Re-apply r175088 for bug fix 13622: Add paired register support for

inline asm with 64-bit data on ARM

Update test case to use -mtriple=arm-linux-gnueabi


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175186 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/ARM/ARMAsmPrinter.cpp

@@ -342,6 +342,11 @@ void ARMAsmPrinter::printOperand(const MachineInstr *MI, int OpNum,
     unsigned Reg = MO.getReg();
     assert(TargetRegisterInfo::isPhysicalRegister(Reg));
     assert(!MO.getSubReg() && "Subregs should be eliminated!");
+    if(ARM::GPRPairRegClass.contains(Reg)) {
+      const MachineFunction &MF = *MI->getParent()->getParent();
+      const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
+      Reg = TRI->getSubReg(Reg, ARM::gsub_0);
+    }
     O << ARMInstPrinter::getRegisterName(Reg);
     break;
   }
@@ -530,14 +535,12 @@ bool ARMAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum,
       const MachineOperand &MO = MI->getOperand(OpNum);
       if (!MO.isReg())
         return true;
-      const TargetRegisterClass &RC = ARM::GPRRegClass;
       const MachineFunction &MF = *MI->getParent()->getParent();
       const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
-
+      unsigned Reg = MO.getReg();
-      unsigned RegIdx = TRI->getEncodingValue(MO.getReg());
+      if(!ARM::GPRPairRegClass.contains(Reg))
-      RegIdx |= 1; //The odd register is also the higher-numbered one of a pair.
+        return false;
-
+      Reg = TRI->getSubReg(Reg, ARM::gsub_1);
-      unsigned Reg = RC.getRegister(RegIdx);
       O << ARMInstPrinter::getRegisterName(Reg);
       return false;
     }

lib/Target/ARM/ARMISelDAGToDAG.cpp

@@ -19,6 +19,7 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/IR/CallingConv.h"
@@ -257,6 +258,8 @@ private:
   // Select special operations if node forms integer ABS pattern
   SDNode *SelectABSOp(SDNode *N);
 
+  SDNode *SelectInlineAsm(SDNode *N);
+
   SDNode *SelectConcatVector(SDNode *N);
 
   SDNode *SelectAtomic64(SDNode *Node, unsigned Opc);
@@ -2552,6 +2555,12 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
 
   switch (N->getOpcode()) {
   default: break;
+  case ISD::INLINEASM: {
+    SDNode *ResNode = SelectInlineAsm(N);
+    if (ResNode)
+      return ResNode;
+    break;
+  }
   case ISD::XOR: {
     // Select special operations if XOR node forms integer ABS pattern
     SDNode *ResNode = SelectABSOp(N);
@@ -3446,6 +3455,138 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
   return SelectCode(N);
 }
 
