[Mpi3-bwcompat] notes from MPI_Count discussion

Jeff Squyres jsquyres at [hidden]
Tue Feb 8 15:11:33 CST 2011



section 2.5.8

...to overcome this *limitation*.

p29:9
MPI_COUNT -> MPI_Count

p123:32
Move "The following auxillary..." down.

p96:16
We don't *return* the value MPI_UNDEFINED; we assign it to the OUT
variable size.

p96:16
also dont return size from function

p102:44
needs to mention get_elements_x

p103:27
missing an "and"

103:35
"returns" is listed twice

p125
there was some allergic reactions to the pack_size changes
drop change on line 4
keep change at line 7

p144:27
recvtype should be pluaral
int recvcounts[] (not *)

p155:12-13
t in front of the j

p155:38
missing a "dot" (multiply) before the extent

--> do we need to fix get_extent / get_true_extent to also return
    count types?  (adam thinks that MPI_Aint is not guaranteed to be
    large enough -- may be problematic for files)  Quincy chimes in
    later that he thinks Adam is right.

p161:45-46
pointer types should be array types

p194:4
indent is screwed up

p194:15-16
t is in front of the j

Follow up with Pavan: he doesn't think that the W collectives 

traw vote: as it stands
- yes: 14
- no: 3 -- dave g, pavan, ...? (adam?)
  - pavan: hates w collectives
  - dave: hates w collectives
  - adam: thinks critical pieces missing (elements) -- and others that might be missing, too
- abstain: 6 -- didn't get all of them, but saw:
  - 2 from OSU (Sayantan and the OSU grad student)
  - new IBM guy
  - ...?


-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/




More information about the Mpi3-bwcompat mailing list