[mips] Add code to do tail call optimization.

Currently, it is enabled only if option "enable-mips-tail-calls" is given and
all of the callee's arguments are passed in registers.

llvm-svn: 166342

llvm/lib/Target/Mips/MipsISelLowering.cpp

@@ -25,6 +25,7 @@
 #include "llvm/GlobalVariable.h"
 #include "llvm/Intrinsics.h"
 #include "llvm/CallingConv.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -32,12 +33,19 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
 
+STATISTIC(NumTailCalls, "Number of tail calls");
+
+static cl::opt<bool>
+EnableMipsTailCalls("enable-mips-tail-calls", cl::Hidden,
+                    cl::desc("MIPS: Enable tail calls."), cl::init(false));
+
 // If I is a shifted mask, set the size (Size) and the first bit of the
 // mask (Pos), and return true.
 // For example, if I is 0x003ff800, (Pos, Size) = (11, 11).
@@ -2871,9 +2879,26 @@ PassByValArg64(SDValue Chain, DebugLoc dl,
   MemOpChains.push_back(Chain);
 }
 
+/// IsEligibleForTailCallOptimization - Check whether the call is eligible
+/// for tail call optimization.
+bool MipsTargetLowering::
+IsEligibleForTailCallOptimization(CallingConv::ID CalleeCC,
+                                  unsigned NextStackOffset) const {
+  if (!EnableMipsTailCalls)
+    return false;
+
+  // Do not tail-call optimize if there is an argument passed on stack.
+  if (IsO32 && (CalleeCC != CallingConv::Fast)) {
+    if (NextStackOffset > 16)
+      return false;
+  } else if (NextStackOffset)
+    return false;
+
+  return true;
+}
+
 /// LowerCall - functions arguments are copied from virtual regs to
 /// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
-/// TODO: isTailCall.
 SDValue
 MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                               SmallVectorImpl<SDValue> &InVals) const {
@@ -2888,9 +2913,6 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   CallingConv::ID CallConv              = CLI.CallConv;
   bool isVarArg                         = CLI.IsVarArg;
 
-  // MIPs target does not yet support tail call optimization.
-  isTailCall = false;
-
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
   const TargetFrameLowering *TFL = MF.getTarget().getFrameLowering();
@@ -2921,11 +2943,20 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   if (IsO32 && (CallConv != CallingConv::Fast))
     NextStackOffset = std::max(NextStackOffset, (unsigned)16);
 
+  // Check if it's really possible to do a tail call.
+  if (isTailCall)
+    isTailCall = IsEligibleForTailCallOptimization(CallConv, NextStackOffset);
+
+  if (isTailCall)
+    ++NumTailCalls;
+
   // Chain is the output chain of the last Load/Store or CopyToReg node.
   // ByValChain is the output chain of the last Memcpy node created for copying
   // byval arguments to the stack.
   SDValue NextStackOffsetVal = DAG.getIntPtrConstant(NextStackOffset, true);
-  Chain = DAG.getCALLSEQ_START(Chain, NextStackOffsetVal);
+
+  if (!isTailCall)
+    Chain = DAG.getCALLSEQ_START(Chain, NextStackOffsetVal);
 
   SDValue StackPtr = DAG.getCopyFromReg(Chain, dl,
                                         IsN64 ? Mips::SP_64 : Mips::SP,
@@ -3135,6 +3166,9 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   if (InFlag.getNode())
     Ops.push_back(InFlag);
 
+  if (isTailCall)
+    return DAG.getNode(MipsISD::TailCall, dl, MVT::Other, &Ops[0], Ops.size());
+
   Chain  = DAG.getNode(MipsISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size());
   InFlag = Chain.getValue(1);
 

llvm/lib/Target/Mips/MipsISelLowering.h

@@ -208,6 +208,11 @@ namespace llvm {
     SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
 
