MPIR Specification - An Official Companion Document for the MPI Standard

[mpiforumbot]

Originally by schulzm on 2010-08-31 23:27:24 -0500

---

MPIR SpecificationContact: Martin Schulz and Jeff Squyres

Main document author: John DelSignore

Description

The document attached to this ticket describes the MPIR process acquisition interface as it is currently used and implemented in a variety of MPI implementations. The tools WG is proposing to provide this document as an officially side document approved by the MPI forum. Based on the discussions during the June MPI forum meeting, such side documents should go through a similar voting process (details tbd.) as any addition to the MPI standard itself, but it will be published on the MPI forum servers as a separate document.

[mpiforumbot]

Originally by schulzm on 2010-09-01 17:48:43 -0500

---

Attachment added: mpi-report.pdf (284.6 KiB)
MPIR Specification document (draft 9/1/10)

[mpiforumbot]

Originally by jsquyres on 2010-09-16 12:32:37 -0500

---

Attachment added: mpir-specification-2010-09-16.pdf (297.9 KiB)

[mpiforumbot]

Originally by jsquyres on 2010-09-16 12:33:24 -0500

---

New version attached after Forum feedback at the Sep 16 2010 Stuttgart meeting. Changes are noted in red; the final version will simply s/red/black/ text.

[mpiforumbot]

Originally by schulzm on 2010-10-04 13:27:29 -0500

---

Attachment added: mpir-specification-2010-10-04.pdf (336.9 KiB)

[mpiforumbot]

Originally by schulzm on 2010-10-04 13:30:08 -0500

---

New version attached with modifications requested by MPI forum at the Sep 2010 meeting in Stuttgart. Based on the discussions at that meeting, this new version counts as officially read and is hence ready for a first vote.

[mpiforumbot]

Originally by asupalov on 2010-10-07 06:16:50 -0500

---

We identified three possible issues in the current spec, see below.

1. The proposal requires the VOLATILE macro to be defined (p.15, top). What happens if VOLATILE expands to nothing? How will the respective compiler understand that it should go down to the memory every time a variable is queried? Not our problem, but still. :)
2. Intel MPI (and most likely MPICH2) has a bit different definition for MPIR_PROCDESC table (p.15, bottom).

Our defition:
typedef struct MPIR_PROCDESC {
const char *host_name;
const char *executable_name;
int pid;
} MPIR_PROCDESC;

Proposal:
typedef struct MPIR_PROCDESC {
char *host_name;
char *executable_name;
int pid;
} MPIR_PROCDESC;

Our definition seems to be more strict. I don't think this will cause any problem, but you may want to review this and allow for the above extension, as we'd not like to change the source code unless absolutely necessary.

1. Same issue for MPIR_debug_state (p. 17, bottom).

Our definition:
volatile int MPIR_debug_state

Proposal definition:
int MPIR_debug_state

This may be material, as the MPIR_debug_state is set by the program and inspected by the tool. We should probably add VOLATILE to the proposal.

[mpiforumbot]

Originally by jsquyres on 2010-10-12 07:09:08 -0500

---

Attachment added: mpir-specification-10-11-2010.pdf (280.5 KiB)
Added a missing VOLATILE, got first vote

[mpiforumbot]

Originally by jsquyres on 2010-10-12 07:11:09 -0500

---

New attachment added: we missed a "VOLATILE" in the text. Alexander S. and John D. cited some Really Old Stuff that included this volatile, so clearly this is just a typo that we missed it.

The Forum thought it would still be ok to vote on this even though the "VOLATILE" was missed in the first reading because a) this is not part of the official MPI standards document, and b) therefore it's ok to bend the rules for an obvious typo. Hence, we made the change to the document and then it passed its first vote.

[mpiforumbot]

Originally by jsquyres on 2010-12-06 19:20:30 -0600

---

Passed 2nd vote at Forum meeting in San Jose, Dec 6, 2010.

[mpiforumbot]

Originally by jsquyres on 2010-12-06 19:21:12 -0600

---

Setting to "Ticket Complete" because this is not a normal MPI ticket that requires additional text changes, etc.

We're done!