[X86][SSE] Allow matchVectorShuffleWithUNPCK to recognise UNDEF inputs

Add support for specifying an UNPCK input as UNDEF

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@295061 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index f62befed91fd50590cbeddeaf73d38f31647db4d..2288ce8941f52a3f7b48d4909f6c5bde09f89601 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8224,25 +8224,37 @@ static SDValue lowerVectorShuffleToEXPAND(const SDLoc &DL, MVT VT,
 
 static bool matchVectorShuffleWithUNPCK(MVT VT, SDValue &V1, SDValue &V2,
                                         unsigned &UnpackOpcode, bool IsUnary,
-                                        ArrayRef<int> TargetMask) {
+                                        ArrayRef<int> TargetMask,
+                                        SelectionDAG &DAG) {
+  int NumElts = VT.getVectorNumElements();
+  int NumEltsInLane = 128 / VT.getScalarSizeInBits();
+
+  bool Undef1 = true, Undef2 = true;
+  for (int i = 0; (i != NumElts) && (Undef1 || Undef2); i += 2) {
+    Undef1 &= (SM_SentinelUndef == TargetMask[i + 0]);
+    Undef2 &= (SM_SentinelUndef == TargetMask[i + 1]);
+  }
+
   // Attempt to match the target mask against the unpack lo/hi mask patterns.
   SmallVector<int, 64> Unpckl, Unpckh;
   createUnpackShuffleMask(VT, Unpckl, /* Lo = */ true, IsUnary);
   if (isTargetShuffleEquivalent(TargetMask, Unpckl)) {
     UnpackOpcode = X86ISD::UNPCKL;
-    V2 = IsUnary ? V1 : V2;
+    V2 = (Undef2 ? DAG.getUNDEF(VT) : (IsUnary ? V1 : V2));
+    V1 = (Undef1 ? DAG.getUNDEF(VT) : V1);
     return true;
   }
 
   createUnpackShuffleMask(VT, Unpckh, /* Lo = */ false, IsUnary);
   if (isTargetShuffleEquivalent(TargetMask, Unpckh)) {
     UnpackOpcode = X86ISD::UNPCKH;
-    V2 = IsUnary ? V1 : V2;
+    V2 = (Undef2 ? DAG.getUNDEF(VT) : (IsUnary ? V1 : V2));
+    V1 = (Undef1 ? DAG.getUNDEF(VT) : V1);
     return true;
   }
 
-  // If a binary shuffle, commute and try again.
   if (!IsUnary) {
+    // If a binary shuffle, commute and try again.
     ShuffleVectorSDNode::commuteMask(Unpckl);
     if (isTargetShuffleEquivalent(TargetMask, Unpckl)) {
       UnpackOpcode = X86ISD::UNPCKL;
@@ -26571,6 +26583,7 @@ static bool matchUnaryPermuteVectorShuffle(MVT MaskVT, ArrayRef<int> Mask,
 // TODO: Investigate sharing more of this with shuffle lowering.
 static bool matchBinaryVectorShuffle(MVT MaskVT, ArrayRef<int> Mask,
                                      bool FloatDomain, SDValue &V1, SDValue &V2,
+                                     SelectionDAG &DAG,
                                      const X86Subtarget &Subtarget,
                                      unsigned &Shuffle, MVT &ShuffleVT,
                                      bool IsUnary) {
@@ -26610,7 +26623,8 @@ static bool matchBinaryVectorShuffle(MVT MaskVT, ArrayRef<int> Mask,
       (MaskVT.is256BitVector() && 32 <= EltSizeInBits && Subtarget.hasAVX()) ||
       (MaskVT.is256BitVector() && Subtarget.hasAVX2()) ||
       (MaskVT.is512BitVector() && Subtarget.hasAVX512())) {
-    if (matchVectorShuffleWithUNPCK(MaskVT, V1, V2, Shuffle, IsUnary, Mask)) {
+    if (matchVectorShuffleWithUNPCK(MaskVT, V1, V2, Shuffle, IsUnary, Mask,
+                                    DAG)) {
       ShuffleVT = MaskVT;
       if (ShuffleVT.is256BitVector() && !Subtarget.hasAVX2())
         ShuffleVT = (32 == EltSizeInBits ? MVT::v8f32 : MVT::v4f64);
@@ -26941,8 +26955,8 @@ static bool combineX86ShuffleChain(ArrayRef<SDValue> Inputs, SDValue Root,
     }
   }
 
