[build] Add support for ASM files in Make + CMake

* Extract out common portion of `lib/Makefile` into `lib/libzstd.mk`.
  Most relevantly, the way we find library files.
* Use `lib/libzstd.mk` in the other Makefiles instead of repeating the
  same code.
* Add a test `tests/test-variants.sh` that checks that the builds of
  `make -C programs allVariants` are correct, and run it in Actions.
* Adds support for ASM files in the CMake build.

The Meson build is not updated because it lists every file in zstd,
and supports ASM off the bat, so the Huffman ASM commit will just add
the ASM file to the list.

The Visual Studios build is not updated because I'm not adding ASM
support to Visual Studios yet.

Author: terrelln

.github/workflows/dev-short-tests.yml

@@ -200,6 +200,15 @@ jobs:
         make clean && make -j all MOREFLAGS="-Werror -DZSTD_NO_INLINE -DZSTD_STRIP_ERROR_STRINGS"
         make clean && make check MOREFLAGS="-Werror -DZSTD_NO_INLINE -DZSTD_STRIP_ERROR_STRINGS"
 
+  test-variants:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - name: make all variants & validate
+      run: |
+        make -j -C programs allVariants MOREFLAGS=-O0
+        ./tests/test-variants.sh
+
 
   qemu-consistency:
     name: QEMU ${{ matrix.name }}
@@ -263,7 +272,6 @@ jobs:
       run: |
         LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check
 
-
 # This test currently fails on Github Actions specifically.
 # Possible reason : TTY emulation.
 # Note that the same test works fine locally and on travisCI.

Makefile

@@ -217,7 +217,7 @@ armbuild: clean
 	CC=arm-linux-gnueabi-gcc CFLAGS="-Werror" $(MAKE) allzstd
 
 aarch64build: clean
-	CC=aarch64-linux-gnu-gcc CFLAGS="-Werror" $(MAKE) allzstd
+	CC=aarch64-linux-gnu-gcc CFLAGS="-Werror -O0" $(MAKE) allzstd
 
 ppcbuild: clean
 	CC=powerpc-linux-gnu-gcc CFLAGS="-m32 -Wno-attributes -Werror" $(MAKE) -j allzstd

build/cmake/lib/CMakeLists.txt

@@ -7,7 +7,7 @@
 # in the COPYING file in the root directory of this source tree).
 # ################################################################
 
-project(libzstd C)
+project(libzstd C ASM)
 
 set(CMAKE_INCLUDE_CURRENT_DIR TRUE)
 option(ZSTD_BUILD_STATIC "BUILD STATIC LIBRARIES" ON)
@@ -22,7 +22,7 @@ include_directories(${LIBRARY_DIR} ${LIBRARY_DIR}/common)
 
 file(GLOB CommonSources ${LIBRARY_DIR}/common/*.c)
 file(GLOB CompressSources ${LIBRARY_DIR}/compress/*.c)
-file(GLOB DecompressSources ${LIBRARY_DIR}/decompress/*.c)
+file(GLOB DecompressSources ${LIBRARY_DIR}/decompress/*.c ${LIBRARY_DIR}/decompress/*.S)
 file(GLOB DictBuilderSources ${LIBRARY_DIR}/dictBuilder/*.c)
 
 set(Sources

contrib/freestanding_lib/freestanding.py

@@ -460,7 +460,8 @@ def _log(self, *args, **kwargs):
         print(*args, **kwargs)
 
     def _copy_file(self, lib_path):
-        if not (lib_path.endswith(".c") or lib_path.endswith(".h")):
+        suffixes = [".c", ".h", ".S"]
+        if not any((lib_path.endswith(suffix) for suffix in suffixes)):
             return
         if lib_path in SKIPPED_FILES:
             self._log(f"\tSkipping file: {lib_path}")

