Code

Support CUDA Graphs#68

Code

Closed

raimis wants to merge 25 commits intoopenmm:masteropenmm/openmm-torch:masterfrom raimis:cuda_graphraimis/openmm-torch:cuda_graphCopy head branch name to clipboard

+199-67

Commits

Commits on Feb 25, 2022

• Separate the graphable operations

  

  Raimondas Galvelis

  committed
  3e08967

  Copy full SHA for 3e08967

• Drop PyTorch 1.7

  

  Raimondas Galvelis

  committed
  0dc4609

  Copy full SHA for 0dc4609

• Make the energy and force tensors persistent

  

  Raimondas Galvelis

  committed
  885d9ee

  Copy full SHA for 885d9ee

• Set the data pointer during the initialization

  

  Raimondas Galvelis

  committed
  49c01f1

  Copy full SHA for 49c01f1

• Move the reseting of forces

  

  Raimondas Galvelis

  committed
  f152cf2

  Copy full SHA for f152cf2

• Implement the graph capturing

  

  Raimondas Galvelis

  committed
  39f1fcb

  Copy full SHA for 39f1fcb

• Implement properties

  

  Raimondas Galvelis

  committed
  c82e450

  Copy full SHA for c82e450

• Fix the capturing logic

  

  Raimondas Galvelis

  committed
  ead466a

  Copy full SHA for ead466a

• Update the Python bindings

  

  Raimondas Galvelis

  committed
  60990e4

  Copy full SHA for 60990e4

• Improve the enrgy and force tests

  

  Raimondas Galvelis

  committed
  ec904a7

  Copy full SHA for ec904a7

• Enable the testing of CUDA Graph

  

  Raimondas Galvelis

  committed
  41eb439

  Copy full SHA for 41eb439

• Clean up

  

  Raimondas Galvelis

  committed
  c4b6b05

  Copy full SHA for c4b6b05

Commits on Mar 1, 2022

• Merge branch 'openmm:master' into cuda_graph

  raimis

  authored
  5438416

  Copy full SHA for 5438416

Commits on Mar 6, 2022

• Fix the memory leak

  

  Raimondas Galvelis

  committed
  2074975

  Copy full SHA for 2074975

Commits on Mar 9, 2022

• Properly handle the output tensors

  

  Raimondas Galvelis

  committed
  4f16838

  Copy full SHA for 4f16838

• Use just one warm up execution

  

  Raimondas Galvelis

  committed
  19e82f1

  Copy full SHA for 19e82f1

Commits on Mar 10, 2022

• Capture separate kernel with and without the force calculation

  

  Raimondas Galvelis

  committed
  e903a69

  Copy full SHA for e903a69

• Conditionally compile the support of CUDA Graphs

  

  Raimondas Galvelis

  committed
  038dd69

  Copy full SHA for 038dd69

Commits on Mar 11, 2022

• Test just energy calculation

  

  Raimondas Galvelis

  committed
  e4a0732

  Copy full SHA for e4a0732

• Update CI

  

  Raimondas Galvelis

  committed
  2550b5a

  Copy full SHA for 2550b5a

Commits on Apr 12, 2022

• Unify the API

  

  Raimondas Galvelis

  committed
  0baac5f

  Copy full SHA for 0baac5f

• Pass the propery map to the constructor

  

  Raimondas Galvelis

  committed
  1a29e7c

  Copy full SHA for 1a29e7c

• Clean up the graph code

  

  Raimondas Galvelis

  committed
  df76c86

  Copy full SHA for df76c86

Commits on Jun 7, 2022

• Merge branch 'master' into cuda_graph

  

  Raimondas Galvelis

  authored
  ce99dff

  Copy full SHA for ce99dff

Commits on Jun 10, 2022

• Merge branch 'master' into cuda_graph

  

  Raimondas Galvelis

  committed
  7b770fb

  Copy full SHA for 7b770fb