fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-06 05:08:08 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

util: Optimize util_is_power_of_two_nonzero

Browse source 
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Suggested-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
This commit is contained in:
Ian Romanick 2017-11-13 13:11:09 -08:00
parent cd18aa1e50
commit ef7a4c9015
1 changed files with 17 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
src/util/bitscan.h
View file

@ -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