandWArg(&argument, "in=")) { filenameTable[filenameIdx] = argument; @@ -375,6 +377,7 @@ int main(int argCount, const char* argv[]) /* for cover */ { + /* for cover (optimizing k and d) */ ZDICT_cover_params_t coverParam; memset(&coverParam, 0, sizeof(coverParam)); coverParam.zParams = zParams; @@ -388,6 +391,7 @@ int main(int argCount, const char* argv[]) goto _cleanup; } + /* for cover (with k and d provided) */ const int coverResult = benchmarkDictBuilder(srcInfo, maxDictSize, NULL, &coverParam, NULL, NULL); DISPLAYLEVEL(2, "k=%u\nd=%u\nsteps=%u\nsplit=%u\n", coverParam.k, coverParam.d, coverParam.steps, (unsigned)(coverParam.splitPoint * 100)); if(coverResult) { @@ -398,29 +402,34 @@ int main(int argCount, const char* argv[]) } /* for fastCover */ - for (unsigned f = 15; f < 25; f++){ + for (f = 15; f < 25; f++){ DISPLAYLEVEL(2, "current f is %u\n", f); - /* for fastCover (optimizing k and d) */ - ZDICT_fastCover_params_t fastParam; - memset(&fastParam, 0, sizeof(fastParam)); - fastParam.zParams = zParams; - fastParam.splitPoint = 1.0; - fastParam.f = f; - fastParam.steps = 40; - fastParam.nbThreads = 1; - const int fastOptResult = benchmarkDictBuilder(srcInfo, maxDictSize, NULL, NULL, NULL, &fastParam); - DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\n", fastParam.k, fastParam.d, fastParam.f, fastParam.steps, (unsigned)(fastParam.splitPoint * 100)); - if(fastOptResult) { - result = 1; - goto _cleanup; - } + for (accel = 1; accel < 11; accel++) { + DISPLAYLEVEL(2, "current accel is %u\n", accel); + /* for fastCover (optimizing k and d) */ + ZDICT_fastCover_params_t fastParam; + memset(&fastParam, 0, sizeof(fastParam)); + fastParam.zParams = zParams; + fastParam.f = f; + fastParam.steps = 40; + fastParam.nbThreads = 1; + fastParam.accel = accel; + const int fastOptResult = benchmarkDictBuilder(srcInfo, maxDictSize, NULL, NULL, NULL, &fastParam); + DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\naccel=%u\n", fastParam.k, fastParam.d, fastParam.f, fastParam.steps, (unsigned)(fastParam.splitPoint * 100), fastParam.accel); + if(fastOptResult) { + result = 1; + goto _cleanup; + } - /* for fastCover (with k and d provided) */ - const int fastResult = benchmarkDictBuilder(srcInfo, maxDictSize, NULL, NULL, NULL, &fastParam); - DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\n", fastParam.k, fastParam.d, fastParam.f, fastParam.steps, (unsigned)(fastParam.splitPoint * 100)); - if(fastResult) { - result = 1; - goto _cleanup; + /* for fastCover (with k and d provided) */ + for (i = 0; i < 5; i++) { + const int fastResult = benchmarkDictBuilder(srcInfo, maxDictSize, NULL, NULL, NULL, &fastParam); + DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\naccel=%u\n", fastParam.k, fastParam.d, fastParam.f, fastParam.steps, (unsigned)(fastParam.splitPoint * 100), fastParam.accel); + if(fastResult) { + result = 1; + goto _cleanup; + } + } } } diff --git a/contrib/experimental_dict_builders/fastCover/fastCover.c b/contrib/experimental_dict_builders/fastCover/fastCover.c index 84d841b10..02c155a81 100644 --- a/contrib/experimental_dict_builders/fastCover/fastCover.c +++ b/contrib/experimental_dict_builders/fastCover/fastCover.c @@ -197,7 +197,7 @@ static FASTCOVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx, bestSegment.end = newEnd; } { - /* Half the frequency of hash value of each dmer covered by the chosen segment. */ + /* Zero the frequency of hash value of each dmer covered by the chosen segment. */ U32 pos; for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) { const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, parameters.f, ctx->d); @@ -300,7 +300,7 @@ static int FASTCOVER_ctx_init(FASTCOVER_ctx_t *ctx, const void *samplesBuffer, if (totalSamplesSize < MAX(d, sizeof(U64)) || totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) { DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n", - (U32)(totalSamplesSize>>20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20)); + (U32)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20)); return 0; } /* Check if there are at least 5 training samples */ diff --git a/contrib/largeNbDicts/.gitignore b/contrib/largeNbDicts/.gitignore new file mode 100644 index 000000000..e77c4e496 --- /dev/null +++ b/contrib/largeNbDicts/.gitignore @@ -0,0 +1,2 @@ +# build artifacts +largeNbDicts diff --git a/contrib/largeNbDicts/Makefile b/contrib/largeNbDicts/Makefile new file mode 100644 index 000000000..624140fab --- /dev/null +++ b/contrib/largeNbDicts/Makefile @@ -0,0 +1,49 @@ +# ################################################################ +# Copyright (c) 2018-present, Yann Collet, Facebook, Inc. +# All rights reserved. +# +# This source code is licensed under both the BSD-style license (found in the +# LICENSE file in the root directory of this source tree) and the GPLv2 (found +# in the COPYING file in the root directory of this source tree). +# ################################################################ + +PROGDIR = ../../programs +LIBDIR = ../../lib + +LIBZSTD = $(LIBDIR)/libzstd.a + +CPPFLAGS+= -I$(LIBDIR) -I$(LIBDIR)/common -I$(LIBDIR)/dictBuilder -I$(PROGDIR) + +CFLAGS ?= -O3 +CFLAGS += -std=gnu99 +DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \ + -Wstrict-aliasing=1 -Wswitch-enum \ + -Wstrict-prototypes -Wundef -Wpointer-arith -Wformat-security \ + -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \ + -Wredundant-decls +CFLAGS += $(DEBUGFLAGS) $(MOREFLAGS) + + +default: largeNbDicts + +all : largeNbDicts + +largeNbDicts: bench.o datagen.o xxhash.o largeNbDicts.c $(LIBZSTD) + $(CC) $(CPPFLAGS) $(CFLAGS) $^ $(LDFLAGS) -o $@ + +.PHONY: $(LIBZSTD) +$(LIBZSTD): + $(MAKE) -C $(LIBDIR) libzstd.a + +bench.o : $(PROGDIR)/bench.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +datagen.o: $(PROGDIR)/datagen.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +xxhash.o : $(LIBDIR)/common/xxhash.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +clean: + $(RM) *.o + $(RM) largeNbDicts diff --git a/contrib/largeNbDicts/README.md b/contrib/largeNbDicts/README.md new file mode 100644 index 000000000..f29bcdfe8 --- /dev/null +++ b/contrib/largeNbDicts/README.md @@ -0,0 +1,25 @@ +largeNbDicts +===================== + +`largeNbDicts` is a benchmark test tool +dedicated to the specific scenario of +dictionary decompression using a very large number of dictionaries. +When dictionaries are constantly changing, they are always "cold", +suffering from increased latency due to cache misses. + +The tool is created in a bid to investigate performance for this scenario, +and experiment mitigation techniques. + +Command line : +``` +largeNbDicts [Options] filename(s) + +Options : +-r : recursively load all files in subdirectories (default: off) +-B# : split input into blocks of size # (default: no split) +-# : use compression level # (default: 3) +-D # : use # as a dictionary (default: create one) +-i# : nb benchmark rounds (default: 6) +--nbDicts=# : set nb of dictionaries to # (default: one per block) +-h : help (this text) +``` diff --git a/contrib/largeNbDicts/largeNbDicts.c b/contrib/largeNbDicts/largeNbDicts.c new file mode 100644 index 000000000..e0bc55380 --- /dev/null +++ b/contrib/largeNbDicts/largeNbDicts.c @@ -0,0 +1,806 @@ +/* + * Copyright (c) 2018-present, Yann Collet, Facebook, Inc. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +/* largeNbDicts + * This is a benchmark test tool + * dedicated to the specific case of dictionary decompression + * using a very large nb of dictionaries + * thus suffering latency from lots of cache misses. + * It's created in a bid to investigate performance and find optimizations. */ + + +/*--- Dependencies ---*/ + +#include /* size_t */ +#include /* malloc, free, abort */ +#include /* fprintf */ +#include /* assert */ + +#include "util.h" +#include "bench.h" +#define ZSTD_STATIC_LINKING_ONLY +#include "zstd.h" +#include "zdict.h" + + +/*--- Constants --- */ + +#define KB *(1<<10) +#define MB *(1<<20) + +#define BLOCKSIZE_DEFAULT 0 /* no slicing into blocks */ +#define DICTSIZE (4 KB) +#define CLEVEL_DEFAULT 3 + +#define BENCH_TIME_DEFAULT_S 6 +#define RUN_TIME_DEFAULT_MS 1000 +#define BENCH_TIME_DEFAULT_MS (BENCH_TIME_DEFAULT_S * RUN_TIME_DEFAULT_MS) + +#define DISPLAY_LEVEL_DEFAULT 3 + +#define BENCH_SIZE_MAX (1200 MB) + + +/*--- Macros ---*/ +#define CONTROL(c) { if (!(c)) abort(); } +#undef MIN +#define MIN(a,b) ((a) < (b) ? (a) : (b)) + + +/*--- Display Macros ---*/ + +#define DISPLAY(...) fprintf(stdout, __VA_ARGS__) +#define DISPLAYLEVEL(l, ...) { if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } } +static int g_displayLevel = DISPLAY_LEVEL_DEFAULT; /* 0 : no display, 1: errors, 2 : + result + interaction + warnings, 3 : + progression, 4 : + information */ + + +/*--- buffer_t ---*/ + +typedef struct { + void* ptr; + size_t size; + size_t capacity; +} buffer_t; + +static const buffer_t kBuffNull = { NULL, 0, 0 }; + +/* @return : kBuffNull if any error */ +static buffer_t createBuffer(size_t capacity) +{ + assert(capacity > 0); + void* const ptr = malloc(capacity); + if (ptr==NULL) return kBuffNull; + + buffer_t buffer; + buffer.ptr = ptr; + buffer.capacity = capacity; + buffer.size = 0; + return buffer; +} + +static void freeBuffer(buffer_t buff) +{ + free(buff.ptr); +} + + +static void fillBuffer_fromHandle(buffer_t* buff, FILE* f) +{ + size_t const readSize = fread(buff->ptr, 1, buff->capacity, f); + buff->size = readSize; +} + + +/* @return : kBuffNull if any error */ +static buffer_t createBuffer_fromFile(const char* fileName) +{ + U64 const fileSize = UTIL_getFileSize(fileName); + size_t const bufferSize = (size_t) fileSize; + + if (fileSize == UTIL_FILESIZE_UNKNOWN) return kBuffNull; + assert((U64)bufferSize == fileSize); /* check overflow */ + + { FILE* const f = fopen(fileName, "rb"); + if (f == NULL) return kBuffNull; + + buffer_t buff = createBuffer(bufferSize); + CONTROL(buff.ptr != NULL); + + fillBuffer_fromHandle(&buff, f); + CONTROL(buff.size == buff.capacity); + + fclose(f); /* do nothing specific if fclose() fails */ + return buff; + } +} + + +/* @return : kBuffNull if any error */ +static buffer_t +createDictionaryBuffer(const char* dictionaryName, + const void* srcBuffer, + const size_t* srcBlockSizes, unsigned nbBlocks, + size_t requestedDictSize) +{ + if (dictionaryName) { + DISPLAYLEVEL(3, "loading dictionary %s \n", dictionaryName); + return createBuffer_fromFile(dictionaryName); /* note : result might be kBuffNull */ + + } else { + + DISPLAYLEVEL(3, "creating dictionary, of target size %u bytes \n", + (unsigned)requestedDictSize); + void* const dictBuffer = malloc(requestedDictSize); + CONTROL(dictBuffer != NULL); + + size_t const dictSize = ZDICT_trainFromBuffer(dictBuffer, requestedDictSize, + srcBuffer, + srcBlockSizes, nbBlocks); + CONTROL(!ZSTD_isError(dictSize)); + + buffer_t result; + result.ptr = dictBuffer; + result.capacity = requestedDictSize; + result.size = dictSize; + return result; + } +} + + +/*! BMK_loadFiles() : + * Loads `buffer`, with content from files listed within `fileNamesTable`. + * Fills `buffer` entirely. + * @return : 0 on success, !=0 on error */ +static int loadFiles(void* buffer, size_t bufferSize, + size_t* fileSizes, + const char* const * fileNamesTable, unsigned nbFiles) +{ + size_t pos = 0, totalSize = 0; + + for (unsigned n=0; n 0); + void* const srcBuffer = malloc(loadedSize); + assert(srcBuffer != NULL); + + assert(nbFiles > 0); + size_t* const fileSizes = (size_t*)calloc(nbFiles, sizeof(*fileSizes)); + assert(fileSizes != NULL); + + /* Load input buffer */ + int const errorCode = loadFiles(srcBuffer, loadedSize, + fileSizes, + fileNamesTable, nbFiles); + assert(errorCode == 0); + + void** sliceTable = (void**)malloc(nbFiles * sizeof(*sliceTable)); + assert(sliceTable != NULL); + + char* const ptr = (char*)srcBuffer; + size_t pos = 0; + unsigned fileNb = 0; + for ( ; (pos < loadedSize) && (fileNb < nbFiles); fileNb++) { + sliceTable[fileNb] = ptr + pos; + pos += fileSizes[fileNb]; + } + assert(pos == loadedSize); + assert(fileNb == nbFiles); + + + buffer_t buffer; + buffer.ptr = srcBuffer; + buffer.capacity = loadedSize; + buffer.size = loadedSize; + + slice_collection_t slices; + slices.slicePtrs = sliceTable; + slices.capacities = fileSizes; + slices.nbSlices = nbFiles; + + buffer_collection_t bc; + bc.buffer = buffer; + bc.slices = slices; + return bc; +} + + + + +/*--- ddict_collection_t ---*/ + +typedef struct { + ZSTD_DDict** ddicts; + size_t nbDDict; +} ddict_collection_t; + +static const ddict_collection_t kNullDDictCollection = { NULL, 0 }; + +static void freeDDictCollection(ddict_collection_t ddictc) +{ + for (size_t dictNb=0; dictNb < ddictc.nbDDict; dictNb++) { + ZSTD_freeDDict(ddictc.ddicts[dictNb]); + } + free(ddictc.ddicts); +} + +/* returns .buffers=NULL if operation fails */ +static ddict_collection_t createDDictCollection(const void* dictBuffer, size_t dictSize, size_t nbDDict) +{ + ZSTD_DDict** const ddicts = malloc(nbDDict * sizeof(ZSTD_DDict*)); + assert(ddicts != NULL); + if (ddicts==NULL) return kNullDDictCollection; + for (size_t dictNb=0; dictNb < nbDDict; dictNb++) { + ddicts[dictNb] = ZSTD_createDDict(dictBuffer, dictSize); + assert(ddicts[dictNb] != NULL); + } + ddict_collection_t ddictc; + ddictc.ddicts = ddicts; + ddictc.nbDDict = nbDDict; + return ddictc; +} + + +/* mess with adresses, so that linear scanning dictionaries != linear address scanning */ +void shuffleDictionaries(ddict_collection_t dicts) +{ + size_t const nbDicts = dicts.nbDDict; + for (size_t r=0; rdctx, + dst, dstCapacity, + src, srcSize, + di->dictionaries.ddicts[di->dictNb]); + + di->dictNb = di->dictNb + 1; + if (di->dictNb >= di->nbDicts) di->dictNb = 0; + + return result; +} + + +static int benchMem(slice_collection_t dstBlocks, + slice_collection_t srcBlocks, + ddict_collection_t dictionaries, + int nbRounds) +{ + assert(dstBlocks.nbSlices == srcBlocks.nbSlices); + + unsigned const ms_per_round = RUN_TIME_DEFAULT_MS; + unsigned const total_time_ms = nbRounds * ms_per_round; + + double bestSpeed = 0.; + + BMK_timedFnState_t* const benchState = + BMK_createTimedFnState(total_time_ms, ms_per_round); + decompressInstructions di = createDecompressInstructions(dictionaries); + + for (;;) { + BMK_runOutcome_t const outcome = BMK_benchTimedFn(benchState, + decompress, &di, + NULL, NULL, + dstBlocks.nbSlices, + (const void* const *)srcBlocks.slicePtrs, srcBlocks.capacities, + dstBlocks.slicePtrs, dstBlocks.capacities, + NULL); + CONTROL(BMK_isSuccessful_runOutcome(outcome)); + + BMK_runTime_t const result = BMK_extract_runTime(outcome); + U64 const dTime_ns = result.nanoSecPerRun; + double const dTime_sec = (double)dTime_ns / 1000000000; + size_t const srcSize = result.sumOfReturn; + double const dSpeed_MBps = (double)srcSize / dTime_sec / (1 MB); + if (dSpeed_MBps > bestSpeed) bestSpeed = dSpeed_MBps; + DISPLAY("Decompression Speed : %.1f MB/s \r", bestSpeed); + fflush(stdout); + if (BMK_isCompleted_TimedFn(benchState)) break; + } + DISPLAY("\n"); + + freeDecompressInstructions(di); + BMK_freeTimedFnState(benchState); + + return 0; /* success */ +} + + +/*! bench() : + * fileName : file to load for benchmarking purpose + * dictionary : optional (can be NULL), file to load as dictionary, + * if none provided : will be calculated on the fly by the program. + * @return : 0 is success, 1+ otherwise */ +int bench(const char** fileNameTable, unsigned nbFiles, + const char* dictionary, + size_t blockSize, int clevel, + unsigned nbDictMax, unsigned nbBlocks, + int nbRounds) +{ + int result = 0; + + DISPLAYLEVEL(3, "loading %u files... \n", nbFiles); + buffer_collection_t const srcs = createBufferCollection_fromFiles(fileNameTable, nbFiles); + CONTROL(srcs.buffer.ptr != NULL); + buffer_t srcBuffer = srcs.buffer; + size_t const srcSize = srcBuffer.size; + DISPLAYLEVEL(3, "created src buffer of size %.1f MB \n", + (double)srcSize / (1 MB)); + + slice_collection_t const srcSlices = splitSlices(srcs.slices, blockSize, nbBlocks); + nbBlocks = (unsigned)(srcSlices.nbSlices); + DISPLAYLEVEL(3, "split input into %u blocks ", nbBlocks); + if (blockSize) + DISPLAYLEVEL(3, "of max size %u bytes ", (unsigned)blockSize); + DISPLAYLEVEL(3, "\n"); + + + size_t* const dstCapacities = malloc(nbBlocks * sizeof(*dstCapacities)); + CONTROL(dstCapacities != NULL); + size_t dstBufferCapacity = 0; + for (size_t bnb=0; bnb='0') && (**stringPtr <='9')) { + unsigned const max = (((unsigned)(-1)) / 10) - 1; + assert(result <= max); /* check overflow */ + result *= 10, result += **stringPtr - '0', (*stringPtr)++ ; + } + if ((**stringPtr=='K') || (**stringPtr=='M')) { + unsigned const maxK = ((unsigned)(-1)) >> 10; + assert(result <= maxK); /* check overflow */ + result <<= 10; + if (**stringPtr=='M') { + assert(result <= maxK); /* check overflow */ + result <<= 10; + } + (*stringPtr)++; /* skip `K` or `M` */ + if (**stringPtr=='i') (*stringPtr)++; + if (**stringPtr=='B') (*stringPtr)++; + } + return result; +} + +/** 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. + */ +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; +} + + +int usage(const char* exeName) +{ + DISPLAY (" \n"); + DISPLAY (" %s [Options] filename(s) \n", exeName); + DISPLAY (" \n"); + DISPLAY ("Options : \n"); + DISPLAY ("-r : recursively load all files in subdirectories (default: off) \n"); + DISPLAY ("-B# : split input into blocks of size # (default: no split) \n"); + DISPLAY ("-# : use compression level # (default: %u) \n", CLEVEL_DEFAULT); + DISPLAY ("-D # : use # as a dictionary (default: create one) \n"); + DISPLAY ("-i# : nb benchmark rounds (default: %u) \n", BENCH_TIME_DEFAULT_S); + DISPLAY ("--nbBlocks=#: use # blocks for bench (default: one per file) \n"); + DISPLAY ("--nbDicts=# : create # dictionaries for bench (default: one per block) \n"); + DISPLAY ("-h : help (this text) \n"); + return 0; +} + +int bad_usage(const char* exeName) +{ + DISPLAY (" bad usage : \n"); + usage(exeName); + return 1; +} + +int main (int argc, const char** argv) +{ + int recursiveMode = 0; + int nbRounds = BENCH_TIME_DEFAULT_S; + const char* const exeName = argv[0]; + + if (argc < 2) return bad_usage(exeName); + + const char** nameTable = (const char**)malloc(argc * sizeof(const char*)); + assert(nameTable != NULL); + unsigned nameIdx = 0; + + const char* dictionary = NULL; + int cLevel = CLEVEL_DEFAULT; + size_t blockSize = BLOCKSIZE_DEFAULT; + unsigned nbDicts = 0; /* determine nbDicts automatically: 1 dictionary per block */ + unsigned nbBlocks = 0; /* determine nbBlocks automatically, from source and blockSize */ + + for (int argNb = 1; argNb < argc ; argNb++) { + const char* argument = argv[argNb]; + if (!strcmp(argument, "-h")) { free(nameTable); return usage(exeName); } + if (!strcmp(argument, "-r")) { recursiveMode = 1; continue; } + if (!strcmp(argument, "-D")) { argNb++; assert(argNb < argc); dictionary = argv[argNb]; continue; } + if (longCommandWArg(&argument, "-i")) { nbRounds = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--dictionary=")) { dictionary = argument; continue; } + if (longCommandWArg(&argument, "-B")) { blockSize = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--blockSize=")) { blockSize = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--nbDicts=")) { nbDicts = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--nbBlocks=")) { nbBlocks = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--clevel=")) { cLevel = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "-")) { cLevel = readU32FromChar(&argument); continue; } + /* anything that's not a command is a filename */ + nameTable[nameIdx++] = argument; + } + + const char** filenameTable = nameTable; + unsigned nbFiles = nameIdx; + char* buffer_containing_filenames = NULL; + + if (recursiveMode) { +#ifndef UTIL_HAS_CREATEFILELIST + assert(0); /* missing capability, do not run */ +#endif + filenameTable = UTIL_createFileList(nameTable, nameIdx, &buffer_containing_filenames, &nbFiles, 1 /* follow_links */); + } + + int result = bench(filenameTable, nbFiles, dictionary, blockSize, cLevel, nbDicts, nbBlocks, nbRounds); + + free(buffer_containing_filenames); + free(nameTable); + + return result; +} diff --git a/contrib/seekable_format/examples/seekable_compression.c b/contrib/seekable_format/examples/seekable_compression.c index 9485bf26f..9a331a895 100644 --- a/contrib/seekable_format/examples/seekable_compression.c +++ b/contrib/seekable_format/examples/seekable_compression.c @@ -101,7 +101,7 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve free(buffOut); } -static const char* createOutFilename_orDie(const char* filename) +static char* createOutFilename_orDie(const char* filename) { size_t const inL = strlen(filename); size_t const outL = inL + 5; @@ -109,7 +109,7 @@ static const char* createOutFilename_orDie(const char* filename) memset(outSpace, 0, outL); strcat(outSpace, filename); strcat(outSpace, ".zst"); - return (const char*)outSpace; + return (char*)outSpace; } int main(int argc, const char** argv) { @@ -124,8 +124,9 @@ int main(int argc, const char** argv) { { const char* const inFileName = argv[1]; unsigned const frameSize = (unsigned)atoi(argv[2]); - const char* const outFileName = createOutFilename_orDie(inFileName); + char* const outFileName = createOutFilename_orDie(inFileName); compressFile_orDie(inFileName, outFileName, 5, frameSize); + free(outFileName); } return 0; diff --git a/contrib/seekable_format/examples/seekable_decompression.c b/contrib/seekable_format/examples/seekable_decompression.c index 9cd232922..7050e0fa5 100644 --- a/contrib/seekable_format/examples/seekable_decompression.c +++ b/contrib/seekable_format/examples/seekable_decompression.c @@ -84,7 +84,7 @@ static void fseek_orDie(FILE* file, long int offset, int origin) { } -static void decompressFile_orDie(const char* fname, unsigned startOffset, unsigned endOffset) +static void decompressFile_orDie(const char* fname, off_t startOffset, off_t endOffset) { FILE* const fin = fopen_orDie(fname, "rb"); FILE* const fout = stdout; @@ -129,8 +129,8 @@ int main(int argc, const char** argv) { const char* const inFilename = argv[1]; - unsigned const startOffset = (unsigned) atoi(argv[2]); - unsigned const endOffset = (unsigned) atoi(argv[3]); + off_t const startOffset = atoll(argv[2]); + off_t const endOffset = atoll(argv[3]); decompressFile_orDie(inFilename, startOffset, endOffset); } diff --git a/contrib/seekable_format/zstdseek_decompress.c b/contrib/seekable_format/zstdseek_decompress.c index b006ff834..b4c48754e 100644 --- a/contrib/seekable_format/zstdseek_decompress.c +++ b/contrib/seekable_format/zstdseek_decompress.c @@ -56,6 +56,7 @@ #include /* malloc, free */ #include /* FILE* */ +#include #define XXH_STATIC_LINKING_ONLY #define XXH_NAMESPACE ZSTD_ @@ -112,7 +113,7 @@ static int ZSTD_seekable_read_buff(void* opaque, void* buffer, size_t n) static int ZSTD_seekable_seek_buff(void* opaque, long long offset, int origin) { - buffWrapper_t* buff = (buffWrapper_t*) opaque; + buffWrapper_t* const buff = (buffWrapper_t*) opaque; unsigned long long newOffset; switch (origin) { case SEEK_SET: @@ -124,6 +125,8 @@ static int ZSTD_seekable_seek_buff(void* opaque, long long offset, int origin) case SEEK_END: newOffset = (unsigned long long)buff->size - offset; break; + default: + assert(0); /* not possible */ } if (newOffset > buff->size) { return -1; @@ -310,8 +313,8 @@ static size_t ZSTD_seekable_loadSeekTable(ZSTD_seekable* zs) /* compute cumulative positions */ for (; idx < numFrames; idx++) { if (pos + sizePerEntry > SEEKABLE_BUFF_SIZE) { - U32 const toRead = MIN(remaining, SEEKABLE_BUFF_SIZE); U32 const offset = SEEKABLE_BUFF_SIZE - pos; + U32 const toRead = MIN(remaining, SEEKABLE_BUFF_SIZE - offset); memmove(zs->inBuff, zs->inBuff + pos, offset); /* move any data we haven't read yet */ CHECK_IO(src.read(src.opaque, zs->inBuff+offset, toRead)); remaining -= toRead; diff --git a/doc/zstd_compression_format.md b/doc/zstd_compression_format.md index 0b79f959f..c57b58269 100644 --- a/doc/zstd_compression_format.md +++ b/doc/zstd_compression_format.md @@ -16,7 +16,7 @@ Distribution of this document is unlimited. ### Version -0.2.8 (30/05/18) +0.2.9 (05/09/18) Introduction @@ -1192,6 +1192,8 @@ Number_of_Bits = Weight ? (Max_Number_of_Bits + 1 - Weight) : 0 The last symbol's `Weight` is deduced from previously decoded ones, by completing to the nearest power of 2. This power of 2 gives `Max_Number_of_Bits`, the depth of the current tree. +`Max_Number_of_Bits` must be <= 11, +otherwise the representation is considered corrupted. __Example__ : Let's presume the following Huffman tree must be described : @@ -1216,12 +1218,12 @@ It gives the following series of weights : | `Weight` | 4 | 3 | 2 | 0 | 1 | The decoder will do the inverse operation : -having collected weights of literals from `0` to `4`, +having collected weights of literal symbols from `0` to `4`, it knows the last literal, `5`, is present with a non-zero weight. The weight of `5` can be determined by advancing to the next power of 2. The sum of `2^(Weight-1)` (excluding 0's) is : `8 + 4 + 2 + 0 + 1 = 15`. -Nearest power of 2 is 16. +Nearest larger power of 2 value is 16. Therefore, `Max_Number_of_Bits = 4` and `Weight[5] = 16-15 = 1`. #### Huffman Tree header @@ -1233,18 +1235,24 @@ which describes how the series of weights is encoded. the series of weights is compressed using FSE (see below). The length of the FSE-compressed series is equal to `headerByte` (0-127). -- if `headerByte` >= 128 : this is a direct representation, - where each `Weight` is written directly as a 4 bits field (0-15). - They are encoded forward, 2 weights to a byte with the first weight taking - the top four bits and the second taking the bottom four (e.g. the following - operations could be used to read the weights: - `Weight[0] = (Byte[0] >> 4), Weight[1] = (Byte[0] & 0xf)`, etc.). - The full representation occupies `Ceiling(Number_of_Symbols/2)` bytes, - meaning it uses only full bytes even if `Number_of_Symbols` is odd. - `Number_of_Symbols = headerByte - 127`. - Note that maximum `Number_of_Symbols` is 255-127 = 128. - If any literal has a value > 128, raw header mode is not possible. - In such case, it's necessary to use FSE compression. +- if `headerByte` >= 128 : + + the series of weights uses a direct representation, + where each `Weight` is encoded directly as a 4 bits field (0-15). + + They are encoded forward, 2 weights to a byte, + first weight taking the top four bits and second one taking the bottom four. + * e.g. the following operations could be used to read the weights: + `Weight[0] = (Byte[0] >> 4), Weight[1] = (Byte[0] & 0xf)`, etc. + + The full representation occupies `Ceiling(Number_of_Weights/2)` bytes, + meaning it uses only full bytes even if `Number_of_Weights` is odd. + + `Number_of_Weights = headerByte - 127`. + * Note that maximum `Number_of_Weights` is 255-127 = 128, + therefore, only up to 128 `Weight` can be encoded using direct representation. + * Since the last non-zero `Weight` is _not_ encoded, + this scheme is compatible with alphabet sizes of up to 129 symbols, + hence including literal symbol 128. + * If any literal symbol > 128 has a non-zero `Weight`, + direct representation is not possible. + In such case, it's necessary to use FSE compression. #### Finite State Entropy (FSE) compression of Huffman weights @@ -1621,6 +1629,7 @@ or at least provide a meaningful error code explaining for which reason it canno Version changes --------------- +- 0.2.9 : clarifications for huffman weights direct representation, by Ulrich Kunitz - 0.2.8 : clarifications for IETF RFC discuss - 0.2.7 : clarifications from IETF RFC review, by Vijay Gurbani and Nick Terrell - 0.2.6 : fixed an error in huffman example, by Ulrich Kunitz diff --git a/doc/zstd_manual.html b/doc/zstd_manual.html index 48b9ca7f2..e5aade5b2 100644 --- a/doc/zstd_manual.html +++ b/doc/zstd_manual.html @@ -35,22 +35,34 @@