-  if (matchBinaryVectorShuffle(MaskVT, Mask, FloatDomain, V1, V2, Subtarget,
-                               Shuffle, ShuffleVT, UnaryShuffle)) {
+  if (matchBinaryVectorShuffle(MaskVT, Mask, FloatDomain, V1, V2, DAG,
+                               Subtarget, Shuffle, ShuffleVT, UnaryShuffle)) {
     if (Depth == 1 && Root.getOpcode() == Shuffle)
       return false; // Nothing to do!
     if (IsEVEXShuffle && (NumRootElts != ShuffleVT.getVectorNumElements()))

diff --git a/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index 77864904261a9f147feb83fd199c9e3f280c41f1..bebf2bae245a89d3e527f90c37796e299f2b6b0e 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -665,12 +665,10 @@ define <32 x i8> @combine_pshufb_not_as_pshufw(<32 x i8> %a0) {
 define <32 x i8> @combine_pshufb_as_unpacklo_undef(<32 x i8> %a0) {
 ; X32-LABEL: combine_pshufb_as_unpacklo_undef:
 ; X32:       # BB#0:
-; X32-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: combine_pshufb_as_unpacklo_undef:
 ; X64:       # BB#0:
-; X64-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; X64-NEXT:    retq
   %1 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 undef, i8 0, i8 undef, i8 1, i8 undef, i8 2, i8 undef, i8 3, i8 undef, i8 4, i8 undef, i8 5, i8 undef, i8 6, i8 undef, i8 7, i8 undef, i8 16, i8 undef, i8 17, i8 undef, i8 18, i8 undef, i8 19, i8 undef, i8 20, i8 undef, i8 21, i8 undef, i8 22, i8 undef, i8 23>)
   %2 = shufflevector <32 x i8> %1, <32 x i8> undef, <32 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6, i32 8, i32 8, i32 10, i32 10, i32 12, i32 12, i32 14, i32 14, i32 16, i32 16, i32 18, i32 18, i32 20, i32 20, i32 22, i32 22, i32 24, i32 24, i32 26, i32 26, i32 28, i32 28, i32 30, i32 30>

diff --git a/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll b/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
index 94d9f81214bfa6b6fbf5fb5a7eb6f3d9c2f84b69..8d9fb07a66b018612cb9dc6c55db9a66f20cd365 100644 (file)
--- a/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
@@ -474,15 +474,9 @@ define <16 x i8> @combine_pshufb_as_unary_unpckhwd(<16 x i8> %a0) {
 }
 
 define <8 x i16> @combine_pshufb_as_unpacklo_undef(<16 x i8> %a0) {
-; SSE-LABEL: combine_pshufb_as_unpacklo_undef:
-; SSE:       # BB#0:
-; SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE-NEXT:    retq
-;
-; AVX-LABEL: combine_pshufb_as_unpacklo_undef:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; AVX-NEXT:    retq
+; ALL-LABEL: combine_pshufb_as_unpacklo_undef:
+; ALL:       # BB#0:
+; ALL-NEXT:    retq
   %1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> <i8 undef, i8 undef, i8 0, i8 1, i8 undef, i8 undef, i8 2, i8 3, i8 undef, i8 undef, i8 4, i8 5, i8 undef, i8 undef, i8 6, i8 7>)
   %2 = bitcast <16 x i8> %1 to <8 x i16>
   %3 = shufflevector <8 x i16> %2, <8 x i16> undef, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
@@ -490,15 +484,9 @@ define <8 x i16> @combine_pshufb_as_unpacklo_undef(<16 x i8> %a0) {
 }
 
 define <16 x i8> @combine_pshufb_as_unpackhi_undef(<16 x i8> %a0) {
-; SSE-LABEL: combine_pshufb_as_unpackhi_undef:
-; SSE:       # BB#0:
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    retq
-;
-; AVX-LABEL: combine_pshufb_as_unpackhi_undef:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX-NEXT:    retq
+; ALL-LABEL: combine_pshufb_as_unpackhi_undef:
+; ALL:       # BB#0:
+; ALL-NEXT:    retq
   %1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> <i8 8, i8 undef, i8 9, i8 undef, i8 10, i8 undef, i8 11, i8 undef, i8 12, i8 undef, i8 13, i8 undef, i8 14, i8 undef, i8 15, i8 undef>)
   %2 = shufflevector <16 x i8> %1, <16 x i8> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, i32 15>
   ret <16 x i8> %2