diff options
Diffstat (limited to 'lib/Target')
-rw-r--r-- | lib/Target/ARM/ARMFrameInfo.cpp | 23 | ||||
-rw-r--r-- | lib/Target/ARM/ARMISelLowering.cpp | 73 | ||||
-rw-r--r-- | lib/Target/ARM/ARMISelLowering.h | 2 | ||||
-rw-r--r-- | lib/Target/X86/README.txt | 18 | ||||
-rw-r--r-- | lib/Target/X86/X86FrameInfo.cpp | 15 | ||||
-rw-r--r-- | lib/Target/X86/X86ISelLowering.cpp | 41 | ||||
-rw-r--r-- | lib/Target/X86/X86ISelLowering.h | 2 |
7 files changed, 112 insertions, 62 deletions
diff --git a/lib/Target/ARM/ARMFrameInfo.cpp b/lib/Target/ARM/ARMFrameInfo.cpp index e2531d04d0..58a7c61d90 100644 --- a/lib/Target/ARM/ARMFrameInfo.cpp +++ b/lib/Target/ARM/ARMFrameInfo.cpp @@ -374,16 +374,19 @@ void ARMFrameInfo::emitEpilogue(MachineFunction &MF, MachineOperand &JumpTarget = MBBI->getOperand(0); // Jump to label or value in register. - if (RetOpcode == ARM::TCRETURNdi) { - BuildMI(MBB, MBBI, dl, - TII.get(STI.isThumb() ? ARM::TAILJMPdt : ARM::TAILJMPd)). - addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), - JumpTarget.getTargetFlags()); - } else if (RetOpcode == ARM::TCRETURNdiND) { - BuildMI(MBB, MBBI, dl, - TII.get(STI.isThumb() ? ARM::TAILJMPdNDt : ARM::TAILJMPdND)). - addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), - JumpTarget.getTargetFlags()); + if (RetOpcode == ARM::TCRETURNdi || RetOpcode == ARM::TCRETURNdiND) { + unsigned TCOpcode = (RetOpcode == ARM::TCRETURNdi) + ? (STI.isThumb() ? ARM::TAILJMPdt : ARM::TAILJMPd) + : (STI.isThumb() ? ARM::TAILJMPdNDt : ARM::TAILJMPdND); + MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(TCOpcode)); + if (JumpTarget.isGlobal()) + MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), + JumpTarget.getTargetFlags()); + else { + assert(JumpTarget.isSymbol()); + MIB.addExternalSymbol(JumpTarget.getSymbolName(), + JumpTarget.getTargetFlags()); + } } else if (RetOpcode == ARM::TCRETURNri) { BuildMI(MBB, MBBI, dl, TII.get(ARM::TAILJMPr)). addReg(JumpTarget.getReg(), RegState::Kill); diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index da274de5a0..3c90704b7e 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -1519,30 +1519,13 @@ ARMTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee, // whether LR is going to be used. Probably the right approach is to // generate the tail call here and turn it back into CALL/RET in // emitEpilogue if LR is used. - if (Subtarget->isThumb1Only()) - return false; - - // For the moment, we can only do this to functions defined in this - // compilation, or to indirect calls. A Thumb B to an ARM function, - // or vice versa, is not easily fixed up in the linker unlike BL. - // (We could do this by loading the address of the callee into a register; - // that is an extra instruction over the direct call and burns a register - // as well, so is not likely to be a win.) - - // It might be safe to remove this restriction on non-Darwin. // Thumb1 PIC calls to external symbols use BX, so they can be tail calls, // but we need to make sure there are enough registers; the only valid // registers are the 4 used for parameters. We don't currently do this // case. - if (isa<ExternalSymbolSDNode>(Callee)) - return false; - - if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) { - const GlobalValue *GV = G->getGlobal(); - if (GV->isDeclaration() || GV->isWeakForLinker()) - return false; - } + if (Subtarget->isThumb1Only()) + return false; // If the calling conventions do not match, then we'd better make sure the // results are returned in the same way as what the caller expects. @@ -1720,6 +1703,58 @@ ARMTargetLowering::LowerReturn(SDValue Chain, return result; } +bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N) const { + if (N->getNumValues() != 1) + return false; + if (!N->hasNUsesOfValue(1, 0)) + return false; + + unsigned NumCopies = 0; + SDNode* Copies[2]; + SDNode *Use = *N->use_begin(); + if (Use->getOpcode() == ISD::CopyToReg) { + Copies[NumCopies++] = Use; + } else if (Use->getOpcode() == ARMISD::VMOVRRD) { + // f64 returned in a pair of GPRs. + for (SDNode::use_iterator UI = Use->use_begin(), UE = Use->use_end(); + UI != UE; ++UI) { + if (UI->getOpcode() != ISD::CopyToReg) + return false; + Copies[UI.getUse().getResNo()] = *UI; + ++NumCopies; + } + } else if (Use->getOpcode() == ISD::BITCAST) { + // f32 returned in a single GPR. + if (!Use->hasNUsesOfValue(1, 0)) + return false; + Use = *Use->use_begin(); + if (Use->getOpcode() != ISD::CopyToReg || !Use->hasNUsesOfValue(1, 0)) + return false; + Copies[NumCopies++] = Use; + } else { + return false; + } + + if (NumCopies != 1 && NumCopies != 2) + return false; + for (unsigned i = 0; i < NumCopies; ++i) { + SDNode *Copy = Copies[i]; + for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end(); + UI != UE; ++UI) { + if (UI->getOpcode() == ISD::CopyToReg) { + SDNode *Use = *UI; + if (Use == Copies[0] || Use == Copies[1]) + continue; + return false; + } + if (UI->getOpcode() != ARMISD::RET_FLAG) + return false; + } + } + + return true; +} + // ConstantPool, JumpTable, GlobalAddress, and ExternalSymbol are lowered as // their target counterpart wrapped in the