+    /// IsEligibleForTailCallOptimization - Check whether the call is eligible
+    /// for tail call optimization.
+    bool IsEligibleForTailCallOptimization(CallingConv::ID CalleeCC,
+                                           unsigned NextStackOffset) const;
+
     virtual SDValue
       LowerFormalArguments(SDValue Chain,
                            CallingConv::ID CallConv, bool isVarArg,

llvm/test/CodeGen/Mips/tailcall.ll

@@ -0,0 +1,100 @@
+; RUN: llc -march=mipsel -relocation-model=pic -enable-mips-tail-calls < %s | \
+; RUN: FileCheck %s -check-prefix=PIC32
+; RUN: llc -march=mipsel -relocation-model=static \
+; RUN: -enable-mips-tail-calls < %s | FileCheck %s -check-prefix=STATIC32
+; RUN: llc -march=mips64el -mcpu=mips64r2 -mattr=+n64 -enable-mips-tail-calls \
+; RUN: < %s | FileCheck %s -check-prefix=N64
+
+@g0 = common global i32 0, align 4
+@g1 = common global i32 0, align 4
+@g2 = common global i32 0, align 4
+@g3 = common global i32 0, align 4
+@g4 = common global i32 0, align 4
+@g5 = common global i32 0, align 4
+@g6 = common global i32 0, align 4
+@g7 = common global i32 0, align 4
+@g8 = common global i32 0, align 4
+@g9 = common global i32 0, align 4
+
+define i32 @caller1(i32 %a0) nounwind {
+entry:
+; PIC32-NOT: jalr
+; STATIC32-NOT: jal
+; N64-NOT: jalr
+
+  %call = tail call i32 @callee1(i32 1, i32 1, i32 1, i32 %a0) nounwind
+  ret i32 %call
+}
+
+declare i32 @callee1(i32, i32, i32, i32)
+
+define i32 @caller2(i32 %a0, i32 %a1, i32 %a2, i32 %a3) nounwind {
+entry:
+; PIC32: jalr
+; STATIC32: jal
+; N64-NOT: jalr
+
+  %call = tail call i32 @callee2(i32 1, i32 %a0, i32 %a1, i32 %a2, i32 %a3) nounwind
+  ret i32 %call
+}
+
+declare i32 @callee2(i32, i32, i32, i32, i32)
+
+define i32 @caller3(i32 %a0, i32 %a1, i32 %a2, i32 %a3, i32 %a4) nounwind {
+entry:
+; PIC32: jalr
+; STATIC32: jal
+; N64-NOT: jalr
+
+  %call = tail call i32 @callee3(i32 1, i32 1, i32 1, i32 %a0, i32 %a1, i32 %a2, i32 %a3, i32 %a4) nounwind
+  ret i32 %call
+}
+
+declare i32 @callee3(i32, i32, i32, i32, i32, i32, i32, i32)
+
+define i32 @caller4(i32 %a0, i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6, i32 %a7) nounwind {
+entry:
+; PIC32: jalr
+; STATIC32: jal
+; N64: jalr
+
+  %call = tail call i32 @callee4(i32 1, i32 %a0, i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6, i32 %a7) nounwind
+  ret i32 %call
+}
+
+declare i32 @callee4(i32, i32, i32, i32, i32, i32, i32, i32, i32)
+
+define i32 @caller5() nounwind readonly {
+entry:
+; PIC32-NOT: jalr
+; STATIC32-NOT: jal
+; N64-NOT: jalr
+
+  %0 = load i32* @g0, align 4
+  %1 = load i32* @g1, align 4
+  %2 = load i32* @g2, align 4
+  %3 = load i32* @g3, align 4
+  %4 = load i32* @g4, align 4
+  %5 = load i32* @g5, align 4
+  %6 = load i32* @g6, align 4
+  %7 = load i32* @g7, align 4
+  %8 = load i32* @g8, align 4
+  %9 = load i32* @g9, align 4
+  %call = tail call fastcc i32 @callee5(i32 %0, i32 %1, i32 %2, i32 %3, i32 %4, i32 %5, i32 %6, i32 %7, i32 %8, i32 %9)
+  ret i32 %call
+}
+
+define internal fastcc i32 @callee5(i32 %a0, i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6, i32 %a7, i32 %a8, i32 %a9) nounwind readnone noinline {
+entry:
+  %add = add nsw i32 %a1, %a0
+  %add1 = add nsw i32 %add, %a2
+  %add2 = add nsw i32 %add1, %a3
+  %add3 = add nsw i32 %add2, %a4
+  %add4 = add nsw i32 %add3, %a5
+  %add5 = add nsw i32 %add4, %a6
+  %add6 = add nsw i32 %add5, %a7
+  %add7 = add nsw i32 %add6, %a8
+  %add8 = add nsw i32 %add7, %a9
+  ret i32 %add8
+}
+