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`).
This commit is contained in:
Antonio Sanchez
2021-03-25 12:37:25 -07:00
committed by Antonio Sánchez
parent af1247fbc1
commit 78ee3d6261
3 changed files with 26 additions and 10 deletions

View File

@@ -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)))