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

UBSAN: use appropriate SSE intrinsics for loading 4 and 8 bytes

Browse Source 
libeigen/eigen!2346
This commit is contained in:
Charles Schlosser
2026-03-27 19:54:10 +00:00
parent 9939a4c6e3
commit eb4b2eeffa
1 changed files with 7 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -1398,7 +1398,7 @@ template <typename Packet>
EIGEN_STRONG_INLINE Packet ploadl(const typename unpacket_traits<Packet>::type* from);
template <>
EIGEN_STRONG_INLINE Packet4f ploadl<Packet4f>(const float* from) {
EIGEN_DEBUG_UNALIGNED_LOAD return _mm_castpd_ps(_mm_load_sd(reinterpret_cast<const double*>(from)));
EIGEN_DEBUG_UNALIGNED_LOAD return _mm_castsi128_ps(_mm_loadu_si64(reinterpret_cast<const void*>(from)));
}
template <>
EIGEN_STRONG_INLINE Packet2d ploadl<Packet2d>(const double* from) {
@@ -1419,7 +1419,7 @@ EIGEN_STRONG_INLINE Packet2d ploads<Packet2d>(const double* from) {
template <>
EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from) {
return vec4f_swizzle1(_mm_castpd_ps(_mm_load_sd(reinterpret_cast<const double*>(from))), 0, 0, 1, 1);
return vec4f_swizzle1(_mm_castsi128_ps(_mm_loadu_si64(reinterpret_cast<const void*>(from))), 0, 0, 1, 1);
}
template <>
EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double* from) {
@@ -1446,7 +1446,7 @@ EIGEN_STRONG_INLINE Packet4ui ploaddup<Packet4ui>(const uint32_t* from) {
// {b0, b0, b1, b1, b2, b2, b3, b3, b4, b4, b5, b5, b6, b6, b7, b7}
template <>
EIGEN_STRONG_INLINE Packet16b ploaddup<Packet16b>(const bool* from) {
__m128i tmp = _mm_castpd_si128(pload1<Packet2d>(reinterpret_cast<const double*>(from)));
__m128i tmp = _mm_loadu_si64(reinterpret_cast<const void*>(from));
return _mm_unpacklo_epi8(tmp, tmp);
}
@@ -1454,7 +1454,10 @@ EIGEN_STRONG_INLINE Packet16b ploaddup<Packet16b>(const bool* from) {
// {b0, b0 b0, b0, b1, b1, b1, b1, b2, b2, b2, b2, b3, b3, b3, b3}
template <>
EIGEN_STRONG_INLINE Packet16b ploadquad<Packet16b>(const bool* from) {
__m128i tmp = _mm_castps_si128(pload1<Packet4f>(reinterpret_cast<const float*>(from)));
EIGEN_USING_STD(memcpy);
int val;
memcpy(&val, from, sizeof(int));
__m128i tmp = _mm_cvtsi32_si128(val);
tmp = _mm_unpacklo_epi8(tmp, tmp);
return _mm_unpacklo_epi16(tmp, tmp);
}