Vectorization of the loop may result to huge amount of runtime checks and if host machine has limited stack size, Constant Folding will hit the stack overflow and compilation will crash.
The change modifies the algorithm so that in such environment stack overflow won't happen.
Details
Details
- Reviewers
arsenm
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
Can you share how the produced constant expression looks like? Ideally we would not create it in the first place, which we can now do by removing some constant expression operator. I'd be interested in knowing what to target here.