[llvm] fac1729 - [X86] lowerV64I8Shuffle - don't use lowerShuffleWithPERMV until we've tried simpler options

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-03-31T18:05:51+01:00
New Revision: fac17299243b8bf87266aa47984b398ce72089f2

URL: https://github.com/llvm/llvm-project/commit/fac17299243b8bf87266aa47984b398ce72089f2
DIFF: https://github.com/llvm/llvm-project/commit/fac17299243b8bf87266aa47984b398ce72089f2.diff

LOG: [X86] lowerV64I8Shuffle - don't use lowerShuffleWithPERMV until we've tried simpler options

Shuffle combining will still lower to this with better fast cross lane checks.

Noticed while triaging Issue #54658

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 80fec0a54ef69..a8c43895e7850 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -18699,10 +18699,6 @@ static SDValue lowerV64I8Shuffle(const SDLoc &DL, ArrayRef<int> Mask,
                                               Zeroable, Subtarget, DAG))
     return PSHUFB;
 
-  // VBMI can use VPERMV/VPERMV3 byte shuffles.
-  if (Subtarget.hasVBMI())
-    return lowerShuffleWithPERMV(DL, MVT::v64i8, Mask, V1, V2, Subtarget, DAG);
-
   // Try to create an in-lane repeating shuffle mask and then shuffle the
   // results into the target lanes.
   if (SDValue V = lowerShuffleAsRepeatedMaskAndLanePermute(
@@ -18724,6 +18720,10 @@ static SDValue lowerV64I8Shuffle(const SDLoc &DL, ArrayRef<int> Mask,
             DL, MVT::v64i8, V1, V2, Mask, Subtarget, DAG))
       return Result;
 
+  // VBMI can use VPERMV/VPERMV3 byte shuffles.
+  if (Subtarget.hasVBMI())
+    return lowerShuffleWithPERMV(DL, MVT::v64i8, Mask, V1, V2, Subtarget, DAG);
+
   // FIXME: Implement direct support for this type!
   return splitAndLowerShuffle(DL, MVT::v64i8, V1, V2, Mask, DAG);
 }