Re: thanks

[Asher Langton <langton2 at llnl dot gov>]

At 10:55 PM -0400 7/29/05, Jack Howarth wrote:
> Asher,
>    I was able to build gcc-4.1-20050728 with your patch and the
> GTY patch for gfc_array_range_type without any additional gfortran
> regressions popping up. I found that building the cp_test.f90
> test program with "gfortran -fcray-pointer" created a binary
> that executed with no failures. However while I saw the same
> at -O1, I found at -O2 and -O3, the resulting binary had the
> following errors...

Try this updated test program.  I changed the each comparison to a 
function call with a side-effect, and that seems to prevent the 
optimization.  The failures aren't really a bug in gfortran -- either 
front or back end; the optimizer just isn't aware of the aliasing 
introduced by Cray pointers.  When I was putting together the test 
suite, every commercial compiler except IBM's failed on at least some 
of the tests without any extra optimization flags, and even xlf90 
failed at -O4.  So while this behavior isn't ideal, at least gfortran 
is consistent with every other implementation.


-Asher

Attachment: cp_test.f90
Description: Binary data