[PATCH] D143785: [X86] Add Extend shuffle pattern to vNf32 shuffles.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:16939
+
+    if (SimpleOnly && (UseHiV1 || UseHiV2))
+      return false;
----------------
SimpleOnly is guaranteed to be true here

Maybe just `return !(UseHiV1 || UseHiV2);`


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:17210
+  int Size = Mask.size();
+  int LaneSize = Size / 2;
+  InLaneMask = SmallVector<int>(Mask);
----------------
Don't hardcode the LaneSize calculation - it prevents us using for other vector sizes in the future.


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:17211
+  int LaneSize = Size / 2;
+  InLaneMask = SmallVector<int>(Mask);
+  for (int i = 0; i < Size; ++i) {
----------------
InLaneMask.assign(Mask.begin(), Mask.end()); 


================
Comment at: llvm/lib/Target/X86/X86ISelLowering.cpp:18410
 
+  {
+    SmallVector<int> InLaneMask;
----------------
Move if(!Subtarget.hasAVX2()) up here to avoid unnecessary computeInLaneShuffleMask calls


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143785/new/

https://reviews.llvm.org/D143785