diff options
| author | Ian Romanick <[email protected]> | 2017-11-13 13:11:09 -0800 |
|---|---|---|
| committer | Ian Romanick <[email protected]> | 2018-03-29 14:09:29 -0700 |
| commit | ef7a4c90155cb180a4f0d289594ef11da9a2b21a (patch) | |
| tree | 01d8ad661bd8e380bffe568c6cc8db472f76404c | |
| parent | cd18aa1e50e4668c8d992816bcdb74376dd1064c (diff) | |
util: Optimize util_is_power_of_two_nonzero
Signed-off-by: Ian Romanick <[email protected]>
Suggested-by: Matt Turner <[email protected]>
Reviewed-by: Eduardo Lima Mitev <[email protected]>
| -rw-r--r-- | src/util/bitscan.h | 17 |
1 files changed, 17 insertions, 0 deletions
diff --git a/src/util/bitscan.h b/src/util/bitscan.h index a3f2d414bd6..5cc75f0beba 100644 --- a/src/util/bitscan.h +++ b/src/util/bitscan.h @@ -38,6 +38,10 @@ #include <intrin.h> #endif +#if defined(__POPCNT__) +#include <popcntintrin.h> +#endif + #include "c99_compat.h" #ifdef __cplusplus @@ -127,7 +131,20 @@ util_is_power_of_two_or_zero(unsigned v) static inline bool util_is_power_of_two_nonzero(unsigned v) { + /* __POPCNT__ is different from HAVE___BUILTIN_POPCOUNT. The latter + * indicates the existence of the __builtin_popcount function. The former + * indicates that _mm_popcnt_u32 exists and is a native instruction. + * + * The other alternative is to use SSE 4.2 compile-time flags. This has + * two drawbacks. First, there is currently no build infrastructure for + * SSE 4.2 (only 4.1), so that would have to be added. Second, some AMD + * CPUs support POPCNT but not SSE 4.2 (e.g., Barcelona). + */ +#ifdef __POPCNT__ + return _mm_popcnt_u32(v) == 1; +#else return v != 0 && (v & (v - 1)) == 0; +#endif } /* For looping over a bitmask when you want to loop over consecutive bits |