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

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.

3 files changed, 28 insertions, 11 deletions

diff --git a/Eigen/src/Core/GenericPacketMath.h b/Eigen/src/Core/GenericPacketMath.h
index ead0253df..802def51d 100644
--- a/Eigen/src/Core/GenericPacketMath.h
+++ b/Eigen/src/Core/GenericPacketMath.h
@@ -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); }
 
 /***************************************************************************
diff --git a/Eigen/src/Core/arch/CUDA/MathFunctions.h b/Eigen/src/Core/arch/CUDA/MathFunctions.h
index 6e84d3af8..6822700f8 100644
--- a/Eigen/src/Core/arch/CUDA/MathFunctions.h
+++ b/Eigen/src/Core/arch/CUDA/MathFunctions.h
@@ -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
 
diff --git a/Eigen/src/Core/arch/CUDA/PacketMath.h b/Eigen/src/Core/arch/CUDA/PacketMath.h
index d2563030b..25d964600 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMath.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMath.h
@@ -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;