Introduction

-  zstd, short for Zstandard, is a fast lossless compression algorithm,
-  targeting real-time compression scenarios at zlib-level and better compression ratios.
-  The zstd compression library provides in-memory compression and decompression functions.
-  The library supports compression levels from 1 up to ZSTD_maxCLevel() which is currently 22.
-  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
+  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting
+  real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression
+  functions.
+
+  The library supports regular compression levels from 1 up to ZSTD_maxCLevel(),
+  which is currently 22. Levels >= 20, labeled `--ultra`, should be used with
+  caution, as they require more memory. The library also offers negative
+  compression levels, which extend the range of speed vs. ratio preferences.
+  The lower the level, the faster the speed (at the cost of compression).
+
   Compression can be done in:
     - a single step (described as Simple API)
     - a single step, reusing a context (described as Explicit context)
     - unbounded multiple steps (described as Streaming compression)
-  The compression ratio achievable on small data can be highly improved using a dictionary in:
-    - a single step (described as Simple dictionary API)
-    - a single step, reusing a dictionary (described as Bulk-processing dictionary API)
 
-  Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
-  Advanced experimental APIs shall never be used with a dynamic library.
-  They are not "stable", their definition may change in the future. Only static linking is allowed.
+  The compression ratio achievable on small data can be highly improved using
+  a dictionary. Dictionary compression can be performed in:
+    - a single step (described as Simple dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing
+      dictionary API)
+
+  Advanced experimental functions can be accessed using
+  `#define ZSTD_STATIC_LINKING_ONLY` before including zstd.h.
+
+  Advanced experimental APIs should never be used with a dynamically-linked
+  library. They are not "stable"; their definitions or signatures may change in
+  the future. Only static linking is allowed.

Version


@@ -181,7 +193,8 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
 
  When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
   ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
   ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
-  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict 
+  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict
+  Note : A ZSTD_CDict can be created with an empty dictionary, but it is inefficient for small data. 
 


 
 
size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
@@ -195,7 +208,9 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
 
  Compression using a digested Dictionary.
   Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
   Note that compression level is decided during dictionary creation.
-  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) 
+  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no)
+  Note : ZSTD_compress_usingCDict() can be used with a ZSTD_CDict created from an empty dictionary.
+         But it is inefficient for small data, and it is recommended to use ZSTD_compressCCtx(). 
 


 
 
ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
@@ -965,16 +980,21 @@ size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx,
                            const void* prefix, size_t prefixSize,
                            ZSTD_dictContentType_e dictContentType);
 
  Reference a prefix (single-usage dictionary) for next compression job.
-  Decompression need same prefix to properly regenerate data.
-  Prefix is **only used once**. Tables are discarded at end of compression job (ZSTD_e_end).
+  Decompression will need same prefix to properly regenerate data.
+  Compressing with a prefix is similar in outcome as performing a diff and compressing it,
+  but performs much faster, especially during decompression (compression speed is tunable with compression level).
+  Note that prefix is **only used once**. Tables are discarded at end of compression job (ZSTD_e_end).
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
   Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
   Note 1 : Prefix buffer is referenced. It **must** outlive compression job.
            Its contain must remain unmodified up to end of compression (ZSTD_e_end).
-  Note 2 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+  Note 2 : If the intention is to diff some large src data blob with some prior version of itself,
+           ensure that the window size is large enough to contain the entire source.
+           See ZSTD_p_windowLog.
+  Note 3 : Referencing a prefix involves building tables, which are dependent on compression parameters.
            It's a CPU consuming operation, with non-negligible impact on latency.
            If there is a need to use same prefix multiple times, consider loadDictionary instead.
-  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+  Note 4 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
            Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode. 
 


 
@@ -1141,6 +1161,8 @@ size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx,
                         const void* prefix, size_t prefixSize,
                         ZSTD_dictContentType_e dictContentType);
 
  Reference a prefix (single-usage dictionary) for next compression job.
+  This is the reverse operation of ZSTD_CCtx_refPrefix(),
+  and must use the same prefix as the one used during compression.
   Prefix is **only used once**. Reference is discarded at end of frame.
   End of frame is reached when ZSTD_DCtx_decompress_generic() returns 0.
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
diff --git a/examples/multiple_streaming_compression.c b/examples/multiple_streaming_compression.c
index e395aefbc..4308a2e4d 100644
--- a/examples/multiple_streaming_compression.c
+++ b/examples/multiple_streaming_compression.c
@@ -158,7 +158,8 @@ int main(int argc, const char** argv)
     }
 
     freeResources(ress);
-    /* success */
+    free(ofnBuffer);
+
     printf("compressed %i files \n", argc-1);
 
     return 0;
diff --git a/examples/streaming_compression.c b/examples/streaming_compression.c
index f76364d84..9287ff398 100644
--- a/examples/streaming_compression.c
+++ b/examples/streaming_compression.c
@@ -73,7 +73,11 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve
     ZSTD_CStream* const cstream = ZSTD_createCStream();
     if (cstream==NULL) { fprintf(stderr, "ZSTD_createCStream() error \n"); exit(10); }
     size_t const initResult = ZSTD_initCStream(cstream, cLevel);
-    if (ZSTD_isError(initResult)) { fprintf(stderr, "ZSTD_initCStream() error : %s \n", ZSTD_getErrorName(initResult)); exit(11); }
+    if (ZSTD_isError(initResult)) {
+        fprintf(stderr, "ZSTD_initCStream() error : %s \n",
+                    ZSTD_getErrorName(initResult));
+        exit(11);
+    }
 
     size_t read, toRead = buffInSize;
     while( (read = fread_orDie(buffIn, toRead, fin)) ) {
@@ -81,7 +85,11 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve
         while (input.pos < input.size) {
             ZSTD_outBuffer output = { buffOut, buffOutSize, 0 };
             toRead = ZSTD_compressStream(cstream, &output , &input);   /* toRead is guaranteed to be <= ZSTD_CStreamInSize() */
-            if (ZSTD_isError(toRead)) { fprintf(stderr, "ZSTD_compressStream() error : %s \n", ZSTD_getErrorName(toRead)); exit(12); }
+            if (ZSTD_isError(toRead)) {
+                fprintf(stderr, "ZSTD_compressStream() error : %s \n",
+                                ZSTD_getErrorName(toRead));
+                exit(12);
+            }
             if (toRead > buffInSize) toRead = buffInSize;   /* Safely handle case when `buffInSize` is manually changed to a value < ZSTD_CStreamInSize()*/
             fwrite_orDie(buffOut, output.pos, fout);
         }
@@ -100,15 +108,15 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve
 }
 
 
-static const char* createOutFilename_orDie(const char* filename)
+static char* createOutFilename_orDie(const char* filename)
 {
     size_t const inL = strlen(filename);
     size_t const outL = inL + 5;
-    void* outSpace = malloc_orDie(outL);
+    void* const outSpace = malloc_orDie(outL);
     memset(outSpace, 0, outL);
     strcat(outSpace, filename);
     strcat(outSpace, ".zst");
-    return (const char*)outSpace;
+    return (char*)outSpace;
 }
 
 int main(int argc, const char** argv)
@@ -124,8 +132,10 @@ int main(int argc, const char** argv)
 
     const char* const inFilename = argv[1];
 
-    const char* const outFilename = createOutFilename_orDie(inFilename);
+    char* const outFilename = createOutFilename_orDie(inFilename);
     compressFile_orDie(inFilename, outFilename, 1);
 
+    free(outFilename);   /* not strictly required, since program execution stops there,
+                          * but some static analyzer main complain otherwise */
     return 0;
 }
diff --git a/lib/BUCK b/lib/BUCK
index dbe8885fc..bd93b082a 100644
--- a/lib/BUCK
+++ b/lib/BUCK
@@ -69,6 +69,7 @@ cxx_library(
     ]),
     headers=subdir_glob([
         ('dictBuilder', 'divsufsort.h'),
+        ('dictBuilder', 'cover.h'),
     ]),
     srcs=glob(['dictBuilder/*.c']),
     deps=[':common'],
diff --git a/lib/Makefile b/lib/Makefile
index 01689c6d5..cf8e45b0f 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -23,7 +23,7 @@ ifeq ($(OS),Windows_NT)   # MinGW assumed
 CPPFLAGS   += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
 endif
 CFLAGS  ?= -O3
-DEBUGFLAGS = -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
+DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
             -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
             -Wstrict-prototypes -Wundef -Wpointer-arith -Wformat-security \
             -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
diff --git a/lib/common/compiler.h b/lib/common/compiler.h
index 595bac204..07f875e4d 100644
--- a/lib/common/compiler.h
+++ b/lib/common/compiler.h
@@ -89,20 +89,37 @@
 #endif
 
