Match more patterns to movt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@84751 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 65 insertions, 17 deletions

diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index c39de0a..6e97693 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -1427,6 +1427,43 @@ SDNode *ARMDAGToDAGISel::Select(SDValue Op) {
       }
     }
     break;
+  case ISD::AND: {
+    // (and (or x, c2), c1) and top 16-bits of c1 and c2 match, lower 16-bits
+    // of c1 are 0xffff, and lower 16-bit of c2 are 0. That is, the top 16-bits
+    // are entirely contributed by c2 and lower 16-bits are entirely contributed
+    // by x. That's equal to (or (and x, 0xffff), (and c1, 0xffff0000)).
+    // Select it to: "movt x, ((c1 & 0xffff) >> 16)
+    EVT VT = Op.getValueType();
+    if (VT != MVT::i32)
+      break;
+    unsigned Opc = (Subtarget->isThumb() && Subtarget->hasThumb2())
+      ? ARM::t2MOVTi16
+      : (Subtarget->hasV6T2Ops() ? ARM::MOVTi16 : 0);
+    if (!Opc)
+      break;
+    SDValue N0 = Op.getOperand(0), N1 = Op.getOperand(1);
+    ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
+    if (!N1C)
+      break;
+    if (N0.getOpcode() == ISD::OR && N0.getNode()->hasOneUse()) {
+      SDValue N2 = N0.getOperand(1);
+      ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2);
+      if (!N2C)
+        break;
+      unsigned N1CVal = N1C->getZExtValue();
+      unsigned N2CVal = N2C->getZExtValue();
+      if ((N1CVal & 0xffff0000U) == (N2CVal & 0xffff0000U) &&
+          (N1CVal & 0xffffU) == 0xffffU &&
+          (N2CVal & 0xffffU) == 0x0U) {
+        SDValue Imm16 = CurDAG->getTargetConstant((N2CVal & 0xFFFF0000U) >> 16,
+                                                  MVT::i32);
+        SDValue Ops[] = { N0.getOperand(0), Imm16,
+                          getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
+        return CurDAG->getMachineNode(Opc, dl, VT, Ops, 4);
+      }
+    }
+    break;
+  }
   case ARMISD::FMRRD:
     return CurDAG->getMachineNode(ARM::FMRRD, dl, MVT::i32, MVT::i32,
                                   Op.getOperand(0), getAL(CurDAG),
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 1752880..8693a8a 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -3022,7 +3022,6 @@ static SDValue PerformSUBCombine(SDNode *N,
   return SDValue();
 }
 
-
 /// PerformFMRRDCombine - Target-specific dag combine xforms for ARMISD::FMRRD.
 static SDValue PerformFMRRDCombine(SDNode *N,
                                    TargetLowering::DAGCombinerInfo &DCI) {
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index 3e0691c..e6a0bfa 100644
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -980,6 +980,9 @@ def MOVTi16 : AI1<0b1010, (outs GPR:$dst), (ins GPR:$src, i32imm:$imm),
   let Inst{25} = 1;
 }
 
+def : ARMPat<(or GPR:$src, 0xffff0000), (MOVTi16 GPR:$src, 0xffff)>,
+      Requires<[IsARM, HasV6T2]>;
+
 let Uses = [CPSR] in
 def MOVrx : AsI1<0b1101, (outs GPR:$dst), (ins GPR:$src), Pseudo, IIC_iMOVsi,
                  "mov", " $dst, $src, rrx",
diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td
index 0750dcc..b151c99 100644
--- a/lib/Target/ARM/ARMInstrThumb2.td
+++ b/lib/Target/ARM/ARMInstrThumb2.td
@@ -666,6 +666,8 @@ def t2MOVTi16 : T2I<(outs GPR:$dst), (ins GPR:$src, i32imm:$imm), IIC_iMOVi,
                     [(set GPR:$dst,
                           (or (and GPR:$src, 0xffff), lo16AllZero:$imm))]>;
 
+def : T2Pat<(or GPR:$src, 0xffff0000), (t2MOVTi16 GPR:$src, 0xffff)>;
+
 //===----------------------------------------------------------------------===//
 //  Extend Instructions.
 //
diff --git a/test/CodeGen/ARM/movt.ll b/test/CodeGen/ARM/movt.ll
new file mode 100644
index 0000000..e82aca0
--- /dev/null
+++ b/test/CodeGen/ARM/movt.ll
@@ -0,0 +1,19 @@
+; RUN: llc < %s -march=arm -mattr=+thumb2 | FileCheck %s
+; rdar://7317664
+
+define i32 @t(i32 %X) nounwind {
+; CHECK: t:
+; CHECK: movt r0, #65535
+entry:
+	%0 = or i32 %X, -65536
+	ret i32 %0
+}
+
+define i32 @t2(i32 %X) nounwind {
+; CHECK: t2:
+; CHECK: movt r0, #65534
+entry:
+	%0 = or i32 %X, -131072
+	%1 = and i32 %0, -65537
+	ret i32 %1
+}
diff --git a/test/CodeGen/Thumb2/thumb2-mov2.ll b/test/CodeGen/Thumb2/thumb2-mov2.ll
index a02f4f0..32d0075 100644
--- a/test/CodeGen/Thumb2/thumb2-mov2.ll
+++ b/test/CodeGen/Thumb2/thumb2-mov2.ll
@@ -2,10 +2,7 @@
 
 define i32 @t2MOVTi16_ok_1(i32 %a) {
 ; CHECK: t2MOVTi16_ok_1:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = and i32 %a, 65535
     %2 = shl i32 1234, 16
     %3 = or  i32 %1, %2
@@ -15,10 +12,7 @@ define i32 @t2MOVTi16_ok_1(i32 %a) {
 
 define i32 @t2MOVTi16_test_1(i32 %a) {
 ; CHECK: t2MOVTi16_test_1:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3
@@ -31,10 +25,7 @@ define i32 @t2MOVTi16_test_1(i32 %a) {
 
 define i32 @t2MOVTi16_test_2(i32 %a) {
 ; CHECK: t2MOVTi16_test_2:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3
@@ -48,10 +39,7 @@ define i32 @t2MOVTi16_test_2(i32 %a) {
 
 define i32 @t2MOVTi16_test_3(i32 %a) {
 ; CHECK: t2MOVTi16_test_3:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3