ddidderr/zstd-rs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge branch 'dev' into tomerge2051

Browse Source
This commit is contained in:
Yann Collet
2021-12-01 15:29:49 -08:00
parent c8767a4c74 9b97fdf74f
commit 3f64b31585
146 changed files with 8594 additions and 4430 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/Makefile
+45 -84
View File
@@ -100,18 +100,7 @@ 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_SHORT ?= 0
ZSTD_FORCE_DECOMPRESS_LONG ?= 0
ZSTD_NO_INLINE ?= 0
ZSTD_STRIP_ERROR_STRINGS ?= 0
# Input variables for libzstd.mk
ifeq ($(ZSTD_LIB_COMPRESSION), 0)
ZSTD_LIB_DICTBUILDER = 0
ZSTD_LIB_DEPRECATED = 0
@@ -122,86 +111,46 @@ ifeq ($(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)
ZSTD_FILES += $(ZSTD_DICTBUILDER_FILES)
endif
ifneq ($(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_SHORT), 0)
CFLAGS += -DZSTD_FORCE_DECOMPRESS_SHORT
endif
ifneq ($(ZSTD_FORCE_DECOMPRESS_LONG), 0)
CFLAGS += -DZSTD_FORCE_DECOMPRESS_LONG
endif
ifneq ($(ZSTD_NO_INLINE), 0)
CFLAGS += -DZSTD_NO_INLINE
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 $$?), 0)
ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
endif
endif
CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
ZSTD_LOCAL_SRC := $(notdir $(ZSTD_FILES))
ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_SRC:.c=.o)
ZSTD_LOCAL_OBJ0 := $(ZSTD_LOCAL_SRC:.c=.o)
ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_OBJ0:.S=.o)
ZSTD_SUBDIR := common compress decompress dictBuilder legacy deprecated
vpath %.c $(ZSTD_SUBDIR)
VERSION := $(ZSTD_VERSION)
UNAME := $(shell uname)
# Note: by default, the static library is built single-threaded and dynamic library is built
# multi-threaded. It is possible to force multi or single threaded builds by appending
# -mt or -nomt to the build target (like lib-mt for multi-threaded, lib-nomt for single-threaded).
.PHONY: default
default: lib-release
ifndef BUILD_DIR
ifeq ($(UNAME), Darwin)
ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 0)
HASH ?= md5
endif
else ifeq ($(UNAME), FreeBSD)
HASH ?= gmd5sum
else ifeq ($(UNAME), NetBSD)
HASH ?= md5 -n
else ifeq ($(UNAME), OpenBSD)
HASH ?= md5
CPPFLAGS_DYNLIB = -DZSTD_MULTITHREAD # dynamic library build defaults to multi-threaded
LDFLAGS_DYNLIB = -pthread
CPPFLAGS_STATLIB = # static library build defaults to single-threaded
ifeq ($(findstring GCC,$(CCVER)),GCC)
decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
endif
HASH ?= md5sum
HASH_DIR = conf_$(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
endif
endif # BUILD_DIR
# macOS linker doesn't support -soname, and use different extension
@@ -212,19 +161,16 @@ ifeq ($(UNAME), Darwin)
SHARED_EXT_VER = $(LIBVER).$(SHARED_EXT)
SONAME_FLAGS = -install_name $(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR) -compatibility_version $(LIBVER_MAJOR) -current_version $(LIBVER)
else
SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
ifeq ($(UNAME), AIX)
SONAME_FLAGS =
else
SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
endif
SHARED_EXT = so
SHARED_EXT_MAJOR = $(SHARED_EXT).$(LIBVER_MAJOR)
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 +179,13 @@ 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 +296,14 @@ $(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 $@
@@ -374,7 +335,7 @@ clean:
#-----------------------------------------------------------------------------
# make install is validated only for below listed environments
#-----------------------------------------------------------------------------
ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku AIX))
all: libzstd.pc
lib/README.md
+7 -1
View File
@@ -125,7 +125,7 @@ The file structure is designed to make this selection manually achievable for an
`ZSTD_getErrorName` (implied by `ZSTD_LIB_MINIFY`).
Finally, when integrating into your application, make sure you're doing link-
time optimation and unused symbol garbage collection (via some combination of,
time optimization and unused symbol garbage collection (via some combination of,
e.g., `-flto`, `-ffat-lto-objects`, `-fuse-linker-plugin`,
`-ffunction-sections`, `-fdata-sections`, `-fmerge-all-constants`,
`-Wl,--gc-sections`, `-Wl,-z,norelro`, and an archiver that understands
@@ -155,6 +155,12 @@ The file structure is designed to make this selection manually achievable for an
- The build macro `ZSTD_NO_INTRINSICS` can be defined to disable all explicit intrinsics.
Compiler builtins are still used.
- The build macro `ZSTD_DECODER_INTERNAL_BUFFER` can be set to control
the amount of extra memory used during decompression to store literals.
This defaults to 64kB. Reducing this value reduces the memory footprint of
`ZSTD_DCtx` decompression contexts,
but might also result in a small decompression speed cost.
#### Windows : using MinGW+MSYS to create DLL
lib/common/bitstream.h
+24 -9
View File
@@ -143,10 +143,16 @@ MEM_STATIC unsigned BIT_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
# if STATIC_BMI2 == 1
return _lzcnt_u32(val) ^ 31;
return _lzcnt_u32(val) ^ 31;
# else
unsigned long r = 0;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
if (val != 0) {
unsigned long r;
_BitScanReverse(&r, val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
@@ -293,22 +299,22 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
switch(srcSize)
{
case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
/* fall-through */
ZSTD_FALLTHROUGH;
case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
/* fall-through */
ZSTD_FALLTHROUGH;
case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
/* fall-through */
ZSTD_FALLTHROUGH;
case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
/* fall-through */
ZSTD_FALLTHROUGH;
case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
/* fall-through */
ZSTD_FALLTHROUGH;
case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) << 8;
/* fall-through */
ZSTD_FALLTHROUGH;
default: break;
}
@@ -332,7 +338,16 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 c
U32 const regMask = sizeof(bitContainer)*8 - 1;
/* if start > regMask, bitstream is corrupted, and result is undefined */
assert(nbBits < BIT_MASK_SIZE);
/* x86 transform & ((1 << nbBits) - 1) to bzhi instruction, it is better
* than accessing memory. When bmi2 instruction is not present, we consider
* such cpus old (pre-Haswell, 2013) and their performance is not of that
* importance.
*/
#if defined(__x86_64__) || defined(_M_X86)
return (bitContainer >> (start & regMask)) & ((((U64)1) << nbBits) - 1);
#else
return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
#endif
}
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
lib/common/compiler.h
+70 -4
View File
@@ -40,7 +40,7 @@
/**
On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
This explictly marks such functions as __cdecl so that the code will still compile
This explicitly marks such functions as __cdecl so that the code will still compile
if a CC other than __cdecl has been made the default.
*/
#if defined(_MSC_VER)
@@ -101,6 +101,13 @@
# define TARGET_ATTRIBUTE(target)
#endif
/* Target attribute for BMI2 dynamic dispatch.
* Enable lzcnt, bmi, and bmi2.
* We test for bmi1 & bmi2. lzcnt is included in bmi1.
*/
#define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2")
/* Enable runtime BMI2 dispatch based on the CPU.
* Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
*/
@@ -108,7 +115,7 @@
#if ((defined(__clang__) && __has_attribute(__target__)) \
|| (defined(__GNUC__) \
&& (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
&& (defined(__x86_64__) || defined(_M_X86)) \
&& (defined(__x86_64__) || defined(_M_X64)) \
&& !defined(__BMI2__)
# define DYNAMIC_BMI2 1
#else
@@ -150,8 +157,9 @@
}
/* vectorization
* older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */
#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__)
* older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
* and some compilers, like Intel ICC and MCST LCC, do not support it at all. */
#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__)
# if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
# define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
# else
@@ -197,6 +205,22 @@
#define STATIC_BMI2 0
#endif
/* compile time determination of SIMD support */
#if !defined(ZSTD_NO_INTRINSICS)
# if defined(__SSE2__) || defined(_M_AMD64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
# define ZSTD_ARCH_X86_SSE2
# endif
# if defined(__ARM_NEON) || defined(_M_ARM64)
# define ZSTD_ARCH_ARM_NEON
# endif
#
# if defined(ZSTD_ARCH_X86_SSE2)
# include <emmintrin.h>
# elif defined(ZSTD_ARCH_ARM_NEON)
# include <arm_neon.h>
# endif
#endif
/* compat. with non-clang compilers */
#ifndef __has_builtin
# define __has_builtin(x) 0
@@ -207,6 +231,39 @@
# define __has_feature(x) 0
#endif
/* C-language Attributes are added in C23. */
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
# define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
#else
# define ZSTD_HAS_C_ATTRIBUTE(x) 0
#endif
/* Only use C++ attributes in C++. Some compilers report support for C++
* attributes when compiling with C.
*/
#if defined(__cplusplus) && defined(__has_cpp_attribute)
# define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
#else
# define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
#endif
/* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
* - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
* - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
* - Else: __attribute__((__fallthrough__))
*/
#ifndef ZSTD_FALLTHROUGH
# if ZSTD_HAS_C_ATTRIBUTE(fallthrough)
# define ZSTD_FALLTHROUGH [[fallthrough]]
# elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough)
# define ZSTD_FALLTHROUGH [[fallthrough]]
# elif __has_attribute(__fallthrough__)
# define ZSTD_FALLTHROUGH __attribute__((__fallthrough__))
# else
# define ZSTD_FALLTHROUGH
# endif
#endif
/* detects whether we are being compiled under msan */
#ifndef ZSTD_MEMORY_SANITIZER
# if __has_feature(memory_sanitizer)
@@ -216,6 +273,15 @@
# endif
#endif
/* detects whether we are being compiled undef dfsan */
#ifndef ZSTD_DATAFLOW_SANITIZER
# if __has_feature(dataflow_sanitizer)
# define ZSTD_DATAFLOW_SANITIZER 1
# else
# define ZSTD_DATAFLOW_SANITIZER 0
# endif
#endif
#if ZSTD_MEMORY_SANITIZER
/* Not all platforms that support msan provide sanitizers/msan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
lib/common/entropy_common.c
+11 -5
View File
@@ -43,8 +43,14 @@ static U32 FSE_ctz(U32 val)
assert(val != 0);
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
return _BitScanForward(&r, val) ? (unsigned)r : 0;
if (val != 0) {
unsigned long r;
_BitScanForward(&r, val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* GCC Intrinsic */
return __builtin_ctz(val);
# elif defined(__ICCARM__) /* IAR Intrinsic */
@@ -217,7 +223,7 @@ static size_t FSE_readNCount_body_default(
}
#if DYNAMIC_BMI2
TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2(
BMI2_TARGET_ATTRIBUTE static size_t FSE_readNCount_body_bmi2(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
@@ -299,7 +305,7 @@ HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
weightTotal = 0;
{ U32 n; for (n=0; n<oSize; n++) {
if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
if (huffWeight[n] > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
rankStats[huffWeight[n]]++;
weightTotal += (1 << huffWeight[n]) >> 1;
} }
@@ -337,7 +343,7 @@ static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* r
}
#if DYNAMIC_BMI2
static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
static BMI2_TARGET_ATTRIBUTE size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
lib/common/error_private.h
+79
View File
@@ -22,6 +22,8 @@ extern "C" {
* Dependencies
******************************************/
#include "../zstd_errors.h" /* enum list */
#include "compiler.h"
#include "debug.h"
#include "zstd_deps.h" /* size_t */
@@ -73,6 +75,83 @@ ERR_STATIC const char* ERR_getErrorName(size_t code)
return ERR_getErrorString(ERR_getErrorCode(code));
}
/**
* Ignore: this is an internal helper.
*
* This is a helper function to help force C99-correctness during compilation.
* Under strict compilation modes, variadic macro arguments can't be empty.
* However, variadic function arguments can be. Using a function therefore lets
* us statically check that at least one (string) argument was passed,
* independent of the compilation flags.
*/
static INLINE_KEYWORD UNUSED_ATTR
void _force_has_format_string(const char *format, ...) {
(void)format;
}
/**
* Ignore: this is an internal helper.
*
* We want to force this function invocation to be syntactically correct, but
* we don't want to force runtime evaluation of its arguments.
*/
#define _FORCE_HAS_FORMAT_STRING(...) \
if (0) { \
_force_has_format_string(__VA_ARGS__); \
}
#define ERR_QUOTE(str) #str
/**
* Return the specified error if the condition evaluates to true.
*
* In debug modes, prints additional information.
* In order to do that (particularly, printing the conditional that failed),
* this can't just wrap RETURN_ERROR().
*/
#define RETURN_ERROR_IF(cond, err, ...) \
if (cond) { \
RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
__FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
}
/**
* Unconditionally return the specified error.
*
* In debug modes, prints additional information.
*/
#define RETURN_ERROR(err, ...) \
do { \
RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
__FILE__, __LINE__, ERR_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
} while(0);
/**
* If the provided expression evaluates to an error code, returns that error code.
*
* In debug modes, prints additional information.
*/
#define FORWARD_IF_ERROR(err, ...) \
do { \
size_t const err_code = (err); \
if (ERR_isError(err_code)) { \
RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
__FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return err_code; \
} \
} while(0);
#if defined (__cplusplus)
}
#endif
lib/common/fse.h
+2 -1
View File
@@ -336,8 +336,9 @@ size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
/* FSE_buildCTable_wksp() :
* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
* `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
* See FSE_buildCTable_wksp() for breakdown of workspace usage.
*/
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (maxSymbolValue + 2 + (1ull << (tableLog - 2)))
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (((maxSymbolValue + 2) + (1ull << (tableLog)))/2 + sizeof(U64)/sizeof(U32) /* additional 8 bytes for potential table overwrite */)
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
lib/common/fse_decompress.c
+1 -1
View File
@@ -365,7 +365,7 @@ static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, co
}
#if DYNAMIC_BMI2
TARGET_ATTRIBUTE("bmi2") static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
BMI2_TARGET_ATTRIBUTE static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
{
return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
}
lib/common/huf.h
+24 -22
View File
@@ -89,9 +89,9 @@ HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
/** HUF_compress4X_wksp() :
* Same as HUF_compress2(), but uses externally allocated `workSpace`.
* `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
#define HUF_WORKSPACE_SIZE ((6 << 10) + 256)
#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
* `workspace` must be at least as large as HUF_WORKSPACE_SIZE */
#define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
@@ -116,11 +116,11 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
/* *** Constants *** */
#define HUF_TABLELOG_MAX 12 /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
#define HUF_TABLELOG_MAX 12 /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
#define HUF_TABLELOG_DEFAULT 11 /* default tableLog value when none specified */
#define HUF_SYMBOLVALUE_MAX 255
#define HUF_TABLELOG_ABSOLUTEMAX 15 /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
#define HUF_TABLELOG_ABSOLUTEMAX 12 /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
#if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
# error "HUF_TABLELOG_MAX is too large !"
#endif
@@ -136,15 +136,11 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
/* static allocation of HUF's Compression Table */
/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
struct HUF_CElt_s {
U16 val;
BYTE nbBits;
}; /* typedef'd to HUF_CElt */
typedef struct HUF_CElt_s HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_U32(maxSymbolValue) ((maxSymbolValue)+1) /* Use tables of U32, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
typedef size_t HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_ST(maxSymbolValue) ((maxSymbolValue)+2) /* Use tables of size_t, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
HUF_CElt name[HUF_CTABLE_SIZE_U32(maxSymbolValue)] /* no final ; */
HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */
/* static allocation of HUF's DTable */
typedef U32 HUF_DTable;
@@ -194,6 +190,7 @@ size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSym
size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
size_t HUF_compress4X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2);
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
@@ -206,12 +203,13 @@ typedef enum {
* Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
* If it uses hufTable it does not modify hufTable or repeat.
* If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
* If preferRepeat then the old table will always be used if valid. */
* If preferRepeat then the old table will always be used if valid.
* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize, /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible);
/** HUF_buildCTable_wksp() :
* Same as HUF_buildCTable(), but using externally allocated scratch buffer.
@@ -249,11 +247,10 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
* Loading a CTable saved with HUF_writeCTable() */
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
/** HUF_getNbBits() :
/** HUF_getNbBitsFromCTable() :
* Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
* Note 1 : is not inlined, as HUF_CElt definition is private
* Note 2 : const void* used, so that it can provide a statically allocated table as argument (which uses type U32) */
U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue);
* Note 1 : is not inlined, as HUF_CElt definition is private */
U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
/*
* HUF_decompress() does the following:
@@ -305,18 +302,20 @@ size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* c
/* ====================== */
size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize); /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize); /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U64 U64 */
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
size_t HUF_compress1X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2);
/** HUF_compress1X_repeat() :
* Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
* If it uses hufTable it does not modify hufTable or repeat.
* If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
* If preferRepeat then the old table will always be used if valid. */
* If preferRepeat then the old table will always be used if valid.
* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize, /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible);
size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /* single-symbol decoder */
#ifndef HUF_FORCE_DECOMPRESS_X1
@@ -354,6 +353,9 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
#endif
#ifndef HUF_FORCE_DECOMPRESS_X1
size_t HUF_readDTableX2_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
#endif
#endif /* HUF_STATIC_LINKING_ONLY */
lib/common/mem.h
+14
View File
@@ -153,8 +153,22 @@ MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
MEM_STATIC unsigned MEM_isLittleEndian(void)
{
#if defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
return 1;
#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
return 0;
#elif defined(__clang__) && __LITTLE_ENDIAN__
return 1;
#elif defined(__clang__) && __BIG_ENDIAN__
return 0;
#elif defined(_MSC_VER) && (_M_AMD64 || _M_IX86)
return 1;
#elif defined(__DMC__) && defined(_M_IX86)
return 1;
#else
const union { U32 u; BYTE c[4]; } one = { 1 }; /* don't use static : performance detrimental */
return one.c[0];
#endif
}
#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
lib/common/zstd_internal.h
+85 -86
View File
@@ -19,10 +19,8 @@
/*-*************************************
* Dependencies
***************************************/
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
#include <arm_neon.h>
#endif
#include "compiler.h"
#include "cpu.h"
#include "mem.h"
#include "debug.h" /* assert, DEBUGLOG, RAWLOG, g_debuglevel */
#include "error_private.h"
@@ -60,81 +58,7 @@ extern "C" {
#undef MAX
#define MIN(a,b) ((a)<(b) ? (a) : (b))
#define MAX(a,b) ((a)>(b) ? (a) : (b))
/**
* Ignore: this is an internal helper.
*
* This is a helper function to help force C99-correctness during compilation.
* Under strict compilation modes, variadic macro arguments can't be empty.
* However, variadic function arguments can be. Using a function therefore lets
* us statically check that at least one (string) argument was passed,
* independent of the compilation flags.
*/
static INLINE_KEYWORD UNUSED_ATTR
void _force_has_format_string(const char *format, ...) {
(void)format;
}
/**
* Ignore: this is an internal helper.
*
* We want to force this function invocation to be syntactically correct, but
* we don't want to force runtime evaluation of its arguments.
*/
#define _FORCE_HAS_FORMAT_STRING(...) \
if (0) { \
_force_has_format_string(__VA_ARGS__); \
}
/**
* Return the specified error if the condition evaluates to true.
*
* In debug modes, prints additional information.
* In order to do that (particularly, printing the conditional that failed),
* this can't just wrap RETURN_ERROR().
*/
#define RETURN_ERROR_IF(cond, err, ...) \
if (cond) { \
RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
__FILE__, __LINE__, ZSTD_QUOTE(cond), ZSTD_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
}
/**
* Unconditionally return the specified error.
*
* In debug modes, prints additional information.
*/
#define RETURN_ERROR(err, ...) \
do { \
RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
__FILE__, __LINE__, ZSTD_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
} while(0);
/**
* If the provided expression evaluates to an error code, returns that error code.
*
* In debug modes, prints additional information.
*/
#define FORWARD_IF_ERROR(err, ...) \
do { \
size_t const err_code = (err); \
if (ERR_isError(err_code)) { \
RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
__FILE__, __LINE__, ZSTD_QUOTE(err), ERR_getErrorName(err_code)); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return err_code; \
} \
} while(0);
#define BOUNDED(min,val,max) (MAX(min,MIN(val,max)))
/*-*************************************
@@ -247,19 +171,25 @@ static UNUSED_ATTR const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
* Shared functions to include for inlining
*********************************************/
static void ZSTD_copy8(void* dst, const void* src) {
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
#if defined(ZSTD_ARCH_ARM_NEON)
vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
#else
ZSTD_memcpy(dst, src, 8);
#endif
}
#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
/* Need to use memmove here since the literal buffer can now be located within
the dst buffer. In circumstances where the op "catches up" to where the
literal buffer is, there can be partial overlaps in this call on the final
copy if the literal is being shifted by less than 16 bytes. */
static void ZSTD_copy16(void* dst, const void* src) {
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
#if defined(ZSTD_ARCH_ARM_NEON)
vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
#elif defined(ZSTD_ARCH_X86_SSE2)
_mm_storeu_si128((__m128i*)dst, _mm_loadu_si128((const __m128i*)src));
#else
ZSTD_memcpy(dst, src, 16);
ZSTD_memmove(dst, src, 16);
#endif
}
#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
@@ -288,8 +218,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
BYTE* op = (BYTE*)dst;
BYTE* const oend = op + length;
assert(diff >= 8 || (ovtype == ZSTD_no_overlap && diff <= -WILDCOPY_VECLEN));
if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
/* Handle short offset copies. */
do {
@@ -436,8 +364,14 @@ MEM_STATIC U32 ZSTD_highbit32(U32 val) /* compress, dictBuilder, decodeCorpus
# if STATIC_BMI2 == 1
return _lzcnt_u32(val)^31;
# else
unsigned long r=0;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
if (val != 0) {
unsigned long r;
_BitScanReverse(&r, val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* GCC Intrinsic */
return __builtin_clz (val) ^ 31;
@@ -456,6 +390,63 @@ MEM_STATIC U32 ZSTD_highbit32(U32 val) /* compress, dictBuilder, decodeCorpus
}
}
/**
* Counts the number of trailing zeros of a `size_t`.
* Most compilers should support CTZ as a builtin. A backup
* implementation is provided if the builtin isn't supported, but
* it may not be terribly efficient.
*/
MEM_STATIC unsigned ZSTD_countTrailingZeros(size_t val)
{
if (MEM_64bits()) {
# if defined(_MSC_VER) && defined(_WIN64)
# if STATIC_BMI2
return _tzcnt_u64(val);
# else
if (val != 0) {
unsigned long r;
_BitScanForward64(&r, (U64)val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return __builtin_ctzll((U64)val);
# else
static const int DeBruijnBytePos[64] = { 0, 1, 2, 7, 3, 13, 8, 19,
4, 25, 14, 28, 9, 34, 20, 56,
5, 17, 26, 54, 15, 41, 29, 43,
10, 31, 38, 35, 21, 45, 49, 57,
63, 6, 12, 18, 24, 27, 33, 55,
16, 53, 40, 42, 30, 37, 44, 48,
62, 11, 23, 32, 52, 39, 36, 47,
61, 22, 51, 46, 60, 50, 59, 58 };
return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
# endif
} else { /* 32 bits */
# if defined(_MSC_VER)
if (val != 0) {
unsigned long r;
_BitScanForward(&r, (U32)val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return __builtin_ctz((U32)val);
# else
static const int DeBruijnBytePos[32] = { 0, 1, 28, 2, 29, 14, 24, 3,
30, 22, 20, 15, 25, 17, 4, 8,
31, 27, 13, 23, 21, 19, 16, 7,
26, 12, 18, 6, 11, 5, 10, 9 };
return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
# endif
}
}
/* ZSTD_invalidateRepCodes() :
* ensures next compression will not use repcodes from previous block.
@@ -482,6 +473,14 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
const void* src, size_t srcSize);
/**
* @returns true iff the CPU supports dynamic BMI2 dispatch.
*/
MEM_STATIC int ZSTD_cpuSupportsBmi2(void)
{
ZSTD_cpuid_t cpuid = ZSTD_cpuid();
return ZSTD_cpuid_bmi1(cpuid) && ZSTD_cpuid_bmi2(cpuid);
}
#if defined (__cplusplus)
}
lib/common/zstd_trace.h
+12 -3
View File
@@ -17,10 +17,19 @@ extern "C" {
#include <stddef.h>
/* weak symbol support */
#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && defined(__GNUC__) && \
/* weak symbol support
* For now, enable conservatively:
* - Only GNUC
* - Only ELF
* - Only x86-64 and i386
* Also, explicitly disable on platforms known not to work so they aren't
* forgotten in the future.
*/
#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && \
defined(__GNUC__) && defined(__ELF__) && \
(defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86)) && \
!defined(__APPLE__) && !defined(_WIN32) && !defined(__MINGW32__) && \
!defined(__CYGWIN__)
!defined(__CYGWIN__) && !defined(_AIX)
# define ZSTD_HAVE_WEAK_SYMBOLS 1
#else
# define ZSTD_HAVE_WEAK_SYMBOLS 0
lib/compress/clevels.h
+135
View File
@@ -0,0 +1,135 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_CLEVELS_H
#define ZSTD_CLEVELS_H
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_compressionParameters */
#include "../zstd.h"
/*-===== Pre-defined compression levels =====-*/
#define ZSTD_MAX_CLEVEL 22
#define ZSTD_MAX_32BIT_CLEVEL 21
#ifdef __GNUC__
__attribute__((__unused__))
#endif
static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
{ /* "default" - for any srcSize > 256 KB */
/* W, C, H, S, L, TL, strat */
{ 19, 12, 13, 1, 6, 1, ZSTD_fast }, /* base for negative levels */
{ 19, 13, 14, 1, 7, 0, ZSTD_fast }, /* level 1 */
{ 20, 15, 16, 1, 6, 0, ZSTD_fast }, /* level 2 */
{ 21, 16, 17, 1, 5, 0, ZSTD_dfast }, /* level 3 */
{ 21, 18, 18, 1, 5, 0, ZSTD_dfast }, /* level 4 */
{ 21, 18, 19, 4, 5, 2, ZSTD_greedy }, /* level 5 */
{ 21, 19, 20, 5, 5, 4, ZSTD_greedy }, /* level 6 */
{ 21, 19, 20, 4, 5, 8, ZSTD_lazy }, /* level 7 */
{ 21, 19, 20, 5, 5, 16, ZSTD_lazy }, /* level 8 */
{ 21, 20, 21, 4, 5, 16, ZSTD_lazy2 }, /* level 9 */
{ 22, 21, 22, 4, 5, 16, ZSTD_lazy2 }, /* level 10 */
{ 22, 21, 22, 5, 5, 16, ZSTD_lazy2 }, /* level 11 */
{ 22, 21, 22, 6, 5, 32, ZSTD_lazy2 }, /* level 12 */
{ 22, 22, 22, 4, 5, 32, ZSTD_btlazy2 }, /* level 13 */
{ 22, 22, 23, 5, 5, 32, ZSTD_btlazy2 }, /* level 14 */
{ 22, 23, 23, 6, 5, 32, ZSTD_btlazy2 }, /* level 15 */
{ 22, 22, 22, 5, 5, 48, ZSTD_btopt }, /* level 16 */
{ 23, 23, 22, 5, 4, 64, ZSTD_btopt }, /* level 17 */
{ 23, 23, 22, 6, 3, 64, ZSTD_btultra }, /* level 18 */
{ 23, 24, 22, 7, 3,256, ZSTD_btultra2}, /* level 19 */
{ 25, 25, 23, 7, 3,256, ZSTD_btultra2}, /* level 20 */
{ 26, 26, 24, 7, 3,512, ZSTD_btultra2}, /* level 21 */
{ 27, 27, 25, 9, 3,999, ZSTD_btultra2}, /* level 22 */
},
{ /* for srcSize <= 256 KB */
/* W, C, H, S, L, T, strat */
{ 18, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 18, 13, 14, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 18, 14, 14, 1, 5, 0, ZSTD_dfast }, /* level 2 */
{ 18, 16, 16, 1, 4, 0, ZSTD_dfast }, /* level 3 */
{ 18, 16, 17, 3, 5, 2, ZSTD_greedy }, /* level 4.*/
{ 18, 17, 18, 5, 5, 2, ZSTD_greedy }, /* level 5.*/
{ 18, 18, 19, 3, 5, 4, ZSTD_lazy }, /* level 6.*/
{ 18, 18, 19, 4, 4, 4, ZSTD_lazy }, /* level 7 */
{ 18, 18, 19, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 18, 18, 19, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 18, 18, 19, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 18, 18, 19, 5, 4, 12, ZSTD_btlazy2 }, /* level 11.*/
{ 18, 19, 19, 7, 4, 12, ZSTD_btlazy2 }, /* level 12.*/
{ 18, 18, 19, 4, 4, 16, ZSTD_btopt }, /* level 13 */
{ 18, 18, 19, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 18, 18, 19, 6, 3,128, ZSTD_btopt }, /* level 15.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra2}, /* level 18.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 18, 19, 19, 10, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 18, 19, 19, 12, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 18, 19, 19, 13, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 128 KB */
/* W, C, H, S, L, T, strat */
{ 17, 12, 12, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 17, 12, 13, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 17, 13, 15, 1, 5, 0, ZSTD_fast }, /* level 2 */
{ 17, 15, 16, 2, 5, 0, ZSTD_dfast }, /* level 3 */
{ 17, 17, 17, 2, 4, 0, ZSTD_dfast }, /* level 4 */
{ 17, 16, 17, 3, 4, 2, ZSTD_greedy }, /* level 5 */
{ 17, 16, 17, 3, 4, 4, ZSTD_lazy }, /* level 6 */
{ 17, 16, 17, 3, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 17, 16, 17, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 17, 16, 17, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 17, 16, 17, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 17, 17, 17, 5, 4, 8, ZSTD_btlazy2 }, /* level 11 */
{ 17, 18, 17, 7, 4, 12, ZSTD_btlazy2 }, /* level 12 */
{ 17, 18, 17, 3, 4, 12, ZSTD_btopt }, /* level 13.*/
{ 17, 18, 17, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 17, 18, 17, 6, 3,256, ZSTD_btopt }, /* level 15.*/
{ 17, 18, 17, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 17, 18, 17, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 17, 18, 17, 10, 3,512, ZSTD_btultra }, /* level 18.*/
{ 17, 18, 17, 5, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 17, 18, 17, 7, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 17, 18, 17, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 17, 18, 17, 11, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 16 KB */
/* W, C, H, S, L, T, strat */
{ 14, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 14, 14, 15, 1, 5, 0, ZSTD_fast }, /* level 1 */
{ 14, 14, 15, 1, 4, 0, ZSTD_fast }, /* level 2 */
{ 14, 14, 15, 2, 4, 0, ZSTD_dfast }, /* level 3 */
{ 14, 14, 14, 4, 4, 2, ZSTD_greedy }, /* level 4 */
{ 14, 14, 14, 3, 4, 4, ZSTD_lazy }, /* level 5.*/
{ 14, 14, 14, 4, 4, 8, ZSTD_lazy2 }, /* level 6 */
{ 14, 14, 14, 6, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 14, 14, 14, 8, 4, 8, ZSTD_lazy2 }, /* level 8.*/
{ 14, 15, 14, 5, 4, 8, ZSTD_btlazy2 }, /* level 9.*/
{ 14, 15, 14, 9, 4, 8, ZSTD_btlazy2 }, /* level 10.*/
{ 14, 15, 14, 3, 4, 12, ZSTD_btopt }, /* level 11.*/
{ 14, 15, 14, 4, 3, 24, ZSTD_btopt }, /* level 12.*/
{ 14, 15, 14, 5, 3, 32, ZSTD_btultra }, /* level 13.*/
{ 14, 15, 15, 6, 3, 64, ZSTD_btultra }, /* level 14.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra }, /* level 15.*/
{ 14, 15, 15, 5, 3, 48, ZSTD_btultra2}, /* level 16.*/
{ 14, 15, 15, 6, 3,128, ZSTD_btultra2}, /* level 17.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra2}, /* level 18.*/
{ 14, 15, 15, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 14, 15, 15, 8, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 14, 15, 15, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 14, 15, 15, 10, 3,999, ZSTD_btultra2}, /* level 22.*/
},
};
#endif /* ZSTD_CLEVELS_H */
lib/compress/fse_compress.c
+63 -27
View File
@@ -75,13 +75,14 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
U32 const step = FSE_TABLESTEP(tableSize);
U32 const maxSV1 = maxSymbolValue+1;
U32* cumul = (U32*)workSpace;
FSE_FUNCTION_TYPE* tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSymbolValue + 2));
U16* cumul = (U16*)workSpace; /* size = maxSV1 */
FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1)); /* size = tableSize */
U32 highThreshold = tableSize-1;
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* Must be 4 byte aligned */
assert(((size_t)workSpace & 1) == 0); /* Must be 2 bytes-aligned */
if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
/* CTable header */
tableU16[-2] = (U16) tableLog;
@@ -98,20 +99,61 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
/* symbol start positions */
{ U32 u;
cumul[0] = 0;
for (u=1; u <= maxSymbolValue+1; u++) {
for (u=1; u <= maxSV1; u++) {
if (normalizedCounter[u-1]==-1) { /* Low proba symbol */
cumul[u] = cumul[u-1] + 1;
tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
} else {
cumul[u] = cumul[u-1] + normalizedCounter[u-1];
assert(normalizedCounter[u-1] >= 0);
cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
assert(cumul[u] >= cumul[u-1]); /* no overflow */
} }
cumul[maxSymbolValue+1] = tableSize+1;
cumul[maxSV1] = (U16)(tableSize+1);
}
/* Spread symbols */
{ U32 position = 0;
if (highThreshold == tableSize - 1) {
/* Case for no low prob count symbols. Lay down 8 bytes at a time
* to reduce branch misses since we are operating on a small block
*/
BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
{ U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
assert(n>=0);
pos += (size_t)n;
}
}
/* Spread symbols across the table. Lack of lowprob symbols means that
* we don't need variable sized inner loop, so we can unroll the loop and
* reduce branch misses.
*/
{ size_t position = 0;
size_t s;
size_t const unroll = 2; /* Experimentally determined optimal unroll */
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableSymbol[uPosition] = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0); /* Must have initialized all positions */
}
} else {
U32 position = 0;
U32 symbol;
for (symbol=0; symbol<=maxSymbolValue; symbol++) {
for (symbol=0; symbol<maxSV1; symbol++) {
int nbOccurrences;
int const freq = normalizedCounter[symbol];
for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
@@ -120,7 +162,6 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
while (position > highThreshold)
position = (position + step) & tableMask; /* Low proba area */
} }
assert(position==0); /* Must have initialized all positions */
}
@@ -144,16 +185,17 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
case -1:
case 1:
symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
symbolTT[s].deltaFindState = total - 1;
assert(total <= INT_MAX);
symbolTT[s].deltaFindState = (int)(total - 1);
total ++;
break;
default :
{
U32 const maxBitsOut = tableLog - BIT_highbit32 (normalizedCounter[s]-1);
U32 const minStatePlus = normalizedCounter[s] << maxBitsOut;
assert(normalizedCounter[s] > 1);
{ U32 const maxBitsOut = tableLog - BIT_highbit32 ((U32)normalizedCounter[s]-1);
U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
symbolTT[s].deltaFindState = total - normalizedCounter[s];
total += normalizedCounter[s];
symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
total += (unsigned)normalizedCounter[s];
} } } }
#if 0 /* debug : symbol costs */
@@ -164,32 +206,26 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
symbol, normalizedCounter[symbol],
FSE_getMaxNbBits(symbolTT, symbol),
(double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
}
}
} }
#endif
return 0;
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
{
FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE]; /* memset() is not necessary, even if static analyzer complain about it */
return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
}
#endif
#ifndef FSE_COMMONDEFS_ONLY
/*-**************************************************************
* FSE NCount encoding
****************************************************************/
size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
{
size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+ 4 /* bitCount initialized at 4 */
+ 2 /* first two symbols may use one additional bit each */) / 8)
+ 1 /* round up to whole nb bytes */
+ 2 /* additional two bytes for bitstream flush */;
return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND; /* maxSymbolValue==0 ? use default */
}
lib/compress/huf_compress.c
+538 -105
View File
@@ -53,6 +53,28 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
/* *******************************************************
* HUF : Huffman block compression
*********************************************************/
#define HUF_WORKSPACE_MAX_ALIGNMENT 8
static void* HUF_alignUpWorkspace(void* workspace, size_t* workspaceSizePtr, size_t align)
{
size_t const mask = align - 1;
size_t const rem = (size_t)workspace & mask;
size_t const add = (align - rem) & mask;
BYTE* const aligned = (BYTE*)workspace + add;
assert((align & (align - 1)) == 0); /* pow 2 */
assert(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
if (*workspaceSizePtr >= add) {
assert(add < align);
assert(((size_t)aligned & mask) == 0);
*workspaceSizePtr -= add;
return aligned;
} else {
*workspaceSizePtr = 0;
return NULL;
}
}
/* HUF_compressWeights() :
* Same as FSE_compress(), but dedicated to huff0's weights compression.
* The use case needs much less stack memory.
@@ -75,7 +97,7 @@ static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightT
unsigned maxSymbolValue = HUF_TABLELOG_MAX;
U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)workspace;
HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, sizeof(U32));
if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
@@ -106,6 +128,40 @@ static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightT
return (size_t)(op-ostart);
}
static size_t HUF_getNbBits(HUF_CElt elt)
{
return elt & 0xFF;
}
static size_t HUF_getNbBitsFast(HUF_CElt elt)
{
return elt;
}
static size_t HUF_getValue(HUF_CElt elt)
{
return elt & ~0xFF;
}
static size_t HUF_getValueFast(HUF_CElt elt)
{
return elt;
}
static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
{
assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
*elt = nbBits;
}
static void HUF_setValue(HUF_CElt* elt, size_t value)
{
size_t const nbBits = HUF_getNbBits(*elt);
if (nbBits > 0) {
assert((value >> nbBits) == 0);
*elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
}
}
typedef struct {
HUF_CompressWeightsWksp wksp;
@@ -117,9 +173,10 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
void* workspace, size_t workspaceSize)
{
HUF_CElt const* const ct = CTable + 1;
BYTE* op = (BYTE*)dst;
U32 n;
HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)workspace;
HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, sizeof(U32));
/* check conditions */
if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
@@ -130,9 +187,10 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
for (n=1; n<huffLog+1; n++)
wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
for (n=0; n<maxSymbolValue; n++)
wksp->huffWeight[n] = wksp->bitsToWeight[CTable[n].nbBits];
wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
/* attempt weights compression by FSE */
if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
{ CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
if ((hSize>1) & (hSize < maxSymbolValue/2)) { /* FSE compressed */
op[0] = (BYTE)hSize;
@@ -166,6 +224,7 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1]; /* large enough for values from 0 to 16 */
U32 tableLog = 0;
U32 nbSymbols = 0;
HUF_CElt* const ct = CTable + 1;
/* get symbol weights */
CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
@@ -175,6 +234,8 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
CTable[0] = tableLog;
/* Prepare base value per rank */
{ U32 n, nextRankStart = 0;
for (n=1; n<=tableLog; n++) {
@@ -186,13 +247,13 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
/* fill nbBits */
{ U32 n; for (n=0; n<nbSymbols; n++) {
const U32 w = huffWeight[n];
CTable[n].nbBits = (BYTE)(tableLog + 1 - w) & -(w != 0);
HUF_setNbBits(ct + n, (BYTE)(tableLog + 1 - w) & -(w != 0));
} }
/* fill val */
{ U16 nbPerRank[HUF_TABLELOG_MAX+2] = {0}; /* support w=0=>n=tableLog+1 */
U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
{ U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[CTable[n].nbBits]++; }
{ U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[HUF_getNbBits(ct[n])]++; }
/* determine stating value per rank */
valPerRank[tableLog+1] = 0; /* for w==0 */
{ U16 min = 0;
@@ -202,18 +263,18 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
min >>= 1;
} }
/* assign value within rank, symbol order */
{ U32 n; for (n=0; n<nbSymbols; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
{ U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
}
*maxSymbolValuePtr = nbSymbols - 1;
return readSize;
}
U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue)
U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
{
const HUF_CElt* table = (const HUF_CElt*)symbolTable;
const HUF_CElt* ct = CTable + 1;
assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
return table[symbolValue].nbBits;
return (U32)HUF_getNbBits(ct[symbolValue]);
}
@@ -367,22 +428,118 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
}
typedef struct {
U32 base;
U32 curr;
U16 base;
U16 curr;
} rankPos;
typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
#define RANK_POSITION_TABLE_SIZE 32
/* Number of buckets available for HUF_sort() */
#define RANK_POSITION_TABLE_SIZE 192
typedef struct {
huffNodeTable huffNodeTbl;
rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
} HUF_buildCTable_wksp_tables;
/* 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.
