Details
- Reviewers
nicolasvasilache ftynse rriddle wsmoses
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Would you mind providing a commit description in addition to the one-line summary? A good description explains _why_ this new functionality is useful, rather than what the change does.
mlir/lib/Dialect/SCF/SCF.cpp | ||
---|---|---|
719 | Nit: could you put the results of all std::get<N>(it) into named variables? It would make the code more readable. | |
mlir/test/Dialect/SCF/canonicalize.mlir | ||
523 | Nit: could this match .* instead of the specific regexp? | |
526 | I don't think it's important for this code to be on the next line, so let's just use CHECK and avoid over-constraining the tests. | |
536 | It looks like this tests for a sequence of canonicalizations. Could we in addition have a test that exercises only the new canonicalization you are adding? This helps bisecting problems. |
mlir/lib/Dialect/SCF/SCF.cpp | ||
---|---|---|
720 | Is there a reason why this has to be constant for this to apply. This looks like LICM to me and otherwise doesn't need the constantop req? |
This is a wrong approach in using the pattern rewrite see: https://mlir.llvm.org/docs/PatternRewriter/#restrictions. I will close this.
Nit: could you put the results of all std::get<N>(it) into named variables? It would make the code more readable.