mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-03-07 18:27:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Define pcast<> for SSE types even when AVX is enabled. (otherwise float are silently reinterpreted as int instead of being converted)

Browse Source
This commit is contained in:
Gael Guennebaud 2018-05-29 20:46:46 +02:00
parent 49262dfee6
commit 647b724a36
2 changed files with 15 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Eigen/Core
View File

@ -413,6 +413,7 @@ using std::ptrdiff_t;
#include "src/Core/arch/AVX/MathFunctions.h"
#include "src/Core/arch/AVX/Complex.h"
#include "src/Core/arch/AVX/TypeCasting.h"
#include "src/Core/arch/SSE/TypeCasting.h"
#elif defined EIGEN_VECTORIZE_SSE
#include "src/Core/arch/SSE/PacketMath.h"
#include "src/Core/arch/SSE/MathFunctions.h"

28
Eigen/src/Core/arch/SSE/TypeCasting.h
View File

@ -14,6 +14,7 @@ namespace Eigen {
namespace internal {
#ifndef EIGEN_VECTORIZE_AVX
template <>
struct type_casting_traits<float, int> {
enum {
@ -23,11 +24,6 @@ struct type_casting_traits<float, int> {
};
};
template<> EIGEN_STRONG_INLINE Packet4i pcast<Packet4f, Packet4i>(const Packet4f& a) {
return _mm_cvttps_epi32(a);
}
template <>
struct type_casting_traits<int, float> {
enum {
@ -37,11 +33,6 @@ struct type_casting_traits<int, float> {
};
};
template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4i, Packet4f>(const Packet4i& a) {
return _mm_cvtepi32_ps(a);
}
template <>
struct type_casting_traits<double, float> {
enum {
@ -51,10 +42,6 @@ struct type_casting_traits<double, float> {
};
};
template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet2d, Packet4f>(const Packet2d& a, const Packet2d& b) {
return _mm_shuffle_ps(_mm_cvtpd_ps(a), _mm_cvtpd_ps(b), (1 << 2) | (1 << 6));
}
template <>
struct type_casting_traits<float, double> {
enum {
@ -63,6 +50,19 @@ struct type_casting_traits<float, double> {
TgtCoeffRatio = 2
};
};
#endif
template<> EIGEN_STRONG_INLINE Packet4i pcast<Packet4f, Packet4i>(const Packet4f& a) {
return _mm_cvttps_epi32(a);
}
template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4i, Packet4f>(const Packet4i& a) {
return _mm_cvtepi32_ps(a);
}
template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet2d, Packet4f>(const Packet2d& a, const Packet2d& b) {
return _mm_shuffle_ps(_mm_cvtpd_ps(a), _mm_cvtpd_ps(b), (1 << 2) | (1 << 6));
}
template<> EIGEN_STRONG_INLINE Packet2d pcast<Packet4f, Packet2d>(const Packet4f& a) {
// Simply discard the second half of the input