contrib/linux-kernel/test/Makefile

@@ -18,9 +18,13 @@ CPPFLAGS += -DZSTD_ASAN_DONT_POISON_WORKSPACE
 LINUX_ZSTD_MODULE     := $(wildcard $(LINUX_ZSTDLIB)/*.c)
 LINUX_ZSTD_COMMON     := $(wildcard $(LINUX_ZSTDLIB)/common/*.c)
 LINUX_ZSTD_COMPRESS   := $(wildcard $(LINUX_ZSTDLIB)/compress/*.c)
-LINUX_ZSTD_DECOMPRESS := $(wildcard $(LINUX_ZSTDLIB)/decompress/*.c)
+LINUX_ZSTD_DECOMPRESS := $(wildcard $(LINUX_ZSTDLIB)/decompress/*.c $(LINUX_ZSTDLIB)/decompress/*.S)
 LINUX_ZSTD_FILES      := $(LINUX_ZSTD_MODULE) $(LINUX_ZSTD_COMMON) $(LINUX_ZSTD_COMPRESS) $(LINUX_ZSTD_DECOMPRESS)
-LINUX_ZSTD_OBJECTS    := $(LINUX_ZSTD_FILES:.c=.o)
+LINUX_ZSTD_OBJECTS0   := $(LINUX_ZSTD_FILES:.c=.o)
+LINUX_ZSTD_OBJECTS    := $(LINUX_ZSTD_OBJECTS0:.S=.o)
+
+%.o: %.S
+	$(CC) -c $(CPPFLAGS) $(CFLAGS) $^ -o $@
 
 liblinuxzstd.a: $(LINUX_ZSTD_OBJECTS)
 	$(AR) $(ARFLAGS) $@ $^

contrib/pzstd/Makefile

@@ -57,19 +57,6 @@ LD_COMMAND  = $(CXX) $^          $(ALL_LDFLAGS) $(LIBS) -pthread -o $@
 CC_COMMAND  = $(CC)  $(DEPFLAGS) $(ALL_CFLAGS)   -c $<  -o $@
 CXX_COMMAND = $(CXX) $(DEPFLAGS) $(ALL_CXXFLAGS) -c $<  -o $@
 
-# Get a list of all zstd files so we rebuild the static library when we need to
-ZSTDCOMMON_FILES := $(wildcard $(ZSTDDIR)/common/*.c) \
-                    $(wildcard $(ZSTDDIR)/common/*.h)
-ZSTDCOMP_FILES   := $(wildcard $(ZSTDDIR)/compress/*.c) \
-                    $(wildcard $(ZSTDDIR)/compress/*.h)
-ZSTDDECOMP_FILES := $(wildcard $(ZSTDDIR)/decompress/*.c) \
-                    $(wildcard $(ZSTDDIR)/decompress/*.h)
-ZSTDPROG_FILES   := $(wildcard $(PROGDIR)/*.c) \
-                    $(wildcard $(PROGDIR)/*.h)
-ZSTD_FILES       := $(wildcard $(ZSTDDIR)/*.h) \
-                    $(ZSTDDECOMP_FILES) $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES) \
-                    $(ZSTDPROG_FILES)
-
 # List all the pzstd source files so we can determine their dependencies
 PZSTD_SRCS  := $(wildcard *.cpp)
 PZSTD_TESTS := $(wildcard test/*.cpp)
@@ -189,7 +176,8 @@ roundtrip: test/RoundTripTest$(EXT)
 
 # Use the static library that zstd builds for simplicity and
 # so we get the compiler options correct
-$(ZSTDDIR)/libzstd.a: $(ZSTD_FILES)
+.PHONY: $(ZSTDDIR)/libzstd.a
+$(ZSTDDIR)/libzstd.a:
 	CFLAGS="$(ALL_CFLAGS)" LDFLAGS="$(ALL_LDFLAGS)" $(MAKE) -C $(ZSTDDIR) libzstd.a
 
 # Rules to build the tests

lib/Makefile

@@ -8,110 +8,13 @@
 # You may select, at your option, one of the above-listed licenses.
 # ################################################################
 
