diff options
| author | Chenxi Mao <chenxi.mao2013@gmail.com> | 2024-01-09 17:23:11 +0800 |
|---|---|---|
| committer | Jia Tan <jiat0218@gmail.com> | 2024-01-27 21:49:26 +0800 |
| commit | 849d0f282a6a890c5cf5a0e0f02980b12d9ebb0f (patch) | |
| tree | 2f8da4c7fcf61a8b9f6425f3075727c972e6bf94 /src/liblzma/check/crc_common.h | |
| parent | Bump version number for 5.5.1alpha. (diff) | |
| download | xz-849d0f282a6a890c5cf5a0e0f02980b12d9ebb0f.tar.xz | |
Speed up CRC32 calculation on ARM64
The CRC32 instructions in ARM64 can calculate the CRC32 result
for 8 bytes in a single operation, making the use of ARM64
instructions much faster compared to the general CRC32 algorithm.
Optimized CRC32 will be enabled if ARM64 has CRC extension
running on Linux.
Signed-off-by: Chenxi Mao <chenxi.mao2013@gmail.com>
Diffstat (limited to 'src/liblzma/check/crc_common.h')
| -rw-r--r-- | src/liblzma/check/crc_common.h | 16 |
1 files changed, 10 insertions, 6 deletions
diff --git a/src/liblzma/check/crc_common.h b/src/liblzma/check/crc_common.h index 417d88bb..7c7f098d 100644 --- a/src/liblzma/check/crc_common.h +++ b/src/liblzma/check/crc_common.h @@ -52,29 +52,33 @@ #undef CRC_GENERIC #undef CRC_ARCH_OPTIMIZED #undef CRC_X86_CLMUL +#undef CRC32_ARM64 #undef CRC_USE_IFUNC #undef CRC_USE_GENERIC_FOR_SMALL_INPUTS -// If CLMUL cannot be used then only the generic slice-by-eight (CRC32) -// or slice-by-four (CRC64) is built. -#if !defined(HAVE_USABLE_CLMUL) -# define CRC_GENERIC 1 - // If CLMUL is allowed unconditionally in the compiler options then the // generic version can be omitted. Note that this doesn't work with MSVC // as I don't know how to detect the features here. // // NOTE: Keep this this in sync with crc32_table.c. -#elif (defined(__SSSE3__) && defined(__SSE4_1__) && defined(__PCLMUL__)) \ +#if (defined(__SSSE3__) && defined(__SSE4_1__) && defined(__PCLMUL__)) \ || (defined(__e2k__) && __iset__ >= 6) # define CRC_ARCH_OPTIMIZED 1 # define CRC_X86_CLMUL 1 +#elif (defined(__aarch64__)) +# define CRC_ARCH_OPTIMIZED 1 +# define CRC32_ARM64 1 +// If CLMUL cannot be used then only the generic slice-by-eight (CRC32) +// or slice-by-four (CRC64) is built. +#elif !defined(HAVE_USABLE_CLMUL) +# define CRC_GENERIC 1 // Otherwise build both and detect at runtime which version to use. #else # define CRC_GENERIC 1 # define CRC_ARCH_OPTIMIZED 1 # define CRC_X86_CLMUL 1 +# define CRC32_ARM64 1 # ifdef HAVE_FUNC_ATTRIBUTE_IFUNC # define CRC_USE_IFUNC 1 |