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

Merged in ebrevdo/eigen (pull request PR-169)

Browse Source 
Bugfixes to cuda tests, igamma & igammac implemented, & tests for digamma, igamma, igammac on CPU & GPU.
This commit is contained in:
Christoph Hertzberg
2016-03-16 21:59:08 +01:00
parent ab9b749b45 f1f7181f53
commit 46aa9772fc
15 changed files with 858 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
Eigen/src/Core/arch/CUDA/MathFunctions.h
View File

@@ -117,6 +117,42 @@ double2 perfc<double2>(const double2& a)
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pigamma<float4>(const float4& a, const float4& x)
{
using numext::igamma;
return make_float4(
igamma(a.x, x.x),
igamma(a.y, x.y),
igamma(a.z, x.z),
igamma(a.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pigamma<double2>(const double2& a, const double2& x)
{
using numext::igamma;
return make_double2(igamma(a.x, x.x), igamma(a.y, x.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pigammac<float4>(const float4& a, const float4& x)
{
using numext::igammac;
return make_float4(
igammac(a.x, x.x),
igammac(a.y, x.y),
igammac(a.z, x.z),
igammac(a.w, x.w));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pigammac<double2>(const double2& a, const double2& x)
{
using numext::igammac;
return make_double2(igammac(a.x, x.x), igammac(a.y, x.y));
}
#endif
} // end namespace internal

5
Eigen/src/Core/arch/CUDA/PacketMath.h
View File

@@ -42,6 +42,8 @@ template<> struct packet_traits<float> : default_packet_traits
HasDiGamma = 1,
HasErf = 1,
HasErfc = 1,
HasIgamma = 1,
HasIGammac = 1,
HasBlend = 0,
};
@@ -66,6 +68,8 @@ template<> struct packet_traits<double> : default_packet_traits
HasDiGamma = 1,
HasErf = 1,
HasErfc = 1,
HasIGamma = 1,
HasIGammac = 1,
HasBlend = 0,
};
@@ -308,7 +312,6 @@ template<> EIGEN_DEVICE_FUNC inline double2 pabs<double2>(const double2& a) {
return make_double2(fabs(a.x), fabs(a.y));
}
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<float4,4>& kernel) {
double tmp = kernel.packet[0].y;