devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize predux_any<Packet4f>

Browse Source 
libeigen/eigen!2277
This commit is contained in:
onalante-ebay
2026-03-12 16:15:16 +00:00
committed by Rasmus Munk Larsen
parent 8190c82cb4
commit 3a2ba7c434
1 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@@ -3988,8 +3988,16 @@ EIGEN_STRONG_INLINE uint64_t predux_max<Packet2ul>(const Packet2ul& a) {
template <>
EIGEN_STRONG_INLINE bool predux_any(const Packet4f& x) {
uint32x2_t tmp = vorr_u32(vget_low_u32(vreinterpretq_u32_f32(x)), vget_high_u32(vreinterpretq_u32_f32(x)));
return vget_lane_u32(vpmax_u32(tmp, tmp), 0);
uint32x4_t u = vreinterpretq_u32_f32(x);
#if EIGEN_ARCH_ARM64
return vget_lane_u64(vreinterpret_u64_u16(vmovn_u32(u)), 0);
#else
uint32x2_t tmp = vorr_u32(vget_low_u32(u), vget_high_u32(u));
uint32_t a, b;
// GCC and Clang refuse to emit this instruction.
asm("vmov %0, %1, %P2" : "=r"(a), "=r"(b) : "w"(tmp));
return a | b;
#endif
}
// Helpers for ptranspose.