Code

GPU support with CuPy#37

Code

Open

brownbaerchen wants to merge 21 commits intompi4py:mastermpi4py/mpi4py-fft:masterfrom

brownbaerchen:cupy_implementationbrownbaerchen/mpi4py-fft:cupy_implementationCopy head branch name to clipboard

Commits

Commits on Dec 8, 2023

• Implemented CuPy version of `DistArray`

  

  Thomas

  committed
  a50137f

  Copy full SHA for a50137f

• Cleanup and tests

  

  Thomas

  committed
  e770b56

  Copy full SHA for e770b56

Commits on Dec 12, 2023

• Added `cupy` backend for FFTs

  

  Thomas

  committed
  424fdb1

  Copy full SHA for 424fdb1

• Removed `useCuPy` argument from `NewDistArray`

  

  Thomas

  committed
  44be7bc

  Copy full SHA for 44be7bc

• Fixed tests

  

  Thomas

  committed
  90551b4

  Copy full SHA for 90551b4

Commits on Dec 14, 2023

• Added cupyx-scipy backend

  

  Thomas

  committed
  202f30c

  Copy full SHA for 202f30c

Commits on Dec 18, 2023

• Implemented NCCL for Alltoallw operations

  

  Thomas

  committed
  e5d4b56

  Copy full SHA for e5d4b56

• Minor refactor

  

  Thomas

  committed
  d0a493e

  Copy full SHA for d0a493e

Commits on Dec 20, 2023

• Changed `Alltoallw` communication pattern

  

  Thomas

  committed
  6b47e36

  Copy full SHA for 6b47e36

Commits on Jan 4, 2024

• Added custom MPI based Alltoallw implementation

  

  Thomas

  committed
  21805de

  Copy full SHA for 21805de

Commits on Jan 12, 2024

• Cleanup

  

  Thomas

  committed
  ceccc7d

  Copy full SHA for ceccc7d

Commits on Feb 13, 2024

• Overlapped all communication within a single Alltoallw in NCCL backend

  

  Thomas

  committed
  c3c20ff

  Copy full SHA for c3c20ff

• Cosmetic changes

  

  Thomas

  committed
  405fb4d

  Copy full SHA for 405fb4d

Commits on Feb 23, 2024

• Swapped `array[...]=data` for `cupy.copyto`

  

  Thomas

  committed
  b69a303

  Copy full SHA for b69a303

Commits on Feb 27, 2024

• Removed unnecessary rescaling in CuPy backend

  

  Thomas

  committed
  b9ac0f7

  Copy full SHA for b9ac0f7

Commits on Feb 29, 2024

• Use CUDA graphs in NCCL Alltoallw

  

  Thomas

  committed
  77b96f3

  Copy full SHA for 77b96f3

Commits on Apr 16, 2024

• Removed dependence on CuPy

  

  Thomas

  committed
  75f0343

  Copy full SHA for 75f0343

Commits on May 24, 2024

• Merge remote-tracking branch 'upstream/master' into cupy_implementation

  

  Thomas

  committed
  a920dca

  Copy full SHA for a920dca

Commits on Oct 12, 2024

• Fixes

  

  Thomas

  committed
  7330b6b

  Copy full SHA for 7330b6b

Commits on Apr 4, 2025

• Merge remote-tracking branch 'upstream/master' into cupy_implementation

  brownbaerchen

  committed
  5500da4

  Copy full SHA for 5500da4

• Merge branch 'cupy_implementation' of github.com:brownbaerchen/mpi4py-fft into cupy_implementation

  brownbaerchen

  committed
  a7aeec6

  Copy full SHA for a7aeec6