Re: bug with f77 on ppc egcs-1.1 prerelease

[Alan Reynolds <alan1 at bellatlantic dot net>]

I've been looking into this problem now for about 3 weeks.  It has changed
a
little as a result of some changes made to the reload code and the machine

description and has finally settled down to the reported failure.  Someone

also reported the same bug except it occured in a C++ compilation and I've
been
looking at it because it is also appearing in an Ada program (gnatprep).
I'm certain the
problem is in reload.  The problem generally doesn't occur when using
-O0 and -Os.  The problem seems to be correlated with very large stack
frames due to the presence of large array variable declarations.  The
failure usually occurs in
shorten_branches while it is getting the length of a base-offset
load/store that has the
> 32767 offset.

Anyway, I've been swamped of late and haven't had a chance to look at it
seriously for a week.  I should be able to make headway on it tomorrow
night.

Have fun,
Al Reynolds@Home

Craig Burley wrote:

> Greg, thanks so much for taking the time and trouble to test this!
> Hopefully, somebody with access to a ppc machine will look into it.
> (I believe the egcs team is already considering fixing some regressions
> on ppc machines before shipping 1.1.  It would be fortuitous if, in
> doing so, they fixed this bug!)
>
>         tq vm, (burley)