*/
#define RANK_POSITION_MAX_COUNT_LOG 32
#define RANK_POSITION_LOG_BUCKETS_BEGIN (RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */
#define RANK_POSITION_DISTINCT_COUNT_CUTOFF RANK_POSITION_LOG_BUCKETS_BEGIN + BIT_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */
/* Return the appropriate bucket index for a given count. See definition of
* RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
*/
static U32 HUF_getIndex(U32 const count) {
return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
? count
: BIT_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
}
/* Helper swap function for HUF_quickSortPartition() */
static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
nodeElt tmp = *a;
*a = *b;
*b = tmp;
}
/* Returns 0 if the huffNode array is not sorted by descending count */
MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
U32 i;
for (i = 1; i < maxSymbolValue1; ++i) {
if (huffNode[i].count > huffNode[i-1].count) {
return 0;
}
}
return 1;
}
/* Insertion sort by descending order */
HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
int i;
int const size = high-low+1;
huffNode += low;
for (i = 1; i < size; ++i) {
nodeElt const key = huffNode[i];
int j = i - 1;
while (j >= 0 && huffNode[j].count < key.count) {
huffNode[j + 1] = huffNode[j];
j--;
}
huffNode[j + 1] = key;
}
}
/* Pivot helper function for quicksort. */
static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
/* Simply select rightmost element as pivot. "Better" selectors like
* median-of-three don't experimentally appear to have any benefit.
*/
U32 const pivot = arr[high].count;
int i = low - 1;
int j = low;
for ( ; j < high; j++) {
if (arr[j].count > pivot) {
i++;
HUF_swapNodes(&arr[i], &arr[j]);
}
}
HUF_swapNodes(&arr[i + 1], &arr[high]);
return i + 1;
}
/* Classic quicksort by descending with partially iterative calls
* to reduce worst case callstack size.
*/
static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
int const kInsertionSortThreshold = 8;
if (high - low < kInsertionSortThreshold) {
HUF_insertionSort(arr, low, high);
return;
}
while (low < high) {
int const idx = HUF_quickSortPartition(arr, low, high);
if (idx - low < high - idx) {
HUF_simpleQuickSort(arr, low, idx - 1);
low = idx + 1;
} else {
HUF_simpleQuickSort(arr, idx + 1, high);
high = idx - 1;
}
}
}
/**
* HUF_sort():
* Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
* This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
*
* @param[out] huffNode Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
* Must have (maxSymbolValue + 1) entries.
@@ -390,44 +547,52 @@ typedef struct {
* @param[in] maxSymbolValue Maximum symbol value.
* @param rankPosition This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
*/
static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValue, rankPos* rankPosition)
{
int n;
int const maxSymbolValue1 = (int)maxSymbolValue + 1;
static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
U32 n;
U32 const maxSymbolValue1 = maxSymbolValue+1;
/* Compute base and set curr to base.
* For symbol s let lowerRank = BIT_highbit32(count[n]+1) and rank = lowerRank + 1.
* Then 2^lowerRank <= count[n]+1 <= 2^rank.
* For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
* See HUF_getIndex to see bucketing strategy.
* We attribute each symbol to lowerRank's base value, because we want to know where
* each rank begins in the output, so for rank R we want to count ranks R+1 and above.
*/
ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
for (n = 0; n < maxSymbolValue1; ++n) {
U32 lowerRank = BIT_highbit32(count[n] + 1);
U32 lowerRank = HUF_getIndex(count[n]);
assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
rankPosition[lowerRank].base++;
}
assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
/* Set up the rankPosition table */
for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
rankPosition[n-1].base += rankPosition[n].base;
rankPosition[n-1].curr = rankPosition[n-1].base;
}
/* Sort */
/* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
for (n = 0; n < maxSymbolValue1; ++n) {
U32 const c = count[n];
U32 const r = BIT_highbit32(c+1) + 1;
U32 pos = rankPosition[r].curr++;
/* Insert into the correct position in the rank.
* We have at most 256 symbols, so this insertion should be fine.
*/
while ((pos > rankPosition[r].base) && (c > huffNode[pos-1].count)) {
huffNode[pos] = huffNode[pos-1];
pos--;
}
U32 const r = HUF_getIndex(c) + 1;
U32 const pos = rankPosition[r].curr++;
assert(pos < maxSymbolValue1);
huffNode[pos].count = c;
huffNode[pos].byte = (BYTE)n;
}
}
/* Sort each bucket. */
for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
U32 const bucketSize = rankPosition[n].curr-rankPosition[n].base;
U32 const bucketStartIdx = rankPosition[n].base;
if (bucketSize > 1) {
assert(bucketStartIdx < maxSymbolValue1);
HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
}
}
assert(HUF_isSorted(huffNode, maxSymbolValue1));
}
/** HUF_buildCTable_wksp() :
* Same as HUF_buildCTable(), but using externally allocated scratch buffer.
@@ -490,6 +655,7 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
*/
static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
{
HUF_CElt* const ct = CTable + 1;
/* fill result into ctable (val, nbBits) */
int n;
U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
@@ -505,20 +671,20 @@ static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, i
min >>= 1;
} }
for (n=0; n<alphabetSize; n++)
CTable[huffNode[n].byte].nbBits = huffNode[n].nbBits; /* push nbBits per symbol, symbol order */
HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits); /* push nbBits per symbol, symbol order */
for (n=0; n<alphabetSize; n++)
CTable[n].val = valPerRank[CTable[n].nbBits]++; /* assign value within rank, symbol order */
HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); /* assign value within rank, symbol order */
CTable[0] = maxNbBits;
}
size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
size_t HUF_buildCTable_wksp (HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
{
HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, sizeof(U32));
nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
nodeElt* const huffNode = huffNode0+1;
int nonNullRank;
/* safety checks */
if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
return ERROR(workSpace_tooSmall);
if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
@@ -536,96 +702,334 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbo
maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC); /* check fit into table */
HUF_buildCTableFromTree(tree, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
return maxNbBits;
}
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
{
HUF_CElt const* ct = CTable + 1;
size_t nbBits = 0;
int s;
for (s = 0; s <= (int)maxSymbolValue; ++s) {
nbBits += CTable[s].nbBits * count[s];
nbBits += HUF_getNbBits(ct[s]) * count[s];
}
return nbBits >> 3;
}
int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
HUF_CElt const* ct = CTable + 1;
int bad = 0;
int s;
for (s = 0; s <= (int)maxSymbolValue; ++s) {
bad |= (count[s] != 0) & (CTable[s].nbBits == 0);
bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
}
return !bad;
}
size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
FORCE_INLINE_TEMPLATE void
HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
/** HUF_CStream_t:
* Huffman uses its own BIT_CStream_t implementation.
* There are three major differences from BIT_CStream_t:
* 1. HUF_addBits() takes a HUF_CElt (size_t) which is
* the pair (nbBits, value) in the format:
* format:
* - Bits [0, 4) = nbBits
* - Bits [4, 64 - nbBits) = 0
* - Bits [64 - nbBits, 64) = value
* 2. The bitContainer is built from the upper bits and
* right shifted. E.g. to add a new value of N bits
* you right shift the bitContainer by N, then or in
* the new value into the N upper bits.
* 3. The bitstream has two bit containers. You can add
* bits to the second container and merge them into
* the first container.
*/
#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
typedef struct {
size_t bitContainer[2];
size_t bitPos[2];
BYTE* startPtr;
BYTE* ptr;
BYTE* endPtr;
} HUF_CStream_t;
/**! HUF_initCStream():
* Initializes the bitstream.
* @returns 0 or an error code.
*/
static size_t HUF_initCStream(HUF_CStream_t* bitC,
void* startPtr, size_t dstCapacity)
{
BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
ZSTD_memset(bitC, 0, sizeof(*bitC));
bitC->startPtr = (BYTE*)startPtr;
bitC->ptr = bitC->startPtr;
bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
return 0;
}
#define HUF_FLUSHBITS(s) BIT_flushBits(s)
/*! HUF_addBits():
* Adds the symbol stored in HUF_CElt elt to the bitstream.
*
* @param elt The element we're adding. This is a (nbBits, value) pair.
* See the HUF_CStream_t docs for the format.
* @param idx Insert into the bitstream at this idx.
* @param kFast This is a template parameter. If the bitstream is guaranteed
* to have at least 4 unused bits after this call it may be 1,
* otherwise it must be 0. HUF_addBits() is faster when fast is set.
*/
FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
{
assert(idx <= 1);
assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
/* This is efficient on x86-64 with BMI2 because shrx
* only reads the low 6 bits of the register. The compiler
* knows this and elides the mask. When fast is set,
* every operation can use the same value loaded from elt.
*/
bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
/* We only read the low 8 bits of bitC->bitPos[idx] so it
* doesn't matter that the high bits have noise from the value.
*/
bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
/* The last 4-bits of elt are dirty if fast is set,
* so we must not be overwriting bits that have already been
* inserted into the bit container.
*/
#if DEBUGLEVEL >= 1
{
size_t const nbBits = HUF_getNbBits(elt);
size_t const dirtyBits = nbBits == 0 ? 0 : BIT_highbit32((U32)nbBits) + 1;
(void)dirtyBits;
/* Middle bits are 0. */
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
}
#define HUF_FLUSHBITS_1(stream) \
if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
{
bitC->bitContainer[1] = 0;
bitC->bitPos[1] = 0;
}
/*! HUF_mergeIndex1() :
* Merges the bit container @ index 1 into the bit container @ index 0
* and zeros the bit container @ index 1.
*/
FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
{
assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
bitC->bitContainer[0] |= bitC->bitContainer[1];
bitC->bitPos[0] += bitC->bitPos[1];
assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
}
/*! HUF_flushBits() :
* Flushes the bits in the bit container @ index 0.
*
* @post bitPos will be < 8.
* @param kFast If kFast is set then we must know a-priori that
* the bit container will not overflow.
*/
FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
{
/* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
size_t const nbBits = bitC->bitPos[0] & 0xFF;
size_t const nbBytes = nbBits >> 3;
/* The top nbBits bits of bitContainer are the ones we need. */
size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
/* Mask bitPos to account for the bytes we consumed. */
bitC->bitPos[0] &= 7;
assert(nbBits > 0);
assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
assert(bitC->ptr <= bitC->endPtr);
MEM_writeLEST(bitC->ptr, bitContainer);
bitC->ptr += nbBytes;
assert(!kFast || bitC->ptr <= bitC->endPtr);
if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
/* bitContainer doesn't need to be modified because the leftover
* bits are already the top bitPos bits. And we don't care about
* noise in the lower values.
*/
}
/*! HUF_endMark()
* @returns The Huffman stream end mark: A 1-bit value = 1.
*/
static HUF_CElt HUF_endMark(void)
{
HUF_CElt endMark;
HUF_setNbBits(&endMark, 1);
HUF_setValue(&endMark, 1);
return endMark;
}
/*! HUF_closeCStream() :
* @return Size of CStream, in bytes,
* or 0 if it could not fit into dstBuffer */
static size_t HUF_closeCStream(HUF_CStream_t* bitC)
{
HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
HUF_flushBits(bitC, /* kFast */ 0);
{
size_t const nbBits = bitC->bitPos[0] & 0xFF;
if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
return (bitC->ptr - bitC->startPtr) + (nbBits > 0);
}
}
FORCE_INLINE_TEMPLATE void
HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
{
HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
}
FORCE_INLINE_TEMPLATE void
HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
const BYTE* ip, size_t srcSize,
const HUF_CElt* ct,
int kUnroll, int kFastFlush, int kLastFast)
{
/* Join to kUnroll */
int n = (int)srcSize;
int rem = n % kUnroll;
if (rem > 0) {
for (; rem > 0; --rem) {
HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
}
HUF_flushBits(bitC, kFastFlush);
}
assert(n % kUnroll == 0);
/* Join to 2 * kUnroll */
if (n % (2 * kUnroll)) {
int u;
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
HUF_flushBits(bitC, kFastFlush);
n -= kUnroll;
}
assert(n % (2 * kUnroll) == 0);
for (; n>0; n-= 2 * kUnroll) {
/* Encode kUnroll symbols into the bitstream @ index 0. */
int u;
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
HUF_flushBits(bitC, kFastFlush);
/* Encode kUnroll symbols into the bitstream @ index 1.
* This allows us to start filling the bit container
* without any data dependencies.
*/
HUF_zeroIndex1(bitC);
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
/* Merge bitstream @ index 1 into the bitstream @ index 0 */
HUF_mergeIndex1(bitC);
HUF_flushBits(bitC, kFastFlush);
}
assert(n == 0);
}
/**
* Returns a tight upper bound on the output space needed by Huffman
* with 8 bytes buffer to handle over-writes. If the output is at least
* this large we don't need to do bounds checks during Huffman encoding.
*/
static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
{
return ((srcSize * tableLog) >> 3) + 8;
}
#define HUF_FLUSHBITS_2(stream) \
if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
FORCE_INLINE_TEMPLATE size_t
HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable)
{
U32 const tableLog = (U32)CTable[0];
HUF_CElt const* ct = CTable + 1;
const BYTE* ip = (const BYTE*) src;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
size_t n;
BIT_CStream_t bitC;
HUF_CStream_t bitC;
/* init */
if (dstSize < 8) return 0; /* not enough space to compress */
{ size_t const initErr = BIT_initCStream(&bitC, op, (size_t)(oend-op));
{ size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
if (HUF_isError(initErr)) return 0; }
n = srcSize & ~3; /* join to mod 4 */
switch (srcSize & 3)
{
case 3 : HUF_encodeSymbol(&bitC, ip[n+ 2], CTable);
HUF_FLUSHBITS_2(&bitC);
/* fall-through */
case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
HUF_FLUSHBITS_1(&bitC);
/* fall-through */
case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
HUF_FLUSHBITS(&bitC);
/* fall-through */
case 0 : /* fall-through */
default: break;
if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
else {
if (MEM_32bits()) {
switch (tableLog) {
case 11:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 10: ZSTD_FALLTHROUGH;
case 9: ZSTD_FALLTHROUGH;
case 8:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
case 7: ZSTD_FALLTHROUGH;
default:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
}
} else {
switch (tableLog) {
case 11:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 10:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
case 9:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 8:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 7:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 6: ZSTD_FALLTHROUGH;
default:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
}
}
}
assert(bitC.ptr <= bitC.endPtr);
for (; n>0; n-=4) { /* note : n&3==0 at this stage */
HUF_encodeSymbol(&bitC, ip[n- 1], CTable);
HUF_FLUSHBITS_1(&bitC);
HUF_encodeSymbol(&bitC, ip[n- 2], CTable);
HUF_FLUSHBITS_2(&bitC);
HUF_encodeSymbol(&bitC, ip[n- 3], CTable);
HUF_FLUSHBITS_1(&bitC);
HUF_encodeSymbol(&bitC, ip[n- 4], CTable);
HUF_FLUSHBITS(&bitC);
}
return BIT_closeCStream(&bitC);
return HUF_closeCStream(&bitC);
}
#if DYNAMIC_BMI2
static TARGET_ATTRIBUTE("bmi2") size_t
static BMI2_TARGET_ATTRIBUTE size_t
HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable)
@@ -667,9 +1071,13 @@ HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
{
return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
return HUF_compress1X_usingCTable_bmi2(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
}
size_t HUF_compress1X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2)
{
return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, bmi2);
}
static size_t
HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
@@ -689,8 +1097,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
if (cSize==0) return 0;
assert(cSize <= 65535);
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart, (U16)cSize);
op += cSize;
}
@@ -698,8 +1105,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
ip += segmentSize;
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
if (cSize==0) return 0;
assert(cSize <= 65535);
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart+2, (U16)cSize);
op += cSize;
}
@@ -707,8 +1113,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
ip += segmentSize;
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
if (cSize==0) return 0;
assert(cSize <= 65535);
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart+4, (U16)cSize);
op += cSize;
}
@@ -717,7 +1122,7 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
assert(op <= oend);
assert(ip <= iend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, bmi2) );
if (cSize==0) return 0;
if (cSize == 0 || cSize > 65535) return 0;
op += cSize;
}
@@ -726,7 +1131,12 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
{
return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
return HUF_compress4X_usingCTable_bmi2(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
}
size_t HUF_compress4X_usingCTable_bmi2(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int bmi2)
{
return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, bmi2);
}
typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
@@ -750,35 +1160,38 @@ static size_t HUF_compressCTable_internal(
typedef struct {
unsigned count[HUF_SYMBOLVALUE_MAX + 1];
HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
union {
HUF_buildCTable_wksp_tables buildCTable_wksp;
HUF_WriteCTableWksp writeCTable_wksp;
U32 hist_wksp[HIST_WKSP_SIZE_U32];
} wksps;
} HUF_compress_tables_t;
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10 /* Must be >= 2 */
/* HUF_compress_internal() :
* `workSpace_align4` must be aligned on 4-bytes boundaries,
* and occupies the same space as a table of HUF_WORKSPACE_SIZE_U32 unsigned */
* and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
static size_t
HUF_compress_internal (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
HUF_nbStreams_e nbStreams,
void* workSpace_align4, size_t wkspSize,
void* workSpace, size_t wkspSize,
HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
const int bmi2)
const int bmi2, unsigned suspectUncompressible)
{
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace_align4;
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, sizeof(size_t));
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
HUF_STATIC_ASSERT(sizeof(*table) <= HUF_WORKSPACE_SIZE);
assert(((size_t)workSpace_align4 & 3) == 0); /* must be aligned on 4-bytes boundaries */
HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
/* checks & inits */
if (wkspSize < HUF_WORKSPACE_SIZE) return ERROR(workSpace_tooSmall);
if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
if (!srcSize) return 0; /* Uncompressed */
if (!dstSize) return 0; /* cannot fit anything within dst budget */
if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong); /* current block size limit */
@@ -794,8 +1207,23 @@ HUF_compress_internal (void* dst, size_t dstSize,
nbStreams, oldHufTable, bmi2);
}
/* If uncompressible data is suspected, do a smaller sampling first */
DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
if (suspectUncompressible && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
size_t largestTotal = 0;
{ unsigned maxSymbolValueBegin = maxSymbolValue;
CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
largestTotal += largestBegin;
}
{ unsigned maxSymbolValueEnd = maxSymbolValue;
CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
largestTotal += largestEnd;
}
if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0; /* heuristic : probably not compressible enough */
}
/* Scan input and build symbol stats */
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace_align4, wkspSize) );
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; } /* single symbol, rle */
if (largest <= (srcSize >> 7)+4) return 0; /* heuristic : probably not compressible enough */
}
@@ -820,9 +1248,12 @@ HUF_compress_internal (void* dst, size_t dstSize,
&table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
CHECK_F(maxBits);
huffLog = (U32)maxBits;
/* Zero unused symbols in CTable, so we can check it for validity */
ZSTD_memset(table->CTable + (maxSymbolValue + 1), 0,
sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
}
/* Zero unused symbols in CTable, so we can check it for validity */
{
size_t const ctableSize = HUF_CTABLE_SIZE_ST(maxSymbolValue);
size_t const unusedSize = sizeof(table->CTable) - ctableSize * sizeof(HUF_CElt);
ZSTD_memset(table->CTable + ctableSize, 0, unusedSize);
}
/* Write table description header */
@@ -859,19 +1290,20 @@ size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_singleStream,
workSpace, wkspSize,
NULL, NULL, 0, 0 /*bmi2*/);
NULL, NULL, 0, 0 /*bmi2*/, 0);
}
size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
void* workSpace, size_t wkspSize,
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat,
int bmi2, unsigned suspectUncompressible)
{
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_singleStream,
workSpace, wkspSize, hufTable,
repeat, preferRepeat, bmi2);
repeat, preferRepeat, bmi2, suspectUncompressible);
}
/* HUF_compress4X_repeat():
@@ -885,22 +1317,23 @@ size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_fourStreams,
workSpace, wkspSize,
NULL, NULL, 0, 0 /*bmi2*/);
NULL, NULL, 0, 0 /*bmi2*/, 0);
}
/* HUF_compress4X_repeat():
* compress input using 4 streams.
* consider skipping quickly
* re-use an existing huffman compression table */
size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
void* workSpace, size_t wkspSize,
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2, unsigned suspectUncompressible)
{
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_fourStreams,
workSpace, wkspSize,
hufTable, repeat, preferRepeat, bmi2);
hufTable, repeat, preferRepeat, bmi2, suspectUncompressible);
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
@@ -918,7 +1351,7 @@ size_t HUF_compress1X (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog)
{
unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
U64 workSpace[HUF_WORKSPACE_SIZE_U64];
return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
}
@@ -926,7 +1359,7 @@ size_t HUF_compress2 (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog)
{
unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
U64 workSpace[HUF_WORKSPACE_SIZE_U64];
return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
}
lib/compress/zstd_compress.c
+190 -274
View File
@@ -12,7 +12,6 @@
* Dependencies
***************************************/
#include "../common/zstd_deps.h" /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
#include "../common/cpu.h"
#include "../common/mem.h"
#include "hist.h" /* HIST_countFast_wksp */
#define FSE_STATIC_LINKING_ONLY /* FSE_encodeSymbol */
@@ -42,6 +41,18 @@
# define ZSTD_COMPRESS_HEAPMODE 0
#endif
/*!
* ZSTD_HASHLOG3_MAX :
* Maximum size of the hash table dedicated to find 3-bytes matches,
* in log format, aka 17 => 1 << 17 == 128Ki positions.
* This structure is only used in zstd_opt.
* Since allocation is centralized for all strategies, it has to be known here.
* The actual (selected) size of the hash table is then stored in ZSTD_matchState_t.hashLog3,
* so that zstd_opt.c doesn't need to know about this constant.
*/
#ifndef ZSTD_HASHLOG3_MAX
# define ZSTD_HASHLOG3_MAX 17
#endif
/*-*************************************
* Helper functions
@@ -72,10 +83,10 @@ struct ZSTD_CDict_s {
ZSTD_customMem customMem;
U32 dictID;
int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
ZSTD_useRowMatchFinderMode_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
* row-based matchfinder. Unless the cdict is reloaded, we will use
* the same greedy/lazy matchfinder at compression time.
*/
ZSTD_paramSwitch_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
* row-based matchfinder. Unless the cdict is reloaded, we will use
* the same greedy/lazy matchfinder at compression time.
*/
}; /* typedef'd to ZSTD_CDict within "zstd.h" */
ZSTD_CCtx* ZSTD_createCCtx(void)
@@ -88,7 +99,7 @@ static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
assert(cctx != NULL);
ZSTD_memset(cctx, 0, sizeof(*cctx));
cctx->customMem = memManager;
cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
cctx->bmi2 = ZSTD_cpuSupportsBmi2();
{ size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
assert(!ZSTD_isError(err));
(void)err;
@@ -214,35 +225,42 @@ static int ZSTD_rowMatchFinderSupported(const ZSTD_strategy strategy) {
/* Returns true if the strategy and useRowMatchFinder mode indicate that we will use the row based matchfinder
* for this compression.
*/
static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_useRowMatchFinderMode_e mode) {
assert(mode != ZSTD_urm_auto);
return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_urm_enableRowMatchFinder);
static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_paramSwitch_e mode) {
assert(mode != ZSTD_ps_auto);
return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_ps_enable);
}
/* Returns row matchfinder usage enum given an initial mode and cParams */
static ZSTD_useRowMatchFinderMode_e ZSTD_resolveRowMatchFinderMode(ZSTD_useRowMatchFinderMode_e mode,
const ZSTD_compressionParameters* const cParams) {
#if !defined(ZSTD_NO_INTRINSICS) && (defined(__SSE2__) || defined(__ARM_NEON))
/* Returns row matchfinder usage given an initial mode and cParams */
static ZSTD_paramSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
#if defined(ZSTD_ARCH_X86_SSE2) || defined(ZSTD_ARCH_ARM_NEON)
int const kHasSIMD128 = 1;
#else
int const kHasSIMD128 = 0;
#endif
if (mode != ZSTD_urm_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
mode = ZSTD_urm_disableRowMatchFinder;
if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
mode = ZSTD_ps_disable;
if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
if (kHasSIMD128) {
if (cParams->windowLog > 14) mode = ZSTD_urm_enableRowMatchFinder;
if (cParams->windowLog > 14) mode = ZSTD_ps_enable;
} else {
if (cParams->windowLog > 17) mode = ZSTD_urm_enableRowMatchFinder;
if (cParams->windowLog > 17) mode = ZSTD_ps_enable;
}
return mode;
}
/* Returns block splitter usage (generally speaking, when using slower/stronger compression modes) */
static ZSTD_paramSwitch_e ZSTD_resolveBlockSplitterMode(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
if (mode != ZSTD_ps_auto) return mode;
return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17) ? ZSTD_ps_enable : ZSTD_ps_disable;
}
/* Returns 1 if the arguments indicate that we should allocate a chainTable, 0 otherwise */
static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
const ZSTD_paramSwitch_e useRowMatchFinder,
const U32 forDDSDict) {
assert(useRowMatchFinder != ZSTD_urm_auto);
assert(useRowMatchFinder != ZSTD_ps_auto);
/* We always should allocate a chaintable if we are allocating a matchstate for a DDS dictionary matchstate.
* We do not allocate a chaintable if we are using ZSTD_fast, or are using the row-based matchfinder.
*/
@@ -253,16 +271,10 @@ static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
* enable long distance matching (wlog >= 27, strategy >= btopt).
* Returns 0 otherwise.
*/
static U32 ZSTD_CParams_shouldEnableLdm(const ZSTD_compressionParameters* const cParams) {
return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27;
}
/* Returns 1 if compression parameters are such that we should
* enable blockSplitter (wlog >= 17, strategy >= btopt).
* Returns 0 otherwise.
*/
static U32 ZSTD_CParams_useBlockSplitter(const ZSTD_compressionParameters* const cParams) {
return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17;
static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
if (mode != ZSTD_ps_auto) return mode;
return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
}
static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
@@ -274,20 +286,13 @@ static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
cctxParams.cParams = cParams;
/* Adjust advanced params according to cParams */
if (ZSTD_CParams_shouldEnableLdm(&cParams)) {
DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including LDM into cctx params");
cctxParams.ldmParams.enableLdm = 1;
/* LDM is enabled by default for optimal parser and window size >= 128MB */
cctxParams.ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams.ldmParams.enableLdm, &cParams);
if (cctxParams.ldmParams.enableLdm == ZSTD_ps_enable) {
ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
assert(cctxParams.ldmParams.hashRateLog < 32);
}
if (ZSTD_CParams_useBlockSplitter(&cParams)) {
DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including block splitting into cctx params");
cctxParams.splitBlocks = 1;
}
cctxParams.useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.useBlockSplitter, &cParams);
cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
assert(!ZSTD_checkCParams(cParams));
return cctxParams;
@@ -348,7 +353,10 @@ static void ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams, ZSTD_par
*/
cctxParams->compressionLevel = compressionLevel;
cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, &params->cParams);
DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d", cctxParams->useRowMatchFinder);
cctxParams->useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->useBlockSplitter, &params->cParams);
cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, &params->cParams);
DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
cctxParams->useRowMatchFinder, cctxParams->useBlockSplitter, cctxParams->ldmParams.enableLdm);
}
size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
@@ -518,9 +526,9 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
return bounds;
case ZSTD_c_literalCompressionMode:
ZSTD_STATIC_ASSERT(ZSTD_lcm_auto < ZSTD_lcm_huffman && ZSTD_lcm_huffman < ZSTD_lcm_uncompressed);
bounds.lowerBound = ZSTD_lcm_auto;
bounds.upperBound = ZSTD_lcm_uncompressed;
ZSTD_STATIC_ASSERT(ZSTD_ps_auto < ZSTD_ps_enable && ZSTD_ps_enable < ZSTD_ps_disable);
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_targetCBlockSize:
@@ -549,14 +557,14 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
bounds.upperBound = 1;
return bounds;
case ZSTD_c_splitBlocks:
bounds.lowerBound = 0;
bounds.upperBound = 1;
case ZSTD_c_useBlockSplitter:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_useRowMatchFinder:
bounds.lowerBound = (int)ZSTD_urm_auto;
bounds.upperBound = (int)ZSTD_urm_enableRowMatchFinder;
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_deterministicRefPrefix:
@@ -625,7 +633,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_splitBlocks:
case ZSTD_c_useBlockSplitter:
case ZSTD_c_useRowMatchFinder:
case ZSTD_c_deterministicRefPrefix:
default:
@@ -680,7 +688,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_splitBlocks:
case ZSTD_c_useBlockSplitter:
case ZSTD_c_useRowMatchFinder:
case ZSTD_c_deterministicRefPrefix:
break;
@@ -780,7 +788,7 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
}
case ZSTD_c_literalCompressionMode : {
const ZSTD_literalCompressionMode_e lcm = (ZSTD_literalCompressionMode_e)value;
const ZSTD_paramSwitch_e lcm = (ZSTD_paramSwitch_e)value;
BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
CCtxParams->literalCompressionMode = lcm;
return CCtxParams->literalCompressionMode;
@@ -835,7 +843,7 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
return CCtxParams->enableDedicatedDictSearch;
case ZSTD_c_enableLongDistanceMatching :
CCtxParams->ldmParams.enableLdm = (value!=0);
CCtxParams->ldmParams.enableLdm = (ZSTD_paramSwitch_e)value;
return CCtxParams->ldmParams.enableLdm;
case ZSTD_c_ldmHashLog :
@@ -857,8 +865,8 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
return CCtxParams->ldmParams.bucketSizeLog;
case ZSTD_c_ldmHashRateLog :
RETURN_ERROR_IF(value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN,
parameter_outOfBound, "Param out of bounds!");
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
CCtxParams->ldmParams.hashRateLog = value;
return CCtxParams->ldmParams.hashRateLog;
@@ -894,14 +902,14 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
CCtxParams->validateSequences = value;
return CCtxParams->validateSequences;
case ZSTD_c_splitBlocks:
BOUNDCHECK(ZSTD_c_splitBlocks, value);
CCtxParams->splitBlocks = value;
return CCtxParams->splitBlocks;
case ZSTD_c_useBlockSplitter:
BOUNDCHECK(ZSTD_c_useBlockSplitter, value);
CCtxParams->useBlockSplitter = (ZSTD_paramSwitch_e)value;
return CCtxParams->useBlockSplitter;
case ZSTD_c_useRowMatchFinder:
BOUNDCHECK(ZSTD_c_useRowMatchFinder, value);
CCtxParams->useRowMatchFinder = (ZSTD_useRowMatchFinderMode_e)value;
CCtxParams->useRowMatchFinder = (ZSTD_paramSwitch_e)value;
return CCtxParams->useRowMatchFinder;
case ZSTD_c_deterministicRefPrefix:
@@ -1032,8 +1040,8 @@ size_t ZSTD_CCtxParams_getParameter(
case ZSTD_c_validateSequences :
*value = (int)CCtxParams->validateSequences;
break;
case ZSTD_c_splitBlocks :
*value = (int)CCtxParams->splitBlocks;
case ZSTD_c_useBlockSplitter :
*value = (int)CCtxParams->useBlockSplitter;
break;
case ZSTD_c_useRowMatchFinder :
*value = (int)CCtxParams->useRowMatchFinder;
@@ -1324,7 +1332,7 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
break;
case ZSTD_cpm_createCDict:
/* Assume a small source size when creating a dictionary
* with an unkown source size.
* with an unknown source size.
*/
if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
srcSize = minSrcSize;
@@ -1398,7 +1406,7 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
srcSizeHint = CCtxParams->srcSizeHint;
}
cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize, mode);
if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
if (CCtxParams->ldmParams.enableLdm == ZSTD_ps_enable) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
assert(!ZSTD_checkCParams(cParams));
/* srcSizeHint == 0 means 0 */
@@ -1407,7 +1415,7 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
static size_t
ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
const ZSTD_paramSwitch_e useRowMatchFinder,
const U32 enableDedicatedDictSearch,
const U32 forCCtx)
{
@@ -1440,7 +1448,7 @@ ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
/* tables are guaranteed to be sized in multiples of 64 bytes (or 16 uint32_t) */
ZSTD_STATIC_ASSERT(ZSTD_HASHLOG_MIN >= 4 && ZSTD_WINDOWLOG_MIN >= 4 && ZSTD_CHAINLOG_MIN >= 4);
assert(useRowMatchFinder != ZSTD_urm_auto);
assert(useRowMatchFinder != ZSTD_ps_auto);
DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
(U32)chainSize, (U32)hSize, (U32)h3Size);
@@ -1451,12 +1459,12 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
const ZSTD_compressionParameters* cParams,
const ldmParams_t* ldmParams,
const int isStatic,
const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
const ZSTD_paramSwitch_e useRowMatchFinder,
const size_t buffInSize,
const size_t buffOutSize,
const U64 pledgedSrcSize)
{
size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << cParams->windowLog), pledgedSrcSize));
size_t const windowSize = (size_t) BOUNDED(1ULL, 1ULL << cParams->windowLog, pledgedSrcSize);
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
U32 const divider = (cParams->minMatch==3) ? 3 : 4;
size_t const maxNbSeq = blockSize / divider;
@@ -1469,7 +1477,7 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
size_t const ldmSpace = ZSTD_ldm_getTableSize(*ldmParams);
size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(*ldmParams, blockSize);
size_t const ldmSeqSpace = ldmParams->enableLdm ?
size_t const ldmSeqSpace = ldmParams->enableLdm == ZSTD_ps_enable ?
ZSTD_cwksp_aligned_alloc_size(maxNbLdmSeq * sizeof(rawSeq)) : 0;
@@ -1496,8 +1504,8 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
{
ZSTD_compressionParameters const cParams =
ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
&cParams);
ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
&cParams);
RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
/* estimateCCtxSize is for one-shot compression. So no buffers should
@@ -1514,9 +1522,9 @@ size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
/* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
size_t noRowCCtxSize;
size_t rowCCtxSize;
initialParams.useRowMatchFinder = ZSTD_urm_disableRowMatchFinder;
initialParams.useRowMatchFinder = ZSTD_ps_disable;
noRowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
initialParams.useRowMatchFinder = ZSTD_urm_enableRowMatchFinder;
initialParams.useRowMatchFinder = ZSTD_ps_enable;
rowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
return MAX(noRowCCtxSize, rowCCtxSize);
} else {
@@ -1561,7 +1569,7 @@ size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
size_t const outBuffSize = (params->outBufferMode == ZSTD_bm_buffered)
? ZSTD_compressBound(blockSize) + 1
: 0;
ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, &params->cParams);
ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, &params->cParams);
return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
&cParams, &params->ldmParams, 1, useRowMatchFinder, inBuffSize, outBuffSize,
@@ -1576,9 +1584,9 @@ size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
/* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
size_t noRowCCtxSize;
size_t rowCCtxSize;
initialParams.useRowMatchFinder = ZSTD_urm_disableRowMatchFinder;
initialParams.useRowMatchFinder = ZSTD_ps_disable;
noRowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
initialParams.useRowMatchFinder = ZSTD_urm_enableRowMatchFinder;
initialParams.useRowMatchFinder = ZSTD_ps_enable;
rowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
return MAX(noRowCCtxSize, rowCCtxSize);
} else {
@@ -1713,7 +1721,7 @@ static size_t
ZSTD_reset_matchState(ZSTD_matchState_t* ms,
ZSTD_cwksp* ws,
const ZSTD_compressionParameters* cParams,
const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
const ZSTD_paramSwitch_e useRowMatchFinder,
const ZSTD_compResetPolicy_e crp,
const ZSTD_indexResetPolicy_e forceResetIndex,
const ZSTD_resetTarget_e forWho)
@@ -1728,7 +1736,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
assert(useRowMatchFinder != ZSTD_urm_auto);
assert(useRowMatchFinder != ZSTD_ps_auto);
if (forceResetIndex == ZSTDirp_reset) {
ZSTD_window_init(&ms->window);
ZSTD_cwksp_mark_tables_dirty(ws);
@@ -1774,8 +1782,8 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
if (ms->tagTable) ZSTD_memset(ms->tagTable, 0, tagTableSize);
}
{ /* Switch to 32-entry rows if searchLog is 5 (or more) */
U32 const rowLog = cParams->searchLog < 5 ? 4 : 5;
assert(cParams->hashLog > rowLog);
U32 const rowLog = BOUNDED(4, cParams->searchLog, 6);
assert(cParams->hashLog >= rowLog);
ms->rowHashLog = cParams->hashLog - rowLog;
}
}
@@ -1824,8 +1832,8 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
ZSTD_buffered_policy_e const zbuff)
{
ZSTD_cwksp* const ws = &zc->workspace;
DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d",
(U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder);
DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d useBlockSplitter=%d",
(U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder, (int)params->useBlockSplitter);
assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
zc->isFirstBlock = 1;
@@ -1836,8 +1844,10 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
zc->appliedParams = *params;
params = &zc->appliedParams;
assert(params->useRowMatchFinder != ZSTD_urm_auto);
if (params->ldmParams.enableLdm) {
assert(params->useRowMatchFinder != ZSTD_ps_auto);
assert(params->useBlockSplitter != ZSTD_ps_auto);
assert(params->ldmParams.enableLdm != ZSTD_ps_auto);
if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
/* Adjust long distance matching parameters */
ZSTD_ldm_adjustParameters(&zc->appliedParams.ldmParams, &params->cParams);
assert(params->ldmParams.hashLog >= params->ldmParams.bucketSizeLog);
@@ -1937,7 +1947,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
/* ldm bucketOffsets table */
if (params->ldmParams.enableLdm) {
if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
/* TODO: avoid memset? */
size_t const numBuckets =
((size_t)1) << (params->ldmParams.hashLog -
@@ -1964,7 +1974,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
ZSTD_resetTarget_CCtx), "");
/* ldm hash table */
if (params->ldmParams.enableLdm) {
if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
/* TODO: avoid memset? */
size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
@@ -1976,8 +1986,8 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
zc->ldmState.loadedDictEnd = 0;
}
assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace, resizeWorkspace));
DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace, resizeWorkspace));
zc->initialized = 1;
@@ -2115,7 +2125,7 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
}
ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
assert(params.useRowMatchFinder != ZSTD_urm_auto);
assert(params.useRowMatchFinder != ZSTD_ps_auto);
/* copy tables */
{ size_t const chainSize = ZSTD_allocateChainTable(cdict_cParams->strategy, cdict->useRowMatchFinder, 0 /* DDS guaranteed disabled */)
@@ -2209,8 +2219,12 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
{ ZSTD_CCtx_params params = dstCCtx->requestedParams;
/* Copy only compression parameters related to tables. */
params.cParams = srcCCtx->appliedParams.cParams;
assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_urm_auto);
assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_ps_auto);
assert(srcCCtx->appliedParams.useBlockSplitter != ZSTD_ps_auto);
assert(srcCCtx->appliedParams.ldmParams.enableLdm != ZSTD_ps_auto);
params.useRowMatchFinder = srcCCtx->appliedParams.useRowMatchFinder;
params.useBlockSplitter = srcCCtx->appliedParams.useBlockSplitter;
params.ldmParams = srcCCtx->appliedParams.ldmParams;
params.fParams = fParams;
ZSTD_resetCCtx_internal(dstCCtx, &params, pledgedSrcSize,
/* loadedDictSize */ 0,
@@ -2296,6 +2310,8 @@ ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerVa
int const nbRows = (int)size / ZSTD_ROWSIZE;
int cellNb = 0;
int rowNb;
/* Protect special index values < ZSTD_WINDOW_START_INDEX. */
U32 const reducerThreshold = reducerValue + ZSTD_WINDOW_START_INDEX;
assert((size & (ZSTD_ROWSIZE-1)) == 0); /* multiple of ZSTD_ROWSIZE */
assert(size < (1U<<31)); /* can be casted to int */
@@ -2315,12 +2331,17 @@ ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerVa
for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
int column;
for (column=0; column<ZSTD_ROWSIZE; column++) {
if (preserveMark) {
U32 const adder = (table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) ? reducerValue : 0;
table[cellNb] += adder;
U32 newVal;
if (preserveMark && table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) {
/* This write is pointless, but is required(?) for the compiler
* to auto-vectorize the loop. */
newVal = ZSTD_DUBT_UNSORTED_MARK;
} else if (table[cellNb] < reducerThreshold) {
newVal = 0;
} else {
newVal = table[cellNb] - reducerValue;
}
if (table[cellNb] < reducerValue) table[cellNb] = 0;
else table[cellNb] -= reducerValue;
table[cellNb] = newVal;
cellNb++;
} }
}
@@ -2399,11 +2420,13 @@ static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
/* ZSTD_blockSplitterEnabled():
* Returns if block splitting param is being used
* If used, compression will do best effort to split a block in order to improve compression ratio.
