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 Eigen's building of sharded tests that use CUDA & more igamma/igammac bugfixes.

Browse Source 
0. Prior to this PR, not a single sharded CUDA test was actually being *run*.
Fixed that.

GPU tests are still failing for igamma/igammac.

1. Add calls for igamma/igammac to TensorBase
2. Fix up CUDA-specific calls of igamma/igammac
3. Add unit tests for digamma, igamma, igammac in CUDA.
This commit is contained in:
Eugene Brevdo
2016-03-07 14:08:56 -08:00
parent 0b9e0abc96
commit 5707004d6b
7 changed files with 268 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/GenericPacketMath.h
View File

@@ -460,11 +460,11 @@ template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
Packet perfc(const Packet& a) { using numext::erfc; return erfc(a); }
/** \internal \returns the incomplete gamma function igamma(\a a, \a x) */
template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Packet pigamma(const Packet& a, const Packet& x) { using numext::igamma; return igamma(a, x); }
/** \internal \returns the complementary incomplete gamma function igammac(\a a, \a x) */
template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Packet pigammac(const Packet& a, const Packet& x) { using numext::igammac; return igammac(a, x); }
/***************************************************************************

34
Eigen/src/Core/arch/CUDA/MathFunctions.h
View File

@@ -116,24 +116,42 @@ double2 perfc<double2>(const double2& a)
return make_double2(erfc(a.x), erfc(a.y));
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pigamma<float4>(const float4& a, const float4& x)
{
using numext::pigamma;
using numext::igamma;
return make_float4(
pigamma(a.x, x.x),
pigamma(a.y, x.y),
pigamma(a.z, x.z),
pigamma(a.w, x.w));
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 pigammac<double2>(const double2& a, const double& x)
double2 pigamma<double2>(const double2& a, const double2& x)
{
using numext::pigammac;
return make_double2(pigammac(a.x, x.x), pigammac(a.y, x.y));
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

1
Eigen/src/Core/arch/CUDA/PacketMath.h
View File

@@ -284,7 +284,6 @@ 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;