-# Note: by default, the static library is built single-threaded and dynamic library is built
-# multi-threaded. It is possible to force multi or single threaded builds by appending
-# -mt or -nomt to the build target (like lib-mt for multi-threaded, lib-nomt for single-threaded).
-.PHONY: default
-default: lib-release
-
-# define silent mode as default (verbose mode with V=1 or VERBOSE=1)
-$(V)$(VERBOSE).SILENT:
-
-# When cross-compiling from linux to windows,
-# one might need to specify TARGET_SYSTEM as "Windows."
-# Building from Fedora fails without it.
-# (but Ubuntu and Debian don't need to set anything)
-TARGET_SYSTEM ?= $(OS)
-
-# Version numbers
-LIBVER_MAJOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MAJOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_MINOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MINOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_PATCH_SCRIPT:=`sed -n '/define ZSTD_VERSION_RELEASE/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_SCRIPT:= $(LIBVER_MAJOR_SCRIPT).$(LIBVER_MINOR_SCRIPT).$(LIBVER_PATCH_SCRIPT)
-LIBVER_MAJOR := $(shell echo $(LIBVER_MAJOR_SCRIPT))
-LIBVER_MINOR := $(shell echo $(LIBVER_MINOR_SCRIPT))
-LIBVER_PATCH := $(shell echo $(LIBVER_PATCH_SCRIPT))
-LIBVER := $(shell echo $(LIBVER_SCRIPT))
-VERSION?= $(LIBVER)
-CCVER := $(shell $(CC) --version)
-
-# ZSTD_LIB_MINIFY is a helper variable that
-# configures a bunch of other variables to space-optimized defaults.
-ZSTD_LIB_MINIFY ?= 0
-ifneq ($(ZSTD_LIB_MINIFY), 0)
-  HAVE_CC_OZ ?= $(shell echo "" | $(CC) -Oz -x c -c - -o /dev/null 2> /dev/null && echo 1 || echo 0)
-  ZSTD_LEGACY_SUPPORT ?= 0
-  ZSTD_LIB_DEPRECATED ?= 0
-  HUF_FORCE_DECOMPRESS_X1 ?= 1
-  ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT ?= 1
-  ZSTD_NO_INLINE ?= 1
-  ZSTD_STRIP_ERROR_STRINGS ?= 1
-ifneq ($(HAVE_CC_OZ), 0)
-    # Some compilers (clang) support an even more space-optimized setting.
-    CFLAGS += -Oz
-else
-    CFLAGS += -Os
-endif
-  CFLAGS += -fno-stack-protector -fomit-frame-pointer -fno-ident \
-            -DDYNAMIC_BMI2=0 -DNDEBUG
-else
-  CFLAGS += -O3
-endif
-
-DEBUGLEVEL ?= 0
-CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
-ifeq ($(TARGET_SYSTEM),Windows_NT)   # MinGW assumed
-  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
-endif
-DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
-            -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
-            -Wstrict-prototypes -Wundef -Wpointer-arith \
-            -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
-            -Wredundant-decls -Wmissing-prototypes -Wc++-compat
-CFLAGS   += $(DEBUGFLAGS) $(MOREFLAGS)
-FLAGS     = $(CPPFLAGS) $(CFLAGS)
-
-CPPFLAGS_DYNLIB  = -DZSTD_MULTITHREAD # dynamic library build defaults to multi-threaded
-LDFLAGS_DYNLIB   = -pthread
-CPPFLAGS_STATLIB =                    # static library build defaults to single-threaded
-
-HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
-GREP_OPTIONS ?=
-ifeq ($HAVE_COLORNEVER, 1)
-  GREP_OPTIONS += --color=never
-endif
-GREP = grep $(GREP_OPTIONS)
-SED_ERE_OPT ?= -E
-
-ZSTDCOMMON_FILES := $(sort $(wildcard common/*.c))
-ZSTDCOMP_FILES := $(sort $(wildcard compress/*.c))
-ZSTDDECOMP_FILES := $(sort $(wildcard decompress/*.c))
-ZDICT_FILES := $(sort $(wildcard dictBuilder/*.c))
-ZDEPR_FILES := $(sort $(wildcard deprecated/*.c))
-ZSTD_FILES := $(ZSTDCOMMON_FILES)
-
-ifeq ($(findstring GCC,$(CCVER)),GCC)
-decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
-endif
-
 # Modules
 ZSTD_LIB_COMPRESSION ?= 1
 ZSTD_LIB_DECOMPRESSION ?= 1
 ZSTD_LIB_DICTBUILDER ?= 1
 ZSTD_LIB_DEPRECATED ?= 0
 
