From 78ee3d6261ec6971fcd8fdda6ee040962edde1dc Mon Sep 17 00:00:00 2001
From: Antonio Sanchez <cantonios@google.com>
Date: Thu, 25 Mar 2021 12:37:25 -0700
Subject: Fix CUDA constexpr issues for numeric_limits.

Some CUDA/HIP constants fail on device with `constexpr` since they
internally rely on non-constexpr functions, e.g.
```
\#define CUDART_INF_F            __int_as_float(0x7f800000)
```
This fails for cuda-clang (though passes with nvcc). These constants are
currently used by `device::numeric_limits`.  For portability, we
need to remove `constexpr` from the affected functions.

For C++11 or higher, we should be able to rely on the `std::numeric_limits`
versions anyways, since the methods themselves are now `constexpr`, so
should be supported on device (clang/hipcc natively, nvcc with
`--expr-relaxed-constexpr`).
---
 Eigen/src/Core/util/Macros.h |  2 --
 Eigen/src/Core/util/Meta.h   | 16 ++++++++--------
 2 files changed, 8 insertions(+), 10 deletions(-)

(limited to 'Eigen')

diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 183300928..bdc0de0ea 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -763,8 +763,6 @@
     #if EIGEN_MAX_CPP_VER>=14 && (EIGEN_COMP_CXXVER >= 11 && (EIGEN_COMP_CLANG || EIGEN_COMP_NVCC >= 70500))
       #define EIGEN_HAS_CONSTEXPR 1
     #endif
-  #elif defined(EIGEN_HIPCC)
-  // Skip const_expr on the HIP platform
   #elif EIGEN_MAX_CPP_VER>=14 && (__has_feature(cxx_relaxed_constexpr) || (EIGEN_COMP_CXXVER >= 14) || \
     (EIGEN_GNUC_AT_LEAST(4,8) && (EIGEN_COMP_CXXVER >= 11)) || \
     (EIGEN_COMP_CLANG >= 306 && (EIGEN_COMP_CXXVER >= 11)))
diff --git a/Eigen/src/Core/util/Meta.h b/Eigen/src/Core/util/Meta.h
index 85d35fc4e..f66325f89 100755
--- a/Eigen/src/Core/util/Meta.h
+++ b/Eigen/src/Core/util/Meta.h
@@ -275,7 +275,7 @@ template<bool Condition, typename T=void> struct enable_if;
 template<typename T> struct enable_if<true,T>
 { typedef T type; };
 
-#if defined(EIGEN_GPU_COMPILE_PHASE)
+#if defined(EIGEN_GPU_COMPILE_PHASE) && !EIGEN_HAS_CXX11
 #if !defined(__FLT_EPSILON__)
 #define __FLT_EPSILON__ FLT_EPSILON
 #define __DBL_EPSILON__ DBL_EPSILON
@@ -296,7 +296,7 @@ template<> struct numeric_limits<float>
 {
   EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static float epsilon() { return __FLT_EPSILON__; }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static float (max)() {
   #if defined(EIGEN_CUDA_ARCH)
     return CUDART_MAX_NORMAL_F;
@@ -306,7 +306,7 @@ template<> struct numeric_limits<float>
   }
   EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static float (min)() { return FLT_MIN; }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static float infinity() {
   #if defined(EIGEN_CUDA_ARCH)
     return CUDART_INF_F;
@@ -314,7 +314,7 @@ template<> struct numeric_limits<float>
     return HIPRT_INF_F;
   #endif
   }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static float quiet_NaN() {
   #if defined(EIGEN_CUDA_ARCH)
     return CUDART_NAN_F;
@@ -331,7 +331,7 @@ template<> struct numeric_limits<double>
   static double (max)() { return DBL_MAX; }
   EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static double (min)() { return DBL_MIN; }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static double infinity() {
   #if defined(EIGEN_CUDA_ARCH)
     return CUDART_INF;
@@ -339,7 +339,7 @@ template<> struct numeric_limits<double>
     return HIPRT_INF;
   #endif
   }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static double quiet_NaN() {
   #if defined(EIGEN_CUDA_ARCH)
     return CUDART_NAN;
@@ -414,7 +414,7 @@ template<> struct numeric_limits<bool>
 
 }
 
-#endif
+#endif // defined(EIGEN_GPU_COMPILE_PHASE) && !EIGEN_HAS_CXX11
 
 /** \internal
   * A base class do disable default copy ctor and copy assignment operator.
@@ -761,7 +761,7 @@ template<typename T> EIGEN_DEVICE_FUNC   void swap(T &a, T &b) { T tmp = b; b =
 template<typename T> EIGEN_STRONG_INLINE void swap(T &a, T &b) { std::swap(a,b); }
 #endif
 
-#if defined(EIGEN_GPU_COMPILE_PHASE)
+#if defined(EIGEN_GPU_COMPILE_PHASE) && !EIGEN_HAS_CXX11
 using internal::device::numeric_limits;
 #else
 using std::numeric_limits;
-- 
cgit v1.2.3