[PATCH] D104122: [SLP]Improve vectorization of stores.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev added inline comments.


================
Comment at: llvm/test/Transforms/SLPVectorizer/X86/stores-non-ordered.ll:23
+; CHECK-NEXT:    [[TMP13:%.*]] = mul <2 x i32> [[TMP3]], [[TMP9]]
+; CHECK-NEXT:    [[TMP14:%.*]] = mul <2 x i32> [[TMP6]], [[TMP12]]
 ; CHECK-NEXT:    br label [[BLOCK1:%.*]]
----------------
RKSimon wrote:
> We've managed to increase the number of gathers - which can''t be good....
Not exactly. Before we had 2 masked gathers and 2 gathers of loads. Now we have 4 masked gathers instead, which still can be lowered to gathers if masked gather is not profitable.
Actually, need to improve the cost model for masked gathers and their emission, they do not work correctly in SLP.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104122/new/

https://reviews.llvm.org/D104122