-# Legacy support
-ZSTD_LEGACY_SUPPORT ?= 5
-ZSTD_LEGACY_MULTITHREADED_API ?= 0
-
-# Build size optimizations
-HUF_FORCE_DECOMPRESS_X1 ?= 0
-HUF_FORCE_DECOMPRESS_X2 ?= 0
-ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT ?= 0
-ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG ?= 0
-ZSTD_NO_INLINE ?= 0
-ZSTD_STRIP_ERROR_STRINGS ?= 0
-
+# Input variables for libzstd.mk
 ifeq ($(ZSTD_LIB_COMPRESSION), 0)
   ZSTD_LIB_DICTBUILDER = 0
   ZSTD_LIB_DEPRECATED = 0
@@ -122,86 +25,46 @@ ifeq ($(ZSTD_LIB_DECOMPRESSION), 0)
   ZSTD_LIB_DEPRECATED = 0
 endif
 
+include libzstd.mk
+
+ZSTD_FILES := $(ZSTD_COMMON_FILES) $(ZSTD_LEGACY_FILES)
+
 ifneq ($(ZSTD_LIB_COMPRESSION), 0)
-  ZSTD_FILES += $(ZSTDCOMP_FILES)
+  ZSTD_FILES += $(ZSTD_COMPRESS_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DECOMPRESSION), 0)
-  ZSTD_FILES += $(ZSTDDECOMP_FILES)
+  ZSTD_FILES += $(ZSTD_DECOMPRESS_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DEPRECATED), 0)
-  ZSTD_FILES += $(ZDEPR_FILES)
+  ZSTD_FILES += $(ZSTD_DEPRECATED_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DICTBUILDER), 0)
-  ZSTD_FILES += $(ZDICT_FILES)
-endif
-
-ifneq ($(HUF_FORCE_DECOMPRESS_X1), 0)
-  CFLAGS += -DHUF_FORCE_DECOMPRESS_X1
-endif
-
-ifneq ($(HUF_FORCE_DECOMPRESS_X2), 0)
-  CFLAGS += -DHUF_FORCE_DECOMPRESS_X2
-endif
-
-ifneq ($(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT), 0)
-  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
-endif
-
-ifneq ($(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG), 0)
-  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
-endif
-
-ifneq ($(ZSTD_NO_INLINE), 0)
-  CFLAGS += -DZSTD_NO_INLINE
+  ZSTD_FILES += $(ZSTD_DICTBUILDER_FILES)
 endif
 
-ifneq ($(ZSTD_STRIP_ERROR_STRINGS), 0)
-  CFLAGS += -DZSTD_STRIP_ERROR_STRINGS
-endif
-
-ifneq ($(ZSTD_LEGACY_MULTITHREADED_API), 0)
-  CFLAGS += -DZSTD_LEGACY_MULTITHREADED_API
-endif
-
-ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
-ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-  ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
-endif
-endif
-CPPFLAGS  += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
-
 ZSTD_LOCAL_SRC := $(notdir $(ZSTD_FILES))
-ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_SRC:.c=.o)
+ZSTD_LOCAL_OBJ0 := $(ZSTD_LOCAL_SRC:.c=.o)
+ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_OBJ0:.S=.o)
 
-ZSTD_SUBDIR := common compress decompress dictBuilder legacy deprecated
-vpath %.c $(ZSTD_SUBDIR)
+VERSION := $(ZSTD_VERSION)
 
-UNAME := $(shell uname)
+# Note: by default, the static library is built single-threaded and dynamic library is built
+# multi-threaded. It is possible to force multi or single threaded builds by appending
+# -mt or -nomt to the build target (like lib-mt for multi-threaded, lib-nomt for single-threaded).
+.PHONY: default
+default: lib-release
+
+CPPFLAGS_DYNLIB  = -DZSTD_MULTITHREAD # dynamic library build defaults to multi-threaded
+LDFLAGS_DYNLIB   = -pthread
+CPPFLAGS_STATLIB =                    # static library build defaults to single-threaded
 
