This happens when we find identical specializations. Despite the call site pointing to the right clone, we do emit a dead copy (compute.3 in this example).
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/test/Transforms/FunctionSpecialization/identical-specializations.ll | ||
---|---|---|
3 | Don't we avoid old PM syntax in new tests? |
Comment Actions
Chages from last revision:
- used the command line option syntax of the new pass manager
- added a recursive call to demonstrate the issue found in D135463
But the same ought to happen for:
void g(int x, int y) { ... g(x, 2); ... }
...
g(1, 2);
Don't we avoid old PM syntax in new tests?