+SDNode *ARMDAGToDAGISel::SelectInlineAsm(SDNode *N){
+  std::vector<SDValue> AsmNodeOperands;
+  unsigned Flag, Kind;
+  bool Changed = false;
+  unsigned NumOps = N->getNumOperands();
+
+  ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(
+      N->getOperand(InlineAsm::Op_AsmString));
+  StringRef AsmString = StringRef(S->getSymbol());
+
+  // Normally, i64 data is bounded to two arbitrary GRPs for "%r" constraint.
+  // However, some instrstions (e.g. ldrexd/strexd in ARM mode) require
+  // (even/even+1) GPRs and use %n and %Hn to refer to the individual regs
+  // respectively. Since there is no constraint to explicitly specify a
+  // reg pair, we search %H operand inside the asm string. If it is found, the
+  // transformation below enforces a GPRPair reg class for "%r" for 64-bit data.
+  if (AsmString.find(":H}") == StringRef::npos)
+    return NULL;
+
+  DebugLoc dl = N->getDebugLoc();
+  SDValue Glue = N->getOperand(NumOps-1);
+
+  // Glue node will be appended late.
+  for(unsigned i = 0; i < NumOps -1; ++i) {
+    SDValue op = N->getOperand(i);
+    AsmNodeOperands.push_back(op);
+
+    if (i < InlineAsm::Op_FirstOperand)
+      continue;
+
+    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(i))) {
+      Flag = C->getZExtValue();
+      Kind = InlineAsm::getKind(Flag);
+    }
+    else
+      continue;
+
+    if (Kind != InlineAsm::Kind_RegUse && Kind != InlineAsm::Kind_RegDef
+        && Kind != InlineAsm::Kind_RegDefEarlyClobber)
+      continue;
+
+    unsigned RegNum = InlineAsm::getNumOperandRegisters(Flag);
+    unsigned RC;
+    bool HasRC = InlineAsm::hasRegClassConstraint(Flag, RC);
+    if (!HasRC || RC != ARM::GPRRegClassID || RegNum != 2)
+      continue;
+
+    assert((i+2 < NumOps-1) && "Invalid number of operands in inline asm");
+    SDValue V0 = N->getOperand(i+1);
+    SDValue V1 = N->getOperand(i+2);
+    unsigned Reg0 = cast<RegisterSDNode>(V0)->getReg();
+    unsigned Reg1 = cast<RegisterSDNode>(V1)->getReg();
+    SDValue PairedReg;
+    MachineRegisterInfo &MRI = MF->getRegInfo();
+
+    if (Kind == InlineAsm::Kind_RegDef ||
+        Kind == InlineAsm::Kind_RegDefEarlyClobber) {
+      // Replace the two GPRs with 1 GPRPair and copy values from GPRPair to
+      // the original GPRs.
+
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      SDValue Chain = SDValue(N,0);
+
+      SDNode *GU = N->getGluedUser();
+      SDValue RegCopy = CurDAG->getCopyFromReg(Chain, dl, GPVR, MVT::Untyped,
+                                               Chain.getValue(1));
+
+      // Extract values from a GPRPair reg and copy to the original GPR reg.
+      SDValue Sub0 = CurDAG->getTargetExtractSubreg(ARM::gsub_0, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue Sub1 = CurDAG->getTargetExtractSubreg(ARM::gsub_1, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue T0 = CurDAG->getCopyToReg(Sub0, dl, Reg0, Sub0,
+                                        RegCopy.getValue(1));
+      SDValue T1 = CurDAG->getCopyToReg(Sub1, dl, Reg1, Sub1, T0.getValue(1));
+
+      // Update the original glue user.
+      std::vector<SDValue> Ops(GU->op_begin(), GU->op_end()-1);
+      Ops.push_back(T1.getValue(1));
+      CurDAG->UpdateNodeOperands(GU, &Ops[0], Ops.size());
+      GU = T1.getNode();
+    }
+    else {
+      // For Kind  == InlineAsm::Kind_RegUse, we first copy two GPRs into a
+      // GPRPair and then pass the GPRPair to the inline asm.
+      SDValue Chain = AsmNodeOperands[InlineAsm::Op_InputChain];
+
+      // As REG_SEQ doesn't take RegisterSDNode, we copy them first.
+      SDValue T0 = CurDAG->getCopyFromReg(Chain, dl, Reg0, MVT::i32,
+                                          Chain.getValue(1));
+      SDValue T1 = CurDAG->getCopyFromReg(Chain, dl, Reg1, MVT::i32,
+                                          T0.getValue(1));
+      SDValue Pair = SDValue(createGPRPairNode(MVT::Untyped, T0, T1), 0);
+
+      // Copy REG_SEQ into a GPRPair-typed VR and replace the original two
+      // i32 VRs of inline asm with it.
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      Chain = CurDAG->getCopyToReg(T1, dl, GPVR, Pair, T1.getValue(1));
+
+      AsmNodeOperands[InlineAsm::Op_InputChain] = Chain;
+      Glue = Chain.getValue(1);
+    }
+
+    Changed = true;
+
+    if(PairedReg.getNode()) {
+      Flag = InlineAsm::getFlagWord(Kind, 1 /* RegNum*/);
+      Flag = InlineAsm::getFlagWordForRegClass(Flag, ARM::GPRPairRegClassID);
+      // Replace the current flag.
+      AsmNodeOperands[AsmNodeOperands.size() -1] = CurDAG->getTargetConstant(
+          Flag, MVT::i32);
+      // Add the new register node and skip the original two GPRs.
+      AsmNodeOperands.push_back(PairedReg);
+      // Skip the next two GPRs.
+      i += 2;
+    }
+  }
+
+  AsmNodeOperands.push_back(Glue);
+  if (!Changed)
+    return NULL;
+
+  SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(),
+      CurDAG->getVTList(MVT::Other, MVT::Glue), &AsmNodeOperands[0],
+                        AsmNodeOperands.size());
+  New->setNodeId(-1);
+  return New.getNode();
+}
+
+
 bool ARMDAGToDAGISel::
 SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
                              std::vector<SDValue> &OutOps) {

