cudaGetLastError not compiling

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello again. I’m afraid I’ve run into a strange little issue. I cannot seem to compile code containing cudaGetLastError as it says
NVFORTRAN-S-0155-Illegal call from host code to device subprogram __pgi_get_last_error

I’m using the cudafor package at the top and I’m compiling with
nvfortran -Mcuda -o exampleFile exampleFile.f90
and I’m using nvfortran 20.11-0 LLVM 64-bit target on x86-64 Linux -tp skylake

The code more or less looks like this

program exampleProgram
use cudafor
use exampleModule

integer :: errorID
call exampleKernal<<<>>>
errorID = cudaGetLastError()
write(,) cudaGetErrorString(cudaGetLastError())
end program exampleProgram

was cudaGetLastError depreciated since 2017 and can no longer be used from host code?

2

do you have

use cudadevice

anywhere in your code?

3

It’s in the module file because I was under the impression it was needed to use clock64. Is that not correct?

4

I don’t think it is required. And it is the source of this problem. Just try removing it.

5

Looks like it’s compiling, so I believe that did it.