-
Notifications
You must be signed in to change notification settings - Fork 2.8k
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
[ROCm] Global (average) Pooling unusable. #15482
Comments
Minimal MIOpenDriver command to reproduce the issue Crash Issue:
Incorrect Issue:
|
@cloudhan are you interested to get a proper fix in MIOpen? |
Why not? |
@cloudhan Ok, I'll keep in informed. |
The correctness issue in MIOpen is fixed in ROCm/MIOpen#2118. |
FYI this is fixed for Forward/Inference pooling in ROCm/MIOpen#2163. |
FYI this is fixed for Backward pooling (except workspace index mask mode for Max pooling) in ROCm/MIOpen#2372. |
Describe the issue
To reproduce
To reproduce a crash
Run the following single node model
will create the following error:
The maxmium shape it can run is
This fix the problem by switching the global pool to use reduction instead.
This problem impact the usability of ROCm EP of our internal users.
To reproduce an incorrect result
Urgency
Must be fixed
Platform
Linux
OS Version
no apply
ONNX Runtime Installation
Built from Source
ONNX Runtime Version or Commit ID
d49a8de
ONNX Runtime API
Python
Architecture
X64
Execution Provider
Other / Unknown
Execution Provider Library Version
ROCm 5.4.2
The text was updated successfully, but these errors were encountered: