If target returns false in TTI.prefersVectorizedAddressing(), it means the address registers will not need to be extracted. Therefore, there should be no operands scalarization overhead for a load instruction.
Test for SystemZ.
Differential D52351
[LoopVectorizer] Fix in getScalarizationOverhead() jonpa on Sep 21 2018, 4:53 AM. Authored by
Details
Diff Detail Event TimelineComment Actions Thanks for review. Updated just the test to make it more clear that the address computation is in fact scalar (the multiply). Comment Actions Thanks for review. I am waiting with committing this until https://reviews.llvm.org/D52417 is approved, which is a simple fix I found necessary after applying this patch. |