d-smirnov commented on code in PR #13369:
URL: https://github.com/apache/tvm/pull/13369#discussion_r1024613292
##########
tests/python/contrib/test_ethosu/test_networks.py:
##########
@@ -142,18 +141,8 @@ def
test_networks_with_usmp_and_cascader_wo_striping(accel_type, model_url, work
@pytest.mark.parametrize(
"accel_type, model_url, workspace_size",
[
- # Checks the same test case multiple times to make sure its not flaky
- ("ethos-u55-256", MOBILENET_V1_URL, 1010000),
- ("ethos-u55-256", MOBILENET_V1_URL, 1010000),
- ("ethos-u55-256", MOBILENET_V1_URL, 1010000),
- ("ethos-u55-256", MOBILENET_V1_URL, 1010000),
- ("ethos-u55-256", MOBILENET_V1_URL, 1010000),
- # Checks the same test case multiple times to make sure its not flaky
- ("ethos-u55-256", MOBILENET_V2_URL, 1400000),
- ("ethos-u55-256", MOBILENET_V2_URL, 1400000),
- ("ethos-u55-256", MOBILENET_V2_URL, 1400000),
- ("ethos-u55-256", MOBILENET_V2_URL, 1400000),
- ("ethos-u55-256", MOBILENET_V2_URL, 1400000),
+ ("ethos-u55-256", MOBILENET_V1_URL, 1005312),
+ ("ethos-u55-256", MOBILENET_V2_URL, 1162368),
Review Comment:
Looks like I can confirm instability with the cascade scheduling as in two
subsequent runs of
`tests/python/contrib/test_ethosu/test_networks.py::test_networks_with_usmp_and_cascader_with_striping[ethos-u55-256-model_url1-...`
the _hill_climb_ allocator was supplied at first run with 8 and, at second,
with 7 buffers to allocate, which resulted in different allocated sizes
(1374688 and 1162448 bytes correspondingly, in both cases all the buffers
"conflicting" with each other).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
