add cudaDeviceCount flag to the request requirements

[ndonyapour]

Hello,

I have been attempting to execute a CUDA-based workflow using the --parallel flag. However, I encountered the following error message:

ERROR Got workflow error: 'cudaDeviceCount'
Traceback (most recent call last):
  File "/home/donyapourn2/actions-runner/_work/workflow-inference-compiler/workflow-inference-compiler/3/envs/globalwic/lib/pypy3.9/site-packages/cwltool/executors.py", line 314, in _runner
    job.run(runtime_context, TMPDIR_LOCK)
  File "/home/donyapourn2/actions-runner/_work/workflow-inference-compiler/workflow-inference-compiler/3/envs/globalwic/lib/pypy3.9/site-packages/cwltool/job.py", line 823, in run
    self._setup(runtimeContext)
  File "/home/donyapourn2/actions-runner/_work/workflow-inference-compiler/workflow-inference-compiler/3/envs/globalwic/lib/pypy3.9/site-packages/cwltool/job.py", line 181, in _setup
    count = cuda_check(cuda_req, math.ceil(self.builder.resources["cudaDeviceCount"]))
KeyError: 'cudaDeviceCount'

Upon investigating the code, it appears that the evalResources function fails to include cudaDeviceCount in the request. I have made necessary changes to fix the error.

[codecov]

Codecov Report

Merging #1895 (e855fed) into main (509ffb9) will increase coverage by 0.00%.
The diff coverage is 75.00%.

@@           Coverage Diff           @@
##             main    #1895   +/-   ##
=======================================
  Coverage   83.95%   83.96%           
=======================================
  Files          46       46           
  Lines        8152     8163   +11     
  Branches     2168     2168           
=======================================
+ Hits         6844     6854   +10     
- Misses        838      839    +1     
  Partials      470      470           

Files Changed	Coverage Δ
cwltool/executors.py	84.51% <71.42%> (+0.23%)	⬆️
cwltool/process.py	92.81% <100.00%> (+0.01%)	⬆️

[mr-c]

Thank you @ndonyapour !

I'm going to see if I can come up with a unit test for this, before merging.

[mr-c]

@ndonyapour can you test my changes with a real workflow using cudaDeviceCountMin and cwltool --parallel?

[ndonyapour]

yes, I tested your changes and it worked!