AndrewZhaoLuo commented on code in PR #13163:
URL: https://github.com/apache/tvm/pull/13163#discussion_r1002097212
##########
python/tvm/topi/image/grid_sample.py:
##########
@@ -200,13 +200,13 @@ def _bilinear_sample(n, c, h, w):
def _nearest_sample(n, c, h, w):
y, x = _compute_source_index(n, h, w)
- y_new = te.round(y).astype("int32")
- x_new = te.round(x).astype("int32")
+ y_new = te.nearbyint(y).astype("int32")
Review Comment:
Can you comment a little about this change?
The difference is now behavior can change by setting the rounding mode. Is
this needed?
##########
tests/python/frontend/onnx/test_forward.py:
##########
@@ -5383,6 +5375,12 @@ def verify_eyelike(indata, dynamic=False):
]
}
+flaky_target_skips = {
Review Comment:
Looking at this test:
https://ci.tlcpack.ai/blue/organizations/jenkins/tvm/detail/PR-13163/2/tests
It appears the issue is the `nearbyint` rounding, which is only supported
for llvm targets.
What is the expected rounding mode when using nearbyint. Perhaps you can
just use the corresponding rounding function directly.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
