Issue |
136425
|
Summary |
Widened pointers not optimized back to scalar for vectorized interleaved accesses on RISC-V
|
Labels |
backend:RISC-V,
missed-optimization
|
Assignees |
|
Reporter |
lukel97
|
https://compiler-explorer.com/z/bcnocK139
In the loop f, the loop vectorizer is able to use a scalar address throughout the segmented accesses.
In g, where not every segmented field is written back to, it's kept around as a widened vector of pointers, even though we only use the first element of it.
`legalizeAndOptimizeInductions` in VPlanTransforms.cpp may not be catching this case.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs