Fix use-after-free bug in peephole pass#6037
Conversation
| if (!destroyed) { | ||
| for (Block * sub_block : n->blocks()) { | ||
| PeepholeOptimize(sub_block); | ||
| } |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
@pytorchbot add to whitelist |
|
Hi:
Thanks. |
ezyang
left a comment
There was a problem hiding this comment.
Want to know if a test is necessary
|
@ezyang This is very likely covered by tests in onnx-fb-universe, however its CI is not running in ASAN mode. Considering the those tests are going to be moved soon, probably not worth beefing up its CI setup? |
|
I wouldn't suggest updating onnx-fb-universe, but I am wondering why the peephole optimizer (it's NOT onnx) isn't being sufficiently exercised in |
|
Yeah I know it's optimizing the trace. I'm saying there are plenty of export tests in onnx-fb-universe and they are going to be moved and after the move there will be asan build of them. |
* Fix use after free bug in peephole pass * Move the loop befor the switch
@ezyang @apaszke