marcoabreu commented on a change in pull request #11630: [WIP] Fix flaky test
test_deconvolution
URL: https://github.com/apache/incubator-mxnet/pull/11630#discussion_r201808718
##########
File path: tests/python/unittest/common.py
##########
@@ -94,6 +95,23 @@ def random_seed(seed=None):
random.seed(next_seed)
+def assert_raises_cudnn_disabled(assertion_error=False):
+ def test_helper(orig_test):
+ @make_decorator(orig_test)
+ def test_new(*args, **kwargs):
+ cudnn_disabled = (os.getenv('CUDNN_DISABLED') == "ON")
+ if not cudnn_disabled or mx.context.current_context().device_type
== 'cpu':
+ orig_test(*args, **kwargs)
+ else:
+ if assertion_error:
Review comment:
Agree, let's not block that PR. But please add a comment here that this is a
temporary solution and link it with the issue.
Would you mind if we start a thread on dev@ to discuss this topic? I think
we should have a thorough analysis on this topic. I agree that this might cause
a breaking change, but on the other hand, our users might have worked with
wrong operators the entire time which might have caused unknown side effects -
something that's probably even worse than a detailed error message. I'd like
to understand what calculation we're exactly doing if cudnn is not supported.
What are we falling back to and why is that fallback incorrect.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