* At the time this function is called, the parameter must be finalized.
* Returns 1 if true, 0 otherwise. */
static int ZSTD_blockSplitterEnabled(ZSTD_CCtx_params* cctxParams)
{
DEBUGLOG(5, "ZSTD_blockSplitterEnabled(splitBlocks=%d)", cctxParams->splitBlocks);
return (cctxParams->splitBlocks != 0);
DEBUGLOG(5, "ZSTD_blockSplitterEnabled (useBlockSplitter=%d)", cctxParams->useBlockSplitter);
assert(cctxParams->useBlockSplitter != ZSTD_ps_auto);
return (cctxParams->useBlockSplitter == ZSTD_ps_enable);
}
/* Type returned by ZSTD_buildSequencesStatistics containing finalized symbol encoding types
@@ -2546,6 +2569,7 @@ ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
* compresses both literals and sequences
* Returns compressed size of block, or a zstd error.
*/
#define SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO 20
MEM_STATIC size_t
ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
@@ -2580,15 +2604,19 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
/* Compress literals */
{ const BYTE* const literals = seqStorePtr->litStart;
size_t const numSequences = seqStorePtr->sequences - seqStorePtr->sequencesStart;
size_t const numLiterals = seqStorePtr->lit - seqStorePtr->litStart;
/* Base suspicion of uncompressibility on ratio of literals to sequences */
unsigned const suspectUncompressible = (numSequences == 0) || (numLiterals / numSequences >= SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO);
size_t const litSize = (size_t)(seqStorePtr->lit - literals);
size_t const cSize = ZSTD_compressLiterals(
&prevEntropy->huf, &nextEntropy->huf,
cctxParams->cParams.strategy,
ZSTD_disableLiteralsCompression(cctxParams),
ZSTD_literalsCompressionIsDisabled(cctxParams),
op, dstCapacity,
literals, litSize,
entropyWorkspace, entropyWkspSize,
bmi2);
bmi2, suspectUncompressible);
FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
assert(cSize <= dstCapacity);
op += cSize;
@@ -2693,7 +2721,7 @@ ZSTD_entropyCompressSeqStore(seqStore_t* seqStorePtr,
/* ZSTD_selectBlockCompressor() :
* Not static, but internal use only (used by long distance matcher)
* assumption : strat is a valid strategy */
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRowMatchFinderMode_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
{
static const ZSTD_blockCompressor blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
{ ZSTD_compressBlock_fast /* default for 0 */,
@@ -2758,7 +2786,7 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRow
ZSTD_compressBlock_lazy2_dedicatedDictSearch_row }
};
DEBUGLOG(4, "Selecting a row-based matchfinder");
assert(useRowMatchFinder != ZSTD_urm_auto);
assert(useRowMatchFinder != ZSTD_ps_auto);
selectedCompressor = rowBasedBlockCompressors[(int)dictMode][(int)strat - (int)ZSTD_greedy];
} else {
selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
@@ -2825,7 +2853,7 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
}
if (zc->externSeqStore.pos < zc->externSeqStore.size) {
assert(!zc->appliedParams.ldmParams.enableLdm);
assert(zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_disable);
/* Updates ldmSeqStore.pos */
lastLLSize =
ZSTD_ldm_blockCompress(&zc->externSeqStore,
@@ -2834,7 +2862,7 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
zc->appliedParams.useRowMatchFinder,
src, srcSize);
assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
} else if (zc->appliedParams.ldmParams.enableLdm) {
} else if (zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
rawSeqStore_t ldmSeqStore = kNullRawSeqStore;
ldmSeqStore.seq = zc->ldmSequences;
@@ -3027,7 +3055,7 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_hufCTablesMetadata_t* hufMetadata,
const int disableLiteralsCompression,
const int literalsCompressionIsDisabled,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
@@ -3045,7 +3073,7 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
/* Prepare nextEntropy assuming reusing the existing table */
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralsCompression) {
if (literalsCompressionIsDisabled) {
DEBUGLOG(5, "set_basic - disabled");
hufMetadata->hType = set_basic;
return 0;
@@ -3192,7 +3220,7 @@ size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
&prevEntropy->huf, &nextEntropy->huf,
&entropyMetadata->hufMetadata,
ZSTD_disableLiteralsCompression(cctxParams),
ZSTD_literalsCompressionIsDisabled(cctxParams),
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
entropyMetadata->fseMetadata.fseTablesSize =
@@ -3319,19 +3347,20 @@ static size_t ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
*
* Returns the estimated compressed size of the seqStore, or a zstd error.
*/
static size_t ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, const ZSTD_CCtx* zc) {
ZSTD_entropyCTablesMetadata_t entropyMetadata;
static size_t ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, ZSTD_CCtx* zc) {
ZSTD_entropyCTablesMetadata_t* entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
DEBUGLOG(6, "ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize()");
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
&entropyMetadata,
entropyMetadata,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
return ZSTD_estimateBlockSize(seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
(size_t)(seqStore->sequences - seqStore->sequencesStart),
&zc->blockState.nextCBlock->entropy, &entropyMetadata, zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
(int)(entropyMetadata.hufMetadata.hType == set_compressed), 1);
&zc->blockState.nextCBlock->entropy, entropyMetadata, zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
(int)(entropyMetadata->hufMetadata.hType == set_compressed), 1);
}
/* Returns literals bytes represented in a seqStore */
@@ -3474,6 +3503,7 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
if (isPartition)
ZSTD_seqStore_resolveOffCodes(dRep, cRep, seqStore, (U32)(seqStore->sequences - seqStore->sequencesStart));
RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "Block header doesn't fit");
cSeqsSize = ZSTD_entropyCompressSeqStore(seqStore,
&zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
@@ -3499,9 +3529,6 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
return 0;
}
if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
if (cSeqsSize == 0) {
cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
FORWARD_IF_ERROR(cSize, "Nocompress block failed");
@@ -3518,6 +3545,10 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
cSize = ZSTD_blockHeaderSize + cSeqsSize;
DEBUGLOG(4, "Writing out compressed block, size: %zu", cSize);
}
if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
return cSize;
}
@@ -3528,7 +3559,6 @@ typedef struct {
} seqStoreSplits;
#define MIN_SEQUENCES_BLOCK_SPLITTING 300
#define MAX_NB_SPLITS 196
/* Helper function to perform the recursive search for block splits.
* Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
@@ -3539,29 +3569,31 @@ typedef struct {
* In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
* In practice, recursion depth usually doesn't go beyond 4.
*
* Furthermore, the number of splits is capped by MAX_NB_SPLITS. At MAX_NB_SPLITS == 196 with the current existing blockSize
* Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS. At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
* maximum of 128 KB, this value is actually impossible to reach.
*/
static void ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
const ZSTD_CCtx* zc, const seqStore_t* origSeqStore) {
seqStore_t fullSeqStoreChunk;
seqStore_t firstHalfSeqStore;
seqStore_t secondHalfSeqStore;
ZSTD_CCtx* zc, const seqStore_t* origSeqStore) {
seqStore_t* fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
seqStore_t* firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
seqStore_t* secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
size_t estimatedOriginalSize;
size_t estimatedFirstHalfSize;
size_t estimatedSecondHalfSize;
size_t midIdx = (startIdx + endIdx)/2;
if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= MAX_NB_SPLITS) {
if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= ZSTD_MAX_NB_BLOCK_SPLITS) {
DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences");
return;
}
ZSTD_deriveSeqStoreChunk(&fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
ZSTD_deriveSeqStoreChunk(&firstHalfSeqStore, origSeqStore, startIdx, midIdx);
ZSTD_deriveSeqStoreChunk(&secondHalfSeqStore, origSeqStore, midIdx, endIdx);
estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&fullSeqStoreChunk, zc);
estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&firstHalfSeqStore, zc);
estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&secondHalfSeqStore, zc);
DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
DEBUGLOG(4, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
ZSTD_deriveSeqStoreChunk(fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
ZSTD_deriveSeqStoreChunk(firstHalfSeqStore, origSeqStore, startIdx, midIdx);
ZSTD_deriveSeqStoreChunk(secondHalfSeqStore, origSeqStore, midIdx, endIdx);
estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(fullSeqStoreChunk, zc);
estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(firstHalfSeqStore, zc);
estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(secondHalfSeqStore, zc);
DEBUGLOG(4, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
return;
@@ -3601,12 +3633,12 @@ static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, s
size_t cSize = 0;
const BYTE* ip = (const BYTE*)src;
BYTE* op = (BYTE*)dst;
U32 partitions[MAX_NB_SPLITS];
size_t i = 0;
size_t srcBytesTotal = 0;
U32* partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
seqStore_t* nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
seqStore_t* currSeqStore = &zc->blockSplitCtx.currSeqStore;
size_t numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
seqStore_t nextSeqStore;
seqStore_t currSeqStore;
/* If a block is split and some partitions are emitted as RLE/uncompressed, then repcode history
* may become invalid. In order to reconcile potentially invalid repcodes, we keep track of two
@@ -3626,6 +3658,7 @@ static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, s
repcodes_t cRep;
ZSTD_memcpy(dRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
ZSTD_memcpy(cRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
ZSTD_memset(nextSeqStore, 0, sizeof(seqStore_t));
DEBUGLOG(4, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
(unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
@@ -3643,36 +3676,36 @@ static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, s
return cSizeSingleBlock;
}
ZSTD_deriveSeqStoreChunk(&currSeqStore, &zc->seqStore, 0, partitions[0]);
ZSTD_deriveSeqStoreChunk(currSeqStore, &zc->seqStore, 0, partitions[0]);
for (i = 0; i <= numSplits; ++i) {
size_t srcBytes;
size_t cSizeChunk;
U32 const lastPartition = (i == numSplits);
U32 lastBlockEntireSrc = 0;
srcBytes = ZSTD_countSeqStoreLiteralsBytes(&currSeqStore) + ZSTD_countSeqStoreMatchBytes(&currSeqStore);
srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
srcBytesTotal += srcBytes;
if (lastPartition) {
/* This is the final partition, need to account for possible last literals */
srcBytes += blockSize - srcBytesTotal;
lastBlockEntireSrc = lastBlock;
} else {
ZSTD_deriveSeqStoreChunk(&nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
ZSTD_deriveSeqStoreChunk(nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
}
cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, &currSeqStore,
cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, currSeqStore,
&dRep, &cRep,
op, dstCapacity,
ip, srcBytes,
lastBlockEntireSrc, 1 /* isPartition */);
DEBUGLOG(5, "Estimated size: %zu actual size: %zu", ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&currSeqStore, zc), cSizeChunk);
DEBUGLOG(5, "Estimated size: %zu actual size: %zu", ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
ip += srcBytes;
op += cSizeChunk;
dstCapacity -= cSizeChunk;
cSize += cSizeChunk;
currSeqStore = nextSeqStore;
*currSeqStore = *nextSeqStore;
assert(cSizeChunk <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
}
/* cRep and dRep may have diverged during the compression. If so, we use the dRep repcodes
@@ -3690,6 +3723,7 @@ static size_t ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
U32 nbSeq;
size_t cSize;
DEBUGLOG(4, "ZSTD_compressBlock_splitBlock");
assert(zc->appliedParams.useBlockSplitter == ZSTD_ps_enable);
{ const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
@@ -3704,7 +3738,6 @@ static size_t ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
nbSeq = (U32)(zc->seqStore.sequences - zc->seqStore.sequencesStart);
}
assert(zc->appliedParams.splitBlocks == 1);
cSize = ZSTD_compressBlock_splitBlock_internal(zc, dst, dstCapacity, src, srcSize, lastBlock, nbSeq);
FORWARD_IF_ERROR(cSize, "Splitting blocks failed!");
return cSize;
@@ -3746,12 +3779,6 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
zc->bmi2);
if (zc->seqCollector.collectSequences) {
ZSTD_copyBlockSequences(zc);
return 0;
}
if (frame &&
/* We don't want to emit our first block as a RLE even if it qualifies because
* doing so will cause the decoder (cli only) to throw a "should consume all input error."
@@ -3915,6 +3942,7 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
ZSTD_overflowCorrectIfNeeded(
ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
ZSTD_window_enforceMaxDist(&ms->window, ip, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
/* Ensure hash/chain table insertion resumes no sooner than lowlimit */
if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
@@ -3991,7 +4019,9 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
if (!singleSegment) op[pos++] = windowLogByte;
switch(dictIDSizeCode)
{
default: assert(0); /* impossible */
default:
assert(0); /* impossible */
ZSTD_FALLTHROUGH;
case 0 : break;
case 1 : op[pos] = (BYTE)(dictID); pos++; break;
case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
@@ -3999,7 +4029,9 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
}
switch(fcsCode)
{
default: assert(0); /* impossible */
default:
assert(0); /* impossible */
ZSTD_FALLTHROUGH;
case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
@@ -4047,7 +4079,7 @@ size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSe
{
RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong,
"wrong cctx stage");
RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm,
RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable,
parameter_unsupported,
"incompatible with ldm");
cctx->externSeqStore.seq = seq;
@@ -4088,7 +4120,7 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
ms->forceNonContiguous = 0;
ms->nextToUpdate = ms->window.dictLimit;
}
if (cctx->appliedParams.ldmParams.enableLdm) {
if (cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
ZSTD_window_update(&cctx->ldmState.window, src, srcSize, /* forceNonContiguous */ 0);
}
@@ -4157,7 +4189,7 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
{
const BYTE* ip = (const BYTE*) src;
const BYTE* const iend = ip + srcSize;
int const loadLdmDict = params->ldmParams.enableLdm && ls != NULL;
int const loadLdmDict = params->ldmParams.enableLdm == ZSTD_ps_enable && ls != NULL;
/* Assert that we the ms params match the params we're being given */
ZSTD_assertEqualCParams(params->cParams, ms->cParams);
@@ -4214,8 +4246,8 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
assert(ms->chainTable != NULL);
ZSTD_dedicatedDictSearch_lazy_loadDictionary(ms, iend-HASH_READ_SIZE);
} else {
assert(params->useRowMatchFinder != ZSTD_urm_auto);
if (params->useRowMatchFinder == ZSTD_urm_enableRowMatchFinder) {
assert(params->useRowMatchFinder != ZSTD_ps_auto);
if (params->useRowMatchFinder == ZSTD_ps_enable) {
size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog) * sizeof(U16);
ZSTD_memset(ms->tagTable, 0, tagTableSize);
ZSTD_row_update(ms, iend-HASH_READ_SIZE);
@@ -4715,7 +4747,7 @@ size_t ZSTD_estimateCDictSize_advanced(
+ ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
/* enableDedicatedDictSearch == 1 ensures that CDict estimation will not be too small
* in case we are using DDS with row-hash. */
+ ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_urm_auto, &cParams),
+ ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams),
/* enableDedicatedDictSearch */ 1, /* forCCtx */ 0)
+ (dictLoadMethod == ZSTD_dlm_byRef ? 0
: ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
@@ -4792,7 +4824,7 @@ static size_t ZSTD_initCDict_internal(
static ZSTD_CDict* ZSTD_createCDict_advanced_internal(size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_compressionParameters cParams,
ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
ZSTD_paramSwitch_e useRowMatchFinder,
U32 enableDedicatedDictSearch,
ZSTD_customMem customMem)
{
@@ -4947,7 +4979,7 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
ZSTD_dictContentType_e dictContentType,
ZSTD_compressionParameters cParams)
{
ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_urm_auto, &cParams);
ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams);
/* enableDedicatedDictSearch == 1 ensures matchstate is not too small in case this CDict will be used for DDS + row hash */
size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0);
size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
@@ -5403,7 +5435,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
zcs->outBuffFlushedSize = 0;
zcs->streamStage = zcss_flush; /* pass-through to flush stage */
}
/* fall-through */
ZSTD_FALLTHROUGH;
case zcss_flush:
DEBUGLOG(5, "flush stage");
assert(zcs->appliedParams.outBufferMode == ZSTD_bm_buffered);
@@ -5524,17 +5556,8 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
dictSize, mode);
}
if (ZSTD_CParams_shouldEnableLdm(&params.cParams)) {
/* Enable LDM by default for optimal parser and window size >= 128MB */
DEBUGLOG(4, "LDM enabled by default (window size >= 128MB, strategy >= btopt)");
params.ldmParams.enableLdm = 1;
}
if (ZSTD_CParams_useBlockSplitter(&params.cParams)) {
DEBUGLOG(4, "Block splitter enabled by default (window size >= 128K, strategy >= btopt)");
params.splitBlocks = 1;
}
params.useBlockSplitter = ZSTD_resolveBlockSplitterMode(params.useBlockSplitter, &params.cParams);
params.ldmParams.enableLdm = ZSTD_resolveEnableLdm(params.ldmParams.enableLdm, &params.cParams);
params.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params.useRowMatchFinder, &params.cParams);
#ifdef ZSTD_MULTITHREAD
@@ -6140,119 +6163,12 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
/*-===== Pre-defined compression levels =====-*/
#include "clevels.h"
#define ZSTD_MAX_CLEVEL 22
int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
int ZSTD_maxCLevel(void) { return MEM_32bits() ? ZSTD_MAX_32BIT_CLEVEL : ZSTD_MAX_CLEVEL; }
int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
int ZSTD_defaultCLevel(void) { return ZSTD_CLEVEL_DEFAULT; }
static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
{ /* "default" - for any srcSize > 256 KB */
/* W, C, H, S, L, TL, strat */
{ 19, 12, 13, 1, 6, 1, ZSTD_fast }, /* base for negative levels */
{ 19, 13, 14, 1, 7, 0, ZSTD_fast }, /* level 1 */
{ 20, 15, 16, 1, 6, 0, ZSTD_fast }, /* level 2 */
{ 21, 16, 17, 1, 5, 0, ZSTD_dfast }, /* level 3 */
{ 21, 18, 18, 1, 5, 0, ZSTD_dfast }, /* level 4 */
{ 21, 18, 19, 2, 5, 2, ZSTD_greedy }, /* level 5 */
{ 21, 19, 19, 3, 5, 4, ZSTD_greedy }, /* level 6 */
{ 21, 19, 19, 3, 5, 8, ZSTD_lazy }, /* level 7 */
{ 21, 19, 19, 3, 5, 16, ZSTD_lazy2 }, /* level 8 */
{ 21, 19, 20, 4, 5, 16, ZSTD_lazy2 }, /* level 9 */
{ 22, 20, 21, 4, 5, 16, ZSTD_lazy2 }, /* level 10 */
{ 22, 21, 22, 4, 5, 16, ZSTD_lazy2 }, /* level 11 */
{ 22, 21, 22, 5, 5, 16, ZSTD_lazy2 }, /* level 12 */
{ 22, 21, 22, 5, 5, 32, ZSTD_btlazy2 }, /* level 13 */
{ 22, 22, 23, 5, 5, 32, ZSTD_btlazy2 }, /* level 14 */
{ 22, 23, 23, 6, 5, 32, ZSTD_btlazy2 }, /* level 15 */
{ 22, 22, 22, 5, 5, 48, ZSTD_btopt }, /* level 16 */
{ 23, 23, 22, 5, 4, 64, ZSTD_btopt }, /* level 17 */
{ 23, 23, 22, 6, 3, 64, ZSTD_btultra }, /* level 18 */
{ 23, 24, 22, 7, 3,256, ZSTD_btultra2}, /* level 19 */
{ 25, 25, 23, 7, 3,256, ZSTD_btultra2}, /* level 20 */
{ 26, 26, 24, 7, 3,512, ZSTD_btultra2}, /* level 21 */
{ 27, 27, 25, 9, 3,999, ZSTD_btultra2}, /* level 22 */
},
{ /* for srcSize <= 256 KB */
/* W, C, H, S, L, T, strat */
{ 18, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 18, 13, 14, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 18, 14, 14, 1, 5, 0, ZSTD_dfast }, /* level 2 */
{ 18, 16, 16, 1, 4, 0, ZSTD_dfast }, /* level 3 */
{ 18, 16, 17, 2, 5, 2, ZSTD_greedy }, /* level 4.*/
{ 18, 18, 18, 3, 5, 2, ZSTD_greedy }, /* level 5.*/
{ 18, 18, 19, 3, 5, 4, ZSTD_lazy }, /* level 6.*/
{ 18, 18, 19, 4, 4, 4, ZSTD_lazy }, /* level 7 */
{ 18, 18, 19, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 18, 18, 19, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 18, 18, 19, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 18, 18, 19, 5, 4, 12, ZSTD_btlazy2 }, /* level 11.*/
{ 18, 19, 19, 7, 4, 12, ZSTD_btlazy2 }, /* level 12.*/
{ 18, 18, 19, 4, 4, 16, ZSTD_btopt }, /* level 13 */
{ 18, 18, 19, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 18, 18, 19, 6, 3,128, ZSTD_btopt }, /* level 15.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra2}, /* level 18.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 18, 19, 19, 10, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 18, 19, 19, 12, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 18, 19, 19, 13, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 128 KB */
/* W, C, H, S, L, T, strat */
{ 17, 12, 12, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 17, 12, 13, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 17, 13, 15, 1, 5, 0, ZSTD_fast }, /* level 2 */
{ 17, 15, 16, 2, 5, 0, ZSTD_dfast }, /* level 3 */
{ 17, 17, 17, 2, 4, 0, ZSTD_dfast }, /* level 4 */
{ 17, 16, 17, 3, 4, 2, ZSTD_greedy }, /* level 5 */
{ 17, 17, 17, 3, 4, 4, ZSTD_lazy }, /* level 6 */
{ 17, 17, 17, 3, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 17, 17, 17, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 17, 17, 17, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 17, 17, 17, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 17, 17, 17, 5, 4, 8, ZSTD_btlazy2 }, /* level 11 */
{ 17, 18, 17, 7, 4, 12, ZSTD_btlazy2 }, /* level 12 */
{ 17, 18, 17, 3, 4, 12, ZSTD_btopt }, /* level 13.*/
{ 17, 18, 17, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 17, 18, 17, 6, 3,256, ZSTD_btopt }, /* level 15.*/
{ 17, 18, 17, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 17, 18, 17, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 17, 18, 17, 10, 3,512, ZSTD_btultra }, /* level 18.*/
{ 17, 18, 17, 5, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 17, 18, 17, 7, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 17, 18, 17, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 17, 18, 17, 11, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 16 KB */
/* W, C, H, S, L, T, strat */
{ 14, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 14, 14, 15, 1, 5, 0, ZSTD_fast }, /* level 1 */
{ 14, 14, 15, 1, 4, 0, ZSTD_fast }, /* level 2 */
{ 14, 14, 15, 2, 4, 0, ZSTD_dfast }, /* level 3 */
{ 14, 14, 14, 4, 4, 2, ZSTD_greedy }, /* level 4 */
{ 14, 14, 14, 3, 4, 4, ZSTD_lazy }, /* level 5.*/
{ 14, 14, 14, 4, 4, 8, ZSTD_lazy2 }, /* level 6 */
{ 14, 14, 14, 6, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 14, 14, 14, 8, 4, 8, ZSTD_lazy2 }, /* level 8.*/
{ 14, 15, 14, 5, 4, 8, ZSTD_btlazy2 }, /* level 9.*/
{ 14, 15, 14, 9, 4, 8, ZSTD_btlazy2 }, /* level 10.*/
{ 14, 15, 14, 3, 4, 12, ZSTD_btopt }, /* level 11.*/
{ 14, 15, 14, 4, 3, 24, ZSTD_btopt }, /* level 12.*/
{ 14, 15, 14, 5, 3, 32, ZSTD_btultra }, /* level 13.*/
{ 14, 15, 15, 6, 3, 64, ZSTD_btultra }, /* level 14.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra }, /* level 15.*/
{ 14, 15, 15, 5, 3, 48, ZSTD_btultra2}, /* level 16.*/
{ 14, 15, 15, 6, 3,128, ZSTD_btultra2}, /* level 17.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra2}, /* level 18.*/
{ 14, 15, 15, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 14, 15, 15, 8, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 14, 15, 15, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 14, 15, 15, 10, 3,999, ZSTD_btultra2}, /* level 22.*/
},
};
static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(int const compressionLevel, size_t const dictSize)
{
ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, 0, dictSize, ZSTD_cpm_createCDict);
@@ -6345,7 +6261,7 @@ static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel,
/* row */
if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
else if (compressionLevel < 0) row = 0; /* entry 0 is baseline for fast mode */
else if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
else if (compressionLevel > ZSTD_maxCLevel()) row = ZSTD_maxCLevel();
else row = compressionLevel;
{ ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
lib/compress/zstd_compress_internal.h
+102 -44
View File
@@ -63,7 +63,7 @@ typedef struct {
} ZSTD_localDict;
typedef struct {
HUF_CElt CTable[HUF_CTABLE_SIZE_U32(255)];
HUF_CElt CTable[HUF_CTABLE_SIZE_ST(255)];
HUF_repeat repeatMode;
} ZSTD_hufCTables_t;
@@ -179,7 +179,7 @@ typedef struct {
U32 offCodeSumBasePrice; /* to compare to log2(offreq) */
ZSTD_OptPrice_e priceType; /* prices can be determined dynamically, or follow a pre-defined cost structure */
const ZSTD_entropyCTables_t* symbolCosts; /* pre-calculated dictionary statistics */
ZSTD_literalCompressionMode_e literalCompressionMode;
ZSTD_paramSwitch_e literalCompressionMode;
} optState_t;
typedef struct {
@@ -199,6 +199,8 @@ typedef struct {
*/
} ZSTD_window_t;
#define ZSTD_WINDOW_START_INDEX 2
typedef struct ZSTD_matchState_t ZSTD_matchState_t;
#define ZSTD_ROW_HASH_CACHE_SIZE 8 /* Size of prefetching hash cache for row-based matchfinder */
@@ -264,7 +266,7 @@ typedef struct {
} ldmState_t;
typedef struct {
U32 enableLdm; /* 1 if enable long distance matching */
ZSTD_paramSwitch_e enableLdm; /* ZSTD_ps_enable to enable LDM. ZSTD_ps_auto by default */
U32 hashLog; /* Log size of hashTable */
U32 bucketSizeLog; /* Log bucket size for collision resolution, at most 8 */
U32 minMatchLength; /* Minimum match length */
@@ -295,7 +297,7 @@ struct ZSTD_CCtx_params_s {
* There is no guarantee that hint is close to actual source size */
ZSTD_dictAttachPref_e attachDictPref;
ZSTD_literalCompressionMode_e literalCompressionMode;
ZSTD_paramSwitch_e literalCompressionMode;
/* Multithreading: used to pass parameters to mtctx */
int nbWorkers;
@@ -318,10 +320,10 @@ struct ZSTD_CCtx_params_s {
int validateSequences;
/* Block splitting */
int splitBlocks;
ZSTD_paramSwitch_e useBlockSplitter;
/* Param for deciding whether to use row-based matchfinder */
ZSTD_useRowMatchFinderMode_e useRowMatchFinder;
ZSTD_paramSwitch_e useRowMatchFinder;
/* Always load a dictionary in ext-dict mode (not prefix mode)? */
int deterministicRefPrefix;
@@ -343,6 +345,22 @@ typedef enum {
ZSTDb_buffered
} ZSTD_buffered_policy_e;
/**
* Struct that contains all elements of block splitter that should be allocated
* in a wksp.
*/
#define ZSTD_MAX_NB_BLOCK_SPLITS 196
typedef struct {
seqStore_t fullSeqStoreChunk;
seqStore_t firstHalfSeqStore;
seqStore_t secondHalfSeqStore;
seqStore_t currSeqStore;
seqStore_t nextSeqStore;
U32 partitions[ZSTD_MAX_NB_BLOCK_SPLITS];
ZSTD_entropyCTablesMetadata_t entropyMetadata;
} ZSTD_blockSplitCtx;
struct ZSTD_CCtx_s {
ZSTD_compressionStage_e stage;
int cParamsChanged; /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
@@ -374,7 +392,7 @@ struct ZSTD_CCtx_s {
ZSTD_blockState_t blockState;
U32* entropyWorkspace; /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
/* Wether we are streaming or not */
/* Whether we are streaming or not */
ZSTD_buffered_policy_e bufferedPolicy;
/* streaming */
@@ -408,6 +426,9 @@ struct ZSTD_CCtx_s {
#if ZSTD_TRACE
ZSTD_TraceCtx traceCtx;
#endif
/* Workspace for block splitter */
ZSTD_blockSplitCtx blockSplitCtx;
};
typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
@@ -442,7 +463,7 @@ typedef enum {
typedef size_t (*ZSTD_blockCompressor) (
ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRowMatchFinderMode_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
@@ -549,17 +570,17 @@ MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
return (srcSize >> minlog) + 2;
}
MEM_STATIC int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParams)
MEM_STATIC int ZSTD_literalsCompressionIsDisabled(const ZSTD_CCtx_params* cctxParams)
{
switch (cctxParams->literalCompressionMode) {
case ZSTD_lcm_huffman:
case ZSTD_ps_enable:
return 0;
case ZSTD_lcm_uncompressed:
case ZSTD_ps_disable:
return 1;
default:
assert(0 /* impossible: pre-validated */);
/* fall-through */
case ZSTD_lcm_auto:
ZSTD_FALLTHROUGH;
case ZSTD_ps_auto:
return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
}
}
@@ -651,8 +672,14 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
# if STATIC_BMI2
return _tzcnt_u64(val) >> 3;
# else
unsigned long r = 0;
return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
if (val != 0) {
unsigned long r;
_BitScanForward64(&r, (U64)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (__builtin_ctzll((U64)val) >> 3);
@@ -669,8 +696,14 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
# endif
} else { /* 32 bits */
# if defined(_MSC_VER)
unsigned long r=0;
return _BitScanForward( &r, (U32)val ) ? (unsigned)(r >> 3) : 0;
if (val != 0) {
unsigned long r;
_BitScanForward(&r, (U32)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_ctz((U32)val) >> 3);
# else
@@ -687,8 +720,14 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
# if STATIC_BMI2
return _lzcnt_u64(val) >> 3;
# else
unsigned long r = 0;
return _BitScanReverse64(&r, (U64)val) ? (unsigned)(r >> 3) : 0;
if (val != 0) {
unsigned long r;
_BitScanReverse64(&r, (U64)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (__builtin_clzll(val) >> 3);
@@ -702,8 +741,14 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
# endif
} else { /* 32 bits */
# if defined(_MSC_VER)
unsigned long r = 0;
return _BitScanReverse( &r, (unsigned long)val ) ? (unsigned)(r >> 3) : 0;
if (val != 0) {
unsigned long r;
_BitScanReverse(&r, (unsigned long)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_clz((U32)val) >> 3);
# else
@@ -884,9 +929,9 @@ MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
{
return window.dictLimit == 1 &&
window.lowLimit == 1 &&
(window.nextSrc - window.base) == 1;
return window.dictLimit == ZSTD_WINDOW_START_INDEX &&
window.lowLimit == ZSTD_WINDOW_START_INDEX &&
(window.nextSrc - window.base) == ZSTD_WINDOW_START_INDEX;
}
/**
@@ -937,7 +982,9 @@ MEM_STATIC U32 ZSTD_window_canOverflowCorrect(ZSTD_window_t const window,
{
U32 const cycleSize = 1u << cycleLog;
U32 const curr = (U32)((BYTE const*)src - window.base);
U32 const minIndexToOverflowCorrect = cycleSize + MAX(maxDist, cycleSize);
U32 const minIndexToOverflowCorrect = cycleSize
+ MAX(maxDist, cycleSize)
+ ZSTD_WINDOW_START_INDEX;
/* Adjust the min index to backoff the overflow correction frequency,
* so we don't waste too much CPU in overflow correction. If this
@@ -1012,10 +1059,14 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
U32 const cycleSize = 1u << cycleLog;
U32 const cycleMask = cycleSize - 1;
U32 const curr = (U32)((BYTE const*)src - window->base);
U32 const currentCycle0 = curr & cycleMask;
/* Exclude zero so that newCurrent - maxDist >= 1. */
U32 const currentCycle1 = currentCycle0 == 0 ? cycleSize : currentCycle0;
U32 const newCurrent = currentCycle1 + MAX(maxDist, cycleSize);
U32 const currentCycle = curr & cycleMask;
/* Ensure newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX. */
U32 const currentCycleCorrection = currentCycle < ZSTD_WINDOW_START_INDEX
? MAX(cycleSize, ZSTD_WINDOW_START_INDEX)
: 0;
U32 const newCurrent = currentCycle
+ currentCycleCorrection
+ MAX(maxDist, cycleSize);
U32 const correction = curr - newCurrent;
/* maxDist must be a power of two so that:
* (newCurrent & cycleMask) == (curr & cycleMask)
@@ -1031,14 +1082,20 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
window->base += correction;
window->dictBase += correction;
if (window->lowLimit <= correction) window->lowLimit = 1;
else window->lowLimit -= correction;
if (window->dictLimit <= correction) window->dictLimit = 1;
else window->dictLimit -= correction;
if (window->lowLimit < correction + ZSTD_WINDOW_START_INDEX) {
window->lowLimit = ZSTD_WINDOW_START_INDEX;
} else {
window->lowLimit -= correction;
}
if (window->dictLimit < correction + ZSTD_WINDOW_START_INDEX) {
window->dictLimit = ZSTD_WINDOW_START_INDEX;
} else {
window->dictLimit -= correction;
}
/* Ensure we can still reference the full window. */
assert(newCurrent >= maxDist);
assert(newCurrent - maxDist >= 1);
assert(newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX);
/* Ensure that lowLimit and dictLimit didn't underflow. */
assert(window->lowLimit <= newCurrent);
assert(window->dictLimit <= newCurrent);
@@ -1149,11 +1206,12 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
ZSTD_memset(window, 0, sizeof(*window));
window->base = (BYTE const*)"";
window->dictBase = (BYTE const*)"";
window->dictLimit = 1; /* start from 1, so that 1st position is valid */
window->lowLimit = 1; /* it ensures first and later CCtx usages compress the same */
window->nextSrc = window->base + 1; /* see issue #1241 */
window->base = (BYTE const*)" ";
window->dictBase = (BYTE const*)" ";
ZSTD_STATIC_ASSERT(ZSTD_DUBT_UNSORTED_MARK < ZSTD_WINDOW_START_INDEX); /* Start above ZSTD_DUBT_UNSORTED_MARK */
window->dictLimit = ZSTD_WINDOW_START_INDEX; /* start from >0, so that 1st position is valid */
window->lowLimit = ZSTD_WINDOW_START_INDEX; /* it ensures first and later CCtx usages compress the same */
window->nextSrc = window->base + ZSTD_WINDOW_START_INDEX; /* see issue #1241 */
window->nbOverflowCorrections = 0;
}
@@ -1206,15 +1264,15 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
*/
MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
{
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.lowLimit;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.lowLimit;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
/* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
* is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
* valid for the entire block. So this check is sufficient to find the lowest valid match index.
*/
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
return matchLowest;
}
lib/compress/zstd_compress_literals.c
+4 -3
View File
@@ -73,7 +73,8 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
void* entropyWorkspace, size_t entropyWorkspaceSize,
const int bmi2)
const int bmi2,
unsigned suspectUncompressible)
{
size_t const minGain = ZSTD_minGain(srcSize, strategy);
size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
@@ -105,11 +106,11 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
HUF_compress1X_repeat(
ostart+lhSize, dstCapacity-lhSize, src, srcSize,
HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
(HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) :
(HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2, suspectUncompressible) :
HUF_compress4X_repeat(
ostart+lhSize, dstCapacity-lhSize, src, srcSize,
HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
(HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
(HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2, suspectUncompressible);
if (repeat != HUF_repeat_none) {
/* reused the existing table */
DEBUGLOG(5, "Reusing previous huffman table");
lib/compress/zstd_compress_literals.h
+3 -1
View File
@@ -18,12 +18,14 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
/* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_strategy strategy, int disableLiteralCompression,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
void* entropyWorkspace, size_t entropyWorkspaceSize,
const int bmi2);
const int bmi2,
unsigned suspectUncompressible);
#endif /* ZSTD_COMPRESS_LITERALS_H */
lib/compress/zstd_compress_sequences.c
+5 -4
View File
@@ -275,10 +275,11 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
assert(nbSeq_1 > 1);
assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
(void)entropyWorkspaceSize;
FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
{ size_t const NCountSize = FSE_writeNCount(op, oend - op, wksp->norm, max, tableLog); /* overflow protected */
FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
assert(oend >= op);
{ size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog); /* overflow protected */
FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
return NCountSize;
}
}
@@ -398,7 +399,7 @@ ZSTD_encodeSequences_default(
#if DYNAMIC_BMI2
static TARGET_ATTRIBUTE("bmi2") size_t
static BMI2_TARGET_ATTRIBUTE size_t
ZSTD_encodeSequences_bmi2(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
lib/compress/zstd_compress_superblock.c
+2 -1
View File
@@ -132,6 +132,7 @@ static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef*
const seqDef* sp = sstart;
size_t matchLengthSum = 0;
size_t litLengthSum = 0;
(void)(litLengthSum); /* suppress unused variable warning on some environments */
while (send-sp > 0) {
ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
litLengthSum += seqLen.litLength;
@@ -474,7 +475,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
/* I think there is an optimization opportunity here.
* Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
* since it recalculates estimate from scratch.
* For example, it would recount literal distribution and symbol codes everytime.
* For example, it would recount literal distribution and symbol codes every time.
*/
cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
&nextCBlock->entropy, entropyMetadata,
lib/compress/zstd_cwksp.h
+1 -1
View File
@@ -219,7 +219,7 @@ MEM_STATIC size_t ZSTD_cwksp_aligned_alloc_size(size_t size) {
MEM_STATIC size_t ZSTD_cwksp_slack_space_required(void) {
/* For alignment, the wksp will always allocate an additional n_1=[1, 64] bytes
* to align the beginning of tables section, as well as another n_2=[0, 63] bytes
* to align the beginning of the aligned secion.
* to align the beginning of the aligned section.
