From a4e7f05a28d3454d63c5e3217b9ce6b72c741b82 Mon Sep 17 00:00:00 2001
From: Chandler Carruth <chandlerc@gmail.com>
Date: Thu, 10 Jul 2014 11:09:29 +0000
Subject: [PATCH] [x86] Add another combine that is particularly useful for the
 new vector shuffle lowering: match shuffle patterns equivalent to an unpcklwd
 or unpckhwd instruction.

This allows us to use generic lowering code for v8i16 shuffles and match
the unpack pattern late.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@212705 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp         | 41 ++++++++++++++++++++++
 test/CodeGen/X86/vector-shuffle-128-v16.ll | 14 ++++++--
 2 files changed, 52 insertions(+), 3 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 245cb414ca9..5ccff20e294 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -18694,6 +18694,47 @@ static SDValue PerformTargetShuffleCombine(SDValue N, SelectionDAG &DAG,
       return DAG.getNode(ISD::BITCAST, DL, MVT::v8i16, V);
     }
 
+    // Look for shuffle patterns which can be implemented as a single unpack.
+    // FIXME: This doesn't handle the location of the PSHUFD generically, and
+    // only works when we have a PSHUFD followed by two half-shuffles.
+    if (Mask[0] == Mask[1] && Mask[2] == Mask[3] &&
+        (V.getOpcode() == X86ISD::PSHUFLW ||
+         V.getOpcode() == X86ISD::PSHUFHW) &&
+        V.getOpcode() != N.getOpcode() &&
+        V.hasOneUse()) {
+      SDValue D = V.getOperand(0);
+      while (D.getOpcode() == ISD::BITCAST && D.hasOneUse())
+        D = D.getOperand(0);
+      if (D.getOpcode() == X86ISD::PSHUFD && D.hasOneUse()) {
+        SmallVector<int, 4> VMask = getPSHUFShuffleMask(V);
+        SmallVector<int, 4> DMask = getPSHUFShuffleMask(D);
+        int NOffset = N.getOpcode() == X86ISD::PSHUFLW ? 0 : 4;
+        int VOffset = V.getOpcode() == X86ISD::PSHUFLW ? 0 : 4;
+        int WordMask[8];
+        for (int i = 0; i < 4; ++i) {
+          WordMask[i + NOffset] = Mask[i] + NOffset;
+          WordMask[i + VOffset] = VMask[i] + VOffset;
+        }
+        // Map the word mask through the DWord mask.
+        int MappedMask[8];
+        for (int i = 0; i < 8; ++i)
+          MappedMask[i] = 2 * DMask[WordMask[i] / 2] + WordMask[i] % 2;
+        const int UnpackLoMask[] = {0, 0, 1, 1, 2, 2, 3, 3};
+        const int UnpackHiMask[] = {4, 4, 5, 5, 6, 6, 7, 7};
+        if (std::equal(std::begin(MappedMask), std::end(MappedMask),
+                       std::begin(UnpackLoMask)) ||
+            std::equal(std::begin(MappedMask), std::end(MappedMask),
+                       std::begin(UnpackHiMask))) {
+          // We can replace all three shuffles with an unpack.
+          V = DAG.getNode(ISD::BITCAST, DL, MVT::v8i16, D.getOperand(0));
+          DCI.AddToWorklist(V.getNode());
+          return DAG.getNode(MappedMask[0] == 0 ? X86ISD::UNPCKL
+                                                : X86ISD::UNPCKH,
+                             DL, MVT::v8i16, V, V);
+        }
+      }
+    }
+
     break;
 
   case X86ISD::PSHUFD:
diff --git a/test/CodeGen/X86/vector-shuffle-128-v16.ll b/test/CodeGen/X86/vector-shuffle-128-v16.ll
index 71fac00021f..4da7e42caab 100644
--- a/test/CodeGen/X86/vector-shuffle-128-v16.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-v16.ll
@@ -44,14 +44,22 @@ define <16 x i8> @shuffle_v16i8_00_00_00_00_01_01_01_01_02_02_02_02_03_03_03_03(
 ; CHECK-SSE2-LABEL: @shuffle_v16i8_00_00_00_00_01_01_01_01_02_02_02_02_03_03_03_03
 ; CHECK-SSE2:       # BB#0:
 ; CHECK-SSE2-NEXT:    punpcklbw %xmm0, %xmm0
-; CHECK-SSE2-NEXT:    pshufd {{.*}} # xmm0 = xmm0[0,1,2,1]
-; CHECK-SSE2-NEXT:    pshuflw {{.*}} # xmm0 = xmm0[0,0,1,1,4,5,6,7]
-; CHECK-SSE2-NEXT:    pshufhw {{.*}} # xmm0 = xmm0[0,1,2,3,6,6,7,7]
+; CHECK-SSE2-NEXT:    punpcklwd %xmm0, %xmm0
 ; CHECK-SSE2-NEXT:    retq
   %shuffle = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
   ret <16 x i8> %shuffle
 }
 
+define <16 x i8> @shuffle_v16i8_04_04_04_04_05_05_05_05_06_06_06_06_07_07_07_07(<16 x i8> %a, <16 x i8> %b) {
+; CHECK-SSE2-LABEL: @shuffle_v16i8_04_04_04_04_05_05_05_05_06_06_06_06_07_07_07_07
+; CHECK-SSE2:       # BB#0:
+; CHECK-SSE2-NEXT:    punpcklbw %xmm0, %xmm0
+; CHECK-SSE2-NEXT:    punpckhwd %xmm0, %xmm0
+; CHECK-SSE2-NEXT:    retq
+  %shuffle = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7>
+  ret <16 x i8> %shuffle
+}
+
 define <16 x i8> @shuffle_v16i8_00_00_00_00_04_04_04_04_08_08_08_08_12_12_12_12(<16 x i8> %a, <16 x i8> %b) {
 ; CHECK-SSE2-LABEL: @shuffle_v16i8_00_00_00_00_04_04_04_04_08_08_08_08_12_12_12_12
 ; CHECK-SSE2:       # BB#0: