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

Merge pull request #4265 from pps83/static-bmi2-check

Browse Source 
Check `STATIC_BMI2` instead of `STATIC_BMI2 == 1`
This commit is contained in:
Yann Collet
2025-01-31 14:39:20 -08:00
committed by GitHub
parent 086ddcd9ba 0cda0100ea
commit 26a2b5d5df
2 changed files with 93 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/common/bits.h
+92 -95
View File
@@ -28,29 +28,29 @@ MEM_STATIC unsigned ZSTD_countTrailingZeros32_fallback(U32 val)
MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val) MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val)
{ {
assert(val != 0); assert(val != 0);
# if defined(_MSC_VER) #if defined(_MSC_VER)
# if STATIC_BMI2 == 1 # if STATIC_BMI2
return (unsigned)_tzcnt_u32(val); return (unsigned)_tzcnt_u32(val);
# else # else
if (val != 0) { if (val != 0) {
unsigned long r; unsigned long r;
_BitScanForward(&r, val); _BitScanForward(&r, val);
return (unsigned)r; return (unsigned)r;
} else { } else {
/* Should not reach this code path */ __assume(0); /* Should not reach this code path */
__assume(0); }
} # endif
# endif #elif defined(__GNUC__) && (__GNUC__ >= 4)
# elif defined(__GNUC__) && (__GNUC__ >= 4) return (unsigned)__builtin_ctz(val);
return (unsigned)__builtin_ctz(val); #elif defined(__ICCARM__)
# elif defined(__ICCARM__) return (unsigned)__builtin_ctz(val);
return (unsigned)__builtin_ctz(val); #else
# else return ZSTD_countTrailingZeros32_fallback(val);
return ZSTD_countTrailingZeros32_fallback(val); #endif
# endif
} }
MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) { MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val)
{
assert(val != 0); assert(val != 0);
{ {
static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29, static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29,
@@ -69,92 +69,89 @@ MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) {
MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val) MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val)
{ {
assert(val != 0); assert(val != 0);
# if defined(_MSC_VER) #if defined(_MSC_VER)
# if STATIC_BMI2 == 1 # if STATIC_BMI2
return (unsigned)_lzcnt_u32(val); return (unsigned)_lzcnt_u32(val);
# else # else
if (val != 0) { if (val != 0) {
unsigned long r; unsigned long r;
_BitScanReverse(&r, val); _BitScanReverse(&r, val);
return (unsigned)(31 - r); return (unsigned)(31 - r);
} else { } else {
/* Should not reach this code path */ __assume(0); /* Should not reach this code path */
__assume(0); }
} # endif
# endif #elif defined(__GNUC__) && (__GNUC__ >= 4)
# elif defined(__GNUC__) && (__GNUC__ >= 4) return (unsigned)__builtin_clz(val);
return (unsigned)__builtin_clz(val); #elif defined(__ICCARM__)
# elif defined(__ICCARM__) return (unsigned)__builtin_clz(val);
return (unsigned)__builtin_clz(val); #else
# else return ZSTD_countLeadingZeros32_fallback(val);
return ZSTD_countLeadingZeros32_fallback(val); #endif
# endif
} }
MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val) MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val)
{ {
assert(val != 0); assert(val != 0);
# if defined(_MSC_VER) && defined(_WIN64) #if defined(_MSC_VER) && defined(_WIN64)
# if STATIC_BMI2 == 1 # if STATIC_BMI2
return (unsigned)_tzcnt_u64(val); return (unsigned)_tzcnt_u64(val);
# else # else
if (val != 0) { if (val != 0) {
unsigned long r; unsigned long r;
_BitScanForward64(&r, val); _BitScanForward64(&r, val);
return (unsigned)r; return (unsigned)r;
} else { } else {
/* Should not reach this code path */ __assume(0); /* Should not reach this code path */
__assume(0); }
} # endif
# endif #elif defined(__GNUC__) && (__GNUC__ >= 4) && defined(__LP64__)
# elif defined(__GNUC__) && (__GNUC__ >= 4) && defined(__LP64__) return (unsigned)__builtin_ctzll(val);
return (unsigned)__builtin_ctzll(val); #elif defined(__ICCARM__)
# elif defined(__ICCARM__) return (unsigned)__builtin_ctzll(val);
return (unsigned)__builtin_ctzll(val); #else
# else {
{ U32 mostSignificantWord = (U32)(val >> 32);
U32 mostSignificantWord = (U32)(val >> 32); U32 leastSignificantWord = (U32)val;
U32 leastSignificantWord = (U32)val; if (leastSignificantWord == 0) {
if (leastSignificantWord == 0) { return 32 + ZSTD_countTrailingZeros32(mostSignificantWord);
return 32 + ZSTD_countTrailingZeros32(mostSignificantWord); } else {
} else { return ZSTD_countTrailingZeros32(leastSignificantWord);
return ZSTD_countTrailingZeros32(leastSignificantWord);
}
} }
# endif }
#endif
} }
MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val) MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val)
{ {
assert(val != 0); assert(val != 0);
# if defined(_MSC_VER) && defined(_WIN64) #if defined(_MSC_VER) && defined(_WIN64)
# if STATIC_BMI2 == 1 # if STATIC_BMI2
return (unsigned)_lzcnt_u64(val); return (unsigned)_lzcnt_u64(val);
# else # else
if (val != 0) { if (val != 0) {
unsigned long r; unsigned long r;
_BitScanReverse64(&r, val); _BitScanReverse64(&r, val);
return (unsigned)(63 - r); return (unsigned)(63 - r);
} else { } else {
/* Should not reach this code path */ __assume(0); /* Should not reach this code path */
__assume(0); }
} # endif
# endif #elif defined(__GNUC__) && (__GNUC__ >= 4)
# elif defined(__GNUC__) && (__GNUC__ >= 4) return (unsigned)(__builtin_clzll(val));
return (unsigned)(__builtin_clzll(val)); #elif defined(__ICCARM__)
# elif defined(__ICCARM__) return (unsigned)(__builtin_clzll(val));
return (unsigned)(__builtin_clzll(val)); #else
# else {
{ U32 mostSignificantWord = (U32)(val >> 32);
U32 mostSignificantWord = (U32)(val >> 32); U32 leastSignificantWord = (U32)val;
U32 leastSignificantWord = (U32)val; if (mostSignificantWord == 0) {
if (mostSignificantWord == 0) { return 32 + ZSTD_countLeadingZeros32(leastSignificantWord);
return 32 + ZSTD_countLeadingZeros32(leastSignificantWord); } else {
} else { return ZSTD_countLeadingZeros32(mostSignificantWord);
return ZSTD_countLeadingZeros32(mostSignificantWord);
}
} }
# endif }
#endif
} }
MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val) MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val)
lib/common/bitstream.h
+1 -1
View File
@@ -161,7 +161,7 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
FORCE_INLINE_TEMPLATE BitContainerType BIT_getLowerBits(BitContainerType bitContainer, U32 const nbBits) FORCE_INLINE_TEMPLATE BitContainerType BIT_getLowerBits(BitContainerType bitContainer, U32 const nbBits)
{ {
#if defined(STATIC_BMI2) && STATIC_BMI2 == 1 && !defined(ZSTD_NO_INTRINSICS) #if STATIC_BMI2 && !defined(ZSTD_NO_INTRINSICS)
# if (defined(__x86_64__) || defined(_M_X64)) && !defined(__ILP32__) # if (defined(__x86_64__) || defined(_M_X64)) && !defined(__ILP32__)
return _bzhi_u64(bitContainer, nbBits); return _bzhi_u64(bitContainer, nbBits);
# else # else