Details
Details
- Reviewers
llvm-commits ChuanqiXu fhahn nikic efriedma
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
This is a proof of concept for RFC: Should we enable Function Specialization?, not ready for review.
llvm/lib/Transforms/Utils/SCCPSolver.cpp | ||
---|---|---|
1713 | This is something weren't checking before. There can be self-referencing PHI nodes which get replaced and then we are visiting them, which results in a crash. | |
1718–1724 | Also this should happen first and then visit the users, otherwise the solver keeps invalid lattice state in the meantime. |
Comment Actions
- rebased
- remove lattice value of deleted globals (fixes https://reviews.llvm.org/D128827)
Comment Actions
Changes from last revision:
- replace ssa_copy intrinsics on each iteration of Function Specialization
- Clear the Solver's worklist before each iteration of Function Specialization
This is something weren't checking before. There can be self-referencing PHI nodes which get replaced and then we are visiting them, which results in a crash.