Keep phis contiguous after inverting a phi's incoming values - #3126
Open
ZuseZ4 wants to merge 1 commit into
Open
Keep phis contiguous after inverting a phi's incoming values#3126ZuseZ4 wants to merge 1 commit into
ZuseZ4 wants to merge 1 commit into
Conversation
In invertPointerM's PHINode case, inverting an incoming value can emit instructions into the phi's own block above the shadow phi just created -- materializing the shadow of a value defined in the loop header inserts an alloca, a store and byte-wise zeroing between the canonical IV phi and the shadow phi. That leaves the block's phis no longer contiguous at the block start, and the resulting IR is rejected: Instruction does not dominate all uses! PHI nodes not grouped at top of basic block! function failed verification One misplacement produces both complaints: the cache store for the primal phi is then emitted at getFirstNonPHI(), which now precedes that phi's definition. Re-group the block's phis at the block start once the incoming values have been inverted. Note this restores the invariant rather than preventing the insertion; the cleaner fix is to correct whichever builder positions the shadow-materialization machinery above the phis. Reduced test case reproducing the failure, and the instrumented-build evidence localizing it to this site, are in the librint repro/ notes (e11-reduced.ll, e11-verifier-phi.rs). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
In invertPointerM's PHINode case, inverting an incoming value can emit instructions into the phi's own block above the shadow phi just created -- materializing the shadow of a value defined in the loop header inserts an alloca, a store and byte-wise zeroing between the canonical IV phi and the shadow phi. That leaves the block's phis no longer contiguous at the block start, and the resulting IR is rejected:
Instruction does not dominate all uses!
PHI nodes not grouped at top of basic block!
function failed verification
One misplacement produces both complaints: the cache store for the primal phi is then emitted at getFirstNonPHI(), which now precedes that phi's definition.
Re-group the block's phis at the block start once the incoming values have been inverted.
Note this restores the invariant rather than preventing the insertion; the cleaner fix is to correct whichever builder positions the shadow-materialization machinery above the phis.
Workaround for #3029, though you might want to find the proper cause @wsmoses