Code

Testing: RMA AxpyInterface#34

Code

Open

sg0 wants to merge 112 commits intoelemental:masterelemental/Elemental:masterfrom sg0:testingsg0/Elemental:testingCopy head branch name to clipboard

+9,601-2,226

Commits

Commits on Jun 12, 2014

• modified axpy:localtoglobal code to mimic nblocking consensus as shown in alg2 of T.Hoefler DSDE paper

  sg0

  committed
  969d18d

  Copy full SHA for 969d18d

• forgot to add mpi-3 macros around a variable

  sg0

  committed
  b8e2e8d

  Copy full SHA for b8e2e8d

• preposting receives, alternative to iprobe

  sg0

  committed
  61071e7

  Copy full SHA for 61071e7

Commits on Jun 17, 2014

• contains code to avoid probe overhead for short messages, Jeffs rma interface code sync from his forked repo and defined EL_MPI_EXPERIMENTAL macro

  sg0

  committed
  2d52d96

  Copy full SHA for 2d52d96

Commits on Jun 26, 2014

• committing after a week, removed prepost related stuff because of poor performance; added non-blocking consensus stuff and dummy rma code, code in rmainterface is just some cut-paste, would work on…

  Show description for 28faf25
  sg0

  committed
  28faf25

  Copy full SHA for 28faf25

• merging with upstream

  sg0

  committed
  a9c658c

  Copy full SHA for a9c658c

• forgot a const

  sg0

  committed
  b3e2ec9

  Copy full SHA for b3e2ec9

• populating/fixing rmainterface...

  sg0

  committed
  1f9eb06

  Copy full SHA for 1f9eb06

Commits on Jun 27, 2014

• intermediate commits

  sg0

  committed
  e636c1d

  Copy full SHA for e636c1d

Commits on Jun 30, 2014

• updated macro names - use-barrier to use-nbc, also added provisions to use wait instead of test in nbc version

  sg0

  committed
  aa88f1b

  Copy full SHA for aa88f1b

• comments

  sg0

  committed
  e4a5d6c

  Copy full SHA for e4a5d6c

• forgot to add a constructor

  sg0

  committed
  d64331e

  Copy full SHA for d64331e

Commits on Jul 1, 2014

• added a requestfree function to free spurious requests, also added a variant that waits instead of test in nbc...found a bug - when axpy dim < npes the nbc variant deadlocks...fixing it after some …

  Show description for 6c844d7
  sg0

  committed
  6c844d7

  Copy full SHA for 6c844d7

Commits on Jul 2, 2014

• fixing put

  sg0

  committed
  e12bd08

  Copy full SHA for e12bd08

Commits on Jul 7, 2014

• modifications to fix errors in p/g/a

  sg0

  committed
  f55685f

  Copy full SHA for f55685f

• intermediate commit, cleaning up rma interface

  sg0

  committed
  18598b7

  Copy full SHA for 18598b7

• removed spurious instantiations

  sg0

  committed
  5432b3f

  Copy full SHA for 5432b3f

• added some checks in p/g/a

  sg0

  committed
  8a36dd5

  Copy full SHA for 8a36dd5

• intermediate commit

  sg0

  committed
  26034bd

  Copy full SHA for 26034bd

• added a comment in Get

  sg0

  committed
  46648f9

  Copy full SHA for 46648f9

• fixed dtype of mpi-get, updated comment

  sg0

  committed
  e160ce6

  Copy full SHA for e160ce6

Commits on Jul 8, 2014

• intermediate commit - introduce another variable to track attachments

  sg0

  committed
  3fa7135

  Copy full SHA for 3fa7135

• bugs wrt get-put flags

  sg0

  committed
  54eba3e

  Copy full SHA for 54eba3e

• fixing detach

  sg0

  committed
  cd00d7d

  Copy full SHA for cd00d7d

• added another flag to track detach, as dtor will call detach again, plus some comments

  sg0

  committed
  add6b7d

  Copy full SHA for add6b7d

• temporarily reverting back to something that works...the idea is, when nbc is turned on, we could bypass everything related to handleoms and save memory

  sg0

  committed
  d03441f

  Copy full SHA for d03441f

Commits on Jul 9, 2014

• added rmainterface test in line with axpyinterface, removed all header packing-unpacking logic from rmainterface

  sg0

  committed
  2a914b2

  Copy full SHA for 2a914b2

Commits on Jul 11, 2014

• called flush local after p/g/a, modified put-getvector types and initialization...intermediate commit, acc still does not work because of incorrect indices and window disp

  sg0

  committed
  a759aba

  Copy full SHA for a759aba

• added disp in mpi functions

  sg0

  committed
  7c169ed

  Copy full SHA for 7c169ed

Commits on Jul 16, 2014

• figured out the displacement, Acc contains correct displacement, modify others following the same

  sg0

  committed
  15263ff

  Copy full SHA for 15263ff

• addding interface with no local flush, IMO this is temporary, for some testing

  sg0

  committed
  194298e

  Copy full SHA for 194298e

• fixed acc and flush...next is get

  sg0

  committed
  8cbc72f

  Copy full SHA for 8cbc72f

• trying to fix get, tracked some problem in acc, will fix it soon

  sg0

  committed
  77918b2

  Copy full SHA for 77918b2

• removed op from acc in rmainterface as well as mpi, mpi has overloaded functions for op...btw acc is still not working, trying to fix it now

  sg0

  committed
  51efcae

  Copy full SHA for 51efcae

• intermediate commit to fixing get

  sg0

  committed
  da57bb1

  Copy full SHA for da57bb1

• fixed get

  sg0

  committed
  be45ddf

  Copy full SHA for be45ddf

• trying to fix acc...

  sg0

  committed
  dfe393e

  Copy full SHA for dfe393e

Commits on Jul 18, 2014

• modified one-sided interface significantly, using typemap everywhere instead of mpi_byte...added overloaded/modified interfaces

  sg0

  committed
  b0411ba

  Copy full SHA for b0411ba

Commits on Jul 21, 2014

• fixing rma interface - intmd commit

  sg0

  committed
  c18aab4

  Copy full SHA for c18aab4

• removed const_cast and changed type of put-get vectors from byte to T

  sg0

  committed
  3a4ca35

  Copy full SHA for 3a4ca35

Commits on Jul 22, 2014

• turning off codepath not required for nb consensus

  sg0

  committed
  47862e0

  Copy full SHA for 47862e0

Commits on Jul 23, 2014

• some modifications for nb consensus

  sg0

  committed
  27c8da6

  Copy full SHA for 27c8da6

• restructure detach for nb consensus

  sg0

  committed
  ec43dd3

  Copy full SHA for ec43dd3

• add a progress routine

  sg0

  committed
  b70731e

  Copy full SHA for b70731e

• added strided/vector api, untested...removed scale from put

  sg0

  committed
  5fc36e9

  Copy full SHA for 5fc36e9

Commits on Jul 24, 2014

• committing test codes, further updates upon testing would ensue

  sg0

  committed
  f76e507

  Copy full SHA for f76e507

• some more testing, better debug prints

  sg0

  committed
  5de32c4

  Copy full SHA for 5de32c4

• added HF proxy using original AxpyInterface, got to test well

  sg0

  committed
  9f29de0

  Copy full SHA for 9f29de0

• following AXPYinterface syntax

  sg0

  committed
  c6a067b

  Copy full SHA for c6a067b

Commits on Jul 25, 2014

• exploring an alternate design of nbc, where notification is sent immediately after sends, as opposed to setting flag in detach.

  sg0

  committed
  79cddb9

  Copy full SHA for 79cddb9

Commits on Jul 28, 2014

• made a modification in nb consensus, unless there is a strong reason to send to all PEs a notification about issends being complete and recv before doing anything in detach, not doing it. Furthermo…

  Show description for cbd7ff5
  sg0

  committed
  cbd7ff5

  Copy full SHA for cbd7ff5

• commented out some of the stuff I had for nbc, it will get completely removed eventually....added a LocalAcc function to simulate AxpyGLOBAL_TO_LOCAL...modified the RMA test so that it uses LocalAc…

  Show description for e88c840
  sg0

  committed
  e88c840

  Copy full SHA for e88c840

Commits on Jul 29, 2014

• intermediate commit, removed code paths that required request objects for nbc version (to save memory), updated rma tests

  sg0

  committed
  c8ac446

  Copy full SHA for c8ac446

• removed scale as a parameter from RmaInterface, added a macro to enable derived type creation...

