Skip to content

JIT: Tolerate edge/successor count mismatch at OSR entry in profile reconstruction #127363

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AndyAyersMS merged 2 commits into from
Apr 24, 2026
Merged

JIT: Tolerate edge/successor count mismatch at OSR entry in profile reconstruction #127363

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9e9a1c9
JIT: Tolerate edge/successor count mismatch at OSR entry in profile r…
AndyAyersMS Apr 23, 2026
50df6b7
Address review feedback: fix typos and clarify JITDUMP message
AndyAyersMS Apr 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 34 additions & 5 deletions src/coreclr/jit/fgprofile.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3917,17 +3917,46 @@ void EfficientEdgeCountReconstructor::PropagateOSREntryEdges(BasicBlock* block,
assert(pseudoEdge != nullptr);
}

assert(nEdges == nSucc);

if ((info->m_weight == BB_ZERO_WEIGHT) || (successorWeight == BB_ZERO_WEIGHT))
// We may not have have the same number of model edges and flow edges.
//
// As in PropagateEdges, this can happen because some BBJ_LEAVE blocks may have
// been missed during our spanning tree walk since we don't know where all the
// finallies can return to just yet (specifically, in WalkSpanningTree, we may
// not add the target of a BBJ_LEAVE to the worklist). Worst case those missed
// blocks dominate other blocks so we can't limit the screening here to specific
// BBJ kinds.
//
// Handle those specially, and also the zero-weight cases, by just assuming
Comment thread
AndyAyersMS marked this conversation as resolved.
Outdated
// equally likely successors.
//
// (TODO: use synthesis here)
//
if ((nEdges != nSucc) || (info->m_weight == BB_ZERO_WEIGHT) || (successorWeight == BB_ZERO_WEIGHT))
{
JITDUMP("\nPropagate: OSR entry block or successor weight is zero\n");
EntryWeightZero();
JITDUMP("\nPropagate: OSR entry block %s, setting outgoing likelihoods heuristically\n",
(nEdges != nSucc) ? "has inaccurate flow model" : "or successor weight is zero");
Comment thread
AndyAyersMS marked this conversation as resolved.
Outdated

weight_t const equalLikelihood = 1.0 / nSucc;

for (FlowEdge* const succEdge : block->SuccEdges())
{
BasicBlock* const succBlock = succEdge->getDestinationBlock();
JITDUMP("Setting likelihood of " FMT_BB " -> " FMT_BB " to " FMT_WT " (heur)\n", block->bbNum,
succBlock->bbNum, equalLikelihood);
succEdge->setLikelihood(equalLikelihood);
}

if ((info->m_weight == BB_ZERO_WEIGHT) || (successorWeight == BB_ZERO_WEIGHT))
{
EntryWeightZero();
}

return;
}

// Transfer model edge weight onto the FlowEdges as likelihoods.
//
assert(nEdges == nSucc);
JITDUMP("Normalizing OSR successor likelihoods with factor 1/" FMT_WT "\n", successorWeight);

for (Edge* edge = info->m_outgoingEdges; edge != nullptr; edge = edge->m_nextOutgoingEdge)
Expand Down
Loading