 /* prefetch
- * can be disabled, by declaring NO_PREFETCH macro */
+ * can be disabled, by declaring NO_PREFETCH macro
+ * All prefetch invocations use a single default locality 2,
+ * generating instruction prefetcht1,
+ * which, according to Intel, means "load data into L2 cache".
+ * This is a good enough "middle ground" for the time being,
+ * though in theory, it would be better to specialize locality depending on data being prefetched.
+ * Tests could not determine any sensible difference based on locality value. */
 #if defined(NO_PREFETCH)
-#  define PREFETCH(ptr)   /* disabled */
+#  define PREFETCH(ptr)     (void)(ptr)  /* disabled */
 #else
 #  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
 #    include    /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-#    define PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
+#    define PREFETCH(ptr)   _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
 #  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
-#    define PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
+#    define PREFETCH(ptr)   __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
 #  else
-#    define PREFETCH(ptr)   /* disabled */
+#    define PREFETCH(ptr)   (void)(ptr)  /* disabled */
 #  endif
 #endif  /* NO_PREFETCH */
 
+#define CACHELINE_SIZE 64
+
+#define PREFETCH_AREA(p, s)  {            \
+    const char* const _ptr = (const char*)(p);  \
+    size_t const _size = (size_t)(s);     \
+    size_t _pos;                          \
+    for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) {  \
+        PREFETCH(_ptr + _pos);            \
+    }                                     \
+}
+
 /* disable warnings */
 #ifdef _MSC_VER    /* Visual Studio */
 #  include                     /* For Visual 2005 */
diff --git a/lib/common/cpu.h b/lib/common/cpu.h
index 88e0ebf44..eeb428ad5 100644
--- a/lib/common/cpu.h
+++ b/lib/common/cpu.h
@@ -36,7 +36,7 @@ MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
     U32 f1d = 0;
     U32 f7b = 0;
     U32 f7c = 0;
-#ifdef _MSC_VER
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
     int reg[4];
     __cpuid((int*)reg, 0);
     {
diff --git a/lib/common/zstd_internal.h b/lib/common/zstd_internal.h
index b4c1af53f..e75adfa61 100644
--- a/lib/common/zstd_internal.h
+++ b/lib/common/zstd_internal.h
@@ -79,8 +79,7 @@ static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
 static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
 
-#define ZSTD_FRAMEIDSIZE 4
-static const size_t ZSTD_frameIdSize = ZSTD_FRAMEIDSIZE;  /* magic number size */
+#define ZSTD_FRAMEIDSIZE 4   /* magic number size */
 
 #define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
 static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
@@ -193,6 +192,8 @@ typedef struct {
     BYTE* llCode;
     BYTE* mlCode;
     BYTE* ofCode;
+    size_t maxNbSeq;
+    size_t maxNbLit;
     U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
     U32   longLengthPos;
 } seqStore_t;
diff --git a/lib/compress/fse_compress.c b/lib/compress/fse_compress.c
index 70daae3bc..4075db217 100644
--- a/lib/compress/fse_compress.c
+++ b/lib/compress/fse_compress.c
@@ -83,7 +83,9 @@
  * wkspSize should be sized to handle worst case situation, which is `1< wkspSize) return ERROR(tableLog_tooLarge);
     tableU16[-2] = (U16) tableLog;
     tableU16[-1] = (U16) maxSymbolValue;
-    assert(tableLog < 16);   /* required for the threshold strategy to work */
+    assert(tableLog < 16);   /* required for threshold strategy to work */
 
     /* For explanations on how to distribute symbol values over the table :
-    *  http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+     * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+     #ifdef __clang_analyzer__
+     memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     #endif
 
     /* symbol start positions */
     {   U32 u;
@@ -124,13 +130,15 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
         U32 symbol;
         for (symbol=0; symbol<=maxSymbolValue; symbol++) {
             int nbOccurences;
-            for (nbOccurences=0; nbOccurences highThreshold) position = (position + step) & tableMask;   /* Low proba area */
+                while (position > highThreshold)
+                    position = (position + step) & tableMask;   /* Low proba area */
         }   }
 
-        if (position!=0) return ERROR(GENERIC);   /* Must have gone through all positions */
+        assert(position==0);  /* Must have initialized all positions */
     }
 
     /* Build table */
@@ -201,9 +209,10 @@ size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
-static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
-                                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
-                                       unsigned writeIsSafe)
+static size_t
+FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                         unsigned writeIsSafe)
 {
     BYTE* const ostart = (BYTE*) header;
     BYTE* out = ostart;
@@ -212,13 +221,12 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     const int tableSize = 1 << tableLog;
     int remaining;
     int threshold;
-    U32 bitStream;
-    int bitCount;
-    unsigned charnum = 0;
-    int previous0 = 0;
+    U32 bitStream = 0;
+    int bitCount = 0;
+    unsigned symbol = 0;
+    unsigned const alphabetSize = maxSymbolValue + 1;
+    int previousIs0 = 0;
 
-    bitStream = 0;
-    bitCount  = 0;
     /* Table Size */
     bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
     bitCount  += 4;
@@ -228,48 +236,53 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     threshold = tableSize;
     nbBits = tableLog+1;
 
-    while (remaining>1) {  /* stops at 1 */
-        if (previous0) {
-            unsigned start = charnum;
-            while (!normalizedCounter[charnum]) charnum++;
-            while (charnum >= start+24) {
+    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
+        if (previousIs0) {
+            unsigned start = symbol;
+            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
+            if (symbol == alphabetSize) break;   /* incorrect distribution */
+            while (symbol >= start+24) {
                 start+=24;
                 bitStream += 0xFFFFU << bitCount;
-                if ((!writeIsSafe) && (out > oend-2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend-2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE) bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out+=2;
                 bitStream>>=16;
             }
-            while (charnum >= start+3) {
+            while (symbol >= start+3) {
                 start+=3;
                 bitStream += 3 << bitCount;
                 bitCount += 2;
             }
-            bitStream += (charnum-start) << bitCount;
+            bitStream += (symbol-start) << bitCount;
             bitCount += 2;
             if (bitCount>16) {
-                if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend - 2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE)bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out += 2;
                 bitStream >>= 16;
                 bitCount -= 16;
         }   }
-        {   int count = normalizedCounter[charnum++];
-            int const max = (2*threshold-1)-remaining;
+        {   int count = normalizedCounter[symbol++];
+            int const max = (2*threshold-1) - remaining;
             remaining -= count < 0 ? -count : count;
             count++;   /* +1 for extra accuracy */
-            if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            if (count>=threshold)
+                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
             bitStream += count << bitCount;
             bitCount  += nbBits;
             bitCount  -= (count>=1; }
         }
         if (bitCount>16) {
-            if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            if ((!writeIsSafe) && (out > oend - 2))
+                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
             out[0] = (BYTE)bitStream;
             out[1] = (BYTE)(bitStream>>8);
             out += 2;
@@ -277,19 +290,23 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
             bitCount -= 16;
     }   }
 
+    if (remaining != 1)
+        return ERROR(GENERIC);  /* incorrect normalized distribution */
+    assert(symbol <= alphabetSize);
+
     /* flush remaining bitStream */
-    if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    if ((!writeIsSafe) && (out > oend - 2))
+        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
     out[0] = (BYTE)bitStream;
     out[1] = (BYTE)(bitStream>>8);
     out+= (bitCount+7) /8;
 
-    if (charnum > maxSymbolValue + 1) return ERROR(GENERIC);
-
     return (out-ostart);
 }
 
 
-size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
@@ -297,7 +314,7 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalized
     if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
         return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
 
-    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1);
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
 }
 
 
diff --git a/lib/compress/hist.h b/lib/compress/hist.h
index 788470da7..8b1991a90 100644
--- a/lib/compress/hist.h
+++ b/lib/compress/hist.h
@@ -50,7 +50,7 @@
 size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
                   const void* src, size_t srcSize);
 
-unsigned HIST_isError(size_t code);  /*< tells if a return value is an error code */
+unsigned HIST_isError(size_t code);  /**< tells if a return value is an error code */
 
 
 /* --- advanced histogram functions --- */
diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
index e09eef3d1..513156604 100644
--- a/lib/compress/zstd_compress.c
+++ b/lib/compress/zstd_compress.c
@@ -805,7 +805,7 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
         U32    const divider = (cParams.searchLength==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const tokenSpace = WILDCOPY_OVERLENGTH + blockSize + 11*maxNbSeq;
         size_t const entropySpace = HUF_WORKSPACE_SIZE;
         size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
         size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
@@ -949,33 +949,51 @@ typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
 /* ZSTD_sufficientBuff() :
  * check internal buffers exist for streaming if buffPol == ZSTDb_buffered .
  * Note : they are assumed to be correctly sized if ZSTD_equivalentCParams()==1 */
-static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t blockSize1,
+static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t maxNbSeq1,
+                            size_t maxNbLit1,
                             ZSTD_buffered_policy_e buffPol2,
                             ZSTD_compressionParameters cParams2,
                             U64 pledgedSrcSize)
 {
     size_t const windowSize2 = MAX(1, (size_t)MIN(((U64)1 << cParams2.windowLog), pledgedSrcSize));
     size_t const blockSize2 = MIN(ZSTD_BLOCKSIZE_MAX, windowSize2);
+    size_t const maxNbSeq2 = blockSize2 / ((cParams2.searchLength == 3) ? 3 : 4);
+    size_t const maxNbLit2 = blockSize2;
     size_t const neededBufferSize2 = (buffPol2==ZSTDb_buffered) ? windowSize2 + blockSize2 : 0;
-    DEBUGLOG(4, "ZSTD_sufficientBuff: is windowSize2=%u <= wlog1=%u",
-                (U32)windowSize2, cParams2.windowLog);
-    DEBUGLOG(4, "ZSTD_sufficientBuff: is blockSize2=%u <= blockSize1=%u",
-                (U32)blockSize2, (U32)blockSize1);
-    return (blockSize2 <= blockSize1) /* seqStore space depends on blockSize */
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is neededBufferSize2=%u <= bufferSize1=%u",
+                (U32)neededBufferSize2, (U32)bufferSize1);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is maxNbSeq2=%u <= maxNbSeq1=%u",
+                (U32)maxNbSeq2, (U32)maxNbSeq1);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is maxNbLit2=%u <= maxNbLit1=%u",
+                (U32)maxNbLit2, (U32)maxNbLit1);
+    return (maxNbLit2 <= maxNbLit1)
+         & (maxNbSeq2 <= maxNbSeq1)
          & (neededBufferSize2 <= bufferSize1);
 }
 
 /** Equivalence for resetCCtx purposes */
 static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
                                  ZSTD_CCtx_params params2,
-                                 size_t buffSize1, size_t blockSize1,
+                                 size_t buffSize1,
+                                 size_t maxNbSeq1, size_t maxNbLit1,
                                  ZSTD_buffered_policy_e buffPol2,
                                  U64 pledgedSrcSize)
 {
     DEBUGLOG(4, "ZSTD_equivalentParams: pledgedSrcSize=%u", (U32)pledgedSrcSize);
-    return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
-           ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams) &&
-           ZSTD_sufficientBuff(buffSize1, blockSize1, buffPol2, params2.cParams, pledgedSrcSize);
+    if (!ZSTD_equivalentCParams(params1.cParams, params2.cParams)) {
+      DEBUGLOG(4, "ZSTD_equivalentCParams() == 0");
+      return 0;
+    }
+    if (!ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams)) {
+      DEBUGLOG(4, "ZSTD_equivalentLdmParams() == 0");
+      return 0;
+    }
+    if (!ZSTD_sufficientBuff(buffSize1, maxNbSeq1, maxNbLit1, buffPol2,
+                             params2.cParams, pledgedSrcSize)) {
+      DEBUGLOG(4, "ZSTD_sufficientBuff() == 0");
+      return 0;
+    }
+    return 1;
 }
 
 static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
@@ -1103,8 +1121,9 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
     if (crp == ZSTDcrp_continue) {
         if (ZSTD_equivalentParams(zc->appliedParams, params,
-                                zc->inBuffSize, zc->blockSize,
-                                zbuff, pledgedSrcSize)) {
+                                  zc->inBuffSize,
+                                  zc->seqStore.maxNbSeq, zc->seqStore.maxNbLit,
+                                  zbuff, pledgedSrcSize)) {
             DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%zu)",
                         zc->appliedParams.cParams.windowLog, zc->blockSize);
             zc->workSpaceOversizedDuration += (zc->workSpaceOversizedDuration > 0);   /* if it was too large, it still is */
@@ -1125,7 +1144,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
         U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const tokenSpace = WILDCOPY_OVERLENGTH + blockSize + 11*maxNbSeq;
         size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
         size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
         size_t const matchStateSize = ZSTD_sizeof_matchState(¶ms.cParams, /* forCCtx */ 1);
@@ -1165,7 +1184,6 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                 if (zc->workSpace == NULL) return ERROR(memory_allocation);
                 zc->workSpaceSize = neededSpace;
                 zc->workSpaceOversizedDuration = 0;
-                ptr = zc->workSpace;
 
                 /* Statically sized space.
                  * entropyWorkspace never moves,
@@ -1216,13 +1234,18 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, ¶ms.cParams, crp, /* forCCtx */ 1);
 
         /* sequences storage */
+        zc->seqStore.maxNbSeq = maxNbSeq;
         zc->seqStore.sequencesStart = (seqDef*)ptr;
         ptr = zc->seqStore.sequencesStart + maxNbSeq;
         zc->seqStore.llCode = (BYTE*) ptr;
         zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
         zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
         zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
-        ptr = zc->seqStore.litStart + blockSize;
+        /* ZSTD_wildcopy() is used to copy into the literals buffer,
+         * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
+         */
+        zc->seqStore.maxNbLit = blockSize;
+        ptr = zc->seqStore.litStart + blockSize + WILDCOPY_OVERLENGTH;
 
         /* ldm bucketOffsets table */
         if (params.ldmParams.enableLdm) {
@@ -1341,8 +1364,7 @@ static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
         }
 
         /* copy dictionary offsets */
-        {
-            ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+        {   ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
             ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
             dstMatchState->window       = srcMatchState->window;
             dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
@@ -1666,6 +1688,7 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
     BYTE* const mlCodeTable = seqStorePtr->mlCode;
     U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     U32 u;
+    assert(nbSeq <= seqStorePtr->maxNbSeq);
     for (u=0; u= maxCSize) return 0;  /* block not compressed */
     }
 
-    /* We check that dictionaries have offset codes available for the first
-     * block. After the first block, the offcode table might not have large
-     * enough codes to represent the offsets in the data.
-     */
-    if (nextEntropy->fse.offcode_repeatMode == FSE_repeat_valid)
-        nextEntropy->fse.offcode_repeatMode = FSE_repeat_check;
-
     return cSize;
 }
 
@@ -2399,12 +2415,20 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
                                 &zc->appliedParams,
                                 dst, dstCapacity,
                                 srcSize, zc->entropyWorkspace, zc->bmi2);
-        if (ZSTD_isError(cSize) || cSize == 0) return cSize;
-        /* confirm repcodes and entropy tables */
-        {   ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
+        if (!ZSTD_isError(cSize) && cSize != 0) {
+            /* confirm repcodes and entropy tables */
+            ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
             zc->blockState.prevCBlock = zc->blockState.nextCBlock;
             zc->blockState.nextCBlock = tmp;
         }
+
+        /* We check that dictionaries have offset codes available for the first
+         * block. After the first block, the offcode table might not have large
+         * enough codes to represent the offsets in the data.
+         */
+        if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+            zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
         return cSize;
     }
 }
diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h
index d31542c69..1c95b7de9 100644
--- a/lib/compress/zstd_compress_internal.h
+++ b/lib/compress/zstd_compress_internal.h
@@ -314,8 +314,10 @@ MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const v
                pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offsetCode);
     }
 #endif
+    assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
     /* copy Literals */
-    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + 128 KB);
+    assert(seqStorePtr->maxNbLit <= 128 KB);
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
     ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
     seqStorePtr->lit += litLength;
 
diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
index 476cdc148..c4b9bb13b 100644
--- a/lib/compress/zstd_opt.c
+++ b/lib/compress/zstd_opt.c
@@ -970,7 +970,7 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
             U32 seqPos = cur;
 
             DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
-                        last_pos, cur);
+                        last_pos, cur); (void)last_pos;
             assert(storeEnd < ZSTD_OPT_NUM);
             DEBUGLOG(6, "last sequence copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
                         storeEnd, lastSequence.litlen, lastSequence.mlen, lastSequence.off);
diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c
index cc1af58e6..244690cdc 100644
--- a/lib/compress/zstdmt_compress.c
+++ b/lib/compress/zstdmt_compress.c
@@ -320,7 +320,8 @@ static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
 
 static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
 {
-    ZSTDMT_seqPool* seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    if (seqPool == NULL) return NULL;
     ZSTDMT_setNbSeq(seqPool, 0);
     return seqPool;
 }
diff --git a/lib/decompress/huf_decompress.c b/lib/decompress/huf_decompress.c
index a696261bd..83ecaff01 100644
--- a/lib/decompress/huf_decompress.c
+++ b/lib/decompress/huf_decompress.c
@@ -533,9 +533,9 @@ static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
     }
 }
 
-size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src,
-                             size_t srcSize, void* workSpace,
-                             size_t wkspSize)
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
+                       const void* src, size_t srcSize,
+                             void* workSpace, size_t wkspSize)
 {
     U32 tableLog, maxW, sizeOfSort, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index 8f4589d13..1382c9c7f 100644
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -40,7 +40,6 @@
 #  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_DEFAULTMAX) + 1)
 #endif
 
-
 /*!
  *  NO_FORWARD_PROGRESS_MAX :
  *  maximum allowed nb of calls to ZSTD_decompressStream() and ZSTD_decompress_generic()
@@ -52,11 +51,13 @@
 #  define ZSTD_NO_FORWARD_PROGRESS_MAX 16
 #endif
 
+
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
 #include       /* memcpy, memmove, memset */
-#include "cpu.h"
+#include "compiler.h"    /* prefetch */
+#include "cpu.h"         /* bmi2 */
 #include "mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
@@ -68,6 +69,9 @@
 #  include "zstd_legacy.h"
 #endif
 
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
+
 
 /*-*************************************
 *  Errors
@@ -110,11 +114,10 @@ typedef struct {
 #define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
 
 typedef struct {
-    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];
-    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];
-    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
     HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
-    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
     U32 rep[ZSTD_REP_NUM];
 } ZSTD_entropyDTables_t;
 
@@ -125,6 +128,7 @@ struct ZSTD_DCtx_s
     const ZSTD_seqSymbol* OFTptr;
     const HUF_DTable* HUFptr;
     ZSTD_entropyDTables_t entropy;
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];   /* space needed when building huffman tables */
     const void* previousDstEnd;   /* detect continuity */
     const void* prefixStart;      /* start of current segment */
     const void* virtualStart;     /* virtual start of previous segment if it was just before current one */
@@ -138,7 +142,6 @@ struct ZSTD_DCtx_s
     U32 fseEntropy;
     XXH64_state_t xxhState;
     size_t headerSize;
-    U32 dictID;
     ZSTD_format_e format;
     const BYTE* litPtr;
     ZSTD_customMem customMem;
@@ -147,9 +150,13 @@ struct ZSTD_DCtx_s
     size_t staticSize;
     int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
 
