[llvm] [VPlan] Don't convert widen recipes to VP intrinsics in EVL transform (PR #126177)

Alexey Bataev via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 13 12:55:57 PST 2025


alexey-bataev wrote:

> I could easily miss some decision of the EVL vectorizer as I rarely check what's going on there. So if you and @alexey-bataev @Mel-Chen @arcbbb @ppenzin @preames agreed to come back to proposal to convert instructions to vp-intrinsics after vectorizer that change seems ok. Otherwise I do see that change a step backward: it's known that at the beginning EVL vectorization won't be perfect and passes such InstCombine need to be updated. But overall hardware that have vl-dependent RVV implementation will benefit of having proper EVL set.

The patch relies on VLOptimizer, which (is supposed, at least) should optimize VL for the instructions

https://github.com/llvm/llvm-project/pull/126177


More information about the llvm-commits mailing list