slai-nick opened a new issue, #15227:
URL: https://github.com/apache/tvm/issues/15227
### Expected behavior
No segfault happening during `MergeCompositeFunctions` pass.
### Actual behavior
A segfault occurs, I didn't get any stack trace info unfortunately:
```shell
!!!!!!! TVM encountered a Segfault !!!!!!!
Stack trace:
zsh: segmentation fault python bug.py
```
### Environment
- MacOS Ventura (13.3.1) with M1 chip (also tested on windows + docker,
using an image based on cli_minimal)
- TVM unity branch, pulled as of today
### Steps to reproduce
```python
import numpy as np
from torch import hub, fx
from tvm.relax.frontend.torch import from_fx
from tvm.relax import transform
from tvm.relax.dpl import is_op, wildcard
model = hub.load('pytorch/vision:v0.10.0', 'resnet18',
pretrained=False).eval()
traced_model = fx.symbolic_trace(model)
input_info = [((1, 3, 224, 224), "float")]
ir_mod = from_fx(traced_model, input_info)
patterns = [
("foo.matmul", is_op("relax.matmul")(wildcard(), wildcard())),
]
ir_mod = transform.FuseOpsByPattern(patterns)(ir_mod)
# segault with the following line:
ir_mod = transform.MergeCompositeFunctions()(ir_mod)
```
### Triage
* relax:transform
--
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]
