Hexagon: ArePredicatesComplement should not restrict itself to TFRs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181803 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 63 insertions, 5 deletions

diff --git a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
index c508d124b3..59b4fabe01 100644
--- a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
+++ b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
@@ -837,16 +837,38 @@ bool HexagonPacketizerList::RestrictingDepExistInPacket (MachineInstr* MI,
 }
 
 
+/// Gets the predicate register of a predicated instruction.
+unsigned getPredicatedRegister(MachineInstr *MI, const HexagonInstrInfo *QII) {
+  /// We use the following rule: The first predicate register that is a use is
+  /// the predicate register of a predicated instruction.
+
+  assert(QII->isPredicated(MI) && "Must be predicated instruction");
+
+  for (MachineInstr::mop_iterator OI = MI->operands_begin(),
+       OE = MI->operands_end(); OI != OE; ++OI) {
+    MachineOperand &Op = *OI;
+    if (Op.isReg() && Op.getReg() && Op.isUse() &&
+        Hexagon::PredRegsRegClass.contains(Op.getReg()))
+      return Op.getReg();
+  }
+
+  llvm_unreachable("Unknown instruction operand layout");
+
+  return 0;
+}
+
 // Given two predicated instructions, this function detects whether
 // the predicates are complements
 bool HexagonPacketizerList::ArePredicatesComplements (MachineInstr* MI1,
      MachineInstr* MI2, std::map <MachineInstr*, SUnit*> MIToSUnit) {
 
   const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
-  // Currently can only reason about conditional transfers
-  if (!QII->isConditionalTransfer(MI1) || !QII->isConditionalTransfer(MI2)) {
+
+  // If we don't know the predicate sense of the instructions bail out early, we
+  // need it later.
+  if (getPredicateSense(MI1, QII) == PK_Unknown ||
+      getPredicateSense(MI2, QII) == PK_Unknown)
     return false;
-  }
 
   // Scheduling unit for candidate
   SUnit* SU = MIToSUnit[MI1];
@@ -885,9 +907,9 @@ bool HexagonPacketizerList::ArePredicatesComplements (MachineInstr* MI1,
         // there already exist anti dep on the same pred in
         // the packet.
         if (PacketSU->Succs[i].getSUnit() == SU &&
+            PacketSU->Succs[i].getKind() == SDep::Data &&
             Hexagon::PredRegsRegClass.contains(
               PacketSU->Succs[i].getReg()) &&
-            PacketSU->Succs[i].getKind() == SDep::Data &&
             // Here I know that *VIN is predicate setting instruction
             // with true data dep to candidate on the register
             // we care about - c) in the above example.
@@ -908,7 +930,11 @@ bool HexagonPacketizerList::ArePredicatesComplements (MachineInstr* MI1,
   // that the predicate sense is different
   // We also need to differentiate .old vs. .new:
   // !p0 is not complimentary to p0.new
-  return ((MI1->getOperand(1).getReg() == MI2->getOperand(1).getReg()) &&
+  unsigned PReg1 = getPredicatedRegister(MI1, QII);
+  unsigned PReg2 = getPredicatedRegister(MI2, QII);
+  return ((PReg1 == PReg2) &&
+          Hexagon::PredRegsRegClass.contains(PReg1) &&
+          Hexagon::PredRegsRegClass.contains(PReg2) &&
           (getPredicateSense(MI1, QII) != getPredicateSense(MI2, QII)) &&
           (QII->isDotNewInst(MI1) == QII->isDotNewInst(MI2)));
 }
diff --git a/test/CodeGen/Hexagon/packetize_cond_inst.ll b/test/CodeGen/Hexagon/packetize_cond_inst.ll
new file mode 100644
index 0000000000..a48a9f62ec
--- /dev/null
+++ b/test/CodeGen/Hexagon/packetize_cond_inst.ll
@@ -0,0 +1,32 @@
+; RUN: llc -mcpu=hexagonv4 -tail-dup-size=1 < %s | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a0:0-n16:32"
+target triple = "hexagon-unknown--elf"
+
+; Make sure we put the two conditionally executed adds in a packet.
+; ifcnv_add:
+;     {
+;       p0 = cmp.gt(r2, r1)
+;       if (!p0.new) r0 = add(r2, r1)
+;       if (p0.new) r0 = add(r0, #10)
+;     }
+; CHECK: cmp
+; CHECK-NEXT: add
+; CHECH-NEXT: add
+define i32 @ifcnv_add(i32, i32, i32) nounwind readnone {
+  %4 = icmp sgt i32 %2, %1
+  br i1 %4, label %5, label %7
+
+; <label>:5                                       ; preds = %3
+  %6 = add nsw i32 %0, 10
+  br label %9
+
+; <label>:7                                       ; preds = %3
+  %8 = add nsw i32 %2, %1
+  br label %9
+
+; <label>:9                                       ; preds = %7, %5
+  %10 = phi i32 [ %6, %5 ], [ %8, %7 ]
+  %11 = add nsw i32 %10, 1
+  ret i32 %11
+}