GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-04 18:54:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Tests added and AVX512 bug fixed for pcmp_lt_or_nan

Browse Source
This commit is contained in:
Jakub Lichman 2021-04-25 20:58:56 +00:00
parent 1115f5462e
commit d87648a6be
2 changed files with 34 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@ -487,7 +487,7 @@ template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt(const Packet16f& a, const Packe
} }
template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt_or_nan(const Packet16f& a, const Packet16f& b) { template<> EIGEN_STRONG_INLINE Packet16f pcmp_lt_or_nan(const Packet16f& a, const Packet16f& b) {
__mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_NGT_UQ); __mmask16 mask = _mm512_cmp_ps_mask(a, b, _CMP_NGE_UQ);
return _mm512_castsi512_ps( return _mm512_castsi512_ps(
_mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu)); _mm512_mask_set1_epi32(_mm512_set1_epi32(0), mask, 0xffffffffu));
} }
@ -518,7 +518,7 @@ EIGEN_STRONG_INLINE Packet8d pcmp_lt(const Packet8d& a, const Packet8d& b) {
} }
template <> template <>
EIGEN_STRONG_INLINE Packet8d pcmp_lt_or_nan(const Packet8d& a, const Packet8d& b) { EIGEN_STRONG_INLINE Packet8d pcmp_lt_or_nan(const Packet8d& a, const Packet8d& b) {
__mmask8 mask = _mm512_cmp_pd_mask(a, b, _CMP_NGT_UQ); __mmask8 mask = _mm512_cmp_pd_mask(a, b, _CMP_NGE_UQ);
return _mm512_castsi512_pd( return _mm512_castsi512_pd(
_mm512_mask_set1_epi64(_mm512_set1_epi64(0), mask, 0xffffffffffffffffu)); _mm512_mask_set1_epi64(_mm512_set1_epi64(0), mask, 0xffffffffffffffffu));
} }

32
test/packetmath.cpp
View File

@ -279,6 +279,36 @@ void packetmath_boolean_mask_ops() {
CHECK_CWISE2_IF(true, internal::pcmp_eq, internal::pcmp_eq); CHECK_CWISE2_IF(true, internal::pcmp_eq, internal::pcmp_eq);
} }
template <typename Scalar, typename Packet>
void packetmath_boolean_mask_ops_real() {
const int PacketSize = internal::unpacket_traits<Packet>::size;
const int size = 2 * PacketSize;
EIGEN_ALIGN_MAX Scalar data1[size];
EIGEN_ALIGN_MAX Scalar data2[size];
EIGEN_ALIGN_MAX Scalar ref[size];
for (int i = 0; i < PacketSize; ++i) {
data1[i] = internal::random<Scalar>();
data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
}
CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
//Test (-0) <=/< (0) for signed operations
for (int i = 0; i < PacketSize; ++i) {
data1[i] = Scalar(-0.0);
data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
}
CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
//Test NaN
for (int i = 0; i < PacketSize; ++i) {
data1[i] = NumTraits<Scalar>::quiet_NaN();
data1[i + PacketSize] = internal::random<bool>() ? data1[i] : Scalar(0);
}
CHECK_CWISE2_IF(true, internal::pcmp_lt_or_nan, internal::pcmp_lt_or_nan);
}
template <typename Scalar, typename Packet> template <typename Scalar, typename Packet>
void packetmath_boolean_mask_ops_notcomplex() { void packetmath_boolean_mask_ops_notcomplex() {
const int PacketSize = internal::unpacket_traits<Packet>::size; const int PacketSize = internal::unpacket_traits<Packet>::size;
@ -610,6 +640,8 @@ void packetmath_real() {
CHECK_CWISE1_EXACT_IF(PacketTraits::HasFloor, numext::floor, internal::pfloor); CHECK_CWISE1_EXACT_IF(PacketTraits::HasFloor, numext::floor, internal::pfloor);
CHECK_CWISE1_EXACT_IF(PacketTraits::HasRint, numext::rint, internal::print); CHECK_CWISE1_EXACT_IF(PacketTraits::HasRint, numext::rint, internal::print);
packetmath_boolean_mask_ops_real<Scalar,Packet>();
// Rounding edge cases. // Rounding edge cases.
if (PacketTraits::HasRound || PacketTraits::HasCeil || PacketTraits::HasFloor || PacketTraits::HasRint) { if (PacketTraits::HasRound || PacketTraits::HasCeil || PacketTraits::HasFloor || PacketTraits::HasRint) {
typedef typename internal::make_integer<Scalar>::type IntType; typedef typename internal::make_integer<Scalar>::type IntType;