GerbilSoft/zlib-ng
1
Fork 0
mirror of https://github.com/GerbilSoft/zlib-ng.git synced 2025-06-18 19:45:37 -04:00
Code Issues Actions Packages Projects Releases Wiki Activity

Provide an inline asm fallback for the ARMv8 intrinsics

Browse Source
This commit is contained in:
Cameron Cawley 2024-02-29 21:20:25 +00:00 committed by Hans Kristian Rosbach
parent 721c488aff
commit 7ea78f12c8
4 changed files with 105 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
CMakeLists.txt
View File

@ -713,12 +713,15 @@ if(WITH_OPTIM)
if(WITH_ARMV8)
check_armv8_compiler_flag()
if(HAVE_ARMV8_FLAG)
if(HAVE_ARMV8_INLINE_ASM OR HAVE_ARMV8_INTRIN)
add_definitions(-DARM_CRC32)
set(ARMV8_SRCS ${ARCHDIR}/crc32_armv8.c)
set_property(SOURCE ${ARMV8_SRCS} PROPERTY COMPILE_FLAGS "${ARMV8FLAG} ${NOLTOFLAG}")
list(APPEND ZLIB_ARCH_SRCS ${ARMV8_SRCS})
add_feature_info(ARMV8_CRC 1 "Support ARMv8 optimized CRC hash generation, using \"${ARMV8FLAG}\"")
if(HAVE_ARMV8_INTRIN)
add_definitions(-DARM_CRC32_INTRIN)
endif()
else()
set(WITH_ARMV8 OFF)
endif()

50
arch/arm/acle_intrins.h
View File

@ -14,6 +14,56 @@
#else
# define Z_TARGET_CRC
#endif
#if !defined(ARM_CRC32_INTRIN) && !defined(_MSC_VER)
#ifdef __aarch64__
static inline uint32_t __crc32b(uint32_t __a, uint8_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32b %w0, %w1, %w2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32h(uint32_t __a, uint16_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32h %w0, %w1, %w2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32w(uint32_t __a, uint32_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32w %w0, %w1, %w2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32d(uint32_t __a, uint64_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32x %w0, %w1, %x2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
#else
static inline uint32_t __crc32b(uint32_t __a, uint8_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32b %0, %1, %2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32h(uint32_t __a, uint16_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32h %0, %1, %2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32w(uint32_t __a, uint32_t __b) {
uint32_t __c;
__asm__ __volatile__("crc32w %0, %1, %2" : "=r" (__c) : "r"(__a), "r"(__b));
return __c;
}
static inline uint32_t __crc32d(uint32_t __a, uint64_t __b) {
return __crc32w (__crc32w (__a, __b & 0xffffffffULL), __b >> 32);
}
#endif
#endif
#endif
#ifdef ARM_SIMD

18
cmake/detect-intrinsics.cmake
View File

@ -15,8 +15,22 @@ macro(check_armv8_compiler_flag)
endif()
endif()
endif()
# Check whether compiler supports ARMv8 CRC intrinsics
# Check whether compiler supports ARMv8 inline asm
set(CMAKE_REQUIRED_FLAGS "${ARMV8FLAG} ${NATIVEFLAG} ${ZNOLTOFLAG}")
check_c_source_compiles(
"unsigned int f(unsigned int a, unsigned int b) {
unsigned int c;
#ifdef __aarch64__
__asm__ __volatile__ ( \"crc32w %w0, %w1, %w2\" : \"=r\" (c) : \"r\" (a), \"r\" (b));
#else
__asm__ __volatile__ ( \"crc32w %0, %1, %2\" : \"=r\" (c) : \"r\" (a), \"r\" (b));
#endif
return (int)c;
}
int main(void) { return f(1,2); }"
HAVE_ARMV8_INLINE_ASM
)
# Check whether compiler supports ARMv8 intrinsics
check_c_source_compiles(
"#if defined(_MSC_VER)
#include <intrin.h>
@ -27,7 +41,7 @@ macro(check_armv8_compiler_flag)
return __crc32w(a, b);
}
int main(void) { return 0; }"
HAVE_ARMV8_FLAG
HAVE_ARMV8_INTRIN
)
set(CMAKE_REQUIRED_FLAGS)
endmacro()

44
configure vendored
View File

@ -1121,7 +1121,7 @@ EOF
fi
}
check_armv8_compiler_flag() {
check_armv8_intrinsics() {
# Check whether -march=armv8-a+crc works correctly
cat > $test.c << EOF
int main() { return 0; }
@ -1139,7 +1139,28 @@ EOF
fi
fi
# Check whether compiler supports ARMv8 CRC intrinsics
# Check whether compiler supports ARMv8 inline asm
cat > $test.c << EOF
unsigned int f(unsigned int a, unsigned int b) {
unsigned int c;
#ifdef __aarch64__
__asm__ __volatile__("crc32w %w0, %w1, %w2" : "=r" (c) : "r"(a), "r"(b));
#else
__asm__ __volatile__("crc32w %0, %1, %2" : "=r" (c) : "r"(a), "r"(b));
#endif
return c;
}
int main(void) { return 0; }
EOF
if try ${CC} ${CFLAGS} ${armv8flag} $test.c; then
echo "Checking for ARMv8 inline assembly ... Yes." | tee -a configure.log
HAVE_ARMV8_INLINE_ASM=1
else
echo "Checking for ARMv8 inline assembly ... No." | tee -a configure.log
HAVE_ARMV8_INLINE_ASM=0
fi
# Check whether compiler supports ARMv8 intrinsics
cat > $test.c << EOF
#include <arm_acle.h>
unsigned int f(unsigned int a, unsigned int b) {
@ -1147,12 +1168,12 @@ unsigned int f(unsigned int a, unsigned int b) {
}
int main(void) { return 0; }
EOF
if try ${CC} ${CFLAGS} ${acleflag} $test.c; then
echo "Checking for ARMv8 CRC intrinsics ... Yes." | tee -a configure.log
ARMV8_AVAILABLE=1
if try ${CC} ${CFLAGS} ${armv8flag} $test.c; then
echo "Checking for ARMv8 intrinsics ... Yes." | tee -a configure.log
HAVE_ARMV8_INTRIN=1
else
echo "Checking for ARMv8 CRC intrinsics ... No." | tee -a configure.log
ARMV8_AVAILABLE=0
echo "Checking for ARMv8 intrinsics ... No." | tee -a configure.log
HAVE_ARMV8_INTRIN=0
fi
}
@ -1738,12 +1759,17 @@ EOF
fi
if test $buildarmv8 -eq 1; then
check_armv8_compiler_flag
check_armv8_intrinsics
if test $ARMV8_AVAILABLE -eq 1; then
if test $HAVE_ARMV8_INTRIN -eq 1 || test $HAVE_ARMV8_INLINE_ASM -eq 1; then
CFLAGS="${CFLAGS} -DARM_CRC32"
SFLAGS="${SFLAGS} -DARM_CRC32"
if test $HAVE_ARMV8_INTRIN -eq 1; then
CFLAGS="${CFLAGS} -DARM_CRC32_INTRIN"
SFLAGS="${SFLAGS} -DARM_CRC32_INTRIN"
fi
ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} crc32_armv8.o"
ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} crc32_armv8.lo"
fi