Nvfortran 23.3 problems in passing complex parameters to kernels through modules

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

Hello :)
as in the title, I find problems in passing complex parameters to kernels through modules, with nvfortran 23.3 compiler.
Everything is O.K. with real parameters.
The simplest code showing the problem I could write is as follows:


!*******************************************************************************
!*******************************************************************************
MODULE PASSPAR
Implicit None

 Complex(8), Dimension(2), Parameter :: cmplx_par=[(1._8,2._8),(3._8,4._8)]
 Real(8), Dimension(2), Parameter :: real_par=[5._8,6._8]

END MODULE PASSPAR


!*******************************************************************************
!*******************************************************************************
PROGRAM MAIN
USE CUDAFOR
Implicit None
Integer :: istat

INTERFACE
  Attributes(Global) Subroutine kernel_gpu_writepar()
  USE PASSPAR
  Implicit None
  End Subroutine kernel_gpu_writepar
END INTERFACE


Call kernel_gpu_writepar<<<dim3(1,1,1),dim3(1,1,1)>>>()
istat=CudaDeviceSynchronize()

END PROGRAM MAIN

!*******************************************************************************
!*******************************************************************************
Attributes(Global) Subroutine kernel_gpu_writepar()
USE PASSPAR
Implicit None

write(*,*)
write(*,*) cmplx_par(1)
write(*,*) cmplx_par(2)
write(*,*)
write(*,*) real_par(1)
write(*,*) real_par(2)
write(*,*)

End Subroutine kernel_gpu_writepar

Any help is wellcome, thankyou :)

2

This just looks like a bug. I will file an issue with the compiler group. complex(4) works, as do complex(8) parameters when they are not a part of an array.

3

Hello, bleback, thankyou very much for your reply :) So, I will adopt the workaround you indicated where possible, while waiting for the bug to be fixed in future releases of nvfortran

4

This is fixed in our 23.7 release which is now available for download.

5

Thankyou :)