Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Use vmaxvq_s32 to implement AnyZeroSignBit more efficiently on ARM64

Browse source
This commit is contained in:
Henrik Rydgård 2025-01-19 17:41:34 +01:00
parent a40cf12593
commit a5116e1590
1 changed files with 10 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
Common/Math/CrossSIMD.h
View file

@ -655,17 +655,27 @@ inline void TranslateAndScaleInplace(Mat4F32 &m, Vec4F32 scale, Vec4F32 translat
}
inline bool AnyZeroSignBit(Vec4S32 value) {
#if PPSSPP_ARCH(ARM64_NEON)
// Shortcut on arm64
return vmaxvq_s32(value.v) >= 0;
#else
// Very suboptimal, let's optimize later.
int32x2_t prod = vand_s32(vget_low_s32(value.v), vget_high_s32(value.v));
int mask = vget_lane_s32(prod, 0) & vget_lane_s32(prod, 1);
return (mask & 0x80000000) == 0;
#endif
}
inline bool AnyZeroSignBit(Vec4F32 value) {
int32x4_t ival = vreinterpretq_s32_f32(value.v);
#if PPSSPP_ARCH(ARM64_NEON)
// Shortcut on arm64
return vmaxvq_s32(value.v) >= 0;
#else
int32x2_t prod = vand_s32(vget_low_s32(ival), vget_high_s32(ival));
int mask = vget_lane_s32(prod, 0) & vget_lane_s32(prod, 1);
return (mask & 0x80000000) == 0;
#endif
}