-ifndef BUILD_DIR
-ifeq ($(UNAME), Darwin)
-  ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 0)
-    HASH ?= md5
-  endif
-else ifeq ($(UNAME), FreeBSD)
-  HASH ?= gmd5sum
-else ifeq ($(UNAME), NetBSD)
-  HASH ?= md5 -n
-else ifeq ($(UNAME), OpenBSD)
-  HASH ?= md5
-endif
-HASH ?= md5sum
 
-HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " " )
-HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
-ifeq ($(HAVE_HASH),0)
-  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
-  BUILD_DIR := obj/generic_noconf
+ifeq ($(findstring GCC,$(CCVER)),GCC)
+decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
 endif
-endif # BUILD_DIR
 
 
 # macOS linker doesn't support -soname, and use different extension
@@ -218,13 +81,6 @@ else
   SHARED_EXT_VER = $(SHARED_EXT).$(LIBVER)
 endif
 
-SET_CACHE_DIRECTORY = \
-   +$(MAKE) --no-print-directory $@ \
-    BUILD_DIR=obj/$(HASH_DIR) \
-    CPPFLAGS="$(CPPFLAGS)" \
-    CFLAGS="$(CFLAGS)" \
-    LDFLAGS="$(LDFLAGS)"
-
 
 .PHONY: all
 all: lib
@@ -233,6 +89,13 @@ all: lib
 .PHONY: libzstd.a  # must be run every time
 libzstd.a: CPPFLAGS += $(CPPFLAGS_STATLIB)
 
+SET_CACHE_DIRECTORY = \
+   +$(MAKE) --no-print-directory $@ \
+    BUILD_DIR=obj/$(HASH_DIR) \
+    CPPFLAGS="$(CPPFLAGS)" \
+    CFLAGS="$(CFLAGS)" \
+    LDFLAGS="$(LDFLAGS)"
+
 ifndef BUILD_DIR
 # determine BUILD_DIR from compilation flags
 
@@ -343,6 +206,14 @@ $(ZSTD_STATLIB_DIR)/%.o : %.c $(ZSTD_STATLIB_DIR)/%.d | $(ZSTD_STATLIB_DIR)
 	@echo CC $@
 	$(COMPILE.c) $(DEPFLAGS) $(ZSTD_STATLIB_DIR)/$*.d $(OUTPUT_OPTION) $<
 
+$(ZSTD_DYNLIB_DIR)/%.o : %.S | $(ZSTD_DYNLIB_DIR)
+	@echo AS $@
+	$(COMPILE.c) $(OUTPUT_OPTION) $<
+
+$(ZSTD_STATLIB_DIR)/%.o : %.S | $(ZSTD_STATLIB_DIR)
+	@echo AS $@
+	$(COMPILE.c) $(OUTPUT_OPTION) $<
+
 MKDIR ?= mkdir
 $(BUILD_DIR) $(ZSTD_DYNLIB_DIR) $(ZSTD_STATLIB_DIR):
 	$(MKDIR) -p $@

lib/compress/huf_compress.c

@@ -445,7 +445,7 @@ typedef struct {
 /* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
  * Strategy is to use as many buckets as possible for representing distinct
  * counts while using the remainder to represent all "large" counts.
- * 
+ *
  * To satisfy this requirement for 192 buckets, we can do the following:
  * Let buckets 0-166 represent distinct counts of [0, 166]
  * Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
@@ -517,7 +517,7 @@ static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high)
 }
 
 /* Classic quicksort by descending with partially iterative calls
- * to reduce worst case callstack size. 
+ * to reduce worst case callstack size.
  */
 static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
     int const kInsertionSortThreshold = 8;
@@ -813,6 +813,7 @@ FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int id
         assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
         /* We didn't overwrite any bits in the bit container. */
         assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+        (void)dirtyBits;
     }
 #endif
 }