test/CodeGen/ARM/arm-modifier.ll

@@ -61,8 +61,7 @@ ret void
 define i64 @f4(i64* %val) nounwind {
 entry:
   ;CHECK: f4
-  ;CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], {{r[0-9]?[13579]}}, [r0]
+  ;CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
-  ;CHECK: mov r0, [[REG1]]
   %0 = tail call i64 asm sideeffect "ldrexd $0, ${0:H}, [$1]", "=&r,r,*Qo"(i64* %val, i64* %val) nounwind
   ret i64 %0
 }

test/CodeGen/ARM/inlineasm-64bit.ll

@@ -0,0 +1,54 @@
+; RUN: llc < %s -O3  -mtriple=arm-linux-gnueabi | FileCheck %s
+
+; check if regs are passing correctly
+define void @i64_write(i64* %p, i64 %val) nounwind {
+; CHECK: i64_write:
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; CHECK: strexd [[REG1]], {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
+  %1 = tail call i64 asm sideeffect "1: ldrexd $0, ${0:H}, [$2]\0A strexd $0, $3, ${3:H}, [$2]\0A teq $0, #0\0A bne 1b", "=&r,=*Qo,r,r,~{cc}"(i64* %p, i64* %p, i64 %val) nounwind
+  ret void
+}
+
+; check if register allocation can reuse the registers
+define void @multi_writes(i64* %p, i64 %val1, i64 %val2, i64 %val3, i64 %val4, i64 %val5, i64 %val6) nounwind {
+entry:
+; CHECK: multi_writes:
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; check: strexd {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+
+  tail call void asm sideeffect " strexd $1, ${1:H}, [$0]\0A strexd $2, ${2:H}, [$0]\0A strexd $3, ${3:H}, [$0]\0A strexd $4, ${4:H}, [$0]\0A strexd $5, ${5:H}, [$0]\0A strexd $6, ${6:H}, [$0]\0A", "r,r,r,r,r,r,r"(i64* %p, i64 %val1, i64 %val2, i64 %val3, i64 %val4, i64 %val5, i64 %val6) nounwind
+  %incdec.ptr = getelementptr inbounds i64* %p, i32 1
+  tail call void asm sideeffect " strexd $1, ${1:H}, [$0]\0A strexd $2, ${2:H}, [$0]\0A strexd $3, ${3:H}, [$0]\0A strexd $4, ${4:H}, [$0]\0A strexd $5, ${5:H}, [$0]\0A strexd $6, ${6:H}, [$0]\0A", "r,r,r,r,r,r,r"(i64* %incdec.ptr, i64 %val1, i64 %val2, i64 %val3, i64 %val4, i64 %val5, i64 %val6) nounwind
+  tail call void asm sideeffect " strexd $1, ${1:H}, [$0]\0A strexd $2, ${2:H}, [$0]\0A strexd $3, ${3:H}, [$0]\0A strexd $4, ${4:H}, [$0]\0A strexd $5, ${5:H}, [$0]\0A strexd $6, ${6:H}, [$0]\0A", "r,r,r,r,r,r,r"(i64* %incdec.ptr, i64 %val1, i64 %val2, i64 %val3, i64 %val4, i64 %val5, i64 %val6) nounwind
+  ret void
+}
+
+
+; check if callee-saved registers used by inline asm are saved/restored
+define void @foo(i64* %p, i64 %i) nounwind {
+; CHECK:foo:
+; CHECK: push {{{r[4-9]|r10|r11}}
+; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], {{r[0-9]?[13579]}}, [r{{[0-9]+}}]
+; CHECK: strexd [[REG1]], {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
+; CHECK: pop {{{r[4-9]|r10|r11}}
+  %1 = tail call { i64, i64 } asm sideeffect "@ atomic64_set\0A1: ldrexd $0, ${0:H}, [$3]\0Aldrexd $1, ${1:H}, [$3]\0A strexd $0, $4, ${4:H}, [$3]\0A teq $0, #0\0A bne 1b", "=&r,=&r,=*Qo,r,r,~{cc}"(i64* %p, i64* %p, i64 %i) nounwind
+  ret void
+}