Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

reimplementation of gpu_count #3718

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
effigies merged 3 commits into nipy:master from mauriliogenovese:FIX-gputil-error-on-window
Mar 19, 2025

Conversation

@mauriliogenovese
Copy link
Contributor

@mauriliogenovese mauriliogenovese commented Mar 4, 2025

fix #3717

I tried to write a simpler implementation of getGPUs from gputil package.
I used a separate file to include their license.

This need to be tested under windows with Python 3.12+ and I don't have such system, so it's untested for now.

Copy link

codecov bot commented Mar 4, 2025 •
edited
Loading

Codecov Report

Attention: Patch coverage is 62.50000% with 6 lines in your changes missing coverage. Please review.

Project coverage is 73.05%. Comparing base (fe20d45) to head (c189912).
Report is 20 commits behind head on master.

Files with missing lines Patch % Lines
nipype/utils/gpu_count.py 60.00% 6 Missing ⚠️
Additional details and impacted files 
@@ Coverage Diff @@
## master #3718 +/- ##
==========================================
- Coverage 73.07% 73.05% -0.03% 
==========================================
 Files 1278 1279 +1 
 Lines 59406 59414 +8 
==========================================
- Hits 43411 43404 -7 
- Misses 15995 16010 +15

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

Copy link
Member

effigies commented Mar 5, 2025

We need to remove gputil from the dependencies.

I'm not terribly worried about Windows, as there are many years of development since last time anybody tried to get nipype working on Windows.

Copy link
Contributor Author

mauriliogenovese commented Mar 5, 2025 •
edited
Loading

We need to remove gputil from the dependencies.

Done

I'm not terribly worried about Windows, as there are many years of development since last time anybody tried to get nipype working on Windows.

I meant that I don't have a computer without a nvidia GPU to test if this code fix the error reported, I assumed it was on Windows. So even a test on other OS may be useful.

Comment on lines 28 to 55
from subprocess import Popen, PIPE
import os


def gpu_count():
try:
if platform.system() == "Windows":
nvidia_smi = shutil.which('nvidia-smi')
if nvidia_smi is None:
nvidia_smi = (
"%s\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe"
% os.environ['systemdrive']
)
else:
nvidia_smi = "nvidia-smi"

p = Popen(
[nvidia_smi, "--query-gpu=name", "--format=csv,noheader,nounits"],
stdout=PIPE,
)
stdout, stderror = p.communicate()

output = stdout.decode('UTF-8')
lines = output.split(os.linesep)
num_devices = len(lines) - 1
return num_devices
except:
return 0
Copy link
Member

@effigies effigies Mar 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If the goal is to keep it as close to the source as possible, that's fine. Here are some cleanups for your consideration:

  1. Use shutil.which() unconditionally. We can detect a FileNotFoundError before calling the process and simply return 0.
  2. Use subprocess.run, which is the recommended API. By using text mode, we don't have to handle decoding or newline normalization.
  3. Use targeted errors. FileNotFoundError and PermissionError (both OSErrors) should catch failures to run. UnicodeDecodeError will catch bad output.
  4. Use p.output.splitlines() to get content-full lines. A trailing newline does not produce an empty string at the end of the list.
Suggested change
from subprocess import Popen, PIPE
import os
def gpu_count():
try:
if platform.system() == "Windows":
nvidia_smi = shutil.which('nvidia-smi')
if nvidia_smi is None:
nvidia_smi = (
"%s\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe"
% os.environ['systemdrive']
)
else:
nvidia_smi = "nvidia-smi"
p = Popen(
[nvidia_smi, "--query-gpu=name", "--format=csv,noheader,nounits"],
stdout=PIPE,
)
stdout, stderror = p.communicate()
output = stdout.decode('UTF-8')
lines = output.split(os.linesep)
num_devices = len(lines) - 1
return num_devices
except:
return 0
import subprocess
import os
def gpu_count():
nvidia_smi = shutil.which('nvidia-smi')
if nvidia_smi is None and platform.system() == "Windows":
nvidia_smi = f'{os.environ["systemdrive"]}\\Program Files\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe'
if nvidia_smi is None:
return 0
try:
p = subprocess.run(
[nvidia_smi, "--query-gpu=name", "--format=csv,noheader,nounits"],
stdout=subprocess.PIPE,
text=True,
)
except (OSError, UnicodeDecodeError):
return 0
return len(output.splitlines())

@effigies effigies merged commit 8732bf6 into nipy:master Mar 19, 2025
25 checks passed
effigies added a commit that referenced this pull request Mar 19, 2025 
1.10.0 (March 19, 2025)
New feature release in the 1.10.x series.
This release adds GPUs to multiprocess resource management.
In general, no changes to existing code should be required if the GPU-enabled
interface has a ``use_gpu`` input.
The ``n_gpu_procs`` can be used to set the number of GPU processes that may
be run in parallel, which will override the default of GPUs identified by
``nvidia-smi``, or 1 if no GPUs are detected.
* FIX: Reimplement ``gpu_count()`` (#3718)
* FIX: Avoid 0D array in ``algorithms.misc.merge_rois`` (#3713)
* FIX: Allow nipype.sphinx.ext.apidoc Config to work with Sphinx 8.2.1+ (#3716)
* FIX: Resolve crashes when running workflows with updatehash=True (#3709)
* ENH: Support for gpu queue (#3642)
* ENH: Update to .wci.yml (#3708)
* ENH: Add Workflow Community Initiative (WCI) descriptor (#3608)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@effigies effigies effigies left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

[ENH] Multiproc always looks for GPUs making it impossible to run pipelines on CPU-only machines

AltStyle によって変換されたページ (->オリジナル) /