-    /* streaming */
+    /* dictionary */
     ZSTD_DDict* ddictLocal;
-    const ZSTD_DDict* ddict;
+    const ZSTD_DDict* ddict;     /* set by ZSTD_initDStream_usingDDict(), or ZSTD_DCtx_refDDict() */
+    U32 dictID;
+    int ddictIsCold;             /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
+
+    /* streaming */
     ZSTD_dStreamStage streamStage;
     char*  inBuff;
     size_t inBuffSize;
@@ -185,7 +192,7 @@ size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 static size_t ZSTD_startingInputLength(ZSTD_format_e format)
 {
     size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
-                    ZSTD_frameHeaderSize_prefix - ZSTD_frameIdSize :
+                    ZSTD_frameHeaderSize_prefix - ZSTD_FRAMEIDSIZE :
                     ZSTD_frameHeaderSize_prefix;
     ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
     /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
@@ -200,6 +207,8 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     dctx->ddict       = NULL;
     dctx->ddictLocal  = NULL;
+    dctx->dictEnd     = NULL;
+    dctx->ddictIsCold = 0;
     dctx->inBuff      = NULL;
     dctx->inBuffSize  = 0;
     dctx->outBuffSize = 0;
@@ -278,7 +287,7 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
 unsigned ZSTD_isFrame(const void* buffer, size_t size)
 {
-    if (size < ZSTD_frameIdSize) return 0;
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
     {   U32 const magic = MEM_readLE32(buffer);
         if (magic == ZSTD_MAGICNUMBER) return 1;
         if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
@@ -330,7 +339,9 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
+    memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
     if (srcSize < minInputSize) return minInputSize;
+    if (src==NULL) return ERROR(GENERIC);   /* invalid parameter */
 
     if ( (format != ZSTD_f_zstd1_magicless)
       && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
@@ -339,7 +350,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
             if (srcSize < ZSTD_skippableHeaderSize)
                 return ZSTD_skippableHeaderSize; /* magic number + frame length */
             memset(zfhPtr, 0, sizeof(*zfhPtr));
-            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_frameIdSize);
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
             zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
         }
@@ -451,7 +462,7 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
             size_t skippableSize;
             if (srcSize < ZSTD_skippableHeaderSize)
                 return ERROR(srcSize_wrong);
-            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_frameIdSize)
+            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_FRAMEIDSIZE)
                           + ZSTD_skippableHeaderSize;
             if (srcSize < skippableSize) {
                 return ZSTD_CONTENTSIZE_ERROR;
@@ -540,6 +551,7 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
 static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
                           const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
@@ -572,6 +584,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         case set_repeat:
             if (dctx->litEntropy==0) return ERROR(dictionary_corrupted);
             /* fall-through */
+
         case set_compressed:
             if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
             {   size_t lhSize, litSize, litCSize;
@@ -603,15 +616,20 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                 if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
                 if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
 
+                /* prefetch huffman table if cold */
+                if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
+                    PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
+                }
+
                 if (HUF_isError((litEncType==set_repeat) ?
                                     ( singleStream ?
                                         HUF_decompress1X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) :
                                         HUF_decompress4X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) ) :
                                     ( singleStream ?
                                         HUF_decompress1X1_DCtx_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                         dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2) :
+                                                                         dctx->workspace, sizeof(dctx->workspace), dctx->bmi2) :
                                         HUF_decompress4X_hufOnly_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                           dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2))))
+                                                                           dctx->workspace, sizeof(dctx->workspace), dctx->bmi2))))
                     return ERROR(corruption_detected);
 
                 dctx->litPtr = dctx->litBuffer;
@@ -883,7 +901,8 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
                                  symbolEncodingType_e type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
                                  const U32* baseValue, const U32* nbAdditionalBits,
-                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable)
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
+                                 int ddictIsCold, int nbSeq)
 {
     switch(type)
     {
@@ -902,6 +921,12 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
         return 0;
     case set_repeat:
         if (!flagRepeatTable) return ERROR(corruption_detected);
+        /* prefetch FSE table if used */
+        if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
+            const void* const pStart = *DTablePtr;
+            size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
+            PREFETCH_AREA(pStart, pSize);
+        }
         return 0;
     case set_compressed :
         {   U32 tableLog;
@@ -954,25 +979,25 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
     const BYTE* const istart = (const BYTE* const)src;
     const BYTE* const iend = istart + srcSize;
     const BYTE* ip = istart;
+    int nbSeq;
     DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
 
     /* check */
     if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
 
     /* SeqHead */
-    {   int nbSeq = *ip++;
-        if (!nbSeq) { *nbSeqPtr=0; return 1; }
-        if (nbSeq > 0x7F) {
-            if (nbSeq == 0xFF) {
-                if (ip+2 > iend) return ERROR(srcSize_wrong);
-                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
-            } else {
-                if (ip >= iend) return ERROR(srcSize_wrong);
-                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
-            }
+    nbSeq = *ip++;
+    if (!nbSeq) { *nbSeqPtr=0; return 1; }
+    if (nbSeq > 0x7F) {
+        if (nbSeq == 0xFF) {
+            if (ip+2 > iend) return ERROR(srcSize_wrong);
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+        } else {
+            if (ip >= iend) return ERROR(srcSize_wrong);
+            nbSeq = ((nbSeq-0x80)<<8) + *ip++;
         }
-        *nbSeqPtr = nbSeq;
     }
+    *nbSeqPtr = nbSeq;
 
     /* FSE table descriptors */
     if (ip+4 > iend) return ERROR(srcSize_wrong); /* minimum possible size */
@@ -986,7 +1011,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       LLtype, MaxLL, LLFSELog,
                                                       ip, iend-ip,
                                                       LL_base, LL_bits,
-                                                      LL_defaultDTable, dctx->fseEntropy);
+                                                      LL_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
             ip += llhSize;
         }
@@ -995,7 +1021,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       OFtype, MaxOff, OffFSELog,
                                                       ip, iend-ip,
                                                       OF_base, OF_bits,
-                                                      OF_defaultDTable, dctx->fseEntropy);
+                                                      OF_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
             ip += ofhSize;
         }
@@ -1004,12 +1031,23 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       MLtype, MaxML, MLFSELog,
                                                       ip, iend-ip,
                                                       ML_base, ML_bits,
-                                                      ML_defaultDTable, dctx->fseEntropy);
+                                                      ML_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
             if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
             ip += mlhSize;
         }
     }
 
+    /* prefetch dictionary content */
+    if (dctx->ddictIsCold) {
+        size_t const dictSize = (const char*)dctx->prefixStart - (const char*)dctx->virtualStart;
+        size_t const psmin = MIN(dictSize, (size_t)(64*nbSeq) /* heuristic */ );
+        size_t const pSize = MIN(psmin, 128 KB /* protection */ );
+        const void* const pStart = (const char*)dctx->dictEnd - pSize;
+        PREFETCH_AREA(pStart, pSize);
+        dctx->ddictIsCold = 0;
+    }
+
     return ip-istart;
 }
 
@@ -1676,7 +1714,8 @@ static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
     /* isLongOffset must be true if there are long offsets.
      * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
      * We don't expect that to be the case in 64-bit mode.
-     * In block mode, window size is not known, so we have to be conservative. (note: but it could be evaluated from current-lowLimit)
+     * In block mode, window size is not known, so we have to be conservative.
+     * (note: but it could be evaluated from current-lowLimit)
      */
     ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)));
     DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
@@ -1763,7 +1802,7 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
 #endif
     if ( (srcSize >= ZSTD_skippableHeaderSize)
       && (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START ) {
-        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize);
+        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_FRAMEIDSIZE);
     } else {
         const BYTE* ip = (const BYTE*)src;
         const BYTE* const ipstart = ip;
@@ -1797,7 +1836,6 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
         if (zfh.checksumFlag) {   /* Final frame content checksum */
             if (remainingSize < 4) return ERROR(srcSize_wrong);
             ip += 4;
-            remainingSize -= 4;
         }
 
         return ip - ipstart;
@@ -1885,9 +1923,6 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
     return op-ostart;
 }
 
-static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
-static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
-
 static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                         void* dst, size_t dstCapacity,
                                   const void* src, size_t srcSize,
@@ -1896,6 +1931,8 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
 {
     void* const dststart = dst;
     int moreThan1Frame = 0;
+
+    DEBUGLOG(5, "ZSTD_decompressMultiFrame");
     assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
 
     if (ddict) {
@@ -1932,7 +1969,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                 size_t skippableSize;
                 if (srcSize < ZSTD_skippableHeaderSize)
                     return ERROR(srcSize_wrong);
-                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize)
+                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_FRAMEIDSIZE)
                               + ZSTD_skippableHeaderSize;
                 if (srcSize < skippableSize) return ERROR(srcSize_wrong);
 
@@ -2057,7 +2094,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     case ZSTDds_getFrameHeaderSize :
         assert(src != NULL);
         if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
-            assert(srcSize >= ZSTD_frameIdSize);  /* to read skippable magic number */
+            assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
             if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
                 memcpy(dctx->headerBuffer, src, srcSize);
                 dctx->expected = ZSTD_skippableHeaderSize - srcSize;  /* remaining to load to get full skippable frame header */
@@ -2167,7 +2204,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         assert(src != NULL);
         assert(srcSize <= ZSTD_skippableHeaderSize);
         memcpy(dctx->headerBuffer + (ZSTD_skippableHeaderSize - srcSize), src, srcSize);   /* complete skippable header */
-        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_frameIdSize);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
         dctx->stage = ZSTDds_skipFrame;
         return 0;
 
@@ -2191,21 +2228,27 @@ static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dict
     return 0;
 }
 
-/* ZSTD_loadEntropy() :
- * dict : must point at beginning of a valid zstd dictionary
+/*! ZSTD_loadEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
  * @return : size of entropy tables read */
-static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
+static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy,
+                         const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
     if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
     dictPtr += 8;   /* skip header = magic + dictID */
 
-
-    {   size_t const hSize = HUF_readDTableX2_wksp(
-            entropy->hufTable, dictPtr, dictEnd - dictPtr,
-            entropy->workspace, sizeof(entropy->workspace));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, OFTable) == offsetof(ZSTD_entropyDTables_t, LLTable) + sizeof(entropy->LLTable));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, MLTable) == offsetof(ZSTD_entropyDTables_t, OFTable) + sizeof(entropy->OFTable));
+    ZSTD_STATIC_ASSERT(sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable) >= HUF_DECOMPRESS_WORKSPACE_SIZE);
+    {   void* const workspace = &entropy->LLTable;   /* use fse tables as temporary workspace; implies fse tables are grouped together */
+        size_t const workspaceSize = sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable);
+        size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize);
         if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
         dictPtr += hSize;
     }
@@ -2216,7 +2259,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
         if (offcodeMaxValue > MaxOff) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->OFTable,
+        ZSTD_buildFSETable( entropy->OFTable,
                             offcodeNCount, offcodeMaxValue,
                             OF_base, OF_bits,
                             offcodeLog);
@@ -2229,7 +2272,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (matchlengthMaxValue > MaxML) return ERROR(dictionary_corrupted);
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->MLTable,
+        ZSTD_buildFSETable( entropy->MLTable,
                             matchlengthNCount, matchlengthMaxValue,
                             ML_base, ML_bits,
                             matchlengthLog);
@@ -2242,7 +2285,7 @@ static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (litlengthMaxValue > MaxLL) return ERROR(dictionary_corrupted);
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
-        ZSTD_buildFSETable(entropy->LLTable,
+        ZSTD_buildFSETable( entropy->LLTable,
                             litlengthNCount, litlengthMaxValue,
                             LL_base, LL_bits,
                             litlengthLog);
@@ -2268,7 +2311,7 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
         if (magic != ZSTD_MAGIC_DICTIONARY) {
             return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
     }   }
-    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 
     /* load entropy tables */
     {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
@@ -2282,7 +2325,6 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
     return ZSTD_refDictContent(dctx, dict, dictSize);
 }
 
-/* Note : this function cannot fail */
 size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 {
     assert(dctx != NULL);
@@ -2328,42 +2370,53 @@ struct ZSTD_DDict_s {
 
 static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
 {
+    assert(ddict != NULL);
     return ddict->dictContent;
 }
 
 static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
 {
+    assert(ddict != NULL);
     return ddict->dictSize;
 }
 
-size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    CHECK_F( ZSTD_decompressBegin(dstDCtx) );
-    if (ddict) {   /* support begin on NULL */
-        dstDCtx->dictID = ddict->dictID;
-        dstDCtx->prefixStart = ddict->dictContent;
-        dstDCtx->virtualStart = ddict->dictContent;
-        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
-        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
+    DEBUGLOG(4, "ZSTD_decompressBegin_usingDDict");
+    assert(dctx != NULL);
+    if (ddict) {
+        dctx->ddictIsCold = (dctx->dictEnd != (const char*)ddict->dictContent + ddict->dictSize);
+        DEBUGLOG(4, "DDict is %s",
+                    dctx->ddictIsCold ? "~cold~" : "hot!");
+    }
+    CHECK_F( ZSTD_decompressBegin(dctx) );
+    if (ddict) {   /* NULL ddict is equivalent to no dictionary */
+        dctx->dictID = ddict->dictID;
+        dctx->prefixStart = ddict->dictContent;
+        dctx->virtualStart = ddict->dictContent;
+        dctx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+        dctx->previousDstEnd = dctx->dictEnd;
         if (ddict->entropyPresent) {
-            dstDCtx->litEntropy = 1;
-            dstDCtx->fseEntropy = 1;
-            dstDCtx->LLTptr = ddict->entropy.LLTable;
-            dstDCtx->MLTptr = ddict->entropy.MLTable;
-            dstDCtx->OFTptr = ddict->entropy.OFTable;
-            dstDCtx->HUFptr = ddict->entropy.hufTable;
-            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
-            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
-            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
+            dctx->litEntropy = 1;
+            dctx->fseEntropy = 1;
+            dctx->LLTptr = ddict->entropy.LLTable;
+            dctx->MLTptr = ddict->entropy.MLTable;
+            dctx->OFTptr = ddict->entropy.OFTable;
+            dctx->HUFptr = ddict->entropy.hufTable;
+            dctx->entropy.rep[0] = ddict->entropy.rep[0];
+            dctx->entropy.rep[1] = ddict->entropy.rep[1];
+            dctx->entropy.rep[2] = ddict->entropy.rep[2];
         } else {
-            dstDCtx->litEntropy = 0;
-            dstDCtx->fseEntropy = 0;
+            dctx->litEntropy = 0;
+            dctx->fseEntropy = 0;
         }
     }
     return 0;
 }
 
-static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e dictContentType)
+static size_t
+ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict,
+                         ZSTD_dictContentType_e dictContentType)
 {
     ddict->dictID = 0;
     ddict->entropyPresent = 0;
@@ -2381,10 +2434,12 @@ static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e
             return 0;   /* pure content mode */
         }
     }
-    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_frameIdSize);
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_FRAMEIDSIZE);
 
     /* load entropy tables */
-    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
+    CHECK_E( ZSTD_loadEntropy(&ddict->entropy,
+                              ddict->dictContent, ddict->dictSize),
+             dictionary_corrupted );
     ddict->entropyPresent = 1;
     return 0;
 }
@@ -2398,6 +2453,7 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
     if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
         ddict->dictBuffer = NULL;
         ddict->dictContent = dict;
+        if (!dict) dictSize = 0;
     } else {
         void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
         ddict->dictBuffer = internalBuffer;
@@ -2422,14 +2478,15 @@ ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
     if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
 
     {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
-        if (!ddict) return NULL;
+        if (ddict == NULL) return NULL;
         ddict->cMem = customMem;
-
-        if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, dictLoadMethod, dictContentType) )) {
-            ZSTD_freeDDict(ddict);
-            return NULL;
-        }
-
+        {   size_t const initResult = ZSTD_initDDict_internal(ddict,
+                                            dict, dictSize,
+                                            dictLoadMethod, dictContentType);
+            if (ZSTD_isError(initResult)) {
+                ZSTD_freeDDict(ddict);
+                return NULL;
+        }   }
         return ddict;
     }
 }
@@ -2456,23 +2513,25 @@ ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize
 
 
 const ZSTD_DDict* ZSTD_initStaticDDict(
-                                void* workspace, size_t workspaceSize,
+                                void* sBuffer, size_t sBufferSize,
                                 const void* dict, size_t dictSize,
                                 ZSTD_dictLoadMethod_e dictLoadMethod,
                                 ZSTD_dictContentType_e dictContentType)
 {
-    size_t const neededSpace =
-            sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
-    ZSTD_DDict* const ddict = (ZSTD_DDict*)workspace;
-    assert(workspace != NULL);
+    size_t const neededSpace = sizeof(ZSTD_DDict)
+                             + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)sBuffer;
+    assert(sBuffer != NULL);
     assert(dict != NULL);
-    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
-    if (workspaceSize < neededSpace) return NULL;
+    if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
+    if (sBufferSize < neededSpace) return NULL;
     if (dictLoadMethod == ZSTD_dlm_byCopy) {
         memcpy(ddict+1, dict, dictSize);  /* local copy */
         dict = ddict+1;
     }
-    if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, ZSTD_dlm_byRef, dictContentType) ))
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType) ))
         return NULL;
     return ddict;
 }
@@ -2510,7 +2569,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
 {
     if (dictSize < 8) return 0;
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
-    return MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+    return MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 }
 
 /*! ZSTD_getDictID_fromDDict() :
@@ -2586,12 +2645,15 @@ size_t ZSTD_freeDStream(ZSTD_DStream* zds)
 }
 
 
-/* *** Initialization *** */
+/* ***  Initialization  *** */
 
 size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
 size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
 
-size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
+                                   const void* dict, size_t dictSize,
+                                         ZSTD_dictLoadMethod_e dictLoadMethod,
+                                         ZSTD_dictContentType_e dictContentType)
 {
     if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
     ZSTD_freeDDict(dctx->ddictLocal);
@@ -2645,13 +2707,6 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
     return ZSTD_initDStream_usingDict(zds, NULL, 0);
 }
 
-size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
-{
-    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
-    dctx->ddict = ddict;
-    return 0;
-}
-
 /* ZSTD_initDStream_usingDDict() :
  * ddict will just be referenced, and must outlive decompression session
  * this function cannot fail */
@@ -2690,6 +2745,13 @@ size_t ZSTD_setDStreamParameter(ZSTD_DStream* dctx,
     return 0;
 }
 
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->ddict = ddict;
+    return 0;
+}
+
 size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
 {
     if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
@@ -2855,7 +2917,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             CHECK_F(ZSTD_decompressBegin_usingDDict(zds, zds->ddict));
 
             if ((MEM_readLE32(zds->headerBuffer) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
-                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_frameIdSize);
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
                 zds->stage = ZSTDds_skipFrame;
             } else {
                 CHECK_F(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
diff --git a/lib/dictBuilder/cover.c b/lib/dictBuilder/cover.c
index e32991652..6b4af69d2 100644
--- a/lib/dictBuilder/cover.c
+++ b/lib/dictBuilder/cover.c
@@ -29,6 +29,7 @@
 #include "mem.h" /* read */
 #include "pool.h"
 #include "threading.h"
+#include "cover.h"
 #include "zstd_internal.h" /* includes zstd.h */
 #ifndef ZDICT_STATIC_LINKING_ONLY
 #define ZDICT_STATIC_LINKING_ONLY
@@ -185,7 +186,7 @@ static void COVER_map_remove(COVER_map_t *map, U32 key) {
 }
 
 /**
- * Destroyes a map that is inited with COVER_map_init().
+ * Destroys a map that is inited with COVER_map_init().
  */
 static void COVER_map_destroy(COVER_map_t *map) {
   if (map->data) {
@@ -223,7 +224,7 @@ static COVER_ctx_t *g_ctx = NULL;
 /**
  * Returns the sum of the sample sizes.
  */
-static size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
   size_t sum = 0;
   unsigned i;
   for (i = 0; i < nbSamples; ++i) {
@@ -380,14 +381,6 @@ static void COVER_group(COVER_ctx_t *ctx, const void *group,
   ctx->suffix[dmerId] = freq;
 }
 
-/**
- * A segment is a range in the source as well as the score of the segment.
- */
-typedef struct {
-  U32 begin;
-  U32 end;
-  U32 score;
-} COVER_segment_t;
 
 /**
  * Selects the best segment in an epoch.
@@ -737,28 +730,65 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   }
 }
 
