[gmx-developers] GROMACS 4.0.1 still broken on Bluegene

[Roland Schulz]

On Sun, Nov 9, 2008 at 6:51 PM, Erik Lindahl <lindahl at cbr.su.se> wrote:

>
> On Nov 9, 2008, at 11:01 PM, Roland Schulz wrote:
>
>
> I'm not sure. But I think it is little bit different.
>
> Parameters with *different* types are *not allowed* to point to the same
> memory.
> Parameters with *the same* types are *allowed* to point to the same
> memory.
> Parameters with *the same* types declared *restricted* are *not allowed*to point to the same memory.
>
>
> You're probably right, and that makes it even more safe. We had to work a
> bit to fulfill this, in particular in cases where we (for good reasons,
> alignment manipulation) need to covert a pointer to a void, add to it, and
> then move it back to the pointer type (which the Intel compiler considered
> breaking aliasing rules).
>

I think what would help in finding compiler problems is to fix the few
places where the compiler produce warnings and then compile with -Werror by
default. This way people will write bug reports when the use different
compilers. Warning will be overlooked.

Currently I get the following 3 warning for gcc 4.3.1:
futil.c: In function 'get_libdir':
futil.c:412: warning: the address of 'bin_name' will always evaluate as
'true'

gen_ad.c: In function 'gen_pad':
gen_ad.c:512: warning: array subscript is above array bounds
gen_ad.c:514: warning: array subscript is above array bounds
gen_ad.c:516: warning: array subscript is above array bounds
gen_ad.c:517: warning: array subscript is above array bounds
gen_ad.c:516: warning: array subscript is above array bounds
gen_ad.c:514: warning: array subscript is above array bounds
gen_ad.c:512: warning: array subscript is above array bounds
gen_ad.c:517: warning: array subscript is above array bounds
gen_ad.c:512: warning: array subscript is above array bounds
gen_ad.c:514: warning: array subscript is above array bounds
gen_ad.c:516: warning: array subscript is above array bounds
gen_ad.c:517: warning: array subscript is above array bounds

resall.c: In function 'read_atype':
resall.c:73: warning: the address of 'buf' will always evaluate as 'true'
resall.c:77: warning: the address of 'buf' will always evaluate as 'true'

BTW: Of course -Wall with -fstrick-align gives warnings not with
-fnostrick-align. I had that the other way around in the last mail.

Roland


-- 
ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20081109/57af32f8/attachment.html>