Strange compilation failure....

Hi. I have a piece of device code that is calling a device function:

          Z(1) =-1.0
          Z(2) = Zi

!          Z = (/ -1.0, Zi /)

          lnL  = r8_coulog( Ep, n, T, A, Z, Bmag )

It works fine. If however, I comment out the two lines that set Z, and uncomment the line that sets Z as a single line of code:

!          Z(1) =-1.0
!          Z(2) = Zi

          Z = (/ -1.0, Zi /)

          lnL  = r8_coulog( Ep, n, T, A, Z, Bmag )

I get the following compiler error:

pgfortran -fast wrapper.cuf prec_mod.cuf rand_mod.cuf curk4_mod.cuf MersenneTwister_kernel.o -mp
/tmp/pgcudaforLRogVcmMvbuE.gpu(141): error: identifier "z_c_0" is undefined

1 error detected in the compilation of "/tmp/pgnvd0SogCGIz-c5r.nv0".
PGF90-F-0000-Internal compiler error. pgnvd job exited with nonzero status code       0 (curk4_mod.cuf: 1243)
PGF90/x86-64 Linux 10.2-1: compilation aborted

Rather disturbingly, I do have some code that uses Z_C_0 as a parameter, but it’s on a different machine on a different cluster, so I don’t think this is a problem with my code. Also, the code works OK with the cuda emulator.

Any ideas?

Hi Rob,

Looks like a compiler bug to me. I was able to recreate the problem here and have filed a technical problem report (TPR#17080) and sent it to our engineers.

Thanks for the report,


Some late book-keeping. TPR 17080 was fixed in the 11.1 release of the compilers.