Added atomic 64 min/max/umin/umax instrinsics support in the ARM backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168886 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 163 insertions, 12 deletions

diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index 66fb83ca7b..de9c8eb2d8 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -3327,6 +3327,15 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     return SelectAtomic64(N, ARM::ATOMSWAP6432);
   case ARMISD::ATOMCMPXCHG64_DAG:
     return SelectAtomic64(N, ARM::ATOMCMPXCHG6432);
+
+  case ARMISD::ATOMMIN64_DAG:
+    return SelectAtomic64(N, ARM::ATOMMIN6432);
+  case ARMISD::ATOMUMIN64_DAG:
+    return SelectAtomic64(N, ARM::ATOMUMIN6432);
+  case ARMISD::ATOMMAX64_DAG:
+    return SelectAtomic64(N, ARM::ATOMMAX6432);
+  case ARMISD::ATOMUMAX64_DAG:
+    return SelectAtomic64(N, ARM::ATOMUMAX6432);
   }
 
   return SelectCode(N);
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index b2af378977..d139a568a7 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -695,7 +695,11 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::ATOMIC_LOAD_AND,  MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_LOAD_OR,   MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_LOAD_XOR,  MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_SWAP,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_SWAP,      MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_MIN,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_MAX,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_CMP_SWAP,  MVT::i64, Custom);
     // Automatically insert fences (dmb ist) around ATOMIC_SWAP etc.
     setInsertFencesForAtomic(true);
@@ -5406,6 +5410,18 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
   case ISD::ATOMIC_CMP_SWAP:
     ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMCMPXCHG64_DAG);
     return;
+  case ISD::ATOMIC_LOAD_MIN:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMMIN64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_UMIN:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMUMIN64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_MAX:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMMAX64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_UMAX:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMUMAX64_DAG);
+    return;
   }
   if (Res.getNode())
     Results.push_back(Res);
@@ -5745,7 +5761,8 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI,
 MachineBasicBlock *
 ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
                                       unsigned Op1, unsigned Op2,
-                                      bool NeedsCarry, bool IsCmpxchg) const {
+                                      bool NeedsCarry, bool IsCmpxchg,
+                                      bool IsMinMax, ARMCC::CondCodes CC) const {
   // This also handles ATOMIC_SWAP, indicated by Op1==0.
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
 
@@ -5774,16 +5791,15 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
 
   MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
   MachineBasicBlock *contBB = 0, *cont2BB = 0;
-  if (IsCmpxchg) {
+  if (IsCmpxchg || IsMinMax)
     contBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  if (IsCmpxchg)
     cont2BB = MF->CreateMachineBasicBlock(LLVM_BB);
-  }
   MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+
   MF->insert(It, loopMBB);
-  if (IsCmpxchg) {
-    MF->insert(It, contBB);
-    MF->insert(It, cont2BB);
-  }
+  if (IsCmpxchg || IsMinMax) MF->insert(It, contBB);
+  if (IsCmpxchg) MF->insert(It, cont2BB);
   MF->insert(It, exitMBB);
 
   // Transfer the remainder of BB and its successor edges to exitMBB.
@@ -5821,6 +5837,22 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
   // Load
   unsigned GPRPair0 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
   unsigned GPRPair1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+  unsigned GPRPair2;
+  if (IsMinMax) {
+    //We need an extra double register for doing min/max.
+    unsigned undef = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    GPRPair2 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), undef);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
+      .addReg(undef)
+      .addReg(vallo)
+      .addImm(ARM::gsub_0);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), GPRPair2)
+      .addReg(r1)
+      .addReg(valhi)
+      .addImm(ARM::gsub_1);
+  }
 
   AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc))
                  .addReg(GPRPair0, RegState::Define).addReg(ptr));
@@ -5866,7 +5898,8 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
         .addReg(NeedsCarry ? ARM::CPSR : 0, getDefRegState(NeedsCarry));
     unsigned tmpRegHi = MRI.createVirtualRegister(TRC);
     AddDefaultPred(BuildMI(BB, dl, TII->get(Op2), tmpRegHi)
-                   .addReg(desthi).addReg(valhi)).addReg(0);
+                   .addReg(desthi).addReg(valhi))
+        .addReg(IsMinMax ? ARM::CPSR : 0, getDefRegState(IsMinMax));
 
     unsigned UndefPair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
     BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), UndefPair);
@@ -5893,10 +5926,20 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
       .addReg(valhi)
       .addImm(ARM::gsub_1);
   }
+  unsigned GPRPairStore = GPRPair1;
+  if (IsMinMax) {
+    // Compare and branch to exit block.
+    BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
+      .addMBB(exitMBB).addImm(CC).addReg(ARM::CPSR);
+    BB->addSuccessor(exitMBB);
+    BB->addSuccessor(contBB);
+    BB = contBB;
+    GPRPairStore = GPRPair2;
+  }
 
