diff options
Diffstat (limited to 'lib/Target/ARM')
59 files changed, 3233 insertions, 2553 deletions
diff --git a/lib/Target/ARM/ARM.h b/lib/Target/ARM/ARM.h index 4679f74..08dc340 100644 --- a/lib/Target/ARM/ARM.h +++ b/lib/Target/ARM/ARM.h @@ -16,24 +16,29 @@ #define TARGET_ARM_H #include "ARMBaseInfo.h" +#include "MCTargetDesc/ARMMCTargetDesc.h" +#include "llvm/Support/DataTypes.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Target/TargetMachine.h" #include <cassert> namespace llvm { +class ARMAsmPrinter; class ARMBaseTargetMachine; class FunctionPass; class JITCodeEmitter; -class formatted_raw_ostream; -class MCCodeEmitter; -class TargetAsmBackend; class MachineInstr; -class ARMAsmPrinter; +class MCCodeEmitter; class MCInst; +class MCInstrInfo; +class MCObjectWriter; +class MCSubtargetInfo; +class TargetAsmBackend; +class formatted_raw_ostream; -MCCodeEmitter *createARMMCCodeEmitter(const Target &, - TargetMachine &TM, +MCCodeEmitter *createARMMCCodeEmitter(const MCInstrInfo &MCII, + const MCSubtargetInfo &STI, MCContext &Ctx); TargetAsmBackend *createARMAsmBackend(const Target &, const std::string &); @@ -53,11 +58,15 @@ FunctionPass *createMLxExpansionPass(); FunctionPass *createThumb2ITBlockPass(); FunctionPass *createThumb2SizeReductionPass(); -extern Target TheARMTarget, TheThumbTarget; - void LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI, ARMAsmPrinter &AP); +/// createARMMachObjectWriter - Construct an ARM Mach-O object writer. +MCObjectWriter *createARMMachObjectWriter(raw_ostream &OS, + bool Is64Bit, + uint32_t CPUType, + uint32_t CPUSubtype); + } // end namespace llvm; #endif diff --git a/lib/Target/ARM/ARM.td b/lib/Target/ARM/ARM.td index 6af5f85..cf333cc 100644 --- a/lib/Target/ARM/ARM.td +++ b/lib/Target/ARM/ARM.td @@ -16,18 +16,26 @@ include "llvm/Target/Target.td" +//===----------------------------------------------------------------------===// +// ARM Subtarget state. +// + +def ModeThumb : SubtargetFeature<"thumb-mode", "InThumbMode", "true", + "Thumb mode">; //===----------------------------------------------------------------------===// // ARM Subtarget features. // -def FeatureVFP2 : SubtargetFeature<"vfp2", "ARMFPUType", "VFPv2", +def FeatureVFP2 : SubtargetFeature<"vfp2", "HasVFPv2", "true", "Enable VFP2 instructions">; -def FeatureVFP3 : SubtargetFeature<"vfp3", "ARMFPUType", "VFPv3", - "Enable VFP3 instructions">; -def FeatureNEON : SubtargetFeature<"neon", "ARMFPUType", "NEON", - "Enable NEON instructions">; -def FeatureThumb2 : SubtargetFeature<"thumb2", "ThumbMode", "Thumb2", +def FeatureVFP3 : SubtargetFeature<"vfp3", "HasVFPv3", "true", + "Enable VFP3 instructions", + [FeatureVFP2]>; +def FeatureNEON : SubtargetFeature<"neon", "HasNEON", "true", + "Enable NEON instructions", + [FeatureVFP3]>; +def FeatureThumb2 : SubtargetFeature<"thumb2", "HasThumb2", "true", "Enable Thumb2 instructions">; def FeatureNoARM : SubtargetFeature<"noarm", "NoARM", "true", "Does not support ARM mode execution">; @@ -75,32 +83,32 @@ def FeatureAvoidPartialCPSR : SubtargetFeature<"avoid-partial-cpsr", "AvoidCPSRPartialUpdate", "true", "Avoid CPSR partial update for OOO execution">; +/// Some M architectures don't have the DSP extension (v7E-M vs. v7M) +def FeatureDSPThumb2 : SubtargetFeature<"t2dsp", "Thumb2DSP", "true", + "Supports v7 DSP instructions in Thumb2.">; + // Multiprocessing extension. def FeatureMP : SubtargetFeature<"mp", "HasMPExtension", "true", "Supports Multiprocessing extension">; -// ARM architectures. -def ArchV4T : SubtargetFeature<"v4t", "ARMArchVersion", "V4T", - "ARM v4T">; -def ArchV5T : SubtargetFeature<"v5t", "ARMArchVersion", "V5T", - "ARM v5T">; -def ArchV5TE : SubtargetFeature<"v5te", "ARMArchVersion", "V5TE", - "ARM v5TE, v5TEj, v5TExp">; -def ArchV6 : SubtargetFeature<"v6", "ARMArchVersion", "V6", - "ARM v6">; -def ArchV6M : SubtargetFeature<"v6m", "ARMArchVersion", "V6M", - "ARM v6m", - [FeatureNoARM, FeatureDB]>; -def ArchV6T2 : SubtargetFeature<"v6t2", "ARMArchVersion", "V6T2", - "ARM v6t2", - [FeatureThumb2]>; -def ArchV7A : SubtargetFeature<"v7a", "ARMArchVersion", "V7A", - "ARM v7A", - [FeatureThumb2, FeatureNEON, FeatureDB]>; -def ArchV7M : SubtargetFeature<"v7m", "ARMArchVersion", "V7M", - "ARM v7M", - [FeatureThumb2, FeatureNoARM, FeatureDB, - FeatureHWDiv]>; +// ARM ISAs. +def HasV4TOps : SubtargetFeature<"v4t", "HasV4TOps", "true", + "Support ARM v4T instructions">; +def HasV5TOps : SubtargetFeature<"v5t", "HasV5TOps", "true", + "Support ARM v5T instructions", + [HasV4TOps]>; +def HasV5TEOps : SubtargetFeature<"v5te", "HasV5TEOps", "true", + "Support ARM v5TE, v5TEj, and v5TExp instructions", + [HasV5TOps]>; +def HasV6Ops : SubtargetFeature<"v6", "HasV6Ops", "true", + "Support ARM v6 instructions", + [HasV5TEOps]>; +def HasV6T2Ops : SubtargetFeature<"v6t2", "HasV6T2Ops", "true", + "Support ARM v6t2 instructions", + [HasV6Ops, FeatureThumb2, FeatureDSPThumb2]>; +def HasV7Ops : SubtargetFeature<"v7", "HasV7Ops", "true", + "Support ARM v7 instructions", + [HasV6T2Ops]>; //===----------------------------------------------------------------------===// // ARM Processors supported. @@ -109,8 +117,6 @@ def ArchV7M : SubtargetFeature<"v7m", "ARMArchVersion", "V7M", include "ARMSchedule.td" // ARM processor families. -def ProcOthers : SubtargetFeature<"others", "ARMProcFamily", "Others", - "One of the other ARM processor families">; def ProcA8 : SubtargetFeature<"a8", "ARMProcFamily", "CortexA8", "Cortex-A8 ARM processors", [FeatureSlowFPBrcc, FeatureNEONForFP, @@ -135,64 +141,76 @@ def : ProcNoItin<"strongarm1100", []>; def : ProcNoItin<"strongarm1110", []>; // V4T Processors. -def : ProcNoItin<"arm7tdmi", [ArchV4T]>; -def : ProcNoItin<"arm7tdmi-s", [ArchV4T]>; -def : ProcNoItin<"arm710t", [ArchV4T]>; -def : ProcNoItin<"arm720t", [ArchV4T]>; -def : ProcNoItin<"arm9", [ArchV4T]>; -def : ProcNoItin<"arm9tdmi", [ArchV4T]>; -def : ProcNoItin<"arm920", [ArchV4T]>; -def : ProcNoItin<"arm920t", [ArchV4T]>; -def : ProcNoItin<"arm922t", [ArchV4T]>; -def : ProcNoItin<"arm940t", [ArchV4T]>; -def : ProcNoItin<"ep9312", [ArchV4T]>; +def : ProcNoItin<"arm7tdmi", [HasV4TOps]>; +def : ProcNoItin<"arm7tdmi-s", [HasV4TOps]>; +def : ProcNoItin<"arm710t", [HasV4TOps]>; +def : ProcNoItin<"arm720t", [HasV4TOps]>; +def : ProcNoItin<"arm9", [HasV4TOps]>; +def : ProcNoItin<"arm9tdmi", [HasV4TOps]>; +def : ProcNoItin<"arm920", [HasV4TOps]>; +def : ProcNoItin<"arm920t", [HasV4TOps]>; +def : ProcNoItin<"arm922t", [HasV4TOps]>; +def : ProcNoItin<"arm940t", [HasV4TOps]>; +def : ProcNoItin<"ep9312", [HasV4TOps]>; // V5T Processors. -def : ProcNoItin<"arm10tdmi", [ArchV5T]>; -def : ProcNoItin<"arm1020t", [ArchV5T]>; +def : ProcNoItin<"arm10tdmi", [HasV5TOps]>; +def : ProcNoItin<"arm1020t", [HasV5TOps]>; // V5TE Processors. -def : ProcNoItin<"arm9e", [ArchV5TE]>; -def : ProcNoItin<"arm926ej-s", [ArchV5TE]>; -def : ProcNoItin<"arm946e-s", [ArchV5TE]>; -def : ProcNoItin<"arm966e-s", [ArchV5TE]>; -def : ProcNoItin<"arm968e-s", [ArchV5TE]>; -def : ProcNoItin<"arm10e", [ArchV5TE]>; -def : ProcNoItin<"arm1020e", [ArchV5TE]>; -def : ProcNoItin<"arm1022e", [ArchV5TE]>; -def : ProcNoItin<"xscale", [ArchV5TE]>; -def : ProcNoItin<"iwmmxt", [ArchV5TE]>; +def : ProcNoItin<"arm9e", [HasV5TEOps]>; +def : ProcNoItin<"arm926ej-s", [HasV5TEOps]>; +def : ProcNoItin<"arm946e-s", [HasV5TEOps]>; +def : ProcNoItin<"arm966e-s", [HasV5TEOps]>; +def : ProcNoItin<"arm968e-s", [HasV5TEOps]>; +def : ProcNoItin<"arm10e", [HasV5TEOps]>; +def : ProcNoItin<"arm1020e", [HasV5TEOps]>; +def : ProcNoItin<"arm1022e", [HasV5TEOps]>; +def : ProcNoItin<"xscale", [HasV5TEOps]>; +def : ProcNoItin<"iwmmxt", [HasV5TEOps]>; // V6 Processors. -def : Processor<"arm1136j-s", ARMV6Itineraries, [ArchV6]>; -def : Processor<"arm1136jf-s", ARMV6Itineraries, [ArchV6, FeatureVFP2, +def : Processor<"arm1136j-s", ARMV6Itineraries, [HasV6Ops]>; +def : Processor<"arm1136jf-s", ARMV6Itineraries, [HasV6Ops, FeatureVFP2, FeatureHasSlowFPVMLx]>; -def : Processor<"arm1176jz-s", ARMV6Itineraries, [ArchV6]>; -def : Processor<"arm1176jzf-s", ARMV6Itineraries, [ArchV6, FeatureVFP2, +def : Processor<"arm1176jz-s", ARMV6Itineraries, [HasV6Ops]>; +def : Processor<"arm1176jzf-s", ARMV6Itineraries, [HasV6Ops, FeatureVFP2, FeatureHasSlowFPVMLx]>; -def : Processor<"mpcorenovfp", ARMV6Itineraries, [ArchV6]>; -def : Processor<"mpcore", ARMV6Itineraries, [ArchV6, FeatureVFP2, +def : Processor<"mpcorenovfp", ARMV6Itineraries, [HasV6Ops]>; +def : Processor<"mpcore", ARMV6Itineraries, [HasV6Ops, FeatureVFP2, FeatureHasSlowFPVMLx]>; // V6M Processors. -def : Processor<"cortex-m0", ARMV6Itineraries, [ArchV6M]>; +def : Processor<"cortex-m0", ARMV6Itineraries, [HasV6Ops, FeatureNoARM, + FeatureDB]>; // V6T2 Processors. -def : Processor<"arm1156t2-s", ARMV6Itineraries, [ArchV6T2]>; -def : Processor<"arm1156t2f-s", ARMV6Itineraries, [ArchV6T2, FeatureVFP2, +def : Processor<"arm1156t2-s", ARMV6Itineraries, [HasV6T2Ops]>; +def : Processor<"arm1156t2f-s", ARMV6Itineraries, [HasV6T2Ops, FeatureVFP2, FeatureHasSlowFPVMLx]>; -// V7 Processors. +// V7a Processors. def : Processor<"cortex-a8", CortexA8Itineraries, - [ArchV7A, ProcA8]>; + [ProcA8, HasV7Ops, FeatureNEON, FeatureDB, + FeatureDSPThumb2]>; def : Processor<"cortex-a9", CortexA9Itineraries, - [ArchV7A, ProcA9]>; + [ProcA9, HasV7Ops, FeatureNEON, FeatureDB, + FeatureDSPThumb2]>; def : Processor<"cortex-a9-mp", CortexA9Itineraries, - [ArchV7A, ProcA9, FeatureMP]>; + [ProcA9, HasV7Ops, FeatureNEON, FeatureDB, + FeatureDSPThumb2, FeatureMP]>; // V7M Processors. -def : ProcNoItin<"cortex-m3", [ArchV7M]>; -def : ProcNoItin<"cortex-m4", [ArchV7M, FeatureVFP2, FeatureVFPOnlySP]>; +def : ProcNoItin<"cortex-m3", [HasV7Ops, + FeatureThumb2, FeatureNoARM, FeatureDB, + FeatureHWDiv]>; + +// V7EM Processors. +def : ProcNoItin<"cortex-m4", [HasV7Ops, + FeatureThumb2, FeatureNoARM, FeatureDB, + FeatureHWDiv, FeatureDSPThumb2, + FeatureT2XtPk, FeatureVFP2, + FeatureVFPOnlySP]>; //===----------------------------------------------------------------------===// // Register File Description diff --git a/lib/Target/ARM/ARMAsmBackend.cpp b/lib/Target/ARM/ARMAsmBackend.cpp index 618a2b5..5e438a9 100644 --- a/lib/Target/ARM/ARMAsmBackend.cpp +++ b/lib/Target/ARM/ARMAsmBackend.cpp @@ -28,14 +28,6 @@ using namespace llvm; namespace { -class ARMMachObjectWriter : public MCMachObjectTargetWriter { -public: - ARMMachObjectWriter(bool Is64Bit, uint32_t CPUType, - uint32_t CPUSubtype) - : MCMachObjectTargetWriter(Is64Bit, CPUType, CPUSubtype, - /*UseAggressiveSymbolFolding=*/true) {} -}; - class ARMELFObjectWriter : public MCELFObjectTargetWriter { public: ARMELFObjectWriter(Triple::OSType OSType) @@ -182,7 +174,8 @@ static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) { Value >>= 16; // Fallthrough case ARM::fixup_t2_movw_lo16: - case ARM::fixup_t2_movt_hi16_pcrel: + case ARM::fixup_t2_movt_hi16_pcrel: //FIXME: Shouldn't this be shifted like + // the other hi16 fixup? case ARM::fixup_t2_movw_lo16_pcrel: { unsigned Hi4 = (Value & 0xF000) >> 12; unsigned i = (Value & 0x800) >> 11; @@ -192,8 +185,10 @@ static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) { // inst{26} = i; // inst{14-12} = Mid3; // inst{7-0} = Lo8; - assert ((((int64_t)Value) >= -0x8000) && (((int64_t)Value) <= 0x7fff) && - "Out of range pc-relative fixup value!"); + // The value comes in as the whole thing, not just the portion required + // for this fixup, so we need to mask off the bits not handled by this + // portion (lo vs. hi). + Value &= 0xffff; Value = (Hi4 << 16) | (i << 26) | (Mid3 << 12) | (Lo8); uint64_t swapped = (Value & 0xFFFF0000) >> 16; swapped |= (Value & 0x0000FFFF) << 16; @@ -423,12 +418,9 @@ public: : ARMAsmBackend(T), Subtype(st) { } MCObjectWriter *createObjectWriter(raw_ostream &OS) const { - return createMachObjectWriter(new ARMMachObjectWriter( - /*Is64Bit=*/false, - object::mach::CTM_ARM, - Subtype), - OS, - /*IsLittleEndian=*/true); + return createARMMachObjectWriter(OS, /*Is64Bit=*/false, + object::mach::CTM_ARM, + Subtype); } void ApplyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize, @@ -505,7 +497,13 @@ TargetAsmBackend *llvm::createARMAsmBackend(const Target &T, Triple TheTriple(TT); if (TheTriple.isOSDarwin()) { - if (TheTriple.getArchName() == "armv6" || + if (TheTriple.getArchName() == "armv4t" || + TheTriple.getArchName() == "thumbv4t") + return new DarwinARMAsmBackend(T, object::mach::CSARM_V4T); + else if (TheTriple.getArchName() == "armv5e" || + TheTriple.getArchName() == "thumbv5e") + return new DarwinARMAsmBackend(T, object::mach::CSARM_V5TEJ); + else if (TheTriple.getArchName() == "armv6" || TheTriple.getArchName() == "thumbv6") return new DarwinARMAsmBackend(T, object::mach::CSARM_V6); return new DarwinARMAsmBackend(T, object::mach::CSARM_V7); diff --git a/lib/Target/ARM/ARMAsmPrinter.cpp b/lib/Target/ARM/ARMAsmPrinter.cpp index eb73902..dbc3ee4 100644 --- a/lib/Target/ARM/ARMAsmPrinter.cpp +++ b/lib/Target/ARM/ARMAsmPrinter.cpp @@ -654,7 +654,7 @@ void ARMAsmPrinter::emitAttributes() { } /* TODO: ARMBuildAttrs::Allowed is not completely accurate, - * since NEON can have 1 (allowed) or 2 (fused MAC operations) */ + * since NEON can have 1 (allowed) or 2 (MAC operations) */ if (Subtarget->hasNEON()) { AttrEmitter->EmitAttribute(ARMBuildAttrs::Advanced_SIMD_arch, ARMBuildAttrs::Allowed); @@ -1010,19 +1010,16 @@ void ARMAsmPrinter::EmitUnwindingInstruction(const MachineInstr *MI) { MI->dump(); assert(0 && "Unsupported opcode for unwinding information"); case ARM::MOVr: - case ARM::tMOVgpr2gpr: - case ARM::tMOVgpr2tgpr: Offset = 0; break; case ARM::ADDri: Offset = -MI->getOperand(2).getImm(); break; case ARM::SUBri: - case ARM::t2SUBrSPi: - Offset = MI->getOperand(2).getImm(); + Offset = MI->getOperand(2).getImm(); break; case ARM::tSUBspi: - Offset = MI->getOperand(2).getImm()*4; + Offset = MI->getOperand(2).getImm()*4; break; case ARM::tADDspi: case ARM::tADDrSPi: @@ -1072,39 +1069,18 @@ void ARMAsmPrinter::EmitUnwindingInstruction(const MachineInstr *MI) { extern cl::opt<bool> EnableARMEHABI; +// Simple pseudo-instructions have their lowering (with expansion to real +// instructions) auto-generated. +#include "ARMGenMCPseudoLowering.inc" + void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { - unsigned Opc = MI->getOpcode(); - switch (Opc) { - default: break; - case ARM::B: { - // B is just a Bcc with an 'always' predicate. - MCInst TmpInst; - LowerARMMachineInstrToMCInst(MI, TmpInst, *this); - TmpInst.setOpcode(ARM::Bcc); - // Add predicate operands. - TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); - TmpInst.addOperand(MCOperand::CreateReg(0)); - OutStreamer.EmitInstruction(TmpInst); - return; - } - case ARM::LDMIA_RET: { - // LDMIA_RET is just a normal LDMIA_UPD instruction that targets PC and as - // such has additional code-gen properties and scheduling information. - // To emit it, we just construct as normal and set the opcode to LDMIA_UPD. - MCInst TmpInst; - LowerARMMachineInstrToMCInst(MI, TmpInst, *this); - TmpInst.setOpcode(ARM::LDMIA_UPD); - OutStreamer.EmitInstruction(TmpInst); + // Do any auto-generated pseudo lowerings. + if (emitPseudoExpansionLowering(OutStreamer, MI)) return; - } - case ARM::t2ADDrSPi: - case ARM::t2ADDrSPi12: - case ARM::t2SUBrSPi: - case ARM::t2SUBrSPi12: - assert ((MI->getOperand(1).getReg() == ARM::SP) && - "Unexpected source register!"); - break; + // Check for manual lowerings. + unsigned Opc = MI->getOpcode(); + switch (Opc) { case ARM::t2MOVi32imm: assert(0 && "Should be lowered by thumb2it pass"); case ARM::DBG_VALUE: { if (isVerbose() && OutStreamer.hasRawTextSupport()) { @@ -1115,14 +1091,6 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { } return; } - case ARM::tBfar: { - MCInst TmpInst; - TmpInst.setOpcode(ARM::tBL); - TmpInst.addOperand(MCOperand::CreateExpr(MCSymbolRefExpr::Create( - MI->getOperand(0).getMBB()->getSymbol(), OutContext))); - OutStreamer.EmitInstruction(TmpInst); - return; - } case ARM::LEApcrel: case ARM::tLEApcrel: case ARM::t2LEApcrel: { @@ -1153,39 +1121,8 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { OutStreamer.EmitInstruction(TmpInst); return; } - case ARM::MOVPCRX: { - MCInst TmpInst; - TmpInst.setOpcode(ARM::MOVr); - TmpInst.addOperand(MCOperand::CreateReg(ARM::PC)); - TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg())); - // Add predicate operands. - TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); - TmpInst.addOperand(MCOperand::CreateReg(0)); - // Add 's' bit operand (always reg0 for this) - TmpInst.addOperand(MCOperand::CreateReg(0)); - OutStreamer.EmitInstruction(TmpInst); - return; - } // Darwin call instructions are just normal call instructions with different // clobber semantics (they clobber R9). - case ARM::BLr9: - case ARM::BLr9_pred: - case ARM::BLXr9: - case ARM::BLXr9_pred: { - unsigned newOpc; - switch (Opc) { - default: assert(0); - case ARM::BLr9: newOpc = ARM::BL; break; - case ARM::BLr9_pred: newOpc = ARM::BL_pred; break; - case ARM::BLXr9: newOpc = ARM::BLX; break; - case ARM::BLXr9_pred: newOpc = ARM::BLX_pred; break; - } - MCInst TmpInst; - LowerARMMachineInstrToMCInst(MI, TmpInst, *this); - TmpInst.setOpcode(newOpc); - OutStreamer.EmitInstruction(TmpInst); - return; - } case ARM::BXr9_CALL: case ARM::BX_CALL: { { @@ -1215,6 +1152,9 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { TmpInst.setOpcode(ARM::tMOVr); TmpInst.addOperand(MCOperand::CreateReg(ARM::LR)); TmpInst.addOperand(MCOperand::CreateReg(ARM::PC)); + // Add predicate operands. + TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); + TmpInst.addOperand(MCOperand::CreateReg(0)); OutStreamer.EmitInstruction(TmpInst); } { @@ -1445,7 +1385,7 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { case ARM::t2BR_JT: { // Lower and emit the instruction itself, then the jump table following it. MCInst TmpInst; - TmpInst.setOpcode(ARM::tMOVgpr2gpr); + TmpInst.setOpcode(ARM::tMOVr); TmpInst.addOperand(MCOperand::CreateReg(ARM::PC)); TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg())); // Add predicate operands. @@ -1494,7 +1434,7 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { // mov pc, target MCInst TmpInst; unsigned Opc = MI->getOpcode() == ARM::BR_JTr ? - ARM::MOVr : ARM::tMOVgpr2gpr; + ARM::MOVr : ARM::tMOVr; TmpInst.setOpcode(Opc); TmpInst.addOperand(MCOperand::CreateReg(ARM::PC)); TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg())); @@ -1507,7 +1447,7 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { OutStreamer.EmitInstruction(TmpInst); // Make sure the Thumb jump table is 4-byte aligned. - if (Opc == ARM::tMOVgpr2gpr) + if (Opc == ARM::tMOVr) EmitAlignment(2); // Output the data for the jump table itself @@ -1599,11 +1539,12 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { MCSymbol *Label = GetARMSJLJEHLabel(); { MCInst TmpInst; - TmpInst.setOpcode(ARM::tMOVgpr2tgpr); + TmpInst.setOpcode(ARM::tMOVr); TmpInst.addOperand(MCOperand::CreateReg(ValReg)); TmpInst.addOperand(MCOperand::CreateReg(ARM::PC)); - // 's' bit operand - TmpInst.addOperand(MCOperand::CreateReg(ARM::CPSR)); + // Predicate. + TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); + TmpInst.addOperand(MCOperand::CreateReg(0)); OutStreamer.AddComment("eh_setjmp begin"); OutStreamer.EmitInstruction(TmpInst); } @@ -1817,7 +1758,7 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { } { MCInst TmpInst; - TmpInst.setOpcode(ARM::tMOVtgpr2gpr); + TmpInst.setOpcode(ARM::tMOVr); TmpInst.addOperand(MCOperand::CreateReg(ARM::SP)); TmpInst.addOperand(MCOperand::CreateReg(ScratchReg)); // Predicate. @@ -1858,75 +1799,6 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { } return; } - // Tail jump branches are really just branch instructions with additional - // code-gen attributes. Convert them to the canonical form here. - case ARM::TAILJMPd: - case ARM::TAILJMPdND: { - MCInst TmpInst, TmpInst2; - // Lower the instruction as-is to get the operands properly converted. - LowerARMMachineInstrToMCInst(MI, TmpInst2, *this); - TmpInst.setOpcode(ARM::Bcc); - TmpInst.addOperand(TmpInst2.getOperand(0)); - // Add predicate operands. - TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); - TmpInst.addOperand(MCOperand::CreateReg(0)); - OutStreamer.AddComment("TAILCALL"); - OutStreamer.EmitInstruction(TmpInst); - return; - } - case ARM::tTAILJMPd: - case ARM::tTAILJMPdND: { - MCInst TmpInst, TmpInst2; - LowerARMMachineInstrToMCInst(MI, TmpInst2, *this); - // The Darwin toolchain doesn't support tail call relocations of 16-bit - // branches. - TmpInst.setOpcode(Opc == ARM::tTAILJMPd ? ARM::t2B : ARM::tB); - TmpInst.addOperand(TmpInst2.getOperand(0)); - OutStreamer.AddComment("TAILCALL"); - OutStreamer.EmitInstruction(TmpInst); - return; - } - case ARM::TAILJMPrND: - case ARM::tTAILJMPrND: - case ARM::TAILJMPr: - case ARM::tTAILJMPr: { - unsigned newOpc = (Opc == ARM::TAILJMPr || Opc == ARM::TAILJMPrND) - ? ARM::BX : ARM::tBX; - MCInst TmpInst; - TmpInst.setOpcode(newOpc); - TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg())); - // Predicate. - TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL)); - TmpInst.addOperand(MCOperand::CreateReg(0)); - OutStreamer.AddComment("TAILCALL"); - OutStreamer.EmitInstruction(TmpInst); - return; - } - - // These are the pseudos created to comply with stricter operand restrictions - // on ARMv5. Lower them now to "normal" instructions, since all the - // restrictions are already satisfied. - case ARM::MULv5: - EmitPatchedInstruction(MI, ARM::MUL); - return; - case ARM::MLAv5: - EmitPatchedInstruction(MI, ARM::MLA); - return; - case ARM::SMULLv5: - EmitPatchedInstruction(MI, ARM::SMULL); - return; - case ARM::UMULLv5: - EmitPatchedInstruction(MI, ARM::UMULL); - return; - case ARM::SMLALv5: - EmitPatchedInstruction(MI, ARM::SMLAL); - return; - case ARM::UMLALv5: - EmitPatchedInstruction(MI, ARM::UMLAL); - return; - case ARM::UMAALv5: - EmitPatchedInstruction(MI, ARM::UMAAL); - return; } MCInst TmpInst; @@ -1944,11 +1816,10 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) { //===----------------------------------------------------------------------===// static MCInstPrinter *createARMMCInstPrinter(const Target &T, - TargetMachine &TM, unsigned SyntaxVariant, const MCAsmInfo &MAI) { if (SyntaxVariant == 0) - return new ARMInstPrinter(TM, MAI); + return new ARMInstPrinter(MAI); return 0; } diff --git a/lib/Target/ARM/ARMAsmPrinter.h b/lib/Target/ARM/ARMAsmPrinter.h index 5f9169e..7741fc4 100644 --- a/lib/Target/ARM/ARMAsmPrinter.h +++ b/lib/Target/ARM/ARMAsmPrinter.h @@ -21,6 +21,8 @@ namespace llvm { +class MCOperand; + namespace ARM { enum DW_ISA { DW_ISA_ARM_thumb = 1, @@ -72,6 +74,9 @@ public: void EmitStartOfAsmFile(Module &M); void EmitEndOfAsmFile(Module &M); + // lowerOperand - Convert a MachineOperand into the equivalent MCOperand. + bool lowerOperand(const MachineOperand &MO, MCOperand &MCOp); + private: // Helpers for EmitStartOfAsmFile() and EmitEndOfAsmFile() void emitAttributes(); @@ -84,6 +89,10 @@ private: void EmitUnwindingInstruction(const MachineInstr *MI); + // emitPseudoExpansionLowering - tblgen'erated. + bool emitPseudoExpansionLowering(MCStreamer &OutStreamer, + const MachineInstr *MI); + public: void PrintDebugValueComment(const MachineInstr *MI, raw_ostream &OS); @@ -100,6 +109,7 @@ public: llvm::ARM::DW_ISA_ARM_thumb : llvm::ARM::DW_ISA_ARM_arm; } + MCOperand GetSymbolRef(const MachineOperand &MO, const MCSymbol *Symbol); MCSymbol *GetARMSetPICJumpTableLabel2(unsigned uid, unsigned uid2, const MachineBasicBlock *MBB) const; MCSymbol *GetARMJTIPICJumpTableLabel2(unsigned uid, unsigned uid2) const; @@ -107,7 +117,7 @@ public: MCSymbol *GetARMSJLJEHLabel(void) const; MCSymbol *GetARMGVSymbol(const GlobalValue *GV); - + /// EmitMachineConstantPoolValue - Print a machine constantpool value to /// the .s file. virtual void EmitMachineConstantPoolValue(MachineConstantPoolValue *MCPV); diff --git a/lib/Target/ARM/ARMBaseInfo.h b/lib/Target/ARM/ARMBaseInfo.h index 36edbad..458f7dd 100644 --- a/lib/Target/ARM/ARMBaseInfo.h +++ b/lib/Target/ARM/ARMBaseInfo.h @@ -17,20 +17,12 @@ #ifndef ARMBASEINFO_H #define ARMBASEINFO_H +#include "MCTargetDesc/ARMMCTargetDesc.h" #include "llvm/Support/ErrorHandling.h" // Note that the following auto-generated files only defined enum types, and // so are safe to include here. -// Defines symbolic names for ARM registers. This defines a mapping from -// register name to register number. -// -#include "ARMGenRegisterNames.inc" - -// Defines symbolic names for the ARM instructions. -// -#include "ARMGenInstrNames.inc" - namespace llvm { // Enums corresponding to ARM condition codes diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp index 44a3976..649bd7d 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.cpp +++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp @@ -18,7 +18,6 @@ #include "ARMHazardRecognizer.h" #include "ARMMachineFunctionInfo.h" #include "ARMRegisterInfo.h" -#include "ARMGenInstrInfo.inc" #include "llvm/Constants.h" #include "llvm/Function.h" #include "llvm/GlobalValue.h" @@ -31,10 +30,15 @@ #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/MC/MCAsmInfo.h" +#include "llvm/Support/BranchProbability.