*
* n_1 + n_2 == 64 bytes if the cwksp is freshly allocated, due to tables and
* aligneds being sized in multiples of 64 bytes.
lib/compress/zstd_double_fast.c
+289 -114
View File
@@ -48,10 +48,216 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
FORCE_INLINE_TEMPLATE
size_t ZSTD_compressBlock_doubleFast_generic(
size_t ZSTD_compressBlock_doubleFast_noDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize, U32 const mls /* template */)
{
ZSTD_compressionParameters const* cParams = &ms->cParams;
U32* const hashLong = ms->hashTable;
const U32 hBitsL = cParams->hashLog;
U32* const hashSmall = ms->chainTable;
const U32 hBitsS = cParams->chainLog;
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const BYTE* anchor = istart;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
/* presumes that, if there is a dictionary, it must be using Attach mode */
const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
const BYTE* const prefixLowest = base + prefixLowestIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
U32 offset_1=rep[0], offset_2=rep[1];
U32 offsetSaved = 0;
size_t mLength;
U32 offset;
U32 curr;
/* how many positions to search before increasing step size */
const size_t kStepIncr = 1 << kSearchStrength;
/* the position at which to increment the step size if no match is found */
const BYTE* nextStep;
size_t step; /* the current step size */
size_t hl0; /* the long hash at ip */
size_t hl1; /* the long hash at ip1 */
U32 idxl0; /* the long match index for ip */
U32 idxl1; /* the long match index for ip1 */
const BYTE* matchl0; /* the long match for ip */
const BYTE* matchs0; /* the short match for ip */
const BYTE* matchl1; /* the long match for ip1 */
const BYTE* ip = istart; /* the current position */
const BYTE* ip1; /* the next position */
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
/* init */
ip += ((ip - prefixLowest) == 0);
{
U32 const current = (U32)(ip - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
U32 const maxRep = current - windowLow;
if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
}
/* Outer Loop: one iteration per match found and stored */
while (1) {
step = 1;
nextStep = ip + kStepIncr;
ip1 = ip + step;
if (ip1 > ilimit) {
goto _cleanup;
}
hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
idxl0 = hashLong[hl0];
matchl0 = base + idxl0;
/* Inner Loop: one iteration per search / position */
do {
const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
const U32 idxs0 = hashSmall[hs0];
curr = (U32)(ip-base);
matchs0 = base + idxs0;
hashLong[hl0] = hashSmall[hs0] = curr; /* update hash tables */
/* check noDict repcode */
if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
ip++;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
goto _match_stored;
}
hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
if (idxl0 > prefixLowestIndex) {
/* check prefix long match */
if (MEM_read64(matchl0) == MEM_read64(ip)) {
mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
offset = (U32)(ip-matchl0);
while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
goto _match_found;
}
}
idxl1 = hashLong[hl1];
matchl1 = base + idxl1;
if (idxs0 > prefixLowestIndex) {
/* check prefix short match */
if (MEM_read32(matchs0) == MEM_read32(ip)) {
goto _search_next_long;
}
}
if (ip1 >= nextStep) {
PREFETCH_L1(ip1 + 64);
PREFETCH_L1(ip1 + 128);
step++;
nextStep += kStepIncr;
}
ip = ip1;
ip1 += step;
hl0 = hl1;
idxl0 = idxl1;
matchl0 = matchl1;
#if defined(__aarch64__)
PREFETCH_L1(ip+256);
#endif
} while (ip1 <= ilimit);
_cleanup:
/* save reps for next block */
rep[0] = offset_1 ? offset_1 : offsetSaved;
rep[1] = offset_2 ? offset_2 : offsetSaved;
/* Return the last literals size */
return (size_t)(iend - anchor);
_search_next_long:
/* check prefix long +1 match */
if (idxl1 > prefixLowestIndex) {
if (MEM_read64(matchl1) == MEM_read64(ip1)) {
ip = ip1;
mLength = ZSTD_count(ip+8, matchl1+8, iend) + 8;
offset = (U32)(ip-matchl1);
while (((ip>anchor) & (matchl1>prefixLowest)) && (ip[-1] == matchl1[-1])) { ip--; matchl1--; mLength++; } /* catch up */
goto _match_found;
}
}
/* if no long +1 match, explore the short match we found */
mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
offset = (U32)(ip - matchs0);
while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* catch up */
/* fall-through */
_match_found: /* requires ip, offset, mLength */
offset_2 = offset_1;
offset_1 = offset;
if (step < 4) {
/* It is unsafe to write this value back to the hashtable when ip1 is
* greater than or equal to the new ip we will have after we're done
* processing this match. Rather than perform that test directly
* (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
* more predictable test. The minmatch even if we take a short match is
* 4 bytes, so as long as step, the distance between ip and ip1
* (initially) is less than 4, we know ip1 < new ip. */
hashLong[hl1] = (U32)(ip1 - base);
}
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
_match_stored:
/* match found */
ip += mLength;
anchor = ip;
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
}
/* check immediate repcode */
while ( (ip <= ilimit)
&& ( (offset_2>0)
& (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
/* store sequence */
size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
ip += rLength;
anchor = ip;
continue; /* faster when present ... (?) */
}
}
}
}
FORCE_INLINE_TEMPLATE
size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize,
U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
U32 const mls /* template */)
{
ZSTD_compressionParameters const* cParams = &ms->cParams;
U32* const hashLong = ms->hashTable;
@@ -72,54 +278,30 @@ size_t ZSTD_compressBlock_doubleFast_generic(
U32 offsetSaved = 0;
const ZSTD_matchState_t* const dms = ms->dictMatchState;
const ZSTD_compressionParameters* const dictCParams =
dictMode == ZSTD_dictMatchState ?
&dms->cParams : NULL;
const U32* const dictHashLong = dictMode == ZSTD_dictMatchState ?
dms->hashTable : NULL;
const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
dms->chainTable : NULL;
const U32 dictStartIndex = dictMode == ZSTD_dictMatchState ?
dms->window.dictLimit : 0;
const BYTE* const dictBase = dictMode == ZSTD_dictMatchState ?
dms->window.base : NULL;
const BYTE* const dictStart = dictMode == ZSTD_dictMatchState ?
dictBase + dictStartIndex : NULL;
const BYTE* const dictEnd = dictMode == ZSTD_dictMatchState ?
dms->window.nextSrc : NULL;
const U32 dictIndexDelta = dictMode == ZSTD_dictMatchState ?
prefixLowestIndex - (U32)(dictEnd - dictBase) :
0;
const U32 dictHBitsL = dictMode == ZSTD_dictMatchState ?
dictCParams->hashLog : hBitsL;
const U32 dictHBitsS = dictMode == ZSTD_dictMatchState ?
dictCParams->chainLog : hBitsS;
const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
const U32* const dictHashLong = dms->hashTable;
const U32* const dictHashSmall = dms->chainTable;
const U32 dictStartIndex = dms->window.dictLimit;
const BYTE* const dictBase = dms->window.base;
const BYTE* const dictStart = dictBase + dictStartIndex;
const BYTE* const dictEnd = dms->window.nextSrc;
const U32 dictIndexDelta = prefixLowestIndex - (U32)(dictEnd - dictBase);
const U32 dictHBitsL = dictCParams->hashLog;
const U32 dictHBitsS = dictCParams->chainLog;
const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
/* if a dictionary is attached, it must be within window range */
if (dictMode == ZSTD_dictMatchState) {
assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
}
assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
/* init */
ip += (dictAndPrefixLength == 0);
if (dictMode == ZSTD_noDict) {
U32 const curr = (U32)(ip - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
U32 const maxRep = curr - windowLow;
if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
}
if (dictMode == ZSTD_dictMatchState) {
/* dictMatchState repCode checks don't currently handle repCode == 0
* disabling. */
assert(offset_1 <= dictAndPrefixLength);
assert(offset_2 <= dictAndPrefixLength);
}
/* dictMatchState repCode checks don't currently handle repCode == 0
* disabling. */
assert(offset_1 <= dictAndPrefixLength);
assert(offset_2 <= dictAndPrefixLength);
/* Main Search Loop */
while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */
@@ -135,15 +317,13 @@ size_t ZSTD_compressBlock_doubleFast_generic(
const BYTE* matchLong = base + matchIndexL;
const BYTE* match = base + matchIndexS;
const U32 repIndex = curr + 1 - offset_1;
const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
&& repIndex < prefixLowestIndex) ?
const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
dictBase + (repIndex - dictIndexDelta) :
base + repIndex;
hashLong[h2] = hashSmall[h] = curr; /* update hash tables */
/* check dictMatchState repcode */
if (dictMode == ZSTD_dictMatchState
&& ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
/* check repcode */
if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
@@ -152,15 +332,6 @@ size_t ZSTD_compressBlock_doubleFast_generic(
goto _match_stored;
}
/* check noDict repcode */
if ( dictMode == ZSTD_noDict
&& ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
ip++;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
goto _match_stored;
}
if (matchIndexL > prefixLowestIndex) {
/* check prefix long match */
if (MEM_read64(matchLong) == MEM_read64(ip)) {
@@ -169,7 +340,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
goto _match_found;
}
} else if (dictMode == ZSTD_dictMatchState) {
} else {
/* check dictMatchState long match */
U32 const dictMatchIndexL = dictHashLong[dictHL];
const BYTE* dictMatchL = dictBase + dictMatchIndexL;
@@ -187,7 +358,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
if (MEM_read32(match) == MEM_read32(ip)) {
goto _search_next_long;
}
} else if (dictMode == ZSTD_dictMatchState) {
} else {
/* check dictMatchState short match */
U32 const dictMatchIndexS = dictHashSmall[dictHS];
match = dictBase + dictMatchIndexS;
@@ -220,7 +391,7 @@ _search_next_long:
while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
goto _match_found;
}
} else if (dictMode == ZSTD_dictMatchState) {
} else {
/* check dict long +1 match */
U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
@@ -234,7 +405,7 @@ _search_next_long:
} } }
/* if no long +1 match, explore the short match we found */
if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
if (matchIndexS < prefixLowestIndex) {
mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
offset = (U32)(curr - matchIndexS);
while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
@@ -244,8 +415,6 @@ _search_next_long:
while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
}
/* fall-through */
_match_found:
offset_2 = offset_1;
offset_1 = offset;
@@ -268,43 +437,27 @@ _match_stored:
}
/* check immediate repcode */
if (dictMode == ZSTD_dictMatchState) {
while (ip <= ilimit) {
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
&& repIndex2 < prefixLowestIndex ?
dictBase + repIndex2 - dictIndexDelta :
base + repIndex2;
if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
ip += repLength2;
anchor = ip;
continue;
}
break;
} }
if (dictMode == ZSTD_noDict) {
while ( (ip <= ilimit)
&& ( (offset_2>0)
& (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
/* store sequence */
size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
ip += rLength;
while (ip <= ilimit) {
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
dictBase + repIndex2 - dictIndexDelta :
base + repIndex2;
if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
ip += repLength2;
anchor = ip;
continue; /* faster when present ... (?) */
} } }
continue;
}
break;
}
}
} /* while (ip < ilimit) */
/* save reps for next block */
@@ -315,6 +468,24 @@ _match_stored:
return (size_t)(iend - anchor);
}
#define ZSTD_GEN_DFAST_FN(dictMode, mls) \
static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls( \
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \
void const* src, size_t srcSize) \
{ \
return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
}
ZSTD_GEN_DFAST_FN(noDict, 4)
ZSTD_GEN_DFAST_FN(noDict, 5)
ZSTD_GEN_DFAST_FN(noDict, 6)
ZSTD_GEN_DFAST_FN(noDict, 7)
ZSTD_GEN_DFAST_FN(dictMatchState, 4)
ZSTD_GEN_DFAST_FN(dictMatchState, 5)
ZSTD_GEN_DFAST_FN(dictMatchState, 6)
ZSTD_GEN_DFAST_FN(dictMatchState, 7)
size_t ZSTD_compressBlock_doubleFast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -325,13 +496,13 @@ size_t ZSTD_compressBlock_doubleFast(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
}
}
@@ -345,13 +516,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
}
}
@@ -387,7 +558,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
/* if extDict is invalidated due to maxDistance, switch to "regular" variant */
if (prefixStartIndex == dictStartIndex)
return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
/* Search Loop */
while (ip < ilimit) { /* < instead of <=, because (ip+1) */
@@ -409,7 +580,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
hashSmall[hSmall] = hashLong[hLong] = curr; /* update hash table */
if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
& (offset_1 < curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
& (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
@@ -477,7 +648,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
& (offset_2 < current2 - dictStartIndex))
& (offset_2 <= current2 - dictStartIndex))
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
@@ -500,6 +671,10 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
return (size_t)(iend - anchor);
}
ZSTD_GEN_DFAST_FN(extDict, 4)
ZSTD_GEN_DFAST_FN(extDict, 5)
ZSTD_GEN_DFAST_FN(extDict, 6)
ZSTD_GEN_DFAST_FN(extDict, 7)
size_t ZSTD_compressBlock_doubleFast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -510,12 +685,12 @@ size_t ZSTD_compressBlock_doubleFast_extDict(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
}
}
lib/compress/zstd_fast.c
+258 -108
View File
@@ -43,8 +43,54 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
}
/**
* If you squint hard enough (and ignore repcodes), the search operation at any
* given position is broken into 4 stages:
*
* 1. Hash (map position to hash value via input read)
* 2. Lookup (map hash val to index via hashtable read)
* 3. Load (map index to value at that position via input read)
* 4. Compare
*
* Each of these steps involves a memory read at an address which is computed
* from the previous step. This means these steps must be sequenced and their
* latencies are cumulative.
*
* Rather than do 1->2->3->4 sequentially for a single position before moving
* onto the next, this implementation interleaves these operations across the
* next few positions:
*
* R = Repcode Read & Compare
* H = Hash
* T = Table Lookup
* M = Match Read & Compare
*
* Pos | Time -->
* ----+-------------------
* N | ... M
* N+1 | ... TM
* N+2 | R H T M
* N+3 | H TM
* N+4 | R H T M
* N+5 | H ...
* N+6 | R ...
*
* This is very much analogous to the pipelining of execution in a CPU. And just
* like a CPU, we have to dump the pipeline when we find a match (i.e., take a
* branch).
*
* When this happens, we throw away our current state, and do the following prep
* to re-enter the loop:
*
* Pos | Time -->
* ----+-------------------
* N | H T
* N+1 | H
*
* This is also the work we do at the beginning to enter the loop initially.
*/
FORCE_INLINE_TEMPLATE size_t
ZSTD_compressBlock_fast_generic(
ZSTD_compressBlock_fast_noDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize,
U32 const mls)
@@ -53,135 +99,229 @@ ZSTD_compressBlock_fast_generic(
U32* const hashTable = ms->hashTable;
U32 const hlog = cParams->hashLog;
/* support stepSize of 0 */
size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
size_t const stepSize = cParams->targetLength + !(cParams->targetLength);
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
/* We check ip0 (ip + 0) and ip1 (ip + 1) each loop */
const BYTE* ip0 = istart;
const BYTE* ip1;
const BYTE* anchor = istart;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
const U32 prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
const BYTE* const prefixStart = base + prefixStartIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
U32 offset_1=rep[0], offset_2=rep[1];
const BYTE* anchor = istart;
const BYTE* ip0 = istart;
const BYTE* ip1;
const BYTE* ip2;
const BYTE* ip3;
U32 current0;
U32 rep_offset1 = rep[0];
U32 rep_offset2 = rep[1];
U32 offsetSaved = 0;
/* init */
size_t hash0; /* hash for ip0 */
size_t hash1; /* hash for ip1 */
U32 idx; /* match idx for ip0 */
U32 mval; /* src value at match idx */
U32 offcode;
const BYTE* match0;
size_t mLength;
size_t step;
const BYTE* nextStep;
const size_t kStepIncr = (1 << (kSearchStrength - 1));
DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
ip0 += (ip0 == prefixStart);
ip1 = ip0 + 1;
{ U32 const curr = (U32)(ip0 - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
U32 const maxRep = curr - windowLow;
if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
if (rep_offset2 > maxRep) offsetSaved = rep_offset2, rep_offset2 = 0;
if (rep_offset1 > maxRep) offsetSaved = rep_offset1, rep_offset1 = 0;
}
/* Main Search Loop */
#ifdef __INTEL_COMPILER
/* From intel 'The vector pragma indicates that the loop should be
* vectorized if it is legal to do so'. Can be used together with
* #pragma ivdep (but have opted to exclude that because intel
* warns against using it).*/
#pragma vector always
#endif
while (ip1 < ilimit) { /* < instead of <=, because check at ip0+2 */
size_t mLength;
BYTE const* ip2 = ip0 + 2;
size_t const h0 = ZSTD_hashPtr(ip0, hlog, mls);
U32 const val0 = MEM_read32(ip0);
size_t const h1 = ZSTD_hashPtr(ip1, hlog, mls);
U32 const val1 = MEM_read32(ip1);
U32 const current0 = (U32)(ip0-base);
U32 const current1 = (U32)(ip1-base);
U32 const matchIndex0 = hashTable[h0];
U32 const matchIndex1 = hashTable[h1];
BYTE const* repMatch = ip2 - offset_1;
const BYTE* match0 = base + matchIndex0;
const BYTE* match1 = base + matchIndex1;
U32 offcode;
/* start each op */
_start: /* Requires: ip0 */
#if defined(__aarch64__)
PREFETCH_L1(ip0+256);
#endif
step = stepSize;
nextStep = ip0 + kStepIncr;
hashTable[h0] = current0; /* update hash table */
hashTable[h1] = current1; /* update hash table */
/* calculate positions, ip0 - anchor == 0, so we skip step calc */
ip1 = ip0 + stepSize;
ip2 = ip1 + stepSize;
ip3 = ip2 + stepSize;
assert(ip0 + 1 == ip1);
if (ip3 >= ilimit) {
goto _cleanup;
}
if ((offset_1 > 0) & (MEM_read32(repMatch) == MEM_read32(ip2))) {
mLength = (ip2[-1] == repMatch[-1]) ? 1 : 0;
ip0 = ip2 - mLength;
match0 = repMatch - mLength;
mLength += 4;
hash0 = ZSTD_hashPtr(ip0, hlog, mls);
hash1 = ZSTD_hashPtr(ip1, hlog, mls);
idx = hashTable[hash0];
do {
/* load repcode match for ip[2]*/
const U32 rval = MEM_read32(ip2 - rep_offset1);
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
/* check repcode at ip[2] */
if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
ip0 = ip2;
match0 = ip0 - rep_offset1;
mLength = ip0[-1] == match0[-1];
ip0 -= mLength;
match0 -= mLength;
offcode = 0;
mLength += 4;
goto _match;
}
if ((matchIndex0 > prefixStartIndex) && MEM_read32(match0) == val0) {
/* found a regular match */
/* load match for ip[0] */
if (idx >= prefixStartIndex) {
mval = MEM_read32(base + idx);
} else {
mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
}
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
goto _offset;
}
if ((matchIndex1 > prefixStartIndex) && MEM_read32(match1) == val1) {
/* found a regular match after one literal */
ip0 = ip1;
match0 = match1;
/* lookup ip[1] */
idx = hashTable[hash1];
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip3;
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
/* load match for ip[0] */
if (idx >= prefixStartIndex) {
mval = MEM_read32(base + idx);
} else {
mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
}
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
goto _offset;
}
{ size_t const step = ((size_t)(ip0-anchor) >> (kSearchStrength - 1)) + stepSize;
assert(step >= 2);
ip0 += step;
ip1 += step;
continue;
/* lookup ip[1] */
idx = hashTable[hash1];
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* calculate step */
if (ip2 >= nextStep) {
PREFETCH_L1(ip1 + 64);
PREFETCH_L1(ip1 + 128);
step++;
nextStep += kStepIncr;
}
_offset: /* Requires: ip0, match0 */
/* Compute the offset code */
offset_2 = offset_1;
offset_1 = (U32)(ip0-match0);
offcode = offset_1 + ZSTD_REP_MOVE;
mLength = 4;
/* Count the backwards match length */
while (((ip0>anchor) & (match0>prefixStart))
&& (ip0[-1] == match0[-1])) { ip0--; match0--; mLength++; } /* catch up */
_match: /* Requires: ip0, match0, offcode */
/* Count the forward length */
mLength += ZSTD_count(ip0+mLength, match0+mLength, iend);
ZSTD_storeSeq(seqStore, (size_t)(ip0-anchor), anchor, iend, offcode, mLength-MINMATCH);
/* match found */
ip0 += mLength;
anchor = ip0;
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip2 + step;
ip3 = ip2 + step;
} while (ip3 < ilimit);
if (ip0 <= ilimit) {
/* Fill Table */
assert(base+current0+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2; /* here because current+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
if (offset_2 > 0) { /* offset_2==0 means offset_2 is invalidated */
while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - offset_2)) ) {
/* store sequence */
size_t const rLength = ZSTD_count(ip0+4, ip0+4-offset_2, iend) + 4;
{ U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
ip0 += rLength;
ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
anchor = ip0;
continue; /* faster when present (confirmed on gcc-8) ... (?) */
} } }
ip1 = ip0 + 1;
}
_cleanup:
/* Note that there are probably still a couple positions we could search.
* However, it seems to be a meaningful performance hit to try to search
* them. So let's not. */
/* save reps for next block */
rep[0] = offset_1 ? offset_1 : offsetSaved;
rep[1] = offset_2 ? offset_2 : offsetSaved;
rep[0] = rep_offset1 ? rep_offset1 : offsetSaved;
rep[1] = rep_offset2 ? rep_offset2 : offsetSaved;
/* Return the last literals size */
return (size_t)(iend - anchor);
_offset: /* Requires: ip0, idx */
/* Compute the offset code. */
match0 = base + idx;
rep_offset2 = rep_offset1;
rep_offset1 = (U32)(ip0-match0);
offcode = rep_offset1 + ZSTD_REP_MOVE;
mLength = 4;
/* Count the backwards match length. */
while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
ip0--;
match0--;
mLength++;
}
_match: /* Requires: ip0, match0, offcode */
/* Count the forward length. */
mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);
ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength - MINMATCH);
ip0 += mLength;
anchor = ip0;
/* write next hash table entry */
if (ip1 < ip0) {
hashTable[hash1] = (U32)(ip1 - base);
}
/* Fill table and check for immediate repcode. */
if (ip0 <= ilimit) {
/* Fill Table */
assert(base+current0+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2; /* here because current+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
/* store sequence */
size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
{ U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
ip0 += rLength;
ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
anchor = ip0;
continue; /* faster when present (confirmed on gcc-8) ... (?) */
} } }
goto _start;
}
#define ZSTD_GEN_FAST_FN(dictMode, mls) \
static size_t ZSTD_compressBlock_fast_##dictMode##_##mls( \
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \
void const* src, size_t srcSize) \
{ \
return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
}
ZSTD_GEN_FAST_FN(noDict, 4)
ZSTD_GEN_FAST_FN(noDict, 5)
ZSTD_GEN_FAST_FN(noDict, 6)
ZSTD_GEN_FAST_FN(noDict, 7)
size_t ZSTD_compressBlock_fast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -193,13 +333,13 @@ size_t ZSTD_compressBlock_fast(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4);
return ZSTD_compressBlock_fast_noDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5);
return ZSTD_compressBlock_fast_noDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6);
return ZSTD_compressBlock_fast_noDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7);
return ZSTD_compressBlock_fast_noDict_7(ms, seqStore, rep, src, srcSize);
}
}
@@ -351,6 +491,12 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
return (size_t)(iend - anchor);
}
ZSTD_GEN_FAST_FN(dictMatchState, 4)
ZSTD_GEN_FAST_FN(dictMatchState, 5)
ZSTD_GEN_FAST_FN(dictMatchState, 6)
ZSTD_GEN_FAST_FN(dictMatchState, 7)
size_t ZSTD_compressBlock_fast_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
@@ -361,13 +507,13 @@ size_t ZSTD_compressBlock_fast_dictMatchState(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 4);
return ZSTD_compressBlock_fast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 5);
return ZSTD_compressBlock_fast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 6);
return ZSTD_compressBlock_fast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 7);
return ZSTD_compressBlock_fast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
}
}
@@ -402,7 +548,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
/* switch to "regular" variant if extDict is invalidated due to maxDistance */
if (prefixStartIndex == dictStartIndex)
return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, mls);
return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
/* Search Loop */
while (ip < ilimit) { /* < instead of <=, because (ip+1) */
@@ -418,7 +564,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */
& (offset_1 < curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
& (offset_1 <= curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
@@ -453,7 +599,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 < curr - dictStartIndex)) /* intentional overflow */
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 <= curr - dictStartIndex)) /* intentional overflow */
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
@@ -475,6 +621,10 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
return (size_t)(iend - anchor);
}
ZSTD_GEN_FAST_FN(extDict, 4)
ZSTD_GEN_FAST_FN(extDict, 5)
ZSTD_GEN_FAST_FN(extDict, 6)
ZSTD_GEN_FAST_FN(extDict, 7)
size_t ZSTD_compressBlock_fast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -485,12 +635,12 @@ size_t ZSTD_compressBlock_fast_extDict(
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
return ZSTD_compressBlock_fast_extDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
return ZSTD_compressBlock_fast_extDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
return ZSTD_compressBlock_fast_extDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
return ZSTD_compressBlock_fast_extDict_7(ms, seqStore, rep, src, srcSize);
}
}
lib/compress/zstd_lazy.c
+406 -487
View File
@@ -61,7 +61,7 @@ ZSTD_updateDUBT(ZSTD_matchState_t* ms,
* assumption : curr >= btlow == (curr - btmask)
* doesn't fail */
static void
ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
U32 curr, const BYTE* inputEnd,
U32 nbCompares, U32 btLow,
const ZSTD_dictMode_e dictMode)
@@ -93,7 +93,7 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
assert(curr >= btLow);
assert(ip < iend); /* condition for ZSTD_count */
while (nbCompares-- && (matchIndex > windowLow)) {
for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
assert(matchIndex < curr);
@@ -151,7 +151,7 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
static size_t
ZSTD_DUBT_findBetterDictMatch (
ZSTD_matchState_t* ms,
const ZSTD_matchState_t* ms,
const BYTE* const ip, const BYTE* const iend,
size_t* offsetPtr,
size_t bestLength,
@@ -185,7 +185,7 @@ ZSTD_DUBT_findBetterDictMatch (
(void)dictMode;
assert(dictMode == ZSTD_dictMatchState);
while (nbCompares-- && (dictMatchIndex > dictLowLimit)) {
for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) {
U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
const BYTE* match = dictBase + dictMatchIndex;
@@ -309,7 +309,7 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
matchIndex = hashTable[h];
hashTable[h] = curr; /* Update Hash Table */
while (nbCompares-- && (matchIndex > windowLow)) {
for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
const BYTE* match;
@@ -357,6 +357,7 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
*smallerPtr = *largerPtr = 0;
assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
if (dictMode == ZSTD_dictMatchState && nbCompares) {
bestLength = ZSTD_DUBT_findBetterDictMatch(
ms, ip, iend,
@@ -390,54 +391,6 @@ ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offsetPtr, mls, dictMode);
}
static size_t
ZSTD_BtFindBestMatch_selectMLS ( ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
case 7 :
case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
}
}
static size_t ZSTD_BtFindBestMatch_dictMatchState_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
case 7 :
case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
}
}
static size_t ZSTD_BtFindBestMatch_extDict_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
case 7 :
case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
}
}
/***********************************
* Dedicated dict search
***********************************/
@@ -450,7 +403,7 @@ void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const B
U32* const chainTable = ms->chainTable;
U32 const chainSize = 1 << ms->cParams.chainLog;
U32 idx = ms->nextToUpdate;
U32 const minChain = chainSize < target ? target - chainSize : idx;
U32 const minChain = chainSize < target - idx ? target - chainSize : idx;
U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
U32 const cacheSize = bucketSize - 1;
U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
@@ -692,7 +645,7 @@ U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
/* inlining is important to hardwire a hot branch (template emulation) */
FORCE_INLINE_TEMPLATE
size_t ZSTD_HcFindBestMatch_generic (
size_t ZSTD_HcFindBestMatch(
ZSTD_matchState_t* ms,
const BYTE* const ip, const BYTE* const iLimit,
size_t* offsetPtr,
@@ -758,6 +711,7 @@ size_t ZSTD_HcFindBestMatch_generic (
matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
}
assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
if (dictMode == ZSTD_dedicatedDictSearch) {
ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms,
ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
@@ -797,310 +751,80 @@ size_t ZSTD_HcFindBestMatch_generic (
return ml;
}
FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
case 7 :
case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
}
}
static size_t ZSTD_HcFindBestMatch_dictMatchState_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
case 7 :
case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
}
}
static size_t ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dedicatedDictSearch);
case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dedicatedDictSearch);
case 7 :
case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dedicatedDictSearch);
}
}
FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
case 7 :
case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
}
}
/* *********************************
* (SIMD) Row-based matchfinder
***********************************/
/* Constants for row-based hash */
#define ZSTD_ROW_HASH_TAG_OFFSET 1 /* byte offset of hashes in the match state's tagTable from the beginning of a row */
#define ZSTD_ROW_HASH_TAG_BITS 8 /* nb bits to use for the tag */
#define ZSTD_ROW_HASH_TAG_OFFSET 16 /* byte offset of hashes in the match state's tagTable from the beginning of a row */
#define ZSTD_ROW_HASH_TAG_BITS 8 /* nb bits to use for the tag */
#define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
#define ZSTD_ROW_HASH_MAX_ENTRIES 64 /* absolute maximum number of entries per row, for all configurations */
#define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1)
typedef U32 ZSTD_VecMask; /* Clarifies when we are interacting with a U32 representing a mask of matches */
#if !defined(ZSTD_NO_INTRINSICS) && defined(__SSE2__) /* SIMD SSE version */
#include <emmintrin.h>
typedef __m128i ZSTD_Vec128;
/* Returns a 128-bit container with 128-bits from src */
static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
return _mm_loadu_si128((ZSTD_Vec128 const*)src);
}
/* Returns a ZSTD_Vec128 with the byte "val" packed 16 times */
static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
return _mm_set1_epi8((char)val);
}
/* Do byte-by-byte comparison result of x and y. Then collapse 128-bit resultant mask
* into a 32-bit mask that is the MSB of each byte.
* */
static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
return (ZSTD_VecMask)_mm_movemask_epi8(_mm_cmpeq_epi8(x, y));
}
typedef struct {
__m128i fst;
__m128i snd;
} ZSTD_Vec256;
static ZSTD_Vec256 ZSTD_Vec256_read(const void* const ptr) {
ZSTD_Vec256 v;
v.fst = ZSTD_Vec128_read(ptr);
v.snd = ZSTD_Vec128_read((ZSTD_Vec128 const*)ptr + 1);
return v;
}
static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
ZSTD_Vec256 v;
v.fst = ZSTD_Vec128_set8(val);
v.snd = ZSTD_Vec128_set8(val);
return v;
}
static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
ZSTD_VecMask fstMask;
ZSTD_VecMask sndMask;
fstMask = ZSTD_Vec128_cmpMask8(x.fst, y.fst);
sndMask = ZSTD_Vec128_cmpMask8(x.snd, y.snd);
return fstMask | (sndMask << 16);
}
#elif !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON) /* SIMD ARM NEON Version */
#include <arm_neon.h>
typedef uint8x16_t ZSTD_Vec128;
static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
return vld1q_u8((const BYTE* const)src);
}
static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
return vdupq_n_u8(val);
}
/* Mimics '_mm_movemask_epi8()' from SSE */
static U32 ZSTD_vmovmaskq_u8(ZSTD_Vec128 val) {
/* Shift out everything but the MSB bits in each byte */
uint16x8_t highBits = vreinterpretq_u16_u8(vshrq_n_u8(val, 7));
/* Merge the even lanes together with vsra (right shift and add) */
uint32x4_t paired16 = vreinterpretq_u32_u16(vsraq_n_u16(highBits, highBits, 7));
uint64x2_t paired32 = vreinterpretq_u64_u32(vsraq_n_u32(paired16, paired16, 14));
uint8x16_t paired64 = vreinterpretq_u8_u64(vsraq_n_u64(paired32, paired32, 28));
/* Extract the low 8 bits from each lane, merge */
return vgetq_lane_u8(paired64, 0) | ((U32)vgetq_lane_u8(paired64, 8) << 8);
}
static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
return (ZSTD_VecMask)ZSTD_vmovmaskq_u8(vceqq_u8(x, y));
}
typedef struct {
uint8x16_t fst;
uint8x16_t snd;
} ZSTD_Vec256;
static ZSTD_Vec256 ZSTD_Vec256_read(const void* const ptr) {
ZSTD_Vec256 v;
v.fst = ZSTD_Vec128_read(ptr);
v.snd = ZSTD_Vec128_read((ZSTD_Vec128 const*)ptr + 1);
return v;
}
static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
ZSTD_Vec256 v;
v.fst = ZSTD_Vec128_set8(val);
v.snd = ZSTD_Vec128_set8(val);
return v;
}
static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
ZSTD_VecMask fstMask;
ZSTD_VecMask sndMask;
fstMask = ZSTD_Vec128_cmpMask8(x.fst, y.fst);
sndMask = ZSTD_Vec128_cmpMask8(x.snd, y.snd);
return fstMask | (sndMask << 16);
}
#else /* Scalar fallback version */
#define VEC128_NB_SIZE_T (16 / sizeof(size_t))
typedef struct {
size_t vec[VEC128_NB_SIZE_T];
} ZSTD_Vec128;
static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
ZSTD_Vec128 ret;
ZSTD_memcpy(ret.vec, src, VEC128_NB_SIZE_T*sizeof(size_t));
return ret;
}
static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
ZSTD_Vec128 ret = { {0} };
int startBit = sizeof(size_t) * 8 - 8;
for (;startBit >= 0; startBit -= 8) {
unsigned j = 0;
for (;j < VEC128_NB_SIZE_T; ++j) {
ret.vec[j] |= ((size_t)val << startBit);
}
}
return ret;
}
/* Compare x to y, byte by byte, generating a "matches" bitfield */
static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
ZSTD_VecMask res = 0;
unsigned i = 0;
unsigned l = 0;
for (; i < VEC128_NB_SIZE_T; ++i) {
const size_t cmp1 = x.vec[i];
const size_t cmp2 = y.vec[i];
unsigned j = 0;
for (; j < sizeof(size_t); ++j, ++l) {
if (((cmp1 >> j*8) & 0xFF) == ((cmp2 >> j*8) & 0xFF)) {
res |= ((U32)1 << (j+i*sizeof(size_t)));
}
}
}
return res;
}
#define VEC256_NB_SIZE_T 2*VEC128_NB_SIZE_T
typedef struct {
size_t vec[VEC256_NB_SIZE_T];
} ZSTD_Vec256;
static ZSTD_Vec256 ZSTD_Vec256_read(const void* const src) {
ZSTD_Vec256 ret;
ZSTD_memcpy(ret.vec, src, VEC256_NB_SIZE_T*sizeof(size_t));
return ret;
}
static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
ZSTD_Vec256 ret = { {0} };
int startBit = sizeof(size_t) * 8 - 8;
for (;startBit >= 0; startBit -= 8) {
unsigned j = 0;
for (;j < VEC256_NB_SIZE_T; ++j) {
ret.vec[j] |= ((size_t)val << startBit);
}
}
return ret;
}
/* Compare x to y, byte by byte, generating a "matches" bitfield */
static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
ZSTD_VecMask res = 0;
unsigned i = 0;
unsigned l = 0;
for (; i < VEC256_NB_SIZE_T; ++i) {
const size_t cmp1 = x.vec[i];
const size_t cmp2 = y.vec[i];
unsigned j = 0;
for (; j < sizeof(size_t); ++j, ++l) {
if (((cmp1 >> j*8) & 0xFF) == ((cmp2 >> j*8) & 0xFF)) {
res |= ((U32)1 << (j+i*sizeof(size_t)));
}
}
}
return res;
}
#endif /* !defined(ZSTD_NO_INTRINSICS) && defined(__SSE2__) */
typedef U64 ZSTD_VecMask; /* Clarifies when we are interacting with a U64 representing a mask of matches */
/* ZSTD_VecMask_next():
* Starting from the LSB, returns the idx of the next non-zero bit.
* Basically counting the nb of trailing zeroes.
*/
static U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
return _BitScanForward(&r, val) ? (U32)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (U32)__builtin_ctz(val);
assert(val != 0);
# if defined(_MSC_VER) && defined(_WIN64)
if (val != 0) {
unsigned long r;
_BitScanForward64(&r, val);
return (U32)(r);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif (defined(__GNUC__) && ((__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))
if (sizeof(size_t) == 4) {
U32 mostSignificantWord = (U32)(val >> 32);
U32 leastSignificantWord = (U32)val;
if (leastSignificantWord == 0) {
return 32 + (U32)__builtin_ctz(mostSignificantWord);
} else {
return (U32)__builtin_ctz(leastSignificantWord);
}
} else {
return (U32)__builtin_ctzll(val);
}
# else
/* Software ctz version: http://graphics.stanford.edu/~seander/bithacks.html#ZerosOnRightMultLookup */
static const U32 multiplyDeBruijnBitPosition[32] =
{
0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
};
return multiplyDeBruijnBitPosition[((U32)((v & -(int)v) * 0x077CB531U)) >> 27];
/* Software ctz version: http://aggregate.org/MAGIC/#Trailing%20Zero%20Count
* and: https://stackoverflow.com/questions/2709430/count-number-of-bits-in-a-64-bit-long-big-integer
*/
val = ~val & (val - 1ULL); /* Lowest set bit mask */
val = val - ((val >> 1) & 0x5555555555555555);
val = (val & 0x3333333333333333ULL) + ((val >> 2) & 0x3333333333333333ULL);
return (U32)((((val + (val >> 4)) & 0xF0F0F0F0F0F0F0FULL) * 0x101010101010101ULL) >> 56);
# endif
}
/* ZSTD_VecMask_rotateRight():
* Rotates a bitfield to the right by "rotation" bits.
* If the rotation is greater than totalBits, the returned mask is 0.
/* ZSTD_rotateRight_*():
* Rotates a bitfield to the right by "count" bits.
* https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
*/
FORCE_INLINE_TEMPLATE ZSTD_VecMask
ZSTD_VecMask_rotateRight(ZSTD_VecMask mask, U32 const rotation, U32 const totalBits) {
if (rotation == 0)
return mask;
switch (totalBits) {
default:
assert(0);
case 16:
return (mask >> rotation) | (U16)(mask << (16 - rotation));
case 32:
return (mask >> rotation) | (U32)(mask << (32 - rotation));
}
FORCE_INLINE_TEMPLATE
U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
assert(count < 64);
count &= 0x3F; /* for fickle pattern recognition */
return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
}
FORCE_INLINE_TEMPLATE
U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
assert(count < 32);
count &= 0x1F; /* for fickle pattern recognition */
return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
}
FORCE_INLINE_TEMPLATE
U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
assert(count < 16);
count &= 0x0F; /* for fickle pattern recognition */
return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
}
/* ZSTD_row_nextIndex():
@@ -1126,20 +850,24 @@ MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
*/
FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, U16 const* tagTable, U32 const relRow, U32 const rowLog) {
PREFETCH_L1(hashTable + relRow);
if (rowLog == 5) {
if (rowLog >= 5) {
PREFETCH_L1(hashTable + relRow + 16);
/* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */
}
PREFETCH_L1(tagTable + relRow);
assert(rowLog == 4 || rowLog == 5);
if (rowLog == 6) {
PREFETCH_L1(tagTable + relRow + 32);
}
assert(rowLog == 4 || rowLog == 5 || rowLog == 6);
assert(ZSTD_isAligned(hashTable + relRow, 64)); /* prefetched hash row always 64-byte aligned */
assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on a multiple of 32 or 64 bytes */
assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */
}
/* ZSTD_row_fillHashCache():
* Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
* but not beyond iLimit.
*/
static void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
FORCE_INLINE_TEMPLATE void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
U32 const rowLog, U32 const mls,
U32 idx, const BYTE* const iLimit)
{
@@ -1179,35 +907,65 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTab
}
}
/* ZSTD_row_update_internal():
* Inserts the byte at ip into the appropriate position in the hash table.
* Determines the relative row, and the position within the {16, 32} entry row to insert at.
/* ZSTD_row_update_internalImpl():
* Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
*/
FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
U32 const mls, U32 const rowLog,
U32 const rowMask, U32 const useCache)
FORCE_INLINE_TEMPLATE void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
U32 updateStartIdx, U32 const updateEndIdx,
U32 const mls, U32 const rowLog,
U32 const rowMask, U32 const useCache)
{
U32* const hashTable = ms->hashTable;
U16* const tagTable = ms->tagTable;
U32 const hashLog = ms->rowHashLog;
const BYTE* const base = ms->window.base;
const U32 target = (U32)(ip - base);
U32 idx = ms->nextToUpdate;
DEBUGLOG(6, "ZSTD_row_update_internal(): nextToUpdate=%u, current=%u", idx, target);
for (; idx < target; ++idx) {
U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, idx, hashLog, rowLog, mls)
: (U32)ZSTD_hashPtr(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls)
: (U32)ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
U32* const row = hashTable + relRow;
BYTE* tagRow = (BYTE*)(tagTable + relRow); /* Though tagTable is laid out as a table of U16, each tag is only 1 byte.
Explicit cast allows us to get exact desired position within each row */
U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
assert(hash == ZSTD_hashPtr(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls));
assert(hash == ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls));
((BYTE*)tagRow)[pos + ZSTD_ROW_HASH_TAG_OFFSET] = hash & ZSTD_ROW_HASH_TAG_MASK;
row[pos] = idx;
row[pos] = updateStartIdx;
}
}
/* ZSTD_row_update_internal():
* Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
* Skips sections of long matches as is necessary.