-/**
- * COVER_best_t is used for two purposes:
- * 1. Synchronizing threads.
- * 2. Saving the best parameters and dictionary.
- *
- * All of the methods except COVER_best_init() are thread safe if zstd is
- * compiled with multithreaded support.
- */
-typedef struct COVER_best_s {
-  ZSTD_pthread_mutex_t mutex;
-  ZSTD_pthread_cond_t cond;
-  size_t liveJobs;
-  void *dict;
-  size_t dictSize;
-  ZDICT_cover_params_t parameters;
-  size_t compressedSize;
-} COVER_best_t;
+
+
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                    const size_t *samplesSizes, const BYTE *samples,
+                                    size_t *offsets,
+                                    size_t nbTrainSamples, size_t nbSamples,
+                                    BYTE *const dict, size_t dictBufferCapacity) {
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Pointers */
+  ZSTD_CCtx *cctx;
+  ZSTD_CDict *cdict;
+  void *dst;
+  /* Local variables */
+  size_t dstCapacity;
+  size_t i;
+  /* Allocate dst with enough space to compress the maximum sized sample */
+  {
+    size_t maxSampleSize = 0;
+    i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+    for (; i < nbSamples; ++i) {
+      maxSampleSize = MAX(samplesSizes[i], maxSampleSize);
+    }
+    dstCapacity = ZSTD_compressBound(maxSampleSize);
+    dst = malloc(dstCapacity);
+  }
+  /* Create the cctx and cdict */
+  cctx = ZSTD_createCCtx();
+  cdict = ZSTD_createCDict(dict, dictBufferCapacity,
+                           parameters.zParams.compressionLevel);
+  if (!dst || !cctx || !cdict) {
+    goto _compressCleanup;
+  }
+  /* Compress each sample and sum their sizes (or error) */
+  totalCompressedSize = dictBufferCapacity;
+  i = parameters.splitPoint < 1.0 ? nbTrainSamples : 0;
+  for (; i < nbSamples; ++i) {
+    const size_t size = ZSTD_compress_usingCDict(
+        cctx, dst, dstCapacity, samples + offsets[i],
+        samplesSizes[i], cdict);
+    if (ZSTD_isError(size)) {
+      totalCompressedSize = ERROR(GENERIC);
+      goto _compressCleanup;
+    }
+    totalCompressedSize += size;
+  }
+_compressCleanup:
+  ZSTD_freeCCtx(cctx);
+  ZSTD_freeCDict(cdict);
+  if (dst) {
+    free(dst);
+  }
+  return totalCompressedSize;
+}
+
 
 /**
  * Initialize the `COVER_best_t`.
  */
-static void COVER_best_init(COVER_best_t *best) {
+void COVER_best_init(COVER_best_t *best) {
   if (best==NULL) return; /* compatible with init on NULL */
   (void)ZSTD_pthread_mutex_init(&best->mutex, NULL);
   (void)ZSTD_pthread_cond_init(&best->cond, NULL);
@@ -772,7 +802,7 @@ static void COVER_best_init(COVER_best_t *best) {
 /**
  * Wait until liveJobs == 0.
  */
-static void COVER_best_wait(COVER_best_t *best) {
+void COVER_best_wait(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -786,7 +816,7 @@ static void COVER_best_wait(COVER_best_t *best) {
 /**
  * Call COVER_best_wait() and then destroy the COVER_best_t.
  */
-static void COVER_best_destroy(COVER_best_t *best) {
+void COVER_best_destroy(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -802,7 +832,7 @@ static void COVER_best_destroy(COVER_best_t *best) {
  * Called when a thread is about to be launched.
  * Increments liveJobs.
  */
-static void COVER_best_start(COVER_best_t *best) {
+void COVER_best_start(COVER_best_t *best) {
   if (!best) {
     return;
   }
@@ -816,7 +846,7 @@ static void COVER_best_start(COVER_best_t *best) {
  * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
  * If this dictionary is the best so far save it and its parameters.
  */
-static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
                               ZDICT_cover_params_t parameters, void *dict,
                               size_t dictSize) {
   if (!best) {
@@ -847,10 +877,10 @@ static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
       best->parameters = parameters;
       best->compressedSize = compressedSize;
     }
-    ZSTD_pthread_mutex_unlock(&best->mutex);
     if (liveJobs == 0) {
       ZSTD_pthread_cond_broadcast(&best->cond);
     }
+    ZSTD_pthread_mutex_unlock(&best->mutex);
   }
 }
 
@@ -904,51 +934,10 @@ static void COVER_tryParameters(void *opaque) {
     }
   }
   /* Check total compressed size */
-  {
-    /* Pointers */
-    ZSTD_CCtx *cctx;
-    ZSTD_CDict *cdict;
-    void *dst;
-    /* Local variables */
-    size_t dstCapacity;
-    size_t i;
-    /* Allocate dst with enough space to compress the maximum sized sample */
-    {
-      size_t maxSampleSize = 0;
-      i = parameters.splitPoint < 1.0 ? ctx->nbTrainSamples : 0;
-      for (; i < ctx->nbSamples; ++i) {
-        maxSampleSize = MAX(ctx->samplesSizes[i], maxSampleSize);
-      }
-      dstCapacity = ZSTD_compressBound(maxSampleSize);
-      dst = malloc(dstCapacity);
-    }
-    /* Create the cctx and cdict */
-    cctx = ZSTD_createCCtx();
-    cdict = ZSTD_createCDict(dict, dictBufferCapacity,
-                             parameters.zParams.compressionLevel);
-    if (!dst || !cctx || !cdict) {
-      goto _compressCleanup;
-    }
-    /* Compress each sample and sum their sizes (or error) */
-    totalCompressedSize = dictBufferCapacity;
-    i = parameters.splitPoint < 1.0 ? ctx->nbTrainSamples : 0;
-    for (; i < ctx->nbSamples; ++i) {
-      const size_t size = ZSTD_compress_usingCDict(
-          cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
-          ctx->samplesSizes[i], cdict);
-      if (ZSTD_isError(size)) {
-        totalCompressedSize = ERROR(GENERIC);
-        goto _compressCleanup;
-      }
-      totalCompressedSize += size;
-    }
-  _compressCleanup:
-    ZSTD_freeCCtx(cctx);
-    ZSTD_freeCDict(cdict);
-    if (dst) {
-      free(dst);
-    }
-  }
+  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
+                                                       ctx->samples, ctx->offsets,
+                                                       ctx->nbTrainSamples, ctx->nbSamples,
+                                                       dict, dictBufferCapacity);
 
 _cleanup:
   COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
diff --git a/lib/dictBuilder/cover.h b/lib/dictBuilder/cover.h
new file mode 100644
index 000000000..82e2e1cea
--- /dev/null
+++ b/lib/dictBuilder/cover.h
@@ -0,0 +1,83 @@
+#include   /* fprintf */
+#include  /* malloc, free, qsort */
+#include  /* memset */
+#include    /* clock */
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+/**
+ * COVER_best_t is used for two purposes:
+ * 1. Synchronizing threads.
+ * 2. Saving the best parameters and dictionary.
+ *
+ * All of the methods except COVER_best_init() are thread safe if zstd is
+ * compiled with multithreaded support.
+ */
+typedef struct COVER_best_s {
+  ZSTD_pthread_mutex_t mutex;
+  ZSTD_pthread_cond_t cond;
+  size_t liveJobs;
+  void *dict;
+  size_t dictSize;
+  ZDICT_cover_params_t parameters;
+  size_t compressedSize;
+} COVER_best_t;
+
+/**
+ * A segment is a range in the source as well as the score of the segment.
+ */
+typedef struct {
+  U32 begin;
+  U32 end;
+  U32 score;
+} COVER_segment_t;
+
+/**
+ *  Checks total compressed size of a dictionary
+ */
+size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
+                                      const size_t *samplesSizes, const BYTE *samples,
+                                      size_t *offsets,
+                                      size_t nbTrainSamples, size_t nbSamples,
+                                      BYTE *const dict, size_t dictBufferCapacity);
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) ;
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+void COVER_best_init(COVER_best_t *best);
+
+/**
+ * Wait until liveJobs == 0.
+ */
+void COVER_best_wait(COVER_best_t *best);
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+void COVER_best_destroy(COVER_best_t *best);
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+void COVER_best_start(COVER_best_t *best);
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+                       ZDICT_cover_params_t parameters, void *dict,
+                       size_t dictSize);
diff --git a/lib/dictBuilder/divsufsort.c b/lib/dictBuilder/divsufsort.c
index 60cceb088..ead922044 100644
--- a/lib/dictBuilder/divsufsort.c
+++ b/lib/dictBuilder/divsufsort.c
@@ -1637,7 +1637,7 @@ construct_SA(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else {
           assert(((s == 0) && (T[s] == c1)) || (s < 0));
@@ -1701,7 +1701,7 @@ construct_BWT(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else if(s != 0) {
           *j = ~s;
@@ -1785,7 +1785,7 @@ construct_BWT_indexes(const unsigned char *T, int *SA,
             if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
             k = SA + BUCKET_B(c2 = c0, c1);
           }
-          assert(k < j);
+          assert(k < j); assert(k != NULL);
           *k-- = s;
         } else if(s != 0) {
           *j = ~s;
diff --git a/lib/dictBuilder/fastcover.c b/lib/dictBuilder/fastcover.c
new file mode 100644
index 000000000..6ce8c8809
--- /dev/null
+++ b/lib/dictBuilder/fastcover.c
@@ -0,0 +1,701 @@
+/*-*************************************
+*  Dependencies
+***************************************/
+#include   /* fprintf */
+#include  /* malloc, free, qsort */
+#include  /* memset */
+#include    /* clock */
+
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "cover.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
+#define FASTCOVER_MAX_F 31
+#define FASTCOVER_MAX_ACCEL 10
+#define DEFAULT_SPLITPOINT 0.75
+#define DEFAULT_F 20
+#define DEFAULT_ACCEL 1
+
+
+/*-*************************************
+*  Console display
+***************************************/
+static int g_displayLevel = 2;
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+
+
+/*-*************************************
+* Hash Functions
+***************************************/
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+
+/**
+ * Hash the d-byte value pointed to by p and mod 2^f
+ */
+static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
+  if (d == 6) {
+    return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
+  }
+  return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
+}
+
+
+/*-*************************************
+* Acceleration
+***************************************/
+typedef struct {
+  unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
+  unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
+} FASTCOVER_accel_t;
+
+
+static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
+  { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
+  { 100, 0 },   /* accel = 1 */
+  { 50, 1 },   /* accel = 2 */
+  { 34, 2 },   /* accel = 3 */
+  { 25, 3 },   /* accel = 4 */
+  { 20, 4 },   /* accel = 5 */
+  { 17, 5 },   /* accel = 6 */
+  { 14, 6 },   /* accel = 7 */
+  { 13, 7 },   /* accel = 8 */
+  { 11, 8 },   /* accel = 9 */
+  { 10, 9 },   /* accel = 10 */
+};
+
+
+/*-*************************************
+* Context
+***************************************/
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
+  size_t nbDmers;
+  U32 *freqs;
+  unsigned d;
+  unsigned f;
+  FASTCOVER_accel_t accelParams;
+} FASTCOVER_ctx_t;
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of all dmers with hash value d.
+ * Let S_i be hash value of the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer with hash value d is in the dictionay we set F(d) = 0.
+ */
+static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
+                                              U32 *freqs, U32 begin, U32 end,
+                                              ZDICT_cover_params_t parameters,
+                                              U16* segmentFreqs) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 f = ctx->f;
+  const U32 dmersInK = k - d + 1;
+
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+
+  /* Reset the activeDmers in the segment */
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* Get hash value of current dmer */
+    const size_t index = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
+
+    /* Add frequency of this index to score if this is the first occurence of index in active segment */
+    if (segmentFreqs[index] == 0) {
+      activeSegment.score += freqs[index];
+    }
+    /* Increment end of segment and segmentFreqs*/
+    activeSegment.end += 1;
+    segmentFreqs[index] += 1;
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      /* Get hash value of the dmer to be eliminated from active segment */
+      const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+      segmentFreqs[delIndex] -= 1;
+      /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
+      if (segmentFreqs[delIndex] == 0) {
+        activeSegment.score -= freqs[delIndex];
+      }
+      /* Increment start of segment */
+      activeSegment.begin += 1;
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+
+  /* Zero out rest of segmentFreqs array */
+  while (activeSegment.begin < end) {
+    const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+    segmentFreqs[delIndex] -= 1;
+    activeSegment.begin += 1;
+  }
+
+  {
+    /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
+      freqs[i] = 0;
+    }
+  }
+
+  return bestSegment;
+}
+
+
+static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
+                                     size_t maxDictSize, unsigned f,
+                                     unsigned accel) {
+  /* k, d, and f are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* d has to be 6 or 8 */
+  if (parameters.d != 6 && parameters.d != 8) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  /* 0 < f <= FASTCOVER_MAX_F*/
+  if (f > FASTCOVER_MAX_F || f == 0) {
+    return 0;
+  }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
+    return 0;
+  }
+  /* 0 < accel <= 10 */
+  if (accel > 10 || accel == 0) {
+    return 0;
+  }
+  return 1;
+}
+
+
+/**
+ * Clean up a context initialized with `FASTCOVER_ctx_init()`.
+ */
+static void FASTCOVER_ctx_destroy(FASTCOVER_ctx_t *ctx) {
+  if (!ctx) {
+    return;
+  }
+
+  free(ctx->freqs);
+  ctx->freqs = NULL;
+
+  free(ctx->offsets);
+  ctx->offsets = NULL;
+}
+
+
+/**
+ * Calculate for frequency of hash value of each dmer in ctx->samples
+ */
+static void FASTCOVER_computeFrequency(U32 *freqs, FASTCOVER_ctx_t *ctx){
+  const unsigned f = ctx->f;
+  const unsigned d = ctx->d;
+  const unsigned skip = ctx->accelParams.skip;
+  const unsigned readLength = MAX(d, 8);
+  size_t start; /* start of current dmer */
+  size_t i;
+  for (i = 0; i < ctx->nbTrainSamples; i++) {
+    size_t currSampleStart = ctx->offsets[i];
+    size_t currSampleEnd = ctx->offsets[i+1];
+    start = currSampleStart;
+    while (start + readLength <= currSampleEnd) {
+      const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
+      freqs[dmerIndex]++;
+      start = start + skip + 1;
+    }
+  }
+}
+
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can used multiple
+ * times.
+ * Returns 1 on success or zero on error.
+ * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
+ */
+static int FASTCOVER_ctx_init(FASTCOVER_ctx_t *ctx, const void *samplesBuffer,
+                              const size_t *samplesSizes, unsigned nbSamples,
+                              unsigned d, double splitPoint, unsigned f,
+                              FASTCOVER_accel_t accelParams) {
+  const BYTE *const samples = (const BYTE *)samplesBuffer;
+  const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+  /* Split samples into testing and training sets */
+  const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+  const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+  const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+  const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
+  /* Checks */
+  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+      totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
+    DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                 (U32)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
+    return 0;
+  }
+  /* Check if there are at least 5 training samples */
+  if (nbTrainSamples < 5) {
+    DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
+    return 0;
+  }
+  /* Check if there's testing sample */
+  if (nbTestSamples < 1) {
+    DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
+    return 0;
+  }
+  /* Zero the context */
+  memset(ctx, 0, sizeof(*ctx));
+  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+               (U32)trainingSamplesSize);
+  DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+               (U32)testSamplesSize);
+
+  ctx->samples = samples;
+  ctx->samplesSizes = samplesSizes;
+  ctx->nbSamples = nbSamples;
+  ctx->nbTrainSamples = nbTrainSamples;
+  ctx->nbTestSamples = nbTestSamples;
+  ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
+  ctx->d = d;
+  ctx->f = f;
+  ctx->accelParams = accelParams;
+
+  /* The offsets of each file */
+  ctx->offsets = (size_t *)malloc((nbSamples + 1) * sizeof(size_t));
+  if (!ctx->offsets) {
+    DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
+    FASTCOVER_ctx_destroy(ctx);
+    return 0;
+  }
+
+  /* Fill offsets from the samplesSizes */
+  {
+    U32 i;
+    ctx->offsets[0] = 0;
+    for (i = 1; i <= nbSamples; ++i) {
+      ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+    }
+  }
+
+  /* Initialize frequency array of size 2^f */
+  ctx->freqs = (U32 *)calloc(((U64)1 << f), sizeof(U32));
+
+  DISPLAYLEVEL(2, "Computing frequencies\n");
+  FASTCOVER_computeFrequency(ctx->freqs, ctx);
+
+  return 1;
+}
+
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t FASTCOVER_buildDictionary(const FASTCOVER_ctx_t *ctx, U32 *freqs,
+                                        void *dictBuffer, size_t dictBufferCapacity,
+                                        ZDICT_cover_params_t parameters, U16* segmentFreqs){
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data up into epochs of equal size.
+   * We will select at least one segment from each epoch.
+   */
+  const U32 epochs = MAX(1, (U32)(dictBufferCapacity / parameters.k));
+  const U32 epochSize = (U32)(ctx->nbDmers / epochs);
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
+               epochSize);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs) {
+    const U32 epochBegin = (U32)(epoch * epochSize);
+    const U32 epochEnd = epochBegin + epochSize;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = FASTCOVER_selectSegment(
+        ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
+
+    /* If the segment covers no dmers, then we are out of content */
+    if (segment.score == 0) {
+      break;
+    }
+
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (U32)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+
+/**
+ * Parameters for FASTCOVER_tryParameters().
+ */
+typedef struct FASTCOVER_tryParameters_data_s {
+  const FASTCOVER_ctx_t *ctx;
+  COVER_best_t *best;
+  size_t dictBufferCapacity;
+  ZDICT_cover_params_t parameters;
+} FASTCOVER_tryParameters_data_t;
+
+
+/**
+ * Tries a set of parameters and updates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void FASTCOVER_tryParameters(void *opaque) {
+  /* Save parameters as local variables */
+  FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
+  const FASTCOVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Initialize array to keep track of frequency of dmer within activeSegment */
+  U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
+  /* Allocate space for hash table, dict, and freqs */
+  BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
+  if (!segmentFreqs || !dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
+  /* Build the dictionary */
+  {
+    const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
+                                                  parameters, segmentFreqs);
+    const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
+    dictBufferCapacity = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, nbFinalizeSamples, parameters.zParams);
+    if (ZDICT_isError(dictBufferCapacity)) {
+      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+      goto _cleanup;
+    }
+  }
+  /* Check total compressed size */
+  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
+                                                       ctx->samples, ctx->offsets,
+                                                       ctx->nbTrainSamples, ctx->nbSamples,
+                                                       dict, dictBufferCapacity);
+_cleanup:
+  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
+                    dictBufferCapacity);
+  free(data);
+  free(segmentFreqs);
+  free(dict);
+  free(freqs);
+}
+
+
+
+static void FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
+                                          ZDICT_cover_params_t *coverParams) {
+    coverParams->k = fastCoverParams.k;
+    coverParams->d = fastCoverParams.d;
+    coverParams->steps = fastCoverParams.steps;
+    coverParams->nbThreads = fastCoverParams.nbThreads;
+    coverParams->splitPoint = fastCoverParams.splitPoint;
+    coverParams->zParams = fastCoverParams.zParams;
+}
+
+
+static void FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
+                                          ZDICT_fastCover_params_t *fastCoverParams,
+                                          unsigned f, unsigned accel) {
+    fastCoverParams->k = coverParams.k;
+    fastCoverParams->d = coverParams.d;
+    fastCoverParams->steps = coverParams.steps;
+    fastCoverParams->nbThreads = coverParams.nbThreads;
+    fastCoverParams->splitPoint = coverParams.splitPoint;
+    fastCoverParams->f = f;
+    fastCoverParams->accel = accel;
+    fastCoverParams->zParams = coverParams.zParams;
+}
+
+
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_fastCover(
+    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
+    const size_t *samplesSizes, unsigned nbSamples, ZDICT_fastCover_params_t parameters) {
+    BYTE* const dict = (BYTE*)dictBuffer;
+    FASTCOVER_ctx_t ctx;
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* Initialize global data */
+    g_displayLevel = parameters.zParams.notificationLevel;
+    /* Assign splitPoint and f if not provided */
+    parameters.splitPoint = 1.0;
+    parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
+    parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
+    /* Convert to cover parameter */
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(parameters, &coverParams);
+    /* Checks */
+    if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
+                                   parameters.accel)) {
+      DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+      return ERROR(GENERIC);
+    }
+    if (nbSamples == 0) {
+      DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
+      return ERROR(GENERIC);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    /* Assign corresponding FASTCOVER_accel_t to accelParams*/
+    accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
+    /* Initialize context */
+    if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                            coverParams.d, parameters.splitPoint, parameters.f,
+                            accelParams)) {
+      DISPLAYLEVEL(1, "Failed to initialize context\n");
+      return ERROR(GENERIC);
+    }
+    /* Build the dictionary */
+    DISPLAYLEVEL(2, "Building dictionary\n");
+    {
+      /* Initialize array to keep track of frequency of dmer within activeSegment */
+      U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
+      const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
+                                                dictBufferCapacity, coverParams, segmentFreqs);
+      const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
+      const size_t dictionarySize = ZDICT_finalizeDictionary(
+          dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+          samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
+      if (!ZSTD_isError(dictionarySize)) {
+          DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                      (U32)dictionarySize);
+      }
+      FASTCOVER_ctx_destroy(&ctx);
+      free(segmentFreqs);
+      return dictionarySize;
+    }
+}
+
+
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(
+    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
+    const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_fastCover_params_t *parameters) {
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* constants */
+    const unsigned nbThreads = parameters->nbThreads;
+    const double splitPoint =
+        parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
+    const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+    const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+    const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+    const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+    const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+    const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+    const unsigned kIterations =
+        (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+    const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
+    const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
+    /* Local variables */
+    const int displayLevel = parameters->zParams.notificationLevel;
+    unsigned iteration = 1;
+    unsigned d;
+    unsigned k;
+    COVER_best_t best;
+    POOL_ctx *pool = NULL;
+    /* Checks */
+    if (splitPoint <= 0 || splitPoint > 1) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
+      return ERROR(GENERIC);
+    }
+    if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
+      return ERROR(GENERIC);
+    }
+    if (kMinK < kMaxD || kMaxK < kMinK) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
+      return ERROR(GENERIC);
+    }
+    if (nbSamples == 0) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
+      return ERROR(GENERIC);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    if (nbThreads > 1) {
+      pool = POOL_create(nbThreads, 1);
+      if (!pool) {
+        return ERROR(memory_allocation);
+      }
+    }
+    /* Initialization */
+    COVER_best_init(&best);
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(*parameters, &coverParams);
+    accelParams = FASTCOVER_defaultAccelParameters[accel];
+    /* Turn down global display level to clean up display at level 2 and below */
+    g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+    /* Loop through d first because each new value needs a new context */
+    LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                      kIterations);
+    for (d = kMinD; d <= kMaxD; d += 2) {
+      /* Initialize the context for this value of d */
+      FASTCOVER_ctx_t ctx;
+      LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+      if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams)) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return ERROR(GENERIC);
+      }
+      /* Loop through k reusing the same context */
+      for (k = kMinK; k <= kMaxK; k += kStepSize) {
+        /* Prepare the arguments */
+        FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
+            sizeof(FASTCOVER_tryParameters_data_t));
+        LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+        if (!data) {
+          LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+          COVER_best_destroy(&best);
+          FASTCOVER_ctx_destroy(&ctx);
+          POOL_free(pool);
+          return ERROR(GENERIC);
+        }
+        data->ctx = &ctx;
+        data->best = &best;
+        data->dictBufferCapacity = dictBufferCapacity;
+        data->parameters = coverParams;
+        data->parameters.k = k;
+        data->parameters.d = d;
+        data->parameters.splitPoint = splitPoint;
+        data->parameters.steps = kSteps;
+        data->parameters.zParams.notificationLevel = g_displayLevel;
+        /* Check the parameters */
+        if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
+                                       data->ctx->f, accel)) {
+          DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+          free(data);
+          continue;
+        }
+        /* Call the function and pass ownership of data to it */
+        COVER_best_start(&best);
+        if (pool) {
+          POOL_add(pool, &FASTCOVER_tryParameters, data);
+        } else {
+          FASTCOVER_tryParameters(data);
+        }
+        /* Print status */
+        LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                           (U32)((iteration * 100) / kIterations));
+        ++iteration;
+      }
+      COVER_best_wait(&best);
+      FASTCOVER_ctx_destroy(&ctx);
+    }
+    LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+    /* Fill the output buffer and parameters with output of the best parameters */
+    {
+      const size_t dictSize = best.dictSize;
+      if (ZSTD_isError(best.compressedSize)) {
+        const size_t compressedSize = best.compressedSize;
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return compressedSize;
+      }
+      FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
+      memcpy(dictBuffer, best.dict, dictSize);
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return dictSize;
+    }
+
+}
diff --git a/lib/dictBuilder/zdict.c b/lib/dictBuilder/zdict.c
index b8a51789a..eb1a5b669 100644
--- a/lib/dictBuilder/zdict.c
+++ b/lib/dictBuilder/zdict.c
@@ -698,7 +698,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     short litLengthNCount[MaxLL+1];
     U32 repOffset[MAXREPOFFSET];
     offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
-    EStats_ress_t esr;
+    EStats_ress_t esr = { NULL, NULL, NULL };
     ZSTD_parameters params;
     U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
     size_t pos = 0, errorCode;
@@ -863,8 +863,8 @@ _cleanup:
 