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/ADT/STLExtras.h" + +#define GET_INSTRINFO_CTOR +#include "ARMGenInstrInfo.inc" + using namespace llvm; static cl::opt<bool> @@ -74,7 +78,7 @@ static const ARM_MLxEntry ARM_MLxTable[] = { }; ARMBaseInstrInfo::ARMBaseInstrInfo(const ARMSubtarget& STI) - : TargetInstrInfoImpl(ARMInsts, array_lengthof(ARMInsts)), + : ARMGenInstrInfo(ARM::ADJCALLSTACKDOWN, ARM::ADJCALLSTACKUP), Subtarget(STI) { for (unsigned i = 0, e = array_lengthof(ARM_MLxTable); i != e; ++i) { if (!MLxEntryMap.insert(std::make_pair(ARM_MLxTable[i].MLxOpc, i)).second) @@ -136,9 +140,9 @@ ARMBaseInstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI, MachineInstr *UpdateMI = NULL; MachineInstr *MemMI = NULL; unsigned AddrMode = (TSFlags & ARMII::AddrModeMask); - const TargetInstrDesc &TID = MI->getDesc(); - unsigned NumOps = TID.getNumOperands(); - bool isLoad = !TID.mayStore(); + const MCInstrDesc &MCID = MI->getDesc(); + unsigned NumOps = MCID.getNumOperands(); + bool isLoad = !MCID.mayStore(); const MachineOperand &WB = isLoad ? MI->getOperand(1) : MI->getOperand(0); const MachineOperand &Base = MI->getOperand(2); const MachineOperand &Offset = MI->getOperand(NumOps-3); @@ -475,8 +479,8 @@ SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1, bool ARMBaseInstrInfo::DefinesPredicate(MachineInstr *MI, std::vector<MachineOperand> &Pred) const { // FIXME: This confuses implicit_def with optional CPSR def. - const TargetInstrDesc &TID = MI->getDesc(); - if (!TID.getImplicitDefs() && !TID.hasOptionalDef()) + const MCInstrDesc &MCID = MI->getDesc(); + if (!MCID.getImplicitDefs() && !MCID.hasOptionalDef()) return false; bool Found = false; @@ -495,11 +499,11 @@ bool ARMBaseInstrInfo::DefinesPredicate(MachineInstr *MI, /// By default, this returns true for every instruction with a /// PredicateOperand. bool ARMBaseInstrInfo::isPredicable(MachineInstr *MI) const { - const TargetInstrDesc &TID = MI->getDesc(); - if (!TID.isPredicable()) + const MCInstrDesc &MCID = MI->getDesc(); + if (!MCID.isPredicable()) return false; - if ((TID.TSFlags & ARMII::DomainMask) == ARMII::DomainNEON) { + if ((MCID.TSFlags & ARMII::DomainMask) == ARMII::DomainNEON) { ARMFunctionInfo *AFI = MI->getParent()->getParent()->getInfo<ARMFunctionInfo>(); return AFI->isThumb2Function(); @@ -524,35 +528,23 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const { const MachineFunction *MF = MBB.getParent(); const MCAsmInfo *MAI = MF->getTarget().getMCAsmInfo(); - // Basic size info comes from the TSFlags field. - const TargetInstrDesc &TID = MI->getDesc(); - uint64_t TSFlags = TID.TSFlags; + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.getSize()) + return MCID.getSize(); - unsigned Opc = MI->getOpcode(); - switch ((TSFlags & ARMII::SizeMask) >> ARMII::SizeShift) { - default: { // If this machine instr is an inline asm, measure it. if (MI->getOpcode() == ARM::INLINEASM) return getInlineAsmLength(MI->getOperand(0).getSymbolName(), *MAI); if (MI->isLabel()) return 0; + unsigned Opc = MI->getOpcode(); switch (Opc) { - default: - llvm_unreachable("Unknown or unset size field for instr!"); case TargetOpcode::IMPLICIT_DEF: case TargetOpcode::KILL: case TargetOpcode::PROLOG_LABEL: case TargetOpcode::EH_LABEL: case TargetOpcode::DBG_VALUE: return 0; - } - break; - } - case ARMII::Size8Bytes: return 8; // ARM instruction x 2. - case ARMII::Size4Bytes: return 4; // ARM / Thumb2 instruction. - case ARMII::Size2Bytes: return 2; // Thumb1 instruction. - case ARMII::SizeSpecial: { - switch (Opc) { case ARM::MOVi16_ga_pcrel: case ARM::MOVTi16_ga_pcrel: case ARM::t2MOVi16_ga_pcrel: @@ -588,9 +580,9 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const { // entry is one byte; TBH two byte each. unsigned EntrySize = (Opc == ARM::t2TBB_JT) ? 1 : ((Opc == ARM::t2TBH_JT) ? 2 : 4); - unsigned NumOps = TID.getNumOperands(); + unsigned NumOps = MCID.getNumOperands(); MachineOperand JTOP = - MI->getOperand(NumOps - (TID.isPredicable() ? 3 : 2)); + MI->getOperand(NumOps - (MCID.isPredicable() ? 3 : 2)); unsigned JTI = JTOP.getIndex(); const MachineJumpTableInfo *MJTI = MF->getJumpTableInfo(); assert(MJTI != 0); @@ -616,8 +608,6 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const { // Otherwise, pseudo-instruction sizes are zero. return 0; } - } - } return 0; // Not reached } @@ -647,7 +637,7 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB, else if (ARM::DPRRegClass.contains(DestReg, SrcReg)) Opc = ARM::VMOVD; else if (ARM::QPRRegClass.contains(DestReg, SrcReg)) - Opc = ARM::VMOVQ; + Opc = ARM::VORRq; else if (ARM::QQPRRegClass.contains(DestReg, SrcReg)) Opc = ARM::VMOVQQ; else if (ARM::QQQQPRRegClass.contains(DestReg, SrcReg)) @@ -657,6 +647,8 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB, MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(Opc), DestReg); MIB.addReg(SrcReg, getKillRegState(KillSrc)); + if (Opc == ARM::VORRq) + MIB.addReg(SrcReg, getKillRegState(KillSrc)); if (Opc != ARM::VMOVQQ && Opc != ARM::VMOVQQQQ) AddDefaultPred(MIB); } @@ -788,7 +780,7 @@ ARMBaseInstrInfo::isStoreToStackSlot(const MachineInstr *MI, break; case ARM::STRi12: case ARM::t2STRi12: - case ARM::tSpill: + case ARM::tSTRspi: case ARM::VSTRD: case ARM::VSTRS: if (MI->getOperand(1).isFI() && @@ -923,7 +915,7 @@ ARMBaseInstrInfo::isLoadFromStackSlot(const MachineInstr *MI, break; case ARM::LDRi12: case ARM::t2LDRi12: - case ARM::tRestore: + case ARM::tLDRspi: case ARM::VLDRD: case ARM::VLDRS: if (MI->getOperand(1).isFI() && @@ -1269,20 +1261,20 @@ bool ARMBaseInstrInfo::isSchedulingBoundary(const MachineInstr *MI, return false; } -bool ARMBaseInstrInfo::isProfitableToIfCvt(MachineBasicBlock &MBB, - unsigned NumCycles, - unsigned ExtraPredCycles, - float Probability, - float Confidence) const { +bool ARMBaseInstrInfo:: +isProfitableToIfCvt(MachineBasicBlock &MBB, + unsigned NumCycles, unsigned ExtraPredCycles, + const BranchProbability &Probability) const { if (!NumCycles) return false; // Attempt to estimate the relative costs of predication versus branching. - float UnpredCost = Probability * NumCycles; - UnpredCost += 1.0; // The branch itself - UnpredCost += (1.0 - Confidence) * Subtarget.getMispredictionPenalty(); + unsigned UnpredCost = Probability.getNumerator() * NumCycles; + UnpredCost /= Probability.getDenominator(); + UnpredCost += 1; // The branch itself + UnpredCost += Subtarget.getMispredictionPenalty() / 10; - return (float)(NumCycles + ExtraPredCycles) < UnpredCost; + return (NumCycles + ExtraPredCycles) <= UnpredCost; } bool ARMBaseInstrInfo:: @@ -1290,16 +1282,23 @@ isProfitableToIfCvt(MachineBasicBlock &TMBB, unsigned TCycles, unsigned TExtra, MachineBasicBlock &FMBB, unsigned FCycles, unsigned FExtra, - float Probability, float Confidence) const { + const BranchProbability &Probability) const { if (!TCycles || !FCycles) return false; // Attempt to estimate the relative costs of predication versus branching. - float UnpredCost = Probability * TCycles + (1.0 - Probability) * FCycles; - UnpredCost += 1.0; // The branch itself - UnpredCost += (1.0 - Confidence) * Subtarget.getMispredictionPenalty(); - - return (float)(TCycles + FCycles + TExtra + FExtra) < UnpredCost; + unsigned TUnpredCost = Probability.getNumerator() * TCycles; + TUnpredCost /= Probability.getDenominator(); + + uint32_t Comp = Probability.getDenominator() - Probability.getNumerator(); + unsigned FUnpredCost = Comp * FCycles; + FUnpredCost /= Probability.getDenominator(); + + unsigned UnpredCost = TUnpredCost + FUnpredCost; + UnpredCost += 1; // The branch itself + UnpredCost += Subtarget.getMispredictionPenalty() / 10; + + return (TCycles + FCycles + TExtra + FExtra) <= UnpredCost; } /// getInstrPredicate - If instruction is predicated, returns its predicate @@ -1363,7 +1362,7 @@ bool llvm::rewriteARMFrameIndex(MachineInstr &MI, unsigned FrameRegIdx, unsigned FrameReg, int &Offset, const ARMBaseInstrInfo &TII) { unsigned Opcode = MI.getOpcode(); - const TargetInstrDesc &Desc = MI.getDesc(); + const MCInstrDesc &Desc = MI.getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); bool isSub = false; @@ -1803,7 +1802,7 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData, if (!ItinData || ItinData->isEmpty()) return 1; - const TargetInstrDesc &Desc = MI->getDesc(); + const MCInstrDesc &Desc = MI->getDesc(); unsigned Class = Desc.getSchedClass(); unsigned UOps = ItinData->Itineraries[Class].NumMicroOps; if (UOps) @@ -1906,10 +1905,10 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData, int ARMBaseInstrInfo::getVLDMDefCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefClass, unsigned DefIdx, unsigned DefAlign) const { - int RegNo = (int)(DefIdx+1) - DefTID.getNumOperands() + 1; + int RegNo = (int)(DefIdx+1) - DefMCID.getNumOperands() + 1; if (RegNo <= 0) // Def is the address writeback. return ItinData->getOperandCycle(DefClass, DefIdx); @@ -1924,7 +1923,7 @@ ARMBaseInstrInfo::getVLDMDefCycle(const InstrItineraryData *ItinData, DefCycle = RegNo; bool isSLoad = false; - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: break; case ARM::VLDMSIA: case ARM::VLDMSIA_UPD: @@ -1947,10 +1946,10 @@ ARMBaseInstrInfo::getVLDMDefCycle(const InstrItineraryData *ItinData, int ARMBaseInstrInfo::getLDMDefCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefClass, unsigned DefIdx, unsigned DefAlign) const { - int RegNo = (int)(DefIdx+1) - DefTID.getNumOperands() + 1; + int RegNo = (int)(DefIdx+1) - DefMCID.getNumOperands() + 1; if (RegNo <= 0) // Def is the address writeback. return ItinData->getOperandCycle(DefClass, DefIdx); @@ -1982,10 +1981,10 @@ ARMBaseInstrInfo::getLDMDefCycle(const InstrItineraryData *ItinData, int ARMBaseInstrInfo::getVSTMUseCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseClass, unsigned UseIdx, unsigned UseAlign) const { - int RegNo = (int)(UseIdx+1) - UseTID.getNumOperands() + 1; + int RegNo = (int)(UseIdx+1) - UseMCID.getNumOperands() + 1; if (RegNo <= 0) return ItinData->getOperandCycle(UseClass, UseIdx); @@ -1999,7 +1998,7 @@ ARMBaseInstrInfo::getVSTMUseCycle(const InstrItineraryData *ItinData, UseCycle = RegNo; bool isSStore = false; - switch (UseTID.getOpcode()) { + switch (UseMCID.getOpcode()) { default: break; case ARM::VSTMSIA: case ARM::VSTMSIA_UPD: @@ -2022,10 +2021,10 @@ ARMBaseInstrInfo::getVSTMUseCycle(const InstrItineraryData *ItinData, int ARMBaseInstrInfo::getSTMUseCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseClass, unsigned UseIdx, unsigned UseAlign) const { - int RegNo = (int)(UseIdx+1) - UseTID.getNumOperands() + 1; + int RegNo = (int)(UseIdx+1) - UseMCID.getNumOperands() + 1; if (RegNo <= 0) return ItinData->getOperandCycle(UseClass, UseIdx); @@ -2051,14 +2050,14 @@ ARMBaseInstrInfo::getSTMUseCycle(const InstrItineraryData *ItinData, int ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefIdx, unsigned DefAlign, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseIdx, unsigned UseAlign) const { - unsigned DefClass = DefTID.getSchedClass(); - unsigned UseClass = UseTID.getSchedClass(); + unsigned DefClass = DefMCID.getSchedClass(); + unsigned UseClass = UseMCID.getSchedClass(); - if (DefIdx < DefTID.getNumDefs() && UseIdx < UseTID.getNumOperands()) + if (DefIdx < DefMCID.getNumDefs() && UseIdx < UseMCID.getNumOperands()) return ItinData->getOperandLatency(DefClass, DefIdx, UseClass, UseIdx); // This may be a def / use of a variable_ops instruction, the operand @@ -2066,7 +2065,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, // figure it out. int DefCycle = -1; bool LdmBypass = false; - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: DefCycle = ItinData->getOperandCycle(DefClass, DefIdx); break; @@ -2077,7 +2076,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, case ARM::VLDMSIA: case ARM::VLDMSIA_UPD: case ARM::VLDMSDB_UPD: - DefCycle = getVLDMDefCycle(ItinData, DefTID, DefClass, DefIdx, DefAlign); + DefCycle = getVLDMDefCycle(ItinData, DefMCID, DefClass, DefIdx, DefAlign); break; case ARM::LDMIA_RET: @@ -2098,7 +2097,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, case ARM::t2LDMIA_UPD: case ARM::t2LDMDB_UPD: LdmBypass = 1; - DefCycle = getLDMDefCycle(ItinData, DefTID, DefClass, DefIdx, DefAlign); + DefCycle = getLDMDefCycle(ItinData, DefMCID, DefClass, DefIdx, DefAlign); break; } @@ -2107,7 +2106,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, DefCycle = 2; int UseCycle = -1; - switch (UseTID.getOpcode()) { + switch (UseMCID.getOpcode()) { default: UseCycle = ItinData->getOperandCycle(UseClass, UseIdx); break; @@ -2118,7 +2117,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, case ARM::VSTMSIA: case ARM::VSTMSIA_UPD: case ARM::VSTMSDB_UPD: - UseCycle = getVSTMUseCycle(ItinData, UseTID, UseClass, UseIdx, UseAlign); + UseCycle = getVSTMUseCycle(ItinData, UseMCID, UseClass, UseIdx, UseAlign); break; case ARM::STMIA: @@ -2137,7 +2136,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, case ARM::t2STMDB: case ARM::t2STMIA_UPD: case ARM::t2STMDB_UPD: - UseCycle = getSTMUseCycle(ItinData, UseTID, UseClass, UseIdx, UseAlign); + UseCycle = getSTMUseCycle(ItinData, UseMCID, UseClass, UseIdx, UseAlign); break; } @@ -2150,7 +2149,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, if (LdmBypass) { // It's a variable_ops instruction so we can't use DefIdx here. Just use // first def operand. - if (ItinData->hasPipelineForwarding(DefClass, DefTID.getNumOperands()-1, + if (ItinData->hasPipelineForwarding(DefClass, DefMCID.getNumOperands()-1, UseClass, UseIdx)) --UseCycle; } else if (ItinData->hasPipelineForwarding(DefClass, DefIdx, @@ -2170,11 +2169,11 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, DefMI->isRegSequence() || DefMI->isImplicitDef()) return 1; - const TargetInstrDesc &DefTID = DefMI->getDesc(); + const MCInstrDesc &DefMCID = DefMI->getDesc(); if (!ItinData || ItinData->isEmpty()) - return DefTID.mayLoad() ? 3 : 1; + return DefMCID.mayLoad() ? 3 : 1; - const TargetInstrDesc &UseTID = UseMI->getDesc(); + const MCInstrDesc &UseMCID = UseMI->getDesc(); const MachineOperand &DefMO = DefMI->getOperand(DefIdx); if (DefMO.getReg() == ARM::CPSR) { if (DefMI->getOpcode() == ARM::FMSTAT) { @@ -2183,7 +2182,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, } // CPSR set and branch can be paired in the same cycle. - if (UseTID.isBranch()) + if (UseMCID.isBranch()) return 0; } @@ -2191,14 +2190,14 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, ? (*DefMI->memoperands_begin())->getAlignment() : 0; unsigned UseAlign = UseMI->hasOneMemOperand() ? (*UseMI->memoperands_begin())->getAlignment() : 0; - int Latency = getOperandLatency(ItinData, DefTID, DefIdx, DefAlign, - UseTID, UseIdx, UseAlign); + int Latency = getOperandLatency(ItinData, DefMCID, DefIdx, DefAlign, + UseMCID, UseIdx, UseAlign); if (Latency > 1 && (Subtarget.isCortexA8() || Subtarget.isCortexA9())) { // FIXME: Shifter op hack: no shift (i.e. [r +/- r]) or [r + r << 2] // variants are one cycle cheaper. - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: break; case ARM::LDRrs: case ARM::LDRBrs: { @@ -2223,7 +2222,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, } if (DefAlign < 8 && Subtarget.isCortexA9()) - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: break; case ARM::VLD1q8: case ARM::VLD1q16: @@ -2327,37 +2326,37 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, if (!DefNode->isMachineOpcode()) return 1; - const TargetInstrDesc &DefTID = get(DefNode->getMachineOpcode()); + const MCInstrDesc &DefMCID = get(DefNode->getMachineOpcode()); - if (isZeroCost(DefTID.Opcode)) + if (isZeroCost(DefMCID.Opcode)) return 0; if (!ItinData || ItinData->isEmpty()) - return DefTID.mayLoad() ? 3 : 1; + return DefMCID.mayLoad() ? 3 : 1; if (!UseNode->isMachineOpcode()) { - int Latency = ItinData->getOperandCycle(DefTID.getSchedClass(), DefIdx); + int Latency = ItinData->getOperandCycle(DefMCID.getSchedClass(), DefIdx); if (Subtarget.isCortexA9()) return Latency <= 2 ? 1 : Latency - 1; else return Latency <= 3 ? 1 : Latency - 2; } - const TargetInstrDesc &UseTID = get(UseNode->getMachineOpcode()); + const MCInstrDesc &UseMCID = get(UseNode->getMachineOpcode()); const MachineSDNode *DefMN = dyn_cast<MachineSDNode>(DefNode); unsigned DefAlign = !DefMN->memoperands_empty() ? (*DefMN->memoperands_begin())->getAlignment() : 0; const MachineSDNode *UseMN = dyn_cast<MachineSDNode>(UseNode); unsigned UseAlign = !UseMN->memoperands_empty() ? (*UseMN->memoperands_begin())->getAlignment() : 0; - int Latency = getOperandLatency(ItinData, DefTID, DefIdx, DefAlign, - UseTID, UseIdx, UseAlign); + int Latency = getOperandLatency(ItinData, DefMCID, DefIdx, DefAlign, + UseMCID, UseIdx, UseAlign); if (Latency > 1 && (Subtarget.isCortexA8() || Subtarget.isCortexA9())) { // FIXME: Shifter op hack: no shift (i.e. [r +/- r]) or [r + r << 2] // variants are one cycle cheaper. - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: break; case ARM::LDRrs: case ARM::LDRBrs: { @@ -2384,7 +2383,7 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData, } if (DefAlign < 8 && Subtarget.isCortexA9()) - switch (DefTID.getOpcode()) { + switch (DefMCID.getOpcode()) { default: break; case ARM::VLD1q8Pseudo: case ARM::VLD1q16Pseudo: @@ -2503,10 +2502,10 @@ int ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData, if (!ItinData || ItinData->isEmpty()) return 1; - const TargetInstrDesc &TID = MI->getDesc(); - unsigned Class = TID.getSchedClass(); + const MCInstrDesc &MCID = MI->getDesc(); + unsigned Class = MCID.getSchedClass(); unsigned UOps = ItinData->Itineraries[Class].NumMicroOps; - if (PredCost && TID.hasImplicitDefOfPhysReg(ARM::CPSR)) + if (PredCost && MCID.hasImplicitDefOfPhysReg(ARM::CPSR)) // When predicated, CPSR is an additional source operand for CPSR updating // instructions, this apparently increases their latencies. *PredCost = 1; diff --git a/lib/Target/ARM/ARMBaseInstrInfo.h b/lib/Target/ARM/ARMBaseInstrInfo.h index 9a2faf8..507e897 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.h +++ b/lib/Target/ARM/ARMBaseInstrInfo.h @@ -20,6 +20,9 @@ #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/SmallSet.h" +#define GET_INSTRINFO_HEADER +#include "ARMGenInstrInfo.inc" + namespace llvm { class ARMSubtarget; class ARMBaseRegisterInfo; @@ -36,24 +39,16 @@ namespace ARMII { // This four-bit field describes the addressing mode used. AddrModeMask = 0x1f, // The AddrMode enums are declared in ARMBaseInfo.h - // Size* - Flags to keep track of the size of an instruction. - SizeShift = 5, - SizeMask = 7 << SizeShift, - SizeSpecial = 1, // 0 byte pseudo or special case. - Size8Bytes = 2, - Size4Bytes = 3, - Size2Bytes = 4, - // IndexMode - Unindex, pre-indexed, or post-indexed are valid for load // and store ops only. Generic "updating" flag is used for ld/st multiple. // The index mode enums are declared in ARMBaseInfo.h - IndexModeShift = 8, + IndexModeShift = 5, IndexModeMask = 3 << IndexModeShift, //===------------------------------------------------------------------===// // Instruction encoding formats. // - FormShift = 10, + FormShift = 7, FormMask = 0x3f << FormShift, // Pseudo instructions @@ -126,15 +121,15 @@ namespace ARMII { // UnaryDP - Indicates this is a unary data processing instruction, i.e. // it doesn't have a Rn operand. - UnaryDP = 1 << 16, + UnaryDP = 1 << 13, // Xform16Bit - Indicates this Thumb2 instruction may be transformed into // a 16-bit Thumb instruction if certain conditions are met. - Xform16Bit = 1 << 17, + Xform16Bit = 1 << 14, //===------------------------------------------------------------------===// // Code domain. - DomainShift = 18, + DomainShift = 15, DomainMask = 7 << DomainShift, DomainGeneral = 0 << DomainShift, DomainVFP = 1 << DomainShift, @@ -172,7 +167,7 @@ namespace ARMII { }; } -class ARMBaseInstrInfo : public TargetInstrInfoImpl { +class ARMBaseInstrInfo : public ARMGenInstrInfo { const ARMSubtarget &Subtarget; protected: @@ -291,8 +286,8 @@ public: int64_t &Offset1, int64_t &Offset2)const; /// shouldScheduleLoadsNear - This is a used by the pre-regalloc scheduler to - /// determine (in conjunction with areLoadsFromSameBasePtr) if two loads should - /// be scheduled togther. On some targets if two loads are loading from + /// determine (in conjunction with areLoadsFromSameBasePtr) if two loads + /// should be scheduled togther. On some targets if two loads are loading from /// addresses in the same cache line, it's better if they are scheduled /// together. This function takes two integers that represent the load offsets /// from the common base address. It returns true if it decides it's desirable @@ -308,18 +303,18 @@ public: virtual bool isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumCycles, unsigned ExtraPredCycles, - float Prob, float Confidence) const; + const BranchProbability &Probability) const; virtual bool isProfitableToIfCvt(MachineBasicBlock &TMBB, unsigned NumT, unsigned ExtraT, MachineBasicBlock &FMBB, unsigned NumF, unsigned ExtraF, - float Probability, float Confidence) const; + const BranchProbability &Probability) const; virtual bool isProfitableToDupForIfCvt(MachineBasicBlock &MBB, unsigned NumCycles, - float Probability, - float Confidence) const { + const BranchProbability + &Probability) const { return NumCycles == 1; } @@ -353,25 +348,25 @@ public: SDNode *UseNode, unsigned UseIdx) const; private: int getVLDMDefCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefClass, unsigned DefIdx, unsigned DefAlign) const; int getLDMDefCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefClass, unsigned DefIdx, unsigned DefAlign) const; int getVSTMUseCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseClass, unsigned UseIdx, unsigned UseAlign) const; int getSTMUseCycle(const InstrItineraryData *ItinData, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseClass, unsigned UseIdx, unsigned UseAlign) const; int getOperandLatency(const InstrItineraryData *ItinData, - const TargetInstrDesc &DefTID, + const MCInstrDesc &DefMCID, unsigned DefIdx, unsigned DefAlign, - const TargetInstrDesc &UseTID, + const MCInstrDesc &UseMCID, unsigned UseIdx, unsigned UseAlign) const; int getInstrLatency(const InstrItineraryData *ItinData, diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp index 4ab37f6..ba42295 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp +++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp @@ -40,6 +40,9 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/Support/CommandLine.h" +#define GET_REGINFO_TARGET_DESC +#include "ARMGenRegisterInfo.inc" + using namespace llvm; static cl::opt<bool> @@ -54,8 +57,7 @@ EnableBasePointer("arm-use-base-pointer", cl::Hidden, cl::init(true), ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii, const ARMSubtarget &sti) - : ARMGenRegisterInfo(ARM::ADJCALLSTACKDOWN, ARM::ADJCALLSTACKUP), - TII(tii), STI(sti), + : ARMGenRegisterInfo(), TII(tii), STI(sti), FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11), BasePtr(ARM::R6) { } @@ -100,6 +102,12 @@ getReservedRegs(const MachineFunction &MF) const { // Some targets reserve R9. if (STI.isR9Reserved()) Reserved.set(ARM::R9); + // Reserve D16-D31 if the subtarget doesn't support them. + if (!STI.hasVFP3() || STI.hasD16()) { + assert(ARM::D31 == ARM::D16 + 15); + for (unsigned i = 0; i != 16; ++i) + Reserved.set(ARM::D16 + i); + } return Reserved; } @@ -387,12 +395,12 @@ ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, } } -/// getAllocationOrder - Returns the register allocation order for a specified -/// register class in the form of a pair of TargetRegisterClass iterators. -std::pair<TargetRegisterClass::iterator,TargetRegisterClass::iterator> -ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC, - unsigned HintType, unsigned HintReg, - const MachineFunction &MF) const { +/// getRawAllocationOrder - Returns the register allocation order for a +/// specified register class with a target-dependent hint. +ArrayRef<unsigned> +ARMBaseRegisterInfo::getRawAllocationOrder(const TargetRegisterClass *RC, + unsigned HintType, unsigned HintReg, + const MachineFunction &MF) const { const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); // Alternative register allocation orders when favoring even / odd registers // of register pairs. @@ -469,70 +477,54 @@ ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC, // We only support even/odd hints for GPR and rGPR. if (RC != ARM::GPRRegisterClass && RC != ARM::rGPRRegisterClass) - return std::make_pair(RC->allocation_order_begin(MF), - RC->allocation_order_end(MF)); + return RC->getRawAllocationOrder(MF); if (HintType == ARMRI::RegPairEven) { if (isPhysicalRegister(HintReg) && getRegisterPairEven(HintReg, MF) == 0) // It's no longer possible to fulfill this hint. Return the default // allocation order. - return std::make_pair(RC->allocation_order_begin(MF), - RC->allocation_order_end(MF)); + return RC->getRawAllocationOrder(MF); if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) - return std::make_pair(GPREven1, - GPREven1 + (sizeof(GPREven1)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven1); else - return std::make_pair(GPREven4, - GPREven4 + (sizeof(GPREven4)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven4); } else if (FramePtr == ARM::R7) { if (!STI.isR9Reserved()) - return std::make_pair(GPREven2, - GPREven2 + (sizeof(GPREven2)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven2); else - return std::make_pair(GPREven5, - GPREven5 + (sizeof(GPREven5)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven5); } else { // FramePtr == ARM::R11 if (!STI.isR9Reserved()) - return std::make_pair(GPREven3, - GPREven3 + (sizeof(GPREven3)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven3); else - return std::make_pair(GPREven6, - GPREven6 + (sizeof(GPREven6)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPREven6); } } else if (HintType == ARMRI::RegPairOdd) { if (isPhysicalRegister(HintReg) && getRegisterPairOdd(HintReg, MF) == 0) // It's no longer possible to fulfill this hint. Return the default // allocation order. - return std::make_pair(RC->allocation_order_begin(MF), - RC->allocation_order_end(MF)); + return RC->getRawAllocationOrder(MF); if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) - return std::make_pair(GPROdd1, - GPROdd1 + (sizeof(GPROdd1)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd1); else - return std::make_pair(GPROdd4, - GPROdd4 + (sizeof(GPROdd4)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd4); } else if (FramePtr == ARM::R7) { if (!STI.isR9Reserved()) - return std::make_pair(GPROdd2, - GPROdd2 + (sizeof(GPROdd2)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd2); else - return std::make_pair(GPROdd5, - GPROdd5 + (sizeof(GPROdd5)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd5); } else { // FramePtr == ARM::R11 if (!STI.isR9Reserved()) - return std::make_pair(GPROdd3, - GPROdd3 + (sizeof(GPROdd3)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd3); else - return std::make_pair(GPROdd6, - GPROdd6 + (sizeof(GPROdd6)/sizeof(unsigned))); + return ArrayRef<unsigned>(GPROdd6); } } - return std::make_pair(RC->allocation_order_begin(MF), - RC->allocation_order_end(MF)); + return RC->getRawAllocationOrder(MF); } /// ResolveRegAllocHint - Resolves the specified register allocation hint @@ -965,7 +957,7 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB, int64_t ARMBaseRegisterInfo:: getFrameIndexInstrOffset(const MachineInstr *MI, int Idx) const { - const TargetInstrDesc &Desc = MI->getDesc(); + const MCInstrDesc &Desc = MI->getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); int64_t InstrOffs = 0;; int Scale = 1; @@ -1115,11 +1107,11 @@ materializeFrameBaseRegister(MachineBasicBlock *MBB, if (Ins != MBB->end()) DL = Ins->getDebugLoc(); - const TargetInstrDesc &TID = TII.get(ADDriOpc); + const MCInstrDesc &MCID = TII.get(ADDriOpc); MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); - MRI.constrainRegClass(BaseReg, TID.OpInfo[0].getRegClass(this)); + MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this)); - MachineInstrBuilder MIB = BuildMI(*MBB, Ins, DL, TID, BaseReg) + MachineInstrBuilder MIB = BuildMI(*MBB, Ins, DL, MCID, BaseReg) .addFrameIndex(FrameIdx).addImm(Offset); if (!AFI->isThumb1OnlyFunction()) @@ -1155,7 +1147,7 @@ ARMBaseRegisterInfo::resolveFrameIndex(MachineBasicBlock::iterator I, bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, int64_t