GUFA: Don't misoptimize struct/array atomics#8484
Conversation
There was a problem hiding this comment.
I guess we should put these new test cases in a separate "rmw" test file for now so we can ignore it selectively in the fuzzer.
There was a problem hiding this comment.
Oh, are there more known issues? I was hoping this was it...
There was a problem hiding this comment.
Are you fuzzing in a special way? I don't see stuff...
There was a problem hiding this comment.
Oh, I'm fuzzing with a patch that makes the fuzzer generate new RMW operations from scratch. Since that's finding problems, I assume the same problems could be found from initial contents, even if it might take a lot longer.
| if (multibyte) { | ||
| valueContents = PossibleContents::fromType(value->type); | ||
| } |
There was a problem hiding this comment.
Are these multibyte changes a drive-by fix? Are they NFC? This isn't clear to me.
There was a problem hiding this comment.
This uses the new code added for the rmw instrs, as mentioned in the description,
Simplify ArrayStore similarly: remove special multibyte handling and
just pass in an unknown value of the type.
The old code manually saw "multibyte" and applied an unknown value. This sends
it in without a special multibyte param.
2 participants
For now, just write an unknown value of the proper type.
Simplify ArrayStore similarly: remove special multibyte handling and
just pass in an unknown value of the type.
Fixes #8473