Looks great to me. Still wonder how we manage to get all the test up to
now ...
george.
On Tue, 6 Dec 2005, Jeff Squyres wrote:
> George -- can you review r8379?
>
>
> On Dec 6, 2005, at 9:30 AM, Tim S. Woodall wrote:
>
> > Most definately. Any large sends to self would not deliver
> > the correct data, exhibited in NPB2.3.
> >
> > Tim
> >
> >
> > Jeff Squyres wrote:
> >> Tim --
> >>
> >> Does this need to come over to v1.0?
> >>
> >>
> >> On Dec 5, 2005, at 6:36 PM, twood...@osl.iu.edu wrote:
> >>
> >>
> >>> Author: twoodall
> >>> Date: 2005-12-05 18:36:33 -0500 (Mon, 05 Dec 2005)
> >>> New Revision: 8379
> >>>
> >>> Modified:
> >>> trunk/ompi/mca/btl/self/btl_self.c
> >>> Log:
> >>> fix send to self for large messages
> >>>
> >>>
> >>> Modified: trunk/ompi/mca/btl/self/btl_self.c
> >>> ===================================================================
> >>> --- trunk/ompi/mca/btl/self/btl_self.c 2005-12-03 15:38:42 UTC (rev
> >>> 8378)
> >>> +++ trunk/ompi/mca/btl/self/btl_self.c 2005-12-05 23:36:33 UTC (rev
> >>> 8379)
> >>> @@ -201,7 +201,7 @@
> >>> int rc;
> >>>
> >>> /* non-contigous data */
> >>> - if(ompi_convertor_need_buffers(convertor) || max_data <
> >>> mca_btl_self.btl_max_send_size ) {
> >>> + if(ompi_convertor_need_buffers(convertor) || max_data <
> >>> mca_btl_self.btl_max_send_size || reserve != 0) {
> >>> MCA_BTL_SELF_FRAG_ALLOC_SEND(frag, rc);
> >>> if(NULL == frag) {
> >>> return NULL;
> >>> @@ -237,7 +237,7 @@
> >>> return NULL;
> >>> }
> >>> frag->segment.seg_addr.pval = iov.iov_base;
> >>> - frag->segment.seg_len = reserve + max_data;
> >>> + frag->segment.seg_len = max_data;
> >>> frag->base.des_src = &frag->segment;
> >>> frag->base.des_src_cnt = 1;
> >>> frag->base.des_dst = NULL;
> >>> @@ -326,7 +326,7 @@
> >>> mca_btl_base_segment_t* dst = des->des_dst;
> >>> size_t src_cnt = des->des_src_cnt;
> >>> size_t dst_cnt = des->des_dst_cnt;
> >>> - unsigned char* src_addr = dst->seg_addr.pval;
> >>> + unsigned char* src_addr = src->seg_addr.pval;
> >>> size_t src_len = src->seg_len;
> >>> unsigned char* dst_addr = dst->seg_addr.pval;
> >>> size_t dst_len = dst->seg_len;
> >>>
> >>> _______________________________________________
> >>> svn-full mailing list
> >>> svn-f...@open-mpi.org
> >>> http://www.open-mpi.org/mailman/listinfo.cgi/svn-full
> >>
> >>
> >>
> >> --
> >> {+} Jeff Squyres
> >> {+} The Open MPI Project
> >> {+} http://www.open-mpi.org/
> >>
> >>
> >>
> >> _______________________________________________
> >> devel mailing list
> >> de...@open-mpi.org
> >> http://www.open-mpi.org/mailman/listinfo.cgi/devel
> >>
> > _______________________________________________
> > devel mailing list
> > de...@open-mpi.org
> > http://www.open-mpi.org/mailman/listinfo.cgi/devel
>
>
> --
> {+} Jeff Squyres
> {+} The Open MPI Project
> {+} http://www.open-mpi.org/
>
>
>
> _______________________________________________
> devel mailing list
> de...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>
"We must accept finite disappointment, but we must never lose infinite
hope."
Martin Luther King
