[X86] SimplifyDemandedVectorElts - attempt to recombine target shuffle using DemandedElts mask (reapplied)

If we don't demand all elements, then attempt to combine to a simpler shuffle.

At the moment we can only do this if Depth == 0 as combineX86ShufflesRecursively uses Depth to track whether the shuffle has really changed or not - we'll need to change this before we can properly start merging combineX86ShufflesRecursively into SimplifyDemandedVectorElts. 

The insertps-combine.ll regression is because XFormVExtractWithShuffleIntoLoad can't see through shuffles of different widths - this will be fixed in a follow-up commit.

Reapplying this as rL368307 had to be reverted as part of rL368660 to revert rL368276

llvm-svn: 368662

1 files changed, 17 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index a82f2138562..66d7b76a98d 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -33920,6 +33920,23 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
       return true;
   }
 
+  // If we don't demand all elements, then attempt to combine to a simpler
+  // shuffle.
+  // TODO: Handle other depths, but first we need to handle the fact that
+  // it might combine to the same shuffle.
+  if (!DemandedElts.isAllOnesValue() && Depth == 0) {
+    SmallVector<int, 32> DemandedMask(NumElts, SM_SentinelUndef);
+    for (int i = 0; i != NumElts; ++i)
+      if (DemandedElts[i])
+        DemandedMask[i] = i;
+
+    SDValue NewShuffle = combineX86ShufflesRecursively(
+        {Op}, 0, Op, DemandedMask, {}, Depth, /*HasVarMask*/ false,
+        /*AllowVarMask*/ true, TLO.DAG, Subtarget);
+    if (NewShuffle)
+      return TLO.CombineTo(Op, NewShuffle);
+  }
+
   // Extract known zero/undef elements.
   // TODO - Propagate input undef/zero elts.
   for (int i = 0; i != NumElts; ++i) {