Offset) const { - const TargetInstrDesc &Desc = MI->getDesc(); + const MCInstrDesc &Desc = MI->getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); unsigned i = 0; @@ -1291,11 +1283,5 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, } // Update the original instruction to use the scratch register. MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true); - if (MI.getOpcode() == ARM::t2ADDrSPi) - MI.setDesc(TII.get(ARM::t2ADDri)); - else if (MI.getOpcode() == ARM::t2SUBrSPi) - MI.setDesc(TII.get(ARM::t2SUBri)); } } - -#include "ARMGenRegisterInfo.inc" diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.h b/lib/Target/ARM/ARMBaseRegisterInfo.h index c60d75a..b4b4059 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.h +++ b/lib/Target/ARM/ARMBaseRegisterInfo.h @@ -16,7 +16,9 @@ #include "ARM.h" #include "llvm/Target/TargetRegisterInfo.h" -#include "ARMGenRegisterInfo.h.inc" + +#define GET_REGINFO_HEADER +#include "ARMGenRegisterInfo.inc" namespace llvm { class ARMSubtarget; @@ -134,10 +136,9 @@ public: unsigned getRegPressureLimit(const TargetRegisterClass *RC, MachineFunction &MF) const; - std::pair<TargetRegisterClass::iterator,TargetRegisterClass::iterator> - getAllocationOrder(const TargetRegisterClass *RC, - unsigned HintType, unsigned HintReg, - const MachineFunction &MF) const; + ArrayRef<unsigned> getRawAllocationOrder(const TargetRegisterClass *RC, + unsigned HintType, unsigned HintReg, + const MachineFunction &MF) const; unsigned ResolveRegAllocHint(unsigned Type, unsigned Reg, const MachineFunction &MF) const; diff --git a/lib/Target/ARM/ARMCodeEmitter.cpp b/lib/Target/ARM/ARMCodeEmitter.cpp index 16d4ca5..d6fca62 100644 --- a/lib/Target/ARM/ARMCodeEmitter.cpp +++ b/lib/Target/ARM/ARMCodeEmitter.cpp @@ -96,13 +96,13 @@ namespace { void addPCLabel(unsigned LabelID); void emitPseudoInstruction(const MachineInstr &MI); unsigned getMachineSoRegOpValue(const MachineInstr &MI, - const TargetInstrDesc &TID, + const MCInstrDesc &MCID, const MachineOperand &MO, unsigned OpIdx); unsigned getMachineSoImmOpValue(unsigned SoImm); unsigned getAddrModeSBit(const MachineInstr &MI, - const TargetInstrDesc &TID) const; + const MCInstrDesc &MCID) const; void emitDataProcessingInstruction(const MachineInstr &MI, unsigned ImplicitRd = 0, @@ -443,9 +443,9 @@ unsigned ARMCodeEmitter::getMachineOpValue(const MachineInstr &MI, else if (MO.isSymbol()) emitExternalSymbolAddress(MO.getSymbolName(), ARM::reloc_arm_branch); else if (MO.isCPI()) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // For VFP load, the immediate offset is multiplied by 4. - unsigned Reloc = ((TID.TSFlags & ARMII::FormMask) == ARMII::VFPLdStFrm) + unsigned Reloc = ((MCID.TSFlags & ARMII::FormMask) == ARMII::VFPLdStFrm) ? ARM::reloc_arm_vfp_cp_entry : ARM::reloc_arm_cp_entry; emitConstPoolAddress(MO.getIndex(), Reloc); } else if (MO.isJTI()) @@ -757,7 +757,7 @@ void ARMCodeEmitter::emitMOVi2piecesInstruction(const MachineInstr &MI) { void ARMCodeEmitter::emitLEApcrelJTInstruction(const MachineInstr &MI) { // It's basically add r, pc, (LJTI - $+8) - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Emit the 'add' instruction. unsigned Binary = 0x4 << 21; // add: Insts{24-21} = 0b0100 @@ -766,7 +766,7 @@ void ARMCodeEmitter::emitLEApcrelJTInstruction(const MachineInstr &MI) { Binary |= II->getPredicate(&MI) << ARMII::CondShift; // Encode S bit if MI modifies CPSR. - Binary |= getAddrModeSBit(MI, TID); + Binary |= getAddrModeSBit(MI, MCID); // Encode Rd. Binary |= getMachineOpValue(MI, 0) << ARMII::RegRdShift; @@ -912,7 +912,7 @@ void ARMCodeEmitter::emitPseudoInstruction(const MachineInstr &MI) { } unsigned ARMCodeEmitter::getMachineSoRegOpValue(const MachineInstr &MI, - const TargetInstrDesc &TID, + const MCInstrDesc &MCID, const MachineOperand &MO, unsigned OpIdx) { unsigned Binary = getMachineOpValue(MI, MO); @@ -982,8 +982,8 @@ unsigned ARMCodeEmitter::getMachineSoImmOpValue(unsigned SoImm) { } unsigned ARMCodeEmitter::getAddrModeSBit(const MachineInstr &MI, - const TargetInstrDesc &TID) const { - for (unsigned i = MI.getNumOperands(), e = TID.getNumOperands(); i >= e; --i){ + const MCInstrDesc &MCID) const { + for (unsigned i = MI.getNumOperands(), e = MCID.getNumOperands(); i >= e; --i){ const MachineOperand &MO = MI.getOperand(i-1); if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) return 1 << ARMII::S_BitShift; @@ -994,7 +994,7 @@ unsigned ARMCodeEmitter::getAddrModeSBit(const MachineInstr &MI, void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, unsigned ImplicitRd, unsigned ImplicitRn) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1003,10 +1003,10 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, Binary |= II->getPredicate(&MI) << ARMII::CondShift; // Encode S bit if MI modifies CPSR. - Binary |= getAddrModeSBit(MI, TID); + Binary |= getAddrModeSBit(MI, MCID); // Encode register def if there is one. - unsigned NumDefs = TID.getNumDefs(); + unsigned NumDefs = MCID.getNumDefs(); unsigned OpIdx = 0; if (NumDefs) Binary |= getMachineOpValue(MI, OpIdx++) << ARMII::RegRdShift; @@ -1014,7 +1014,7 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, // Special handling for implicit use (e.g. PC). Binary |= (getARMRegisterNumbering(ImplicitRd) << ARMII::RegRdShift); - if (TID.Opcode == ARM::MOVi16) { + if (MCID.Opcode == ARM::MOVi16) { // Get immediate from MI. unsigned Lo16 = getMovi32Value(MI, MI.getOperand(OpIdx), ARM::reloc_arm_movw); @@ -1023,14 +1023,14 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, Binary |= ((Lo16 >> 12) & 0xF) << 16; emitWordLE(Binary); return; - } else if(TID.Opcode == ARM::MOVTi16) { + } else if(MCID.Opcode == ARM::MOVTi16) { unsigned Hi16 = (getMovi32Value(MI, MI.getOperand(OpIdx), ARM::reloc_arm_movt) >> 16); Binary |= Hi16 & 0xFFF; Binary |= ((Hi16 >> 12) & 0xF) << 16; emitWordLE(Binary); return; - } else if ((TID.Opcode == ARM::BFC) || (TID.Opcode == ARM::BFI)) { + } else if ((MCID.Opcode == ARM::BFC) || (MCID.Opcode == ARM::BFI)) { uint32_t v = ~MI.getOperand(2).getImm(); int32_t lsb = CountTrailingZeros_32(v); int32_t msb = (32 - CountLeadingZeros_32(v)) - 1; @@ -1039,7 +1039,7 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, Binary |= (lsb & 0x1F) << 7; emitWordLE(Binary); return; - } else if ((TID.Opcode == ARM::UBFX) || (TID.Opcode == ARM::SBFX)) { + } else if ((MCID.Opcode == ARM::UBFX) || (MCID.Opcode == ARM::SBFX)) { // Encode Rn in Instr{0-3} Binary |= getMachineOpValue(MI, OpIdx++); @@ -1054,11 +1054,11 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, } // If this is a two-address operand, skip it. e.g. MOVCCr operand 1. - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; // Encode first non-shifter register operand if there is one. - bool isUnary = TID.TSFlags & ARMII::UnaryDP; + bool isUnary = MCID.TSFlags & ARMII::UnaryDP; if (!isUnary) { if (ImplicitRn) // Special handling for implicit use (e.g. PC). @@ -1071,9 +1071,9 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, // Encode shifter operand. const MachineOperand &MO = MI.getOperand(OpIdx); - if ((TID.TSFlags & ARMII::FormMask) == ARMII::DPSoRegFrm) { + if ((MCID.TSFlags & ARMII::FormMask) == ARMII::DPSoRegFrm) { // Encode SoReg. - emitWordLE(Binary | getMachineSoRegOpValue(MI, TID, MO, OpIdx)); + emitWordLE(Binary | getMachineSoRegOpValue(MI, MCID, MO, OpIdx)); return; } @@ -1092,9 +1092,9 @@ void ARMCodeEmitter::emitDataProcessingInstruction(const MachineInstr &MI, void ARMCodeEmitter::emitLoadStoreInstruction(const MachineInstr &MI, unsigned ImplicitRd, unsigned ImplicitRn) { - const TargetInstrDesc &TID = MI.getDesc(); - unsigned Form = TID.TSFlags & ARMII::FormMask; - bool IsPrePost = (TID.TSFlags & ARMII::IndexModeMask) != 0; + const MCInstrDesc &MCID = MI.getDesc(); + unsigned Form = MCID.TSFlags & ARMII::FormMask; + bool IsPrePost = (MCID.TSFlags & ARMII::IndexModeMask) != 0; // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1134,7 +1134,7 @@ void ARMCodeEmitter::emitLoadStoreInstruction(const MachineInstr &MI, Binary |= getMachineOpValue(MI, OpIdx++) << ARMII::RegRnShift; // If this is a two-address operand, skip it. e.g. LDR_PRE. - if (!Skipped && TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (!Skipped && MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; const MachineOperand &MO2 = MI.getOperand(OpIdx); @@ -1170,9 +1170,9 @@ void ARMCodeEmitter::emitLoadStoreInstruction(const MachineInstr &MI, void ARMCodeEmitter::emitMiscLoadStoreInstruction(const MachineInstr &MI, unsigned ImplicitRn) { - const TargetInstrDesc &TID = MI.getDesc(); - unsigned Form = TID.TSFlags & ARMII::FormMask; - bool IsPrePost = (TID.TSFlags & ARMII::IndexModeMask) != 0; + const MCInstrDesc &MCID = MI.getDesc(); + unsigned Form = MCID.TSFlags & ARMII::FormMask; + bool IsPrePost = (MCID.TSFlags & ARMII::IndexModeMask) != 0; // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1194,7 +1194,7 @@ void ARMCodeEmitter::emitMiscLoadStoreInstruction(const MachineInstr &MI, Binary |= getMachineOpValue(MI, OpIdx++) << ARMII::RegRdShift; // Skip LDRD and STRD's second operand. - if (TID.Opcode == ARM::LDRD || TID.Opcode == ARM::STRD) + if (MCID.Opcode == ARM::LDRD || MCID.Opcode == ARM::STRD) ++OpIdx; // Set second operand @@ -1205,7 +1205,7 @@ void ARMCodeEmitter::emitMiscLoadStoreInstruction(const MachineInstr &MI, Binary |= getMachineOpValue(MI, OpIdx++) << ARMII::RegRnShift; // If this is a two-address operand, skip it. e.g. LDRH_POST. - if (!Skipped && TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (!Skipped && MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; const MachineOperand &MO2 = MI.getOperand(OpIdx); @@ -1255,8 +1255,8 @@ static unsigned getAddrModeUPBits(unsigned Mode) { } void ARMCodeEmitter::emitLoadStoreMultipleInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); - bool IsUpdating = (TID.TSFlags & ARMII::IndexModeMask) != 0; + const MCInstrDesc &MCID = MI.getDesc(); + bool IsUpdating = (MCID.TSFlags & ARMII::IndexModeMask) != 0; // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1295,7 +1295,7 @@ void ARMCodeEmitter::emitLoadStoreMultipleInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitMulFrmInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1304,12 +1304,12 @@ void ARMCodeEmitter::emitMulFrmInstruction(const MachineInstr &MI) { Binary |= II->getPredicate(&MI) << ARMII::CondShift; // Encode S bit if MI modifies CPSR. - Binary |= getAddrModeSBit(MI, TID); + Binary |= getAddrModeSBit(MI, MCID); // 32x32->64bit operations have two destination registers. The number // of register definitions will tell us if that's what we're dealing with. unsigned OpIdx = 0; - if (TID.getNumDefs() == 2) + if (MCID.getNumDefs() == 2) Binary |= getMachineOpValue (MI, OpIdx++) << ARMII::RegRdLoShift; // Encode Rd @@ -1323,16 +1323,16 @@ void ARMCodeEmitter::emitMulFrmInstruction(const MachineInstr &MI) { // Many multiple instructions (e.g. MLA) have three src operands. Encode // it as Rn (for multiply, that's in the same offset as RdLo. - if (TID.getNumOperands() > OpIdx && - !TID.OpInfo[OpIdx].isPredicate() && - !TID.OpInfo[OpIdx].isOptionalDef()) + if (MCID.getNumOperands() > OpIdx && + !MCID.OpInfo[OpIdx].isPredicate() && + !MCID.OpInfo[OpIdx].isOptionalDef()) Binary |= getMachineOpValue(MI, OpIdx) << ARMII::RegRdLoShift; emitWordLE(Binary); } void ARMCodeEmitter::emitExtendInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1361,15 +1361,15 @@ void ARMCodeEmitter::emitExtendInstruction(const MachineInstr &MI) { // Encode rot imm (0, 8, 16, or 24) if it has a rotate immediate operand. if (MI.getOperand(OpIdx).isImm() && - !TID.OpInfo[OpIdx].isPredicate() && - !TID.OpInfo[OpIdx].isOptionalDef()) + !MCID.OpInfo[OpIdx].isPredicate() && + !MCID.OpInfo[OpIdx].isOptionalDef()) Binary |= (getMachineOpValue(MI, OpIdx) / 8) << ARMII::ExtRotImmShift; emitWordLE(Binary); } void ARMCodeEmitter::emitMiscArithInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1378,7 +1378,7 @@ void ARMCodeEmitter::emitMiscArithInstruction(const MachineInstr &MI) { Binary |= II->getPredicate(&MI) << ARMII::CondShift; // PKH instructions are finished at this point - if (TID.Opcode == ARM::PKHBT || TID.Opcode == ARM::PKHTB) { + if (MCID.Opcode == ARM::PKHBT || MCID.Opcode == ARM::PKHTB) { emitWordLE(Binary); return; } @@ -1389,9 +1389,9 @@ void ARMCodeEmitter::emitMiscArithInstruction(const MachineInstr &MI) { Binary |= getMachineOpValue(MI, OpIdx++) << ARMII::RegRdShift; const MachineOperand &MO = MI.getOperand(OpIdx++); - if (OpIdx == TID.getNumOperands() || - TID.OpInfo[OpIdx].isPredicate() || - TID.OpInfo[OpIdx].isOptionalDef()) { + if (OpIdx == MCID.getNumOperands() || + MCID.OpInfo[OpIdx].isPredicate() || + MCID.OpInfo[OpIdx].isOptionalDef()) { // Encode Rm and it's done. Binary |= getMachineOpValue(MI, MO); emitWordLE(Binary); @@ -1406,7 +1406,7 @@ void ARMCodeEmitter::emitMiscArithInstruction(const MachineInstr &MI) { // Encode shift_imm. unsigned ShiftAmt = MI.getOperand(OpIdx).getImm(); - if (TID.Opcode == ARM::PKHTB) { + if (MCID.Opcode == ARM::PKHTB) { assert(ShiftAmt != 0 && "PKHTB shift_imm is 0!"); if (ShiftAmt == 32) ShiftAmt = 0; @@ -1418,7 +1418,7 @@ void ARMCodeEmitter::emitMiscArithInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitSaturateInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGen. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1431,11 +1431,11 @@ void ARMCodeEmitter::emitSaturateInstruction(const MachineInstr &MI) { // Encode saturate bit position. unsigned Pos = MI.getOperand(1).getImm(); - if (TID.Opcode == ARM::SSAT || TID.Opcode == ARM::SSAT16) + if (MCID.Opcode == ARM::SSAT || MCID.Opcode == ARM::SSAT16) Pos -= 1; assert((Pos < 16 || (Pos < 32 && - TID.Opcode != ARM::SSAT16 && - TID.Opcode != ARM::USAT16)) && + MCID.Opcode != ARM::SSAT16 && + MCID.Opcode != ARM::USAT16)) && "saturate bit position out of range"); Binary |= Pos << 16; @@ -1443,7 +1443,7 @@ void ARMCodeEmitter::emitSaturateInstruction(const MachineInstr &MI) { Binary |= getMachineOpValue(MI, 2); // Encode shift_imm. - if (TID.getNumOperands() == 4) { + if (MCID.getNumOperands() == 4) { unsigned ShiftOp = MI.getOperand(3).getImm(); ARM_AM::ShiftOpc Opc = ARM_AM::getSORegShOp(ShiftOp); if (Opc == ARM_AM::asr) @@ -1459,9 +1459,9 @@ void ARMCodeEmitter::emitSaturateInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitBranchInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); - if (TID.Opcode == ARM::TPsoft) { + if (MCID.Opcode == ARM::TPsoft) { llvm_unreachable("ARM::TPsoft FIXME"); // FIXME } @@ -1498,20 +1498,20 @@ void ARMCodeEmitter::emitInlineJumpTable(unsigned JTIndex) { } void ARMCodeEmitter::emitMiscBranchInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Handle jump tables. - if (TID.Opcode == ARM::BR_JTr || TID.Opcode == ARM::BR_JTadd) { + if (MCID.Opcode == ARM::BR_JTr || MCID.Opcode == ARM::BR_JTadd) { // First emit a ldr pc, [] instruction. emitDataProcessingInstruction(MI, ARM::PC); // Then emit the inline jump table. unsigned JTIndex = - (TID.Opcode == ARM::BR_JTr) + (MCID.Opcode == ARM::BR_JTr) ? MI.getOperand(1).getIndex() : MI.getOperand(2).getIndex(); emitInlineJumpTable(JTIndex); return; - } else if (TID.Opcode == ARM::BR_JTm) { + } else if (MCID.Opcode == ARM::BR_JTm) { // First emit a ldr pc, [] instruction. emitLoadStoreInstruction(MI, ARM::PC); @@ -1526,7 +1526,7 @@ void ARMCodeEmitter::emitMiscBranchInstruction(const MachineInstr &MI) { // Set the conditional execution predicate Binary |= II->getPredicate(&MI) << ARMII::CondShift; - if (TID.Opcode == ARM::BX_RET || TID.Opcode == ARM::MOVPCLR) + if (MCID.Opcode == ARM::BX_RET || MCID.Opcode == ARM::MOVPCLR) // The return register is LR. Binary |= getARMRegisterNumbering(ARM::LR); else @@ -1579,7 +1579,7 @@ static unsigned encodeVFPRm(const MachineInstr &MI, unsigned OpIdx) { } void ARMCodeEmitter::emitVFPArithInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1596,16 +1596,16 @@ void ARMCodeEmitter::emitVFPArithInstruction(const MachineInstr &MI) { Binary |= encodeVFPRd(MI, OpIdx++); // If this is a two-address operand, skip it, e.g. FMACD. - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; // Encode Dn / Sn. - if ((TID.TSFlags & ARMII::FormMask) == ARMII::VFPBinaryFrm) + if ((MCID.TSFlags & ARMII::FormMask) == ARMII::VFPBinaryFrm) Binary |= encodeVFPRn(MI, OpIdx++); - if (OpIdx == TID.getNumOperands() || - TID.OpInfo[OpIdx].isPredicate() || - TID.OpInfo[OpIdx].isOptionalDef()) { + if (OpIdx == MCID.getNumOperands() || + MCID.OpInfo[OpIdx].isPredicate() || + MCID.OpInfo[OpIdx].isOptionalDef()) { // FCMPEZD etc. has only one operand. emitWordLE(Binary); return; @@ -1618,8 +1618,8 @@ void ARMCodeEmitter::emitVFPArithInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitVFPConversionInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); - unsigned Form = TID.TSFlags & ARMII::FormMask; + const MCInstrDesc &MCID = MI.getDesc(); + unsigned Form = MCID.TSFlags & ARMII::FormMask; // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1709,8 +1709,8 @@ void ARMCodeEmitter::emitVFPLoadStoreInstruction(const MachineInstr &MI) { void ARMCodeEmitter::emitVFPLoadStoreMultipleInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); - bool IsUpdating = (TID.TSFlags & ARMII::IndexModeMask) != 0; + const MCInstrDesc &MCID = MI.getDesc(); + bool IsUpdating = (MCID.TSFlags & ARMII::IndexModeMask) != 0; // Part of binary is determined by TableGn. unsigned Binary = getBinaryCodeForInstr(MI); @@ -1795,8 +1795,8 @@ void ARMCodeEmitter::emitNEONLaneInstruction(const MachineInstr &MI) { unsigned Binary = getBinaryCodeForInstr(MI); unsigned RegTOpIdx, RegNOpIdx, LnOpIdx; - const TargetInstrDesc &TID = MI.getDesc(); - if ((TID.TSFlags & ARMII::FormMask) == ARMII::NGetLnFrm) { + const MCInstrDesc &MCID = MI.getDesc(); + if ((MCID.TSFlags & ARMII::FormMask) == ARMII::NGetLnFrm) { RegTOpIdx = 0; RegNOpIdx = 1; LnOpIdx = 2; @@ -1863,12 +1863,12 @@ void ARMCodeEmitter::emitNEON1RegModImmInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitNEON2RegInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); unsigned Binary = getBinaryCodeForInstr(MI); // Destination register is encoded in Dd; source register in Dm. unsigned OpIdx = 0; Binary |= encodeNEONRd(MI, OpIdx++); - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; Binary |= encodeNEONRm(MI, OpIdx); if (IsThumb) @@ -1878,15 +1878,15 @@ void ARMCodeEmitter::emitNEON2RegInstruction(const MachineInstr &MI) { } void ARMCodeEmitter::emitNEON3RegInstruction(const MachineInstr &MI) { - const TargetInstrDesc &TID = MI.getDesc(); + const MCInstrDesc &MCID = MI.getDesc(); unsigned Binary = getBinaryCodeForInstr(MI); // Destination register is encoded in Dd; source registers in Dn and Dm. unsigned OpIdx = 0; Binary |= encodeNEONRd(MI, OpIdx++); - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; Binary |= encodeNEONRn(MI, OpIdx++); - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) ++OpIdx; Binary |= encodeNEONRm(MI, OpIdx); if (IsThumb) diff --git a/lib/Target/ARM/ARMConstantIslandPass.cpp b/lib/Target/ARM/ARMConstantIslandPass.cpp index baf95a3..f45ebdc 100644 --- a/lib/Target/ARM/ARMConstantIslandPass.cpp +++ b/lib/Target/ARM/ARMConstantIslandPass.cpp @@ -1538,7 +1538,10 @@ bool ARMConstantIslands::UndoLRSpillRestore() { if (MI->getOpcode() == ARM::tPOP_RET && MI->getOperand(2).getReg() == ARM::PC && MI->getNumExplicitOperands() == 3) { - BuildMI(MI->getParent(), MI->getDebugLoc(), TII->get(ARM::tBX_RET)); + // Create the new insn and copy the predicate from the old. + BuildMI(MI->getParent(), MI->getDebugLoc(), TII->get(ARM::tBX_RET)) + .addOperand(MI->getOperand(0)) + .addOperand(MI->getOperand(1)); MI->eraseFromParent(); MadeChange = true; } @@ -1692,9 +1695,9 @@ bool ARMConstantIslands::OptimizeThumb2JumpTables(MachineFunction &MF) { const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables(); for (unsigned i = 0, e = T2JumpTables.size(); i != e; ++i) { MachineInstr *MI = T2JumpTables[i]; - const TargetInstrDesc &TID = MI->getDesc(); - unsigned NumOps = TID.getNumOperands(); - unsigned JTOpIdx = NumOps - (TID.isPredicable() ? 3 : 2); + const MCInstrDesc &MCID = MI->getDesc(); + unsigned NumOps = MCID.getNumOperands(); + unsigned JTOpIdx = NumOps - (MCID.isPredicable() ? 3 : 2); MachineOperand JTOP = MI->getOperand(JTOpIdx); unsigned JTI = JTOP.getIndex(); assert(JTI < JT.size()); @@ -1815,9 +1818,9 @@ bool ARMConstantIslands::ReorderThumb2JumpTables(MachineFunction &MF) { const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables(); for (unsigned i = 0, e = T2JumpTables.size(); i != e; ++i) { MachineInstr *MI = T2JumpTables[i]; - const TargetInstrDesc &TID = MI->getDesc(); - unsigned NumOps = TID.getNumOperands(); - unsigned JTOpIdx = NumOps - (TID.isPredicable() ? 3 : 2); + const MCInstrDesc &MCID = MI->getDesc(); + unsigned NumOps = MCID.getNumOperands(); + unsigned JTOpIdx = NumOps - (MCID.isPredicable() ? 3 : 2); MachineOperand JTOP = MI->getOperand(JTOpIdx); unsigned JTI = JTOP.getIndex(); assert(JTI < JT.size()); diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp index b6b3c75..94b72fd 100644 --- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp +++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp @@ -68,7 +68,7 @@ namespace { void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI, MachineInstrBuilder &UseMI, MachineInstrBuilder &DefMI) { - const TargetInstrDesc &Desc = OldMI.getDesc(); + const MCInstrDesc &Desc = OldMI.getDesc(); for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands(); i != e; ++i) { const MachineOperand &MO = OldMI.getOperand(i); @@ -727,8 +727,10 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, MI.eraseFromParent(); return true; } + case ARM::t2MOVCCr: case ARM::MOVCCr: { - BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVr), + unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVr : ARM::MOVr; + BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), MI.getOperand(1).getReg()) .addReg(MI.getOperand(2).getReg(), getKillRegState(MI.getOperand(2).isKill())) @@ -764,8 +766,10 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, MI.eraseFromParent(); return true; } + case ARM::t2MOVCCi: case ARM::MOVCCi: { - BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVi), + unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVi : ARM::MOVi; + BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), MI.getOperand(1).getReg()) .addImm(MI.getOperand(2).getImm()) .addImm(MI.getOperand(3).getImm()) // 'pred' @@ -837,8 +841,9 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, MI.getOperand(0).getReg()) .addOperand(MI.getOperand(1)) .addReg(0) - .addImm(ARM_AM::getSORegOpc((Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr - : ARM_AM::asr), 1))) + .addImm(ARM_AM::getSORegOpc((Opcode == ARM::MOVsrl_flag ? + ARM_AM::lsr : ARM_AM::asr), + 1))) .addReg(ARM::CPSR, RegState::Define); MI.eraseFromParent(); return true; @@ -856,10 +861,11 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, MI.eraseFromParent(); return true; } + case ARM::tTPsoft: case ARM::TPsoft: { MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), - TII->get(ARM::BL)) + TII->get(Opcode == ARM::tTPsoft ? ARM::tBL : ARM::BL)) .addExternalSymbol("__aeabi_read_tp", 0); MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); @@ -900,10 +906,10 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, const MachineOperand &MO1 = MI.getOperand(1); const GlobalValue *GV = MO1.getGlobal(); unsigned TF = MO1.getTargetFlags(); - bool isARM = (Opcode != ARM::t2MOV_ga_pcrel && Opcode != ARM::t2MOV_ga_dyn); + bool isARM = (Opcode != ARM::t2MOV_ga_pcrel && Opcode!