MINLOC and MAXLOC missing in cuda FORTRAN ???

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hello,

This is quite strange, the reference manual says that minloc and maxloc
are permitted in device subprograms, but when I try to compile this simple
piece of code:

MODULE CUDAMOD

CONTAINS
ATTRIBUTES(DEVICE) SUBROUTINE FINDMAX

real(KIND=4), DIMENSION(3) :: mu
INTEGER, DIMENSION(1) :: ef


mu(1)=1.
mu(2)=2.
mu(3)=3.

ef(1:1)=MINLOC(mu)
ef(1:1)=MAXLOC(mu)



END SUBROUTINE FINDMAX

END MODULE CUDAMOD
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

PROGRAM test
USE CUDAMOD
USE CUDAFOR
integer :: i

i=1


END PROGRAM

I get the following message:

/opt/pgi/linux86-64/11.1/bin/pgf90 -Mcuda=cc13 test.cuf
PGF90-S-0000-Internal compiler error. unsupported procedure 328 (test.cuf: 19)
PGF90-S-0000-Internal compiler error. unsupported procedure 328 (test.cuf: 19)
PGF90-S-0000-Internal compiler error. unsupported procedure 328 (test.cuf: 19)
PGF90-S-0000-Internal compiler error. unsupported procedure 316 (test.cuf: 19)
PGF90-S-0000-Internal compiler error. unsupported procedure 318 (test.cuf: 19)
0 inform, 0 warnings, 5 severes, 0 fatal for findmax

Could it be that those intrinsics were…forgoten???

Benoit.

2

Hi benoit,

I’m not sure if these were forgotten or if there was something preventing their implementation. Either way, we shouldn’t be generating an ICE. I have sent a report to our engineers (TPR#17664) for further investigation.

Thanks,
Mat

3

Hi! Is there still an issue with “MINLOC” and CUDA FORTRAN? On my PGI Accelerator 11.8 x64 version it doesn’t seem to work.

I looked at the following web page:

and request 17664 has not been fixed.

4

and request 17664 has not been fixed.

Correct. Though, I’ve asked our engineers to increase the priority of the TPR.

  • Mat