JIT: enhance SSA checker to check some PHI properties #85533
Conversation
* Ensure PhiArgs have the right local number. * Ensure PhiArgs have unique `gtPredBB` (for most blocks) Handler entries may have multiple PhiArgs with the same `gtPredBB`. This is by design, to model exceptional flow from the middle of the block to a handler. Jump threading relies on there being just a single PhiArg per pred. So exclude handler entry blocks from jump threading. Prep work for possibly changing SSA to produce one PhiArg per pred, instead of one PhiArg per ssa def.
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue Details
Handler entries may have multiple PhiArgs with the same Jump threading relies on there being just a single PhiArg per pred. So exclude handler entry blocks from jump threading. Prep work for possibly changing SSA to produce one PhiArg per pred, instead of one PhiArg per ssa def.
|
|
@jakobbotsch PTAL No diffs expected. |
src/coreclr/jit/fgdiagnostic.cpp
Outdated
| for (Statement* const stmt : block->Statements()) | ||
| { | ||
| if (!stmt->IsPhiDefnStmt()) | ||
| { | ||
| break; | ||
| } |
There was a problem hiding this comment.
We already have NonPhiStatements(); maybe we should have PhiStatements()? (Presumably somebody else would verify all the PHI statements appear first).
| for (Statement* const stmt : block->Statements()) | |
| { | |
| if (!stmt->IsPhiDefnStmt()) | |
| { | |
| break; | |
| } | |
| for (Statement* const stmt : block->PhiStatements()) | |
| { |
There was a problem hiding this comment.
I should probably check that all the PhiDefs appear first, but then I wouldn't need the partial iterator.
|
Failure looks similar to #85082 but the pattern there is different. |
ghost
locked as resolved and limited conversation to collaborators
gtPredBB(for most blocks)Handler entries may have multiple PhiArgs with the same
gtPredBB. This is by design, to model exceptional flow from the middle of the block to a handler.Jump threading relies on there being just a single PhiArg per pred. So exclude handler entry blocks from jump threading.
Prep work for possibly changing SSA to produce one PhiArg per pred, instead of one PhiArg per ssa def.