[mpich2-dev] dataloop bug

[Rob Ross]

Hi Bob,

Back in the days of BG/L, I thought I recalled Gheorghe A. using this  
flag so that the Segment_pack/unpack calls would allow them to stop  
processing at arbitrary byte boundaries (by default the dataloop code  
stops at element boundaries, such as the end of the last double). I  
was sort of expecting that your team would have continued to use this  
flag in the BG/P implementation. Apparently this isn't the case.

No worries; as long as you're passing the tests you're doing fine. I  
was just surprised.

Regards,

Rob

On Aug 8, 2008, at 10:01 AM, Bob Cernohous wrote:

>
> I don't see that we're using ALL_BYTES?  Did I miss something in the  
> code or config?
>
> Bob Cernohous:  (T/L 553) 507-253-6093
>
> BobC at us.ibm.com
> IBM Rochester, Building 030-2(C335), Department 61L
> 3605 Hwy 52 North, Rochester,  MN 55901-7829
>
> > Chaos reigns within.
> > Reflect, repent, and reboot.
> > Order shall return.
>
>
>
> Rob Ross <rross at mcs.anl.gov>
> Sent by: owner-mpich2-dev at mcs.anl.gov
> 08/07/2008 10:01 PM
> Please respond to
> mpich2-dev at mcs.anl.gov
>
> To
> mpich2-dev at mcs.anl.gov
> cc
> Subject
> Re: [mpich2-dev] dataloop bug
>
>
>
>
>
> Hi David,
>
> I think maybe I just responded to mpich2-dev the other day, which
> might explain why you didn't get my response.
>
> I'm a little surprised that this wasn't found by the BG guys, because
> I thought they were also using ALL_BYTES, but from a less-than-cursory
> glance at the code, this appears to fix a real bug.
>
> Thanks David,
>
> Rob
>
> On Aug 7, 2008, at 8:51 PM, David Gingold wrote:
>
> > Rob Ross says that I should send this patch along.  It's a bug that
> > shows up only when ALL_BYTES is turned on in the dataloop code.  (In
> > the stock code it's not turned on.)
> >
> > Thanks for the help today, Rob!
> >
> > -dg
> >
> > ....
> >
> > --- dataloop_create_indexed.c.~1~                 2008-08-05  
> 15:09:54.619072000 -0400
> > +++ dataloop_create_indexed.c                 2008-08-07  
> 14:05:46.049021000 -0400
> > @@ -266,6 +266,8 @@
> >                      /* increase block lengths so they are in  
> bytes */
> >                      tmp_blklen_array[i] *= old_extent;
> >                  }
> > +
> > +                 new_dlp->loop_params.i_t.total_blocks *=  
> old_extent;
> >     }
> >
> >     *dlp_p     = new_dlp;
> >
>
>