larroy commented on a change in pull request #11245:
[MXNET-244][MXNET-523][ARM] improvements to ARMv7 based builds.
URL: https://github.com/apache/incubator-mxnet/pull/11245#discussion_r194971684
##########
File path: tests/python/unittest/test_ndarray.py
##########
@@ -711,9 +711,8 @@ def get_values(ensure_unique):
k=dat_size*dat_size*dat_size*dat_size, is_ascend=False)
assert_almost_equal(nd_ret_argsort, gt)
- # test topk with a big shape
- a = mx.nd.arange(0, 54686454, step=1, repeat=1)
- assert_almost_equal(a.topk(k=54686454).asnumpy(), a.asnumpy()[::-1])
+ a = mx.nd.arange(0, 1024, step=1, repeat=1)
+ assert_almost_equal(a.topk(k=1024).asnumpy(), a.asnumpy()[::-1])
Review comment:
I timeboxed the time to write a CPP test for this, it's not trivial, and
it's out of the scope of my work.
For me, having this test adjusted to a reasonable parameter is a necessary
condition to have ARM support offered. I would suggest to file a ticket to do
this task in the future, move it to nightly, or write it yourself. It's out of
the scope of ARM support to test for an overflow in topk in the tests for
ndarray which have to run on all platforms. I'm not submitting this PR without
this adjustment to this test, because without this change, ndarray can't be
tested in most ARMv7 platforms which don't have multiple gigabytes of RAM.
Is up to the community to decide what's more important, ARM support or
sneakily testing for an overflow which was fixed. In any case this should be
part of a regression test suite and not part of the main tests which need to
run with reasonable resources.
----------------------------------------------------------------
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
