Samantha Andow

> [@samdow](https://github.com/samdow), do we have an issue created for this, so we can link it for future tracking? No but I'm planning to add this today thanks to talk with...

From offline discussion with @zou3519: - As written (with a Python list as the stack and no threading library), this is definitely not thread safe (tested locally with the threading...

I'm not certain what's happening with the lint but @zou3519 and @ezyang, I added some comments to the real meat of what's changed if you're interested

@pytorchbot merge

@pytorchbot merge

So I don't forget: dynamo failure is an upstream issue. So this will wait on this https://github.com/pytorch/pytorch/pull/85225

@pytorchbot merge

@pytorchbot merge -g

On A100s, seeing 107ms to 111ms ~4% regression

On AWS V100s, I'm seeing 169ms to 174ms which is ~3% regression