  sg0

  committed
  e24ae01

  Copy full SHA for e24ae01

• compacted acc function, got to test this

  sg0

  committed
  1eea9cd

  Copy full SHA for 1eea9cd

• remove const from Get params, in Get matrix always need to be updatable

  sg0

  committed
  85dde6a

  Copy full SHA for 85dde6a

• move winfree and winunlock in try block

  sg0

  committed
  bca226c

  Copy full SHA for bca226c

Commits on Jul 31, 2014

• removing a barrier in detach, probably not required, we'll see

  sg0

  committed
  e7b44c9

  Copy full SHA for e7b44c9

Commits on Aug 4, 2014

• testing an implementation for nbc, this should be slightly fast, if not incorrect

  sg0

  committed
  b5e1ddb

  Copy full SHA for b5e1ddb

Commits on Aug 5, 2014

• minor

  sg0

  committed
  bc0441a

  Copy full SHA for bc0441a

• fixing nbc

  sg0

  committed
  8436525

  Copy full SHA for 8436525

Commits on Aug 6, 2014

• cp axpyinterface to axpyint2...optimized 2 sided implementation

  sg0

  committed
  062565c

  Copy full SHA for 062565c

Commits on Aug 9, 2014

• intermediate commit for axpy 2.0, to fix rma and orig nbc soon

  sg0

  committed
  793a1a9

  Copy full SHA for 793a1a9

• intermediate commits, got to fix nbc next, then start over with axpy2

  sg0

  committed
  db50fa2

  Copy full SHA for db50fa2

Commits on Aug 10, 2014

• non blocking consensus is possible only for local to global in the current form without too much code change...so made the modification in orig axpyinterface, completely untested as of now

  sg0

  committed
  941a054

  Copy full SHA for 941a054

Commits on Aug 11, 2014

• intermediate commit, fixing send/recv

  sg0

  committed
  718ca7e

  Copy full SHA for 718ca7e

Commits on Aug 12, 2014

• fixing axpy 2.0...

  sg0

  committed
  5dc3697

  Copy full SHA for 5dc3697

• the coordinates are wrong (to-do next), but the structure is taking shape

  sg0

  committed
  78eb494

  Copy full SHA for 78eb494

Commits on Aug 13, 2014

• intermediate commit

  sg0

  committed
  c63fda3

  Copy full SHA for c63fda3

Commits on Aug 14, 2014

• packed/unpacked send/receive for local to global...still bugfixing

  sg0

  committed
  fa86f05

  Copy full SHA for fa86f05

• remove all packed send/recv stuff, falling back to send/recving coordinates as the base version for axpy2...and yeah, this works

  sg0

  committed
  c58be27

  Copy full SHA for c58be27

• updated the nbc version of axpy original, this is better than before

  sg0

  committed
  3b055f3

  Copy full SHA for 3b055f3

Commits on Aug 16, 2014

• modify the ssends to isend for small messages

  sg0

  committed
  02d3b14

  Copy full SHA for 02d3b14

Commits on Aug 18, 2014

• redesigned the nextindex logic completely, I got to do the same thing for rma now

  sg0

  committed
  757eb3c

  Copy full SHA for 757eb3c

• flush modified, snipped some junk code...however, we should move away from win_flush in flush, and use other fine grain synch methods, I am going to soon make some changes in regards...

  sg0

  committed
  924b36c

  Copy full SHA for 924b36c

• separate struct to handle data and coord...we send coord and data separately

  sg0

  committed
  e34ad17

  Copy full SHA for e34ad17

• tweaking rma

  sg0

  committed
  94f8f82

  Copy full SHA for 94f8f82

• removing functionality to store put/acc statuses...will use flush always

  sg0

  committed
  f5dba36

  Copy full SHA for f5dba36

• adding local flush after put-acc

  sg0

  committed
  40de290

  Copy full SHA for 40de290

Commits on Aug 22, 2014

• remove flush local from put/acc

  sg0

  committed
  67624b7

  Copy full SHA for 67624b7

Commits on Aug 28, 2014

• add a barrier after nbc stuff in original axpyinterface; replace issend by isend in axpy2, and remove a one byte send for data request in get, instead we send coordinates and use data request tag i…

  Show description for e4e7218
  sg0

  committed
  e4e7218

  Copy full SHA for e4e7218

Commits on Sep 2, 2014

• axpy2: adding put/get/acc functions which are blocking, the nb ones are Iget, Iput, Iacc etc which requires a flush to complete communication; rma: removing flush i,j,and modify implementation so t…