*/
FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
U32 const mls, U32 const rowLog,
U32 const rowMask, U32 const useCache)
{
U32 idx = ms->nextToUpdate;
const BYTE* const base = ms->window.base;
const U32 target = (U32)(ip - base);
const U32 kSkipThreshold = 384;
const U32 kMaxMatchStartPositionsToUpdate = 96;
const U32 kMaxMatchEndPositionsToUpdate = 32;
if (useCache) {
/* Only skip positions when using hash cache, i.e.
* if we are loading a dict, don't skip anything.
* If we decide to skip, then we only update a set number
* of positions at the beginning and end of the match.
*/
if (UNLIKELY(target - idx > kSkipThreshold)) {
U32 const bound = idx + kMaxMatchStartPositionsToUpdate;
ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache);
idx = target - kMaxMatchEndPositionsToUpdate;
ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1);
}
}
assert(target >= idx);
ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache);
ms->nextToUpdate = target;
}
@@ -1216,7 +974,7 @@ FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const
* processing.
*/
void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip) {
const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
const U32 rowMask = (1u << rowLog) - 1;
const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
@@ -1226,24 +984,131 @@ void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip) {
/* Returns a ZSTD_VecMask (U32) that has the nth bit set to 1 if the newly-computed "tag" matches
* the hash at the nth position in a row of the tagTable.
*/
* Each row is a circular buffer beginning at the value of "head". So we must rotate the "matches" bitfield
* to match up with the actual layout of the entries within the hashTable */
FORCE_INLINE_TEMPLATE
ZSTD_VecMask ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 head, const U32 rowEntries) {
ZSTD_VecMask matches = 0;
const BYTE* const src = tagRow + ZSTD_ROW_HASH_TAG_OFFSET;
assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
#if defined(ZSTD_ARCH_X86_SSE2)
if (rowEntries == 16) {
ZSTD_Vec128 hashes = ZSTD_Vec128_read(tagRow + ZSTD_ROW_HASH_TAG_OFFSET);
ZSTD_Vec128 expandedTags = ZSTD_Vec128_set8(tag);
matches = ZSTD_Vec128_cmpMask8(hashes, expandedTags);
const __m128i chunk = _mm_loadu_si128((const __m128i*)(const void*)src);
const __m128i equalMask = _mm_cmpeq_epi8(chunk, _mm_set1_epi8(tag));
const U16 matches = (U16)_mm_movemask_epi8(equalMask);
return ZSTD_rotateRight_U16(matches, head);
} else if (rowEntries == 32) {
ZSTD_Vec256 hashes = ZSTD_Vec256_read(tagRow + ZSTD_ROW_HASH_TAG_OFFSET);
ZSTD_Vec256 expandedTags = ZSTD_Vec256_set8(tag);
matches = ZSTD_Vec256_cmpMask8(hashes, expandedTags);
} else {
assert(0);
const __m128i chunk0 = _mm_loadu_si128((const __m128i*)(const void*)&src[0]);
const __m128i chunk1 = _mm_loadu_si128((const __m128i*)(const void*)&src[16]);
const __m128i equalMask0 = _mm_cmpeq_epi8(chunk0, _mm_set1_epi8(tag));
const __m128i equalMask1 = _mm_cmpeq_epi8(chunk1, _mm_set1_epi8(tag));
const U32 lo = (U32)_mm_movemask_epi8(equalMask0);
const U32 hi = (U32)_mm_movemask_epi8(equalMask1);
return ZSTD_rotateRight_U32((hi << 16) | lo, head);
} else { /* rowEntries == 64 */
const __m128i chunk0 = _mm_loadu_si128((const __m128i*)(const void*)&src[0]);
const __m128i chunk1 = _mm_loadu_si128((const __m128i*)(const void*)&src[16]);
const __m128i chunk2 = _mm_loadu_si128((const __m128i*)(const void*)&src[32]);
const __m128i chunk3 = _mm_loadu_si128((const __m128i*)(const void*)&src[48]);
const __m128i comparisonMask = _mm_set1_epi8(tag);
const __m128i equalMask0 = _mm_cmpeq_epi8(chunk0, comparisonMask);
const __m128i equalMask1 = _mm_cmpeq_epi8(chunk1, comparisonMask);
const __m128i equalMask2 = _mm_cmpeq_epi8(chunk2, comparisonMask);
const __m128i equalMask3 = _mm_cmpeq_epi8(chunk3, comparisonMask);
const U64 mask0 = (U64)_mm_movemask_epi8(equalMask0);
const U64 mask1 = (U64)_mm_movemask_epi8(equalMask1);
const U64 mask2 = (U64)_mm_movemask_epi8(equalMask2);
const U64 mask3 = (U64)_mm_movemask_epi8(equalMask3);
return ZSTD_rotateRight_U64((mask3 << 48) | (mask2 << 32) | (mask1 << 16) | mask0, head);
}
/* Each row is a circular buffer beginning at the value of "head". So we must rotate the "matches" bitfield
to match up with the actual layout of the entries within the hashTable */
return ZSTD_VecMask_rotateRight(matches, head, rowEntries);
#else
# if defined(ZSTD_ARCH_ARM_NEON)
if (MEM_isLittleEndian()) {
if (rowEntries == 16) {
const uint8x16_t chunk = vld1q_u8(src);
const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
const uint16x8_t t0 = vshlq_n_u16(equalMask, 7);
const uint32x4_t t1 = vreinterpretq_u32_u16(vsriq_n_u16(t0, t0, 14));
const uint64x2_t t2 = vreinterpretq_u64_u32(vshrq_n_u32(t1, 14));
const uint8x16_t t3 = vreinterpretq_u8_u64(vsraq_n_u64(t2, t2, 28));
const U16 hi = (U16)vgetq_lane_u8(t3, 8);
const U16 lo = (U16)vgetq_lane_u8(t3, 0);
return ZSTD_rotateRight_U16((hi << 8) | lo, head);
} else if (rowEntries == 32) {
const uint16x8x2_t chunk = vld2q_u16((const U16*)(const void*)src);
const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
const uint8x16_t equalMask0 = vceqq_u8(chunk0, vdupq_n_u8(tag));
const uint8x16_t equalMask1 = vceqq_u8(chunk1, vdupq_n_u8(tag));
const int8x8_t pack0 = vqmovn_s16(vreinterpretq_s16_u8(equalMask0));
const int8x8_t pack1 = vqmovn_s16(vreinterpretq_s16_u8(equalMask1));
const uint8x8_t t0 = vreinterpret_u8_s8(pack0);
const uint8x8_t t1 = vreinterpret_u8_s8(pack1);
const uint8x8_t t2 = vsri_n_u8(t1, t0, 2);
const uint8x8x2_t t3 = vuzp_u8(t2, t0);
const uint8x8_t t4 = vsri_n_u8(t3.val[1], t3.val[0], 4);
const U32 matches = vget_lane_u32(vreinterpret_u32_u8(t4), 0);
return ZSTD_rotateRight_U32(matches, head);
} else { /* rowEntries == 64 */
const uint8x16x4_t chunk = vld4q_u8(src);
const uint8x16_t dup = vdupq_n_u8(tag);
const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
return ZSTD_rotateRight_U64(matches, head);
}
}
# endif
{ /* SWAR */
const size_t chunkSize = sizeof(size_t);
const size_t shiftAmount = ((chunkSize * 8) - chunkSize);
const size_t xFF = ~((size_t)0);
const size_t x01 = xFF / 0xFF;
const size_t x80 = x01 << 7;
const size_t splatChar = tag * x01;
ZSTD_VecMask matches = 0;
int i = rowEntries - chunkSize;
assert((sizeof(size_t) == 4) || (sizeof(size_t) == 8));
if (MEM_isLittleEndian()) { /* runtime check so have two loops */
const size_t extractMagic = (xFF / 0x7F) >> chunkSize;
do {
size_t chunk = MEM_readST(&src[i]);
chunk ^= splatChar;
chunk = (((chunk | x80) - x01) | chunk) & x80;
matches <<= chunkSize;
matches |= (chunk * extractMagic) >> shiftAmount;
i -= chunkSize;
} while (i >= 0);
} else { /* big endian: reverse bits during extraction */
const size_t msb = xFF ^ (xFF >> 1);
const size_t extractMagic = (msb / 0x1FF) | msb;
do {
size_t chunk = MEM_readST(&src[i]);
chunk ^= splatChar;
chunk = (((chunk | x80) - x01) | chunk) & x80;
matches <<= chunkSize;
matches |= ((chunk >> 7) * extractMagic) >> shiftAmount;
i -= chunkSize;
} while (i >= 0);
}
matches = ~matches;
if (rowEntries == 16) {
return ZSTD_rotateRight_U16((U16)matches, head);
} else if (rowEntries == 32) {
return ZSTD_rotateRight_U32((U32)matches, head);
} else {
return ZSTD_rotateRight_U64((U64)matches, head);
}
}
#endif
}
/* The high-level approach of the SIMD row based match finder is as follows:
@@ -1262,7 +1127,7 @@ ZSTD_VecMask ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, con
* - Pick the longest match.
*/
FORCE_INLINE_TEMPLATE
size_t ZSTD_RowFindBestMatch_generic (
size_t ZSTD_RowFindBestMatch(
ZSTD_matchState_t* ms,
const BYTE* const ip, const BYTE* const iLimit,
size_t* offsetPtr,
@@ -1293,11 +1158,13 @@ size_t ZSTD_RowFindBestMatch_generic (
/* DMS/DDS variables that may be referenced laster */
const ZSTD_matchState_t* const dms = ms->dictMatchState;
size_t ddsIdx;
U32 ddsExtraAttempts; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
U32 dmsTag;
U32* dmsRow;
BYTE* dmsTagRow;
/* Initialize the following variables to satisfy static analyzer */
size_t ddsIdx = 0;
U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
U32 dmsTag = 0;
U32* dmsRow = NULL;
BYTE* dmsTagRow = NULL;
if (dictMode == ZSTD_dedicatedDictSearch) {
const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
@@ -1329,7 +1196,7 @@ size_t ZSTD_RowFindBestMatch_generic (
U32* const row = hashTable + relRow;
BYTE* tagRow = (BYTE*)(tagTable + relRow);
U32 const head = *tagRow & rowMask;
U32 matchBuffer[32 /* maximum nb entries per row */];
U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
size_t numMatches = 0;
size_t currMatch = 0;
ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, head, rowEntries);
@@ -1385,6 +1252,7 @@ size_t ZSTD_RowFindBestMatch_generic (
}
}
assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
if (dictMode == ZSTD_dedicatedDictSearch) {
ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms,
ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
@@ -1397,7 +1265,7 @@ size_t ZSTD_RowFindBestMatch_generic (
const U32 dmsIndexDelta = dictLimit - dmsSize;
{ U32 const head = *dmsTagRow & rowMask;
U32 matchBuffer[32 /* maximum nb row entries */];
U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
size_t numMatches = 0;
size_t currMatch = 0;
ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, head, rowEntries);
@@ -1435,84 +1303,168 @@ size_t ZSTD_RowFindBestMatch_generic (
return ml;
}
/* Inlining is important to hardwire a hot branch (template emulation) */
FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_selectMLS (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
const ZSTD_dictMode_e dictMode, size_t* offsetPtr, const U32 rowLog)
{
switch(ms->cParams.minMatch)
{
default : /* includes case 3 */
case 4 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, dictMode, rowLog);
case 5 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, dictMode, rowLog);
case 7 :
case 6 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, dictMode, rowLog);
}
}
FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_selectRowLog (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
switch(cappedSearchLog)
{
default :
case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_noDict, offsetPtr, 4);
case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_noDict, offsetPtr, 5);
}
}
typedef size_t (*searchMax_f)(
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_dictMatchState_selectRowLog(
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
switch(cappedSearchLog)
{
default :
case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dictMatchState, offsetPtr, 4);
case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dictMatchState, offsetPtr, 5);
}
}
/**
* This struct contains the functions necessary for lazy to search.
* Currently, that is only searchMax. However, it is still valuable to have the
* VTable because this makes it easier to add more functions to the VTable later.
*
* TODO: The start of the search function involves loading and calculating a
* bunch of constants from the ZSTD_matchState_t. These computations could be
* done in an initialization function, and saved somewhere in the match state.
* Then we could pass a pointer to the saved state instead of the match state,
* and avoid duplicate computations.
*
* TODO: Move the match re-winding into searchMax. This improves compression
* ratio, and unlocks further simplifications with the next TODO.
*
* TODO: Try moving the repcode search into searchMax. After the re-winding
* and repcode search are in searchMax, there is no more logic in the match
* finder loop that requires knowledge about the dictMode. So we should be
* able to avoid force inlining it, and we can join the extDict loop with
* the single segment loop. It should go in searchMax instead of its own
* function to avoid having multiple virtual function calls per search.
*/
typedef struct {
searchMax_f searchMax;
} ZSTD_LazyVTable;
FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_dedicatedDictSearch_selectRowLog(
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
switch(cappedSearchLog)
{
default :
case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dedicatedDictSearch, offsetPtr, 4);
case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dedicatedDictSearch, offsetPtr, 5);
}
}
#define GEN_ZSTD_BT_VTABLE(dictMode, mls, ...) \
static size_t ZSTD_BtFindBestMatch_##dictMode##_##mls( \
ZSTD_matchState_t* ms, \
const BYTE* ip, const BYTE* const iLimit, \
size_t* offsetPtr) \
{ \
assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \
return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
} \
static const ZSTD_LazyVTable ZSTD_BtVTable_##dictMode##_##mls = { \
ZSTD_BtFindBestMatch_##dictMode##_##mls \
};
FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_extDict_selectRowLog (
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* const iLimit,
size_t* offsetPtr)
{
const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
switch(cappedSearchLog)
{
default :
case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_extDict, offsetPtr, 4);
case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_extDict, offsetPtr, 5);
}
}
#define GEN_ZSTD_HC_VTABLE(dictMode, mls, ...) \
static size_t ZSTD_HcFindBestMatch_##dictMode##_##mls( \
ZSTD_matchState_t* ms, \
const BYTE* ip, const BYTE* const iLimit, \
size_t* offsetPtr) \
{ \
assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \
return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
} \
static const ZSTD_LazyVTable ZSTD_HcVTable_##dictMode##_##mls = { \
ZSTD_HcFindBestMatch_##dictMode##_##mls \
};
#define GEN_ZSTD_ROW_VTABLE(dictMode, mls, rowLog) \
static size_t ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog( \
ZSTD_matchState_t* ms, \
const BYTE* ip, const BYTE* const iLimit, \
size_t* offsetPtr) \
{ \
assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \
assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog); \
return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \
} \
static const ZSTD_LazyVTable ZSTD_RowVTable_##dictMode##_##mls##_##rowLog = { \
ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog \
};
#define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \
X(dictMode, mls, 4) \
X(dictMode, mls, 5) \
X(dictMode, mls, 6)
#define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \
ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4) \
ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5) \
ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6)
#define ZSTD_FOR_EACH_MLS(X, dictMode) \
X(dictMode, 4) \
X(dictMode, 5) \
X(dictMode, 6)
#define ZSTD_FOR_EACH_DICT_MODE(X, ...) \
X(__VA_ARGS__, noDict) \
X(__VA_ARGS__, extDict) \
X(__VA_ARGS__, dictMatchState) \
X(__VA_ARGS__, dedicatedDictSearch)
/* Generate Row VTables for each combination of (dictMode, mls, rowLog) */
ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_VTABLE)
/* Generate Binary Tree VTables for each combination of (dictMode, mls) */
ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_VTABLE)
/* Generate Hash Chain VTables for each combination of (dictMode, mls) */
ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_VTABLE)
#define GEN_ZSTD_BT_VTABLE_ARRAY(dictMode) \
{ \
&ZSTD_BtVTable_##dictMode##_4, \
&ZSTD_BtVTable_##dictMode##_5, \
&ZSTD_BtVTable_##dictMode##_6 \
}
#define GEN_ZSTD_HC_VTABLE_ARRAY(dictMode) \
{ \
&ZSTD_HcVTable_##dictMode##_4, \
&ZSTD_HcVTable_##dictMode##_5, \
&ZSTD_HcVTable_##dictMode##_6 \
}
#define GEN_ZSTD_ROW_VTABLE_ARRAY_(dictMode, mls) \
{ \
&ZSTD_RowVTable_##dictMode##_##mls##_4, \
&ZSTD_RowVTable_##dictMode##_##mls##_5, \
&ZSTD_RowVTable_##dictMode##_##mls##_6 \
}
#define GEN_ZSTD_ROW_VTABLE_ARRAY(dictMode) \
{ \
GEN_ZSTD_ROW_VTABLE_ARRAY_(dictMode, 4), \
GEN_ZSTD_ROW_VTABLE_ARRAY_(dictMode, 5), \
GEN_ZSTD_ROW_VTABLE_ARRAY_(dictMode, 6) \
}
#define GEN_ZSTD_VTABLE_ARRAY(X) \
{ \
X(noDict), \
X(extDict), \
X(dictMatchState), \
X(dedicatedDictSearch) \
}
/* *******************************
* Common parser - lazy strategy
*********************************/
typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
static ZSTD_LazyVTable const* ZSTD_selectLazyVTable(ZSTD_matchState_t const* ms, searchMethod_e searchMethod, ZSTD_dictMode_e dictMode)
{
/* Fill the Hc/Bt VTable arrays with the right functions for the (dictMode, mls) combination. */
ZSTD_LazyVTable const* const hcVTables[4][3] = GEN_ZSTD_VTABLE_ARRAY(GEN_ZSTD_HC_VTABLE_ARRAY);
ZSTD_LazyVTable const* const btVTables[4][3] = GEN_ZSTD_VTABLE_ARRAY(GEN_ZSTD_BT_VTABLE_ARRAY);
/* Fill the Row VTable array with the right functions for the (dictMode, mls, rowLog) combination. */
ZSTD_LazyVTable const* const rowVTables[4][3][3] = GEN_ZSTD_VTABLE_ARRAY(GEN_ZSTD_ROW_VTABLE_ARRAY);
U32 const mls = MAX(4, MIN(6, ms->cParams.minMatch));
U32 const rowLog = MAX(4, MIN(6, ms->cParams.searchLog));
switch (searchMethod) {
case search_hashChain:
return hcVTables[dictMode][mls - 4];
case search_binaryTree:
return btVTables[dictMode][mls - 4];
case search_rowHash:
return rowVTables[dictMode][mls - 4][rowLog - 4];
default:
return NULL;
}
}
FORCE_INLINE_TEMPLATE size_t
ZSTD_compressBlock_lazy_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore,
@@ -1531,9 +1483,6 @@ ZSTD_compressBlock_lazy_generic(
const BYTE* const prefixLowest = base + prefixLowestIndex;
const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
typedef size_t (*searchMax_f)(
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
/**
* This table is indexed first by the four ZSTD_dictMode_e values, and then
@@ -1541,30 +1490,8 @@ ZSTD_compressBlock_lazy_generic(
* that should never occur (extDict modes go to the other implementation
* below and there is no DDSS for binary tree search yet).
*/
const searchMax_f searchFuncs[4][3] = {
{
ZSTD_HcFindBestMatch_selectMLS,
ZSTD_BtFindBestMatch_selectMLS,
ZSTD_RowFindBestMatch_selectRowLog
},
{
NULL,
NULL,
NULL
},
{
ZSTD_HcFindBestMatch_dictMatchState_selectMLS,
ZSTD_BtFindBestMatch_dictMatchState_selectMLS,
ZSTD_RowFindBestMatch_dictMatchState_selectRowLog
},
{
ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS,
NULL,
ZSTD_RowFindBestMatch_dedicatedDictSearch_selectRowLog
}
};
searchMax_f const searchMax = searchFuncs[dictMode][(int)searchMethod];
searchMax_f const searchMax = ZSTD_selectLazyVTable(ms, searchMethod, dictMode)->searchMax;
U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
const int isDMS = dictMode == ZSTD_dictMatchState;
@@ -1955,15 +1882,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
const U32 windowLog = ms->cParams.windowLog;
const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
typedef size_t (*searchMax_f)(
ZSTD_matchState_t* ms,
const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
const searchMax_f searchFuncs[3] = {
ZSTD_HcFindBestMatch_extDict_selectMLS,
ZSTD_BtFindBestMatch_extDict_selectMLS,
ZSTD_RowFindBestMatch_extDict_selectRowLog
};
searchMax_f searchMax = searchFuncs[(int)searchMethod];
searchMax_f const searchMax = ZSTD_selectLazyVTable(ms, searchMethod, ZSTD_extDict)->searchMax;
U32 offset_1 = rep[0], offset_2 = rep[1];
DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
@@ -1995,7 +1914,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */
& (offset_1 < curr+1 - windowLow) ) /* note: we are searching at curr+1 */
& (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */
if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
/* repcode detected we should take it */
const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -2010,7 +1929,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
matchLength = ml2, start = ip, offset=offsetFound;
}
if (matchLength < 4) {
if (matchLength < 4) {
ip += ((ip-anchor) >> kSearchStrength) + 1; /* jump faster over incompressible sections */
continue;
}
@@ -2027,7 +1946,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments */
& (offset_1 < curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
& (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
if (MEM_read32(ip) == MEM_read32(repMatch)) {
/* repcode detected */
const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -2059,7 +1978,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments */
& (offset_1 < curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
& (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
if (MEM_read32(ip) == MEM_read32(repMatch)) {
/* repcode detected */
const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -2106,7 +2025,7 @@ _storeSequence:
const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments */
& (offset_2 < repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
& (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
if (MEM_read32(ip) == MEM_read32(repMatch)) {
/* repcode detected we should take it */
const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
lib/compress/zstd_ldm.c
+4 -4
View File
@@ -159,12 +159,12 @@ size_t ZSTD_ldm_getTableSize(ldmParams_t params)
size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+ ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
return params.enableLdm ? totalSize : 0;
return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
}
size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
{
return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
}
/** ZSTD_ldm_getBucket() :
@@ -478,7 +478,7 @@ static size_t ZSTD_ldm_generateSequences_internal(
*/
if (anchor > ip + hashed) {
ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
/* Continue the outter loop at anchor (ip + hashed == anchor). */
/* Continue the outer loop at anchor (ip + hashed == anchor). */
ip = anchor - hashed;
break;
}
@@ -657,7 +657,7 @@ void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
ZSTD_paramSwitch_e useRowMatchFinder,
void const* src, size_t srcSize)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
lib/compress/zstd_ldm.h
+1 -1
View File
@@ -66,7 +66,7 @@ size_t ZSTD_ldm_generateSequences(
*/
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
ZSTD_paramSwitch_e useRowMatchFinder,
void const* src, size_t srcSize);
/**
lib/compress/zstd_ldm_geartab.h
+4 -1
View File
@@ -11,7 +11,10 @@
#ifndef ZSTD_LDM_GEARTAB_H
#define ZSTD_LDM_GEARTAB_H
static U64 ZSTD_ldm_gearTab[256] = {
#include "../common/compiler.h" /* UNUSED_ATTR */
#include "../common/mem.h" /* U64 */
static UNUSED_ATTR const U64 ZSTD_ldm_gearTab[256] = {
0xf5b8f72c5f77775c, 0x84935f266b7ac412, 0xb647ada9ca730ccc,
0xb065bb4b114fb1de, 0x34584e7e8c3a9fd0, 0x4e97e17c6ae26b05,
0x3a03d743bc99a604, 0xcecd042422c4044f, 0x76de76c58524259e,
lib/compress/zstd_opt.c
+100 -76
View File
@@ -8,13 +8,26 @@
* You may select, at your option, one of the above-listed licenses.
*/
/**
* Disable inlining for the optimal parser for the kernel build.
* It is unlikely to be used in the kernel, and where it is used
* latency shouldn't matter because it is very slow to begin with.
* We prefer a ~180KB binary size win over faster optimal parsing.
*
* TODO(https://github.com/facebook/zstd/issues/2862):
* Improve the code size of the optimal parser in general, so we
* don't need this hack for the kernel build.
*/
#ifdef ZSTD_LINUX_KERNEL
#define ZSTD_NO_INLINE 1
#endif
#include "zstd_compress_internal.h"
#include "hist.h"
#include "zstd_opt.h"
#define ZSTD_LITFREQ_ADD 2 /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
#define ZSTD_FREQ_DIV 4 /* log factor when using previous stats to init next stats */
#define ZSTD_MAX_PRICE (1<<30)
#define ZSTD_PREDEF_THRESHOLD 1024 /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
@@ -24,11 +37,11 @@
* Price functions for optimal parser
***************************************/
#if 0 /* approximation at bit level */
#if 0 /* approximation at bit level (for tests) */
# define BITCOST_ACCURACY 0
# define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
# define WEIGHT(stat) ((void)opt, ZSTD_bitWeight(stat))
#elif 0 /* fractional bit accuracy */
# define WEIGHT(stat, opt) ((void)opt, ZSTD_bitWeight(stat))
#elif 0 /* fractional bit accuracy (for tests) */
# define BITCOST_ACCURACY 8
# define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
# define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))
@@ -66,7 +79,7 @@ MEM_STATIC double ZSTD_fCost(U32 price)
static int ZSTD_compressedLiterals(optState_t const* const optPtr)
{
return optPtr->literalCompressionMode != ZSTD_lcm_uncompressed;
return optPtr->literalCompressionMode != ZSTD_ps_disable;
}
static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
@@ -79,25 +92,46 @@ static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
}
/* ZSTD_downscaleStat() :
* reduce all elements in table by a factor 2^(ZSTD_FREQ_DIV+malus)
* return the resulting sum of elements */
static U32 ZSTD_downscaleStat(unsigned* table, U32 lastEltIndex, int malus)
static U32 sum_u32(const unsigned table[], size_t nbElts)
{
size_t n;
U32 total = 0;
for (n=0; n<nbElts; n++) {
total += table[n];
}
return total;
}
static U32 ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift)
{
U32 s, sum=0;
DEBUGLOG(5, "ZSTD_downscaleStat (nbElts=%u)", (unsigned)lastEltIndex+1);
assert(ZSTD_FREQ_DIV+malus > 0 && ZSTD_FREQ_DIV+malus < 31);
DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)", (unsigned)lastEltIndex+1, (unsigned)shift);
assert(shift < 30);
for (s=0; s<lastEltIndex+1; s++) {
table[s] = 1 + (table[s] >> (ZSTD_FREQ_DIV+malus));
table[s] = 1 + (table[s] >> shift);
sum += table[s];
}
return sum;
}
/* ZSTD_scaleStats() :
* reduce all elements in table is sum too large
* return the resulting sum of elements */
static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
{
U32 const prevsum = sum_u32(table, lastEltIndex+1);
U32 const factor = prevsum >> logTarget;
DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
assert(logTarget < 30);
if (factor <= 1) return prevsum;
return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor));
}
/* ZSTD_rescaleFreqs() :
* if first block (detected by optPtr->litLengthSum == 0) : init statistics
* take hints from dictionary if there is one
* or init from zero, using src for literals stats, or flat 1 for match symbols
* and init from zero if there is none,
* using src for literals stats, and baseline stats for sequence symbols
* otherwise downscale existing stats, to be used as seed for next block.
*/
static void
@@ -126,7 +160,7 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
optPtr->litSum = 0;
for (lit=0; lit<=MaxLit; lit++) {
U32 const scaleLog = 11; /* scale to 2K */
U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);
U32 const bitCost = HUF_getNbBitsFromCTable(optPtr->symbolCosts->huf.CTable, lit);
assert(bitCost <= scaleLog);
optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
optPtr->litSum += optPtr->litFreq[lit];
@@ -174,14 +208,18 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
if (compressedLiterals) {
unsigned lit = MaxLit;
HIST_count_simple(optPtr->litFreq, &lit, src, srcSize); /* use raw first block to init statistics */
optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8);
}
{ unsigned ll;
for (ll=0; ll<=MaxLL; ll++)
optPtr->litLengthFreq[ll] = 1;
{ unsigned const baseLLfreqs[MaxLL+1] = {
4, 2, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1
};
ZSTD_memcpy(optPtr->litLengthFreq, baseLLfreqs, sizeof(baseLLfreqs)); optPtr->litLengthSum = sum_u32(baseLLfreqs, MaxLL+1);
}
optPtr->litLengthSum = MaxLL+1;
{ unsigned ml;
for (ml=0; ml<=MaxML; ml++)
@@ -189,21 +227,25 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
}
optPtr->matchLengthSum = MaxML+1;
{ unsigned of;
for (of=0; of<=MaxOff; of++)
optPtr->offCodeFreq[of] = 1;
{ unsigned const baseOFCfreqs[MaxOff+1] = {
6, 2, 1, 1, 2, 3, 4, 4,
4, 3, 2, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1
};
ZSTD_memcpy(optPtr->offCodeFreq, baseOFCfreqs, sizeof(baseOFCfreqs)); optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
}
optPtr->offCodeSum = MaxOff+1;
}
} else { /* new block : re-use previous statistics, scaled down */
if (compressedLiterals)
optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);
optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);
optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);
optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 11);
optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 11);
optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 11);
}
ZSTD_setBasePrices(optPtr, optLevel);
@@ -338,7 +380,7 @@ MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
/* Update hashTable3 up to ip (excluded)
Assumption : always within prefix (i.e. not within extDict) */
static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
static U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
U32* nextToUpdate3,
const BYTE* const ip)
{
@@ -364,11 +406,13 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
* Binary Tree search
***************************************/
/** ZSTD_insertBt1() : add one or multiple positions to tree.
* ip : assumed <= iend-8 .
* @param ip assumed <= iend-8 .
* @param target The target of ZSTD_updateTree_internal() - we are filling to this position
* @return : nb of positions added */
static U32 ZSTD_insertBt1(
ZSTD_matchState_t* ms,
const ZSTD_matchState_t* ms,
const BYTE* const ip, const BYTE* const iend,
U32 const target,
U32 const mls, const int extDict)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -391,7 +435,10 @@ static U32 ZSTD_insertBt1(
U32* smallerPtr = bt + 2*(curr&btMask);
U32* largerPtr = smallerPtr + 1;
U32 dummy32; /* to be nullified at the end */
U32 const windowLow = ms->window.lowLimit;
/* windowLow is based on target because
* we only need positions that will be in the window at the end of the tree update.
*/
U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
U32 matchEndIdx = curr+8+1;
size_t bestLength = 8;
U32 nbCompares = 1U << cParams->searchLog;
@@ -404,11 +451,12 @@ static U32 ZSTD_insertBt1(
DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
assert(curr <= target);
assert(ip <= iend-8); /* required for h calculation */
hashTable[h] = curr; /* Update Hash Table */
assert(windowLow > 0);
while (nbCompares-- && (matchIndex >= windowLow)) {
for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
assert(matchIndex < curr);
@@ -492,7 +540,7 @@ void ZSTD_updateTree_internal(
idx, target, dictMode);
while(idx < target) {
U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
assert(idx < (U32)(idx + forward));
idx += forward;
}
@@ -635,11 +683,11 @@ U32 ZSTD_insertBtAndGetAllMatches (
return 1;
} } }
/* no dictMatchState lookup: dicts don't have a populated HC3 table */
}
} /* if (mls == 3) */
hashTable[h] = curr; /* Update Hash Table */
while (nbCompares-- && (matchIndex >= matchLow)) {
for (; nbCompares && (matchIndex >= matchLow); --nbCompares) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
const BYTE* match;
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
@@ -672,8 +720,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
| (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
break; /* drop, to preserve bt consistency (miss a little bit of compression) */
}
}
} }
if (match[matchLength] < ip[matchLength]) {
/* match smaller than current */
@@ -692,12 +739,13 @@ U32 ZSTD_insertBtAndGetAllMatches (
*smallerPtr = *largerPtr = 0;
assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
if (dictMode == ZSTD_dictMatchState && nbCompares) {
size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
U32 dictMatchIndex = dms->hashTable[dmsH];
const U32* const dmsBt = dms->chainTable;
commonLengthSmaller = commonLengthLarger = 0;
while (nbCompares-- && (dictMatchIndex > dmsLowLimit)) {
for (; nbCompares && (dictMatchIndex > dmsLowLimit); --nbCompares) {
const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
const BYTE* match = dmsBase + dictMatchIndex;
@@ -718,8 +766,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
if ( (matchLength > ZSTD_OPT_NUM)
| (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
break; /* drop, to guarantee consistency (miss a little bit of compression) */
}
}
} }
if (dictMatchIndex <= dmsBtLow) { break; } /* beyond tree size, stop the search */
if (match[matchLength] < ip[matchLength]) {
@@ -729,9 +776,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
/* match is larger than current */
commonLengthLarger = matchLength;
dictMatchIndex = nextPtr[0];
}
}
}
} } } /* if (dictMode == ZSTD_dictMatchState) */
assert(matchEndIdx > curr+8);
ms->nextToUpdate = matchEndIdx - 8; /* skip repetitive patterns */
@@ -893,17 +938,17 @@ static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_
*/
U32 posOvershoot = currPosInBlock - optLdm->endPosInBlock;
ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
}
}
ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
}
ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
}
/*-*******************************
* Optimal parser
*********************************/
static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
{
return sol.litlen + sol.mlen;
@@ -985,7 +1030,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
* in every price. We include the literal length to avoid negative
* prices when we subtract the previous literal length.
*/
opt[0].price = ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
opt[0].price = (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
/* large match -> immediate encoding */
{ U32 const maxML = matches[nbMatches-1].len;
@@ -1005,7 +1050,8 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
} }
/* set prices for first matches starting position == 0 */
{ U32 const literalsPrice = opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
assert(opt[0].price >= 0);
{ U32 const literalsPrice = (U32)opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
U32 pos;
U32 matchNb;
for (pos = 1; pos < minMatch; pos++) {
@@ -1022,7 +1068,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
opt[pos].mlen = pos;
opt[pos].off = offset;
opt[pos].litlen = litlen;
opt[pos].price = sequencePrice;
opt[pos].price = (int)sequencePrice;
} }
last_pos = pos-1;
}
@@ -1037,9 +1083,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
/* Fix current position with one literal if cheaper */
{ U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
int const price = opt[cur-1].price
+ ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+ ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
- ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
+ (int)ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+ (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
- (int)ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
assert(price < 1000000000); /* overflow check */
if (price <= opt[cur].price) {
DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
@@ -1082,9 +1128,10 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
continue; /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
}
assert(opt[cur].price >= 0);
{ U32 const ll0 = (opt[cur].mlen != 0);
U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
U32 const previousPrice = opt[cur].price;
U32 const previousPrice = (U32)opt[cur].price;
U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
U32 matchNb;
@@ -1124,7 +1171,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
for (mlen = lastML; mlen >= startML; mlen--) { /* scan downward */
U32 const pos = cur + mlen;
int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
int const price = (int)basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
if ((pos > last_pos) || (price < opt[pos].price)) {
DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
@@ -1220,28 +1267,7 @@ size_t ZSTD_compressBlock_btopt(
}
/* used in 2-pass strategy */
static U32 ZSTD_upscaleStat(unsigned* table, U32 lastEltIndex, int bonus)
{
U32 s, sum=0;
assert(ZSTD_FREQ_DIV+bonus >= 0);
for (s=0; s<lastEltIndex+1; s++) {
table[s] <<= ZSTD_FREQ_DIV+bonus;
table[s]--;
sum += table[s];
}
return sum;
}
/* used in 2-pass strategy */
MEM_STATIC void ZSTD_upscaleStats(optState_t* optPtr)
{
if (ZSTD_compressedLiterals(optPtr))
optPtr->litSum = ZSTD_upscaleStat(optPtr->litFreq, MaxLit, 0);
optPtr->litLengthSum = ZSTD_upscaleStat(optPtr->litLengthFreq, MaxLL, 0);
optPtr->matchLengthSum = ZSTD_upscaleStat(optPtr->matchLengthFreq, MaxML, 0);
optPtr->offCodeSum = ZSTD_upscaleStat(optPtr->offCodeFreq, MaxOff, 0);
}
/* ZSTD_initStats_ultra():
* make a first compression pass, just to seed stats with more accurate starting values.
@@ -1272,8 +1298,6 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
ms->window.lowLimit = ms->window.dictLimit;
ms->nextToUpdate = ms->window.dictLimit;
/* re-inforce weight of collected statistics */
ZSTD_upscaleStats(&ms->opt);
}
size_t ZSTD_compressBlock_btultra(
lib/compress/zstdmt_compress.c
+49 -23
View File
@@ -467,7 +467,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
ZSTD_dictContentType_e dictContentType)
{
/* Adjust parameters */
if (params.ldmParams.enableLdm) {
if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
@@ -478,7 +478,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
serialState->nextJobID = 0;
if (params.fParams.checksumFlag)
XXH64_reset(&serialState->xxhState, 0);
if (params.ldmParams.enableLdm) {
if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
ZSTD_customMem cMem = params.customMem;
unsigned const hashLog = params.ldmParams.hashLog;
size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
@@ -564,7 +564,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
/* A future job may error and skip our job */
if (serialState->nextJobID == jobID) {
/* It is now our turn, do any processing necessary */
if (serialState->params.ldmParams.enableLdm) {
if (serialState->params.ldmParams.enableLdm == ZSTD_ps_enable) {
size_t error;
assert(seqStore.seq != NULL && seqStore.pos == 0 &&
seqStore.size == 0 && seqStore.capacity > 0);
@@ -594,7 +594,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
if (seqStore.size > 0) {
size_t const err = ZSTD_referenceExternalSequences(
jobCCtx, seqStore.seq, seqStore.size);
assert(serialState->params.ldmParams.enableLdm);
assert(serialState->params.ldmParams.enableLdm == ZSTD_ps_enable);
assert(!ZSTD_isError(err));
(void)err;
}
@@ -672,7 +672,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
job->dstBuff = dstBuff; /* this value can be read in ZSTDMT_flush, when it copies the whole job */
}
if (jobParams.ldmParams.enableLdm && rawSeqStore.seq == NULL)
if (jobParams.ldmParams.enableLdm == ZSTD_ps_enable && rawSeqStore.seq == NULL)
JOB_ERROR(ERROR(memory_allocation));
/* Don't compute the checksum for chunks, since we compute it externally,
@@ -680,7 +680,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
*/
if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
/* Don't run LDM for the chunks, since we handle it externally */
jobParams.ldmParams.enableLdm = 0;
jobParams.ldmParams.enableLdm = ZSTD_ps_disable;
/* Correct nbWorkers to 0. */
jobParams.nbWorkers = 0;
@@ -807,6 +807,15 @@ typedef struct {
static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
#define RSYNC_LENGTH 32
/* Don't create chunks smaller than the zstd block size.
* This stops us from regressing compression ratio too much,
* and ensures our output fits in ZSTD_compressBound().
*
* If this is shrunk < ZSTD_BLOCKSIZELOG_MIN then
* ZSTD_COMPRESSBOUND() will need to be updated.
*/
#define RSYNC_MIN_BLOCK_LOG ZSTD_BLOCKSIZELOG_MAX
#define RSYNC_MIN_BLOCK_SIZE (1<<RSYNC_MIN_BLOCK_LOG)
typedef struct {
U64 hash;
@@ -1135,7 +1144,7 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
{
unsigned jobLog;
if (params->ldmParams.enableLdm) {
if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
/* In Long Range Mode, the windowLog is typically oversized.
* In which case, it's preferable to determine the jobSize
* based on cycleLog instead. */
@@ -1179,7 +1188,7 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
assert(0 <= overlapRLog && overlapRLog <= 8);
if (params->ldmParams.enableLdm) {
if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
/* In Long Range Mode, the windowLog is typically oversized.
* In which case, it's preferable to determine the jobSize
* based on chainLog instead.
