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

Fix the specialization of pfrexp for AVX to be faster when AVX2/AVX512DQ is not available, and avoid undefined behavior in C++. Also mask off the sign bit when extracting the exponent.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2020-10-15 18:39:58 -07:00
parent 011e0db31d
commit 21edea5edd
2 changed files with 20 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -805,10 +805,11 @@ template<> EIGEN_STRONG_INLINE Packet4f pfrexp<Packet4f>(const Packet4f& a, Pack
template<> EIGEN_STRONG_INLINE Packet2d pfrexp<Packet2d>(const Packet2d& a, Packet2d& exponent) {
const Packet2d cst_1022d = pset1<Packet2d>(1022.0);
const Packet2d cst_half = pset1<Packet2d>(0.5);
const Packet2d cst_inv_mant_mask = pset1frombits<Packet2d>(static_cast<uint64_t>(~0x7ff0000000000000ull));
__m128i a_expo = _mm_srli_epi64(_mm_castpd_si128(a), 52);
const Packet2d cst_exp_mask = pset1frombits<Packet2d>(static_cast<uint64_t>(0x7ff0000000000000ull));
__m128i a_expo = _mm_srli_epi64(_mm_castpd_si128(pand(a, cst_exp_mask)), 52);
exponent = psub(_mm_cvtepi32_pd(vec4i_swizzle1(a_expo, 0, 2, 1, 3)), cst_1022d);
return por(pand(a, cst_inv_mant_mask), cst_half);
const Packet2d cst_mant_mask = pset1frombits<Packet2d>(static_cast<uint64_t>(~0x7ff0000000000000ull));
return por(pand(a, cst_mant_mask), cst_half);
}
template<> EIGEN_STRONG_INLINE Packet4f pldexp<Packet4f>(const Packet4f& a, const Packet4f& exponent) {