 size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
                           const void* customDictContent, size_t dictContentSize,
-                          const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                          ZDICT_params_t params)
+                          const void* samplesBuffer, const size_t* samplesSizes,
+                          unsigned nbSamples, ZDICT_params_t params)
 {
     size_t hSize;
 #define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
@@ -987,8 +987,10 @@ size_t ZDICT_trainFromBuffer_unsafe_legacy(
             U32 const pos = dictList[u].pos;
             U32 const length = dictList[u].length;
             U32 const printedLength = MIN(40, length);
-            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize))
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize)) {
+                free(dictList);
                 return ERROR(GENERIC);   /* should never happen */
+            }
             DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
                          u, length, pos, dictList[u].savings);
             ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
@@ -1078,17 +1080,17 @@ size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
 size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
 {
-    ZDICT_cover_params_t params;
+    ZDICT_fastCover_params_t params;
     DEBUGLOG(3, "ZDICT_trainFromBuffer");
     memset(¶ms, 0, sizeof(params));
     params.d = 8;
     params.steps = 4;
     /* Default to level 6 since no compression level information is available */
-    params.zParams.compressionLevel = 6;
+    params.zParams.compressionLevel = 3;
 #if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
     params.zParams.notificationLevel = DEBUGLEVEL;
 #endif
-    return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
+    return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity,
                                                samplesBuffer, samplesSizes, nbSamples,
                                                ¶ms);
 }
diff --git a/lib/dictBuilder/zdict.h b/lib/dictBuilder/zdict.h
index 4094669d1..d57d59f01 100644
--- a/lib/dictBuilder/zdict.h
+++ b/lib/dictBuilder/zdict.h
@@ -39,7 +39,8 @@ extern "C" {
 
 /*! ZDICT_trainFromBuffer():
  *  Train a dictionary from an array of samples.
- *  Redirect towards ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
+ *  Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4,
+ *  f=20, and accel=1.
  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
  *  The resulting dictionary will be saved into `dictBuffer`.
@@ -52,7 +53,8 @@ extern "C" {
  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
  */
 ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                                    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+                                    const void* samplesBuffer,
+                                    const size_t* samplesSizes, unsigned nbSamples);
 
 
 /*======   Helper functions   ======*/
@@ -84,12 +86,22 @@ typedef struct {
 typedef struct {
     unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
     unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
-    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
-    double splitPoint;           /* Percentage of samples used for training: the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
     ZDICT_params_t zParams;
 } ZDICT_cover_params_t;
 
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned f;                  /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */
+    unsigned accel;              /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
+    ZDICT_params_t zParams;
+} ZDICT_fastCover_params_t;
 
 /*! ZDICT_trainFromBuffer_cover():
  *  Train a dictionary from an array of samples using the COVER algorithm.
@@ -116,9 +128,9 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
  * dictionary constructed with those parameters is stored in `dictBuffer`.
  *
  * All of the parameters d, k, steps are optional.
- * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
  * if steps is zero it defaults to its default value.
- * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
  *
  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
  *           or an error code, which can be tested with ZDICT_isError().
@@ -130,6 +142,48 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
           ZDICT_cover_params_t* parameters);
 
+/*! ZDICT_trainFromBuffer_fastCover():
+ *  Train a dictionary from an array of samples using a modified version of COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  d and k are required.
+ *  All other parameters are optional, will use default values if not provided
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer_fastCover() requires about 1 bytes of memory for each input byte and additionally another 6 * 2^f bytes of memory .
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
+                    size_t dictBufferCapacity, const void *samplesBuffer,
+                    const size_t *samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t parameters);
+
+/*! ZDICT_optimizeTrainFromBuffer_fastCover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations (specifically, k and d combinations)
+ * and picks the best parameters. `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ * All of the parameters d, k, steps, f, and accel are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
+ * If f is zero, default value of 20 is used.
+ * If accel is zero, default value of 1 is used.
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *           or an error code, which can be tested with ZDICT_isError().
+ *           On success `*parameters` contains the parameters selected.
+ * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 1 byte of memory for each input byte and additionally another 6 * 2^f bytes of memory for each thread.
+ */
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
+                    size_t dictBufferCapacity, const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters);
+
 /*! ZDICT_finalizeDictionary():
  * Given a custom content as a basis for dictionary, and a set of samples,
  * finalize dictionary by adding headers and statistics.
diff --git a/lib/legacy/zstd_v04.c b/lib/legacy/zstd_v04.c
index a2e2cfa80..15000db6d 100644
--- a/lib/legacy/zstd_v04.c
+++ b/lib/legacy/zstd_v04.c
@@ -1093,6 +1093,7 @@ static size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, un
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
     /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSE_DECODE_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
     DTableH.tableLog = (U16)tableLog;
     for (s=0; s<=maxSymbolValue; s++)
     {
diff --git a/lib/legacy/zstd_v05.c b/lib/legacy/zstd_v05.c
index a5e1b1ffc..65f07a3b5 100644
--- a/lib/legacy/zstd_v05.c
+++ b/lib/legacy/zstd_v05.c
@@ -1224,6 +1224,7 @@ size_t FSEv05_buildDTable(FSEv05_DTable* dt, const short* normalizedCounter, uns
     if (tableLog > FSEv05_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
     /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSEv05_FUNCTION_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
     DTableH.tableLog = (U16)tableLog;
     for (s=0; s<=maxSymbolValue; s++) {
         if (normalizedCounter[s]==-1) {
@@ -2845,6 +2846,7 @@ size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t*
 
 static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
diff --git a/lib/legacy/zstd_v06.c b/lib/legacy/zstd_v06.c
index 8b068b3e5..ca982c35e 100644
--- a/lib/legacy/zstd_v06.c
+++ b/lib/legacy/zstd_v06.c
@@ -3041,6 +3041,7 @@ size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t*
 
 static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
     return srcSize;
@@ -4006,7 +4007,7 @@ size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
                     if (ZSTDv06_isError(hSize)) return hSize;
                     if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
                         memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
-                        zbd->lhSize += iend-ip; ip = iend; notDone = 0;
+                        zbd->lhSize += iend-ip;
                         *dstCapacityPtr = 0;
                         return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
diff --git a/lib/legacy/zstd_v07.c b/lib/legacy/zstd_v07.c
index 70b170f0f..9f95f62ad 100644
--- a/lib/legacy/zstd_v07.c
+++ b/lib/legacy/zstd_v07.c
@@ -3150,10 +3150,10 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
     const BYTE* ip = (const BYTE*)src;
 
     if (srcSize < ZSTDv07_frameHeaderSize_min) return ZSTDv07_frameHeaderSize_min;
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
     if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
         if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
             if (srcSize < ZSTDv07_skippableHeaderSize) return ZSTDv07_skippableHeaderSize; /* magic number + skippable frame length */
-            memset(fparamsPtr, 0, sizeof(*fparamsPtr));
             fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
             fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
             return 0;
@@ -3175,11 +3175,13 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
         U32 windowSize = 0;
         U32 dictID = 0;
         U64 frameContentSize = 0;
-        if ((fhdByte & 0x08) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits, which must be zero */
+        if ((fhdByte & 0x08) != 0)   /* reserved bits, which must be zero */
+            return ERROR(frameParameter_unsupported);
         if (!directMode) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTDv07_WINDOWLOG_ABSOLUTEMIN;
-            if (windowLog > ZSTDv07_WINDOWLOG_MAX) return ERROR(frameParameter_unsupported);
+            if (windowLog > ZSTDv07_WINDOWLOG_MAX)
+                return ERROR(frameParameter_unsupported);
             windowSize = (1U << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
@@ -3201,7 +3203,8 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
             case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
         }
         if (!windowSize) windowSize = (U32)frameContentSize;
-        if (windowSize > windowSizeMax) return ERROR(frameParameter_unsupported);
+        if (windowSize > windowSizeMax)
+            return ERROR(frameParameter_unsupported);
         fparamsPtr->frameContentSize = frameContentSize;
         fparamsPtr->windowSize = windowSize;
         fparamsPtr->dictID = dictID;
@@ -3220,11 +3223,10 @@ size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src,
                    - frame header not completely provided (`srcSize` too small) */
 unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize)
 {
-    {   ZSTDv07_frameParams fparams;
-        size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
-        if (frResult!=0) return 0;
-        return fparams.frameContentSize;
-    }
+    ZSTDv07_frameParams fparams;
+    size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
+    if (frResult!=0) return 0;
+    return fparams.frameContentSize;
 }
 
 
diff --git a/lib/zstd.h b/lib/zstd.h
index edb107c2b..25441e68c 100644
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -35,26 +35,38 @@ extern "C" {
 #endif
 
 
-/*******************************************************************************************************
+/*******************************************************************************
   Introduction
 
-  zstd, short for Zstandard, is a fast lossless compression algorithm,
-  targeting real-time compression scenarios at zlib-level and better compression ratios.
-  The zstd compression library provides in-memory compression and decompression functions.
-  The library supports compression levels from 1 up to ZSTD_maxCLevel() which is currently 22.
-  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
+  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting
+  real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression
+  functions.
+
+  The library supports regular compression levels from 1 up to ZSTD_maxCLevel(),
+  which is currently 22. Levels >= 20, labeled `--ultra`, should be used with
+  caution, as they require more memory. The library also offers negative
+  compression levels, which extend the range of speed vs. ratio preferences.
+  The lower the level, the faster the speed (at the cost of compression).
+
   Compression can be done in:
     - a single step (described as Simple API)
     - a single step, reusing a context (described as Explicit context)
     - unbounded multiple steps (described as Streaming compression)
-  The compression ratio achievable on small data can be highly improved using a dictionary in:
-    - a single step (described as Simple dictionary API)
-    - a single step, reusing a dictionary (described as Bulk-processing dictionary API)
 
-  Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
-  Advanced experimental APIs shall never be used with a dynamic library.
-  They are not "stable", their definition may change in the future. Only static linking is allowed.
-*********************************************************************************************************/
+  The compression ratio achievable on small data can be highly improved using
+  a dictionary. Dictionary compression can be performed in:
+    - a single step (described as Simple dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing
+      dictionary API)
+
+  Advanced experimental functions can be accessed using
+  `#define ZSTD_STATIC_LINKING_ONLY` before including zstd.h.
+
+  Advanced experimental APIs should never be used with a dynamically-linked
+  library. They are not "stable"; their definitions or signatures may change in
+  the future. Only static linking is allowed.
+*******************************************************************************/
 
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
@@ -211,7 +223,8 @@ typedef struct ZSTD_CDict_s ZSTD_CDict;
  *  When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
  *  ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
  *  ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
- *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict */
+ *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict
+ *  Note : A ZSTD_CDict can be created with an empty dictionary, but it is inefficient for small data. */
 ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
                                          int compressionLevel);
 
@@ -223,7 +236,9 @@ ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
  *  Compression using a digested Dictionary.
  *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
  *  Note that compression level is decided during dictionary creation.
- *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
+ *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no)
+ *  Note : ZSTD_compress_usingCDict() can be used with a ZSTD_CDict created from an empty dictionary.
+ *         But it is inefficient for small data, and it is recommended to use ZSTD_compressCCtx(). */
 ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
                                             void* dst, size_t dstCapacity,
                                       const void* src, size_t srcSize,
@@ -1161,16 +1176,21 @@ ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
 
 /*! ZSTD_CCtx_refPrefix() :
  *  Reference a prefix (single-usage dictionary) for next compression job.
- *  Decompression need same prefix to properly regenerate data.
- *  Prefix is **only used once**. Tables are discarded at end of compression job (ZSTD_e_end).
+ *  Decompression will need same prefix to properly regenerate data.
+ *  Compressing with a prefix is similar in outcome as performing a diff and compressing it,
+ *  but performs much faster, especially during decompression (compression speed is tunable with compression level).
+ *  Note that prefix is **only used once**. Tables are discarded at end of compression job (ZSTD_e_end).
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
  *  Note 1 : Prefix buffer is referenced. It **must** outlive compression job.
  *           Its contain must remain unmodified up to end of compression (ZSTD_e_end).
- *  Note 2 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+ *  Note 2 : If the intention is to diff some large src data blob with some prior version of itself,
+ *           ensure that the window size is large enough to contain the entire source.
+ *           See ZSTD_p_windowLog.
+ *  Note 3 : Referencing a prefix involves building tables, which are dependent on compression parameters.
  *           It's a CPU consuming operation, with non-negligible impact on latency.
  *           If there is a need to use same prefix multiple times, consider loadDictionary instead.
- *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *  Note 4 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
  *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode. */
 ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
                                        const void* prefix, size_t prefixSize);
@@ -1353,6 +1373,8 @@ ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
 
 /*! ZSTD_DCtx_refPrefix() :
  *  Reference a prefix (single-usage dictionary) for next compression job.
+ *  This is the reverse operation of ZSTD_CCtx_refPrefix(),
+ *  and must use the same prefix as the one used during compression.
  *  Prefix is **only used once**. Reference is discarded at end of frame.
  *  End of frame is reached when ZSTD_DCtx_decompress_generic() returns 0.
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
diff --git a/programs/Makefile b/programs/Makefile
index 912f9eff0..f1a96325c 100644
--- a/programs/Makefile
+++ b/programs/Makefile
@@ -132,6 +132,15 @@ else
 LZ4_MSG := $(NO_LZ4_MSG)
 endif
 
+# enable backtrace symbol names for Linux/Darwin
+ALL_SYMBOLS := 0
+ifeq (,$(filter Windows%, $(OS)))
+ifeq ($(ALL_SYMBOLS), 1)
+DEBUGFLAGS_LD+=-rdynamic
+endif
+endif
+
+
 .PHONY: default
 default: zstd-release
 
@@ -144,7 +153,7 @@ allVariants: zstd zstd-compress zstd-decompress zstd-small zstd-nolegacy
 $(ZSTDDECOMP_O): CFLAGS += $(ALIGN_LOOP)
 
 zstd : CPPFLAGS += $(THREAD_CPP) $(ZLIBCPP) $(LZMACPP) $(LZ4CPP)
-zstd : LDFLAGS += $(THREAD_LD) $(ZLIBLD) $(LZMALD) $(LZ4LD)
+zstd : LDFLAGS += $(THREAD_LD) $(ZLIBLD) $(LZMALD) $(LZ4LD) $(DEBUGFLAGS_LD)
 zstd : CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
 zstd : $(ZSTDLIB_FILES) zstdcli.o fileio.o bench.o datagen.o dibio.o
 	@echo "$(THREAD_MSG)"
@@ -158,6 +167,7 @@ endif
 
 .PHONY: zstd-release
 zstd-release: DEBUGFLAGS :=
+zstd-release: DEBUGFLAGS_LD :=
 zstd-release: zstd
 
 zstd32 : CPPFLAGS += $(THREAD_CPP)
diff --git a/programs/README.md b/programs/README.md
index 22a00409c..804cb8b0a 100644
--- a/programs/README.md
+++ b/programs/README.md
@@ -61,6 +61,13 @@ There are however other Makefile targets that create different variations of CLI
   In which case, linking stage will fail if `lz4` library cannot be found.
   This is useful to prevent silent feature disabling.
 
+- __ALL_SYMBOLS__ : `zstd` can display a stack backtrace if the execution
+  generates a runtime exception. By default, this feature may be
+  degraded/disabled on some platforms unless additional compiler directives are
+  applied. When triaging a runtime issue, enabling this feature can provided
+  more context to determine the location of the fault.
+  Example : `make zstd ALL_SYMBOLS=1`
+
 
 #### Aggregation of parameters
 CLI supports aggregation of parameters i.e. `-b1`, `-e18`, and `-i1` can be joined into `-b1e18i1`.
@@ -151,6 +158,7 @@ Advanced arguments :
 Dictionary builder :
 --train ## : create a dictionary from a training set of files
 --train-cover[=k=#,d=#,steps=#,split=#] : use the cover algorithm with optional args
+--train-fastcover[=k=#,d=#,f=#,steps=#,split=#,accel=#] : use the fastcover algorithm with optional args
 --train-legacy[=s=#] : use the legacy algorithm with selectivity (default: 9)
  -o file : `file` is dictionary name (default: dictionary)
 --maxdict=# : limit dictionary to specified size (default: 112640)
diff --git a/programs/bench.c b/programs/bench.c
index 79ef42caa..326c1c1c5 100644
--- a/programs/bench.c
+++ b/programs/bench.c
@@ -63,7 +63,12 @@
 #define MB *(1 <<20)
 #define GB *(1U<<30)
 
-static const size_t maxMemory = (sizeof(size_t)==4)  ?  (2 GB - 64 MB) : (size_t)(1ULL << ((sizeof(size_t)*8)-31));
+#define BMK_RUNTEST_DEFAULT_MS 1000
+
+static const size_t maxMemory = (sizeof(size_t)==4)  ?
+                    /* 32-bit */ (2 GB - 64 MB) :
+                    /* 64-bit */ (size_t)(1ULL << ((sizeof(size_t)*8)-31));
+
 
 /* *************************************
 *  console display
@@ -97,26 +102,26 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
     return errorNum;                                  \
 }
 
-#define EXM_THROW(errorNum, retType, ...)  {          \
+#define RETURN_ERROR(errorNum, retType, ...)  {       \
     retType r;                                        \
     memset(&r, 0, sizeof(retType));                   \
     DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);    \
     DISPLAYLEVEL(1, "Error %i : ", errorNum);         \
     DISPLAYLEVEL(1, __VA_ARGS__);                     \
     DISPLAYLEVEL(1, " \n");                           \
-    r.error = errorNum;                               \
+    r.tag = errorNum;                                 \
     return r;                                         \
 }
 
 /* error without displaying */
-#define EXM_THROW_ND(errorNum, retType, ...)  {       \
+#define RETURN_QUIET_ERROR(errorNum, retType, ...)  { \
     retType r;                                        \
     memset(&r, 0, sizeof(retType));                   \
     DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);    \
     DEBUGOUTPUT("Error %i : ", errorNum);             \
     DEBUGOUTPUT(__VA_ARGS__);                         \
     DEBUGOUTPUT(" \n");                               \
-    r.error = errorNum;                               \
+    r.tag = errorNum;                                 \
     return r;                                         \
 }
 
@@ -124,16 +129,15 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
 *  Benchmark Parameters
 ***************************************/
 
-BMK_advancedParams_t BMK_initAdvancedParams(void) { 
-    BMK_advancedParams_t res = { 
+BMK_advancedParams_t BMK_initAdvancedParams(void) {
+    BMK_advancedParams_t const res = {
         BMK_both, /* mode */
-        BMK_timeMode, /* loopMode */
         BMK_TIMETEST_DEFAULT_S, /* nbSeconds */
         0, /* blockSize */
         0, /* nbWorkers */
         0, /* realTime */
         0, /* additionalParam */
-        0, /* ldmFlag */ 
+        0, /* ldmFlag */
         0, /* ldmMinMatch */
         0, /* ldmHashLog */
         0, /* ldmBuckSizeLog */
@@ -156,20 +160,13 @@ typedef struct {
     size_t resSize;
 } blockParam_t;
 
-struct  BMK_timeState_t{
-    unsigned nbLoops;
-    U64 timeRemaining;
-    UTIL_time_t coolTime;
-    U64 fastestTime;
-};
-
 #undef MIN
 #undef MAX
 #define MIN(a,b)    ((a) < (b) ? (a) : (b))
 #define MAX(a,b)    ((a) > (b) ? (a) : (b))
 
-static void BMK_initCCtx(ZSTD_CCtx* ctx, 
-    const void* dictBuffer, size_t dictBufferSize, int cLevel, 
+static void BMK_initCCtx(ZSTD_CCtx* ctx,
+    const void* dictBuffer, size_t dictBufferSize, int cLevel,
     const ZSTD_compressionParameters* comprParams, const BMK_advancedParams_t* adv) {
     ZSTD_CCtx_reset(ctx);
     ZSTD_CCtx_resetParameters(ctx);
@@ -194,15 +191,15 @@ static void BMK_initCCtx(ZSTD_CCtx* ctx,
     ZSTD_CCtx_loadDictionary(ctx, dictBuffer, dictBufferSize);
 }
 
-
-static void BMK_initDCtx(ZSTD_DCtx* dctx, 
+static void BMK_initDCtx(ZSTD_DCtx* dctx,
     const void* dictBuffer, size_t dictBufferSize) {
     ZSTD_DCtx_reset(dctx);
     ZSTD_DCtx_loadDictionary(dctx, dictBuffer, dictBufferSize);
 }
 
+
 typedef struct {
-    ZSTD_CCtx* ctx;
+    ZSTD_CCtx* cctx;
     const void* dictBuffer;
     size_t dictBufferSize;
     int cLevel;
@@ -212,7 +209,7 @@ typedef struct {
 
 static size_t local_initCCtx(void* payload) {
     BMK_initCCtxArgs* ag = (BMK_initCCtxArgs*)payload;
-    BMK_initCCtx(ag->ctx, ag->dictBuffer, ag->dictBufferSize, ag->cLevel, ag->comprParams, ag->adv);
+    BMK_initCCtx(ag->cctx, ag->dictBuffer, ag->dictBufferSize, ag->cLevel, ag->comprParams, ag->adv);
     return 0;
 }
 
@@ -228,26 +225,24 @@ static size_t local_initDCtx(void* payload) {
     return 0;
 }
 
-/* additional argument is just the context */
+
+/* `addArgs` is the context */
 static size_t local_defaultCompress(
-    const void* srcBuffer, size_t srcSize, 
-    void* dstBuffer, size_t dstSize, 
-    void* addArgs) {
+                    const void* srcBuffer, size_t srcSize,
+                    void* dstBuffer, size_t dstSize,
+                    void* addArgs)
+{
     size_t moreToFlush = 1;
-    ZSTD_CCtx* ctx = (ZSTD_CCtx*)addArgs;
+    ZSTD_CCtx* const cctx = (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;
+    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_compress_generic(ctx, &out, &in, ZSTD_e_end);
+        moreToFlush = ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end);
         if (ZSTD_isError(moreToFlush)) {
             return moreToFlush;
         }
@@ -255,27 +250,23 @@ static size_t local_defaultCompress(
     return out.pos;
 }
 
-/* additional argument is just the context */
+/* `addArgs` is the context */
 static size_t local_defaultDecompress(
-    const void* srcBuffer, size_t srcSize, 
-    void* dstBuffer, size_t dstSize, 
-    void* addArgs) {
+                    const void* srcBuffer, size_t srcSize,
+                    void* dstBuffer, size_t dstCapacity,
+                    void* addArgs)
+{
     size_t moreToFlush = 1;
-    ZSTD_DCtx* dctx = (ZSTD_DCtx*)addArgs;
+    ZSTD_DCtx* const 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;
+    in.src = srcBuffer; in.size = srcSize; in.pos = 0;
+    out.dst = dstBuffer; out.size = dstCapacity; out.pos = 0;
     while (moreToFlush) {
         if(out.pos == out.size) {
             return (size_t)-ZSTD_error_dstSize_tooSmall;
         }
-        moreToFlush = ZSTD_decompress_generic(dctx,
-                            &out, &in);
+        moreToFlush = ZSTD_decompress_generic(dctx, &out, &in);
         if (ZSTD_isError(moreToFlush)) {
             return moreToFlush;
         }
@@ -284,198 +275,307 @@ static size_t local_defaultDecompress(
 
 }
 
-/* initFn will be measured once, bench fn will be measured x times */
-/* benchFn should return error value or out Size */
-/* takes # of blocks and list of size & stuff for each. */
-/* only does looping */
-/* note time per loop could be zero if interval too short */
-BMK_customReturn_t BMK_benchFunction(
-    BMK_benchFn_t benchFn, void* benchPayload,
-    BMK_initFn_t initFn, void* initPayload,
-    size_t blockCount,
-    const void* const * const srcBlockBuffers, const size_t* srcBlockSizes,
-    void* const * const dstBlockBuffers, const size_t* dstBlockCapacities, size_t* blockResult, 
-    unsigned nbLoops) {
-    size_t dstSize = 0;
-    U64 totalTime;
 
-    BMK_customReturn_t retval;
-    UTIL_time_t clockStart;
+/*===  Benchmarking an arbitrary function  ===*/
+
+int BMK_isSuccessful_runOutcome(BMK_runOutcome_t outcome)
+{
+    return outcome.tag == 0;
+}
+
+/* warning : this function will stop program execution if outcome is invalid !
+ *           check outcome validity first, using BMK_isValid_runResult() */
+BMK_runTime_t BMK_extract_runTime(BMK_runOutcome_t outcome)
+{
+    assert(outcome.tag == 0);
+    return outcome.internal_never_use_directly;
+}
+
+static BMK_runOutcome_t BMK_runOutcome_error(void)
+{
+    BMK_runOutcome_t b;
+    memset(&b, 0, sizeof(b));
+    b.tag = 1;
+    return b;
+}
+
+static BMK_runOutcome_t BMK_setValid_runTime(BMK_runTime_t runTime)
+{
+    BMK_runOutcome_t outcome;
+    outcome.tag = 0;
+    outcome.internal_never_use_directly = runTime;
+    return outcome;
+}
+
+
+/* initFn will be measured once, benchFn will be measured `nbLoops` times */
+/* initFn is optional, provide NULL if none */
+/* benchFn must return size_t field compliant with ZSTD_isError for error valuee */
+/* takes # of blocks and list of size & stuff for each. */
+/* can report result of benchFn for each block into blockResult. */
+/* blockResult is optional, provide NULL if this information is not required */
+/* note : time per loop could be zero if run time < timer resolution */
+BMK_runOutcome_t BMK_benchFunction(
+            BMK_benchFn_t benchFn, void* benchPayload,
+            BMK_initFn_t initFn, void* initPayload,
+            size_t blockCount,
+            const void* const * srcBlockBuffers, const size_t* srcBlockSizes,
+            void* const * dstBlockBuffers, const size_t* dstBlockCapacities,
+            size_t* blockResults,
+            unsigned nbLoops)
+{
+    size_t dstSize = 0;
 
     if(!nbLoops) {
-        EXM_THROW_ND(1, BMK_customReturn_t, "nbLoops must be nonzero \n");
+        RETURN_QUIET_ERROR(2, BMK_runOutcome_t, "nbLoops must be nonzero ");
     }
 
-    {
-        size_t i;
+    /* init */
+    {   size_t i;
         for(i = 0; i < blockCount; i++) {
             memset(dstBlockBuffers[i], 0xE5, dstBlockCapacities[i]);  /* warm up and erase result buffer */
         }
 #if 0
-        /* based on testing these seem to lower accuracy of multiple calls of 1 nbLoops vs 1 call of multiple nbLoops 
-         * (Makes former slower)   
+        /* based on testing these seem to lower accuracy of multiple calls of 1 nbLoops vs 1 call of multiple nbLoops
+         * (Makes former slower)
          */
         UTIL_sleepMilli(5);  /* give processor time to other processes */
         UTIL_waitForNextTick();
 #endif
     }
 
-    {      
-        unsigned i, j;
-        clockStart = UTIL_getTime();
-        if(initFn != NULL) { initFn(initPayload); }
-        for(i = 0; i < nbLoops; i++) {
-            for(j = 0; j < blockCount; j++) {
-                size_t res = benchFn(srcBlockBuffers[j], srcBlockSizes[j], dstBlockBuffers[j], dstBlockCapacities[j], benchPayload);
+    /* benchmark */
+    {   UTIL_time_t const clockStart = UTIL_getTime();
+        unsigned loopNb, blockNb;
+        if (initFn != NULL) initFn(initPayload);
+        for (loopNb = 0; loopNb < nbLoops; loopNb++) {
+            for (blockNb = 0; blockNb < blockCount; blockNb++) {
+                size_t const res = benchFn(srcBlockBuffers[blockNb], srcBlockSizes[blockNb],
+                                    dstBlockBuffers[blockNb], dstBlockCapacities[blockNb],
+                                    benchPayload);
                 if(ZSTD_isError(res)) {
-                    EXM_THROW_ND(2, BMK_customReturn_t, "Function benchmarking failed on block %u of size %u : %s  \n",
-                        j, (U32)dstBlockCapacities[j], ZSTD_getErrorName(res));
-                } else if(i == nbLoops - 1) {
+                    RETURN_QUIET_ERROR(2, BMK_runOutcome_t,
+                        "Function benchmark failed on block %u of size %u : %s",
+                        blockNb, (U32)dstBlockCapacities[blockNb], ZSTD_getErrorName(res));
+                } else if (loopNb == 0) {
                     dstSize += res;
-                    if(blockResult != NULL) {
-                        blockResult[j] = res;
-                    }
-                } 
-            }
-        }
-        totalTime = UTIL_clockSpanNano(clockStart);
-    }
+                    if (blockResults != NULL) blockResults[blockNb] = res;
+            }   }
+        }  /* for (loopNb = 0; loopNb < nbLoops; loopNb++) */
 
-    retval.error = 0;
-    retval.result.nanoSecPerRun = totalTime / nbLoops;
-    retval.result.sumOfReturn = dstSize;
-    return retval;
-} 
-
-#define MINUSABLETIME 500000000ULL /* 0.5 seconds in ns */
-
-void BMK_resetTimeState(BMK_timedFnState_t* r, unsigned nbSeconds) {
-    r->nbLoops = 1;
-    r->timeRemaining = (U64)nbSeconds * TIMELOOP_NANOSEC;
-    r->coolTime = UTIL_getTime();
-    r->fastestTime = (U64)(-1LL);
+        {   U64 const totalTime = UTIL_clockSpanNano(clockStart);
+            BMK_runTime_t rt;
+            rt.nanoSecPerRun = totalTime / nbLoops;
+            rt.sumOfReturn = dstSize;
+            return BMK_setValid_runTime(rt);
+    }   }
 }
 
-BMK_timedFnState_t* BMK_createTimeState(unsigned nbSeconds) {
-    BMK_timedFnState_t* r = (BMK_timedFnState_t*)malloc(sizeof(struct BMK_timeState_t));
-    if(r == NULL) {
-        return r;
-    }
-    BMK_resetTimeState(r, nbSeconds);
+
+/* ====  Benchmarking any function, providing intermediate results  ==== */
+
+struct BMK_timedFnState_s {
+    U64 timeSpent_ns;
+    U64 timeBudget_ns;
+    U64 runBudget_ns;
+    BMK_runTime_t fastestRun;
+    unsigned nbLoops;
+    UTIL_time_t coolTime;
+};  /* typedef'd to BMK_timedFnState_t within bench.h */
+
+BMK_timedFnState_t* BMK_createTimedFnState(unsigned total_ms, unsigned run_ms)
+{
+    BMK_timedFnState_t* const r = (BMK_timedFnState_t*)malloc(sizeof(*r));
+    if (r == NULL) return NULL;   /* malloc() error */
+    BMK_resetTimedFnState(r, total_ms, run_ms);
     return r;
 }
 
-void BMK_freeTimeState(BMK_timedFnState_t* state) {
+void BMK_freeTimedFnState(BMK_timedFnState_t* state) {
     free(state);
 }
 
-/* make option for dstBlocks to be */
-BMK_customTimedReturn_t BMK_benchFunctionTimed(
-    BMK_timedFnState_t* cont,
-    BMK_benchFn_t benchFn, void* benchPayload,
-    BMK_initFn_t initFn, void* initPayload,
-    size_t blockCount,
-    const void* const* const srcBlockBuffers, const size_t* srcBlockSizes,
-    void * const * const dstBlockBuffers, const size_t * dstBlockCapacities, size_t* blockResults) 
+void BMK_resetTimedFnState(BMK_timedFnState_t* timedFnState, unsigned total_ms, unsigned run_ms)
 {
-    U64 fastest = cont->fastestTime;
-    int completed = 0;
-    BMK_customTimedReturn_t r;
-    r.completed = 0;
+    if (!total_ms) total_ms = 1 ;
+    if (!run_ms) run_ms = 1;
+    if (run_ms > total_ms) run_ms = total_ms;
+    timedFnState->timeSpent_ns = 0;
+    timedFnState->timeBudget_ns = (U64)total_ms * TIMELOOP_NANOSEC / 1000;
+    timedFnState->runBudget_ns = (U64)run_ms * TIMELOOP_NANOSEC / 1000;
+    timedFnState->fastestRun.nanoSecPerRun = (U64)(-1LL);
+    timedFnState->fastestRun.sumOfReturn = (size_t)(-1LL);
+    timedFnState->nbLoops = 1;
+    timedFnState->coolTime = UTIL_getTime();
+}
+
+/* Tells if nb of seconds set in timedFnState for all runs is spent.
+ * note : this function will return 1 if BMK_benchFunctionTimed() has actually errored. */
+int BMK_isCompleted_TimedFn(const BMK_timedFnState_t* timedFnState)
+{
+    return (timedFnState->timeSpent_ns >= timedFnState->timeBudget_ns);
+}
+
+
+#define MINUSABLETIME  (TIMELOOP_NANOSEC / 2)  /* 0.5 seconds */
+
+BMK_runOutcome_t BMK_benchTimedFn(
+            BMK_timedFnState_t* cont,
+            BMK_benchFn_t benchFn, void* benchPayload,
+            BMK_initFn_t initFn, void* initPayload,
+            size_t blockCount,
+            const void* const* srcBlockBuffers, const size_t* srcBlockSizes,
+            void * const * dstBlockBuffers, const size_t * dstBlockCapacities,
+            size_t* blockResults)
+{
+    U64 const runBudget_ns = cont->runBudget_ns;
+    U64 const runTimeMin_ns = runBudget_ns / 2;
+    int completed = 0;
+    BMK_runTime_t bestRunTime = cont->fastestRun;
+
+    while (!completed) {
+        BMK_runOutcome_t runResult;
 
