JIT: fix bug in jump threading #85942
Conversation
Verify that the purported dominator of a block to be jump threaded dominates all of the block's predecessors. This will initially be true (modulo some odd cases with finally continuations), but if we've jump threaded through the dominator it may not remain true as we do not update dominators when optimizing. Fixes dotnet#85892.
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsVerify that the purported dominator of a block to be jump threaded dominates all of the block's predecessors. This will initially be true (modulo some odd cases with finally continuations), but if we've jump threaded through the dominator it may not remain true as we do not update dominators when optimizing. Fixes #85892.
|
|
@jakobbotsch PTAL 7 or so methods with diffs. Did not check to see if any of them actually were creating bad code. |
|
You might wonder how a fix for a possibly incorrect dominator is to make even more dominance queries. So this may not a final or complete fix. I will keep thinking about what sorts of data we can actually trust.... |
|
/azp run Fuzzlyn |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
Fuzzlyn failures seem unrelated. |
ghost
locked as resolved and limited conversation to collaborators
Verify that the purported dominator of a block to be jump threaded dominates all of the block's predecessors.
This will initially be true (modulo some odd cases with finally continuations), but if we've jump threaded through the dominator it may not remain true as we do not update dominators when optimizing.
Fixes #85892.