@@ -1252,6 +1261,9 @@ size_t ZSTDMT_initCStream_internal(
/* Aim for the targetsectionSize as the average job size. */
U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
/* We refuse to create jobs < RSYNC_MIN_BLOCK_SIZE bytes, so make sure our
* expected job size is at least 4x larger. */
assert(rsyncBits >= RSYNC_MIN_BLOCK_LOG + 2);
DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
mtctx->rsync.hash = 0;
mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
@@ -1263,7 +1275,7 @@ size_t ZSTDMT_initCStream_internal(
ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
{
/* If ldm is enabled we need windowSize space. */
size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
size_t const windowSize = mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable ? (1U << mtctx->params.cParams.windowLog) : 0;
/* Two buffers of slack, plus extra space for the overlap
* This is the minimum slack that LDM works with. One extra because
* flush might waste up to targetSectionSize-1 bytes. Another extra
@@ -1575,7 +1587,7 @@ static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
{
if (mtctx->params.ldmParams.enableLdm) {
if (mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable) {
ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
DEBUGLOG(5, "source [0x%zx, 0x%zx)",
@@ -1678,6 +1690,11 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
if (!mtctx->params.rsyncable)
/* Rsync is disabled. */
return syncPoint;
if (mtctx->inBuff.filled + input.size - input.pos < RSYNC_MIN_BLOCK_SIZE)
/* We don't emit synchronization points if it would produce too small blocks.
* We don't have enough input to find a synchronization point, so don't look.
*/
return syncPoint;
if (mtctx->inBuff.filled + syncPoint.toLoad < RSYNC_LENGTH)
/* Not enough to compute the hash.
* We will miss any synchronization points in this RSYNC_LENGTH byte
@@ -1688,10 +1705,28 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
*/
return syncPoint;
/* Initialize the loop variables. */
if (mtctx->inBuff.filled >= RSYNC_LENGTH) {
/* We have enough bytes buffered to initialize the hash.
if (mtctx->inBuff.filled < RSYNC_MIN_BLOCK_SIZE) {
/* We don't need to scan the first RSYNC_MIN_BLOCK_SIZE positions
* because they can't possibly be a sync point. So we can start
* part way through the input buffer.
*/
pos = RSYNC_MIN_BLOCK_SIZE - mtctx->inBuff.filled;
if (pos >= RSYNC_LENGTH) {
prev = istart + pos - RSYNC_LENGTH;
hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
} else {
assert(mtctx->inBuff.filled >= RSYNC_LENGTH);
prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
hash = ZSTD_rollingHash_compute(prev + pos, (RSYNC_LENGTH - pos));
hash = ZSTD_rollingHash_append(hash, istart, pos);
}
} else {
/* We have enough bytes buffered to initialize the hash,
* and are have processed enough bytes to find a sync point.
* Start scanning at the beginning of the input.
*/
assert(mtctx->inBuff.filled >= RSYNC_MIN_BLOCK_SIZE);
assert(RSYNC_MIN_BLOCK_SIZE >= RSYNC_LENGTH);
pos = 0;
prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
@@ -1705,16 +1740,6 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
syncPoint.flush = 1;
return syncPoint;
}
} else {
/* We don't have enough bytes buffered to initialize the hash, but
* we know we have at least RSYNC_LENGTH bytes total.
* Start scanning after the first RSYNC_LENGTH bytes less the bytes
* already buffered.
*/
pos = RSYNC_LENGTH - mtctx->inBuff.filled;
prev = (BYTE const*)mtctx->inBuff.buffer.start - pos;
hash = ZSTD_rollingHash_compute(mtctx->inBuff.buffer.start, mtctx->inBuff.filled);
hash = ZSTD_rollingHash_append(hash, istart, pos);
}
/* Starting with the hash of the previous RSYNC_LENGTH bytes, roll
* through the input. If we hit a synchronization point, then cut the
@@ -1726,8 +1751,9 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
*/
for (; pos < syncPoint.toLoad; ++pos) {
BYTE const toRemove = pos < RSYNC_LENGTH ? prev[pos] : istart[pos - RSYNC_LENGTH];
/* if (pos >= RSYNC_LENGTH) assert(ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash); */
assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
hash = ZSTD_rollingHash_rotate(hash, toRemove, istart[pos], primePower);
assert(mtctx->inBuff.filled + pos >= RSYNC_MIN_BLOCK_SIZE);
if ((hash & hitMask) == hitMask) {
syncPoint.toLoad = pos + 1;
syncPoint.flush = 1;
lib/decompress/huf_decompress.c
+749 -188
View File
@@ -22,6 +22,13 @@
#define HUF_STATIC_LINKING_ONLY
#include "../common/huf.h"
#include "../common/error_private.h"
#include "../common/zstd_internal.h"
/* **************************************************************
* Constants
****************************************************************/
#define HUF_DECODER_FAST_TABLELOG 11
/* **************************************************************
* Macros
@@ -36,6 +43,54 @@
#error "Cannot force the use of the X1 and X2 decoders at the same time!"
#endif
/* Only use assembly on Linux / MacOS.
* Disable when MSAN is enabled.
*/
#if defined(__linux__) || defined(__linux) || defined(__APPLE__)
# if ZSTD_MEMORY_SANITIZER
# define HUF_ASM_SUPPORTED 0
# elif ZSTD_DATAFLOW_SANITIZER
# define HUF_ASM_SUPPORTED 0
# else
# define HUF_ASM_SUPPORTED 1
# endif
#else
# define HUF_ASM_SUPPORTED 0
#endif
/* HUF_DISABLE_ASM: Disables all ASM implementations. */
#if !defined(HUF_DISABLE_ASM) && \
HUF_ASM_SUPPORTED && \
defined(__x86_64__) && (DYNAMIC_BMI2 || defined(__BMI2__))
# define HUF_ENABLE_ASM_X86_64_BMI2 1
#else
# define HUF_ENABLE_ASM_X86_64_BMI2 0
#endif
#if HUF_ENABLE_ASM_X86_64_BMI2 && DYNAMIC_BMI2
# define HUF_ASM_X86_64_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
#else
# define HUF_ASM_X86_64_BMI2_ATTRS
#endif
#ifdef __cplusplus
# define HUF_EXTERN_C extern "C"
#else
# define HUF_EXTERN_C
#endif
#define HUF_ASM_DECL HUF_EXTERN_C
#if DYNAMIC_BMI2 || (HUF_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__))
# define HUF_NEED_BMI2_FUNCTION 1
#else
# define HUF_NEED_BMI2_FUNCTION 0
#endif
#if !(HUF_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__))
# define HUF_NEED_DEFAULT_FUNCTION 1
#else
# define HUF_NEED_DEFAULT_FUNCTION 0
#endif
/* **************************************************************
* Error Management
@@ -65,7 +120,7 @@
return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable); \
} \
\
static TARGET_ATTRIBUTE("bmi2") size_t fn##_bmi2( \
static BMI2_TARGET_ATTRIBUTE size_t fn##_bmi2( \
void* dst, size_t dstSize, \
const void* cSrc, size_t cSrcSize, \
const HUF_DTable* DTable) \
@@ -107,13 +162,146 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
return dtd;
}
#if HUF_ENABLE_ASM_X86_64_BMI2
static size_t HUF_initDStream(BYTE const* ip) {
BYTE const lastByte = ip[7];
size_t const bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
size_t const value = MEM_readLEST(ip) | 1;
assert(bitsConsumed <= 8);
return value << bitsConsumed;
}
typedef struct {
BYTE const* ip[4];
BYTE* op[4];
U64 bits[4];
void const* dt;
BYTE const* ilimit;
BYTE* oend;
BYTE const* iend[4];
} HUF_DecompressAsmArgs;
/**
* Initializes args for the asm decoding loop.
* @returns 0 on success
* 1 if the fallback implementation should be used.
* Or an error code on failure.
*/
static size_t HUF_DecompressAsmArgs_init(HUF_DecompressAsmArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
{
void const* dt = DTable + 1;
U32 const dtLog = HUF_getDTableDesc(DTable).tableLog;
const BYTE* const ilimit = (const BYTE*)src + 6 + 8;
BYTE* const oend = (BYTE*)dst + dstSize;
/* We're assuming x86-64 BMI2 - assure that this is the case. */
assert(MEM_isLittleEndian() && !MEM_32bits());
/* strict minimum : jump table + 1 byte per stream */
if (srcSize < 10)
return ERROR(corruption_detected);
/* Must have at least 8 bytes per stream because we don't handle initializing smaller bit containers.
* If table log is not correct at this point, fallback to the old decoder.
* On small inputs we don't have enough data to trigger the fast loop, so use the old decoder.
*/
if (dtLog != HUF_DECODER_FAST_TABLELOG)
return 1;
/* Read the jump table. */
{
const BYTE* const istart = (const BYTE*)src;
size_t const length1 = MEM_readLE16(istart);
size_t const length2 = MEM_readLE16(istart+2);
size_t const length3 = MEM_readLE16(istart+4);
size_t const length4 = srcSize - (length1 + length2 + length3 + 6);
args->iend[0] = istart + 6; /* jumpTable */
args->iend[1] = args->iend[0] + length1;
args->iend[2] = args->iend[1] + length2;
args->iend[3] = args->iend[2] + length3;
/* HUF_initDStream() requires this, and this small of an input
* won't benefit from the ASM loop anyways.
* length1 must be >= 16 so that ip[0] >= ilimit before the loop
* starts.
*/
if (length1 < 16 || length2 < 8 || length3 < 8 || length4 < 8)
return 1;
if (length4 > srcSize) return ERROR(corruption_detected); /* overflow */
}
/* ip[] contains the position that is currently loaded into bits[]. */
args->ip[0] = args->iend[1] - sizeof(U64);
args->ip[1] = args->iend[2] - sizeof(U64);
args->ip[2] = args->iend[3] - sizeof(U64);
args->ip[3] = (BYTE const*)src + srcSize - sizeof(U64);
/* op[] contains the output pointers. */
args->op[0] = (BYTE*)dst;
args->op[1] = args->op[0] + (dstSize+3)/4;
args->op[2] = args->op[1] + (dstSize+3)/4;
args->op[3] = args->op[2] + (dstSize+3)/4;
/* No point to call the ASM loop for tiny outputs. */
if (args->op[3] >= oend)
return 1;
/* bits[] is the bit container.
* It is read from the MSB down to the LSB.
* It is shifted left as it is read, and zeros are
* shifted in. After the lowest valid bit a 1 is
* set, so that CountTrailingZeros(bits[]) can be used
* to count how many bits we've consumed.
*/
args->bits[0] = HUF_initDStream(args->ip[0]);
args->bits[1] = HUF_initDStream(args->ip[1]);
args->bits[2] = HUF_initDStream(args->ip[2]);
args->bits[3] = HUF_initDStream(args->ip[3]);
/* If ip[] >= ilimit, it is guaranteed to be safe to
* reload bits[]. It may be beyond its section, but is
* guaranteed to be valid (>= istart).
*/
args->ilimit = ilimit;
args->oend = oend;
args->dt = dt;
return 0;
}
static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressAsmArgs const* args, int stream, BYTE* segmentEnd)
{
/* Validate that we haven't overwritten. */
if (args->op[stream] > segmentEnd)
return ERROR(corruption_detected);
/* Validate that we haven't read beyond iend[].
* Note that ip[] may be < iend[] because the MSB is
* the next bit to read, and we may have consumed 100%
* of the stream, so down to iend[i] - 8 is valid.
*/
if (args->ip[stream] < args->iend[stream] - 8)
return ERROR(corruption_detected);
/* Construct the BIT_DStream_t. */
bit->bitContainer = MEM_readLE64(args->ip[stream]);
bit->bitsConsumed = ZSTD_countTrailingZeros((size_t)args->bits[stream]);
bit->start = (const char*)args->iend[0];
bit->limitPtr = bit->start + sizeof(size_t);
bit->ptr = (const char*)args->ip[stream];
return 0;
}
#endif
#ifndef HUF_FORCE_DECOMPRESS_X2
/*-***************************/
/* single-symbol decoding */
/*-***************************/
typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1; /* single-symbol decoding */
typedef struct { BYTE nbBits; BYTE byte; } HUF_DEltX1; /* single-symbol decoding */
/**
* Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
@@ -122,14 +310,44 @@ typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1; /* single-symbol decodi
static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
U64 D4;
if (MEM_isLittleEndian()) {
D4 = symbol + (nbBits << 8);
} else {
D4 = (symbol << 8) + nbBits;
} else {
D4 = symbol + (nbBits << 8);
}
D4 *= 0x0001000100010001ULL;
return D4;
}
/**
* Increase the tableLog to targetTableLog and rescales the stats.
* If tableLog > targetTableLog this is a no-op.
* @returns New tableLog
*/
static U32 HUF_rescaleStats(BYTE* huffWeight, U32* rankVal, U32 nbSymbols, U32 tableLog, U32 targetTableLog)
{
if (tableLog > targetTableLog)
return tableLog;
if (tableLog < targetTableLog) {
U32 const scale = targetTableLog - tableLog;
U32 s;
/* Increase the weight for all non-zero probability symbols by scale. */
for (s = 0; s < nbSymbols; ++s) {
huffWeight[s] += (BYTE)((huffWeight[s] == 0) ? 0 : scale);
}
/* Update rankVal to reflect the new weights.
* All weights except 0 get moved to weight + scale.
* Weights [1, scale] are empty.
*/
for (s = targetTableLog; s > scale; --s) {
rankVal[s] = rankVal[s - scale];
}
for (s = scale; s > 0; --s) {
rankVal[s] = 0;
}
}
return targetTableLog;
}
typedef struct {
U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
@@ -162,8 +380,12 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), bmi2);
if (HUF_isError(iSize)) return iSize;
/* Table header */
{ DTableDesc dtd = HUF_getDTableDesc(DTable);
U32 const maxTableLog = dtd.maxTableLog + 1;
U32 const targetTableLog = MIN(maxTableLog, HUF_DECODER_FAST_TABLELOG);
tableLog = HUF_rescaleStats(wksp->huffWeight, wksp->rankVal, nbSymbols, tableLog, targetTableLog);
if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge); /* DTable too small, Huffman tree cannot fit in */
dtd.tableType = 0;
dtd.tableLog = (BYTE)tableLog;
@@ -207,7 +429,7 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
/* fill DTable
* We fill all entries of each weight in order.
* That way length is a constant for each iteration of the outter loop.
* That way length is a constant for each iteration of the outer loop.
* We can switch based on the length to a different inner loop which is
* optimized for that particular case.
*/
@@ -304,11 +526,15 @@ HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, cons
BYTE* const pStart = p;
/* up to 4 symbols at a time */
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
if ((pEnd - p) > 3) {
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
}
} else {
BIT_reloadDStream(bitDPtr);
}
/* [0-3] symbols remaining */
@@ -388,33 +614,36 @@ HUF_decompress4X1_usingDTable_internal_body(
U32 endSignal = 1;
if (length4 > cSrcSize) return ERROR(corruption_detected); /* overflow */
if (opStart4 > oend) return ERROR(corruption_detected); /* overflow */
CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
/* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
for ( ; (endSignal) & (op4 < olimit) ; ) {
HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
if ((size_t)(oend - op4) >= sizeof(size_t)) {
for ( ; (endSignal) & (op4 < olimit) ; ) {
HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
}
}
/* check corruption */
@@ -440,6 +669,79 @@ HUF_decompress4X1_usingDTable_internal_body(
}
}
#if HUF_NEED_BMI2_FUNCTION
static BMI2_TARGET_ATTRIBUTE
size_t HUF_decompress4X1_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable) {
return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
}
#endif
#if HUF_NEED_DEFAULT_FUNCTION
static
size_t HUF_decompress4X1_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable) {
return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
}
#endif
#if HUF_ENABLE_ASM_X86_64_BMI2
HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop(HUF_DecompressAsmArgs* args);
static HUF_ASM_X86_64_BMI2_ATTRS
size_t
HUF_decompress4X1_usingDTable_internal_bmi2_asm(
void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize,
const HUF_DTable* DTable)
{
void const* dt = DTable + 1;
const BYTE* const iend = (const BYTE*)cSrc + 6;
BYTE* const oend = (BYTE*)dst + dstSize;
HUF_DecompressAsmArgs args;
{
size_t const ret = HUF_DecompressAsmArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
FORWARD_IF_ERROR(ret, "Failed to init asm args");
if (ret != 0)
return HUF_decompress4X1_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
}
assert(args.ip[0] >= args.ilimit);
HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop(&args);
/* Our loop guarantees that ip[] >= ilimit and that we haven't
* overwritten any op[].
*/
assert(args.ip[0] >= iend);
assert(args.ip[1] >= iend);
assert(args.ip[2] >= iend);
assert(args.ip[3] >= iend);
assert(args.op[3] <= oend);
(void)iend;
/* finish bit streams one by one. */
{
size_t const segmentSize = (dstSize+3) / 4;
BYTE* segmentEnd = (BYTE*)dst;
int i;
for (i = 0; i < 4; ++i) {
BIT_DStream_t bit;
if (segmentSize <= (size_t)(oend - segmentEnd))
segmentEnd += segmentSize;
else
segmentEnd = oend;
FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
/* Decompress and validate that we've produced exactly the expected length. */
args.op[i] += HUF_decodeStreamX1(args.op[i], &bit, segmentEnd, (HUF_DEltX1 const*)dt, HUF_DECODER_FAST_TABLELOG);
if (args.op[i] != segmentEnd) return ERROR(corruption_detected);
}
}
/* decoded size */
return dstSize;
}
#endif /* HUF_ENABLE_ASM_X86_64_BMI2 */
typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
const void *cSrc,
@@ -447,8 +749,28 @@ typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
const HUF_DTable *DTable);
HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
HUF_DGEN(HUF_decompress4X1_usingDTable_internal)
static size_t HUF_decompress4X1_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
# if HUF_ENABLE_ASM_X86_64_BMI2
return HUF_decompress4X1_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
# else
return HUF_decompress4X1_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
# endif
}
#else
(void)bmi2;
#endif
#if HUF_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
return HUF_decompress4X1_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
#else
return HUF_decompress4X1_usingDTable_internal_default(dst, dstSize, cSrc, cSrcSize, DTable);
#endif
}
size_t HUF_decompress1X1_usingDTable(
@@ -518,106 +840,226 @@ size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
/* *************************/
typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2; /* double-symbols decoding */
typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
typedef struct { BYTE symbol; } sortedSymbol_t;
typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
/**
* Constructs a HUF_DEltX2 in a U32.
*/
static U32 HUF_buildDEltX2U32(U32 symbol, U32 nbBits, U32 baseSeq, int level)
{
U32 seq;
DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, sequence) == 0);
DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, nbBits) == 2);
DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, length) == 3);
DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U32));
if (MEM_isLittleEndian()) {
seq = level == 1 ? symbol : (baseSeq + (symbol << 8));
return seq + (nbBits << 16) + ((U32)level << 24);
} else {
seq = level == 1 ? (symbol << 8) : ((baseSeq << 8) + symbol);
return (seq << 16) + (nbBits << 8) + (U32)level;
}
}
/**
* Constructs a HUF_DEltX2.
*/
static HUF_DEltX2 HUF_buildDEltX2(U32 symbol, U32 nbBits, U32 baseSeq, int level)
{
HUF_DEltX2 DElt;
U32 const val = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
DEBUG_STATIC_ASSERT(sizeof(DElt) == sizeof(val));
ZSTD_memcpy(&DElt, &val, sizeof(val));
return DElt;
}
/**
* Constructs 2 HUF_DEltX2s and packs them into a U64.
*/
static U64 HUF_buildDEltX2U64(U32 symbol, U32 nbBits, U16 baseSeq, int level)
{
U32 DElt = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
return (U64)DElt + ((U64)DElt << 32);
}
/**
* Fills the DTable rank with all the symbols from [begin, end) that are each
* nbBits long.
*
* @param DTableRank The start of the rank in the DTable.
* @param begin The first symbol to fill (inclusive).
* @param end The last symbol to fill (exclusive).
* @param nbBits Each symbol is nbBits long.
* @param tableLog The table log.
* @param baseSeq If level == 1 { 0 } else { the first level symbol }
* @param level The level in the table. Must be 1 or 2.
*/
static void HUF_fillDTableX2ForWeight(
HUF_DEltX2* DTableRank,
sortedSymbol_t const* begin, sortedSymbol_t const* end,
U32 nbBits, U32 tableLog,
U16 baseSeq, int const level)
{
U32 const length = 1U << ((tableLog - nbBits) & 0x1F /* quiet static-analyzer */);
const sortedSymbol_t* ptr;
assert(level >= 1 && level <= 2);
switch (length) {
case 1:
for (ptr = begin; ptr != end; ++ptr) {
HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
*DTableRank++ = DElt;
}
break;
case 2:
for (ptr = begin; ptr != end; ++ptr) {
HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
DTableRank[0] = DElt;
DTableRank[1] = DElt;
DTableRank += 2;
}
break;
case 4:
for (ptr = begin; ptr != end; ++ptr) {
U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
DTableRank += 4;
}
break;
case 8:
for (ptr = begin; ptr != end; ++ptr) {
U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
DTableRank += 8;
}
break;
default:
for (ptr = begin; ptr != end; ++ptr) {
U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
HUF_DEltX2* const DTableRankEnd = DTableRank + length;
for (; DTableRank != DTableRankEnd; DTableRank += 8) {
ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
}
}
break;
}
}
/* HUF_fillDTableX2Level2() :
* `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 consumed,
const U32* rankValOrigin, const int minWeight,
const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
U32 nbBitsBaseline, U16 baseSeq, U32* wksp, size_t wkspSize)
static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 targetLog, const U32 consumedBits,
const U32* rankVal, const int minWeight, const int maxWeight1,
const sortedSymbol_t* sortedSymbols, U32 const* rankStart,
U32 nbBitsBaseline, U16 baseSeq)
{
HUF_DEltX2 DElt;
U32* rankVal = wksp;
assert(wkspSize >= HUF_TABLELOG_MAX + 1);
(void)wkspSize;
/* get pre-calculated rankVal */
ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
/* fill skipped values */
/* Fill skipped values (all positions up to rankVal[minWeight]).
* These are positions only get a single symbol because the combined weight
* is too large.
*/
if (minWeight>1) {
U32 i, skipSize = rankVal[minWeight];
MEM_writeLE16(&(DElt.sequence), baseSeq);
DElt.nbBits = (BYTE)(consumed);
DElt.length = 1;
for (i = 0; i < skipSize; i++)
DTable[i] = DElt;
U32 const length = 1U << ((targetLog - consumedBits) & 0x1F /* quiet static-analyzer */);
U64 const DEltX2 = HUF_buildDEltX2U64(baseSeq, consumedBits, /* baseSeq */ 0, /* level */ 1);
int const skipSize = rankVal[minWeight];
assert(length > 1);
assert((U32)skipSize < length);
switch (length) {
case 2:
assert(skipSize == 1);
ZSTD_memcpy(DTable, &DEltX2, sizeof(DEltX2));
break;
case 4:
assert(skipSize <= 4);
ZSTD_memcpy(DTable + 0, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTable + 2, &DEltX2, sizeof(DEltX2));
break;
default:
{
int i;
for (i = 0; i < skipSize; i += 8) {
ZSTD_memcpy(DTable + i + 0, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTable + i + 2, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTable + i + 4, &DEltX2, sizeof(DEltX2));
ZSTD_memcpy(DTable + i + 6, &DEltX2, sizeof(DEltX2));
}
}
}
}
/* fill DTable */
{ U32 s; for (s=0; s<sortedListSize; s++) { /* note : sortedSymbols already skipped */
const U32 symbol = sortedSymbols[s].symbol;
const U32 weight = sortedSymbols[s].weight;
const U32 nbBits = nbBitsBaseline - weight;
const U32 length = 1 << (sizeLog-nbBits);
const U32 start = rankVal[weight];
U32 i = start;
const U32 end = start + length;
MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
DElt.nbBits = (BYTE)(nbBits + consumed);
DElt.length = 2;
do { DTable[i++] = DElt; } while (i<end); /* since length >= 1 */
rankVal[weight] += length;
} }
/* Fill each of the second level symbols by weight. */
{
int w;
for (w = minWeight; w < maxWeight1; ++w) {
int const begin = rankStart[w];
int const end = rankStart[w+1];
U32 const nbBits = nbBitsBaseline - w;
U32 const totalBits = nbBits + consumedBits;
HUF_fillDTableX2ForWeight(
DTable + rankVal[w],
sortedSymbols + begin, sortedSymbols + end,
totalBits, targetLog,
baseSeq, /* level */ 2);
}
}
}
static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
const sortedSymbol_t* sortedList, const U32 sortedListSize,
const sortedSymbol_t* sortedList,
const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
const U32 nbBitsBaseline, U32* wksp, size_t wkspSize)
const U32 nbBitsBaseline)
{
U32* rankVal = wksp;
U32* const rankVal = rankValOrigin[0];
const int scaleLog = nbBitsBaseline - targetLog; /* note : targetLog >= srcLog, hence scaleLog <= 1 */
const U32 minBits = nbBitsBaseline - maxWeight;
U32 s;
int w;
int const wEnd = (int)maxWeight + 1;
assert(wkspSize >= HUF_TABLELOG_MAX + 1);
wksp += HUF_TABLELOG_MAX + 1;
wkspSize -= HUF_TABLELOG_MAX + 1;
/* Fill DTable in order of weight. */
for (w = 1; w < wEnd; ++w) {
int const begin = (int)rankStart[w];
int const end = (int)rankStart[w+1];
U32 const nbBits = nbBitsBaseline - w;
ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
/* fill DTable */
for (s=0; s<sortedListSize; s++) {
const U16 symbol = sortedList[s].symbol;
const U32 weight = sortedList[s].weight;
const U32 nbBits = nbBitsBaseline - weight;
const U32 start = rankVal[weight];
const U32 length = 1 << (targetLog-nbBits);
if (targetLog-nbBits >= minBits) { /* enough room for a second symbol */
U32 sortedRank;
if (targetLog-nbBits >= minBits) {
/* Enough room for a second symbol. */
int start = rankVal[w];
U32 const length = 1U << ((targetLog - nbBits) & 0x1F /* quiet static-analyzer */);
int minWeight = nbBits + scaleLog;
int s;
if (minWeight < 1) minWeight = 1;
sortedRank = rankStart[minWeight];
HUF_fillDTableX2Level2(DTable+start, targetLog-nbBits, nbBits,
rankValOrigin[nbBits], minWeight,
sortedList+sortedRank, sortedListSize-sortedRank,
nbBitsBaseline, symbol, wksp, wkspSize);
/* Fill the DTable for every symbol of weight w.
* These symbols get at least 1 second symbol.
*/
for (s = begin; s != end; ++s) {
HUF_fillDTableX2Level2(
DTable + start, targetLog, nbBits,
rankValOrigin[nbBits], minWeight, wEnd,
sortedList, rankStart,
nbBitsBaseline, sortedList[s].symbol);
start += length;
}
} else {
HUF_DEltX2 DElt;
MEM_writeLE16(&(DElt.sequence), symbol);
DElt.nbBits = (BYTE)(nbBits);
DElt.length = 1;
{ U32 const end = start + length;
U32 u;
for (u = start; u < end; u++) DTable[u] = DElt;
} }
rankVal[weight] += length;
/* Only a single symbol. */
HUF_fillDTableX2ForWeight(
DTable + rankVal[w],
sortedList + begin, sortedList + end,
nbBits, targetLog,
/* baseSeq */ 0, /* level */ 1);
}
}
}
typedef struct {
rankValCol_t rankVal[HUF_TABLELOG_MAX];
U32 rankStats[HUF_TABLELOG_MAX + 1];
U32 rankStart0[HUF_TABLELOG_MAX + 2];
U32 rankStart0[HUF_TABLELOG_MAX + 3];
sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
U32 calleeWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
@@ -627,9 +1069,16 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
{
U32 tableLog, maxW, sizeOfSort, nbSymbols;
return HUF_readDTableX2_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
}
size_t HUF_readDTableX2_wksp_bmi2(HUF_DTable* DTable,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize, int bmi2)
{
U32 tableLog, maxW, nbSymbols;
DTableDesc dtd = HUF_getDTableDesc(DTable);
U32 const maxTableLog = dtd.maxTableLog;
U32 maxTableLog = dtd.maxTableLog;
size_t iSize;
void* dtPtr = DTable+1; /* force compiler to avoid strict-aliasing */
HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
@@ -647,11 +1096,12 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
/* ZSTD_memset(weightList, 0, sizeof(weightList)); */ /* is not necessary, even though some analyzer complain ... */
iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), /* bmi2 */ 0);
iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), bmi2);
if (HUF_isError(iSize)) return iSize;
/* check result */
if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge); /* DTable can't fit code depth */
if (tableLog <= HUF_DECODER_FAST_TABLELOG && maxTableLog > HUF_DECODER_FAST_TABLELOG) maxTableLog = HUF_DECODER_FAST_TABLELOG;
/* find maxWeight */
for (maxW = tableLog; wksp->rankStats[maxW]==0; maxW--) {} /* necessarily finds a solution before 0 */
@@ -664,7 +1114,7 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
rankStart[w] = curr;
}
rankStart[0] = nextRankStart; /* put all 0w symbols at the end of sorted list*/
sizeOfSort = nextRankStart;
rankStart[maxW+1] = nextRankStart;
}
/* sort symbols by weight */
@@ -673,7 +1123,6 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
U32 const w = wksp->weightList[s];
U32 const r = rankStart[w]++;
wksp->sortedSymbol[r].symbol = (BYTE)s;
wksp->sortedSymbol[r].weight = (BYTE)w;
}
rankStart[0] = 0; /* forget 0w symbols; this is beginning of weight(1) */
}
@@ -698,10 +1147,9 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
} } } }
HUF_fillDTableX2(dt, maxTableLog,
wksp->sortedSymbol, sizeOfSort,
wksp->sortedSymbol,
wksp->rankStart0, wksp->rankVal, maxW,
tableLog+1,
wksp->calleeWksp, sizeof(wksp->calleeWksp) / sizeof(U32));
tableLog+1);
dtd.tableLog = (BYTE)maxTableLog;
dtd.tableType = 1;
@@ -714,7 +1162,7 @@ FORCE_INLINE_TEMPLATE U32
HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
{
size_t const val = BIT_lookBitsFast(DStream, dtLog); /* note : dtLog >= 1 */
ZSTD_memcpy(op, dt+val, 2);
ZSTD_memcpy(op, &dt[val].sequence, 2);
BIT_skipBits(DStream, dt[val].nbBits);
return dt[val].length;
}
@@ -723,15 +1171,17 @@ FORCE_INLINE_TEMPLATE U32
HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
{
size_t const val = BIT_lookBitsFast(DStream, dtLog); /* note : dtLog >= 1 */
ZSTD_memcpy(op, dt+val, 1);
if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
else {
ZSTD_memcpy(op, &dt[val].sequence, 1);
if (dt[val].length==1) {
BIT_skipBits(DStream, dt[val].nbBits);
} else {
if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
BIT_skipBits(DStream, dt[val].nbBits);
if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
/* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
} }
}
}
return 1;
}
@@ -753,19 +1203,37 @@ HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
BYTE* const pStart = p;
/* up to 8 symbols at a time */
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
if ((size_t)(pEnd - p) >= sizeof(bitDPtr->bitContainer)) {
if (dtLog <= 11 && MEM_64bits()) {
/* up to 10 symbols at a time */
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-9)) {
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
}
} else {
/* up to 8 symbols at a time */
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
}
}
} else {
BIT_reloadDStream(bitDPtr);
}
/* closer to end : up to 2 symbols at a time */
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
if ((size_t)(pEnd - p) >= 2) {
while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
while (p <= pEnd-2)
HUF_DECODE_SYMBOLX2_0(p, bitDPtr); /* no need to reload : reached the end of DStream */
while (p <= pEnd-2)
HUF_DECODE_SYMBOLX2_0(p, bitDPtr); /* no need to reload : reached the end of DStream */
}
if (p < pEnd)
p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
@@ -799,7 +1267,6 @@ HUF_decompress1X2_usingDTable_internal_body(
/* decoded size */
return dstSize;
}
FORCE_INLINE_TEMPLATE size_t
HUF_decompress4X2_usingDTable_internal_body(
void* dst, size_t dstSize,
@@ -841,57 +1308,60 @@ HUF_decompress4X2_usingDTable_internal_body(
U32 const dtLog = dtd.tableLog;
if (length4 > cSrcSize) return ERROR(corruption_detected); /* overflow */
if (opStart4 > oend) return ERROR(corruption_detected); /* overflow */
CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
/* 16-32 symbols per loop (4-8 symbols per stream) */
for ( ; (endSignal) & (op4 < olimit); ) {
if ((size_t)(oend - op4) >= sizeof(size_t)) {
for ( ; (endSignal) & (op4 < olimit); ) {
#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
#else
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
endSignal = (U32)LIKELY(
(BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
endSignal = (U32)LIKELY((U32)
(BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
& (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
#endif
}
}
/* check corruption */
@@ -915,8 +1385,99 @@ HUF_decompress4X2_usingDTable_internal_body(
}
}
#if HUF_NEED_BMI2_FUNCTION
static BMI2_TARGET_ATTRIBUTE
size_t HUF_decompress4X2_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable) {
return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
}
#endif
#if HUF_NEED_DEFAULT_FUNCTION
static
size_t HUF_decompress4X2_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable) {
return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
}
#endif
#if HUF_ENABLE_ASM_X86_64_BMI2
HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop(HUF_DecompressAsmArgs* args);
static HUF_ASM_X86_64_BMI2_ATTRS size_t
HUF_decompress4X2_usingDTable_internal_bmi2_asm(
void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize,
const HUF_DTable* DTable) {
void const* dt = DTable + 1;
const BYTE* const iend = (const BYTE*)cSrc + 6;
BYTE* const oend = (BYTE*)dst + dstSize;
HUF_DecompressAsmArgs args;
{
size_t const ret = HUF_DecompressAsmArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
FORWARD_IF_ERROR(ret, "Failed to init asm args");
if (ret != 0)
return HUF_decompress4X2_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
}
assert(args.ip[0] >= args.ilimit);
HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop(&args);
/* note : op4 already verified within main loop */
assert(args.ip[0] >= iend);
assert(args.ip[1] >= iend);
assert(args.ip[2] >= iend);
assert(args.ip[3] >= iend);
assert(args.op[3] <= oend);
(void)iend;
/* finish bitStreams one by one */
{
size_t const segmentSize = (dstSize+3) / 4;
BYTE* segmentEnd = (BYTE*)dst;
int i;
for (i = 0; i < 4; ++i) {
BIT_DStream_t bit;
if (segmentSize <= (size_t)(oend - segmentEnd))
segmentEnd += segmentSize;
else
segmentEnd = oend;
FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
args.op[i] += HUF_decodeStreamX2(args.op[i], &bit, segmentEnd, (HUF_DEltX2 const*)dt, HUF_DECODER_FAST_TABLELOG);
if (args.op[i] != segmentEnd)
return ERROR(corruption_detected);
}
}
/* decoded size */
return dstSize;
}
#endif /* HUF_ENABLE_ASM_X86_64_BMI2 */
static size_t HUF_decompress4X2_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
size_t cSrcSize, HUF_DTable const* DTable, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
# if HUF_ENABLE_ASM_X86_64_BMI2
return HUF_decompress4X2_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
# else
return HUF_decompress4X2_usingDTable_internal_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);
# endif
}
#else
(void)bmi2;
#endif
#if HUF_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
return HUF_decompress4X2_usingDTable_internal_bmi2_asm(dst, dstSize, cSrc, cSrcSize, DTable);
#else
return HUF_decompress4X2_usingDTable_internal_default(dst, dstSize, cSrc, cSrcSize, DTable);
#endif
}
HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
HUF_DGEN(HUF_decompress4X2_usingDTable_internal)
size_t HUF_decompress1X2_usingDTable(
void* dst, size_t dstSize,
@@ -1025,25 +1586,25 @@ size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
static const algo_time_t algoTime[16 /* Quantization */][2 /* single, double */] =
{
/* single, double, quad */
{{0,0}, {1,1}, {2,2}}, /* Q==0 : impossible */
{{0,0}, {1,1}, {2,2}}, /* Q==1 : impossible */
{{ 38,130}, {1313, 74}, {2151, 38}}, /* Q == 2 : 12-18% */
{{ 448,128}, {1353, 74}, {2238, 41}}, /* Q == 3 : 18-25% */
{{ 556,128}, {1353, 74}, {2238, 47}}, /* Q == 4 : 25-32% */
{{ 714,128}, {1418, 74}, {2436, 53}}, /* Q == 5 : 32-38% */
{{ 883,128}, {1437, 74}, {2464, 61}}, /* Q == 6 : 38-44% */
{{ 897,128}, {1515, 75}, {2622, 68}}, /* Q == 7 : 44-50% */
{{ 926,128}, {1613, 75}, {2730, 75}}, /* Q == 8 : 50-56% */
{{ 947,128}, {1729, 77}, {3359, 77}}, /* Q == 9 : 56-62% */
{{1107,128}, {2083, 81}, {4006, 84}}, /* Q ==10 : 62-69% */
{{1177,128}, {2379, 87}, {4785, 88}}, /* Q ==11 : 69-75% */
{{1242,128}, {2415, 93}, {5155, 84}}, /* Q ==12 : 75-81% */
{{1349,128}, {2644,106}, {5260,106}}, /* Q ==13 : 81-87% */
{{1455,128}, {2422,124}, {4174,124}}, /* Q ==14 : 87-93% */
{{ 722,128}, {1891,145}, {1936,146}}, /* Q ==15 : 93-99% */
{{0,0}, {1,1}}, /* Q==0 : impossible */
{{0,0}, {1,1}}, /* Q==1 : impossible */
{{ 150,216}, { 381,119}}, /* Q == 2 : 12-18% */
{{ 170,205}, { 514,112}}, /* Q == 3 : 18-25% */
{{ 177,199}, { 539,110}}, /* Q == 4 : 25-32% */
{{ 197,194}, { 644,107}}, /* Q == 5 : 32-38% */
{{ 221,192}, { 735,107}}, /* Q == 6 : 38-44% */
{{ 256,189}, { 881,106}}, /* Q == 7 : 44-50% */
{{ 359,188}, {1167,109}}, /* Q == 8 : 50-56% */
{{ 582,187}, {1570,114}}, /* Q == 9 : 56-62% */
{{ 688,187}, {1712,122}}, /* Q ==10 : 62-69% */
{{ 825,186}, {1965,136}}, /* Q ==11 : 69-75% */
{{ 976,185}, {2131,150}}, /* Q ==12 : 75-81% */
{{1180,186}, {2070,175}}, /* Q ==13 : 81-87% */
{{1377,185}, {1731,202}}, /* Q ==14 : 87-93% */
{{1412,185}, {1695,202}}, /* Q ==15 : 93-99% */
};
#endif
@@ -1070,7 +1631,7 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
U32 const D256 = (U32)(dstSize >> 8);
U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
DTime1 += DTime1 >> 3; /* advantage to algorithm using less memory, to reduce cache eviction */
DTime1 += DTime1 >> 5; /* small advantage to algorithm using less memory, to reduce cache eviction */
return DTime1 < DTime0;
}
#endif
lib/decompress/huf_decompress_amd64.S
+578
View File
@@ -0,0 +1,578 @@
#if !defined(HUF_DISABLE_ASM) && defined(__x86_64__)
/* Stack marking
* ref: https://wiki.gentoo.org/wiki/Hardened/GNU_stack_quickstart
*/
#if defined(__linux__) && defined(__ELF__)
.section .note.GNU-stack,"",%progbits
#endif
/* Calling convention:
*
* %rdi contains the first argument: HUF_DecompressAsmArgs*.
* %rbp isn't maintained (no frame pointer).
* %rsp contains the stack pointer that grows down.
* No red-zone is assumed, only addresses >= %rsp are used.
* All register contents are preserved.
*
* TODO: Support Windows calling convention.
*/
.global HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop
.global HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop
.global _HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop
.global _HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop
.text
/* Sets up register mappings for clarity.
* op[], bits[], dtable & ip[0] each get their own register.
* ip[1,2,3] & olimit alias var[].
* %rax is a scratch register.
*/
#define op0 rsi
#define op1 rbx
#define op2 rcx
#define op3 rdi
#define ip0 r8
#define ip1 r9
#define ip2 r10
#define ip3 r11
#define bits0 rbp
#define bits1 rdx
#define bits2 r12
#define bits3 r13
#define dtable r14
#define olimit r15
/* var[] aliases ip[1,2,3] & olimit
* ip[1,2,3] are saved every iteration.
* olimit is only used in compute_olimit.
