[X86][SSSE3] Bailout of lowerVectorShuffleAsPermuteAndUnpack for shuffle-with-zero (PR40306)

If we have PSHUFB and we're shuffling with a zero vector, then we are better off not doing VECTOR_SHUFFLE(UNPCK()) as we lose track of those zero elements.

llvm-svn: 351103

1 files changed, 14 insertions, 8 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index e2e63eb4c18..032c2106543 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -11964,10 +11964,9 @@ static SDValue lowerVectorShuffleAsInsertPS(const SDLoc &DL, SDValue V1,
 /// because for floating point vectors we have a generalized SHUFPS lowering
 /// strategy that handles everything that doesn't *exactly* match an unpack,
 /// making this clever lowering unnecessary.
-static SDValue lowerVectorShuffleAsPermuteAndUnpack(const SDLoc &DL, MVT VT,
-                                                    SDValue V1, SDValue V2,
-                                                    ArrayRef<int> Mask,
-                                                    SelectionDAG &DAG) {
+static SDValue lowerVectorShuffleAsPermuteAndUnpack(
+    const SDLoc &DL, MVT VT, SDValue V1, SDValue V2, ArrayRef<int> Mask,
+    const X86Subtarget &Subtarget, SelectionDAG &DAG) {
   assert(!VT.isFloatingPoint() &&
          "This routine only supports integer vectors.");
   assert(VT.is128BitVector() &&
@@ -12036,6 +12035,13 @@ static SDValue lowerVectorShuffleAsPermuteAndUnpack(const SDLoc &DL, MVT VT,
     if (SDValue Unpack = TryUnpack(ScalarSize, ScalarSize / OrigScalarSize))
       return Unpack;
 
+  // If we have PSHUFB, and we're shuffling with a zero vector then we're
+  // better off not doing VECTOR_SHUFFLE(UNPCK()) as we lose track of those
+  // zero elements.
+  if (Subtarget.hasSSSE3() && (ISD::isBuildVectorAllZeros(V1.getNode()) ||
+                               ISD::isBuildVectorAllZeros(V2.getNode())))
+    return SDValue();
+
   // If none of the unpack-rooted lowerings worked (or were profitable) try an
   // initial unpack.
   if (NumLoInputs == 0 || NumHiInputs == 0) {
@@ -12549,7 +12555,7 @@ static SDValue lowerV4I32VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
 
     // Try to lower by permuting the inputs into an unpack instruction.
     if (SDValue Unpack = lowerVectorShuffleAsPermuteAndUnpack(
-            DL, MVT::v4i32, V1, V2, Mask, DAG))
+            DL, MVT::v4i32, V1, V2, Mask, Subtarget, DAG))
       return Unpack;
   }
 
@@ -13245,8 +13251,8 @@ static SDValue lowerV8I16VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
     return BitBlend;
 
   // Try to lower by permuting the inputs into an unpack instruction.
-  if (SDValue Unpack = lowerVectorShuffleAsPermuteAndUnpack(DL, MVT::v8i16, V1,
-                                                            V2, Mask, DAG))
+  if (SDValue Unpack = lowerVectorShuffleAsPermuteAndUnpack(
+          DL, MVT::v8i16, V1, V2, Mask, Subtarget, DAG))
     return Unpack;
 
   // If we can't directly blend but can use PSHUFB, that will be better as it
@@ -13534,7 +13540,7 @@ static SDValue lowerV16I8VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
       // shuffles will both be pshufb, in which case we shouldn't bother with
       // this.
       if (SDValue Unpack = lowerVectorShuffleAsPermuteAndUnpack(
-              DL, MVT::v16i8, V1, V2, Mask, DAG))
+              DL, MVT::v16i8, V1, V2, Mask, Subtarget, DAG))
         return Unpack;
 
       // If we have VBMI we can use one VPERM instead of multiple PSHUFBs.