   // Store
   AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), storesuccess)
-                 .addReg(GPRPair1).addReg(ptr));
+                 .addReg(GPRPairStore).addReg(ptr));
   // Cmp+jump
   AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
                  .addReg(storesuccess).addImm(0));
@@ -6894,6 +6937,26 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
     return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
                               isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
                               /*NeedsCarry*/ false, /*IsCmpxchg*/true);
+  case ARM::ATOMMIN6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::LE);
+  case ARM::ATOMMAX6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::GE);
+  case ARM::ATOMUMIN6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::LS);
+  case ARM::ATOMUMAX6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::HS);
 
   case ARM::tMOVCCr_pseudo: {
     // To "insert" a SELECT_CC instruction, we actually have to insert the
diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h
index 9cbf866f9b..a42ff7849e 100644
--- a/lib/Target/ARM/ARMISelLowering.h
+++ b/lib/Target/ARM/ARMISelLowering.h
@@ -232,7 +232,11 @@ namespace llvm {
       ATOMAND64_DAG,
       ATOMNAND64_DAG,
       ATOMSWAP64_DAG,
-      ATOMCMPXCHG64_DAG
+      ATOMCMPXCHG64_DAG,
+      ATOMMIN64_DAG,
+      ATOMUMIN64_DAG,
+      ATOMMAX64_DAG,
+      ATOMUMAX64_DAG
     };
   }
 
@@ -532,7 +536,9 @@ namespace llvm {
                                           unsigned Op1,
                                           unsigned Op2,
                                           bool NeedsCarry = false,
-                                          bool IsCmpxchg = false) const;
+                                          bool IsCmpxchg = false,
+                                          bool IsMinMax = false,
+                                          ARMCC::CondCodes CC = ARMCC::AL) const;
     MachineBasicBlock * EmitAtomicBinaryMinMax(MachineInstr *MI,
                                                MachineBasicBlock *BB,
                                                unsigned Size,
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index 5cf7c3c372..12712c007b 100644
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -1624,6 +1624,18 @@ def ATOMCMPXCHG6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
                                  (ins GPR:$addr, GPR:$cmp1, GPR:$cmp2,
                                       GPR:$set1, GPR:$set2),
                                  NoItinerary, []>;
+def ATOMMIN6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMUMIN6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMMAX6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMUMAX6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
 }
 
 def HINT : AI<(outs), (ins imm0_255:$imm), MiscFrm, NoItinerary,
diff --git a/test/CodeGen/ARM/atomic-64bit.ll b/test/CodeGen/ARM/atomic-64bit.ll
index be51e3c129..69da6221b7 100644
--- a/test/CodeGen/ARM/atomic-64bit.ll
+++ b/test/CodeGen/ARM/atomic-64bit.ll
@@ -126,3 +126,64 @@ define void @test9(i64* %ptr, i64 %val) {
   store atomic i64 %val, i64* %ptr seq_cst, align 8
   ret void
 }
+
+define i64 @test10(i64* %ptr, i64 %val) {
+; CHECK: test10:
+; CHECK: dmb ish
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
+; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
+; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
+; CHECK: ble
+; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK: cmp
+; CHECK: bne
+; CHECK: dmb ish
+  %r = atomicrmw min i64* %ptr, i64 %val seq_cst
+  ret i64 %r
+}
+
+define i64 @test11(i64* %ptr, i64 %val) {
+; CHECK: test11:
+; CHECK: dmb ish
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
+; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
+; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
+; CHECK: bls
+; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK: cmp
+; CHECK: bne
+; CHECK: dmb ish
+  %r = atomicrmw umin i64* %ptr, i64 %val seq_cst
+  ret i64 %r
+}
+
+define i64 @test12(i64* %ptr, i64 %val) {
+; CHECK: test12:
+; CHECK: dmb ish
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
+; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
+; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
+; CHECK: bge
+; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK: cmp
+; CHECK: bne
+; CHECK: dmb ish
+  %r = atomicrmw max i64* %ptr, i64 %val seq_cst
+  ret i64 %r
+}
+
+define i64 @test13(i64* %ptr, i64 %val) {
+; CHECK: test13:
+; CHECK: dmb ish
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
+; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
+; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
+; CHECK: bhs
+; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
+; CHECK: cmp
+; CHECK: bne
+; CHECK: dmb ish
+  %r = atomicrmw umax i64* %ptr, i64 %val seq_cst
+  ret i64 %r
+}
+