-    while(!r.completed && !completed)
-    {
         /* Overheat protection */
         if (UTIL_clockSpanMicro(cont->coolTime) > ACTIVEPERIOD_MICROSEC) {
             DEBUGOUTPUT("\rcooling down ...    \r");
             UTIL_sleep(COOLPERIOD_SEC);
             cont->coolTime = UTIL_getTime();
         }
+
         /* reinitialize capacity */
-        r.result = BMK_benchFunction(benchFn, benchPayload, initFn, initPayload,
-        blockCount, srcBlockBuffers, srcBlockSizes, dstBlockBuffers, dstBlockCapacities, blockResults, cont->nbLoops);
-        if(r.result.error) { /* completed w/ error */
-            r.completed = 1;
-            return r;
+        runResult = BMK_benchFunction(benchFn, benchPayload,
+                                    initFn, initPayload,
+                                    blockCount,
+                                    srcBlockBuffers, srcBlockSizes,
+                                    dstBlockBuffers, dstBlockCapacities,
+                                    blockResults,
+                                    cont->nbLoops);
+
+        if(!BMK_isSuccessful_runOutcome(runResult)) { /* error : move out */
+            return BMK_runOutcome_error();
         }
 
-        {   U64 const loopDuration = r.result.result.nanoSecPerRun * cont->nbLoops;
-            r.completed = (cont->timeRemaining <= loopDuration);
-            cont->timeRemaining -= loopDuration;
-            if (loopDuration > (TIMELOOP_NANOSEC / 100)) { 
-                fastest = MIN(fastest, r.result.result.nanoSecPerRun);
-                if(loopDuration >= MINUSABLETIME) {
-                    r.result.result.nanoSecPerRun = fastest;
-                    cont->fastestTime = fastest;
-                }
-                cont->nbLoops = (U32)(TIMELOOP_NANOSEC / r.result.result.nanoSecPerRun) + 1;
+        {   BMK_runTime_t const newRunTime = BMK_extract_runTime(runResult);
+            U64 const loopDuration_ns = newRunTime.nanoSecPerRun * cont->nbLoops;
+
+            cont->timeSpent_ns += loopDuration_ns;
+
+            /* estimate nbLoops for next run to last approximately 1 second */
+            if (loopDuration_ns > (runBudget_ns / 50)) {
+                U64 const fastestRun_ns = MIN(bestRunTime.nanoSecPerRun, newRunTime.nanoSecPerRun);
+                cont->nbLoops = (U32)(runBudget_ns / fastestRun_ns) + 1;
             } else {
-                const unsigned multiplier = 2;
+                /* previous run was too short : blindly increase workload by x multiplier */
+                const unsigned multiplier = 10;
                 assert(cont->nbLoops < ((unsigned)-1) / multiplier);  /* avoid overflow */
                 cont->nbLoops *= multiplier;
             }
-            if(loopDuration < MINUSABLETIME) { /* don't report results which have time too low */
-                continue;
-            }
 
+            if(loopDuration_ns < runTimeMin_ns) {
+                /* don't report results for which benchmark run time was too small : increased risks of rounding errors */
+                assert(completed == 0);
+                continue;
+            } else {
+                if(newRunTime.nanoSecPerRun < bestRunTime.nanoSecPerRun) {
+                    bestRunTime = newRunTime;
+                }
+                completed = 1;
+            }
         }
-        completed = 1;
-    }
-    return r;
+    }   /* while (!completed) */
+
+    return BMK_setValid_runTime(bestRunTime);
 }
 
-/* benchMem with no allocation */
-static BMK_return_t BMK_benchMemAdvancedNoAlloc(
-    const void ** const srcPtrs, size_t* const srcSizes,
-    void** const cPtrs, size_t* const cCapacities, size_t* const cSizes, 
-    void** const resPtrs, size_t* const resSizes,
-    void** resultBufferPtr, void* compressedBuffer,
-    const size_t maxCompressedSize,
-    BMK_timedFnState_t* timeStateCompress, BMK_timedFnState_t* timeStateDecompress,
 
-    const void* srcBuffer, size_t srcSize,
-    const size_t* fileSizes, unsigned nbFiles,
-    const int cLevel, const ZSTD_compressionParameters* comprParams,
-    const void* dictBuffer, size_t dictBufferSize,
-    ZSTD_CCtx* ctx, ZSTD_DCtx* dctx,
-    int displayLevel, const char* displayName, const BMK_advancedParams_t* adv) 
+/* ================================================================= */
+/*      Benchmark Zstandard, mem-to-mem scenarios                    */
+/* ================================================================= */
+
+int BMK_isSuccessful_benchOutcome(BMK_benchOutcome_t outcome)
 {
-    size_t const blockSize = ((adv->blockSize>=32 && (adv->mode != BMK_decodeOnly)) ? adv->blockSize : srcSize) + (!srcSize); /* avoid div by 0 */
-    BMK_return_t results = { { 0, 0, 0, 0 }, 0 } ;
+    return outcome.tag == 0;
+}
+
+BMK_benchResult_t BMK_extract_benchResult(BMK_benchOutcome_t outcome)
+{
+    assert(outcome.tag == 0);
+    return outcome.internal_never_use_directly;
+}
+
+static BMK_benchOutcome_t BMK_benchOutcome_error(void)
+{
+    BMK_benchOutcome_t b;
+    memset(&b, 0, sizeof(b));
+    b.tag = 1;
+    return b;
+}
+
+static BMK_benchOutcome_t BMK_benchOutcome_setValidResult(BMK_benchResult_t result)
+{
+    BMK_benchOutcome_t b;
+    b.tag = 0;
+    b.internal_never_use_directly = result;
+    return b;
+}
+
+
+/* benchMem with no allocation */
+static BMK_benchOutcome_t BMK_benchMemAdvancedNoAlloc(
+            const void** srcPtrs, size_t* srcSizes,
+            void** cPtrs, size_t* cCapacities, size_t* cSizes,
+            void** resPtrs, size_t* resSizes,
+            void** resultBufferPtr, void* compressedBuffer,
+            size_t maxCompressedSize,
+            BMK_timedFnState_t* timeStateCompress,
+            BMK_timedFnState_t* timeStateDecompress,
+
+            const void* srcBuffer, size_t srcSize,
+            const size_t* fileSizes, unsigned nbFiles,
+            const int cLevel, const ZSTD_compressionParameters* comprParams,
+            const void* dictBuffer, size_t dictBufferSize,
+            ZSTD_CCtx* cctx, ZSTD_DCtx* dctx,
+            int displayLevel, const char* displayName,
+            const BMK_advancedParams_t* adv)
+{
+    size_t const blockSize = ((adv->blockSize>=32 && (adv->mode != BMK_decodeOnly)) ? adv->blockSize : srcSize) + (!srcSize);  /* avoid div by 0 */
+    BMK_benchResult_t benchResult;
     size_t const loadedCompressedSize = srcSize;
     size_t cSize = 0;
     double ratio = 0.;
     U32 nbBlocks;
 
-    if(!ctx || !dctx) 
-        EXM_THROW(31, BMK_return_t, "error: passed in null context");
+    assert(cctx != NULL); assert(dctx != NULL);
 
     /* init */
-    if (strlen(displayName)>17) displayName += strlen(displayName)-17;   /* display last 17 characters */
+    memset(&benchResult, 0, sizeof(benchResult));
+    if (strlen(displayName)>17) displayName += strlen(displayName) - 17;   /* display last 17 characters */
     if (adv->mode == BMK_decodeOnly) {  /* benchmark only decompression : source must be already compressed */
         const char* srcPtr = (const char*)srcBuffer;
         U64 totalDSize64 = 0;
         U32 fileNb;
         for (fileNb=0; fileNb decodedSize) {
-                free(*resultBufferPtr); 
-                EXM_THROW(32, BMK_return_t, "original size is too large");   /* size_t overflow */
+            if (totalDSize64 > decodedSize) {  /* size_t overflow */
+                free(*resultBufferPtr);
+                RETURN_ERROR(32, BMK_benchOutcome_t, "original size is too large");
             }
             cSize = srcSize;
             srcSize = decodedSize;
             ratio = (double)srcSize / (double)cSize;
-        }   
+        }
     }
 
     /* Init data blocks  */
@@ -489,21 +589,21 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
             U32 const blockEnd = nbBlocks + nbBlocksforThisFile;
             for ( ; nbBlocksmode == BMK_decodeOnly) ? thisBlockSize : ZSTD_compressBound(thisBlockSize);
-                resPtrs[nbBlocks] = (void*)resPtr;
+                resPtrs[nbBlocks] = resPtr;
                 resSizes[nbBlocks] = (adv->mode == BMK_decodeOnly) ? (size_t) ZSTD_findDecompressedSize(srcPtr, thisBlockSize) : thisBlockSize;
                 srcPtr += thisBlockSize;
                 cPtr += cCapacities[nbBlocks];
                 resPtr += thisBlockSize;
                 remaining -= thisBlockSize;
-            }   
-        }   
+            }
+        }
     }
 
-    /* warmimg up memory */
+    /* warmimg up `compressedBuffer` */
     if (adv->mode == BMK_decodeOnly) {
         memcpy(compressedBuffer, srcBuffer, loadedCompressedSize);
     } else {
@@ -511,151 +611,105 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
     }
 
     /* Bench */
-    {   
-        U64 const crcOrig = (adv->mode == BMK_decodeOnly) ? 0 : XXH64(srcBuffer, srcSize, 0);
+    {   U64 const crcOrig = (adv->mode == BMK_decodeOnly) ? 0 : XXH64(srcBuffer, srcSize, 0);
 #       define NB_MARKS 4
-        const char* const marks[NB_MARKS] = { " |", " /", " =",  "\\" };
+        const char* marks[NB_MARKS] = { " |", " /", " =", " \\" };
         U32 markNb = 0;
-        DISPLAYLEVEL(2, "\r%79s\r", "");
+        int compressionCompleted = (adv->mode == BMK_decodeOnly);
+        int decompressionCompleted = (adv->mode == BMK_compressOnly);
+        BMK_initCCtxArgs cctxprep;
+        BMK_initDCtxArgs dctxprep;
+        cctxprep.cctx = cctx;
+        cctxprep.dictBuffer = dictBuffer;
+        cctxprep.dictBufferSize = dictBufferSize;
+        cctxprep.cLevel = cLevel;
+        cctxprep.comprParams = comprParams;
+        cctxprep.adv = adv;
+        dctxprep.dctx = dctx;
+        dctxprep.dictBuffer = dictBuffer;
+        dctxprep.dictBufferSize = dictBufferSize;
 
+        DISPLAYLEVEL(2, "\r%70s\r", "");   /* blank line */
         DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->\r", marks[markNb], displayName, (U32)srcSize);
-        {
-            BMK_initCCtxArgs cctxprep;
-            BMK_initDCtxArgs dctxprep;
-            cctxprep.ctx = ctx;
-            cctxprep.dictBuffer = dictBuffer;
-            cctxprep.dictBufferSize = dictBufferSize;
-            cctxprep.cLevel = cLevel;
-            cctxprep.comprParams = comprParams;
-            cctxprep.adv = adv;
-            dctxprep.dctx = dctx;
-            dctxprep.dictBuffer = dictBuffer;
-            dctxprep.dictBufferSize = dictBufferSize;
-            if(adv->loopMode == BMK_timeMode) {
-                BMK_customTimedReturn_t intermediateResultCompress;
-                BMK_customTimedReturn_t intermediateResultDecompress;
-                if(adv->mode == BMK_compressOnly) {
-                    intermediateResultCompress.completed = 0;
-                    intermediateResultDecompress.completed = 1;
-                } else if (adv->mode == BMK_decodeOnly) {
-                    intermediateResultCompress.completed = 1;
-                    intermediateResultDecompress.completed = 0;
-                } else { /* both */
-                    intermediateResultCompress.completed = 0;
-                    intermediateResultDecompress.completed = 0;
-                }
-                while(!(intermediateResultCompress.completed && intermediateResultDecompress.completed)) {
-                    if(!intermediateResultCompress.completed) { 
-                        intermediateResultCompress = BMK_benchFunctionTimed(timeStateCompress, &local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
-                        nbBlocks, srcPtrs, srcSizes, cPtrs, cCapacities, cSizes);
-                        if(intermediateResultCompress.result.error) {
-                            results.error = intermediateResultCompress.result.error;
-                            return results;
-                        }
-                        ratio = (double)(srcSize / intermediateResultCompress.result.result.sumOfReturn);
-                        {   
-                            int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                            results.result.cSpeed = (srcSize * TIMELOOP_NANOSEC / intermediateResultCompress.result.result.nanoSecPerRun);
-                            cSize = intermediateResultCompress.result.result.sumOfReturn;
-                            results.result.cSize = cSize;
-                            ratio = (double)srcSize / results.result.cSize;
-                            markNb = (markNb+1) % NB_MARKS;
-                            DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s\r",
-                                    marks[markNb], displayName, (U32)srcSize, (U32)results.result.cSize,
-                                    ratioAccuracy, ratio,
-                                    results.result.cSpeed < (10 MB) ? 2 : 1, (double)results.result.cSpeed / (1 MB));
-                        }
-                    }
 
-                    if(!intermediateResultDecompress.completed) {
-                        intermediateResultDecompress = BMK_benchFunctionTimed(timeStateDecompress, &local_defaultDecompress, (void*)(dctx), &local_initDCtx, (void*)&dctxprep,
-                        nbBlocks, (const void* const*)cPtrs, cSizes, resPtrs, resSizes, NULL);
-                        if(intermediateResultDecompress.result.error) {
-                            results.error = intermediateResultDecompress.result.error;
-                            return results;
-                        }
-      
-                        {   
-                            int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                            results.result.dSpeed = (srcSize * TIMELOOP_NANOSEC/ intermediateResultDecompress.result.result.nanoSecPerRun);
-                            markNb = (markNb+1) % NB_MARKS;
-                            DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
-                                    marks[markNb], displayName, (U32)srcSize, (U32)results.result.cSize,
-                                    ratioAccuracy, ratio,
-                                    results.result.cSpeed < (10 MB) ? 2 : 1, (double)results.result.cSpeed / (1 MB),
-                                    (double)results.result.dSpeed / (1 MB));
-                        }
-                    }
+        while (!(compressionCompleted && decompressionCompleted)) {
+
+            if (!compressionCompleted) {
+                BMK_runOutcome_t const cOutcome =
+                        BMK_benchTimedFn( timeStateCompress,
+                                        &local_defaultCompress, cctx,
+                                        &local_initCCtx, &cctxprep,
+                                        nbBlocks,
+                                        srcPtrs, srcSizes,
+                                        cPtrs, cCapacities,
+                                        cSizes);
+
+                if (!BMK_isSuccessful_runOutcome(cOutcome)) {
+                    return BMK_benchOutcome_error();
                 }
 
-            } else { //iterMode;
-                if(adv->mode != BMK_decodeOnly) {
+                {   BMK_runTime_t const cResult = BMK_extract_runTime(cOutcome);
+                    cSize = cResult.sumOfReturn;
+                    ratio = (double)srcSize / cSize;
+                    {   BMK_benchResult_t newResult;
+                        newResult.cSpeed = ((U64)srcSize * TIMELOOP_NANOSEC / cResult.nanoSecPerRun);
+                        benchResult.cSize = cSize;
+                        if (newResult.cSpeed > benchResult.cSpeed)
+                            benchResult.cSpeed = newResult.cSpeed;
+                }   }
 
-                    BMK_customReturn_t compressionResults = BMK_benchFunction(&local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
-                        nbBlocks, srcPtrs, srcSizes, cPtrs, cCapacities, cSizes, adv->nbSeconds); 
-                    if(compressionResults.error) {
-                        results.error = compressionResults.error;
-                        return results;
-                    }
-
-                    if(compressionResults.result.nanoSecPerRun == 0) {
-                        results.result.cSpeed = 0;
-                    } else {
-                        results.result.cSpeed = srcSize * TIMELOOP_NANOSEC / compressionResults.result.nanoSecPerRun;
-                    }
-
-                    results.result.cSize = compressionResults.result.sumOfReturn;
-                    {   
-                        int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                        cSize = compressionResults.result.sumOfReturn;
-                        results.result.cSize = cSize;
-                        ratio = (double)srcSize / results.result.cSize;
-                        markNb = (markNb+1) % NB_MARKS;
-                        DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s\r",
-                                marks[markNb], displayName, (U32)srcSize, (U32)results.result.cSize,
-                                ratioAccuracy, ratio,
-                                results.result.cSpeed < (10 MB) ? 2 : 1, (double)results.result.cSpeed / (1 MB));
-                    }
-                }
-                if(adv->mode != BMK_compressOnly) {
-                    BMK_customReturn_t decompressionResults = BMK_benchFunction(
-                        &local_defaultDecompress, (void*)(dctx),
-                        &local_initDCtx, (void*)&dctxprep, nbBlocks,
-                        (const void* const*)cPtrs, cSizes, resPtrs, resSizes, NULL, 
-                        adv->nbSeconds);
-                    if(decompressionResults.error) {
-                        results.error = decompressionResults.error;
-                        return results;
-                    }
-
-                    if(decompressionResults.result.nanoSecPerRun == 0) {
-                        results.result.dSpeed = 0;
-                    } else {
-                        results.result.dSpeed = srcSize * TIMELOOP_NANOSEC / decompressionResults.result.nanoSecPerRun;
-                    }
-
-                    {   
-                        int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                        markNb = (markNb+1) % NB_MARKS;
-                        DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
-                                marks[markNb], displayName, (U32)srcSize, (U32)results.result.cSize,
-                                ratioAccuracy, ratio,
-                                results.result.cSpeed < (10 MB) ? 2 : 1, (double)results.result.cSpeed / (1 MB),
-                                (double)results.result.dSpeed / (1 MB));
-                    }
+                {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
+                    markNb = (markNb+1) % NB_MARKS;
+                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s\r",
+                            marks[markNb], displayName,
+                            (U32)srcSize, (U32)cSize,
+                            ratioAccuracy, ratio,
+                            benchResult.cSpeed < (10 MB) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT);
                 }
+                compressionCompleted = BMK_isCompleted_TimedFn(timeStateCompress);
             }
-        }
+
+            if(!decompressionCompleted) {
+                BMK_runOutcome_t const dOutcome =
+                        BMK_benchTimedFn(timeStateDecompress,
+                                        &local_defaultDecompress, dctx,
+                                        &local_initDCtx, &dctxprep,
+                                        nbBlocks,
+                                        (const void *const *)cPtrs, cSizes,
+                                        resPtrs, resSizes,
+                                        NULL);
+
+                if(!BMK_isSuccessful_runOutcome(dOutcome)) {
+                    return BMK_benchOutcome_error();
+                }
+
+                {   BMK_runTime_t const dResult = BMK_extract_runTime(dOutcome);
+                    U64 const newDSpeed = (srcSize * TIMELOOP_NANOSEC / dResult.nanoSecPerRun);
+                    if (newDSpeed > benchResult.dSpeed)
+                        benchResult.dSpeed = newDSpeed;
+                }
+
+                {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
+                    markNb = (markNb+1) % NB_MARKS;
+                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
+                            marks[markNb], displayName,
+                            (U32)srcSize, (U32)benchResult.cSize,
+                            ratioAccuracy, ratio,
+                            benchResult.cSpeed < (10 MB) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT,
+                            (double)benchResult.dSpeed / MB_UNIT);
+                }
+                decompressionCompleted = BMK_isCompleted_TimedFn(timeStateDecompress);
+            }
+        }   /* while (!(compressionCompleted && decompressionCompleted)) */
 
         /* CRC Checking */
-        {   void* resultBuffer = *resultBufferPtr;
+        {   const BYTE* resultBuffer = (const BYTE*)(*resultBufferPtr);
             U64 const crcCheck = XXH64(resultBuffer, srcSize, 0);
-            /* adv->mode == 0 -> compress + decompress */
             if ((adv->mode == BMK_both) && (crcOrig!=crcCheck)) {
                 size_t u;
                 DISPLAY("!!! WARNING !!! %14s : Invalid Checksum : %x != %x   \n", displayName, (unsigned)crcOrig, (unsigned)crcCheck);
                 for (u=0; u