*/
#define var0 r15
#define var1 r9
#define var2 r10
#define var3 r11
/* 32-bit var registers */
#define vard0 r15d
#define vard1 r9d
#define vard2 r10d
#define vard3 r11d
/* Helper macro: args if idx != 4. */
#define IF_NOT_4_0(...) __VA_ARGS__
#define IF_NOT_4_1(...) __VA_ARGS__
#define IF_NOT_4_2(...) __VA_ARGS__
#define IF_NOT_4_3(...) __VA_ARGS__
#define IF_NOT_4_4(...)
#define IF_NOT_4_(idx, ...) IF_NOT_4_##idx(__VA_ARGS__)
#define IF_NOT_4(idx, ...) IF_NOT_4_(idx, __VA_ARGS__)
/* Calls X(N) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM(X) \
X(0); \
X(1); \
X(2); \
X(3)
/* Calls X(N, idx) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM_WITH_INDEX(X, idx) \
X(0, idx); \
X(1, idx); \
X(2, idx); \
X(3, idx)
/* Define both _HUF_* & HUF_* symbols because MacOS
* C symbols are prefixed with '_' & Linux symbols aren't.
*/
_HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop:
HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop:
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
push %rcx
push %rdx
push %rbp
push %rsi
push %rdi
push %r8
push %r9
push %r10
push %r11
push %r12
push %r13
push %r14
push %r15
/* Read HUF_DecompressAsmArgs* args from %rax */
movq %rdi, %rax
movq 0(%rax), %ip0
movq 8(%rax), %ip1
movq 16(%rax), %ip2
movq 24(%rax), %ip3
movq 32(%rax), %op0
movq 40(%rax), %op1
movq 48(%rax), %op2
movq 56(%rax), %op3
movq 64(%rax), %bits0
movq 72(%rax), %bits1
movq 80(%rax), %bits2
movq 88(%rax), %bits3
movq 96(%rax), %dtable
push %rax /* argument */
push 104(%rax) /* ilimit */
push 112(%rax) /* oend */
push %olimit /* olimit space */
subq $24, %rsp
.L_4X1_compute_olimit:
/* Computes how many iterations we can do safely
* %r15, %rax may be clobbered
* rbx, rdx must be saved
* op3 & ip0 mustn't be clobbered
*/
movq %rbx, 0(%rsp)
movq %rdx, 8(%rsp)
movq 32(%rsp), %rax /* rax = oend */
subq %op3, %rax /* rax = oend - op3 */
/* r15 = (oend - op3) / 5 */
movabsq $-3689348814741910323, %rdx
mulq %rdx
movq %rdx, %r15
shrq $2, %r15
movq %ip0, %rax /* rax = ip0 */
movq 40(%rsp), %rdx /* rdx = ilimit */
subq %rdx, %rax /* rax = ip0 - ilimit */
movq %rax, %rbx /* rbx = ip0 - ilimit */
/* rdx = (ip0 - ilimit) / 7 */
movabsq $2635249153387078803, %rdx
mulq %rdx
subq %rdx, %rbx
shrq %rbx
addq %rbx, %rdx
shrq $2, %rdx
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
/* r15 = r15 * 5 */
leaq (%r15, %r15, 4), %r15
/* olimit = op3 + r15 */
addq %op3, %olimit
movq 8(%rsp), %rdx
movq 0(%rsp), %rbx
/* If (op3 + 20 > olimit) */
movq %op3, %rax /* rax = op3 */
addq $20, %rax /* rax = op3 + 20 */
cmpq %rax, %olimit /* op3 + 20 > olimit */
jb .L_4X1_exit
/* If (ip1 < ip0) go to exit */
cmpq %ip0, %ip1
jb .L_4X1_exit
/* If (ip2 < ip1) go to exit */
cmpq %ip1, %ip2
jb .L_4X1_exit
/* If (ip3 < ip2) go to exit */
cmpq %ip2, %ip3
jb .L_4X1_exit
/* Reads top 11 bits from bits[n]
* Loads dt[bits[n]] into var[n]
*/
#define GET_NEXT_DELT(n) \
movq $53, %var##n; \
shrxq %var##n, %bits##n, %var##n; \
movzwl (%dtable,%var##n,2),%vard##n
/* var[n] must contain the DTable entry computed with GET_NEXT_DELT
* Moves var[n] to %rax
* bits[n] <<= var[n] & 63
* op[n][idx] = %rax >> 8
* %ah is a way to access bits [8, 16) of %rax
*/
#define DECODE_FROM_DELT(n, idx) \
movq %var##n, %rax; \
shlxq %var##n, %bits##n, %bits##n; \
movb %ah, idx(%op##n)
/* Assumes GET_NEXT_DELT has been called.
* Calls DECODE_FROM_DELT then GET_NEXT_DELT if n < 4
*/
#define DECODE(n, idx) \
DECODE_FROM_DELT(n, idx); \
IF_NOT_4(idx, GET_NEXT_DELT(n))
/* // ctz & nbBytes is stored in bits[n]
* // nbBits is stored in %rax
* ctz = CTZ[bits[n]]
* nbBits = ctz & 7
* nbBytes = ctz >> 3
* op[n] += 5
* ip[n] -= nbBytes
* // Note: x86-64 is little-endian ==> no bswap
* bits[n] = MEM_readST(ip[n]) | 1
* bits[n] <<= nbBits
*/
#define RELOAD_BITS(n) \
bsfq %bits##n, %bits##n; \
movq %bits##n, %rax; \
andq $7, %rax; \
shrq $3, %bits##n; \
leaq 5(%op##n), %op##n; \
subq %bits##n, %ip##n; \
movq (%ip##n), %bits##n; \
orq $1, %bits##n; \
shlx %rax, %bits##n, %bits##n
/* Store clobbered variables on the stack */
movq %olimit, 24(%rsp)
movq %ip1, 0(%rsp)
movq %ip2, 8(%rsp)
movq %ip3, 16(%rsp)
/* Call GET_NEXT_DELT for each stream */
FOR_EACH_STREAM(GET_NEXT_DELT)
.p2align 6
.L_4X1_loop_body:
/* Decode 5 symbols in each of the 4 streams (20 total)
* Must have called GET_NEXT_DELT for each stream
*/
FOR_EACH_STREAM_WITH_INDEX(DECODE, 0)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 1)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 2)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 3)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 4)
/* Load ip[1,2,3] from stack (var[] aliases them)
* ip[] is needed for RELOAD_BITS
* Each will be stored back to the stack after RELOAD
*/
movq 0(%rsp), %ip1
movq 8(%rsp), %ip2
movq 16(%rsp), %ip3
/* Reload each stream & fetch the next table entry
* to prepare for the next iteration
*/
RELOAD_BITS(0)
GET_NEXT_DELT(0)
RELOAD_BITS(1)
movq %ip1, 0(%rsp)
GET_NEXT_DELT(1)
RELOAD_BITS(2)
movq %ip2, 8(%rsp)
GET_NEXT_DELT(2)
RELOAD_BITS(3)
movq %ip3, 16(%rsp)
GET_NEXT_DELT(3)
/* If op3 < olimit: continue the loop */
cmp %op3, 24(%rsp)
ja .L_4X1_loop_body
/* Reload ip[1,2,3] from stack */
movq 0(%rsp), %ip1
movq 8(%rsp), %ip2
movq 16(%rsp), %ip3
/* Re-compute olimit */
jmp .L_4X1_compute_olimit
#undef GET_NEXT_DELT
#undef DECODE_FROM_DELT
#undef DECODE
#undef RELOAD_BITS
.L_4X1_exit:
addq $24, %rsp
/* Restore stack (oend & olimit) */
pop %rax /* olimit */
pop %rax /* oend */
pop %rax /* ilimit */
pop %rax /* arg */
/* Save ip / op / bits */
movq %ip0, 0(%rax)
movq %ip1, 8(%rax)
movq %ip2, 16(%rax)
movq %ip3, 24(%rax)
movq %op0, 32(%rax)
movq %op1, 40(%rax)
movq %op2, 48(%rax)
movq %op3, 56(%rax)
movq %bits0, 64(%rax)
movq %bits1, 72(%rax)
movq %bits2, 80(%rax)
movq %bits3, 88(%rax)
/* Restore registers */
pop %r15
pop %r14
pop %r13
pop %r12
pop %r11
pop %r10
pop %r9
pop %r8
pop %rdi
pop %rsi
pop %rbp
pop %rdx
pop %rcx
pop %rbx
pop %rax
ret
_HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop:
HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop:
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
push %rcx
push %rdx
push %rbp
push %rsi
push %rdi
push %r8
push %r9
push %r10
push %r11
push %r12
push %r13
push %r14
push %r15
movq %rdi, %rax
movq 0(%rax), %ip0
movq 8(%rax), %ip1
movq 16(%rax), %ip2
movq 24(%rax), %ip3
movq 32(%rax), %op0
movq 40(%rax), %op1
movq 48(%rax), %op2
movq 56(%rax), %op3
movq 64(%rax), %bits0
movq 72(%rax), %bits1
movq 80(%rax), %bits2
movq 88(%rax), %bits3
movq 96(%rax), %dtable
push %rax /* argument */
push %rax /* olimit */
push 104(%rax) /* ilimit */
movq 112(%rax), %rax
push %rax /* oend3 */
movq %op3, %rax
push %rax /* oend2 */
movq %op2, %rax
push %rax /* oend1 */
movq %op1, %rax
push %rax /* oend0 */
/* Scratch space */
subq $8, %rsp
.L_4X2_compute_olimit:
/* Computes how many iterations we can do safely
* %r15, %rax may be clobbered
* rdx must be saved
* op[1,2,3,4] & ip0 mustn't be clobbered
*/
movq %rdx, 0(%rsp)
/* We can consume up to 7 input bytes each iteration. */
movq %ip0, %rax /* rax = ip0 */
movq 40(%rsp), %rdx /* rdx = ilimit */
subq %rdx, %rax /* rax = ip0 - ilimit */
movq %rax, %r15 /* r15 = ip0 - ilimit */
/* rdx = rax / 7 */
movabsq $2635249153387078803, %rdx
mulq %rdx
subq %rdx, %r15
shrq %r15
addq %r15, %rdx
shrq $2, %rdx
/* r15 = (ip0 - ilimit) / 7 */
movq %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 8(%rsp), %rax /* rax = oend0 */
subq %op0, %rax /* rax = oend0 - op0 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 16(%rsp), %rax /* rax = oend1 */
subq %op1, %rax /* rax = oend1 - op1 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 24(%rsp), %rax /* rax = oend2 */
subq %op2, %rax /* rax = oend2 - op2 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 32(%rsp), %rax /* rax = oend3 */
subq %op3, %rax /* rax = oend3 - op3 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
/* olimit = op3 + 5 * r15 */
movq %r15, %rax
leaq (%op3, %rax, 4), %olimit
addq %rax, %olimit
movq 0(%rsp), %rdx
/* If (op3 + 10 > olimit) */
movq %op3, %rax /* rax = op3 */
addq $10, %rax /* rax = op3 + 10 */
cmpq %rax, %olimit /* op3 + 10 > olimit */
jb .L_4X2_exit
/* If (ip1 < ip0) go to exit */
cmpq %ip0, %ip1
jb .L_4X2_exit
/* If (ip2 < ip1) go to exit */
cmpq %ip1, %ip2
jb .L_4X2_exit
/* If (ip3 < ip2) go to exit */
cmpq %ip2, %ip3
jb .L_4X2_exit
#define DECODE(n, idx) \
movq %bits##n, %rax; \
shrq $53, %rax; \
movzwl 0(%dtable,%rax,4),%r8d; \
movzbl 2(%dtable,%rax,4),%r15d; \
movzbl 3(%dtable,%rax,4),%eax; \
movw %r8w, (%op##n); \
shlxq %r15, %bits##n, %bits##n; \
addq %rax, %op##n
#define RELOAD_BITS(n) \
bsfq %bits##n, %bits##n; \
movq %bits##n, %rax; \
shrq $3, %bits##n; \
andq $7, %rax; \
subq %bits##n, %ip##n; \
movq (%ip##n), %bits##n; \
orq $1, %bits##n; \
shlxq %rax, %bits##n, %bits##n
movq %olimit, 48(%rsp)
.p2align 6
.L_4X2_loop_body:
/* We clobber r8, so store it on the stack */
movq %r8, 0(%rsp)
/* Decode 5 symbols from each of the 4 streams (20 symbols total). */
FOR_EACH_STREAM_WITH_INDEX(DECODE, 0)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 1)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 2)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 3)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 4)
/* Reload r8 */
movq 0(%rsp), %r8
FOR_EACH_STREAM(RELOAD_BITS)
cmp %op3, 48(%rsp)
ja .L_4X2_loop_body
jmp .L_4X2_compute_olimit
#undef DECODE
#undef RELOAD_BITS
.L_4X2_exit:
addq $8, %rsp
/* Restore stack (oend & olimit) */
pop %rax /* oend0 */
pop %rax /* oend1 */
pop %rax /* oend2 */
pop %rax /* oend3 */
pop %rax /* ilimit */
pop %rax /* olimit */
pop %rax /* arg */
/* Save ip / op / bits */
movq %ip0, 0(%rax)
movq %ip1, 8(%rax)
movq %ip2, 16(%rax)
movq %ip3, 24(%rax)
movq %op0, 32(%rax)
movq %op1, 40(%rax)
movq %op2, 48(%rax)
movq %op3, 56(%rax)
movq %bits0, 64(%rax)
movq %bits1, 72(%rax)
movq %bits2, 80(%rax)
movq %bits3, 88(%rax)
/* Restore registers */
pop %r15
pop %r14
pop %r13
pop %r12
pop %r11
pop %r10
pop %r9
pop %r8
pop %rdi
pop %rsi
pop %rbp
pop %rdx
pop %rcx
pop %rbx
pop %rax
ret
#endif
lib/decompress/zstd_decompress.c
+85 -22
View File
@@ -56,7 +56,6 @@
* Dependencies
*********************************************************/
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
#include "../common/cpu.h" /* bmi2 */
#include "../common/mem.h" /* low level memory routines */
#define FSE_STATIC_LINKING_ONLY
#include "../common/fse.h"
@@ -177,12 +176,15 @@ static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet,
static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
DEBUGLOG(4, "Allocating new hash set");
if (!ret)
return NULL;
ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
ret->ddictPtrCount = 0;
if (!ret || !ret->ddictPtrTable) {
if (!ret->ddictPtrTable) {
ZSTD_customFree(ret, customMem);
return NULL;
}
ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
ret->ddictPtrCount = 0;
return ret;
}
@@ -255,11 +257,15 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
dctx->inBuffSize = 0;
dctx->outBuffSize = 0;
dctx->streamStage = zdss_init;
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
dctx->legacyContext = NULL;
dctx->previousLegacyVersion = 0;
#endif
dctx->noForwardProgress = 0;
dctx->oversizedDuration = 0;
dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
#if DYNAMIC_BMI2
dctx->bmi2 = ZSTD_cpuSupportsBmi2();
#endif
dctx->ddictSet = NULL;
ZSTD_DCtx_resetParameters(dctx);
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
@@ -280,8 +286,7 @@ ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
return dctx;
}
ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
{
static ZSTD_DCtx* ZSTD_createDCtx_internal(ZSTD_customMem customMem) {
if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
{ ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
@@ -292,10 +297,15 @@ ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
}
}
ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
{
return ZSTD_createDCtx_internal(customMem);
}
ZSTD_DCtx* ZSTD_createDCtx(void)
{
DEBUGLOG(3, "ZSTD_createDCtx");
return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
}
static void ZSTD_clearDict(ZSTD_DCtx* dctx)
@@ -380,6 +390,19 @@ unsigned ZSTD_isFrame(const void* buffer, size_t size)
return 0;
}
/*! ZSTD_isSkippableFrame() :
* Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
* Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
*/
unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size)
{
if (size < ZSTD_FRAMEIDSIZE) return 0;
{ U32 const magic = MEM_readLE32(buffer);
if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
}
return 0;
}
/** ZSTD_frameHeaderSize_internal() :
* srcSize must be large enough to reach header size fields.
* note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
@@ -466,7 +489,9 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
}
switch(dictIDSizeCode)
{
default: assert(0); /* impossible */
default:
assert(0); /* impossible */
ZSTD_FALLTHROUGH;
case 0 : break;
case 1 : dictID = ip[pos]; pos++; break;
case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
@@ -474,7 +499,9 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
}
switch(fcsID)
{
default: assert(0); /* impossible */
default:
assert(0); /* impossible */
ZSTD_FALLTHROUGH;
case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
@@ -503,7 +530,6 @@ size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t src
return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
}
/** ZSTD_getFrameContentSize() :
* compatible with legacy mode
* @return : decompressed size of the single frame pointed to be `src` if known, otherwise
@@ -544,6 +570,37 @@ static size_t readSkippableFrameSize(void const* src, size_t srcSize)
}
}
/*! ZSTD_readSkippableFrame() :
* Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
*
* The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
* i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START. This can be NULL if the caller is not interested
* in the magicVariant.
*
* Returns an error if destination buffer is not large enough, or if the frame is not skippable.
*
* @return : number of bytes written or a ZSTD error.
*/
ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
const void* src, size_t srcSize)
{
U32 const magicNumber = MEM_readLE32(src);
size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
/* check input validity */
RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
/* deliver payload */
if (skippableContentSize > 0 && dst != NULL)
ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
if (magicVariant != NULL)
*magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
return skippableContentSize;
}
/** ZSTD_findDecompressedSize() :
* compatible with legacy mode
* `srcSize` must be the exact length of some number of ZSTD compressed and/or
@@ -858,7 +915,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
switch(blockProperties.blockType)
{
case bt_compressed:
decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1, not_streaming);
break;
case bt_raw :
decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
@@ -1009,7 +1066,7 @@ static ZSTD_DDict const* ZSTD_getDDict(ZSTD_DCtx* dctx)
switch (dctx->dictUses) {
default:
assert(0 /* Impossible */);
/* fall-through */
ZSTD_FALLTHROUGH;
case ZSTD_dont_use:
ZSTD_clearDict(dctx);
return NULL;
@@ -1031,7 +1088,7 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
{
#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
size_t regenSize;
ZSTD_DCtx* const dctx = ZSTD_createDCtx();
ZSTD_DCtx* const dctx = ZSTD_createDCtx_internal(ZSTD_defaultCMem);
RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
ZSTD_freeDCtx(dctx);
@@ -1065,7 +1122,7 @@ static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t
return dctx->expected;
if (dctx->bType != bt_raw)
return dctx->expected;
return MIN(MAX(inputSize, 1), dctx->expected);
return BOUNDED(1, inputSize, dctx->expected);
}
ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
@@ -1073,7 +1130,9 @@ ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
{
default: /* should not happen */
assert(0);
ZSTD_FALLTHROUGH;
case ZSTDds_getFrameHeaderSize:
ZSTD_FALLTHROUGH;
case ZSTDds_decodeFrameHeader:
return ZSTDnit_frameHeader;
case ZSTDds_decodeBlockHeader:
@@ -1085,6 +1144,7 @@ ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
case ZSTDds_checkChecksum:
return ZSTDnit_checksum;
case ZSTDds_decodeSkippableHeader:
ZSTD_FALLTHROUGH;
case ZSTDds_skipFrame:
return ZSTDnit_skippableFrame;
}
@@ -1168,7 +1228,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
{
case bt_compressed:
DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1, is_streaming);
dctx->expected = 0; /* Streaming not supported */
break;
case bt_raw :
@@ -1493,7 +1553,7 @@ size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
ZSTD_DStream* ZSTD_createDStream(void)
{
DEBUGLOG(3, "ZSTD_createDStream");
return ZSTD_createDStream_advanced(ZSTD_defaultCMem);
return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
}
ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
@@ -1503,7 +1563,7 @@ ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
{
return ZSTD_createDCtx_advanced(customMem);
return ZSTD_createDCtx_internal(customMem);
}
size_t ZSTD_freeDStream(ZSTD_DStream* zds)
@@ -1763,7 +1823,8 @@ size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
{
size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
/* space is needed to store the litbuffer after the output of a given block without stomping the extDict of a previous run, as well as to cover both windows against wildcopy*/
unsigned long long const neededRBSize = windowSize + blockSize + ZSTD_BLOCKSIZE_MAX + (WILDCOPY_OVERLENGTH * 2);
unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
size_t const minRBSize = (size_t) neededSize;
RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
@@ -1897,10 +1958,12 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
DEBUGLOG(5, "stage zdss_init => transparent reset ");
zds->streamStage = zdss_loadHeader;
zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
zds->legacyVersion = 0;
#endif
zds->hostageByte = 0;
zds->expectedOutBuffer = *output;
/* fall-through */
ZSTD_FALLTHROUGH;
case zdss_loadHeader :
DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
@@ -2038,7 +2101,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
zds->outBuffSize = neededOutBuffSize;
} } }
zds->streamStage = zdss_read;
/* fall-through */
ZSTD_FALLTHROUGH;
case zdss_read:
DEBUGLOG(5, "stage zdss_read");
@@ -2057,7 +2120,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
} }
if (ip==iend) { someMoreWork = 0; break; } /* no more input */
zds->streamStage = zdss_load;
/* fall-through */
ZSTD_FALLTHROUGH;
case zdss_load:
{ size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
lib/decompress/zstd_decompress_block.c
+737 -213
View File
@@ -69,15 +69,56 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
}
}
/* Allocate buffer for literals, either overlapping current dst, or split between dst and litExtraBuffer, or stored entirely within litExtraBuffer */
static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
{
if (streaming == not_streaming && dstCapacity > ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH)
{
/* room for litbuffer to fit without read faulting */
dctx->litBuffer = (BYTE*)dst + ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH;
dctx->litBufferEnd = dctx->litBuffer + litSize;
dctx->litBufferLocation = ZSTD_in_dst;
}
else if (litSize > ZSTD_LITBUFFEREXTRASIZE)
{
/* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
if (splitImmediately) {
/* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
}
else {
/* initially this will be stored entirely in dst during huffman decoding, it will partially shifted to litExtraBuffer after */
dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
}
dctx->litBufferLocation = ZSTD_split;
}
else
{
/* fits entirely within litExtraBuffer, so no split is necessary */
dctx->litBuffer = dctx->litExtraBuffer;
dctx->litBufferEnd = dctx->litBuffer + litSize;
dctx->litBufferLocation = ZSTD_not_in_dst;
}
}
/* Hidden declaration for fullbench */
size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
const void* src, size_t srcSize);
const void* src, size_t srcSize,
void* dst, size_t dstCapacity, const streaming_operation streaming);
/*! ZSTD_decodeLiteralsBlock() :
* Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
* in the dstBuffer. If there is room to do so, it will be stored in full in the excess dst space after where the current
* block will be output. Otherwise it will be stored at the end of the current dst blockspace, with a small portion being
* stored in dctx->litExtraBuffer to help keep it "ahead" of the current output write.
*
* @return : nb of bytes read from src (< srcSize )
* note : symbol not declared but exposed for fullbench */
size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
const void* src, size_t srcSize) /* note : srcSize < BLOCKSIZE */
const void* src, size_t srcSize, /* note : srcSize < BLOCKSIZE */
void* dst, size_t dstCapacity, const streaming_operation streaming)
{
DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
@@ -90,7 +131,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
case set_repeat:
DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
/* fall-through */
ZSTD_FALLTHROUGH;
case set_compressed:
RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
@@ -99,6 +140,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
U32 const lhlCode = (istart[0] >> 2) & 3;
U32 const lhc = MEM_readLE32(istart);
size_t hufSuccess;
size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
switch(lhlCode)
{
case 0: case 1: default: /* note : default is impossible, since lhlCode into [0..3] */
@@ -121,8 +163,11 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
break;
}
RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
/* prefetch huffman table if cold */
if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
@@ -133,11 +178,11 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
if (singleStream) {
hufSuccess = HUF_decompress1X_usingDTable_bmi2(
dctx->litBuffer, litSize, istart+lhSize, litCSize,
dctx->HUFptr, dctx->bmi2);
dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
} else {
hufSuccess = HUF_decompress4X_usingDTable_bmi2(
dctx->litBuffer, litSize, istart+lhSize, litCSize,
dctx->HUFptr, dctx->bmi2);
dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
}
} else {
if (singleStream) {
@@ -150,15 +195,22 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
dctx->entropy.hufTable, dctx->litBuffer, litSize,
istart+lhSize, litCSize, dctx->workspace,
sizeof(dctx->workspace), dctx->bmi2);
sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
#endif
} else {
hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
dctx->entropy.hufTable, dctx->litBuffer, litSize,
istart+lhSize, litCSize, dctx->workspace,
sizeof(dctx->workspace), dctx->bmi2);
sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
}
}
if (dctx->litBufferLocation == ZSTD_split)
{
ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
}
RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
@@ -166,13 +218,13 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
dctx->litSize = litSize;
dctx->litEntropy = 1;
if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
return litCSize + lhSize;
}
case set_basic:
{ size_t litSize, lhSize;
U32 const lhlCode = ((istart[0]) >> 2) & 3;
size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
switch(lhlCode)
{
case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */
@@ -189,23 +241,36 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
break;
}
RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) { /* risk reading beyond src buffer with wildcopy */
RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
ZSTD_memcpy(dctx->litBuffer, istart+lhSize, litSize);
if (dctx->litBufferLocation == ZSTD_split)
{
ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize - ZSTD_LITBUFFEREXTRASIZE);
ZSTD_memcpy(dctx->litExtraBuffer, istart + lhSize + litSize - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
}
else
{
ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize);
}
dctx->litPtr = dctx->litBuffer;
dctx->litSize = litSize;
ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
return lhSize+litSize;
}
/* direct reference into compressed stream */
dctx->litPtr = istart+lhSize;
dctx->litSize = litSize;
dctx->litBufferEnd = dctx->litPtr + litSize;
dctx->litBufferLocation = ZSTD_not_in_dst;
return lhSize+litSize;
}
case set_rle:
{ U32 const lhlCode = ((istart[0]) >> 2) & 3;
size_t litSize, lhSize;
size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
switch(lhlCode)
{
case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */
@@ -222,8 +287,19 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
break;
}
RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
if (dctx->litBufferLocation == ZSTD_split)
{
ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize - ZSTD_LITBUFFEREXTRASIZE);
ZSTD_memset(dctx->litExtraBuffer, istart[lhSize], ZSTD_LITBUFFEREXTRASIZE);
}
else
{
ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize);
}
dctx->litPtr = dctx->litBuffer;
dctx->litSize = litSize;
return lhSize+1;
@@ -495,7 +571,7 @@ static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
}
#if DYNAMIC_BMI2
TARGET_ATTRIBUTE("bmi2") static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
BMI2_TARGET_ATTRIBUTE static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog, void* wksp, size_t wkspSize)
@@ -620,7 +696,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
LL_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
ZSTD_DCtx_get_bmi2(dctx));
RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += llhSize;
}
@@ -632,7 +708,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
OF_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
ZSTD_DCtx_get_bmi2(dctx));
RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += ofhSize;
}
@@ -644,7 +720,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
ML_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
ZSTD_DCtx_get_bmi2(dctx));
RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += mlhSize;
}
@@ -713,7 +789,7 @@ HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
* - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
* The src buffer must be before the dst buffer.
*/
static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
static void ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
ptrdiff_t const diff = op - ip;
BYTE* const oend = op + length;
@@ -729,6 +805,7 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
/* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
assert(length >= 8);
ZSTD_overlapCopy8(&op, &ip, diff);
length -= 8;
assert(op - ip >= 8);
assert(op <= oend);
}
@@ -743,12 +820,35 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
assert(oend > oend_w);
ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
ip += oend_w - op;
op = oend_w;
op += oend_w - op;
}
/* Handle the leftovers. */
while (op < oend) *op++ = *ip++;
}
/* ZSTD_safecopyDstBeforeSrc():
* This version allows overlap with dst before src, or handles the non-overlap case with dst after src
* Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length) {
ptrdiff_t const diff = op - ip;
BYTE* const oend = op + length;
if (length < 8 || diff > -8) {
/* Handle short lengths, close overlaps, and dst not before src. */
while (op < oend) *op++ = *ip++;
return;
}
if (op <= oend - WILDCOPY_OVERLENGTH && diff < -WILDCOPY_VECLEN) {
ZSTD_wildcopy(op, ip, oend - WILDCOPY_OVERLENGTH - op, ZSTD_no_overlap);
ip += oend - WILDCOPY_OVERLENGTH - op;
op += oend - WILDCOPY_OVERLENGTH - op;
}
/* Handle the leftovers. */
while (op < oend) *op++ = *ip++;
}
/* ZSTD_execSequenceEnd():
* This version handles cases that are near the end of the output buffer. It requires
* more careful checks to make sure there is no overflow. By separating out these hard
@@ -759,9 +859,9 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
*/
FORCE_NOINLINE
size_t ZSTD_execSequenceEnd(BYTE* op,
BYTE* const oend, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
BYTE* const oend, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
{
BYTE* const oLitEnd = op + sequence.litLength;
size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -784,27 +884,76 @@ size_t ZSTD_execSequenceEnd(BYTE* op,
if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
/* offset beyond prefix */
RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
match = dictEnd - (prefixStart-match);
match = dictEnd - (prefixStart - match);
if (match + sequence.matchLength <= dictEnd) {
ZSTD_memmove(oLitEnd, match, sequence.matchLength);
return sequenceLength;
}
/* span extDict & currentPrefixSegment */
{ size_t const length1 = dictEnd - match;
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
} }
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
}
}
ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
return sequenceLength;
}
/* ZSTD_execSequenceEndSplitLitBuffer():
* This version is intended to be used during instances where the litBuffer is still split. It is kept separate to avoid performance impact for the good case.
*/
FORCE_NOINLINE
size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
{
BYTE* const oLitEnd = op + sequence.litLength;
size_t const sequenceLength = sequence.litLength + sequence.matchLength;
const BYTE* const iLitEnd = *litPtr + sequence.litLength;
const BYTE* match = oLitEnd - sequence.offset;
/* bounds checks : careful of address space overflow in 32-bit mode */
RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
assert(op < op + sequenceLength);
assert(oLitEnd < op + sequenceLength);
/* copy literals */
RETURN_ERROR_IF(op > *litPtr && op < *litPtr + sequence.litLength, dstSize_tooSmall, "output should not catch up to and overwrite literal buffer");
ZSTD_safecopyDstBeforeSrc(op, *litPtr, sequence.litLength);
op = oLitEnd;
*litPtr = iLitEnd;
/* copy Match */
if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
/* offset beyond prefix */
RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
match = dictEnd - (prefixStart - match);
if (match + sequence.matchLength <= dictEnd) {
ZSTD_memmove(oLitEnd, match, sequence.matchLength);
return sequenceLength;
}
/* span extDict & currentPrefixSegment */
{ size_t const length1 = dictEnd - match;
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
}
}
ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
return sequenceLength;
}
HINT_INLINE
size_t ZSTD_execSequence(BYTE* op,
BYTE* const oend, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
BYTE* const oend, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
{
BYTE* const oLitEnd = op + sequence.litLength;
size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -813,6 +962,98 @@ size_t ZSTD_execSequence(BYTE* op,
const BYTE* const iLitEnd = *litPtr + sequence.litLength;
const BYTE* match = oLitEnd - sequence.offset;
assert(op != NULL /* Precondition */);
assert(oend_w < oend /* No underflow */);
/* Handle edge cases in a slow path:
* - Read beyond end of literals
* - Match end is within WILDCOPY_OVERLIMIT of oend
* - 32-bit mode and the match length overflows
*/
if (UNLIKELY(
iLitEnd > litLimit ||
oMatchEnd > oend_w ||
(MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
/* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
assert(op <= oLitEnd /* No overflow */);
assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
assert(oMatchEnd <= oend /* No underflow */);
assert(iLitEnd <= litLimit /* Literal length is in bounds */);
assert(oLitEnd <= oend_w /* Can wildcopy literals */);
assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
/* Copy Literals:
* Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
* We likely don't need the full 32-byte wildcopy.
*/
assert(WILDCOPY_OVERLENGTH >= 16);
ZSTD_copy16(op, (*litPtr));
if (UNLIKELY(sequence.litLength > 16)) {
ZSTD_wildcopy(op + 16, (*litPtr) + 16, sequence.litLength - 16, ZSTD_no_overlap);
}
op = oLitEnd;
*litPtr = iLitEnd; /* update for next sequence */
/* Copy Match */
if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
/* offset beyond prefix -> go into extDict */
RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
match = dictEnd + (match - prefixStart);
if (match + sequence.matchLength <= dictEnd) {
ZSTD_memmove(oLitEnd, match, sequence.matchLength);
return sequenceLength;
}
/* span extDict & currentPrefixSegment */
{ size_t const length1 = dictEnd - match;
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
}
}
/* Match within prefix of 1 or more bytes */
assert(op <= oMatchEnd);
assert(oMatchEnd <= oend_w);
assert(match >= prefixStart);
assert(sequence.matchLength >= 1);
/* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
* without overlap checking.
*/
if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
/* We bet on a full wildcopy for matches, since we expect matches to be
* longer than literals (in general). In silesia, ~10% of matches are longer
* than 16 bytes.
*/
ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
return sequenceLength;
}
assert(sequence.offset < WILDCOPY_VECLEN);
/* Copy 8 bytes and spread the offset to be >= 8. */
ZSTD_overlapCopy8(&op, &match, sequence.offset);
/* If the match length is > 8 bytes, then continue with the wildcopy. */
if (sequence.matchLength > 8) {
assert(op < oMatchEnd);
ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength - 8, ZSTD_overlap_src_before_dst);
}
return sequenceLength;
}
HINT_INLINE
size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
const BYTE** litPtr, const BYTE* const litLimit,
const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
{
BYTE* const oLitEnd = op + sequence.litLength;
size_t const sequenceLength = sequence.litLength + sequence.matchLength;
BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */
const BYTE* const iLitEnd = *litPtr + sequence.litLength;
const BYTE* match = oLitEnd - sequence.offset;
assert(op != NULL /* Precondition */);
assert(oend_w < oend /* No underflow */);
/* Handle edge cases in a slow path:
@@ -824,7 +1065,7 @@ size_t ZSTD_execSequence(BYTE* op,
iLitEnd > litLimit ||
oMatchEnd > oend_w ||
(MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
return ZSTD_execSequenceEndSplitLitBuffer(op, oend, oend_w, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
/* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
assert(op <= oLitEnd /* No overflow */);
@@ -892,6 +1133,7 @@ size_t ZSTD_execSequence(BYTE* op,
return sequenceLength;
}
static void
ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
{
@@ -905,20 +1147,10 @@ ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqS
}
FORCE_INLINE_TEMPLATE void
ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD)
ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16 nextState, U32 nbBits)
{
ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state];
U32 const nbBits = DInfo.nbBits;
size_t const lowBits = BIT_readBits(bitD, nbBits);
DStatePtr->state = DInfo.nextState + lowBits;
}
FORCE_INLINE_TEMPLATE void
ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, ZSTD_seqSymbol const DInfo)
{
U32 const nbBits = DInfo.nbBits;
size_t const lowBits = BIT_readBits(bitD, nbBits);
DStatePtr->state = DInfo.nextState + lowBits;
DStatePtr->state = nextState + lowBits;
}
/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
@@ -937,102 +1169,100 @@ FORCE_INLINE_TEMPLATE seq_t
ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
{
seq_t seq;
ZSTD_seqSymbol const llDInfo = seqState->stateLL.table[seqState->stateLL.state];
ZSTD_seqSymbol const mlDInfo = seqState->stateML.table[seqState->stateML.state];
ZSTD_seqSymbol const ofDInfo = seqState->stateOffb.table[seqState->stateOffb.state];
U32 const llBase = llDInfo.baseValue;
U32 const mlBase = mlDInfo.baseValue;
U32 const ofBase = ofDInfo.baseValue;
BYTE const llBits = llDInfo.nbAdditionalBits;
BYTE const mlBits = mlDInfo.nbAdditionalBits;
BYTE const ofBits = ofDInfo.nbAdditionalBits;
BYTE const totalBits = llBits+mlBits+ofBits;
const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
seq.matchLength = mlDInfo->baseValue;
seq.litLength = llDInfo->baseValue;
{ U32 const ofBase = ofDInfo->baseValue;
BYTE const llBits = llDInfo->nbAdditionalBits;
BYTE const mlBits = mlDInfo->nbAdditionalBits;
BYTE const ofBits = ofDInfo->nbAdditionalBits;
BYTE const totalBits = llBits+mlBits+ofBits;
/* sequence */
{ size_t offset;
if (ofBits > 1) {
ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
assert(ofBits <= MaxOff);
if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
BIT_reloadDStream(&seqState->DStream);
if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32); /* to avoid another reload */
} else {
offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
}
seqState->prevOffset[2] = seqState->prevOffset[1];
seqState->prevOffset[1] = seqState->prevOffset[0];
seqState->prevOffset[0] = offset;
} else {
U32 const ll0 = (llBase == 0);
if (LIKELY((ofBits == 0))) {
if (LIKELY(!ll0))
offset = seqState->prevOffset[0];
else {
offset = seqState->prevOffset[1];
seqState->prevOffset[1] = seqState->prevOffset[0];
seqState->prevOffset[0] = offset;
U16 const llNext = llDInfo->nextState;
U16 const mlNext = mlDInfo->nextState;
U16 const ofNext = ofDInfo->nextState;
U32 const llnbBits = llDInfo->nbBits;
U32 const mlnbBits = mlDInfo->nbBits;
U32 const ofnbBits = ofDInfo->nbBits;
/*
* As gcc has better branch and block analyzers, sometimes it is only
* valuable to mark likelyness for clang, it gives around 3-4% of
* performance.
*/
/* sequence */
{ size_t offset;
#if defined(__clang__)
if (LIKELY(ofBits > 1)) {
#else
if (ofBits > 1) {
#endif
ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
assert(ofBits <= MaxOff);
if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
BIT_reloadDStream(&seqState->DStream);
if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32); /* to avoid another reload */
} else {
offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
}
seqState->prevOffset[2] = seqState->prevOffset[1];
seqState->prevOffset[1] = seqState->prevOffset[0];
seqState->prevOffset[0] = offset;
} else {
offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
{ size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */
if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
seqState->prevOffset[1] = seqState->prevOffset[0];
seqState->prevOffset[0] = offset = temp;
} } }
seq.offset = offset;
}
seq.matchLength = mlBase;
if (mlBits > 0)
seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
BIT_reloadDStream(&seqState->DStream);
if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
BIT_reloadDStream(&seqState->DStream);
/* Ensure there are enough bits to read the rest of data in 64-bit mode. */
ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
seq.litLength = llBase;
if (llBits > 0)
seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
if (MEM_32bits())
BIT_reloadDStream(&seqState->DStream);
DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
(U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
/* ANS state update
* gcc-9.0.0 does 2.5% worse with ZSTD_updateFseStateWithDInfo().
* clang-9.2.0 does 7% worse with ZSTD_updateFseState().
* Naturally it seems like ZSTD_updateFseStateWithDInfo() should be the
* better option, so it is the default for other compilers. But, if you
* measure that it is worse, please put up a pull request.
