diff options
| author | Thiago Macieira <thiago.macieira@intel.com> | 2015-12-03 17:37:02 -0800 |
|---|---|---|
| committer | Thiago Macieira <thiago.macieira@intel.com> | 2015-12-05 00:11:28 +0000 |
| commit | f5f47987ce369aa3f7553e6c0da509461a1ddf1a (patch) | |
| tree | aa2687fdc9cdfff5756ccbdf28098932277823e6 /src/corelib | |
| parent | 2fb42eb4af3444b11e7b1210323637937ef0b005 (diff) | |
qCount{Leading,Trailing}ZeroBits: Use __builtin_clzs for 16-bit
If possible.
The BSF/BSR/TZCNT/LZCNT Intel instruction does not exist for 8-bit. And
it's a good idea to use the 32-bit instruction instead of the 16-bit one
for that case, to avoid the Length Changing Prefix (LCP).
GCC doesn't allow us to use __builtin_cl[tz]s unless BMI is active,
while ICC generates the same code either way (Clang understands
__has_builtin).
Change-Id: I8de47ed6c7be4847b99bffff141c91603c7024dc
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
Diffstat (limited to 'src/corelib')
| -rw-r--r-- | src/corelib/tools/qalgorithms.h | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/src/corelib/tools/qalgorithms.h b/src/corelib/tools/qalgorithms.h index ffa3082d5e..27630aea87 100644 --- a/src/corelib/tools/qalgorithms.h +++ b/src/corelib/tools/qalgorithms.h @@ -620,7 +620,11 @@ Q_DECL_RELAXED_CONSTEXPR inline uint qCountTrailingZeroBits(quint8 v) Q_DECL_NOT Q_DECL_RELAXED_CONSTEXPR inline uint qCountTrailingZeroBits(quint16 v) Q_DECL_NOTHROW { #if defined(Q_CC_GNU) +# if QT_HAS_BUILTIN(__builtin_ctzs) || defined(__BMI__) + return v ? __builtin_ctzs(v) : 16U; +# else return v ? __builtin_ctz(v) : 16U; +# endif #else unsigned int c = 16; // c will be the number of zero bits on the right v &= -signed(v); @@ -679,7 +683,11 @@ Q_DECL_RELAXED_CONSTEXPR inline uint qCountLeadingZeroBits(quint8 v) Q_DECL_NOTH Q_DECL_RELAXED_CONSTEXPR inline uint qCountLeadingZeroBits(quint16 v) Q_DECL_NOTHROW { #if defined(Q_CC_GNU) +# if QT_HAS_BUILTIN(__builtin_clzs) || defined(__BMI__) + return v ? __builtin_clzs(v) : 16U; +# else return v ? __builtin_clz(v)-16U : 16U; +# endif #else v = v | (v >> 1); v = v | (v >> 2); |