xplor-nih gfortran build problem

[Asher Langton]

Jack Howarth wrote:
> Asher,
>     Well I was referring to the cray-pointer patch in general as that is
> where we are getting the support for the LOC call in gfortran. 

They were part of the same patch, but the Cray pointer code is entirely
separate from the loc intrinsic.

> If I recall
> correctly, you had gotten access to the xplor-nih source yourself. Could
> you try building xplor-nih 2.14 with gfortran? 

I had the code, but I don't think I have it any more.  I never managed
to compile xplor-nih myself; as I recall, the compile scripts were quite
complicated.

> You will find that the
> xrama.inp test always fails when xrama.f is compiled any optimization
> greater than -O0. The xrama.f code has a nested set of calls which pass
> heap pointers as their parameters. I tried running the code through the
> guardmalloc library and it doesn't trigger anything. However I know
> heap corruption must be going on since simple actions like adding
> write statements in the middle of the xrama.f routines can dramatically
> change the results.

I'd be surprised if the loc intrinsic is causing that, particularly
since it works correctly at -O0.  There's no reason why (correct)
optimization should affect the result of loc.  Can you distill this to
an example where loc returns an incorrect value?


-Asher