*/
{
#if defined(__GNUC__) && !defined(__clang__)
const int kUseUpdateFseState = 1;
#else
const int kUseUpdateFseState = 0;
#endif
if (kUseUpdateFseState) {
ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream); /* <= 9 bits */
ZSTD_updateFseState(&seqState->stateML, &seqState->DStream); /* <= 9 bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */
ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream); /* <= 8 bits */
} else {
ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llDInfo); /* <= 9 bits */
ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlDInfo); /* <= 9 bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */
ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofDInfo); /* <= 8 bits */
U32 const ll0 = (llDInfo->baseValue == 0);
if (LIKELY((ofBits == 0))) {
offset = seqState->prevOffset[ll0];
seqState->prevOffset[1] = seqState->prevOffset[!ll0];
seqState->prevOffset[0] = offset;
} else {
offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
{ size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */
if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
seqState->prevOffset[1] = seqState->prevOffset[0];
seqState->prevOffset[0] = offset = temp;
} } }
seq.offset = offset;
}
#if defined(__clang__)
if (UNLIKELY(mlBits > 0))
#else
if (mlBits > 0)
#endif
seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
BIT_reloadDStream(&seqState->DStream);
if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
BIT_reloadDStream(&seqState->DStream);
/* Ensure there are enough bits to read the rest of data in 64-bit mode. */
ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
#if defined(__clang__)
if (UNLIKELY(llBits > 0))
#else
if (llBits > 0)
#endif
seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
if (MEM_32bits())
BIT_reloadDStream(&seqState->DStream);
DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
(U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits); /* <= 9 bits */
ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits); /* <= 9 bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */
ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits); /* <= 8 bits */
}
return seq;
@@ -1085,9 +1315,11 @@ MEM_STATIC void ZSTD_assertValidSequence(
#endif
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
FORCE_INLINE_TEMPLATE size_t
DONT_VECTORIZE
ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
const ZSTD_longOffset_e isLongOffset,
@@ -1099,11 +1331,11 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
BYTE* const oend = ostart + maxDstSize;
BYTE* op = ostart;
const BYTE* litPtr = dctx->litPtr;
const BYTE* const litEnd = litPtr + dctx->litSize;
const BYTE* litBufferEnd = dctx->litBufferEnd;
const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
DEBUGLOG(5, "ZSTD_decompressSequences_body");
DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer");
(void)frame;
/* Regen sequences */
@@ -1124,55 +1356,237 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
BIT_DStream_endOfBuffer < BIT_DStream_completed &&
BIT_DStream_completed < BIT_DStream_overflow);
/* decompress without overrunning litPtr begins */
{
seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
/* Align the decompression loop to 32 + 16 bytes.
*
* zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
* speed swings based on the alignment of the decompression loop. This
* performance swing is caused by parts of the decompression loop falling
* out of the DSB. The entire decompression loop should fit in the DSB,
* when it can't we get much worse performance. You can measure if you've
* hit the good case or the bad case with this perf command for some
* compressed file test.zst:
*
* perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
* -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
*
* If you see most cycles served out of the MITE you've hit the bad case.
* If you see most cycles served out of the DSB you've hit the good case.
* If it is pretty even then you may be in an okay case.
*
* This issue has been reproduced on the following CPUs:
* - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
* Use Instruments->Counters to get DSB/MITE cycles.
* I never got performance swings, but I was able to
* go from the good case of mostly DSB to half of the
* cycles served from MITE.
* - Coffeelake: Intel i9-9900k
* - Coffeelake: Intel i7-9700k
*
* I haven't been able to reproduce the instability or DSB misses on any
* of the following CPUS:
* - Haswell
* - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
* - Skylake
*
* Alignment is done for each of the three major decompression loops:
* - ZSTD_decompressSequences_bodySplitLitBuffer - presplit section of the literal buffer
* - ZSTD_decompressSequences_bodySplitLitBuffer - postsplit section of the literal buffer
* - ZSTD_decompressSequences_body
* Alignment choices are made to minimize large swings on bad cases and influence on performance
* from changes external to this code, rather than to overoptimize on the current commit.
*
* If you are seeing performance stability this script can help test.
* It tests on 4 commits in zstd where I saw performance change.
*
* https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
*/
#if defined(__GNUC__) && defined(__x86_64__)
/* Align the decompression loop to 32 + 16 bytes.
*
* zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
* speed swings based on the alignment of the decompression loop. This
* performance swing is caused by parts of the decompression loop falling
* out of the DSB. The entire decompression loop should fit in the DSB,
* when it can't we get much worse performance. You can measure if you've
* hit the good case or the bad case with this perf command for some
* compressed file test.zst:
*
* perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
* -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
*
* If you see most cycles served out of the MITE you've hit the bad case.
* If you see most cycles served out of the DSB you've hit the good case.
* If it is pretty even then you may be in an okay case.
*
* This issue has been reproduced on the following CPUs:
* - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
* Use Instruments->Counters to get DSB/MITE cycles.
* I never got performance swings, but I was able to
* go from the good case of mostly DSB to half of the
* cycles served from MITE.
* - Coffeelake: Intel i9-9900k
* - Coffeelake: Intel i7-9700k
*
* I haven't been able to reproduce the instability or DSB misses on any
* of the following CPUS:
* - Haswell
* - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
* - Skylake
*
* If you are seeing performance stability this script can help test.
* It tests on 4 commits in zstd where I saw performance change.
*
* https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
*/
__asm__(".p2align 6");
__asm__("nop");
__asm__(".p2align 5");
__asm__("nop");
# if __GNUC__ >= 9
/* better for gcc-9 and gcc-10, worse for clang and gcc-8 */
__asm__(".p2align 3");
# else
__asm__(".p2align 4");
__asm__(".p2align 6");
# if __GNUC__ >= 7
/* good for gcc-7, gcc-9, and gcc-11 */
__asm__("nop");
__asm__(".p2align 5");
__asm__("nop");
__asm__(".p2align 4");
# if __GNUC__ == 8 || __GNUC__ == 10
/* good for gcc-8 and gcc-10 */
__asm__("nop");
__asm__(".p2align 3");
# endif
# endif
#endif
/* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
for (; litPtr + sequence.litLength <= dctx->litBufferEnd; ) {
size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
#endif
if (UNLIKELY(ZSTD_isError(oneSeqSize)))
return oneSeqSize;
DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
op += oneSeqSize;
if (UNLIKELY(!--nbSeq))
break;
BIT_reloadDStream(&(seqState.DStream));
sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
}
/* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
if (nbSeq > 0) {
const size_t leftoverLit = dctx->litBufferEnd - litPtr;
if (leftoverLit)
{
RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
sequence.litLength -= leftoverLit;
op += leftoverLit;
}
litPtr = dctx->litExtraBuffer;
litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
dctx->litBufferLocation = ZSTD_not_in_dst;
{
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
#endif
if (UNLIKELY(ZSTD_isError(oneSeqSize)))
return oneSeqSize;
DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
op += oneSeqSize;
if (--nbSeq)
BIT_reloadDStream(&(seqState.DStream));
}
}
}
if (nbSeq > 0) /* there is remaining lit from extra buffer */
{
#if defined(__GNUC__) && defined(__x86_64__)
__asm__(".p2align 6");
__asm__("nop");
# if __GNUC__ != 7
/* worse for gcc-7 better for gcc-8, gcc-9, and gcc-10 and clang */
__asm__(".p2align 4");
__asm__("nop");
__asm__(".p2align 3");
# elif __GNUC__ >= 11
__asm__(".p2align 3");
# else
__asm__(".p2align 5");
__asm__("nop");
__asm__(".p2align 3");
# endif
#endif
for (; ; ) {
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
#endif
if (UNLIKELY(ZSTD_isError(oneSeqSize)))
return oneSeqSize;
DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
op += oneSeqSize;
if (UNLIKELY(!--nbSeq))
break;
BIT_reloadDStream(&(seqState.DStream));
}
}
/* check if reached exact end */
DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
RETURN_ERROR_IF(nbSeq, corruption_detected, "");
RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
/* save reps for next block */
{ U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
}
/* last literal segment */
if (dctx->litBufferLocation == ZSTD_split) /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
{
size_t const lastLLSize = litBufferEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
if (op != NULL) {
ZSTD_memmove(op, litPtr, lastLLSize);
op += lastLLSize;
}
litPtr = dctx->litExtraBuffer;
litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
dctx->litBufferLocation = ZSTD_not_in_dst;
}
{ size_t const lastLLSize = litBufferEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
if (op != NULL) {
ZSTD_memcpy(op, litPtr, lastLLSize);
op += lastLLSize;
}
}
return op-ostart;
}
FORCE_INLINE_TEMPLATE size_t
DONT_VECTORIZE
ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
const ZSTD_longOffset_e isLongOffset,
const int frame)
{
const BYTE* ip = (const BYTE*)seqStart;
const BYTE* const iend = ip + seqSize;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ostart + maxDstSize : dctx->litBuffer;
BYTE* op = ostart;
const BYTE* litPtr = dctx->litPtr;
const BYTE* const litEnd = litPtr + dctx->litSize;
const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
DEBUGLOG(5, "ZSTD_decompressSequences_body");
(void)frame;
/* Regen sequences */
if (nbSeq) {
seqState_t seqState;
dctx->fseEntropy = 1;
{ U32 i; for (i = 0; i < ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
RETURN_ERROR_IF(
ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend - ip)),
corruption_detected, "");
ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
assert(dst != NULL);
ZSTD_STATIC_ASSERT(
BIT_DStream_unfinished < BIT_DStream_completed &&
BIT_DStream_endOfBuffer < BIT_DStream_completed &&
BIT_DStream_completed < BIT_DStream_overflow);
#if defined(__GNUC__) && defined(__x86_64__)
__asm__(".p2align 6");
__asm__("nop");
# if __GNUC__ >= 7
__asm__(".p2align 5");
__asm__("nop");
__asm__(".p2align 3");
# else
__asm__(".p2align 4");
__asm__("nop");
__asm__(".p2align 3");
# endif
#endif
for ( ; ; ) {
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
@@ -1218,6 +1632,16 @@ ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
{
return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
static size_t
ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
const ZSTD_longOffset_e isLongOffset,
const int frame)
{
return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
@@ -1250,10 +1674,10 @@ ZSTD_decompressSequencesLong_body(
const BYTE* ip = (const BYTE*)seqStart;
const BYTE* const iend = ip + seqSize;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + maxDstSize;
BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ostart + maxDstSize;
BYTE* op = ostart;
const BYTE* litPtr = dctx->litPtr;
const BYTE* const litEnd = litPtr + dctx->litSize;
const BYTE* litBufferEnd = dctx->litBufferEnd;
const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
@@ -1289,32 +1713,94 @@ ZSTD_decompressSequencesLong_body(
}
RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
/* decode and decompress */
for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
/* decompress without stomping litBuffer */
for (; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb < nbSeq); seqNb++) {
seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t oneSeqSize;
prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
sequences[seqNb & STORED_SEQS_MASK] = sequence;
op += oneSeqSize;
if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd)
{
/* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
const size_t leftoverLit = dctx->litBufferEnd - litPtr;
if (leftoverLit)
{
RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength -= leftoverLit;
op += leftoverLit;
}
litPtr = dctx->litExtraBuffer;
litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
dctx->litBufferLocation = ZSTD_not_in_dst;
oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
sequences[seqNb & STORED_SEQS_MASK] = sequence;
op += oneSeqSize;
}
else
{
/* lit buffer is either wholly contained in first or second split, or not split at all*/
oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
sequences[seqNb & STORED_SEQS_MASK] = sequence;
op += oneSeqSize;
}
}
RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected, "");
/* finish queue */
seqNb -= seqAdvance;
for ( ; seqNb<nbSeq ; seqNb++) {
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[seqNb&STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
seq_t *sequence = &(sequences[seqNb&STORED_SEQS_MASK]);
if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd)
{
const size_t leftoverLit = dctx->litBufferEnd - litPtr;
if (leftoverLit)
{
RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
sequence->litLength -= leftoverLit;
op += leftoverLit;
}
litPtr = dctx->litExtraBuffer;
litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
dctx->litBufferLocation = ZSTD_not_in_dst;
{
size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
op += oneSeqSize;
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
op += oneSeqSize;
}
}
else
{
size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence->litLength - WILDCOPY_OVERLENGTH, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
op += oneSeqSize;
}
}
/* save reps for next block */
@@ -1322,10 +1808,21 @@ ZSTD_decompressSequencesLong_body(
}
/* last literal segment */
{ size_t const lastLLSize = litEnd - litPtr;
if (dctx->litBufferLocation == ZSTD_split) /* first deplete literal buffer in dst, then copy litExtraBuffer */
{
size_t const lastLLSize = litBufferEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
if (op != NULL) {
ZSTD_memmove(op, litPtr, lastLLSize);
op += lastLLSize;
}
litPtr = dctx->litExtraBuffer;
litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
}
{ size_t const lastLLSize = litBufferEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
if (op != NULL) {
ZSTD_memcpy(op, litPtr, lastLLSize);
ZSTD_memmove(op, litPtr, lastLLSize);
op += lastLLSize;
}
}
@@ -1349,7 +1846,7 @@ ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
#if DYNAMIC_BMI2
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
static TARGET_ATTRIBUTE("bmi2") size_t
static BMI2_TARGET_ATTRIBUTE size_t
DONT_VECTORIZE
ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
@@ -1359,10 +1856,20 @@ ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
{
return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
static BMI2_TARGET_ATTRIBUTE size_t
DONT_VECTORIZE
ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
const ZSTD_longOffset_e isLongOffset,
const int frame)
{
return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
static TARGET_ATTRIBUTE("bmi2") size_t
static BMI2_TARGET_ATTRIBUTE size_t
ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
@@ -1391,11 +1898,25 @@ ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
{
DEBUGLOG(5, "ZSTD_decompressSequences");
#if DYNAMIC_BMI2
if (dctx->bmi2) {
if (ZSTD_DCtx_get_bmi2(dctx)) {
return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif
return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
static size_t
ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
const void* seqStart, size_t seqSize, int nbSeq,
const ZSTD_longOffset_e isLongOffset,
const int frame)
{
DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
#if DYNAMIC_BMI2
if (ZSTD_DCtx_get_bmi2(dctx)) {
return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif
return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
@@ -1415,7 +1936,7 @@ ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
{
DEBUGLOG(5, "ZSTD_decompressSequencesLong");
#if DYNAMIC_BMI2
if (dctx->bmi2) {
if (ZSTD_DCtx_get_bmi2(dctx)) {
return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
}
#endif
@@ -1456,7 +1977,7 @@ ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
size_t
ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize, const int frame)
const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
{ /* blockType == blockCompressed */
const BYTE* ip = (const BYTE*)src;
/* isLongOffset must be true if there are long offsets.
@@ -1471,7 +1992,7 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
/* Decode literals section */
{ size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
{ size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
if (ZSTD_isError(litCSize)) return litCSize;
ip += litCSize;
@@ -1519,7 +2040,10 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
/* else */
return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
if (dctx->litBufferLocation == ZSTD_split)
return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
else
return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
#endif
}
}
@@ -1542,7 +2066,7 @@ size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
{
size_t dSize;
ZSTD_checkContinuity(dctx, dst, dstCapacity);
dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0, not_streaming);
dctx->previousDstEnd = (char*)dst + dSize;
return dSize;
}
lib/decompress/zstd_decompress_block.h
+7 -1
View File
@@ -33,6 +33,12 @@
*/
/* Streaming state is used to inform allocation of the literal buffer */
typedef enum {
not_streaming = 0,
is_streaming = 1
} streaming_operation;
/* ZSTD_decompressBlock_internal() :
* decompress block, starting at `src`,
* into destination buffer `dst`.
@@ -41,7 +47,7 @@
*/
size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize, const int frame);
const void* src, size_t srcSize, const int frame, const streaming_operation streaming);
/* ZSTD_buildFSETable() :
* generate FSE decoding table for one symbol (ll, ml or off)
lib/decompress/zstd_decompress_internal.h
+33 -2
View File
@@ -20,7 +20,7 @@
* Dependencies
*********************************************************/
#include "../common/mem.h" /* BYTE, U16, U32 */
#include "../common/zstd_internal.h" /* ZSTD_seqSymbol */
#include "../common/zstd_internal.h" /* constants : MaxLL, MaxML, MaxOff, LLFSELog, etc. */
@@ -106,6 +106,22 @@ typedef struct {
size_t ddictPtrCount;
} ZSTD_DDictHashSet;
#ifndef ZSTD_DECODER_INTERNAL_BUFFER
# define ZSTD_DECODER_INTERNAL_BUFFER (1 << 16)
#endif
#define ZSTD_LBMIN 64
#define ZSTD_LBMAX (128 << 10)
/* extra buffer, compensates when dst is not large enough to store litBuffer */
#define ZSTD_LITBUFFEREXTRASIZE BOUNDED(ZSTD_LBMIN, ZSTD_DECODER_INTERNAL_BUFFER, ZSTD_LBMAX)
typedef enum {
ZSTD_not_in_dst = 0, /* Stored entirely within litExtraBuffer */
ZSTD_in_dst = 1, /* Stored entirely within dst (in memory after current output write) */
ZSTD_split = 2 /* Split between litExtraBuffer and dst */
} ZSTD_litLocation_e;
struct ZSTD_DCtx_s
{
const ZSTD_seqSymbol* LLTptr;
@@ -136,7 +152,9 @@ struct ZSTD_DCtx_s
size_t litSize;
size_t rleSize;
size_t staticSize;
#if DYNAMIC_BMI2 != 0
int bmi2; /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
#endif
/* dictionary */
ZSTD_DDict* ddictLocal;
@@ -158,16 +176,21 @@ struct ZSTD_DCtx_s
size_t outStart;
size_t outEnd;
size_t lhSize;
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
void* legacyContext;
U32 previousLegacyVersion;
U32 legacyVersion;
#endif
U32 hostageByte;
int noForwardProgress;
ZSTD_bufferMode_e outBufferMode;
ZSTD_outBuffer expectedOutBuffer;
/* workspace */
BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
BYTE* litBuffer;
const BYTE* litBufferEnd;
ZSTD_litLocation_e litBufferLocation;
BYTE litExtraBuffer[ZSTD_LITBUFFEREXTRASIZE + WILDCOPY_OVERLENGTH]; /* literal buffer can be split between storage within dst and within this scratch buffer */
BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
size_t oversizedDuration;
@@ -183,6 +206,14 @@ struct ZSTD_DCtx_s
#endif
}; /* typedef'd to ZSTD_DCtx within "zstd.h" */
MEM_STATIC int ZSTD_DCtx_get_bmi2(const struct ZSTD_DCtx_s *dctx) {
#if DYNAMIC_BMI2 != 0
return dctx->bmi2;
#else
(void)dctx;
return 0;
#endif
}
/*-*******************************************************
* Shared internal functions
lib/dictBuilder/cover.c
+9 -2
View File
@@ -40,6 +40,13 @@
/*-*************************************
* Constants
***************************************/
/**
* There are 32bit indexes used to ref samples, so limit samples size to 4GB
* on 64bit builds.
* For 32bit builds we choose 1 GB.
* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
* contiguous buffer, so 1GB is already a high limit.
*/
#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
#define COVER_DEFAULT_SPLITPOINT 1.0
@@ -47,7 +54,7 @@
* Console display
***************************************/
#ifndef LOCALDISPLAYLEVEL
static int g_displayLevel = 2;
static int g_displayLevel = 0;
#endif
#undef DISPLAY
#define DISPLAY(...) \
@@ -735,7 +742,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
COVER_map_t activeDmers;
parameters.splitPoint = 1.0;
/* Initialize global data */
g_displayLevel = parameters.zParams.notificationLevel;
g_displayLevel = (int)parameters.zParams.notificationLevel;
/* Checks */
if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
DISPLAYLEVEL(1, "Cover parameters incorrect\n");
lib/dictBuilder/fastcover.c
+11 -4
View File
@@ -32,6 +32,13 @@
/*-*************************************
* Constants
***************************************/
/**
* There are 32bit indexes used to ref samples, so limit samples size to 4GB
* on 64bit builds.
* For 32bit builds we choose 1 GB.
* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
* contiguous buffer, so 1GB is already a high limit.
*/
#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
#define FASTCOVER_MAX_F 31
#define FASTCOVER_MAX_ACCEL 10
@@ -44,7 +51,7 @@
* Console display
***************************************/
#ifndef LOCALDISPLAYLEVEL
static int g_displayLevel = 2;
static int g_displayLevel = 0;
#endif
#undef DISPLAY
#define DISPLAY(...) \
@@ -549,7 +556,7 @@ ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
ZDICT_cover_params_t coverParams;
FASTCOVER_accel_t accelParams;
/* Initialize global data */
g_displayLevel = parameters.zParams.notificationLevel;
g_displayLevel = (int)parameters.zParams.notificationLevel;
/* Assign splitPoint and f if not provided */
parameters.splitPoint = 1.0;
parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
@@ -632,7 +639,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
const unsigned shrinkDict = 0;
/* Local variables */
const int displayLevel = parameters->zParams.notificationLevel;
const int displayLevel = (int)parameters->zParams.notificationLevel;
unsigned iteration = 1;
unsigned d;
unsigned k;
@@ -716,7 +723,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
data->parameters.splitPoint = splitPoint;
data->parameters.steps = kSteps;
data->parameters.shrinkDict = shrinkDict;
data->parameters.zParams.notificationLevel = g_displayLevel;
data->parameters.zParams.notificationLevel = (unsigned)g_displayLevel;
/* Check the parameters */
if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
data->ctx->f, accel)) {
lib/dictBuilder/zdict.c
+99 -28
View File
@@ -135,22 +135,32 @@ static unsigned ZDICT_NbCommonBytes (size_t val)
if (MEM_isLittleEndian()) {
if (MEM_64bits()) {
# if defined(_MSC_VER) && defined(_WIN64)
unsigned long r = 0;
_BitScanForward64( &r, (U64)val );
return (unsigned)(r>>3);
if (val != 0) {
unsigned long r;
_BitScanForward64(&r, (U64)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_ctzll((U64)val) >> 3);
return (unsigned)(__builtin_ctzll((U64)val) >> 3);
# else
static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
# endif
} else { /* 32 bits */
# if defined(_MSC_VER)
unsigned long r=0;
_BitScanForward( &r, (U32)val );
return (unsigned)(r>>3);
if (val != 0) {
unsigned long r;
_BitScanForward(&r, (U32)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_ctz((U32)val) >> 3);
return (unsigned)(__builtin_ctz((U32)val) >> 3);
# else
static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
@@ -159,11 +169,16 @@ static unsigned ZDICT_NbCommonBytes (size_t val)
} else { /* Big Endian CPU */
if (MEM_64bits()) {
# if defined(_MSC_VER) && defined(_WIN64)
unsigned long r = 0;
_BitScanReverse64( &r, val );
return (unsigned)(r>>3);
if (val != 0) {
unsigned long r;
_BitScanReverse64(&r, val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_clzll(val) >> 3);
return (unsigned)(__builtin_clzll(val) >> 3);
# else
unsigned r;
const unsigned n32 = sizeof(size_t)*4; /* calculate this way due to compiler complaining in 32-bits mode */
@@ -174,11 +189,16 @@ static unsigned ZDICT_NbCommonBytes (size_t val)
# endif
} else { /* 32 bits */
# if defined(_MSC_VER)
unsigned long r = 0;
_BitScanReverse( &r, (unsigned long)val );
return (unsigned)(r>>3);
if (val != 0) {
unsigned long r;
_BitScanReverse(&r, (unsigned long)val);
return (unsigned)(r >> 3);
} else {
/* Should not reach this code path */
__assume(0);
}
# elif defined(__GNUC__) && (__GNUC__ >= 3)
return (__builtin_clz((U32)val) >> 3);
return (unsigned)(__builtin_clz((U32)val) >> 3);
# else
unsigned r;
if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
@@ -235,7 +255,7 @@ static dictItem ZDICT_analyzePos(
U32 savings[LLIMIT] = {0};
const BYTE* b = (const BYTE*)buffer;
size_t maxLength = LLIMIT;
size_t pos = suffix[start];
size_t pos = (size_t)suffix[start];
U32 end = start;
dictItem solution;
@@ -369,7 +389,7 @@ static dictItem ZDICT_analyzePos(
savings[i] = savings[i-1] + (lengthList[i] * (i-3));
DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f) \n",
(unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / maxLength);
(unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / (double)maxLength);
solution.pos = (U32)pos;
solution.length = (U32)maxLength;
@@ -379,7 +399,7 @@ static dictItem ZDICT_analyzePos(
{ U32 id;
for (id=start; id<end; id++) {
U32 p, pEnd, length;
U32 const testedPos = suffix[id];
U32 const testedPos = (U32)suffix[id];
if (testedPos == pos)
length = solution.length;
else {
@@ -442,7 +462,7 @@ static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const
if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) { /* overlap, existing < new */
/* append */
int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
int const addedLength = (int)eltEnd - (int)(table[u].pos + table[u].length);
table[u].savings += elt.length / 8; /* rough approx bonus */
if (addedLength > 0) { /* otherwise, elt fully included into existing */
table[u].length += addedLength;
@@ -766,6 +786,13 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
pos += fileSizes[u];
}
if (notificationLevel >= 4) {
/* writeStats */
DISPLAYLEVEL(4, "Offset Code Frequencies : \n");
for (u=0; u<=offcodeMax; u++) {
DISPLAYLEVEL(4, "%2u :%7u \n", u, offcodeCount[u]);
} }
/* analyze, build stats, starting with literals */
{ size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
if (HUF_isError(maxNbBits)) {
@@ -872,7 +899,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
#else
/* at this stage, we don't use the result of "most common first offset",
as the impact of statistics is not properly evaluated */
* as the impact of statistics is not properly evaluated */
MEM_writeLE32(dstPtr+0, repStartValue[0]);
MEM_writeLE32(dstPtr+4, repStartValue[1]);
MEM_writeLE32(dstPtr+8, repStartValue[2]);
@@ -888,6 +915,17 @@ _cleanup:
}
/**
* @returns the maximum repcode value
*/
static U32 ZDICT_maxRep(U32 const reps[ZSTD_REP_NUM])
{
U32 maxRep = reps[0];
int r;
for (r = 1; r < ZSTD_REP_NUM; ++r)
maxRep = MAX(maxRep, reps[r]);
return maxRep;
}
size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
const void* customDictContent, size_t dictContentSize,
@@ -899,11 +937,13 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
BYTE header[HBUFFSIZE];
int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
U32 const notificationLevel = params.notificationLevel;
/* The final dictionary content must be at least as large as the largest repcode */
size_t const minContentSize = (size_t)ZDICT_maxRep(repStartValue);
size_t paddingSize;
/* check conditions */
DEBUGLOG(4, "ZDICT_finalizeDictionary");
if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
/* dictionary header */
@@ -927,12 +967,43 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
hSize += eSize;
}
/* copy elements in final buffer ; note : src and dst buffer can overlap */
if (hSize + dictContentSize > dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize;
{ size_t const dictSize = hSize + dictContentSize;
char* dictEnd = (char*)dictBuffer + dictSize;
memmove(dictEnd - dictContentSize, customDictContent, dictContentSize);
memcpy(dictBuffer, header, hSize);
/* Shrink the content size if it doesn't fit in the buffer */
if (hSize + dictContentSize > dictBufferCapacity) {
dictContentSize = dictBufferCapacity - hSize;
}
/* Pad the dictionary content with zeros if it is too small */
if (dictContentSize < minContentSize) {
RETURN_ERROR_IF(hSize + minContentSize > dictBufferCapacity, dstSize_tooSmall,
"dictBufferCapacity too small to fit max repcode");
paddingSize = minContentSize - dictContentSize;
} else {
paddingSize = 0;
}
{
size_t const dictSize = hSize + paddingSize + dictContentSize;
/* The dictionary consists of the header, optional padding, and the content.
* The padding comes before the content because the "best" position in the
* dictionary is the last byte.
*/
BYTE* const outDictHeader = (BYTE*)dictBuffer;
BYTE* const outDictPadding = outDictHeader + hSize;
BYTE* const outDictContent = outDictPadding + paddingSize;
assert(dictSize <= dictBufferCapacity);
assert(outDictContent + dictContentSize == (BYTE*)dictBuffer + dictSize);
/* First copy the customDictContent into its final location.
* `customDictContent` and `dictBuffer` may overlap, so we must
* do this before any other writes into the output buffer.
* Then copy the header & padding into the output buffer.
*/
memmove(outDictContent, customDictContent, dictContentSize);
memcpy(outDictHeader, header, hSize);
memset(outDictPadding, 0, paddingSize);
return dictSize;
}
}
lib/legacy/zstd_v01.c
+1 -2
View File
@@ -343,8 +343,7 @@ FORCE_INLINE unsigned FSE_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r;
_BitScanReverse ( &r, val );
return (unsigned) r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (GCC_VERSION >= 304) /* GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v02.c
+2 -3
View File
@@ -353,9 +353,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BIT_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v03.c
+2 -3
View File
@@ -356,9 +356,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BIT_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v04.c
+2 -3
View File
@@ -627,9 +627,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BIT_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v05.c
+2 -3
View File
@@ -756,9 +756,8 @@ MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BITv05_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v06.c
+2 -3
View File
@@ -860,9 +860,8 @@ MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BITv06_highbit32 ( U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/legacy/zstd_v07.c
+2 -3
View File
@@ -530,9 +530,8 @@ MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, unsigned nbBits);
MEM_STATIC unsigned BITv07_highbit32 (U32 val)
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
_BitScanReverse ( &r, val );
return (unsigned) r;
unsigned long r;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# else /* Software version */
lib/libzstd.mk
+185
View File
@@ -0,0 +1,185 @@
# ################################################################
# Copyright (c) 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.
# ################################################################
##################################################################
# Input Variables
##################################################################
# Zstd lib directory
LIBZSTD ?= ./
# 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
# Assembly support
ZSTD_NO_ASM ?= 0
##################################################################
# libzstd helpers
##################################################################
# Make 4.3 doesn't support '\#' anymore (https://lwn.net/Articles/810071/)
NUM_SYMBOL := \#
# 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_SRC := $(LIBZSTD)/zstd.h
LIBVER_MAJOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MAJOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
LIBVER_MINOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MINOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
LIBVER_PATCH_SCRIPT:=`sed -n '/define ZSTD_VERSION_RELEASE/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
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))
CCVER := $(shell $(CC) --version)
ZSTD_VERSION?= $(LIBVER)
# ZSTD_LIB_MINIFY is a helper variable that
# configures a bunch of other variables to space-optimized defaults.
ZSTD_LIB_MINIFY ?= 0
ifneq ($(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)
LDFLAGS += $(MOREFLAGS)
FLAGS = $(CPPFLAGS) $(CFLAGS) $(LDFLAGS)
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
ZSTD_COMMON_FILES := $(sort $(wildcard $(LIBZSTD)/common/*.c))
ZSTD_COMPRESS_FILES := $(sort $(wildcard $(LIBZSTD)/compress/*.c))
ZSTD_DECOMPRESS_FILES := $(sort $(wildcard $(LIBZSTD)/decompress/*.c))
ZSTD_DICTBUILDER_FILES := $(sort $(wildcard $(LIBZSTD)/dictBuilder/*.c))
ZSTD_DEPRECATED_FILES := $(sort $(wildcard $(LIBZSTD)/deprecated/*.c))
ZSTD_LEGACY_FILES :=
ZSTD_DECOMPRESS_AMD64_ASM_FILES := $(sort $(wildcard $(LIBZSTD)/decompress/*_amd64.S))
ifneq ($(ZSTD_NO_ASM), 0)
CPPFLAGS += -DHUF_DISABLE_ASM
else
# Unconditionally add the ASM files they are disabled by
# macros in the .S file.
ZSTD_DECOMPRESS_FILES += $(ZSTD_DECOMPRESS_AMD64_ASM_FILES)
endif
ifneq ($(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
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 $$?), 0)
ZSTD_LEGACY_FILES += $(shell ls $(LIBZSTD)/legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
endif
endif
CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
UNAME := $(shell uname)
ifndef BUILD_DIR
ifeq ($(UNAME), Darwin)
ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 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
endif
endif # BUILD_DIR
ZSTD_SUBDIR := $(LIBZSTD)/common $(LIBZSTD)/compress $(LIBZSTD)/decompress $(LIBZSTD)/dictBuilder $(LIBZSTD)/legacy $(LIBZSTD)/deprecated
vpath %.c $(ZSTD_SUBDIR)
vpath %.S $(ZSTD_SUBDIR)
lib/modulemap/module.modulemap
+4
View File
@@ -0,0 +1,4 @@
module libzstd [extern_c] {
header "../zstd.h"
export *
}
lib/zdict.h
+4 -4
View File
@@ -46,7 +46,7 @@ extern "C" {
*
* Zstd can use dictionaries to improve compression ratio of small data.
* Traditionally small files don't compress well because there is very little
* repetion in a single sample, since it is small. But, if you are compressing
* repetition in a single sample, since it is small. But, if you are compressing
* many similar files, like a bunch of JSON records that share the same
* structure, you can train a dictionary on ahead of time on some samples of
* these files. Then, zstd can use the dictionary to find repetitions that are
@@ -132,7 +132,7 @@ extern "C" {
*
* # Benchmark levels 1-3 without a dictionary
* zstd -b1e3 -r /path/to/my/files
* # Benchmark levels 1-3 with a dictioanry
* # Benchmark levels 1-3 with a dictionary
* zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary
*
* When should I retrain a dictionary?
@@ -237,7 +237,6 @@ typedef struct {
* is presumed that the most profitable content is at the end of the dictionary,
* since that is the cheapest to reference.
*
* `dictContentSize` must be >= ZDICT_CONTENTSIZE_MIN bytes.
* `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN).
*
* @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`),
@@ -272,8 +271,9 @@ ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
* Use them only in association with static linking.
* ==================================================================================== */
#define ZDICT_CONTENTSIZE_MIN 128
#define ZDICT_DICTSIZE_MIN 256
/* Deprecated: Remove in v1.6.0 */
#define ZDICT_CONTENTSIZE_MIN 128
/*! ZDICT_cover_params_t:
* k and d are the only required parameters.
lib/zstd.h
+67 -37
View File
@@ -74,7 +74,7 @@ extern "C" {
/*------ Version ------*/
#define ZSTD_VERSION_MAJOR 1
#define ZSTD_VERSION_MINOR 5
#define ZSTD_VERSION_RELEASE 0
#define ZSTD_VERSION_RELEASE 1
#define ZSTD_VERSION_NUMBER (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
/*! ZSTD_versionNumber() :
@@ -249,7 +249,7 @@ ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
*
* It's possible to reset all parameters to "default" using ZSTD_CCtx_reset().
*
* This API supercedes all other "advanced" API entry points in the experimental section.
* This API supersedes all other "advanced" API entry points in the experimental section.
* In the future, we expect to remove from experimental API entry points which are redundant with this API.
*/
@@ -419,7 +419,7 @@ typedef enum {
* ZSTD_c_stableOutBuffer
* ZSTD_c_blockDelimiters
* ZSTD_c_validateSequences
* ZSTD_c_splitBlocks
* ZSTD_c_useBlockSplitter
* ZSTD_c_useRowMatchFinder
* Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
* note : never ever use experimentalParam? names directly;
@@ -934,7 +934,7 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
* Advanced dictionary and prefix API (Requires v1.4.0+)
*
* This API allows dictionaries to be used with ZSTD_compress2(),
* ZSTD_compressStream2(), and ZSTD_decompress(). Dictionaries are sticky, and
* ZSTD_compressStream2(), and ZSTD_decompressDCtx(). Dictionaries are sticky, and
* only reset with the context is reset with ZSTD_reset_parameters or
* ZSTD_reset_session_and_parameters. Prefixes are single-use.
******************************************************************************/
@@ -1170,9 +1170,6 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
#define ZSTD_SRCSIZEHINT_MIN 0
#define ZSTD_SRCSIZEHINT_MAX INT_MAX
/* internal */
#define ZSTD_HASHLOG3_MAX 17
/* --- Advanced types --- */
@@ -1315,10 +1312,14 @@ typedef enum {
} ZSTD_literalCompressionMode_e;
typedef enum {
ZSTD_urm_auto = 0, /* Automatically determine whether or not we use row matchfinder */
ZSTD_urm_disableRowMatchFinder = 1, /* Never use row matchfinder */
ZSTD_urm_enableRowMatchFinder = 2 /* Always use row matchfinder when applicable */
} ZSTD_useRowMatchFinderMode_e;
/* Note: This enum controls features which are conditionally beneficial. Zstd typically will make a final
* decision on whether or not to enable the feature (ZSTD_ps_auto), but setting the switch to ZSTD_ps_enable
* or ZSTD_ps_disable allow for a force enable/disable the feature.
*/
ZSTD_ps_auto = 0, /* Let the library automatically determine whether the feature shall be enabled */
ZSTD_ps_enable = 1, /* Force-enable the feature */
ZSTD_ps_disable = 2 /* Do not use the feature */
} ZSTD_paramSwitch_e;
/***************************************
* Frame size functions
@@ -1454,6 +1455,26 @@ ZSTDLIB_STATIC_API size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* ds
ZSTDLIB_STATIC_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
const void* src, size_t srcSize, unsigned magicVariant);
/*! ZSTD_readSkippableFrame() :
* Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
*
* The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
* i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START. This can be NULL if the caller is not interested
* in the magicVariant.
*
* Returns an error if destination buffer is not large enough, or if the frame is not skippable.
*
* @return : number of bytes written or a ZSTD error.
*/
ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
const void* src, size_t srcSize);
/*! ZSTD_isSkippableFrame() :
* Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
*/
ZSTDLIB_API unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
/***************************************
* Memory management
@@ -1581,15 +1602,15 @@ ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_
ZSTD_compressionParameters cParams,
ZSTD_customMem customMem);
/* ! Thread pool :
* These prototypes make it possible to share a thread pool among multiple compression contexts.
* This can limit resources for applications with multiple threads where each one uses
* a threaded compression mode (via ZSTD_c_nbWorkers parameter).
* ZSTD_createThreadPool creates a new thread pool with a given number of threads.
* Note that the lifetime of such pool must exist while being used.
* ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
* to use an internal thread pool).
* ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
/*! Thread pool :
* These prototypes make it possible to share a thread pool among multiple compression contexts.
* This can limit resources for applications with multiple threads where each one uses
* a threaded compression mode (via ZSTD_c_nbWorkers parameter).
* ZSTD_createThreadPool creates a new thread pool with a given number of threads.
* Note that the lifetime of such pool must exist while being used.
* ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
* to use an internal thread pool).
* ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
*/
typedef struct POOL_ctx_s ZSTD_threadPool;
ZSTDLIB_STATIC_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
@@ -1725,9 +1746,15 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
* See the comments on that enum for an explanation of the feature. */
#define ZSTD_c_forceAttachDict ZSTD_c_experimentalParam4
/* Controls how the literals are compressed (default is auto).
* The value must be of type ZSTD_literalCompressionMode_e.
* See ZSTD_literalCompressionMode_e enum definition for details.
/* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never compress literals.
* Set to ZSTD_ps_enable to always compress literals. (Note: uncompressed literals
* may still be emitted if huffman is not beneficial to use.)
*
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* literals compression based on the compression parameters - specifically,
* negative compression levels do not use literal compression.
*/
#define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
@@ -1790,7 +1817,7 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
*
* Note that this means that the CDict tables can no longer be copied into the
* CCtx, so the dict attachment mode ZSTD_dictForceCopy will no longer be
* useable. The dictionary can only be attached or reloaded.
* usable. The dictionary can only be attached or reloaded.
*
* In general, you should expect compression to be faster--sometimes very much
* so--and CDict creation to be slightly slower. Eventually, we will probably
@@ -1879,23 +1906,26 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
*/
#define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
/* ZSTD_c_splitBlocks
* Default is 0 == disabled. Set to 1 to enable block splitting.
/* ZSTD_c_useBlockSplitter
* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never use block splitter.
* Set to ZSTD_ps_enable to always use block splitter.
*
* Will attempt to split blocks in order to improve compression ratio at the cost of speed.
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* block splitting based on the compression parameters.
*/
#define ZSTD_c_splitBlocks ZSTD_c_experimentalParam13
#define ZSTD_c_useBlockSplitter ZSTD_c_experimentalParam13
/* ZSTD_c_useRowMatchFinder
* Default is ZSTD_urm_auto.
* Controlled with ZSTD_useRowMatchFinderMode_e enum.
* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never use row-based matchfinder.
* Set to ZSTD_ps_enable to force usage of row-based matchfinder.
*
* By default, in ZSTD_urm_auto, when finalizing the compression parameters, the library
* will decide at runtime whether to use the row-based matchfinder based on support for SIMD
* instructions as well as the windowLog.
*
* Set to ZSTD_urm_disableRowMatchFinder to never use row-based matchfinder.
* Set to ZSTD_urm_enableRowMatchFinder to force usage of row-based matchfinder.
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* the row-based matchfinder based on support for SIMD instructions and the window log.
* Note that this only pertains to compression strategies: greedy, lazy, and lazy2
*/
#define ZSTD_c_useRowMatchFinder ZSTD_c_experimentalParam14
@@ -2218,7 +2248,7 @@ size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
* This function is DEPRECATED, and equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_refCDict(zcs, cdict);
*
*
* note : cdict will just be referenced, and must outlive compression session
* This prototype will generate compilation warnings.
*/