JIT: Fix profile inconsistency asserts in flow graph optimization#127357
Open
AndyAyersMS wants to merge 4 commits intodotnet:mainfrom
Open
JIT: Fix profile inconsistency asserts in flow graph optimization#127357AndyAyersMS wants to merge 4 commits intodotnet:mainfrom
AndyAyersMS wants to merge 4 commits intodotnet:mainfrom
Conversation
Fix three sources of profile weight inconsistency in fgopt.cpp that cause assert failures under PGO stress modes (JitRandomEdgeCounts, JitRandomGuardedDevirtualization): 1. fgCompactBlock: When compacting a block with a target that had other predecessors removed by earlier transforms, the inherited weight may not match incoming edge weights. Detect this and mark fgPgoConsistent false. 2. fgOptimizeBranchToEmptyUnconditional: When fgRedirectEdge merges two edges of a BBJ_COND (TrueEdge == FalseEdge), the merged edge retains the old likelihood instead of 1.0. Fix by setting likelihood to 1.0 after merge. 3. fgOptimizeBranchToEmptyUnconditional: When decreaseBBProfileWeight would clamp a block's weight to zero, mark fgPgoConsistent false. Also fix three pre-existing misleading JITDUMP messages that said 'Data %s consistent' when marking data as inconsistent. Fixes dotnet#126381. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
github-actions
Bot
added
the
area-CodeGen-coreclr
Contributor
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR addresses assert failures under PGO stress modes by tightening/repairing profile consistency tracking in fgopt.cpp during flow-graph optimizations, and fixes a few misleading JITDUMP messages that reported “consistent” while marking the profile as inconsistent.
Changes:
- In
fgCompactBlock, detects post-compaction mismatches between block weight and summed incoming edge flow and marksfgPgoConsistent = false. - In
fgOptimizeBranchToEmptyUnconditional, fixes the likelihood on a merged conditional edge (true/false now target the same block) by setting it to1.0. - Marks profile inconsistent when
decreaseBBProfileWeightwould clamp a block’s weight, and corrects several “Data %s consistent” JITDUMP strings to say “inconsistent”.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use block->hasProfileWeight() (post-inheritWeight) instead of the cached pre-inheritance value, so the check also fires when an unprofiled block inherits profile weight from a profiled target. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Contributor
There was a problem hiding this comment.
Pull request overview
This PR addresses several sources of PGO/profile-weight inconsistency in CoreCLR JIT flow-graph optimizations (fgopt.cpp) that can trigger asserts under PGO stress modes, and corrects misleading debug dump text when the JIT marks profile data as inconsistent.
Changes:
- In
fgCompactBlock, detect cases where compaction can introduce profile-weight mismatches and markfgPgoConsistent = false. - In
fgOptimizeBranchToEmptyUnconditional, correct likelihood afterfgRedirectEdgemerges conditional edges (ensuring the merged edge has likelihood 1.0), and markfgPgoConsistent = falsewhen profile weight clamping would occur. - Fix JITDUMP messages that previously said “Data %s consistent” while transitioning profile data to an inconsistent state.
hasLikelihood() is #ifdef DEBUG only and would break Release builds. Remove the guard and sum getLikelyWeight() unconditionally, since all edges have likelihoods set by this point in compilation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This was referenced Apr 24, 2026
Open
Member
Author
|
/azp run runtime-coreclr libraries-pgo |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Member
Author
|
@dotnet/jit-contrib PTAL A very small set of diffs |
tannergooding
approved these changes
Apr 24, 2026
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
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.
Fix three sources of profile weight inconsistency in fgopt.cpp that cause assert failures under PGO stress modes (JitRandomEdgeCounts, JitRandomGuardedDevirtualization):
fgCompactBlock: When compacting a block with a target that had other predecessors removed by earlier transforms, the inherited weight may not match incoming edge weights. Detect this and mark fgPgoConsistent false.
fgOptimizeBranchToEmptyUnconditional: When fgRedirectEdge merges two edges of a BBJ_COND (TrueEdge == FalseEdge), the merged edge retains the old likelihood instead of 1.0. Fix by setting likelihood to 1.0 after merge.
fgOptimizeBranchToEmptyUnconditional: When decreaseBBProfileWeight would clamp a block's weight to zero, mark fgPgoConsistent false.
Also fix three pre-existing misleading JITDUMP messages that said 'Data %s consistent' when marking data as inconsistent.
Fixes #126381.