JuliaOptimalTransport / OptimalTransport.jl

Optimal transport algorithms for Julia
https://juliaoptimaltransport.github.io/OptimalTransport.jl/dev
MIT License
93 stars 8 forks source link

Simplify GPU test environment #106

Closed devmotion closed 3 years ago

devmotion commented 3 years ago

This PR is used for debugging (and possibly fixing) the GPU errors we see lately.

I should add that the tests pass locally with CUDA 3.3.0.

coveralls commented 3 years ago

Pull Request Test Coverage Report for Build 960388896


Totals Coverage Status
Change from base Build 928044156: 0.0%
Covered Lines: 462
Relevant Lines: 488

💛 - Coveralls
codecov-commenter commented 3 years ago

Codecov Report

Merging #106 (b04f493) into master (61f9fd7) will not change coverage. The diff coverage is n/a.

Impacted file tree graph

@@           Coverage Diff           @@
##           master     #106   +/-   ##
=======================================
  Coverage   94.67%   94.67%           
=======================================
  Files           7        7           
  Lines         488      488           
=======================================
  Hits          462      462           
  Misses         26       26           

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 61f9fd7...b04f493. Read the comment docs.

devmotion commented 3 years ago

OK, it seems it was caused by a bug in CUDA.jl that is already fixed in the master branch. Since a new version will be released tomorrow AFAIK, let's just wait until the new version is available.

davibarreira commented 3 years ago

I suppose you want to merge this before merging the other PR to guarantee that they all pass the test?

devmotion commented 3 years ago

Yes, and additionally the other PR requires a new version of NNlibCUDA (it is broken with CUDA#master).

davibarreira commented 3 years ago

Have the CUDA issues been solved?

devmotion commented 3 years ago

There hasn't been a new release of CUDA yet.