  Show description for 6b4a6e2
  sg0

  committed
  6b4a6e2

  Copy full SHA for 6b4a6e2

Commits on Sep 10, 2014

• added axpy blocking interface, added waitall in mpi...blocking functions returns incorrect results due to a synch issue, working on it

  sg0

  committed
  55e4d4b

  Copy full SHA for 55e4d4b

Commits on Sep 17, 2014

• added some new functions in mpi, still fixing blocking interfaces...get is completely broken

  sg0

  committed
  af9f3ca

  Copy full SHA for af9f3ca

• three small test cases that tests axpy2 blocking and nb, and rma

  sg0

  committed
  a8a70a8

  Copy full SHA for a8a70a8

• adding a sample makefile for tests, will remove it in future though

  sg0

  committed
  63f2f22

  Copy full SHA for 63f2f22

Commits on Oct 8, 2014

• modified most of axpy2.0, got to fix get tomorrow

  sg0

  committed
  1bfa059

  Copy full SHA for 1bfa059

• fixing handle_ functions

  sg0

  committed
  8d28666

  Copy full SHA for 8d28666

Commits on Oct 9, 2014

• fixed flush implementation...more testing required

  sg0

  committed
  e93dbc6

  Copy full SHA for e93dbc6

• modifying Get

  sg0

  committed
  9c22a83

  Copy full SHA for 9c22a83

Commits on Oct 10, 2014

• fixed get

  sg0

  committed
  628f6b0

  Copy full SHA for 628f6b0

Commits on Oct 31, 2014

• fixed Iget, Get and Flush (for Ig/p/a)

  sg0

  committed
  048875b

  Copy full SHA for 048875b

Commits on Nov 8, 2014

• modified blocking interfaces

  sg0

  committed
  ce206e9

  Copy full SHA for ce206e9

• updated get/acc/put...got to test them with larger PEs

  sg0

  committed
  a9c85ea

  Copy full SHA for a9c85ea

Commits on Nov 10, 2014

• updated readinc to localflush instead of flush remote

  sg0

  committed
  d6eddd6

  Copy full SHA for d6eddd6

Commits on Nov 19, 2014

• introduced rma-blocking interface for put/get and ensured local completion for rma-nb...also introduced collective-flush, Cflush for axpy2, which follows nbc (incorrect usage of which shall lead to…

  Show description for a515315
  sg0

  committed
  a515315

  Copy full SHA for a515315

• forgot to add cflush in header

  sg0

  committed
  2f802f1

  Copy full SHA for 2f802f1

• forgot to add function signatures

  sg0

  committed
  7ba1cce

  Copy full SHA for 7ba1cce

Commits on Nov 20, 2014

• test on data rather than coords

  sg0

  committed
  ef56b30

  Copy full SHA for ef56b30

• added local completion logic

  sg0

  committed
  803305c

  Copy full SHA for 803305c

Commits on Jan 27, 2015

• changed the data structure, now we have coords and matrices as structs...added non-collective flush and some synch functions

  sg0

  committed
  57093d6

  Copy full SHA for 57093d6

Commits on Feb 11, 2015

• modified locally blocking routines Acc/Put - memcpying input Z to an intermediate buffer until I find a clever solution to achieve mpi3_flush_local; updated flush to a noncollective version, introd…

  Show description for 01a44a2
  sg0

  committed
  01a44a2

  Copy full SHA for 01a44a2

Commits on Feb 13, 2015

• added const interfaces, some new functions for rmainterface for request based rma

  sg0

  committed
  5e96051

  Copy full SHA for 5e96051

Commits on Feb 14, 2015

• added some const interfaces, and fixed an error for request based rma ops

  sg0

  committed
  95c9c4b

  Copy full SHA for 95c9c4b

Commits on Feb 23, 2015

• delete test case

  sg0

  committed
  2fb071c

  Copy full SHA for 2fb071c

Commits on Feb 24, 2015

• merge

  sg0

  committed
  85ffff3

  Copy full SHA for 85ffff3

• merge

  sg0

  committed
  6bb2242

  Copy full SHA for 6bb2242

• added a macro for rma axpy, so rma axpy gets enabled when mpi version >=3 and rma axpy macro is enabled

  sg0

  committed
  86f4a80

  Copy full SHA for 86f4a80

• moved axpy related macros from header file to cmake build files

  sg0

  committed
  b235bd2

  Copy full SHA for b235bd2

Commits on Feb 25, 2015

• applied astyle to fix indentation, would do some more commits

  sg0

  committed
  ebb4fa6

  Copy full SHA for ebb4fa6

• final round of indentation checks

  sg0

  committed
  0bf9fc5

  Copy full SHA for 0bf9fc5

• minor indentation

  sg0

  committed
  361de46

  Copy full SHA for 361de46