The ABINIT Users Mailing List ( CLOSED )

[Matthieu Verstraete <mjv500@york.ac.uk>]

Hello all

There are several datastructures in 5.1.2 (still in 5.1.3) which make the 
pgi pghpf compiler (on linux FC4 x86_64 AMD athlon64) crash:

PGHPF-F-0000-Internal compiler error. mk_mem_ptr_shape: extnt not subs 
1307 (../../../src/01managempi/distrb2.F90: 83)
PGHPF/any Linux/x86-64 6.1-4: compilation aborted

The crash has been listed by pgi, with no comments or anything, and no 
solution. All of the incriminated lines involve a pointer in a 
datastructure (MPI or the pseudopotential ones in defs/m_pseudo_types.F90)
There are other ones which are fine, but these look like they are only 
conditionally used and allocated.

rp%data                  in defs/m_pseudo_types.F90: l. 111 119 122 124
mpi_enreg%nplanes_fft    in 01managempi/clnmpi_fft.F90: l. 101
mpi_enreg%ind_fft_planes in 01managempi/clnmpi_fft.F90: l. 102
      ???                in 01managempi/distrb2.F90: l. 83
      ???                in 01managempi/distrb2.F90: l. 90
mpi_enreg%proc_distrb    in 01managempi/distrb2.F90: l. 207
                         in 01managempi/distrb2.F90: l. 212
and mpi_enreg%proc_distrb_para in the same.


OK, now this is messed up: with pgf90 everything looks fine! Could be just 
an internal pgi issue. They must have tried too hard to optimize the 
compiler... However, if you do use pgi, be warned...

Is the code not standard f90 or f95?

Matthieu

--
================================================================
Dr. Matthieu Verstraete                 mailto:mjv500@york.ac.uk
Dept. of Physics, University of York,     tel: +44 1904 43 22 08
Heslington, YO10 5DD York, United Kingdom fax: +44 1904 43 22 14