Code

Added mpi4py RMA support #1373

Code

Open

Com1t wants to merge 29 commits intospcl:mainspcl/dace:mainfrom Com1t:rma_devCom1t/dace:rma_devCopy head branch name to clipboard

+2,910-3

Commits

Commits on Aug 18, 2023

• Added win create library node and its test

  Com1t

  committed
  44e1462

  Copy full SHA for 44e1462

• Added MPI RMA fence library node and its test

  Com1t

  committed
  4ff33c2

  Copy full SHA for 4ff33c2

• Added RMA put library node and its test

  Com1t

  committed
  d3696e0

  Copy full SHA for d3696e0

• Added RMA get library node and its test

  Com1t

  committed
  b92ccbb

  Copy full SHA for b92ccbb

• Removed debug msg in put/get tests

  Show description for 6ebaed5
  Com1t

  committed
  6ebaed5

  Copy full SHA for 6ebaed5

• Updated RMA ops administration in sdfg.py

  Show description for b4e9e65
  Com1t

  committed
  b4e9e65

  Copy full SHA for b4e9e65

• Implemented replacement for MPI RMA win_create, fence, put, get

  Com1t

  committed
  5be0962

  Copy full SHA for 5be0962

Commits on Aug 25, 2023

• Added an extra connector to RMA put,get for ordering

  Com1t

  committed
  ebf8eea

  Copy full SHA for ebf8eea

• Updated amd_rma_ops for better readability

  Com1t

  committed
  6fa9603

  Copy full SHA for 6fa9603

• Added support of different comm world for win_create

  Com1t

  committed
  ce8c173

  Copy full SHA for ce8c173

• Added a synchronization check for RMA put/get

  Com1t

  committed
  1e8606f

  Copy full SHA for 1e8606f

• Updated fence, get, and put tests

  Com1t

  committed
  c4d2ab9

  Copy full SHA for c4d2ab9

Commits on Aug 28, 2023

• Replaced SDFG compile to avoid RuntimeError: Could not load library

  Com1t

  committed
  0a564bb

  Copy full SHA for 0a564bb

• Added RMA accumulate library node, test, and replacement

  Com1t

  committed
  d1f6ba2

  Copy full SHA for d1f6ba2

• Merge branch 'spcl:master' into rma_dev

  Com1t

  authored
  3cbdebe

  Copy full SHA for 3cbdebe

• Added MPI RMA flush, lock, unlock library nodes and their tests for passive sync.

  Com1t

  committed
  e4cc8e3

  Copy full SHA for e4cc8e3

Commits on Aug 29, 2023

• Implemented replacement and their tests for passive sync. nodes

  Com1t

  committed
  f8feae9

  Copy full SHA for f8feae9

Commits on Aug 31, 2023

• Added lock type validator for better MPI compatibility

  Com1t

  committed
  ada74b3

  Copy full SHA for ada74b3

Commits on Sep 6, 2023

• Added a GEMM sample for RMA

  Com1t

  committed
  892dda5

  Copy full SHA for 892dda5

Commits on Sep 7, 2023

• Updated mat_mul to tiled version

  Com1t

  committed
  8149219

  Copy full SHA for 8149219

Commits on Sep 8, 2023

• Implemented distributed version mat_mul.py

  Com1t

  committed
  380b5e4

  Copy full SHA for 380b5e4

• Functionized the distibuted computation in mat_mul.py

  Com1t

  committed
  c847717

  Copy full SHA for c847717

Commits on Sep 11, 2023

• Enabled dace acceleration for mat_mul.py

  Com1t

  committed
  ca65d87

  Copy full SHA for ca65d87

• Added MPI RMA free library node, replacement, and tests for both frontend and backend

  Com1t

  committed
  f7231a0

  Copy full SHA for f7231a0

• Refactored RMA last op checker to a function

  Com1t

  committed
  da62e8e

  Copy full SHA for da62e8e

Commits on Sep 12, 2023

• Added RMA sync logic checking to last op checker

  Com1t

  committed
  54fab2e

  Copy full SHA for 54fab2e

Commits on Sep 14, 2023

• changed the data type of matrix from int32 to float32

  Com1t

  committed
  59f8e72

  Copy full SHA for 59f8e72

• Implemented strong scaling benchmark for mat_mul.py

  Com1t

  committed
  1241422

  Copy full SHA for 1241422

Commits on Sep 15, 2023

• Fixed the window size configuration in window creation

  Com1t

  committed
  e7baaf7

  Copy full SHA for e7baaf7