ARMISD::Wrapper node. Suppose N is // one of the above mentioned nodes. It has to be wrapped because otherwise diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h index 38f7399110..6d11bad357 100644 --- a/lib/Target/ARM/ARMISelLowering.h +++ b/lib/Target/ARM/ARMISelLowering.h @@ -421,6 +421,8 @@ namespace llvm { const SmallVectorImpl<SDValue> &OutVals, DebugLoc dl, SelectionDAG &DAG) const; + virtual bool isUsedByReturnOnly(SDNode *N) const; + SDValue getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC, SDValue &ARMcc, SelectionDAG &DAG, DebugLoc dl) const; SDValue getVFPCmp(SDValue LHS, SDValue RHS, diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt index e67fab1090..a305ae6ec5 100644 --- a/lib/Target/X86/README.txt +++ b/lib/Target/X86/README.txt @@ -895,24 +895,6 @@ compare: //===---------------------------------------------------------------------===// -Linux is missing some basic tail call support: - -#include <math.h> -double foo(double a) { return sin(a); } - -This compiles into this on x86-64 Linux (but not darwin): -foo: - subq $8, %rsp - call sin - addq $8, %rsp - ret -vs: - -foo: - jmp sin - -//===---------------------------------------------------------------------===// - Tail call optimization improvements: Tail call optimization currently pushes all arguments on the top of the stack (their normal place for non-tail call optimized calls) that source from the callers arguments diff --git a/lib/Target/X86/X86FrameInfo.cpp b/lib/Target/X86/X86FrameInfo.cpp index 494e905028..c47b0fa697 100644 --- a/lib/Target/X86/X86FrameInfo.cpp +++ b/lib/Target/X86/X86FrameInfo.cpp @@ -712,10 +712,17 @@ void X86FrameInfo::emitEpilogue(MachineFunction &MF, // Jump to label or value in register. if (RetOpcode == X86::TCRETURNdi || RetOpcode == X86::TCRETURNdi64) { - BuildMI(MBB, MBBI, DL, TII.get((RetOpcode == X86::TCRETURNdi) - ? X86::TAILJMPd : X86::TAILJMPd64)). - addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), - JumpTarget.getTargetFlags()); + MachineInstrBuilder MIB = + BuildMI(MBB, MBBI, DL, TII.get((RetOpcode == X86::TCRETURNdi) + ? X86::TAILJMPd : X86::TAILJMPd64)); + if (JumpTarget.isGlobal()) + MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), + JumpTarget.getTargetFlags()); + else { + assert(JumpTarget.isSymbol()); + MIB.addExternalSymbol(JumpTarget.getSymbolName(), + JumpTarget.getTargetFlags()); + } } else if (RetOpcode == X86::TCRETURNmi || RetOpcode == X86::TCRETURNmi64) { MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII.get((RetOpcode == X86::TCRETURNmi) diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 6d6d83d557..bc6d84eeec 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -1338,6 +1338,23 @@ X86TargetLowering::LowerReturn(SDValue Chain, MVT::Other, &RetOps[0], RetOps.size()); } +bool X86TargetLowering::isUsedByReturnOnly(SDNode *N) const { + if (N->getNumValues() != 1) + return false; + if (!N->hasNUsesOfValue(1, 0)) + return false; + + SDNode *Copy = *N->use_begin(); + if (Copy->getOpcode() != ISD::CopyToReg) + return false; + for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end(); + UI != UE; ++UI) + if (UI->getOpcode() != X86ISD::RET_FLAG) + return false; + + return true; +} + /// LowerCallResult - Lower the result values of a call into the /// appropriate copies out of appropriate physical registers. /// @@ -2142,17 +2159,19 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee, } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) { unsigned char OpFlags = 0; - // On ELF targets, in either X86-64 or X86-32 mode, direct calls to external - // symbols should go through the PLT. - if (Subtarget->isTargetELF() && - getTargetMachine().getRelocationModel() == Reloc::PIC_) { - OpFlags = X86II::MO_PLT; - } else if (Subtarget->isPICStyleStubAny() && - Subtarget->getDarwinVers() < 9) { - // PC-relative references to external symbols should go through $stub, - // unless we're building with the leopard linker or later, which - // automatically synthesizes these stubs. - OpFlags = X86II::MO_DARWIN_STUB; + if (!isTailCall) { + // On ELF targets, in either X86-64 or X86-32 mode, direct calls to + // external symbols should go through the PLT. + if (Subtarget->isTargetELF() && + getTargetMachine().getRelocationModel() == Reloc::PIC_) { + OpFlags = X86II::MO_PLT; + } else if (Subtarget->isPICStyleStubAny() && + Subtarget->getDarwinVers() < 9) { + // PC-relative references to external symbols should go through $stub, + // unless we're building with the leopard linker or later, which + // automatically synthesizes these stubs. + OpFlags = X86II::MO_DARWIN_STUB; + } } Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(), diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h index b521702013..edcdceb39a 100644 --- a/lib/Target/X86/X86ISelLowering.h +++ b/lib/Target/X86/X86ISelLowering.h @@ -805,6 +805,8 @@ namespace llvm { const SmallVectorImpl<SDValue> &OutVals, DebugLoc dl, SelectionDAG &DAG) const; + virtual bool isUsedByReturnOnly(SDNode *N) const; + virtual bool CanLowerReturn(CallingConv::ID CallConv, bool isVarArg, const SmallVectorImpl<ISD::OutputArg> &Outs, |