[CUDA] call to pow is ambiguous

[llvmbot]

Bugzilla Link	26497
Resolution	FIXED
Resolved on	Feb 05, 2016 17:18
Version	trunk
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@hfinkel,@jlebar,@Artem-B

Extended Description

Is this intended?

// pow.cu using namespace std;

global void foo(double* dd) {
dd[0] = pow(dd[1], dd[2]);
}

$ clang++ pow.cu -c

pow.cu:4:11: error: call to 'pow' is ambiguous dd[0] = pow(dd[1], dd[2]); ^~~ /usr/local/cuda/include/math_functions_dbl_ptx3.hpp:350:41: note: candidate function MATH_FUNCTIONS_DBL_PTX3_DECL double pow(double a, double b) ^ /usr/local/google/home/jingyue/Work/llvm/install-git/bin/../lib/clang/3.9.0/include/clang_cuda_cmath.h:186:19: note: candidate function DEVICE double pow(double base, double exp) { return ::pow(base, exp); } ^ /usr/local/google/home/jingyue/Work/llvm/install-git/bin/../lib/clang/3.9.0/include/clang_cuda_cmath.h:187:19: note: candidate function DEVICE double pow(double base, int iexp) { return ::powi(base, iexp); } ^ /usr/local/google/home/jingyue/Work/llvm/install-git/bin/../lib/clang/3.9.0/include/clang_cuda_cmath.h:184:18: note: candidate function DEVICE float pow(float base, float exp) { return ::powf(base, exp); } ^ /usr/local/google/home/jingyue/Work/llvm/install-git/bin/../lib/clang/3.9.0/include/clang_cuda_cmath.h:185:18: note: candidate function DEVICE float pow(float base, int iexp) { return ::powif(base, iexp); } ^ 1 error generated.

[Artem-B]

Should be fixed in r259944

[Artem-B]

No, it was not intentional. The wrapper for math functions should only define overloads for functions with signatures not yet provided by CUDA headers in global scope.

[llvmbot]

This is a failed test reduced from SHOC's s3d benchmark. I noticed this test failing after http://reviews.llvm.org/D16593.