[X86] resolveTargetShuffleInputsAndMask - add repeated input handling.

We were relying on combineX86ShufflesRecursively to handle this - this patch gets it done earlier which should make it easier for other code to use resolveTargetShuffleInputsAndMask.

llvm-svn: 364906

1 files changed, 22 insertions, 7 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index a47a6669849..b6e310694ba 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -7010,7 +7010,7 @@ static bool getFauxShuffleMask(SDValue N, const APInt &DemandedElts,
   return false;
 }
 
-/// Removes unused shuffle source inputs and adjusts the shuffle mask accordingly.
+/// Removes unused/repeated shuffle source inputs and adjusts the shuffle mask.
 static void resolveTargetShuffleInputsAndMask(SmallVectorImpl<SDValue> &Inputs,
                                               SmallVectorImpl<int> &Mask) {
   int MaskWidth = Mask.size();
@@ -7026,13 +7026,28 @@ static void resolveTargetShuffleInputsAndMask(SmallVectorImpl<SDValue> &Inputs,
           M = SM_SentinelUndef;
 
     // Check for unused inputs.
-    if (any_of(Mask, [lo, hi](int i) { return (lo <= i) && (i < hi); })) {
-      UsedInputs.push_back(Inputs[i]);
+    if (none_of(Mask, [lo, hi](int i) { return (lo <= i) && (i < hi); })) {
+      for (int &M : Mask)
+        if (lo <= M)
+          M -= MaskWidth;
       continue;
     }
-    for (int &M : Mask)
-      if (lo <= M)
-        M -= MaskWidth;
+
+    // Check for repeated inputs.
+    bool IsRepeat = false;
+    for (int j = 0, ue = UsedInputs.size(); j != ue; ++j) {
+      if (UsedInputs[j] != Inputs[i])
+        continue;
+      for (int &M : Mask)
+        if (lo <= M)
+          M = (M < hi) ? ((M - lo) + (j * MaskWidth)) : (M - MaskWidth);
+      IsRepeat = true;
+      break;
+    }
+    if (IsRepeat)
+      continue;
+
+    UsedInputs.push_back(Inputs[i]);
   }
   Inputs = UsedInputs;
 }
@@ -32668,7 +32683,7 @@ static SDValue combineX86ShufflesRecursively(
     return getZeroVector(Root.getSimpleValueType(), Subtarget, DAG,
                          SDLoc(Root));
 
-  // Remove unused shuffle source ops.
+  // Remove unused/repeated shuffle source ops.
   resolveTargetShuffleInputsAndMask(Ops, Mask);
   assert(!Ops.empty() && "Shuffle with no inputs detected");