PGF90-F-0000-ICE. Subscript for array is out-of-bounds

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

I’ve hit an ICE that I’m not sure if it’s due to the complex nature of this array declaration or due to something wrong in PGI 10.4. To wit, when I try to compile a certain .f90 file, I get:

PGF90-F-0000-Internal compiler error. Subscript for array is out-of-bounds       0 (modistest.f90: 17)

Now the relevant lines of code to look at this error are:

  1 program modistest
  2 
  3 implicit none
  4 
  5 real, parameter :: min_OpticalThickness = 0.3
  6 
  7 integer, parameter :: numTauHistogramBins = 6
  8 
  9 real :: dummy_real
 10 real, dimension(numTauHistogramBins + 1),      parameter :: &
 11   tauHistogramBoundaries = (/ min_OpticalThickness, 1.3, 3.6, 9.4, 23., 60., huge(dummy_real) /)
 12 
 13 !
 14 ! For output - nominal bin centers and  bin boundaries. On output pressure bins are highest to lowest. 
 15 !
 16 integer :: k, l
 17 real, parameter, dimension(2, numTauHistogramBins) ::   &
 18   nominalTauHistogramBoundaries =                       &
 19       reshape(source = (/ tauHistogramBoundaries(1),    &
 20                           ((tauHistogramBoundaries(k), l = 1, 2), k = 2, numTauHistogramBins), &
 21                           100000.0 /),                    &
 22               shape = (/2,  numTauHistogramBins /) )
 23 
 24 write (*,*) nominalTauHistogramBoundaries
 25 
 26 
 27 end program modistest

Now, I spent a bit of time puzzling over line 17 and I think what it does is reshape this array (let tau(i)= tauHistogramBoundaries(i)):

[ tau(1),  tau(2), tau(2),  tau(3),  tau(3), tau(4), tau(4), tau(5), tau(5), tau(6), tau(6), 100000.0 ]

into this 2x6 array:

( tau(1) tau(2) tau(3) tau(4) tau(5) tau(6)
  tau(2) tau(3) tau(4) tau(5) tau(6) 100000.0 )

So, why is this throwing an ICE?

(I should note that the “real” file this code comes from does compile with ifort 11, so perhaps PGI is a bit strict with this?)

2

Sorry for the double post, but I just replaced the above code-snippet from a large module with a small file that can be run.

3

Hi Matt,

It appears to me that the compiler is having trouble doing the expansion of “(tauHistogramBoundaries(k), l = 1, 2), k = 2, numTauHistogramBins)” within this context. I’ve sent a report to our engineers for further investigation (TPR#16926).

Two possible work-arounds are to either remove “parameter” from “nominalTauHistogramBoundaries” and do the initialization in the executable portion of the code, or manually expand the implied do loop.

For example:

% cat test1.f90


program modistest
   implicit none

   real, parameter :: min_OpticalThickness = 0.3

   integer, parameter :: numTauHistogramBins = 6

   real :: dummy_real
  real, dimension(numTauHistogramBins + 1),      parameter :: &
    tauHistogramBoundaries = (/ min_OpticalThickness, 1.3, 3.6, 9.4, 23., 60., huge(dummy_real) /)

 !
 ! For output - nominal bin centers and  bin boundaries. On output pressure bins are highest to lowest.
 !
 integer :: k, l
 real, parameter, dimension(2, numTauHistogramBins) ::   &
  nominalTauHistogramBoundaries =                       &
       reshape(source = (/ tauHistogramBoundaries(1),    &
!                           ((tauHistogramBoundaries(k), l = 1, 2), k = 2, numTauHistogramBins), &
                          tauHistogramBoundaries(2), tauHistogramBoundaries(2), &
                          tauHistogramBoundaries(3), tauHistogramBoundaries(3), &
                          tauHistogramBoundaries(4), tauHistogramBoundaries(4), &
                          tauHistogramBoundaries(5), tauHistogramBoundaries(5), &
                          tauHistogramBoundaries(numTauHistogramBins),  &
                          tauHistogramBoundaries(numTauHistogramBins),  &
                           100000.0 /),                    &
               shape = (/2,  numTauHistogramBins /) )

 write (*,*) nominalTauHistogramBoundaries


end program modistest

% pgf90 -V10.4 test1.f90
% a.out
   0.3000000        1.300000        1.300000        3.600000
    3.600000        9.400000        9.400000        23.00000
    23.00000        60.00000        60.00000        100000.0

Thanks,
Mat

4

Mat,

I just saw in the TPR list that TPR#16926 was fixed in 10.9 and yet when I run my code snippet above:

pgfortran modistest.f90
PGF90-F-0000-Internal compiler error. Subscript for array is out-of-bounds 0 (modistest.f90: 17)
PGF90/x86-64 Linux 10.9-0: compilation aborted

Is there an option I need to specify to get this to work?

5

Hi Matt,

I must apologize for this since it’s something that I should have caught. The fix for TPR#16926 appears to have been scheduled for inclusion in 10.9 but apparently was not. I have verified this error is resolved in the pre-released 11.0 compiler which is scheduled for release at the beginning of December.

  • Mat
6

Matt,

TPR 16926 - ICE Subscript for array is out-of-bounds

has been corrected in the 11.0 release, available now.

regards,
dave