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 CUDA constexpr issues for numeric_limits.

Browse Source 
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
Download Patch File Download Diff File Expand all files Collapse all files

18
test/gpu_basic.cu
View File

@@ -343,6 +343,21 @@ struct matrix_inverse {
}
};
template<typename T>
struct numeric_limits_test {
EIGEN_DEVICE_FUNC
void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const
{
EIGEN_UNUSED_VARIABLE(in)
int out_idx = i * 5;
out[out_idx++] = numext::numeric_limits<float>::epsilon();
out[out_idx++] = (numext::numeric_limits<float>::max)();
out[out_idx++] = (numext::numeric_limits<float>::min)();
out[out_idx++] = numext::numeric_limits<float>::infinity();
out[out_idx++] = numext::numeric_limits<float>::quiet_NaN();
}
};
template<typename Type1, typename Type2>
bool verifyIsApproxWithInfsNans(const Type1& a, const Type2& b, typename Type1::Scalar* = 0) // Enabled for Eigen's type only
{
@@ -434,6 +449,9 @@ EIGEN_DECLARE_TEST(gpu_basic)
CALL_SUBTEST( run_and_compare_to_gpu(complex_operators<Vector3cf>(), nthreads, cfin, cfout) );
CALL_SUBTEST( test_with_infs_nans(complex_sqrt<Vector3cf>(), nthreads, cfin, cfout) );
// numeric_limits
CALL_SUBTEST( test_with_infs_nans(numeric_limits_test<Vector3f>(), 1, in, out) );
#if defined(__NVCC__)
// FIXME
// These subtests compiles only with nvcc and fail with HIPCC and clang-cuda