Add a new datatype creation routine - MPI_Type_create_hindexed_block

[mpiforumbot]

Originally by chaarawi on 2011-06-14 10:19:26 -0500

---

This is a request to add an additional routine to the standard, named MPI_Type_hindexed_block. Most datatype creation functions have an H_version to take byte displacements instead of integer counts of the old type. The create_indexed_block function does not have the H counterpart and should be added.

Extended Scope

N/A

History

None that I am aware of.

Proposed Solution

Add a new routine:

MPI_Type_create_hindexed_block (int count, int blocklength, MPI_Aint array_of_displacements[], MPI_Datatype old_type, MPI_Datatype *newtype )

The existing MPI_Type_create_indexed_block is basically the same, except that the displacement array is of type int (counts of old_type). It does not have a counterpart for byte displacement as all function do.

MPI_Type_create_Hindexed can be used as an alternative, but that takes an array of blocklengths and would be an ugly solution for having to create a block_length array of 1s.

Impact on Implementations

just add a new routine to the standard, and the implementation should be fairly straightforward.

Impact on Applications / Users

None

Alternative Solutions

Entry for the Change Log

Added MPI_Type_create_hindexed_block

[mpiforumbot]

Originally by RolfRabenseifner on 2011-10-21 15:35:36 -0500

---

The attached file draft_nbcio.pdf is the wrong file. It is for ticket #273.

[mpiforumbot]

Originally by chaarawi on 2011-10-21 15:37:47 -0500

---

Replying to RolfRabenseifner:

The attached file draft_nbcio.pdf is the wrong file. It is for ticket #273.

woops.. doubled clicked instead of 1 click

[mpiforumbot]

Originally by moody20 on 2011-10-25 17:38:12 -0500

---

For the draft, please use the macros so that ticket number shows up with the changes.

Missing table for MPI_COMBINER_HINDEXED_BLOCK around page 35, line 44.

[mpiforumbot]

Originally by moody20 on 2011-10-25 17:39:39 -0500

---

Highlighting for hindexed_block is missing on page 31, line 23.

[mpiforumbot]

Originally by moody20 on 2011-10-25 17:49:59 -0500

---

Using MPI-2.2 as reference for page and line numbers...

Add as non-portable type in section 2.4, page 12, line 22.

Add to datatype decoding constants in Annex a, page 522, line 14.

Be sure that function shows up in language bindings, e.g., A.2 C Bindings, page 533, line 30.

[mpiforumbot]

Originally by chaarawi on 2011-10-27 12:16:29 -0500

---

Attachment added: MPIForum_hindexed_block.pptx (45.7 KiB)
ppt presented at october 11 meeting

[mpiforumbot]

Originally by chaarawi on 2011-12-07 10:06:06 -0600

---

Attachment added: draft_hindexedb.pdf (2123.2 KiB)
Draft for formal reading

[mpiforumbot]

Originally by jsquyres on 2012-01-06 16:56:04 -0600

---

We should not add the C++ value MPI::COMBINER_HINDEXED_BLOCK (p562 in the PDF).

[mpiforumbot]

Originally by bosilca on 2012-03-05 19:00:03 -0600

---

Reviewed positively.

[mpiforumbot]

Originally by koziol on 2012-03-05 20:49:58 -0600

---

Looks good to me.

[mpiforumbot]

Originally by bosilca on 2012-05-29 02:55:30 -0500

---

Attachment added: hindexed_block.patch (11.4 KiB)

[mpiforumbot]

Originally by bosilca on 2012-05-29 02:56:11 -0500

---

Patch based on Open MPI revision 26506.

[mpiforumbot]

Originally by chaarawi on 2012-06-20 09:00:15 -0500

---

Attachment added: ticket_280.diff (5.2 KiB)
diff

[mpiforumbot]

Originally by chaarawi on 2012-06-28 12:11:24 -0500

---

Attachment added: draft_hindexedb_sj.pdf (2349.6 KiB)

[mpiforumbot]

Originally by RolfRabenseifner on 2012-06-28 13:28:04 -0500

---

appLang and terms committed (SVN 1197)[[BR]]
appLang corrected (no new C++ entities) (SVN 1199)

[mpiforumbot]

Originally by jsquyres on 2012-07-03 07:12:59 -0500

---

appLang and T&C commits reviewed by Jeff.

[mpiforumbot]

Originally by jsquyres on 2012-07-03 07:14:28 -0500

---

Sorry -- I incorrectly moved this to "Waiting for PDF reviews", but we're actually still waiting for the datatype chapter commits. So I moved it back to "Passed".

[mpiforumbot]

Originally by balaji on 2012-07-06 23:27:15 -0500

---

Second implementation also added in MPICH2 in r9988.

http://www.mcs.anl.gov/research/projects/mpich2/downloads/tarballs/nightly/trunk/

[mpiforumbot]

Originally by bosilca on 2012-07-11 11:28:08 -0500

---

Applied to the "Datatype" chapter (r1263).

[mpiforumbot]

Originally by moody20 on 2012-07-17 00:42:17 -0500

---

REVIEW (r1270.pdf vs https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/ticket/280/draft_hindexedb_sj.pdf): ok

[mpiforumbot]

Originally by jsquyres on 2012-07-18 09:59:17 -0500

---

Appears to be fully committed. Moving to "Waiting for PDF reviews".

[mpiforumbot]

Originally by jitumoto on 2012-07-18 11:33:14 -0500

---

Reviewed in the r1390 PDF (vs draft_hindexdb_sj.pdf on this ticket) : This ticket is done.