=ARM::t2MOV_ga_dyn); bool isPIC = (Opcode != ARM::MOV_ga_dyn && Opcode != ARM::t2MOV_ga_dyn); unsigned LO16Opc = isARM ? ARM::MOVi16_ga_pcrel : ARM::t2MOVi16_ga_pcrel; - unsigned HI16Opc = isARM ? ARM::MOVTi16_ga_pcrel : ARM::t2MOVTi16_ga_pcrel; + unsigned HI16Opc = isARM ? ARM::MOVTi16_ga_pcrel :ARM::t2MOVTi16_ga_pcrel; unsigned LO16TF = isPIC ? ARMII::MO_LO16_NONLAZY_PIC : ARMII::MO_LO16_NONLAZY; unsigned HI16TF = isPIC @@ -958,15 +964,17 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, unsigned OddSrc = TRI->getSubReg(SrcReg, ARM::qsub_1); MachineInstrBuilder Even = AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), - TII->get(ARM::VMOVQ)) + TII->get(ARM::VORRq)) .addReg(EvenDst, RegState::Define | getDeadRegState(DstIsDead)) + .addReg(EvenSrc, getKillRegState(SrcIsKill)) .addReg(EvenSrc, getKillRegState(SrcIsKill))); MachineInstrBuilder Odd = AddDefaultPred(BuildMI(MBB, MBBI, MI.getDebugLoc(), - TII->get(ARM::VMOVQ)) + TII->get(ARM::VORRq)) .addReg(OddDst, RegState::Define | getDeadRegState(DstIsDead)) + .addReg(OddSrc, getKillRegState(SrcIsKill)) .addReg(OddSrc, getKillRegState(SrcIsKill))); TransferImpOps(MI, Even, Odd); MI.eraseFromParent(); diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp index 5cf73c4..f469d7e 100644 --- a/lib/Target/ARM/ARMFastISel.cpp +++ b/lib/Target/ARM/ARMFastISel.cpp @@ -219,8 +219,8 @@ class ARMFastISel : public FastISel { // we don't care about implicit defs here, just places we'll need to add a // default CCReg argument. Sets CPSR if we're setting CPSR instead of CCR. bool ARMFastISel::DefinesOptionalPredicate(MachineInstr *MI, bool *CPSR) { - const TargetInstrDesc &TID = MI->getDesc(); - if (!TID.hasOptionalDef()) + const MCInstrDesc &MCID = MI->getDesc(); + if (!MCID.hasOptionalDef()) return false; // Look to see if our OptionalDef is defining CPSR or CCR. @@ -234,15 +234,15 @@ bool ARMFastISel::DefinesOptionalPredicate(MachineInstr *MI, bool *CPSR) { } bool ARMFastISel::isARMNEONPred(const MachineInstr *MI) { - const TargetInstrDesc &TID = MI->getDesc(); + const MCInstrDesc &MCID = MI->getDesc(); // If we're a thumb2 or not NEON function we were handled via isPredicable. - if ((TID.TSFlags & ARMII::DomainMask) != ARMII::DomainNEON || + if ((MCID.TSFlags & ARMII::DomainMask) != ARMII::DomainNEON || AFI->isThumb2Function()) return false; - for (unsigned i = 0, e = TID.getNumOperands(); i != e; ++i) - if (TID.OpInfo[i].isPredicate()) + for (unsigned i = 0, e = MCID.getNumOperands(); i != e; ++i) + if (MCID.OpInfo[i].isPredicate()) return true; return false; @@ -278,7 +278,7 @@ ARMFastISel::AddOptionalDefs(const MachineInstrBuilder &MIB) { unsigned ARMFastISel::FastEmitInst_(unsigned MachineInstOpcode, const TargetRegisterClass* RC) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)); return ResultReg; @@ -288,7 +288,7 @@ unsigned ARMFastISel::FastEmitInst_r(unsigned MachineInstOpcode, const TargetRegisterClass *RC, unsigned Op0, bool Op0IsKill) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -308,7 +308,7 @@ unsigned ARMFastISel::FastEmitInst_rr(unsigned MachineInstOpcode, unsigned Op0, bool Op0IsKill, unsigned Op1, bool Op1IsKill) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -331,7 +331,7 @@ unsigned ARMFastISel::FastEmitInst_rrr(unsigned MachineInstOpcode, unsigned Op1, bool Op1IsKill, unsigned Op2, bool Op2IsKill) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -355,7 +355,7 @@ unsigned ARMFastISel::FastEmitInst_ri(unsigned MachineInstOpcode, unsigned Op0, bool Op0IsKill, uint64_t Imm) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -377,7 +377,7 @@ unsigned ARMFastISel::FastEmitInst_rf(unsigned MachineInstOpcode, unsigned Op0, bool Op0IsKill, const ConstantFP *FPImm) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -400,7 +400,7 @@ unsigned ARMFastISel::FastEmitInst_rri(unsigned MachineInstOpcode, unsigned Op1, bool Op1IsKill, uint64_t Imm) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -423,7 +423,7 @@ unsigned ARMFastISel::FastEmitInst_i(unsigned MachineInstOpcode, const TargetRegisterClass *RC, uint64_t Imm) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -442,7 +442,7 @@ unsigned ARMFastISel::FastEmitInst_ii(unsigned MachineInstOpcode, const TargetRegisterClass *RC, uint64_t Imm1, uint64_t Imm2) { unsigned ResultReg = createResultReg(RC); - const TargetInstrDesc &II = TII.get(MachineInstOpcode); + const MCInstrDesc &II = TII.get(MachineInstOpcode); if (II.getNumDefs() >= 1) AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg) @@ -1549,7 +1549,7 @@ bool ARMFastISel::ProcessCallArgs(SmallVectorImpl<Value*> &Args, NumBytes = CCInfo.getNextStackOffset(); // Issue CALLSEQ_START - unsigned AdjStackDown = TM.getRegisterInfo()->getCallFrameSetupOpcode(); + unsigned AdjStackDown = TII.getCallFrameSetupOpcode(); AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(AdjStackDown)) .addImm(NumBytes)); @@ -1647,7 +1647,7 @@ bool ARMFastISel::FinishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs, const Instruction *I, CallingConv::ID CC, unsigned &NumBytes) { // Issue CALLSEQ_END - unsigned AdjStackUp = TM.getRegisterInfo()->getCallFrameDestroyOpcode(); + unsigned AdjStackUp = TII.getCallFrameDestroyOpcode(); AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(AdjStackUp)) .addImm(NumBytes).addImm(0)); diff --git a/lib/Target/ARM/ARMFrameLowering.cpp b/lib/Target/ARM/ARMFrameLowering.cpp index e2e95d4..381b404 100644 --- a/lib/Target/ARM/ARMFrameLowering.cpp +++ b/lib/Target/ARM/ARMFrameLowering.cpp @@ -268,14 +268,14 @@ void ARMFrameLowering::emitPrologue(MachineFunction &MF) const { // bic r4, r4, MaxAlign // mov sp, r4 // FIXME: It will be better just to find spare register here. - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2tgpr), ARM::R4) - .addReg(ARM::SP, RegState::Kill); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), ARM::R4) + .addReg(ARM::SP, RegState::Kill)); AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::t2BICri), ARM::R4) .addReg(ARM::R4, RegState::Kill) .addImm(MaxAlign-1))); - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP) - .addReg(ARM::R4, RegState::Kill); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), ARM::SP) + .addReg(ARM::R4, RegState::Kill)); } AFI->setShouldRestoreSPFromFP(true); @@ -293,9 +293,9 @@ void ARMFrameLowering::emitPrologue(MachineFunction &MF) const { .addReg(ARM::SP) .addImm((unsigned)ARMCC::AL).addReg(0).addReg(0); else - BuildMI(MBB, MBBI, dl, - TII.get(ARM::tMOVgpr2gpr), RegInfo->getBaseRegister()) - .addReg(ARM::SP); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), + RegInfo->getBaseRegister()) + .addReg(ARM::SP)); } // If the frame has variable sized objects then the epilogue must restore @@ -364,8 +364,9 @@ void ARMFrameLowering::emitEpilogue(MachineFunction &MF, "No scratch register to restore SP from FP!"); emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::R4, FramePtr, -NumBytes, ARMCC::AL, 0, TII); - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), ARM::SP) - .addReg(ARM::R4); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), + ARM::SP) + .addReg(ARM::R4)); } } else { // Thumb2 or ARM. @@ -373,8 +374,9 @@ void ARMFrameLowering::emitEpilogue(MachineFunction &MF, BuildMI(MBB, MBBI, dl, TII.get(ARM::MOVr), ARM::SP) .addReg(FramePtr).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0); else - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), ARM::SP) - .addReg(FramePtr); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), + ARM::SP) + .addReg(FramePtr)); } } else if (NumBytes) emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes); @@ -427,6 +429,7 @@ void ARMFrameLowering::emitEpilogue(MachineFunction &MF, // Delete the pseudo instruction TCRETURN. MBB.erase(MBBI); + MBBI = NewMI; } if (VARegSaveSize) @@ -736,20 +739,52 @@ static unsigned GetFunctionSizeInBytes(const MachineFunction &MF, /// estimateStackSize - Estimate and return the size of the frame. /// FIXME: Make generic? static unsigned estimateStackSize(MachineFunction &MF) { - const MachineFrameInfo *FFI = MF.getFrameInfo(); + const MachineFrameInfo *MFI = MF.getFrameInfo(); + const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const TargetRegisterInfo *RegInfo = MF.getTarget().getRegisterInfo(); + unsigned MaxAlign = MFI->getMaxAlignment(); int Offset = 0; - for (int i = FFI->getObjectIndexBegin(); i != 0; ++i) { - int FixedOff = -FFI->getObjectOffset(i); + + // This code is very, very similar to PEI::calculateFrameObjectOffsets(). + // It really should be refactored to share code. Until then, changes + // should keep in mind that there's tight coupling between the two. + + for (int i = MFI->getObjectIndexBegin(); i != 0; ++i) { + int FixedOff = -MFI->getObjectOffset(i); if (FixedOff > Offset) Offset = FixedOff; } - for (unsigned i = 0, e = FFI->getObjectIndexEnd(); i != e; ++i) { - if (FFI->isDeadObjectIndex(i)) + for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) { + if (MFI->isDeadObjectIndex(i)) continue; - Offset += FFI->getObjectSize(i); - unsigned Align = FFI->getObjectAlignment(i); + Offset += MFI->getObjectSize(i); + unsigned Align = MFI->getObjectAlignment(i); // Adjust to alignment boundary Offset = (Offset+Align-1)/Align*Align; + + MaxAlign = std::max(Align, MaxAlign); } + + if (MFI->adjustsStack() && TFI->hasReservedCallFrame(MF)) + Offset += MFI->getMaxCallFrameSize(); + + // Round up the size to a multiple of the alignment. If the function has + // any calls or alloca's, align to the target's StackAlignment value to + // ensure that the callee's frame or the alloca data is suitably aligned; + // otherwise, for leaf functions, align to the TransientStackAlignment + // value. + unsigned StackAlign; + if (MFI->adjustsStack() || MFI->hasVarSizedObjects() || + (RegInfo->needsStackRealignment(MF) && MFI->getObjectIndexEnd() != 0)) + StackAlign = TFI->getStackAlignment(); + else + StackAlign = TFI->getTransientStackAlignment(); + + // If the frame pointer is eliminated, all frame offsets will be relative to + // SP not FP. Align to MaxAlign so this works. + StackAlign = std::max(StackAlign, MaxAlign); + unsigned AlignMask = StackAlign - 1; + Offset = (Offset + AlignMask) & ~uint64_t(AlignMask); + return (unsigned)Offset; } @@ -841,9 +876,14 @@ ARMFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF, if (AFI->getVarArgsRegSaveSize() > 0) MF.getRegInfo().setPhysRegUsed(ARM::LR); - // Spill R4 if Thumb1 epilogue has to restore SP from FP since + // Spill R4 if Thumb1 epilogue has to restore SP from FP. We don't know + // for sure what the stack size will be, but for this, an estimate is good + // enough. If there anything changes it, it'll be a spill, which implies + // we've used all the registers and so R4 is already used, so not marking + // it here will be OK. // FIXME: It will be better just to find spare register here. - if (MFI->hasVarSizedObjects()) + unsigned StackSize = estimateStackSize(MF); + if (MFI->hasVarSizedObjects() || StackSize > 508) MF.getRegInfo().setPhysRegUsed(ARM::R4); } diff --git a/lib/Target/ARM/ARMGlobalMerge.cpp b/lib/Target/ARM/ARMGlobalMerge.cpp index 3f02383..8d77b2d 100644 --- a/lib/Target/ARM/ARMGlobalMerge.cpp +++ b/lib/Target/ARM/ARMGlobalMerge.cpp @@ -128,10 +128,10 @@ bool ARMGlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals, for (size_t i = 0, e = Globals.size(); i != e; ) { size_t j = 0; uint64_t MergedSize = 0; - std::vector<const Type*> Tys; + std::vector<Type*> Tys; std::vector<Constant*> Inits; for (j = i; j != e; ++j) { - const Type *Ty = Globals[j]->getType()->getElementType(); + Type *Ty = Globals[j]->getType()->getElementType(); MergedSize += TD->getTypeAllocSize(Ty); if (MergedSize > MaxOffset) { break; @@ -175,7 +175,9 @@ bool ARMGlobalMerge::doInitialization(Module &M) { continue; // Ignore fancy-aligned globals for now. - if (I->getAlignment() != 0) + unsigned Alignment = I->getAlignment(); + const Type *Ty = I->getType()->getElementType(); + if (Alignment > TD->getABITypeAlignment(Ty)) continue; // Ignore all 'special' globals. @@ -183,7 +185,7 @@ bool ARMGlobalMerge::doInitialization(Module &M) { I->getName().startswith(".llvm.")) continue; - if (TD->getTypeAllocSize(I->getType()->getElementType()) < MaxOffset) { + if (TD->getTypeAllocSize(Ty) < MaxOffset) { const TargetLoweringObjectFile &TLOF = TLI->getObjFileLowering(); if (TLOF.getKindForGlobal(I, TLI->getTargetMachine()).isBSSLocal()) BSSGlobals.push_back(I); diff --git a/lib/Target/ARM/ARMHazardRecognizer.cpp b/lib/Target/ARM/ARMHazardRecognizer.cpp index 517bba8..787f6a2 100644 --- a/lib/Target/ARM/ARMHazardRecognizer.cpp +++ b/lib/Target/ARM/ARMHazardRecognizer.cpp @@ -19,11 +19,11 @@ using namespace llvm; static bool hasRAWHazard(MachineInstr *DefMI, MachineInstr *MI, const TargetRegisterInfo &TRI) { // FIXME: Detect integer instructions properly. - const TargetInstrDesc &TID = MI->getDesc(); - unsigned Domain = TID.TSFlags & ARMII::DomainMask; - if (TID.mayStore()) + const MCInstrDesc &MCID = MI->getDesc(); + unsigned Domain = MCID.TSFlags & ARMII::DomainMask; + if (MCID.mayStore()) return false; - unsigned Opcode = TID.getOpcode(); + unsigned Opcode = MCID.getOpcode(); if (Opcode == ARM::VMOVRS || Opcode == ARM::VMOVRRD) return false; if ((Domain & ARMII::DomainVFP) || (Domain & ARMII::DomainNEON)) @@ -43,15 +43,15 @@ ARMHazardRecognizer::getHazardType(SUnit *SU, int Stalls) { // Look for special VMLA / VMLS hazards. A VMUL / VADD / VSUB following // a VMLA / VMLS will cause 4 cycle stall. - const TargetInstrDesc &TID = MI->getDesc(); - if (LastMI && (TID.TSFlags & ARMII::DomainMask) != ARMII::DomainGeneral) { + const MCInstrDesc &MCID = MI->getDesc(); + if (LastMI && (MCID.TSFlags & ARMII::DomainMask) != ARMII::DomainGeneral) { MachineInstr *DefMI = LastMI; - const TargetInstrDesc &LastTID = LastMI->getDesc(); + const MCInstrDesc &LastMCID = LastMI->getDesc(); // Skip over one non-VFP / NEON instruction. - if (!LastTID.isBarrier() && + if (!LastMCID.isBarrier() && // On A9, AGU and NEON/FPU are muxed. - !(STI.isCortexA9() && (LastTID.mayLoad() || LastTID.mayStore())) && - (LastTID.TSFlags & ARMII::DomainMask) == ARMII::DomainGeneral) { + !(STI.isCortexA9() && (LastMCID.mayLoad() || LastMCID.mayStore())) && + (LastMCID.TSFlags & ARMII::DomainMask) == ARMII::DomainGeneral) { MachineBasicBlock::iterator I = LastMI; if (I != LastMI->getParent()->begin()) { I = llvm::prior(I); diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp index 9ad516d..2c9481b 100644 --- a/lib/Target/ARM/ARMISelDAGToDAG.cpp +++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp @@ -329,10 +329,10 @@ bool ARMDAGToDAGISel::hasNoVMLxHazardUse(SDNode *N) const { if (Use->getOpcode() == ISD::CopyToReg) return true; if (Use->isMachineOpcode()) { - const TargetInstrDesc &TID = TII->get(Use->getMachineOpcode()); - if (TID.mayStore()) + const MCInstrDesc &MCID = TII->get(Use->getMachineOpcode()); + if (MCID.mayStore()) return true; - unsigned Opcode = TID.getOpcode(); + unsigned Opcode = MCID.getOpcode(); if (Opcode == ARM::VMOVRS || Opcode == ARM::VMOVRRD) return true; // vmlx feeding into another vmlx. We actually want to unfold @@ -1354,30 +1354,34 @@ SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) { /// SDNode *ARMDAGToDAGISel::PairSRegs(EVT VT, SDValue V0, SDValue V1) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = + CurDAG->getTargetConstant(ARM::DPR_VFP2RegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 4); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5); } /// PairDRegs - Form a quad register from a pair of D registers. /// SDNode *ARMDAGToDAGISel::PairDRegs(EVT VT, SDValue V0, SDValue V1) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = CurDAG->getTargetConstant(ARM::QPRRegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 4); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5); } /// PairQRegs - Form 4 consecutive D registers from a pair of Q registers. /// SDNode *ARMDAGToDAGISel::PairQRegs(EVT VT, SDValue V0, SDValue V1) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 4); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5); } /// QuadSRegs - Form 4 consecutive S registers. @@ -1385,12 +1389,15 @@ SDNode *ARMDAGToDAGISel::PairQRegs(EVT VT, SDValue V0, SDValue V1) { SDNode *ARMDAGToDAGISel::QuadSRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = + CurDAG->getTargetConstant(ARM::QPR_VFP2RegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::ssub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::ssub_1, MVT::i32); SDValue SubReg2 = CurDAG->getTargetConstant(ARM::ssub_2, MVT::i32); SDValue SubReg3 = CurDAG->getTargetConstant(ARM::ssub_3, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1, V2, SubReg2, V3, SubReg3 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 8); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1, + V2, SubReg2, V3, SubReg3 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9); } /// QuadDRegs - Form 4 consecutive D registers. @@ -1398,12 +1405,14 @@ SDNode *ARMDAGToDAGISel::QuadSRegs(EVT VT, SDValue V0, SDValue V1, SDNode *ARMDAGToDAGISel::QuadDRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::dsub_1, MVT::i32); SDValue SubReg2 = CurDAG->getTargetConstant(ARM::dsub_2, MVT::i32); SDValue SubReg3 = CurDAG->getTargetConstant(ARM::dsub_3, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1, V2, SubReg2, V3, SubReg3 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 8); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1, + V2, SubReg2, V3, SubReg3 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9); } /// QuadQRegs - Form 4 consecutive Q registers. @@ -1411,12 +1420,14 @@ SDNode *ARMDAGToDAGISel::QuadDRegs(EVT VT, SDValue V0, SDValue V1, SDNode *ARMDAGToDAGISel::QuadQRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3) { DebugLoc dl = V0.getNode()->getDebugLoc(); + SDValue RegClass = CurDAG->getTargetConstant(ARM::QQQQPRRegClassID, MVT::i32); SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32); SDValue SubReg1 = CurDAG->getTargetConstant(ARM::qsub_1, MVT::i32); SDValue SubReg2 = CurDAG->getTargetConstant(ARM::qsub_2, MVT::i32); SDValue SubReg3 = CurDAG->getTargetConstant(ARM::qsub_3, MVT::i32); - const SDValue Ops[] = { V0, SubReg0, V1, SubReg1, V2, SubReg2, V3, SubReg3 }; - return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 8); + const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1, + V2, SubReg2, V3, SubReg3 }; + return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9); } /// GetVLDSTAlign - Get the alignment (in bytes) for the alignment operand diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index 339c858..cf8c5ba 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -506,6 +506,9 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setTargetDAGCombine(ISD::VECTOR_SHUFFLE); setTargetDAGCombine(ISD::INSERT_VECTOR_ELT); setTargetDAGCombine(ISD::STORE); + setTargetDAGCombine(ISD::FP_TO_SINT); + setTargetDAGCombine(ISD::FP_TO_UINT); + setTargetDAGCombine(ISD::FDIV); } computeRegisterProperties(); @@ -538,7 +541,8 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand); setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand); } - if (Subtarget->isThumb1Only() || !Subtarget->hasV6Ops()) + if (Subtarget->isThumb1Only() || !Subtarget->hasV6Ops() + || (Subtarget->isThumb2() && !Subtarget->hasThumb2DSP())) setOperationAction(ISD::MULHS, MVT::i32, Expand); setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom); @@ -704,6 +708,9 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::FPOW, MVT::f64, Expand); setOperationAction(ISD::FPOW, MVT::f32, Expand); + setOperationAction(ISD::FMA, MVT::f64, Expand); + setOperationAction(ISD::FMA, MVT::f32, Expand); + // Various VFP goodness if (!UseSoftFloat && !Subtarget->isThumb1Only()) { // int <-> fp are custom expanded into bit_convert + ARMISD ops. @@ -974,12 +981,12 @@ Sched::Preference ARMTargetLowering::getSchedulingPreference(SDNode *N) const { // Load are scheduled for latency even if there instruction itinerary // is not available. const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - const TargetInstrDesc &TID = TII->get(N->getMachineOpcode()); + const MCInstrDesc &MCID = TII->get(N->getMachineOpcode()); - if (TID.getNumDefs() == 0) + if (MCID.getNumDefs() == 0) return Sched::RegPressure; if (!Itins->isEmpty() && - Itins->getOperandCycle(TID.getSchedClass(), 0) > 2) + Itins->getOperandCycle(MCID.getSchedClass(), 0) > 2) return Sched::Latency; return Sched::RegPressure; @@ -1633,7 +1640,11 @@ ARMTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee, return false; // FIXME: Completely disable sibcall for Thumb1 since Thumb1RegisterInfo:: - // emitEpilogue is not ready for them. + // emitEpilogue is not ready for them. Thumb tail calls also use t2B, as + // the Thumb1 16-bit unconditional branch doesn't have sufficient relocation + // support in the assembler and linker to be used. This would need to be + // fixed to fully support tail calls in Thumb1. + // // Doing this is tricky, since the LDM/POP instruction on Thumb doesn't take // LR. This means if we need to reload LR, it takes an extra instructions, // which outweighs the value of the tail call; but here we don't know yet @@ -2281,12 +2292,13 @@ static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG, // ARMv7 with MP extension has PLDW. return Op.getOperand(0); - if (Subtarget->isThumb()) + unsigned isData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue(); + if (Subtarget->isThumb()) { // Invert the bits. isRead = ~isRead & 1; - unsigned isData = Subtarget->isThumb() ? 0 : 1; + isData = ~isData & 1; + } - // Currently there is no intrinsic that matches pli. return DAG.getNode(ARMISD::PRELOAD, dl, MVT::Other, Op.getOperand(0), Op.getOperand(1), DAG.getConstant(isRead, MVT::i32), DAG.getConstant(isData, MVT::i32)); @@ -2742,7 +2754,7 @@ SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { SDValue ARMcc; SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32); SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl); - return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,Cmp); + return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp); } ARMCC::CondCodes CondCode, CondCode2; @@ -5522,12 +5534,108 @@ SDValue combineSelectAndUse(SDNode *N, SDValue Slct, SDValue OtherOp, return SDValue(); } +// AddCombineToVPADDL- For pair-wise add on neon, use the vpaddl instruction +// (only after legalization). +static SDValue AddCombineToVPADDL(SDNode *N, SDValue N0, SDValue N1, + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { + + // Only perform optimization if after legalize, and if NEON is available. We + // also expected both operands to be BUILD_VECTORs. + if (DCI.isBeforeLegalize() || !Subtarget->hasNEON() + || N0.getOpcode() != ISD::BUILD_VECTOR + || N1.getOpcode() != ISD::BUILD_VECTOR) + return SDValue(); + + // Check output type since VPADDL operand elements can only be 8, 16, or 32. + EVT VT = N->getValueType(0); + if (!VT.isInteger() || VT.getVectorElementType() == MVT::i64) + return SDValue(); + + // Check that the vector operands are of the right form. + // N0 and N1 are BUILD_VECTOR nodes with N number of EXTRACT_VECTOR + // operands, where N is the size of the formed vector. + // Each EXTRACT_VECTOR should have the same input vector and odd or even + // index such that we have a pair wise add pattern. + + // Grab the vector that all EXTRACT_VECTOR nodes should be referencing. + if (N0->getOperand(0)->getOpcode() != ISD::EXTRACT_VECTOR_ELT) + return SDValue(); + SDValue Vec = N0->getOperand(0)->getOperand(0); + SDNode *V = Vec.getNode(); + unsigned nextIndex = 0; + + // For each operands to the ADD which are BUILD_VECTORs, + // check to see if each of their operands are an EXTRACT_VECTOR with + // the same vector and appropriate index. + for (unsigned i = 0, e = N0->getNumOperands(); i != e; ++i) { + if (N0->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT + && N1->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT) { + + SDValue ExtVec0 = N0->getOperand(i); + SDValue ExtVec1 = N1->getOperand(i); + + // First operand is the vector, verify its the same. + if (V != ExtVec0->getOperand(0).getNode() || + V != ExtVec1->getOperand(0).getNode()) + return SDValue(); + + // Second is the constant, verify its correct. + ConstantSDNode *C0 = dyn_cast<ConstantSDNode>(ExtVec0->getOperand(1)); + ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(ExtVec1->getOperand(1)); + + // For the constant, we want to see all the even or all the odd. + if (!C0 || !C1 || C0->getZExtValue() != nextIndex + || C1->getZExtValue() != nextIndex+1) + return SDValue(); + + // Increment index. + nextIndex+=2; + } else + return SDValue(); + } + + // Create VPADDL node. + SelectionDAG &DAG = DCI.DAG; + const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + + // Build operand list. + SmallVector<SDValue, 8> Ops; + Ops.push_back(DAG.getConstant(Intrinsic::arm_neon_vpaddls, + TLI.getPointerTy())); + + // Input is the vector. + Ops.push_back(Vec); + + // Get widened type and narrowed type. + MVT widenType; + unsigned numElem = VT.getVectorNumElements(); + switch (VT.getVectorElementType().getSimpleVT().SimpleTy) { + case MVT::i8: widenType = MVT::getVectorVT(MVT::i16, numElem); break; + case MVT::i16: widenType = MVT::getVectorVT(MVT::i32, numElem); break; + case MVT::i32: widenType = MVT::getVectorVT(MVT::i64, numElem); break; + default: + assert(0 && "Invalid vector element type for padd optimization."); + } + + SDValue tmp = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, N->getDebugLoc(), + widenType, &Ops[0], Ops.size()); + return DAG.getNode(ISD::TRUNCATE, N->getDebugLoc(), VT, tmp); +} + /// PerformADDCombineWithOperands - Try DAG combinations for an ADD with /// operands N0 and N1. This is a helper for PerformADDCombine that is /// called with the default operands, and if that fails, with commuted /// operands. static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1, - TargetLowering::DAGCombinerInfo &DCI) { + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget){ + + // Attempt to create vpaddl for this add. + SDValue Result = AddCombineToVPADDL(N, N0, N1, DCI, Subtarget); + if (Result.getNode()) + return Result; + // fold (add (select cc, 0, c), x) -> (select cc, x, (add, x, c)) if (N0.getOpcode() == ISD::SELECT && N0.getNode()->hasOneUse()) { SDValue Result = combineSelectAndUse(N, N0, N1, DCI); @@ -5539,17 +5647,18 @@ static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1, /// PerformADDCombine - Target-specific dag combine xforms for ISD::ADD. /// static SDValue PerformADDCombine(SDNode *N, - TargetLowering::DAGCombinerInfo &DCI) { + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); // First try with the default operand order. - SDValue Result = PerformADDCombineWithOperands(N, N0, N1, DCI); + SDValue Result = PerformADDCombineWithOperands(N, N0, N1, DCI, Subtarget); if (Result.getNode()) return Result; // If that didn't work, try again with the operands commuted. - return PerformADDCombineWithOperands(N, N1, N0, DCI); + return PerformADDCombineWithOperands(N, N1, N0, DCI, Subtarget); } /// PerformSUBCombine - Target-specific dag combine xforms for ISD::SUB. @@ -5588,7 +5697,7 @@ static SDValue PerformVMULCombine(SDNode *N, unsigned Opcode = N0.getOpcode(); if (Opcode != ISD::ADD && Opcode != ISD::SUB && Opcode != ISD::FADD && Opcode != ISD::FSUB) { - Opcode = N0.getOpcode(); + Opcode = N1.getOpcode(); if (Opcode != ISD::ADD && Opcode != ISD::SUB && Opcode != ISD::FADD && Opcode != ISD::FSUB) return SDValue(); @@ -5874,8 +5983,8 @@ static SDValue PerformORCombine(SDNode *N, return SDValue(); } -/// PerformBFICombine - (bfi A, (and B, C1), C2) -> (bfi A, B, C2) iff -/// C1 & C2 == C1. +/// PerformBFICombine - (bfi A, (and B, Mask1), Mask2) -> (bfi A, B, Mask2) iff +/// the bits being cleared by the AND are not demanded by the BFI. static SDValue PerformBFICombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) { SDValue N1 = N->getOperand(1); @@ -5883,9 +5992,12 @@ static SDValue PerformBFICombine(SDNode *N, ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1)); if (!N11C) return SDValue(); - unsigned Mask = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue(); + unsigned InvMask = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue(); + unsigned LSB = CountTrailingZeros_32(~InvMask); + unsigned Width = (32 - CountLeadingZeros_32(~InvMask)) - LSB; + unsigned Mask = (1 << Width)-1; unsigned Mask2 = N11C->getZExtValue(); - if ((Mask & Mask2) == Mask2) + if ((Mask & (~Mask2)) == 0) return DCI.DAG.getNode(ARMISD::BFI, N->getDebugLoc(), N->getValueType(0), N->getOperand(0), N1.getOperand(0), N->getOperand(2)); @@ -6378,7 +6490,105 @@ static SDValue PerformVDUPLANECombine(SDNode *N, return DCI.DAG.getNode(ISD::BITCAST, N->getDebugLoc(), VT, Op); } -/// getVShiftImm - Check if this is a valid build_vector for the immediate +// isConstVecPow2 - Return true if each vector element is a power of 2, all +// elements are the same constant, C, and Log2(C) ranges from 1 to 32. +static bool isConstVecPow2(SDValue ConstVec, bool isSigned, uint64_t &C) +{ + integerPart cN; + integerPart c0 = 0; + for (unsigned I = 0, E = ConstVec.getValueType().getVectorNumElements(); + I != E; I++) { + ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(ConstVec.getOperand(I)); + if (!C) + return false; + + bool isExact; + APFloat APF = C->getValueAPF(); + if (APF.convertToInteger(&cN, 64, isSigned, APFloat::rmTowardZero, &isExact) + != APFloat::opOK || !isExact) + return false; + + c0 = (I == 0) ? cN : c0; + if (!isPowerOf2_64(cN) || c0 != cN || Log2_64(c0) < 1 || Log2_64(c0) > 32) + return false; + } + C = c0; + return true; +} + +/// PerformVCVTCombine - VCVT (floating-point to fixed-point, Advanced SIMD) +/// can replace combinations of VMUL and VCVT (floating-point to integer) +/// when the VMUL has a constant operand that is a power of 2. +/// +/// Example (assume d17 = <float 8.000000e+00, float 8.000000e+00>): +/// vmul.f32 d16, d17, d16 +/// vcvt.s32.f32 d16, d16 +/// becomes: +/// vcvt.s32.f32 d16, d16, #3 +static SDValue PerformVCVTCombine(SDNode *N, + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { + SelectionDAG &DAG = DCI.DAG; + SDValue Op = N->getOperand(0); + + if (!Subtarget->hasNEON() || !Op.getValueType().isVector() || + Op.getOpcode() != ISD::FMUL) + return SDValue(); + + uint64_t C; + SDValue N0 = Op->getOperand(0); + SDValue ConstVec = Op->getOperand(1); + bool isSigned = N->getOpcode() == ISD::FP_TO_SINT; + + if (ConstVec.getOpcode() != ISD::BUILD_VECTOR || + !isConstVecPow2(ConstVec, isSigned, C)) + return SDValue(); + + unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfp2fxs : + Intrinsic::arm_neon_vcvtfp2fxu; + return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, N->getDebugLoc(), + N->getValueType(0), + DAG.getConstant(IntrinsicOpcode, MVT::i32), N0, + DAG.getConstant(Log2_64(C), MVT::i32)); +} + +/// PerformVDIVCombine - VCVT (fixed-point to floating-point, Advanced SIMD) +/// can replace combinations of VCVT (integer to floating-point) and VDIV +/// when the VDIV has a constant operand that is a power of 2. +/// +/// Example (assume d17 = <float 8.000000e+00, float 8.000000e+00>): +/// vcvt.f32.s32 d16, d16 +/// vdiv.f32 d16, d17, d16 +/// becomes: +/// vcvt.f32.s32 d16, d16, #3 +static SDValue PerformVDIVCombine(SDNode *N, + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { + SelectionDAG &DAG = DCI.DAG; + SDValue Op = N->getOperand(0); + unsigned OpOpcode = Op.getNode()->getOpcode(); + + if (!Subtarget->hasNEON() || !N->getValueType(0).isVector() || + (OpOpcode != ISD::SINT_TO_FP && OpOpcode != ISD::UINT_TO_FP)) + return SDValue(); + + uint64_t C; + SDValue ConstVec = N->getOperand(1); + bool isSigned = OpOpcode == ISD::SINT_TO_FP; + + if (ConstVec.getOpcode() != ISD::BUILD_VECTOR || + !isConstVecPow2(ConstVec, isSigned, C)) + return SDValue(); + + unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfxs2fp : + Intrinsic::arm_neon_vcvtfxu2fp; + return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, N->getDebugLoc(), + Op.getValueType(), + DAG.getConstant(IntrinsicOpcode, MVT::i32), + Op.getOperand(0), DAG.getConstant(Log2_64(C), MVT::i32)); +} + +/// Getvshiftimm - Check if this is a valid build_vector for the immediate /// operand of a vector shift operation, where all the elements of the /// build_vector must have the same constant integer value. static bool getVShiftImm(SDValue Op, unsigned ElementBits, int64_t &Cnt) { @@ -6750,11 +6960,75 @@ static SDValue PerformSELECT_CCCombine(SDNode *N, SelectionDAG &DAG, return DAG.getNode(Opcode, N->getDebugLoc(), N->getValueType(0), LHS, RHS); } +/// PerformCMOVCombine - Target-specific DAG combining for ARMISD::CMOV. +SDValue +ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const { + SDValue Cmp = N->getOperand(4); + if (Cmp.getOpcode() != ARMISD::CMPZ) + // Only looking at EQ and NE cases. + return SDValue(); + + EVT VT = N->getValueType(0); + DebugLoc dl = N->getDebugLoc(); + SDValue LHS = Cmp.getOperand(0); + SDValue RHS = Cmp.getOperand(1); + SDValue FalseVal = N->getOperand(0); + SDValue TrueVal = N->getOperand(1); + SDValue ARMcc = N->getOperand(2); + ARMCC::CondCodes CC = (ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue(); + + // Simplify + // mov r1, r0 + // cmp r1, x + // mov r0, y + // moveq r0, x + // to + // cmp r0, x + // movne r0, y + // + // mov r1, r0 + // cmp r1, x + // mov r0, x + // movne r0, y + // to + // cmp r0, x + // movne r0, y + /// FIXME: Turn this into a target neutral optimization? + SDValue Res; + if (CC == ARMCC::NE && FalseVal == RHS) { + Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, TrueVal, ARMcc, + N->getOperand(3), Cmp); + } else if (CC == ARMCC::EQ && TrueVal == RHS) { + SDValue ARMcc; + SDValue NewCmp = getARMCmp(LHS, RHS, ISD::SETNE, ARMcc, DAG, dl); + Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, FalseVal, ARMcc, + N->getOperand(3), NewCmp); + } + + if (Res.getNode()) { + APInt KnownZero, KnownOne; + APInt Mask = APInt::getAllOnesValue(VT.getScalarType().getSizeInBits()); + DAG.ComputeMaskedBits(SDValue(N,0), Mask, KnownZero, KnownOne); + // Capture demanded bits information that would be otherwise lost. + if (KnownZero == 0xfffffffe) + Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res, + DAG.getValueType(MVT::i1)); + else if (KnownZero == 0xffffff00) + Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res, + DAG.getValueType(MVT::i8)); + else if (KnownZero == 0xffff0000) + Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res, + DAG.getValueType(MVT::i16)); + } + + return Res; +} + SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { switch (N->getOpcode()) { default: break; - case ISD::ADD: return PerformADDCombine(N, DCI); + case ISD::ADD: return PerformADDCombine(N, DCI, Subtarget); case ISD::SUB: return PerformSUBCombine(N, DCI); case ISD::MUL: return PerformMULCombine(N, DCI, Subtarget); case ISD::OR: return PerformORCombine(N, DCI, Subtarget); @@ -6767,6 +7041,9 @@ SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N, case ISD::INSERT_VECTOR_ELT: return PerformInsertEltCombine(N, DCI); case ISD::VECTOR_SHUFFLE: return PerformVECTOR_SHUFFLECombine(N, DCI.DAG); case ARMISD::VDUPLANE: return PerformVDUPLANECombine(N, DCI); + case ISD::FP_TO_SINT: + case ISD::FP_TO_UINT: return PerformVCVTCombine(N, DCI, Subtarget); + case ISD::FDIV: return PerformVDIVCombine(N, DCI, Subtarget); case ISD::INTRINSIC_WO_CHAIN: return PerformIntrinsicCombine(N, DCI.DAG); case ISD::SHL: case ISD::SRA: @@ -6775,6 +7052,7 @@ SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N, case ISD::ZERO_EXTEND: case ISD::ANY_EXTEND: return PerformExtendCombine(N, DCI.DAG, Subtarget); case ISD::SELECT_CC: return PerformSELECT_CCCombine(N, DCI.DAG, Subtarget); + case ARMISD::CMOV: return PerformCMOVCombine(N, DCI.DAG); case ARMISD::VLD2DUP: case ARMISD::VLD3DUP: case ARMISD::VLD4DUP: @@ -7277,10 +7555,17 @@ ARMTargetLowering::getConstraintType(const std::string &Constraint) const { default: break; case 'l': return C_RegisterClass; case 'w': return C_RegisterClass; + case 'h': return C_RegisterClass; + case 'x': return C_RegisterClass; + case 't': return C_RegisterClass; + case 'j': return C_Other; // Constant for movw. + } + } else if (Constraint.size() == 2) { + switch (Constraint[0]) { + default: break; + // All 'U+' constraints are addresses. + case 'U': return C_Memory; } - } else { - if (Constraint == "Uv") - return C_Memory; } return TargetLowering::getConstraintType(Constraint); } @@ -7319,26 +7604,43 @@ ARMTargetLowering::getSingleConstraintMatchWeight( return weight; } -std::pair<unsigned, const TargetRegisterClass*> +typedef std::pair<unsigned, const TargetRegisterClass*> RCPair; +RCPair ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const { if (Constraint.size() == 1) { // GCC ARM Constraint Letters switch (Constraint[0]) { - case 'l': + case 'l': // Low regs or general regs. if (Subtarget->isThumb()) - return std::make_pair(0U, ARM::tGPRRegisterClass); + return RCPair(0U, ARM::tGPRRegisterClass); else - return std::make_pair(0U, ARM::GPRRegisterClass); + return RCPair(0U, ARM::GPRRegisterClass); + case 'h': // High regs or no regs. + if (Subtarget->isThumb()) + return RCPair(0U, ARM::hGPRRegisterClass); + break; case 'r': - return std::make_pair(0U, ARM::GPRRegisterClass); + return RCPair(0U, ARM::GPRRegisterClass); case 'w': if (VT == MVT::f32) - return std::make_pair(0U, ARM::SPRRegisterClass); + return RCPair(0U, ARM::SPRRegisterClass); if (VT.getSizeInBits() == 64) - return std::make_pair(0U, ARM::DPRRegisterClass); + return RCPair(0U, ARM::DPRRegisterClass); if (VT.getSizeInBits() == 128) - return std::make_pair(0U, ARM::QPRRegisterClass); + return RCPair(0U, ARM::QPRRegisterClass); + break; + case 'x': + if (VT == MVT::f32) + return RCPair(0U, ARM::SPR_8RegisterClass); + if (VT.getSizeInBits() == 64) + return RCPair(0U, ARM::DPR_8RegisterClass); + if (VT.getSizeInBits() == 128) + return RCPair(0U, ARM::QPR_8RegisterClass); + break; + case 't': + if (VT == MVT::f32) + return RCPair(0U, ARM::SPRRegisterClass); break; } } @@ -7348,47 +7650,6 @@ ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT); } -std::vector<unsigned> ARMTargetLowering:: -getRegClassForInlineAsmConstraint(const std::string &Constraint, - EVT VT) const { - if (Constraint.size() != 1) - return std::vector<unsigned>(); - - switch (Constraint[0]) { // GCC ARM Constraint Letters - default: break; - case 'l': - return make_vector<unsigned>(ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R4, ARM::R5, ARM::R6, ARM::R7, - 0); - case 'r': - return make_vector<unsigned>(ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R4, ARM::R5, ARM::R6, ARM::R7, - ARM::R8, ARM::R9, ARM::R10, ARM::R11, - ARM::R12, ARM::LR, 0); - case 'w': - if (VT == MVT::f32) - return make_vector<unsigned>(ARM::S0, ARM::S1, ARM::S2, ARM::S3, - ARM::S4, ARM::S5, ARM::S6, ARM::S7, - ARM::S8, ARM::S9, ARM::S10, ARM::S11, - ARM::S12,ARM::S13,ARM::S14,ARM::S15, - ARM::S16,ARM::S17,ARM::S18,ARM::S19, - ARM::S20,ARM::S21,ARM::S22,ARM::S23, - ARM::S24,ARM::S25,ARM::S26,ARM::S27, - ARM::S28,ARM::S29,ARM::S30,ARM::S31, 0); - if (VT.getSizeInBits() == 64) - return make_vector<unsigned>(ARM::D0, ARM::D1, ARM::D2, ARM::D3, - ARM::D4, ARM::D5, ARM::D6, ARM::D7, - ARM::D8, ARM::D9, ARM::D10,ARM::D11, - ARM::D12,ARM::D13,ARM::D14,ARM::D15, 0); - if (VT.getSizeInBits() == 128) - return make_vector<unsigned>(ARM::Q0, ARM::Q1, ARM::Q2, ARM::Q3, - ARM::Q4, ARM::Q5, ARM::Q6, ARM::Q7, 0); - break; - } - - return std::vector<unsigned>(); -} - /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops /// vector. If it is invalid, don't add anything to Ops. void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op, @@ -7403,6 +7664,7 @@ void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op, char ConstraintLetter = Constraint[0]; switch (ConstraintLetter) { default: break; + case 'j': case 'I': case 'J': case 'K': case 'L': case 'M': case 'N': case 'O': ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op); @@ -7417,6 +7679,13 @@ void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op, return; switch (ConstraintLetter) { + case 'j': + // Constant suitable for movw, must be between 0 and + // 65535. + if (Subtarget->hasV6T2Ops()) + if (CVal >= 0 && CVal <= 65535) + break; + return; case 'I': if (Subtarget->isThumb1Only()) { // This must be a constant between 0 and 255, for ADD @@ -7685,7 +7954,7 @@ bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info, Info.ptrVal = I.getArgOperand(2); Info.offset = 0; Info.align = 8; - Info.vol = false; + Info.vol = true; Info.readMem = false; Info.writeMem = true; return true; @@ -7696,7 +7965,7 @@ bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info, Info.ptrVal = I.getArgOperand(0); Info.offset = 0; Info.align = 8; - Info.vol = false; + Info.vol = true; Info.readMem = true; Info.writeMem = false; return true; diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h index 21a9a3a..980fb40 100644 --- a/lib/Target/ARM/ARMISelLowering.h +++ b/lib/Target/ARM/ARMISelLowering.h @@ -244,6 +244,7 @@ namespace llvm { EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const; + SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const; virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const; bool isDesirableToTransformToIntegerOp(unsigned Opc, EVT VT) const; @@ -306,9 +307,6 @@ namespace llvm { std::pair<unsigned, const TargetRegisterClass*> getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const; - std::vector<unsigned> - getRegClassForInlineAsmConstraint(const std::string &Constraint, - EVT VT) const; /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops /// vector. If it is invalid, don't add anything to Ops. If hasMemory is diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td index 897d8a5..3ccf22f 100644 --- a/lib/Target/ARM/ARMInstrFormats.td +++ b/lib/Target/ARM/ARMInstrFormats.td @@ -107,16 +107,6 @@ def AddrModeT2_pc : AddrMode<14>; def AddrModeT2_i8s4 : AddrMode<15>; def AddrMode_i12 : AddrMode<16>; -// Instruction size. -class SizeFlagVal<bits<3> val> { - bits<3> Value = val; -} -def SizeInvalid : SizeFlagVal<0>; // Unset. -def SizeSpecial : SizeFlagVal<1>; // Pseudo or special. -def Size8Bytes : SizeFlagVal<2>; -def Size4Bytes : SizeFlagVal<3>; -def Size2Bytes : SizeFlagVal<4>; - // Load / store index mode. class IndexMode<bits<2> val> { bits<2> Value = val; @@ -236,13 +226,13 @@ def shr_imm64 : Operand<i32> { // ARM Instruction templates. // -class InstTemplate<AddrMode am, SizeFlagVal sz, IndexMode im, +class InstTemplate<AddrMode am, int sz, IndexMode im, Format f, Domain d, string cstr, InstrItinClass itin> : Instruction { let Namespace = "ARM"; AddrMode AM = am; - SizeFlagVal SZ = sz; + int Size = sz; IndexMode IM = im; bits<2> IndexModeBits = IM.Value; Format F = f; @@ -256,12 +246,11 @@ class InstTemplate<AddrMode am, SizeFlagVal sz, IndexMode im, // The layout of TSFlags should be kept in sync with ARMBaseInstrInfo.h. let TSFlags{4-0} = AM.Value; - let TSFlags{7-5} = SZ.Value; - let TSFlags{9-8} = IndexModeBits; - let TSFlags{15-10} = Form; - let TSFlags{16} = isUnaryDataProc; - let TSFlags{17} = canXformTo16Bit; - let TSFlags{20-18} = D.Value; + let TSFlags{6-5} = IndexModeBits; + let TSFlags{12-7} = Form; + let TSFlags{13} = isUnaryDataProc; + let TSFlags{14} = canXformTo16Bit; + let TSFlags{17-15} = D.Value; let Constraints = cstr; let Itinerary = itin; @@ -271,53 +260,70 @@ class Encoding { field bits<32> Inst; } -class InstARM<AddrMode am, SizeFlagVal sz, IndexMode im, +class InstARM<AddrMode am, int sz, IndexMode im, Format f, Domain d, string cstr, InstrItinClass itin> : InstTemplate<am, sz, im, f, d, cstr, itin>, Encoding; // This Encoding-less class is used by Thumb1 to specify the encoding bits later // on by adding flavors to specific instructions. -class InstThumb<AddrMode am, SizeFlagVal sz, IndexMode im, +class InstThumb<AddrMode am, int sz, IndexMode im, Format f, Domain d, string cstr, InstrItinClass itin> : InstTemplate<am, sz, im, f, d, cstr, itin>; class PseudoInst<dag oops, dag iops, InstrItinClass itin, list<dag> pattern> - // FIXME: This really should derive from InstTemplate instead, as pseudos - // don't need encoding information. TableGen doesn't like that - // currently. Need to figure out why and fix it. - : InstARM<AddrModeNone, SizeSpecial, IndexModeNone, Pseudo, GenericDomain, - "", itin> { + : InstTemplate<AddrModeNone, 0, IndexModeNone, Pseudo, + GenericDomain, "", itin> { let OutOperandList = oops; let InOperandList = iops; let Pattern = pattern; let isCodeGenOnly = 1; + let isPseudo = 1; } // PseudoInst that's ARM-mode only. -class ARMPseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin, +class ARMPseudoInst<dag oops, dag iops, int sz, InstrItinClass itin, list<dag> pattern> : PseudoInst<oops, iops, itin, pattern> { - let SZ = sz; + let Size = sz; list<Predicate> Predicates = [IsARM]; } // PseudoInst that's Thumb-mode only. -class tPseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin, +class tPseudoInst<dag oops, dag iops, int sz, InstrItinClass itin, list<dag> pattern> : PseudoInst<oops, iops, itin, pattern> { - let SZ = sz; + let Size = sz; list<Predicate> Predicates = [IsThumb]; } // PseudoInst that's Thumb2-mode only. -class t2PseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin, +class t2PseudoInst<dag oops, dag iops, int sz, InstrItinClass itin, list<dag> pattern> : PseudoInst<oops, iops, itin, pattern> { - let SZ = sz; + let Size = sz; list<Predicate> Predicates = [IsThumb2]; } + +class ARMPseudoExpand<dag oops, dag iops, int sz, + InstrItinClass itin, list<dag> pattern, + dag Result> + : ARMPseudoInst<oops, iops, sz, itin, pattern>, + PseudoInstExpansion<Result>; + +class tPseudoExpand<dag oops, dag iops, int sz, + InstrItinClass itin, list<dag> pattern, + dag Result> + : tPseudoInst<oops, iops, sz, itin, pattern>, + PseudoInstExpansion<Result>; + +class t2PseudoExpand<dag oops, dag iops, int sz, + InstrItinClass itin, list<dag> pattern, + dag Result> + : t2PseudoInst<oops, iops, sz, itin, pattern>, + PseudoInstExpansion<Result>; + // Almost all ARM instructions are predicable. -class I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class I<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> @@ -332,7 +338,7 @@ class I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, } // A few are not predicable -class InoP<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class InoP<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> @@ -348,7 +354,7 @@ class InoP<dag oops, dag iops, AddrMode am, SizeFlagVal sz, // Same as I except it can optionally modify CPSR. Note it's modeled as an input // operand since by default it's a zero register. It will become an implicit def // once it's "flipped". -class sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class sI<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> @@ -366,7 +372,7 @@ class sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, } // Special cases -class XI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class XI<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, im, f, GenericDomain, cstr, itin> { @@ -379,31 +385,31 @@ class XI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class AI<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, f, itin, opc, asm, "", pattern>; class AsI<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : sI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + : sI<oops, iops, AddrModeNone, 4, IndexModeNone, f, itin, opc, asm, "", pattern>; class AXI<dag oops, dag iops, Format f, InstrItinClass itin, string asm, list<dag> pattern> - : XI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + : XI<oops, iops, AddrModeNone, 4, IndexModeNone, f, itin, asm, "", pattern>; class AInoP<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : InoP<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + : InoP<oops, iops, AddrModeNone, 4, IndexModeNone, f, itin, opc, asm, "", pattern>; // Ctrl flow instructions class ABI<bits<4> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, BrFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, BrFrm, itin, opc, asm, "", pattern> { let Inst{27-24} = opcod; } class ABXI<bits<4> opcod, dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : XI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, BrFrm, itin, + : XI<oops, iops, AddrModeNone, 4, IndexModeNone, BrFrm, itin, asm, "", pattern> { let Inst{27-24} = opcod; } @@ -411,13 +417,13 @@ class ABXI<bits<4> opcod, dag oops, dag iops, InstrItinClass itin, // BR_JT instructions class JTI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : XI<oops, iops, AddrModeNone, SizeSpecial, IndexModeNone, BrMiscFrm, itin, + : XI<oops, iops, AddrModeNone, 0, IndexModeNone, BrMiscFrm, itin, asm, "", pattern>; // Atomic load/store instructions class AIldrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, LdStExFrm, itin, opc, asm, "", pattern> { bits<4> Rt; bits<4> Rn; @@ -430,7 +436,7 @@ class AIldrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin, } class AIstrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, LdStExFrm, itin, opc, asm, "", pattern> { bits<4> Rd; bits<4> Rt; @@ -460,21 +466,21 @@ class AIswp<bit b, dag oops, dag iops, string opc, list<dag> pattern> // addrmode1 instructions class AI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrMode1, Size4Bytes, IndexModeNone, f, itin, + : I<oops, iops, AddrMode1, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { let Inst{24-21} = opcod; let Inst{27-26} = 0b00; } class AsI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : sI<oops, iops, AddrMode1, Size4Bytes, IndexModeNone, f, itin, + : sI<oops, iops, AddrMode1, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { let Inst{24-21} = opcod; let Inst{27-26} = 0b00; } class AXI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin, string asm, list<dag> pattern> - : XI<oops, iops, AddrMode1, Size4Bytes, IndexModeNone, f, itin, + : XI<oops, iops, AddrMode1, 4, IndexModeNone, f, itin, asm, "", pattern> { let Inst{24-21} = opcod; let Inst{27-26} = 0b00; @@ -486,7 +492,7 @@ class AXI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin, class AI2ldst<bits<3> op, bit isLd, bit isByte, dag oops, dag iops, AddrMode am, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, am, Size4Bytes, IndexModeNone, f, itin, opc, asm, + : I<oops, iops, am, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { let Inst{27-25} = op; let Inst{24} = 1; // 24 == P @@ -499,7 +505,7 @@ class AI2ldst<bits<3> op, bit isLd, bit isByte, dag oops, dag iops, AddrMode am, class AI2ldstidx<bit isLd, bit isByte, bit isPre, dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode2, Size4Bytes, im, f, itin, + : I<oops, iops, AddrMode2, 4, im, f, itin, opc, asm, cstr, pattern> { bits<4> Rt; let Inst{27-26} = 0b01; @@ -547,7 +553,7 @@ class AI2stridxT<bit isByte, bit isPre, dag oops, dag iops, // addrmode3 instructions class AI3ld<bits<4> op, bit op20, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { bits<14> addr; bits<4> Rt; @@ -567,7 +573,7 @@ class AI3ld<bits<4> op, bit op20, dag oops, dag iops, Format f, class AI3ldstidx<bits<4> op, bit op20, bit isLd, bit isPre, dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, im, f, itin, + : I<oops, iops, AddrMode3, 4, im, f, itin, opc, asm, cstr, pattern> { bits<4> Rt; let Inst{27-25} = 0b000; @@ -583,7 +589,7 @@ class AI3ldstidx<bits<4> op, bit op20, bit isLd, bit isPre, dag oops, dag iops, class AI3ldstidxT<bits<4> op, bit op20, bit isLd, bit isPre, dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, im, f, itin, + : I<oops, iops, AddrMode3, 4, im, f, itin, opc, asm, cstr, pattern> { // {13} 1 == imm8, 0 == Rm // {12-9} Rn @@ -627,7 +633,7 @@ class AI3stridx<bits<4> op, bit isByte, bit isPre, dag oops, dag iops, // stores class AI3str<bits<4> op, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { bits<14> addr; bits<4> Rt; @@ -647,7 +653,7 @@ class AI3str<bits<4> op, dag oops, dag iops, Format f, InstrItinClass itin, // Pre-indexed stores class AI3sthpr<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModePre, f, itin, opc, asm, cstr, pattern> { let Inst{4} = 1; let Inst{5} = 1; // H bit @@ -660,7 +666,7 @@ class AI3sthpr<dag oops, dag iops, Format f, InstrItinClass itin, } class AI3stdpr<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModePre, f, itin, opc, asm, cstr, pattern> { let Inst{4} = 1; let Inst{5} = 1; // H bit @@ -675,7 +681,7 @@ class AI3stdpr<dag oops, dag iops, Format f, InstrItinClass itin, // Post-indexed stores class AI3sthpo<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModePost, f, itin, opc, asm, cstr,pattern> { // {13} 1 == imm8, 0 == Rm // {12-9} Rn @@ -701,7 +707,7 @@ class AI3sthpo<dag oops, dag iops, Format f, InstrItinClass itin, } class AI3stdpo<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin, + : I<oops, iops, AddrMode3, 4, IndexModePost, f, itin, opc, asm, cstr, pattern> { let Inst{4} = 1; let Inst{5} = 1; // H bit @@ -716,7 +722,7 @@ class AI3stdpo<dag oops, dag iops, Format f, InstrItinClass itin, // addrmode4 instructions class AXI4<dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, string asm, string cstr, list<dag> pattern> - : XI<oops, iops, AddrMode4, Size4Bytes, im, f, itin, asm, cstr, pattern> { + : XI<oops, iops, AddrMode4, 4, im, f, itin, asm, cstr, pattern> { bits<4> p; bits<16> regs; bits<4> Rn; @@ -730,7 +736,7 @@ class AXI4<dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin, // Unsigned multiply, multiply-accumulate instructions. class AMul1I<bits<7> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, MulFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, MulFrm, itin, opc, asm, "", pattern> { let Inst{7-4} = 0b1001; let Inst{20} = 0; // S bit @@ -738,7 +744,7 @@ class AMul1I<bits<7> opcod, dag oops, dag iops, InstrItinClass itin, } class AsMul1I<bits<7> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : sI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, MulFrm, itin, + : sI<oops, iops, AddrModeNone, 4, IndexModeNone, MulFrm, itin, opc, asm, "", pattern> { let Inst{7-4} = 0b1001; let Inst{27-21} = opcod; @@ -747,7 +753,7 @@ class AsMul1I<bits<7> opcod, dag oops, dag iops, InstrItinClass itin, // Most significant word multiply class AMul2I<bits<7> opcod, bits<4> opc7_4, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, MulFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, MulFrm, itin, opc, asm, "", pattern> { bits<4> Rd; bits<4> Rn; @@ -770,7 +776,7 @@ class AMul2Ia<bits<7> opcod, bits<4> opc7_4, dag oops, dag iops, // SMUL<x><y> / SMULW<y> / SMLA<x><y> / SMLAW<x><y> class AMulxyIbase<bits<7> opcod, bits<2> bit6_5, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, MulFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, MulFrm, itin, opc, asm, "", pattern> { bits<4> Rn; bits<4> Rm; @@ -809,7 +815,7 @@ class AMulxyI64<bits<7> opcod, bits<2> bit6_5, dag oops, dag iops, // Extend instructions. class AExtI<bits<8> opcod, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, ExtFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, ExtFrm, itin, opc, asm, "", pattern> { // All AExtI instructions have Rd and Rm register operands. bits<4> Rd; @@ -824,7 +830,7 @@ class AExtI<bits<8> opcod, dag oops, dag iops, InstrItinClass itin, // Misc Arithmetic instructions. class AMiscA1I<bits<8> opcod, bits<4> opc7_4, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, ArithMiscFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, ArithMiscFrm, itin, opc, asm, "", pattern> { bits<4> Rd; bits<4> Rm; @@ -839,7 +845,7 @@ class AMiscA1I<bits<8> opcod, bits<4> opc7_4, dag oops, dag iops, // PKH instructions class APKHI<bits<8> opcod, bit tb, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, ArithMiscFrm, itin, + : I<oops, iops, AddrModeNone, 4, IndexModeNone, ArithMiscFrm, itin, opc, asm, "", pattern> { bits<4> Rd; bits<4> Rn; @@ -874,7 +880,7 @@ class ARMV6Pat<dag pattern, dag result> : Pat<pattern, result> { // Thumb Instruction Format Definitions. // -class ThumbI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class ThumbI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { let OutOperandList = oops; @@ -886,39 +892,32 @@ class ThumbI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, // TI - Thumb instruction. class TI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : ThumbI<oops, iops, AddrModeNone, Size2Bytes, itin, asm, "", pattern>; + : ThumbI<oops, iops, AddrModeNone, 2, itin, asm, "", pattern>; // Two-address instructions class TIt<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : ThumbI<oops, iops, AddrModeNone, Size2Bytes, itin, asm, "$lhs = $dst", + : ThumbI<oops, iops, AddrModeNone, 2, itin, asm, "$lhs = $dst", pattern>; // tBL, tBX 32-bit instructions class TIx2<bits<5> opcod1, bits<2> opcod2, bit opcod3, dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : ThumbI<oops, iops, AddrModeNone, Size4Bytes, itin, asm, "", pattern>, + : ThumbI<oops, iops, AddrModeNone, 4, itin, asm, "", pattern>, Encoding { let Inst{31-27} = opcod1; let Inst{15-14} = opcod2; let Inst{12} = opcod3; } -// Move to/from coprocessor instructions -class T1Cop<dag oops, dag iops, string asm, list<dag> pattern> - : ThumbI<oops, iops, AddrModeNone, Size4Bytes, NoItinerary, asm, "", pattern>, - Encoding, Requires<[IsThumb, HasV6]> { - let Inst{31-28} = 0b1110; -} - // BR_JT instructions class TJTI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : ThumbI<oops, iops, AddrModeNone, SizeSpecial, itin, asm, "", pattern>; + : ThumbI<oops, iops, AddrModeNone, 0, itin, asm, "", pattern>; // Thumb1 only -class Thumb1I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb1I<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { let OutOperandList = oops; @@ -930,19 +929,19 @@ class Thumb1I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class T1I<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : Thumb1I<oops, iops, AddrModeNone, Size2Bytes, itin, asm, "", pattern>; + : Thumb1I<oops, iops, AddrModeNone, 2, itin, asm, "", pattern>; class T1Ix2<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : Thumb1I<oops, iops, AddrModeNone, Size4Bytes, itin, asm, "", pattern>; + : Thumb1I<oops, iops, AddrModeNone, 4, itin, asm, "", pattern>; // Two-address instructions class T1It<dag oops, dag iops, InstrItinClass itin, string asm, string cstr, list<dag> pattern> - : Thumb1I<oops, iops, AddrModeNone, Size2Bytes, itin, + : Thumb1I<oops, iops, AddrModeNone, 2, itin, asm, cstr, pattern>; // Thumb1 instruction that can either be predicated or set CPSR. -class Thumb1sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb1sI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -955,16 +954,16 @@ class Thumb1sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class T1sI<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1sI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm, "", pattern>; + : Thumb1sI<oops, iops, AddrModeNone, 2, itin, opc, asm, "", pattern>; // Two-address instructions class T1sIt<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1sI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm, + : Thumb1sI<oops, iops, AddrModeNone, 2, itin, opc, asm, "$Rn = $Rdn", pattern>; // Thumb1 instruction that can be predicated. -class Thumb1pI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb1pI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -977,17 +976,17 @@ class Thumb1pI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class T1pI<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1pI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm, "", pattern>; + : Thumb1pI<oops, iops, AddrModeNone, 2, itin, opc, asm, "", pattern>; // Two-address instructions class T1pIt<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1pI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm, + : Thumb1pI<oops, iops, AddrModeNone, 2, itin, opc, asm, "$Rn = $Rdn", pattern>; class T1pIs<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1pI<oops, iops, AddrModeT1_s, Size2Bytes, itin, opc, asm, "", pattern>; + : Thumb1pI<oops, iops, AddrModeT1_s, 2, itin, opc, asm, "", pattern>; class Encoding16 : Encoding { let Inst{31-16} = 0x0000; @@ -1036,7 +1035,7 @@ class T1BranchCond<bits<4> opcode> : Encoding16 { class T1pILdStEncode<bits<3> opcode, dag oops, dag iops, AddrMode am, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1pI<oops, iops, am, Size2Bytes, itin, opc, asm, "", pattern>, + : Thumb1pI<oops, iops, am, 2, itin, opc, asm, "", pattern>, T1LoadStore<0b0101, opcode> { bits<3> Rt; bits<8> addr; @@ -1047,7 +1046,7 @@ class T1pILdStEncode<bits<3> opcode, dag oops, dag iops, AddrMode am, class T1pILdStEncodeImm<bits<4> opA, bit opB, dag oops, dag iops, AddrMode am, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb1pI<oops, iops, am, Size2Bytes, itin, opc, asm, "", pattern>, + : Thumb1pI<oops, iops, am, 2, itin, opc, asm, "", pattern>, T1LoadStore<opA, {opB,?,?}> { bits<3> Rt; bits<8> addr; @@ -1063,7 +1062,7 @@ class T1Misc<bits<7> opcode> : Encoding16 { } // Thumb2I - Thumb2 instruction. Almost all Thumb2 instructions are predicable. -class Thumb2I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb2I<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -1080,7 +1079,7 @@ class Thumb2I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, // // FIXME: This uses unified syntax so {s} comes before {p}. We should make it // more consistent. -class Thumb2sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb2sI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -1095,7 +1094,7 @@ class Thumb2sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, } // Special cases -class Thumb2XI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class Thumb2XI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -1106,7 +1105,7 @@ class Thumb2XI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, list<Predicate> Predicates = [IsThumb2]; } -class ThumbXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class ThumbXI<dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { @@ -1119,22 +1118,22 @@ class ThumbXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class T2I<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeNone, Size4Bytes, itin, opc, asm, "", pattern>; + : Thumb2I<oops, iops, AddrModeNone, 4, itin, opc, asm, "", pattern>; class T2Ii12<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeT2_i12, Size4Bytes, itin, opc, asm, "",pattern>; + : Thumb2I<oops, iops, AddrModeT2_i12, 4, itin, opc, asm, "",pattern>; class T2Ii8<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeT2_i8, Size4Bytes, itin, opc, asm, "", pattern>; + : Thumb2I<oops, iops, AddrModeT2_i8, 4, itin, opc, asm, "", pattern>; class T2Iso<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeT2_so, Size4Bytes, itin, opc, asm, "", pattern>; + : Thumb2I<oops, iops, AddrModeT2_so, 4, itin, opc, asm, "", pattern>; class T2Ipc<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeT2_pc, Size4Bytes, itin, opc, asm, "", pattern>; + : Thumb2I<oops, iops, AddrModeT2_pc, 4, itin, opc, asm, "", pattern>; class T2Ii8s4<bit P, bit W, bit isLoad, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2I<oops, iops, AddrModeT2_i8s4, Size4Bytes, itin, opc, asm, "", + : Thumb2I<oops, iops, AddrModeT2_i8s4, 4, itin, opc, asm, "", pattern> { bits<4> Rt; bits<4> Rt2; @@ -1153,32 +1152,32 @@ class T2Ii8s4<bit P, bit W, bit isLoad, dag oops, dag iops, InstrItinClass itin, class T2sI<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : Thumb2sI<oops, iops, AddrModeNone, Size4Bytes, itin, opc, asm, "", pattern>; + : Thumb2sI<oops, iops, AddrModeNone, 4, itin, opc, asm, "", pattern>; class T2XI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : Thumb2XI<oops, iops, AddrModeNone, Size4Bytes, itin, asm, "", pattern>; + : Thumb2XI<oops, iops, AddrModeNone, 4, itin, asm, "", pattern>; class T2JTI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern> - : Thumb2XI<oops, iops, AddrModeNone, SizeSpecial, itin, asm, "", pattern>; + : Thumb2XI<oops, iops, AddrModeNone, 0, itin, asm, "", pattern>; // Move to/from coprocessor instructions -class T2Cop<dag oops, dag iops, string asm, list<dag> pattern> - : T2XI<oops, iops, NoItinerary, asm, pattern>, Requires<[IsThumb2, HasV6]> { - let Inst{31-28} = 0b1111; +class T2Cop<bits<4> opc, dag oops, dag iops, string asm, list<dag> pattern> + : T2XI <oops, iops, NoItinerary, asm, pattern>, Requires<[IsThumb2]> { + let Inst{31-28} = opc; } // Two-address instructions class T2XIt<dag oops, dag iops, InstrItinClass itin, string asm, string cstr, list<dag> pattern> - : Thumb2XI<oops, iops, AddrModeNone, Size4Bytes, itin, asm, cstr, pattern>; + : Thumb2XI<oops, iops, AddrModeNone, 4, itin, asm, cstr, pattern>; // T2Iidxldst - Thumb2 indexed load / store instructions. class T2Iidxldst<bit signed, bits<2> opcod, bit load, bit pre, dag oops, dag iops, AddrMode am, IndexMode im, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : InstARM<am, Size4Bytes, im, ThumbFrm, GenericDomain, cstr, itin> { + : InstARM<am, 4, im, ThumbFrm, GenericDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); let AsmString = !strconcat(opc, "${p}", asm); @@ -1232,7 +1231,7 @@ class T2Pat<dag pattern, dag result> : Pat<pattern, result> { // // Almost all VFP instructions are predicable. -class VFPI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class VFPI<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, im, f, VFPDomain, cstr, itin> { @@ -1247,7 +1246,7 @@ class VFPI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, } // Special cases -class VFPXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class VFPXI<dag oops, dag iops, AddrMode am, int sz, IndexMode im, Format f, InstrItinClass itin, string asm, string cstr, list<dag> pattern> : InstARM<am, sz, im, f, VFPDomain, cstr, itin> { @@ -1263,7 +1262,7 @@ class VFPXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, class VFPAI<dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : VFPI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + : VFPI<oops, iops, AddrModeNone, 4, IndexModeNone, f, itin, opc, asm, "", pattern> { let PostEncoderMethod = "VFPThumb2PostEncoder"; } @@ -1272,7 +1271,7 @@ class VFPAI<dag oops, dag iops, Format f, InstrItinClass itin, class ADI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : VFPI<oops, iops, AddrMode5, Size4Bytes, IndexModeNone, + : VFPI<oops, iops, AddrMode5, 4, IndexModeNone, VFPLdStFrm, itin, opc, asm, "", pattern> { // Instruction operands. bits<5> Dd; @@ -1298,7 +1297,7 @@ class ADI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops, class ASI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> - : VFPI<oops, iops, AddrMode5, Size4Bytes, IndexModeNone, + : VFPI<oops, iops, AddrMode5, 4, IndexModeNone, VFPLdStFrm, itin, opc, asm, "", pattern> { // Instruction operands. bits<5> Sd; @@ -1324,7 +1323,7 @@ class ASI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops, // VFP Load / store multiple pseudo instructions. class PseudoVFPLdStM<dag oops, dag iops, InstrItinClass itin, string cstr, list<dag> pattern> - : InstARM<AddrMode4, Size4Bytes, IndexModeNone, Pseudo, VFPNeonDomain, + : InstARM<AddrMode4, 4, IndexModeNone, Pseudo, VFPNeonDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); @@ -1335,7 +1334,7 @@ class PseudoVFPLdStM<dag oops, dag iops, InstrItinClass itin, string cstr, // Load / store multiple class AXDI4<dag oops, dag iops, IndexMode im, InstrItinClass itin, string asm, string cstr, list<dag> pattern> - : VFPXI<oops, iops, AddrMode4, Size4Bytes, im, + : VFPXI<oops, iops, AddrMode4, 4, im, VFPLdStMulFrm, itin, asm, cstr, pattern> { // Instruction operands. bits<4> Rn; @@ -1355,7 +1354,7 @@ class AXDI4<dag oops, dag iops, IndexMode im, InstrItinClass itin, class AXSI4<dag oops, dag iops, IndexMode im, InstrItinClass itin, string asm, string cstr, list<dag> pattern> - : VFPXI<oops, iops, AddrMode4, Size4Bytes, im, + : VFPXI<oops, iops, AddrMode4, 4, im, VFPLdStMulFrm, itin, asm, cstr, pattern> { // Instruction operands. bits<4> Rn; @@ -1569,7 +1568,7 @@ class AVConv5I<bits<8> opcod1, bits<4> opcod2, dag oops, dag iops, class NeonI<dag oops, dag iops, AddrMode am, IndexMode im, Format f, InstrItinClass itin, string opc, string dt, string asm, string cstr, list<dag> pattern> - : InstARM<am, Size4Bytes, im, f, NeonDomain, cstr, itin> { + : InstARM<am, 4, im, f, NeonDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); let AsmString = !strconcat(opc, "${p}", ".", dt, "\t", asm); @@ -1581,7 +1580,7 @@ class NeonI<dag oops, dag iops, AddrMode am, IndexMode im, Format f, class NeonXI<dag oops, dag iops, AddrMode am, IndexMode im, Format f, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern> - : InstARM<am, Size4Bytes, im, f, NeonDomain, cstr, itin> { + : InstARM<am, 4, im, f, NeonDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); let AsmString = !strconcat(opc, "${p}", "\t", asm); @@ -1621,7 +1620,7 @@ class NLdStLn<bit op23, bits<2> op21_20, bits<4> op11_8, bits<4> op7_4, } class PseudoNLdSt<dag oops, dag iops, InstrItinClass itin, string cstr> - : InstARM<AddrMode6, Size4Bytes, IndexModeNone, Pseudo, NeonDomain, cstr, + : InstARM<AddrMode6, 4, IndexModeNone, Pseudo, NeonDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); @@ -1630,7 +1629,7 @@ class PseudoNLdSt<dag oops, dag iops, InstrItinClass itin, string cstr> class PseudoNeonI<dag oops, dag iops, InstrItinClass itin, string cstr, list<dag> pattern> - : InstARM<AddrModeNone, Size4Bytes, IndexModeNone, Pseudo, NeonDomain, cstr, + : InstARM<AddrModeNone, 4, IndexModeNone, Pseudo, NeonDomain, cstr, itin> { let OutOperandList = oops; let InOperandList = !con(iops, (ins pred:$p)); @@ -1859,7 +1858,7 @@ class N3VX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, class NVLaneOp<bits<8> opcod1, bits<4> opcod2, bits<2> opcod3, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string dt, string asm, list<dag> pattern> - : InstARM<AddrModeNone, Size4Bytes, IndexModeNone, f, NeonDomain, + : InstARM<AddrModeNone, 4, IndexModeNone, f, NeonDomain, "", itin> { let Inst{27-20} = opcod1; let Inst{11-8} = opcod2; diff --git a/lib/Target/ARM/ARMInstrInfo.cpp b/lib/Target/ARM/ARMInstrInfo.cpp index 6f48d96..adcbf18 100644 --- a/lib/Target/ARM/ARMInstrInfo.cpp +++ b/lib/Target/ARM/ARMInstrInfo.cpp @@ -14,7 +14,6 @@ #include "ARMInstrInfo.h" #include "ARM.h" #include "ARMAddressingModes.h" -#include "ARMGenInstrInfo.inc" #include "ARMMachineFunctionInfo.h" #include "llvm/ADT/STLExtras.h" #include "llvm/CodeGen/LiveVariables.h" diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td index 9af76df..a42dd1a 100644 --- a/lib/Target/ARM/ARMInstrInfo.td +++ b/lib/Target/ARM/ARMInstrInfo.td @@ -62,6 +62,9 @@ def SDT_ARMEH_SJLJ_DispatchSetup: SDTypeProfile<0, 1, [SDTCisInt<0>]>; def SDT_ARMMEMBARRIER : SDTypeProfile<0, 1, [SDTCisInt<0>]>; +def SDT_ARMPREFETCH : SDTypeProfile<0, 3, [SDTCisPtrTy<0>, SDTCisSameAs<1, 2>, + SDTCisInt<1>]>; + def SDT_ARMTCRET : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>; def SDT_ARMBFI : SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisVT<1, i32>, @@ -130,7 +133,7 @@ def ARMMemBarrier : SDNode<"ARMISD::MEMBARRIER", SDT_ARMMEMBARRIER, [SDNPHasChain]>; def ARMMemBarrierMCR : SDNode<"ARMISD::MEMBARRIER_MCR", SDT_ARMMEMBARRIER, [SDNPHasChain]>; -def ARMPreload : SDNode<"ARMISD::PRELOAD", SDTPrefetch, +def ARMPreload : SDNode<"ARMISD::PRELOAD", SDT_ARMPREFETCH, [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>; def ARMrbit : SDNode<"ARMISD::RBIT", SDTIntUnaryOp>; @@ -144,33 +147,48 @@ def ARMbfi : SDNode<"ARMISD::BFI", SDT_ARMBFI>; //===----------------------------------------------------------------------===// // ARM Instruction Predicate Definitions. // -def HasV4T : Predicate<"Subtarget->hasV4TOps()">, AssemblerPredicate; +def HasV4T : Predicate<"Subtarget->hasV4TOps()">, + AssemblerPredicate<"HasV4TOps">; def NoV4T : Predicate<"!Subtarget->hasV4TOps()">; def HasV5T : Predicate<"Subtarget->hasV5TOps()">; -def HasV5TE : Predicate<"Subtarget->hasV5TEOps()">, AssemblerPredicate; -def HasV6 : Predicate<"Subtarget->hasV6Ops()">, AssemblerPredicate; +def HasV5TE : Predicate<"Subtarget->hasV5TEOps()">, + AssemblerPredicate<"HasV5TEOps">; +def HasV6 : Predicate<"Subtarget->hasV6Ops()">, + AssemblerPredicate<"HasV6Ops">; def NoV6 : Predicate<"!Subtarget->hasV6Ops()">; -def HasV6T2 : Predicate<"Subtarget->hasV6T2Ops()">, AssemblerPredicate; +def HasV6T2 : Predicate<"Subtarget->hasV6T2Ops()">, + AssemblerPredicate<"HasV6T2Ops">; def NoV6T2 : Predicate<"!Subtarget->hasV6T2Ops()">; -def HasV7 : Predicate<"Subtarget->hasV7Ops()">, AssemblerPredicate; +def HasV7 : Predicate<"Subtarget->hasV7Ops()">, + AssemblerPredicate<"HasV7Ops">; def NoVFP : Predicate<"!Subtarget->hasVFP2()">; -def HasVFP2 : Predicate<"Subtarget->hasVFP2()">, AssemblerPredicate; -def HasVFP3 : Predicate<"Subtarget->hasVFP3()">, AssemblerPredicate; -def HasNEON : Predicate<"Subtarget->hasNEON()">, AssemblerPredicate; -def HasFP16 : Predicate<"Subtarget->hasFP16()">, AssemblerPredicate; -def HasDivide : Predicate<"Subtarget->hasDivide()">, AssemblerPredicate; +def HasVFP2 : Predicate<"Subtarget->hasVFP2()">, + AssemblerPredicate<"FeatureVFP2">; +def HasVFP3 : Predicate<"Subtarget->hasVFP3()">, + AssemblerPredicate<"FeatureVFP3">; +def HasNEON : Predicate<"Subtarget->hasNEON()">, + AssemblerPredicate<"FeatureNEON">; +def HasFP16 : Predicate<"Subtarget->hasFP16()">, + AssemblerPredicate<"FeatureFP16">; +def HasDivide : Predicate<"Subtarget->hasDivide()">, + AssemblerPredicate<"FeatureHWDiv">; def HasT2ExtractPack : Predicate<"Subtarget->hasT2ExtractPack()">, - AssemblerPredicate; + AssemblerPredicate<"FeatureT2XtPk">; +def HasThumb2DSP : Predicate<"Subtarget->hasThumb2DSP()">, + AssemblerPredicate<"FeatureDSPThumb2">; def HasDB : Predicate<"Subtarget->hasDataBarrier()">, - AssemblerPredicate; + AssemblerPredicate<"FeatureDB">; def HasMP : Predicate<"Subtarget->hasMPExtension()">, - AssemblerPredicate; + AssemblerPredicate<"FeatureMP">; def UseNEONForFP : Predicate<"Subtarget->useNEONForSinglePrecisionFP()">; def DontUseNEONForFP : Predicate<"!Subtarget->useNEONForSinglePrecisionFP()">; -def IsThumb : Predicate<"Subtarget->isThumb()">, AssemblerPredicate; +def IsThumb : Predicate<"Subtarget->isThumb()">, + AssemblerPredicate<"ModeThumb">; def IsThumb1Only : Predicate<"Subtarget->isThumb1Only()">; -def IsThumb2 : Predicate<"Subtarget->isThumb2()">, AssemblerPredicate; -def IsARM : Predicate<"!Subtarget->isThumb()">, AssemblerPredicate; +def IsThumb2 : Predicate<"Subtarget->isThumb2()">, + AssemblerPredicate<"ModeThumb,FeatureThumb2">; +def IsARM : Predicate<"!Subtarget->isThumb()">, + AssemblerPredicate<"!ModeThumb">; def IsDarwin : Predicate<"Subtarget->isTargetDarwin()">; def IsNotDarwin : Predicate<"!Subtarget->isTargetDarwin()">; @@ -237,11 +255,13 @@ def lo16AllZero : PatLeaf<(i32 imm), [{ return (((uint32_t)N->getZExtValue()) & 0xFFFFUL) == 0; }], hi16>; -/// imm0_65535 predicate - True if the 32-bit immediate is in the range -/// [0.65535]. -def imm0_65535 : ImmLeaf<i32, [{ +/// imm0_65535 - An immediate is in the range [0.65535]. +def Imm0_65535AsmOperand: AsmOperandClass { let Name = "Imm0_65535"; } +def imm0_65535 : Operand<i32>, ImmLeaf<i32, [{ return Imm >= 0 && Imm < 65536; -}]>; +}]> { + let ParserMatchClass = Imm0_65535AsmOperand; +} class BinOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$RHS), res>; class UnOpFrag <dag res> : PatFrag<(ops node:$Src), res>; @@ -294,16 +314,19 @@ def fsub_mlx : PatFrag<(ops node:$lhs, node:$rhs),(fsub node:$lhs, node:$rhs),[{ // FIXME: rename brtarget to t2_brtarget def brtarget : Operand<OtherVT> { let EncoderMethod = "getBranchTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } // FIXME: get rid of this one? def uncondbrtarget : Operand<OtherVT> { let EncoderMethod = "getUnconditionalBranchTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } // Branch target for ARM. Handles conditional/unconditional def br_target : Operand<OtherVT> { let EncoderMethod = "getARMBranchTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } // Call target. @@ -311,6 +334,7 @@ def br_target : Operand<OtherVT> { def bltarget : Operand<i32> { // Encoded the same as branch targets. let EncoderMethod = "getBranchTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } // Call target for ARM. Handles conditional/unconditional @@ -318,6 +342,7 @@ def bltarget : Operand<i32> { def bl_target : Operand<i32> { // Encoded the same as branch targets. let EncoderMethod = "getARMBranchTargetOpValue"; + let OperandType = "OPERAND_PCREL"; } @@ -394,14 +419,20 @@ def shift_imm : Operand<i32> { let ParserMatchClass = ShifterAsmOperand; } +def ShiftedRegAsmOperand : AsmOperandClass { + let Name = "ShiftedReg"; +} + // shifter_operand operands: so_reg and so_imm. def so_reg : Operand<i32>, // reg reg imm ComplexPattern<i32, 3, "SelectShifterOperandReg", [shl,srl,sra,rotr]> { let EncoderMethod = "getSORegOpValue"; let PrintMethod = "printSORegOperand"; + let ParserMatchClass = ShiftedRegAsmOperand; let MIOperandInfo = (ops GPR, GPR, shift_imm); } +// FIXME: Does this need to be distinct from so_reg? def shift_so_reg : Operand<i32>, // reg reg imm ComplexPattern<i32, 3, "SelectShiftShifterOperandReg", [shl,srl,sra,rotr]> { @@ -416,7 +447,6 @@ def so_imm : Operand<i32>, ImmLeaf<i32, [{ return ARM_AM::getSOImmVal(Imm) != -1; }]> { let EncoderMethod = "getSOImmOpValue"; - let PrintMethod = "printSOImmOperand"; } // Break so_imm's up into two pieces. This handles immediates with up to 16 @@ -434,6 +464,22 @@ def arm_i32imm : PatLeaf<(imm), [{ return ARM_AM::isSOImmTwoPartVal((unsigned)N->getZExtValue()); }]>; +/// imm0_7 predicate - Immediate in the range [0,31]. +def Imm0_7AsmOperand: AsmOperandClass { let Name = "Imm0_7"; } +def imm0_7 : Operand<i32>, ImmLeaf<i32, [{ + return Imm >= 0 && Imm < 8; +}]> { + let ParserMatchClass = Imm0_7AsmOperand; +} + +/// imm0_15 predicate - Immediate in the range [0,31]. +def Imm0_15AsmOperand: AsmOperandClass { let Name = "Imm0_15"; } +def imm0_15 : Operand<i32>, ImmLeaf<i32, [{ + return Imm >= 0 && Imm < 16; +}]> { + let ParserMatchClass = Imm0_15AsmOperand; +} + /// imm0_31 predicate - True if the 32-bit immediate is in the range [0,31]. def imm0_31 : Operand<i32>, ImmLeaf<i32, [{ return Imm >= 0 && Imm < 32; @@ -673,7 +719,7 @@ include "ARMInstrFormats.td" /// binop that produces a value. multiclass AsI1_bin_irs<bits<4> opcod, string opc, InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, - PatFrag opnode, bit Commutable = 0> { + PatFrag opnode, string baseOpc, bit Commutable = 0> { // The register-immediate version is re-materializable. This is useful // in particular for taking the address of a local. let isReMaterializable = 1 in { @@ -713,6 +759,24 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc, let Inst{15-12} = Rd; let Inst{11-0} = shift; } + + // Assembly aliases for optional destination operand when it's the same + // as the source operand. + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $imm"), + (!cast<Instruction>(!strconcat(baseOpc, "ri")) GPR:$Rdn, GPR:$Rdn, + so_imm:$imm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $Rm"), + (!cast<Instruction>(!strconcat(baseOpc, "rr")) GPR:$Rdn, GPR:$Rdn, + GPR:$Rm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $shift"), + (!cast<Instruction>(!strconcat(baseOpc, "rs")) GPR:$Rdn, GPR:$Rdn, + so_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; } /// AI1_bin_s_irs - Similar to AsI1_bin_irs except it sets the 's' bit so the @@ -909,9 +973,9 @@ multiclass AI_exta_rrot_np<bits<8> opcod, string opc> { } /// AI1_adde_sube_irs - Define instructions and patterns for adde and sube. -let Uses = [CPSR] in { multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode, - bit Commutable = 0> { + string baseOpc, bit Commutable = 0> { + let Uses = [CPSR] in { def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm, IIC_iALUi, opc, "\t$Rd, $Rn, $imm", [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]>, @@ -950,7 +1014,24 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode, let Inst{15-12} = Rd; let Inst{19-16} = Rn; } -} + } + // Assembly aliases for optional destination operand when it's the same + // as the source operand. + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $imm"), + (!cast<Instruction>(!strconcat(baseOpc, "ri")) GPR:$Rdn, GPR:$Rdn, + so_imm:$imm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $Rm"), + (!cast<Instruction>(!strconcat(baseOpc, "rr")) GPR:$Rdn, GPR:$Rdn, + GPR:$Rm, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $shift"), + (!cast<Instruction>(!strconcat(baseOpc, "rs")) GPR:$Rdn, GPR:$Rdn, + so_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsARM]>; } // Carry setting variants @@ -958,15 +1039,15 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode, let usesCustomInserter = 1 in { multiclass AI1_adde_sube_s_irs<PatFrag opnode, bit Commutable = 0> { def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - Size4Bytes, IIC_iALUi, + 4, IIC_iALUi, [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]>; def rr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - Size4Bytes, IIC_iALUr, + 4, IIC_iALUr, [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]> { let isCommutable = Commutable; } def rs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - Size4Bytes, IIC_iALUsr, + 4, IIC_iALUsr, [(set GPR:$Rd, (opnode GPR:$Rn, so_reg:$shift))]>; } } @@ -1116,9 +1197,8 @@ def SEV : AI<(outs), (ins), MiscFrm, NoItinerary, "sev", "", // The i32imm operand $val can be used by a debugger to store more information // about the breakpoint. -def BKPT : AI<(outs), (ins i32imm:$val), MiscFrm, NoItinerary, "bkpt", "\t$val", - [/* For disassembly only; pattern left blank */]>, - Requires<[IsARM]> { +def BKPT : AI<(outs), (ins imm0_65535:$val), MiscFrm, NoItinerary, + "bkpt", "\t$val", []>, Requires<[IsARM]> { bits<16> val; let Inst{3-0} = val{3-0}; let Inst{19-8} = val{15-4}; @@ -1208,9 +1288,8 @@ def SETEND : AXI<(outs),(ins setend_op:$end), MiscFrm, NoItinerary, let Inst{8-0} = 0; } -def DBG : AI<(outs), (ins i32imm:$opt), MiscFrm, NoItinerary, "dbg", "\t$opt", - [/* For disassembly only; pattern left blank */]>, - Requires<[IsARM, HasV7]> { +def DBG : AI<(outs), (ins imm0_15:$opt), MiscFrm, NoItinerary, "dbg", "\t$opt", + []>, Requires<[IsARM, HasV7]> { bits<4> opt; let Inst{27-4} = 0b001100100000111100001111; let Inst{3-0} = opt; @@ -1227,40 +1306,40 @@ def TRAP : AXI<(outs), (ins), MiscFrm, NoItinerary, // Address computation and loads and stores in PIC mode. let isNotDuplicable = 1 in { def PICADD : ARMPseudoInst<(outs GPR:$dst), (ins GPR:$a, pclabel:$cp, pred:$p), - Size4Bytes, IIC_iALUr, + 4, IIC_iALUr, [(set GPR:$dst, (ARMpic_add GPR:$a, imm:$cp))]>; let AddedComplexity = 10 in { def PICLDR : ARMPseudoInst<(outs GPR:$dst), (ins addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iLoad_r, + 4, IIC_iLoad_r, [(set GPR:$dst, (load addrmodepc:$addr))]>; def PICLDRH : ARMPseudoInst<(outs GPR:$Rt), (ins addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iLoad_bh_r, + 4, IIC_iLoad_bh_r, [(set GPR:$Rt, (zextloadi16 addrmodepc:$addr))]>; def PICLDRB : ARMPseudoInst<(outs GPR:$Rt), (ins addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iLoad_bh_r, + 4, IIC_iLoad_bh_r, [(set GPR:$Rt, (zextloadi8 addrmodepc:$addr))]>; def PICLDRSH : ARMPseudoInst<(outs GPR:$Rt), (ins addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iLoad_bh_r, + 4, IIC_iLoad_bh_r, [(set GPR:$Rt, (sextloadi16 addrmodepc:$addr))]>; def PICLDRSB : ARMPseudoInst<(outs GPR:$Rt), (ins addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iLoad_bh_r, + 4, IIC_iLoad_bh_r, [(set GPR:$Rt, (sextloadi8 addrmodepc:$addr))]>; } let AddedComplexity = 10 in { def PICSTR : ARMPseudoInst<(outs), (ins GPR:$src, addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iStore_r, [(store GPR:$src, addrmodepc:$addr)]>; + 4, IIC_iStore_r, [(store GPR:$src, addrmodepc:$addr)]>; def PICSTRH : ARMPseudoInst<(outs), (ins GPR:$src, addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iStore_bh_r, [(truncstorei16 GPR:$src, + 4, IIC_iStore_bh_r, [(truncstorei16 GPR:$src, addrmodepc:$addr)]>; def PICSTRB : ARMPseudoInst<(outs), (ins GPR:$src, addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iStore_bh_r, [(truncstorei8 GPR:$src, addrmodepc:$addr)]>; + 4, IIC_iStore_bh_r, [(truncstorei8 GPR:$src, addrmodepc:$addr)]>; } } // isNotDuplicable = 1 @@ -1282,11 +1361,11 @@ def ADR : AI1<{0,?,?,0}, (outs GPR:$Rd), (ins adrlabel:$label), let Inst{11-0} = label; } def LEApcrel : ARMPseudoInst<(outs GPR:$Rd), (ins i32imm:$label, pred:$p), - Size4Bytes, IIC_iALUi, []>; + 4, IIC_iALUi, []>; def LEApcrelJT : ARMPseudoInst<(outs GPR:$Rd), (ins i32imm:$label, nohash_imm:$id, pred:$p), - Size4Bytes, IIC_iALUi, []>; + 4, IIC_iALUi, []>; //===----------------------------------------------------------------------===// // Control Flow Instructions. @@ -1319,22 +1398,13 @@ let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in { let Inst{3-0} = dst; } - // For disassembly only. - def BX_pred : AXI<(outs), (ins GPR:$dst, pred:$p), BrMiscFrm, IIC_Br, - "bx$p\t$dst", [/* pattern left blank */]>, + def BX_pred : AI<(outs), (ins GPR:$dst), BrMiscFrm, IIC_Br, + "bx", "\t$dst", [/* pattern left blank */]>, Requires<[IsARM, HasV4T]> { bits<4> dst; let Inst{27-4} = 0b000100101111111111110001; let Inst{3-0} = dst; } - - // ARMV4 only - // FIXME: We would really like to define this as a vanilla ARMPat like: - // ARMPat<(brind GPR:$dst), (MOVr PC, GPR:$dst)> - // With that, however, we can't set isBranch, isTerminator, etc.. - def MOVPCRX : ARMPseudoInst<(outs), (ins GPR:$dst), - Size4Bytes, IIC_Br, [(brind GPR:$dst)]>, - Requires<[IsARM, NoV4T]>; } // All calls clobber the non-callee saved registers. SP is marked as @@ -1386,12 +1456,12 @@ let isCall = 1, // ARMv4T // Note: Restrict $func to the tGPR regclass to prevent it being in LR. def BX_CALL : ARMPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size8Bytes, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, + 8, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsARM, HasV4T, IsNotDarwin]>; // ARMv4 def BMOVPCRX_CALL : ARMPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size8Bytes, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, + 8, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsARM, NoV4T, IsNotDarwin]>; } @@ -1401,131 +1471,82 @@ let isCall = 1, // moved above / below calls. Defs = [R0, R1, R2, R3, R9, R12, LR, QQQQ0, QQQQ2, QQQQ3, CPSR, FPSCR], Uses = [R7, SP] in { - def BLr9 : ARMPseudoInst<(outs), (ins bltarget:$func, variable_ops), - Size4Bytes, IIC_Br, - [(ARMcall tglobaladdr:$func)]>, Requires<[IsARM, IsDarwin]>; - - def BLr9_pred : ARMPseudoInst<(outs), - (ins bltarget:$func, pred:$p, variable_ops), - Size4Bytes, IIC_Br, - [(ARMcall_pred tglobaladdr:$func)]>, + def BLr9 : ARMPseudoExpand<(outs), (ins bl_target:$func, variable_ops), + 4, IIC_Br, + [(ARMcall tglobaladdr:$func)], (BL bl_target:$func)>, + Requires<[IsARM, IsDarwin]>; + + def BLr9_pred : ARMPseudoExpand<(outs), + (ins bl_target:$func, pred:$p, variable_ops), + 4, IIC_Br, + [(ARMcall_pred tglobaladdr:$func)], + (BL_pred bl_target:$func, pred:$p)>, Requires<[IsARM, IsDarwin]>; // ARMv5T and above - def BLXr9 : ARMPseudoInst<(outs), (ins GPR:$func, variable_ops), - Size4Bytes, IIC_Br, - [(ARMcall GPR:$func)]>, Requires<[IsARM, HasV5T, IsDarwin]>; - - def BLXr9_pred: ARMPseudoInst<(outs), (ins GPR:$func, pred:$p, variable_ops), - Size4Bytes, IIC_Br, - [(ARMcall_pred GPR:$func)]>, + def BLXr9 : ARMPseudoExpand<(outs), (ins GPR:$func, variable_ops), + 4, IIC_Br, + [(ARMcall GPR:$func)], + (BLX GPR:$func)>, + Requires<[IsARM, HasV5T, IsDarwin]>; + + def BLXr9_pred: ARMPseudoExpand<(outs), (ins GPR:$func, pred:$p,variable_ops), + 4, IIC_Br, + [(ARMcall_pred GPR:$func)], + (BLX_pred GPR:$func, pred:$p)>, Requires<[IsARM, HasV5T, IsDarwin]>; // ARMv4T // Note: Restrict $func to the tGPR regclass to prevent it being in LR. def BXr9_CALL : ARMPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size8Bytes, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, + 8, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsARM, HasV4T, IsDarwin]>; // ARMv4 def BMOVPCRXr9_CALL : ARMPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size8Bytes, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, + 8, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsARM, NoV4T, IsDarwin]>; } -// Tail calls. - -// FIXME: The Thumb versions of these should live in ARMInstrThumb.td -let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { - // Darwin versions. - let Defs = [R0, R1, R2, R3, R9, R12, QQQQ0, QQQQ2, QQQQ3, PC], - Uses = [SP] in { - def TCRETURNdi : PseudoInst<(outs), (ins i32imm:$dst, variable_ops), - IIC_Br, []>, Requires<[IsDarwin]>; - - def TCRETURNri : PseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - IIC_Br, []>, Requires<[IsDarwin]>; - - def TAILJMPd : ARMPseudoInst<(outs), (ins brtarget:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsARM, IsDarwin]>; - - def tTAILJMPd: tPseudoInst<(outs), (ins brtarget:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsThumb, IsDarwin]>; - - def TAILJMPr : ARMPseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsARM, IsDarwin]>; - - def tTAILJMPr : tPseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsThumb, IsDarwin]>; - } - - // Non-Darwin versions (the difference is R9). - let Defs = [R0, R1, R2, R3, R12, QQQQ0, QQQQ2, QQQQ3, PC], - Uses = [SP] in { - def TCRETURNdiND : PseudoInst<(outs), (ins i32imm:$dst, variable_ops), - IIC_Br, []>, Requires<[IsNotDarwin]>; - - def TCRETURNriND : PseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - IIC_Br, []>, Requires<[IsNotDarwin]>; - - def TAILJMPdND : ARMPseudoInst<(outs), (ins brtarget:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsARM, IsNotDarwin]>; - - def tTAILJMPdND : tPseudoInst<(outs), (ins brtarget:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsThumb, IsNotDarwin]>; - - def TAILJMPrND : ARMPseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsARM, IsNotDarwin]>; - def tTAILJMPrND : tPseudoInst<(outs), (ins tcGPR:$dst, variable_ops), - Size4Bytes, IIC_Br, - []>, Requires<[IsThumb, IsNotDarwin]>; +let isBranch = 1, isTerminator = 1 in { + // FIXME: should be able to write a pattern for ARMBrcond, but can't use + // a two-value operand where a dag node expects two operands. :( + def Bcc : ABI<0b1010, (outs), (ins br_target:$target), + IIC_Br, "b", "\t$target", + [/*(ARMbrcond bb:$target, imm:$cc, CCR:$ccr)*/]> { + bits<24> target; + let Inst{23-0} = target; } -} -let isBranch = 1, isTerminator = 1 in { - // B is "predicable" since it's just a Bcc with an 'always' condition. let isBarrier = 1 in { + // B is "predicable" since it's just a Bcc with an 'always' condition. let isPredicable = 1 in // FIXME: We shouldn't need this pseudo at all. Just using Bcc directly // should be sufficient. - def B : ARMPseudoInst<(outs), (ins brtarget:$target), Size4Bytes, IIC_Br, - [(br bb:$target)]>; + // FIXME: Is B really a Barrier? That doesn't seem right. + def B : ARMPseudoExpand<(outs), (ins br_target:$target), 4, IIC_Br, + [(br bb:$target)], (Bcc br_target:$target, (ops 14, zero_reg))>; let isNotDuplicable = 1, isIndirectBranch = 1 in { def BR_JTr : ARMPseudoInst<(outs), (ins GPR:$target, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, + 0, IIC_Br, [(ARMbrjt GPR:$target, tjumptable:$jt, imm:$id)]>; // FIXME: This shouldn't use the generic "addrmode2," but rather be split // into i12 and rs suffixed versions. def BR_JTm : ARMPseudoInst<(outs), (ins addrmode2:$target, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, + 0, IIC_Br, [(ARMbrjt (i32 (load addrmode2:$target)), tjumptable:$jt, imm:$id)]>; def BR_JTadd : ARMPseudoInst<(outs), (ins GPR:$target, GPR:$idx, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, + 0, IIC_Br, [(ARMbrjt (add GPR:$target, GPR:$idx), tjumptable:$jt, imm:$id)]>; } // isNotDuplicable = 1, isIndirectBranch = 1 } // isBarrier = 1 - // FIXME: should be able to write a pattern for ARMBrcond, but can't use - // a two-value operand where a dag node expects two operands. :( - def Bcc : ABI<0b1010, (outs), (ins br_target:$target), - IIC_Br, "b", "\t$target", - [/*(ARMbrcond bb:$target, imm:$cc, CCR:$ccr)*/]> { - bits<24> target; - let Inst{23-0} = target; - } } // BLX (immediate) -- for disassembly only @@ -1538,14 +1559,65 @@ def BLXi : AXI<(outs), (ins br_target:$target), BrMiscFrm, NoItinerary, let Inst{24} = target{0}; } -// Branch and Exchange Jazelle -- for disassembly only +// Branch and Exchange Jazelle def BXJ : ABI<0b0001, (outs), (ins GPR:$func), NoItinerary, "bxj", "\t$func", - [/* For disassembly only; pattern left blank */]> { + [/* pattern left blank */]> { + bits<4> func; let Inst{23-20} = 0b0010; - //let Inst{19-8} = 0xfff; + let Inst{19-8} = 0xfff; let Inst{7-4} = 0b0010; + let Inst{3-0} = func; +} + +// Tail calls. + +let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { + // Darwin versions. + let Defs = [R0, R1, R2, R3, R9, R12, QQQQ0, QQQQ2, QQQQ3, PC], + Uses = [SP] in { + def TCRETURNdi : PseudoInst<(outs), (ins i32imm:$dst, variable_ops), + IIC_Br, []>, Requires<[IsDarwin]>; + + def TCRETURNri : PseudoInst<(outs), (ins tcGPR:$dst, variable_ops), + IIC_Br, []>, Requires<[IsDarwin]>; + + def TAILJMPd : ARMPseudoExpand<(outs), (ins br_target:$dst, variable_ops), + 4, IIC_Br, [], + (Bcc br_target:$dst, (ops 14, zero_reg))>, + Requires<[IsARM, IsDarwin]>; + + def TAILJMPr : ARMPseudoExpand<(outs), (ins tcGPR:$dst, variable_ops), + 4, IIC_Br, [], + (BX GPR:$dst)>, + Requires<[IsARM, IsDarwin]>; + + } + + // Non-Darwin versions (the difference is R9). + let Defs = [R0, R1, R2, R3, R12, QQQQ0, QQQQ2, QQQQ3, PC], + Uses = [SP] in { + def TCRETURNdiND : PseudoInst<(outs), (ins i32imm:$dst, variable_ops), + IIC_Br, []>, Requires<[IsNotDarwin]>; + + def TCRETURNriND : PseudoInst<(outs), (ins tcGPR:$dst, variable_ops), + IIC_Br, []>, Requires<[IsNotDarwin]>; + + def TAILJMPdND : ARMPseudoExpand<(outs), (ins brtarget:$dst, variable_ops), + 4, IIC_Br, [], + (Bcc br_target:$dst, (ops 14, zero_reg))>, + Requires<[IsARM, IsNotDarwin]>; + + def TAILJMPrND : ARMPseudoExpand<(outs), (ins tcGPR:$dst, variable_ops), + 4, IIC_Br, [], + (BX GPR:$dst)>, + Requires<[IsARM, IsNotDarwin]>; + } } + + + + // Secure Monitor Call is a system instruction -- for disassembly only def SMC : ABI<0b0001, (outs), (ins i32imm:$opt), NoItinerary, "smc", "\t$opt", [/* For disassembly only; pattern left blank */]> { @@ -1562,7 +1634,6 @@ def SVC : ABI<0b1111, (outs), (ins i32imm:$svc), IIC_Br, "svc", "\t$svc", let Inst{23-0} = svc; } } -def : MnemonicAlias<"swi", "svc">; // Store Return State is a system instruction -- for disassembly only let isCodeGenOnly = 1 in { // FIXME: This should not use submode! @@ -1908,10 +1979,12 @@ def STRHT: AI3sthpo<(outs GPR:$base_wb), (ins GPR:$Rt, addrmode3:$addr), multiclass arm_ldst_mult<string asm, bit L_bit, Format f, InstrItinClass itin, InstrItinClass itin_upd> { + // IA is the default, so no need for an explicit suffix on the + // mnemonic here. Without it is the cannonical spelling. def IA : AXI4<(outs), (ins GPR:$Rn, pred:$p, reglist:$regs, variable_ops), IndexModeNone, f, itin, - !strconcat(asm, "ia${p}\t$Rn, $regs"), "", []> { + !strconcat(asm, "${p}\t$Rn, $regs"), "", []> { let Inst{24-23} = 0b01; // Increment After let Inst{21} = 0; // No writeback let Inst{20} = L_bit; @@ -1919,7 +1992,7 @@ multiclass arm_ldst_mult<string asm, bit L_bit, Format f, def IA_UPD : AXI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, reglist:$regs, variable_ops), IndexModeUpd, f, itin_upd, - !strconcat(asm, "ia${p}\t$Rn!, $regs"), "$Rn = $wb", []> { + !strconcat(asm, "${p}\t$Rn!, $regs"), "$Rn = $wb", []> { let Inst{24-23} = 0b01; // Increment After let Inst{21} = 1; // Writeback let Inst{20} = L_bit; @@ -1984,17 +2057,14 @@ defm STM : arm_ldst_mult<"stm", 0, LdStMulFrm, IIC_iStore_m, IIC_iStore_mu>; } // neverHasSideEffects -// Load / Store Multiple Mnemonic Aliases -def : MnemonicAlias<"ldm", "ldmia">; -def : MnemonicAlias<"stm", "stmia">; - // FIXME: remove when we have a way to marking a MI with these properties. // FIXME: Should pc be an implicit operand like PICADD, etc? let isReturn = 1, isTerminator = 1, isBarrier = 1, mayLoad = 1, hasExtraDefRegAllocReq = 1, isCodeGenOnly = 1 in -def LDMIA_RET : ARMPseudoInst<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, - reglist:$regs, variable_ops), - Size4Bytes, IIC_iLoad_mBr, []>, +def LDMIA_RET : ARMPseudoExpand<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, + reglist:$regs, variable_ops), + 4, IIC_iLoad_mBr, [], + (LDMIA_UPD GPR:$wb, GPR:$Rn, pred:$p, reglist:$regs)>, RegConstraint<"$Rn = $wb">; //===----------------------------------------------------------------------===// @@ -2164,7 +2234,7 @@ defm UXTAB16 : AI_exta_rrot_np<0b01101100, "uxtab16">; def SBFX : I<(outs GPR:$Rd), (ins GPR:$Rn, imm0_31:$lsb, imm0_31_m1:$width), - AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi, + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "sbfx", "\t$Rd, $Rn, $lsb, $width", "", []>, Requires<[IsARM, HasV6T2]> { bits<4> Rd; @@ -2181,7 +2251,7 @@ def SBFX : I<(outs GPR:$Rd), def UBFX : I<(outs GPR:$Rd), (ins GPR:$Rn, imm0_31:$lsb, imm0_31_m1:$width), - AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi, + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "ubfx", "\t$Rd, $Rn, $lsb, $width", "", []>, Requires<[IsARM, HasV6T2]> { bits<4> Rd; @@ -2202,10 +2272,10 @@ def UBFX : I<(outs GPR:$Rd), defm ADD : AsI1_bin_irs<0b0100, "add", IIC_iALUi, IIC_iALUr, IIC_iALUsr, - BinOpFrag<(add node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(add node:$LHS, node:$RHS)>, "ADD", 1>; defm SUB : AsI1_bin_irs<0b0010, "sub", IIC_iALUi, IIC_iALUr, IIC_iALUsr, - BinOpFrag<(sub node:$LHS, node:$RHS)>>; + BinOpFrag<(sub node:$LHS, node:$RHS)>, "SUB">; // ADD and SUB with 's' bit set. defm ADDS : AI1_bin_s_irs<0b0100, "adds", @@ -2216,9 +2286,11 @@ defm SUBS : AI1_bin_s_irs<0b0010, "subs", BinOpFrag<(subc node:$LHS, node:$RHS)>>; defm ADC : AI1_adde_sube_irs<0b0101, "adc", - BinOpFrag<(adde_dead_carry node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(adde_dead_carry node:$LHS, node:$RHS)>, + "ADC", 1>; defm SBC : AI1_adde_sube_irs<0b0110, "sbc", - BinOpFrag<(sube_dead_carry node:$LHS, node:$RHS)>>; + BinOpFrag<(sube_dead_carry node:$LHS, node:$RHS)>, + "SBC">; // ADC and SUBC with 's' bit set. let usesCustomInserter = 1 in { @@ -2271,13 +2343,13 @@ def RSBrs : AsI1<0b0011, (outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), // NOTE: CPSR def omitted because it will be handled by the custom inserter. let usesCustomInserter = 1 in { def RSBSri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - Size4Bytes, IIC_iALUi, + 4, IIC_iALUi, [(set GPR:$Rd, (subc so_imm:$imm, GPR:$Rn))]>; def RSBSrr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - Size4Bytes, IIC_iALUr, + 4, IIC_iALUr, [/* For disassembly only; pattern left blank */]>; def RSBSrs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - Size4Bytes, IIC_iALUsr, + 4, IIC_iALUsr, [(set GPR:$Rd, (subc so_reg:$shift, GPR:$Rn))]>; } @@ -2325,10 +2397,10 @@ def RSCrs : AsI1<0b0111, (outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), // NOTE: CPSR def omitted because it will be handled by the custom inserter. let usesCustomInserter = 1, Uses = [CPSR] in { def RSCSri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), - Size4Bytes, IIC_iALUi, + 4, IIC_iALUi, [(set GPR:$Rd, (sube_dead_carry so_imm:$imm, GPR:$Rn))]>; def RSCSrs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg:$shift), - Size4Bytes, IIC_iALUsr, + 4, IIC_iALUsr, [(set GPR:$Rd, (sube_dead_carry so_reg:$shift, GPR:$Rn))]>; } @@ -2528,19 +2600,19 @@ def : ARMV6Pat<(int_arm_usat GPR:$a, imm:$pos), (USAT imm:$pos, GPR:$a, 0)>; defm AND : AsI1_bin_irs<0b0000, "and", IIC_iBITi, IIC_iBITr, IIC_iBITsr, - BinOpFrag<(and node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(and node:$LHS, node:$RHS)>, "AND", 1>; defm ORR : AsI1_bin_irs<0b1100, "orr", IIC_iBITi, IIC_iBITr, IIC_iBITsr, - BinOpFrag<(or node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(or node:$LHS, node:$RHS)>, "ORR", 1>; defm EOR : AsI1_bin_irs<0b0001, "eor", IIC_iBITi, IIC_iBITr, IIC_iBITsr, - BinOpFrag<(xor node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(xor node:$LHS, node:$RHS)>, "EOR", 1>; defm BIC : AsI1_bin_irs<0b1110, "bic", IIC_iBITi, IIC_iBITr, IIC_iBITsr, - BinOpFrag<(and node:$LHS, (not node:$RHS))>>; + BinOpFrag<(and node:$LHS, (not node:$RHS))>, "BIC">; def BFC : I<(outs GPR:$Rd), (ins GPR:$src, bf_inv_mask_imm:$imm), - AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi, + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "bfc", "\t$Rd, $imm", "$src = $Rd", [(set GPR:$Rd, (and GPR:$src, bf_inv_mask_imm:$imm))]>, Requires<[IsARM, HasV6T2]> { @@ -2555,7 +2627,7 @@ def BFC : I<(outs GPR:$Rd), (ins GPR:$src, bf_inv_mask_imm:$imm), // A8.6.18 BFI - Bitfield insert (Encoding A1) def BFI : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm), - AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi, + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "bfi", "\t$Rd, $Rn, $imm", "$src = $Rd", [(set GPR:$Rd, (ARMbfi GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm))]>, @@ -2575,7 +2647,7 @@ def BFI : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, bf_inv_mask_imm:$imm), let isAsmParserOnly = 1 in def BFI4p : I<(outs GPR:$Rd), (ins GPR:$src, GPR:$Rn, lsb_pos_imm:$lsb, width_imm:$width), - AddrMode1, Size4Bytes, IndexModeNone, DPFrm, IIC_iUNAsi, + AddrMode1, 4, IndexModeNone, DPFrm, IIC_iUNAsi, "bfi", "\t$Rd, $Rn, $lsb, $width", "$src = $Rd", []>, Requires<[IsARM, HasV6T2]> { bits<4> Rd; @@ -2652,31 +2724,26 @@ class AsMul1I64<bits<7> opcod, dag oops, dag iops, InstrItinClass itin, let Inst{3-0} = Rn; } +// FIXME: The v5 pseudos are only necessary for the additional Constraint +// property. Remove them when it's possible to add those properties +// on an individual MachineInstr, not just an instuction description. let isCommutable = 1 in { -let Constraints = "@earlyclobber $Rd" in -def MULv5: ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, - pred:$p, cc_out:$s), - Size4Bytes, IIC_iMUL32, - [(set GPR:$Rd, (mul GPR:$Rn, GPR:$Rm))]>, - Requires<[IsARM, NoV6]>; - def MUL : AsMul1I32<0b0000000, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), IIC_iMUL32, "mul", "\t$Rd, $Rn, $Rm", [(set GPR:$Rd, (mul GPR:$Rn, GPR:$Rm))]>, Requires<[IsARM, HasV6]> { let Inst{15-12} = 0b0000; } -} let Constraints = "@earlyclobber $Rd" in -def MLAv5: ARMPseudoInst<(outs GPR:$Rd), - (ins GPR:$Rn, GPR:$Rm, GPR:$Ra, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMAC32, - [(set GPR:$Rd, (add (mul GPR:$Rn, GPR:$Rm), GPR:$Ra))]>, - Requires<[IsARM, NoV6]> { - bits<4> Ra; - let Inst{15-12} = Ra; +def MULv5: ARMPseudoExpand<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, + pred:$p, cc_out:$s), + 4, IIC_iMUL32, + [(set GPR:$Rd, (mul GPR:$Rn, GPR:$Rm))], + (MUL GPR:$Rd, GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s)>, + Requires<[IsARM, NoV6]>; } + def MLA : AsMul1I32<0b0000001, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), IIC_iMAC32, "mla", "\t$Rd, $Rn, $Rm, $Ra", [(set GPR:$Rd, (add (mul GPR:$Rn, GPR:$Rm), GPR:$Ra))]>, @@ -2685,6 +2752,14 @@ def MLA : AsMul1I32<0b0000001, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), let Inst{15-12} = Ra; } +let Constraints = "@earlyclobber $Rd" in +def MLAv5: ARMPseudoExpand<(outs GPR:$Rd), + (ins GPR:$Rn, GPR:$Rm, GPR:$Ra, pred:$p, cc_out:$s), + 4, IIC_iMAC32, + [(set GPR:$Rd, (add (mul GPR:$Rn, GPR:$Rm), GPR:$Ra))], + (MLA GPR:$Rd, GPR:$Rn, GPR:$Rm, GPR:$Ra, pred:$p, cc_out:$s)>, + Requires<[IsARM, NoV6]>; + def MLS : AMul1I<0b0000011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), IIC_iMAC32, "mls", "\t$Rd, $Rn, $Rm, $Ra", [(set GPR:$Rd, (sub GPR:$Ra, (mul GPR:$Rn, GPR:$Rm)))]>, @@ -2700,49 +2775,34 @@ def MLS : AMul1I<0b0000011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra), } // Extra precision multiplies with low / high results - let neverHasSideEffects = 1 in { let isCommutable = 1 in { -let Constraints = "@earlyclobber $RdLo,@earlyclobber $RdHi" in { -def SMULLv5 : ARMPseudoInst<(outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMUL64, []>, - Requires<[IsARM, NoV6]>; - -def UMULLv5 : ARMPseudoInst<(outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMUL64, []>, - Requires<[IsARM, NoV6]>; -} - def SMULL : AsMul1I64<0b0000110, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), IIC_iMUL64, + (ins GPR:$Rn, GPR:$Rm), IIC_iMUL64, "smull", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV6]>; def UMULL : AsMul1I64<0b0000100, (outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm), IIC_iMUL64, + (ins GPR:$Rn, GPR:$Rm), IIC_iMUL64, "umull", "\t$RdLo, $RdHi, $Rn, $Rm", []>, Requires<[IsARM, HasV6]>; -} -// Multiply + accumulate let Constraints = "@earlyclobber $RdLo,@earlyclobber $RdHi" in { -def SMLALv5 : ARMPseudoInst<(outs GPR:$RdLo, GPR:$RdHi), - (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMAC64, []>, - Requires<[IsARM, NoV6]>; -def UMLALv5 : ARMPseudoInst<(outs GPR:$RdLo, GPR:$RdHi), +def SMULLv5 : ARMPseudoExpand<(outs GPR:$RdLo, GPR:$RdHi), (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMAC64, []>, + 4, IIC_iMUL64, [], + (SMULL GPR:$RdLo, GPR:$RdHi, GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s)>, Requires<[IsARM, NoV6]>; -def UMAALv5 : ARMPseudoInst<(outs GPR:$RdLo, GPR:$RdHi), + +def UMULLv5 : ARMPseudoExpand<(outs GPR:$RdLo, GPR:$RdHi), (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), - Size4Bytes, IIC_iMAC64, []>, + 4, IIC_iMUL64, [], + (UMULL GPR:$RdLo, GPR:$RdHi, GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s)>, Requires<[IsARM, NoV6]>; - +} } +// Multiply + accumulate def SMLAL : AsMul1I64<0b0000111, (outs GPR:$RdLo, GPR:$RdHi), (ins GPR:$Rn, GPR:$Rm), IIC_iMAC64, "smlal", "\t$RdLo, $RdHi, $Rn, $Rm", []>, @@ -2765,6 +2825,25 @@ def UMAAL : AMul1I <0b0000010, (outs GPR:$RdLo, GPR:$RdHi), let Inst{11-8} = Rm; let Inst{3-0} = Rn; } + +let Constraints = "@earlyclobber $RdLo,@earlyclobber $RdHi" in { +def SMLALv5 : ARMPseudoExpand<(outs GPR:$RdLo, GPR:$RdHi), + (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), + 4, IIC_iMAC64, [], + (SMLAL GPR:$RdLo, GPR:$RdHi, GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s)>, + Requires<[IsARM, NoV6]>; +def UMLALv5 : ARMPseudoExpand<(outs GPR:$RdLo, GPR:$RdHi), + (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), + 4, IIC_iMAC64, [], + (UMLAL GPR:$RdLo, GPR:$RdHi, GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s)>, + Requires<[IsARM, NoV6]>; +def UMAALv5 : ARMPseudoExpand<(outs GPR:$RdLo, GPR:$RdHi), + (ins GPR:$Rn, GPR:$Rm, pred:$p), + 4, IIC_iMAC64, [], + (UMAAL GPR:$RdLo, GPR:$RdHi, GPR:$Rn, GPR:$Rm, pred:$p)>, + Requires<[IsARM, NoV6]>; +} + } // neverHasSideEffects // Most significant word multiply @@ -3005,31 +3084,22 @@ def REV : AMiscA1I<0b01101011, 0b0011, (outs GPR:$Rd), (ins GPR:$Rm), IIC_iUNAr, "rev", "\t$Rd, $Rm", [(set GPR:$Rd, (bswap GPR:$Rm))]>, Requires<[IsARM, HasV6]>; +let AddedComplexity = 5 in def REV16 : AMiscA1I<0b01101011, 0b1011, (outs GPR:$Rd), (ins GPR:$Rm), IIC_iUNAr, "rev16", "\t$Rd, $Rm", - [(set GPR:$Rd, - (or (and (srl GPR:$Rm, (i32 8)), 0xFF), - (or (and (shl GPR:$Rm, (i32 8)), 0xFF00), - (or (and (srl GPR:$Rm, (i32 8)), 0xFF0000), - (and (shl GPR:$Rm, (i32 8)), 0xFF000000)))))]>, + [(set GPR:$Rd, (rotr (bswap GPR:$Rm), (i32 16)))]>, Requires<[IsARM, HasV6]>; +let AddedComplexity = 5 in def REVSH : AMiscA1I<0b01101111, 0b1011, (outs GPR:$Rd), (ins GPR:$Rm), IIC_iUNAr, "revsh", "\t$Rd, $Rm", - [(set GPR:$Rd, - (sext_inreg - (or (srl GPR:$Rm, (i32 8)), - (shl GPR:$Rm, (i32 8))), i16))]>, + [(set GPR:$Rd, (sra (bswap GPR:$Rm), (i32 16)))]>, Requires<[IsARM, HasV6]>; -def : ARMV6Pat<(sext_inreg (or (srl (and GPR:$Rm, 0xFF00), (i32 8)), - (shl GPR:$Rm, (i32 8))), i16), +def : ARMV6Pat<(or (sra (shl GPR:$Rm, (i32 24)), (i32 16)), + (and (srl GPR:$Rm, (i32 8)), 0xFF)), (REVSH GPR:$Rm)>; -// Need the AddedComplexity or else MOVs + REV would be chosen. -let AddedComplexity = 5 in -def : ARMV6Pat<(sra (bswap GPR:$Rm), (i32 16)), (REVSH GPR:$Rm)>; - def lsl_shift_imm : SDNodeXForm<imm, [{ unsigned Sh = ARM_AM::getSORegOpc(ARM_AM::lsl, N->getZExtValue()); return CurDAG->getTargetConstant(Sh, MVT::i32); @@ -3177,26 +3247,26 @@ def BCCZi64 : PseudoInst<(outs), // a two-value operand where a dag node expects two operands. :( let neverHasSideEffects = 1 in { def MOVCCr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, GPR:$Rm, pred:$p), - Size4Bytes, IIC_iCMOVr, + 4, IIC_iCMOVr, [/*(set GPR:$Rd, (ARMcmov GPR:$false, GPR:$Rm, imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; def MOVCCs : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, so_reg:$shift, pred:$p), - Size4Bytes, IIC_iCMOVsr, + 4, IIC_iCMOVsr, [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_reg:$shift, imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; let isMoveImm = 1 in def MOVCCi16 : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, i32imm_hilo16:$imm, pred:$p), - Size4Bytes, IIC_iMOVi, + 4, IIC_iMOVi, []>, RegConstraint<"$false = $Rd">, Requires<[IsARM, HasV6T2]>; let isMoveImm = 1 in def MOVCCi : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, so_imm:$imm, pred:$p), - Size4Bytes, IIC_iCMOVi, + 4, IIC_iCMOVi, [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_imm:$imm, imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; @@ -3204,12 +3274,12 @@ def MOVCCi : ARMPseudoInst<(outs GPR:$Rd), let isMoveImm = 1 in def MOVCCi32imm : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, i32imm:$src, pred:$p), - Size8Bytes, IIC_iCMOVix2, []>, RegConstraint<"$false = $Rd">; + 8, IIC_iCMOVix2, []>, RegConstraint<"$false = $Rd">; let isMoveImm = 1 in def MVNCCi : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$false, so_imm:$imm, pred:$p), - Size4Bytes, IIC_iCMOVi, + 4, IIC_iCMOVi, [/*(set GPR:$Rd, (ARMcmov GPR:$false, so_imm_not:$imm, imm:$cc, CCR:$ccr))*/]>, RegConstraint<"$false = $Rd">; } // neverHasSideEffects @@ -3235,19 +3305,20 @@ def DMB : AInoP<(outs), (ins memb_opt:$opt), MiscFrm, NoItinerary, } def DSB : AInoP<(outs), (ins memb_opt:$opt), MiscFrm, NoItinerary, - "dsb", "\t$opt", - [/* For disassembly only; pattern left blank */]>, + "dsb", "\t$opt", []>, Requires<[IsARM, HasDB]> { bits<4> opt; let Inst{31-4} = 0xf57ff04; let Inst{3-0} = opt; } -// ISB has only full system option -- for disassembly only -def ISB : AInoP<(outs), (ins), MiscFrm, NoItinerary, "isb", "", []>, +// ISB has only full system option +def ISB : AInoP<(outs), (ins memb_opt:$opt), MiscFrm, NoItinerary, + "isb", "\t$opt", []>, Requires<[IsARM, HasDB]> { + bits<4> opt; let Inst{31-4} = 0xf57ff06; - let Inst{3-0} = 0b1111; + let Inst{3-0} = opt; } let usesCustomInserter = 1 in { @@ -3410,8 +3481,8 @@ def SWPB : AIswp<1, (outs GPR:$Rt), (ins GPR:$Rt2, GPR:$Rn), "swpb", // Coprocessor Instructions. // -def CDP : ABI<0b1110, (outs), (ins p_imm:$cop, i32imm:$opc1, - c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), +def CDP : ABI<0b1110, (outs), (ins p_imm:$cop, imm0_15:$opc1, + c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, imm0_7:$opc2), NoItinerary, "cdp", "\t$cop, $opc1, $CRd, $CRn, $CRm, $opc2", [(int_arm_cdp imm:$cop, imm:$opc1, imm:$CRd, imm:$CRn, imm:$CRm, imm:$opc2)]> { @@ -3431,8 +3502,8 @@ def CDP : ABI<0b1110, (outs), (ins p_imm:$cop, i32imm:$opc1, let Inst{23-20} = opc1; } -def CDP2 : ABXI<0b1110, (outs), (ins p_imm:$cop, i32imm:$opc1, - c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), +def CDP2 : ABXI<0b1110, (outs), (ins p_imm:$cop, imm0_15:$opc1, + c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, imm0_7:$opc2), NoItinerary, "cdp2\t$cop, $opc1, $CRd, $CRn, $CRm, $opc2", [(int_arm_cdp2 imm:$cop, imm:$opc1, imm:$CRd, imm:$CRn, imm:$CRm, imm:$opc2)]> { @@ -3455,7 +3526,7 @@ def CDP2 : ABXI<0b1110, (outs), (ins p_imm:$cop, i32imm:$opc1, class ACI<dag oops, dag iops, string opc, string asm, IndexMode im = IndexModeNone> - : InoP<oops, iops, AddrModeNone, Size4Bytes, im, BrFrm, NoItinerary, + : InoP<oops, iops, AddrModeNone, 4, im, BrFrm, NoItinerary, opc, asm, "", [/* For disassembly only; pattern left blank */]> { let Inst{27-25} = 0b110; } @@ -3583,8 +3654,8 @@ class MovRCopro<string opc, bit direction, dag oops, dag iops, def MCR : MovRCopro<"mcr", 0 /* from ARM core register to coprocessor */, (outs), - (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, c_imm:$CRn, - c_imm:$CRm, i32imm:$opc2), + (ins p_imm:$cop, imm0_7:$opc1, GPR:$Rt, c_imm:$CRn, + c_imm:$CRm, imm0_7:$opc2), [(int_arm_mcr imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, imm:$CRm, imm:$opc2)]>; def MRC : MovRCopro<"mrc", 1 /* from coprocessor to ARM core register */, @@ -3620,8 +3691,8 @@ class MovRCopro2<string opc, bit direction, dag oops, dag iops, def MCR2 : MovRCopro2<"mcr2", 0 /* from ARM core register to coprocessor */, (outs), - (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, c_imm:$CRn, - c_imm:$CRm, i32imm:$opc2), + (ins p_imm:$cop, imm0_7:$opc1, GPR:$Rt, c_imm:$CRn, + c_imm:$CRm, imm0_7:$opc2), [(int_arm_mcr2 imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, imm:$CRm, imm:$opc2)]>; def MRC2 : MovRCopro2<"mrc2", 1 /* from coprocessor to ARM core register */, @@ -3635,7 +3706,7 @@ def : ARMV5TPat<(int_arm_mrc2 imm:$cop, imm:$opc1, imm:$CRn, class MovRRCopro<string opc, bit direction, list<dag> pattern = [/* For disassembly only */]> - : ABI<0b1100, (outs), (ins p_imm:$cop, i32imm:$opc1, + : ABI<0b1100, (outs), (ins p_imm:$cop, imm0_15:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), NoItinerary, opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm", pattern> { let Inst{23-21} = 0b010; @@ -3661,7 +3732,7 @@ def MRRC : MovRRCopro<"mrrc", 1 /* from coprocessor to ARM core register */>; class MovRRCopro2<string opc, bit direction, list<dag> pattern = [/* For disassembly only */]> - : ABXI<0b1100, (outs), (ins p_imm:$cop, i32imm:$opc1, + : ABXI<0b1100, (outs), (ins p_imm:$cop, imm0_15:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), NoItinerary, !strconcat(opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm"), pattern> { let Inst{31-28} = 0b1111; @@ -3812,6 +3883,13 @@ def Int_eh_sjlj_dispatchsetup : // Non-Instruction Patterns // +// ARMv4 indirect branch using (MOVr PC, dst) +let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in + def MOVPCRX : ARMPseudoExpand<(outs), (ins GPR:$dst), + 4, IIC_Br, [(brind GPR:$dst)], + (MOVr PC, GPR:$dst, (ops 14, zero_reg), zero_reg)>, + Requires<[IsARM, NoV4T]>; + // Large immediate handling. // 32-bit immediate using two piece so_imms or movw + movt. @@ -3977,3 +4055,22 @@ include "ARMInstrVFP.td" include "ARMInstrNEON.td" +//===----------------------------------------------------------------------===// +// Assembler aliases +// + +// Memory barriers +def : InstAlias<"dmb", (DMB 0xf)>, Requires<[IsARM, HasDB]>; +def : InstAlias<"dsb", (DSB 0xf)>, Requires<[IsARM, HasDB]>; +def : InstAlias<"isb", (ISB 0xf)>, Requires<[IsARM, HasDB]>; + +// System instructions +def : MnemonicAlias<"swi", "svc">; + +// Load / Store Multiple +def : MnemonicAlias<"ldmfd", "ldm">; +def : MnemonicAlias<"ldmia", "ldm">; +def : MnemonicAlias<"stmfd", "stmdb">; +def : MnemonicAlias<"stmia", "stm">; +def : MnemonicAlias<"stmea", "stm">; + diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index 79d95d9..0df62f4 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -175,7 +175,7 @@ class VLDQQWBPseudo<InstrItinClass itin> (ins addrmode6:$addr, am6offset:$offset), itin, "$addr.addr = $wb">; class VLDQQQQPseudo<InstrItinClass itin> - : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src), itin,"">; + : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,"">; class VLDQQQQWBPseudo<InstrItinClass itin> : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb), (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin, @@ -1387,7 +1387,7 @@ class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty, : NLdStLn<1, 0b00, op11_8, op7_4, (outs), (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane), IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "", - [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]> { + [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{ let Rm = 0b1111; } class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp> @@ -3793,7 +3793,8 @@ def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VCNTiD, "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd", - [(set DPR:$Vd, (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>; + [(set DPR:$Vd, + (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>; def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd), (and DPR:$Vm, (vnotd DPR:$Vd)))), @@ -3803,7 +3804,8 @@ def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VCNTiQ, "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd", - [(set QPR:$Vd, (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>; + [(set QPR:$Vd, + (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>; def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd), (and QPR:$Vm, (vnotq QPR:$Vd)))), @@ -4212,17 +4214,12 @@ def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0, // Vector Move Operations. // VMOV : Vector Move (Register) +def : InstAlias<"vmov${p} $Vd, $Vm", + (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>; +def : InstAlias<"vmov${p} $Vd, $Vm", + (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>; let neverHasSideEffects = 1 in { -def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm), - N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> { - let Vn{4-0} = Vm{4-0}; -} -def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm), - N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> { - let Vn{4-0} = Vm{4-0}; -} - // Pseudo vector move instructions for QQ and QQQQ registers. This should // be expanded after register allocation is completed. def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src), @@ -4702,11 +4699,10 @@ def VEXTd32 : VEXTd<"vext", "32", v2i32> { let Inst{11-10} = index{1-0}; let Inst{9-8} = 0b00; } -def VEXTdf : VEXTd<"vext", "32", v2f32> { - let Inst{11-10} = index{1-0}; - let Inst{9-8} = 0b00; - -} +def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn), + (v2f32 DPR:$Vm), + (i32 imm:$index))), + (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>; def VEXTq8 : VEXTq<"vext", "8", v16i8> { let Inst{11-8} = index{3-0}; @@ -4719,10 +4715,10 @@ def VEXTq32 : VEXTq<"vext", "32", v4i32> { let Inst{11-10} = index{1-0}; let Inst{9-8} = 0b00; } -def VEXTqf : VEXTq<"vext", "32", v4f32> { - let Inst{11-10} = index{1-0}; - let Inst{9-8} = 0b00; -} +def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn), + (v4f32 QPR:$Vm), + (i32 imm:$index))), + (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>; // VTRN : Vector Transpose diff --git a/lib/Target/ARM/ARMInstrThumb.td b/lib/Target/ARM/ARMInstrThumb.td index 4777189..bfe83ec 100644 --- a/lib/Target/ARM/ARMInstrThumb.td +++ b/lib/Target/ARM/ARMInstrThumb.td @@ -26,17 +26,14 @@ def imm_comp_XFORM : SDNodeXForm<imm, [{ return CurDAG->getTargetConstant(~((uint32_t)N->getZExtValue()), MVT::i32); }]>; -/// imm0_7 predicate - True if the 32-bit immediate is in the range [0,7]. -def imm0_7 : ImmLeaf<i32, [{ - return Imm >= 0 && Imm < 8; -}]>; def imm0_7_neg : PatLeaf<(i32 imm), [{ return (uint32_t)-N->getZExtValue() < 8; }], imm_neg_XFORM>; -def imm0_255 : ImmLeaf<i32, [{ - return Imm >= 0 && Imm < 256; -}]>; +def imm0_255_asmoperand : AsmOperandClass { let Name = "Imm0_255"; } +def imm0_255 : Operand<i32>, ImmLeaf<i32, [{ return Imm >= 0 && Imm < 256; }]> { + let ParserMatchClass = imm0_255_asmoperand; +} def imm0_255_comp : PatLeaf<(i32 imm), [{ return ~((uint32_t)N->getZExtValue()) < 256; }]>; @@ -74,10 +71,12 @@ def t_adrlabel : Operand<i32> { // Scaled 4 immediate. def t_imm_s4 : Operand<i32> { let PrintMethod = "printThumbS4ImmOperand"; + let OperandType = "OPERAND_IMMEDIATE"; } // Define Thumb specific addressing modes. +let OperandType = "OPERAND_PCREL" in { def t_brtarget : Operand<OtherVT> { let EncoderMethod = "getThumbBRTargetOpValue"; } @@ -97,6 +96,7 @@ def t_bltarget : Operand<i32> { def t_blxtarget : Operand<i32> { let EncoderMethod = "getThumbBLXTargetOpValue"; } +} def MemModeRegThumbAsmOperand : AsmOperandClass { let Name = "MemModeRegThumb"; @@ -360,27 +360,6 @@ def tADDspr : TIt<(outs GPR:$dst), (ins GPR:$lhs, GPR:$rhs), IIC_iALUr, // Control Flow Instructions. // -let isReturn = 1, isTerminator = 1, isBarrier = 1 in { - def tBX_RET : TI<(outs), (ins), IIC_Br, "bx\tlr", - [(ARMretflag)]>, - T1Special<{1,1,0,?}> { - // A6.2.3 & A8.6.25 - let Inst{6-3} = 0b1110; // Rm = lr - let Inst{2-0} = 0b000; - } - - // Alternative return instruction used by vararg functions. - def tBX_RET_vararg : TI<(outs), (ins tGPR:$Rm), - IIC_Br, "bx\t$Rm", - []>, - T1Special<{1,1,0,?}> { - // A6.2.3 & A8.6.25 - bits<4> Rm; - let Inst{6-3} = Rm; - let Inst{2-0} = 0b000; - } -} - // Indirect branches let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in { def tBX : TI<(outs), (ins GPR:$Rm, pred:$p), IIC_Br, "bx${p}\t$Rm", []>, @@ -390,31 +369,16 @@ let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in { let Inst{6-3} = Rm; let Inst{2-0} = 0b000; } - - def tBRIND : TI<(outs), (ins GPR:$Rm), - IIC_Br, - "mov\tpc, $Rm", - [(brind GPR:$Rm)]>, - T1Special<{1,0,?,?}> { - // A8.6.97 - bits<4> Rm; - let Inst{7} = 1; // <Rd> = Inst{7:2-0} = pc - let Inst{6-3} = Rm; - let Inst{2-0} = 0b111; - } } -// FIXME: remove when we have a way to marking a MI with these properties. -let isReturn = 1, isTerminator = 1, isBarrier = 1, mayLoad = 1, - hasExtraDefRegAllocReq = 1 in -def tPOP_RET : T1I<(outs), (ins pred:$p, reglist:$regs, variable_ops), - IIC_iPop_Br, - "pop${p}\t$regs", []>, - T1Misc<{1,1,0,?,?,?,?}> { - // A8.6.121 - bits<16> regs; - let Inst{8} = regs{15}; // registers = P:'0000000':register_list - let Inst{7-0} = regs{7-0}; +let isReturn = 1, isTerminator = 1, isBarrier = 1 in { + def tBX_RET : tPseudoExpand<(outs), (ins pred:$p), 2, IIC_Br, + [(ARMretflag)], (tBX LR, pred:$p)>; + + // Alternative return instruction used by vararg functions. + def tBX_RET_vararg : tPseudoExpand<(outs), (ins tGPR:$Rm, pred:$p), + 2, IIC_Br, [], + (tBX GPR:$Rm, pred:$p)>; } // All calls clobber the non-callee saved registers. SP is marked as a use to @@ -464,7 +428,7 @@ let isCall = 1, // ARMv4T def tBX_CALL : tPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size4Bytes, IIC_Br, + 4, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsThumb, IsThumb1Only, IsNotDarwin]>; } @@ -516,7 +480,7 @@ let isCall = 1, // ARMv4T def tBXr9_CALL : tPseudoInst<(outs), (ins tGPR:$func, variable_ops), - Size4Bytes, IIC_Br, + 4, IIC_Br, [(ARMcall_nolink tGPR:$func)]>, Requires<[IsThumb, IsThumb1Only, IsDarwin]>; } @@ -534,12 +498,12 @@ let isBranch = 1, isTerminator = 1, isBarrier = 1 in { // Just a pseudo for a tBL instruction. Needed to let regalloc know about // the clobber of LR. let Defs = [LR] in - def tBfar : tPseudoInst<(outs), (ins t_bltarget:$target), - Size4Bytes, IIC_Br, []>; + def tBfar : tPseudoExpand<(outs), (ins t_bltarget:$target), + 4, IIC_Br, [], (tBL t_bltarget:$target)>; def tBR_JTr : tPseudoInst<(outs), (ins tGPR:$target, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, + 0, IIC_Br, [(ARMbrjt tGPR:$target, tjumptable:$jt, imm:$id)]> { list<Predicate> Predicates = [IsThumb, IsThumb1Only]; } @@ -583,6 +547,33 @@ let isBranch = 1, isTerminator = 1 in { } } +// Tail calls +let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { + // Darwin versions. + let Defs = [R0, R1, R2, R3, R9, R12, QQQQ0, QQQQ2, QQQQ3, PC], + Uses = [SP] in { + // tTAILJMPd: Darwin version uses a Thumb2 branch (no Thumb1 tail calls + // on Darwin), so it's in ARMInstrThumb2.td. + def tTAILJMPr : tPseudoExpand<(outs), (ins tcGPR:$dst, variable_ops), + 4, IIC_Br, [], + (tBX GPR:$dst, (ops 14, zero_reg))>, + Requires<[IsThumb, IsDarwin]>; + } + // Non-Darwin versions (the difference is R9). + let Defs = [R0, R1, R2, R3, R12, QQQQ0, QQQQ2, QQQQ3, PC], + Uses = [SP] in { + def tTAILJMPdND : tPseudoExpand<(outs), (ins t_brtarget:$dst, variable_ops), + 4, IIC_Br, [], + (tB t_brtarget:$dst)>, + Requires<[IsThumb, IsNotDarwin]>; + def tTAILJMPrND : tPseudoExpand<(outs), (ins tcGPR:$dst, variable_ops), + 4, IIC_Br, [], + (tBX GPR:$dst, (ops 14, zero_reg))>, + Requires<[IsThumb, IsNotDarwin]>; + } +} + + // A8.6.218 Supervisor Call (Software Interrupt) -- for disassembly only // A8.6.16 B: Encoding T1 // If Inst{11-8} == 0b1111 then SEE SVC @@ -685,19 +676,6 @@ def tLDRspi : T1pIs<(outs tGPR:$Rt), (ins t_addrmode_sp:$addr), IIC_iLoad_i, let Inst{7-0} = addr; } -// Special instruction for restore. It cannot clobber condition register -// when it's expanded by eliminateCallFramePseudoInstr(). -let canFoldAsLoad = 1, mayLoad = 1, neverHasSideEffects = 1 in -// FIXME: Pseudo for tLDRspi -def tRestore : T1pIs<(outs tGPR:$dst), (ins t_addrmode_sp:$addr), IIC_iLoad_i, - "ldr", "\t$dst, $addr", []>, - T1LdStSP<{1,?,?}> { - bits<3> Rt; - bits<8> addr; - let Inst{10-8} = Rt; - let Inst{7-0} = addr; -} - // Load tconstpool // FIXME: Use ldr.n to work around a Darwin assembler bug. let canFoldAsLoad = 1, isReMaterializable = 1 in @@ -739,9 +717,9 @@ defm tSTRB : thumb_st_rr_ri_enc<0b010, 0b0111, t_addrmode_rrs1, // A8.6.207 & A8.6.205 defm tSTRH : thumb_st_rr_ri_enc<0b001, 0b1000, t_addrmode_rrs2, - t_addrmode_is2, AddrModeT1_2, - IIC_iStore_bh_r, IIC_iStore_bh_i, "strh", - BinOpFrag<(truncstorei16 node:$LHS, node:$RHS)>>; + t_addrmode_is2, AddrModeT1_2, + IIC_iStore_bh_r, IIC_iStore_bh_i, "strh", + BinOpFrag<(truncstorei16 node:$LHS, node:$RHS)>>; def tSTRspi : T1pIs<(outs), (ins tGPR:$Rt, t_addrmode_sp:$addr), IIC_iStore_i, @@ -754,19 +732,6 @@ def tSTRspi : T1pIs<(outs), (ins tGPR:$Rt, t_addrmode_sp:$addr), IIC_iStore_i, let Inst{7-0} = addr; } -let mayStore = 1, neverHasSideEffects = 1 in -// Special instruction for spill. It cannot clobber condition register when it's -// expanded by eliminateCallFramePseudoInstr(). -// FIXME: Pseudo for tSTRspi -def tSpill : T1pIs<(outs), (ins tGPR:$src, t_addrmode_sp:$addr), IIC_iStore_i, - "str", "\t$src, $addr", []>, - T1LdStSP<{0,?,?}> { - bits<3> Rt; - bits<8> addr; - let Inst{10-8} = Rt; - let Inst{7-0} = addr; -} - //===----------------------------------------------------------------------===// // Load / store multiple Instructions. // @@ -911,7 +876,8 @@ def tADC : // A8.6.2 // Add immediate def tADDi3 : // A8.6.4 T1 - T1sIGenEncodeImm<0b01110, (outs tGPR:$Rd), (ins tGPR:$Rm, i32imm:$imm3), IIC_iALUi, + T1sIGenEncodeImm<0b01110, (outs tGPR:$Rd), (ins tGPR:$Rm, i32imm:$imm3), + IIC_iALUi, "add", "\t$Rd, $Rm, $imm3", [(set tGPR:$Rd, (add tGPR:$Rm, imm0_7:$imm3))]> { bits<3> imm3; @@ -1071,7 +1037,7 @@ def tLSRrr : // A8.6.91 // Move register let isMoveImm = 1 in -def tMOVi8 : T1sI<(outs tGPR:$Rd), (ins i32imm:$imm8), IIC_iMOVi, +def tMOVi8 : T1sI<(outs tGPR:$Rd), (ins imm0_255:$imm8), IIC_iMOVi, "mov", "\t$Rd, $imm8", [(set tGPR:$Rd, imm0_255:$imm8)]>, T1General<{1,0,0,?,?}> { @@ -1082,18 +1048,18 @@ def tMOVi8 : T1sI<(outs tGPR:$Rd), (ins i32imm:$imm8), IIC_iMOVi, let Inst{7-0} = imm8; } -// TODO: A7-73: MOV(2) - mov setting flag. +// A7-73: MOV(2) - mov setting flag. let neverHasSideEffects = 1 in { -// FIXME: Make this predicable. -def tMOVr : T1I<(outs tGPR:$Rd), (ins tGPR:$Rm), IIC_iMOVr, - "mov\t$Rd, $Rm", []>, - T1Special<0b1000> { +def tMOVr : Thumb1pI<(outs GPR:$Rd), (ins GPR:$Rm), AddrModeNone, + 2, IIC_iMOVr, + "mov", "\t$Rd, $Rm", "", []>, + T1Special<{1,0,?,?}> { // A8.6.97 bits<4> Rd; bits<4> Rm; - // Bits {7-6} are encoded by the T1Special value. - let Inst{5-3} = Rm{2-0}; + let Inst{7} = Rd{3}; + let Inst{6-3} = Rm; let Inst{2-0} = Rd{2-0}; } let Defs = [CPSR] in @@ -1106,39 +1072,6 @@ def tMOVSr : T1I<(outs tGPR:$Rd), (ins tGPR:$Rm), IIC_iMOVr, let Inst{5-3} = Rm; let Inst{2-0} = Rd; } - -// FIXME: Make these predicable. -def tMOVgpr2tgpr : T1I<(outs tGPR:$Rd), (ins GPR:$Rm), IIC_iMOVr, - "mov\t$Rd, $Rm", []>, - T1Special<{1,0,0,?}> { - // A8.6.97 - bits<4> Rd; - bits<4> Rm; - // Bit {7} is encoded by the T1Special value. - let Inst{6-3} = Rm; - let Inst{2-0} = Rd{2-0}; -} -def tMOVtgpr2gpr : T1I<(outs GPR:$Rd), (ins tGPR:$Rm), IIC_iMOVr, - "mov\t$Rd, $Rm", []>, - T1Special<{1,0,?,0}> { - // A8.6.97 - bits<4> Rd; - bits<4> Rm; - // Bit {6} is encoded by the T1Special value. - let Inst{7} = Rd{3}; - let Inst{5-3} = Rm{2-0}; - let Inst{2-0} = Rd{2-0}; -} -def tMOVgpr2gpr : T1I<(outs GPR:$Rd), (ins GPR:$Rm), IIC_iMOVr, - "mov\t$Rd, $Rm", []>, - T1Special<{1,0,?,?}> { - // A8.6.97 - bits<4> Rd; - bits<4> Rm; - let Inst{7} = Rd{3}; - let Inst{6-3} = Rm; - let Inst{2-0} = Rd{2-0}; -} } // neverHasSideEffects // Multiply register @@ -1175,31 +1108,16 @@ def tREV16 : // A8.6.135 T1pIMiscEncode<{1,0,1,0,0,1,?}, (outs tGPR:$Rd), (ins tGPR:$Rm), IIC_iUNAr, "rev16", "\t$Rd, $Rm", - [(set tGPR:$Rd, - (or (and (srl tGPR:$Rm, (i32 8)), 0xFF), - (or (and (shl tGPR:$Rm, (i32 8)), 0xFF00), - (or (and (srl tGPR:$Rm, (i32 8)), 0xFF0000), - (and (shl tGPR:$Rm, (i32 8)), 0xFF000000)))))]>, + [(set tGPR:$Rd, (rotr (bswap tGPR:$Rm), (i32 16)))]>, Requires<[IsThumb, IsThumb1Only, HasV6]>; def tREVSH : // A8.6.136 T1pIMiscEncode<{1,0,1,0,1,1,?}, (outs tGPR:$Rd), (ins tGPR:$Rm), IIC_iUNAr, "revsh", "\t$Rd, $Rm", - [(set tGPR:$Rd, - (sext_inreg - (or (srl tGPR:$Rm, (i32 8)), - (shl tGPR:$Rm, (i32 8))), i16))]>, + [(set tGPR:$Rd, (sra (bswap tGPR:$Rm), (i32 16)))]>, Requires<[IsThumb, IsThumb1Only, HasV6]>; -def : T1Pat<(sext_inreg (or (srl (and tGPR:$Rm, 0xFF00), (i32 8)), - (shl tGPR:$Rm, (i32 8))), i16), - (tREVSH tGPR:$Rm)>, - Requires<[IsThumb, IsThumb1Only, HasV6]>; - -def : T1Pat<(sra (bswap tGPR:$Rm), (i32 16)), (tREVSH tGPR:$Rm)>, - Requires<[IsThumb, IsThumb1Only, HasV6]>; - // Rotate right register def tROR : // A8.6.139 T1sItDPEncode<0b0111, (outs tGPR:$Rdn), (ins tGPR:$Rn, tGPR:$Rm), @@ -1294,31 +1212,6 @@ let usesCustomInserter = 1 in // Expanded after instruction selection. NoItinerary, [/*(set tGPR:$dst, (ARMcmov tGPR:$false, tGPR:$true, imm:$cc))*/]>; - -// 16-bit movcc in IT blocks for Thumb2. -let neverHasSideEffects = 1 in { -def tMOVCCr : T1pIt<(outs GPR:$Rdn), (ins GPR:$Rn, GPR:$Rm), IIC_iCMOVr, - "mov", "\t$Rdn, $Rm", []>, - T1Special<{1,0,?,?}> { - bits<4> Rdn; - bits<4> Rm; - let Inst{7} = Rdn{3}; - let Inst{6-3} = Rm; - let Inst{2-0} = Rdn{2-0}; -} - -let isMoveImm = 1 in -def tMOVCCi : T1pIt<(outs tGPR:$Rdn), (ins tGPR:$Rn, i32imm:$Rm), IIC_iCMOVi, - "mov", "\t$Rdn, $Rm", []>, - T1General<{1,0,0,?,?}> { - bits<3> Rdn; - bits<8> Rm; - let Inst{10-8} = Rdn; - let Inst{7-0} = Rm; -} - -} // neverHasSideEffects - // tLEApcrel - Load a pc-relative address into a register without offending the // assembler. @@ -1333,118 +1226,22 @@ def tADR : T1I<(outs tGPR:$Rd), (ins t_adrlabel:$addr, pred:$p), let neverHasSideEffects = 1, isReMaterializable = 1 in def tLEApcrel : tPseudoInst<(outs tGPR:$Rd), (ins i32imm:$label, pred:$p), - Size2Bytes, IIC_iALUi, []>; + 2, IIC_iALUi, []>; def tLEApcrelJT : tPseudoInst<(outs tGPR:$Rd), (ins i32imm:$label, nohash_imm:$id, pred:$p), - Size2Bytes, IIC_iALUi, []>; - -//===----------------------------------------------------------------------===// -// Move between coprocessor and ARM core register -- for disassembly only -// - -class tMovRCopro<string opc, bit direction, dag oops, dag iops, - list<dag> pattern> - : T1Cop<oops, iops, !strconcat(opc, "\t$cop, $opc1, $Rt, $CRn, $CRm, $opc2"), - pattern> { - let Inst{27-24} = 0b1110; - let Inst{20} = direction; - let Inst{4} = 1; - - bits<4> Rt; - bits<4> cop; - bits<3> opc1; - bits<3> opc2; - bits<4> CRm; - bits<4> CRn; - - let Inst{15-12} = Rt; - let Inst{11-8} = cop; - let Inst{23-21} = opc1; - let Inst{7-5} = opc2; - let Inst{3-0} = CRm; - let Inst{19-16} = CRn; -} - -def tMCR : tMovRCopro<"mcr", 0 /* from ARM core register to coprocessor */, - (outs), - (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, c_imm:$CRn, - c_imm:$CRm, i32imm:$opc2), - [(int_arm_mcr imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, - imm:$CRm, imm:$opc2)]>; -def tMRC : tMovRCopro<"mrc", 1 /* from coprocessor to ARM core register */, - (outs GPR:$Rt), - (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), - []>; - -def : Pat<(int_arm_mrc imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2), - (tMRC imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>, - Requires<[IsThumb, HasV6T2]>; - -class tMovRRCopro<string opc, bit direction, - list<dag> pattern = [/* For disassembly only */]> - : T1Cop<(outs), (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), - !strconcat(opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm"), pattern> { - let Inst{27-24} = 0b1100; - let Inst{23-21} = 0b010; - let Inst{20} = direction; - - bits<4> Rt; - bits<4> Rt2; - bits<4> cop; - bits<4> opc1; - bits<4> CRm; - - let Inst{15-12} = Rt; - let Inst{19-16} = Rt2; - let Inst{11-8} = cop; - let Inst{7-4} = opc1; - let Inst{3-0} = CRm; -} - -def tMCRR : tMovRRCopro<"mcrr", 0 /* from ARM core register to coprocessor */, - [(int_arm_mcrr imm:$cop, imm:$opc1, GPR:$Rt, GPR:$Rt2, - imm:$CRm)]>; -def tMRRC : tMovRRCopro<"mrrc", 1 /* from coprocessor to ARM core register */>; - -//===----------------------------------------------------------------------===// -// Other Coprocessor Instructions. For disassembly only. -// -def tCDP : T1Cop<(outs), (ins p_imm:$cop, i32imm:$opc1, - c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), - "cdp\t$cop, $opc1, $CRd, $CRn, $CRm, $opc2", - [(int_arm_cdp imm:$cop, imm:$opc1, imm:$CRd, imm:$CRn, - imm:$CRm, imm:$opc2)]> { - let Inst{27-24} = 0b1110; - - bits<4> opc1; - bits<4> CRn; - bits<4> CRd; - bits<4> cop; - bits<3> opc2; - bits<4> CRm; - - let Inst{3-0} = CRm; - let Inst{4} = 0; - let Inst{7-5} = opc2; - let Inst{11-8} = cop; - let Inst{15-12} = CRd; - let Inst{19-16} = CRn; - let Inst{23-20} = opc1; -} + 2, IIC_iALUi, []>; //===----------------------------------------------------------------------===// // TLS Instructions // // __aeabi_read_tp preserves the registers r1-r3. -let isCall = 1, Defs = [R0, LR], Uses = [SP] in -def tTPsoft : TIx2<0b11110, 0b11, 1, (outs), (ins), IIC_Br, - "bl\t__aeabi_read_tp", - [(set R0, ARMthread_pointer)]> { - // Encoding is 0xf7fffffe. - let Inst = 0xf7fffffe; -} +// This is a pseudo inst so that we can get the encoding right, +// complete with fixup for the aeabi_read_tp function. +let isCall = 1, Defs = [R0, R12, LR, CPSR], Uses = [SP] in +def tTPsoft : tPseudoInst<(outs), (ins), 4, IIC_Br, + [(set R0, ARMthread_pointer)]>; //===----------------------------------------------------------------------===// // SJLJ Exception handling intrinsics @@ -1463,14 +1260,14 @@ def tTPsoft : TIx2<0b11110, 0b11, 1, (outs), (ins), IIC_Br, let Defs = [ R0, R1, R2, R3, R4, R5, R6, R7, R12, CPSR ], hasSideEffects = 1, isBarrier = 1, isCodeGenOnly = 1 in def tInt_eh_sjlj_setjmp : ThumbXI<(outs),(ins tGPR:$src, tGPR:$val), - AddrModeNone, SizeSpecial, NoItinerary, "","", + AddrModeNone, 0, NoItinerary, "","", [(set R0, (ARMeh_sjlj_setjmp tGPR:$src, tGPR:$val))]>; // FIXME: Non-Darwin version(s) let isBarrier = 1, hasSideEffects = 1, isTerminator = 1, isCodeGenOnly = 1, Defs = [ R7, LR, SP ] in def tInt_eh_sjlj_longjmp : XI<(outs), (ins GPR:$src, GPR:$scratch), - AddrModeNone, SizeSpecial, IndexModeNone, + AddrModeNone, 0, IndexModeNone, Pseudo, NoItinerary, "", "", [(ARMeh_sjlj_longjmp GPR:$src, GPR:$scratch)]>, Requires<[IsThumb, IsDarwin]>; @@ -1583,3 +1380,18 @@ def tLDRpci_pic : PseudoInst<(outs GPR:$dst), (ins i32imm:$addr, pclabel:$cp), [(set GPR:$dst, (ARMpic_add (load (ARMWrapper tconstpool:$addr)), imm:$cp))]>, Requires<[IsThumb, IsThumb1Only]>; + +// Pseudo-instruction for merged POP and return. +// FIXME: remove when we have a way to marking a MI with these properties. +let isReturn = 1, isTerminator = 1, isBarrier = 1, mayLoad = 1, + hasExtraDefRegAllocReq = 1 in +def tPOP_RET : tPseudoExpand<(outs), (ins pred:$p, reglist:$regs, variable_ops), + 2, IIC_iPop_Br, [], + (tPOP pred:$p, reglist:$regs)>; + +// Indirect branch using "mov pc, $Rm" +let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in { + def tBRIND : tPseudoExpand<(outs), (ins GPR:$Rm, pred:$p), + 2, IIC_Br, [(brind GPR:$Rm)], + (tMOVr PC, GPR:$Rm, pred:$p)>; +} diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td index 598660c..c2c6cbc 100644 --- a/lib/Target/ARM/ARMInstrThumb2.td +++ b/lib/Target/ARM/ARMInstrThumb2.td @@ -44,9 +44,11 @@ def t2_so_imm_neg_XFORM : SDNodeXForm<imm, [{ // t2_so_imm - Match a 32-bit immediate operand, which is an // 8-bit immediate rotated by an arbitrary number of bits, or an 8-bit // immediate splatted into multiple bytes of the word. +def t2_so_imm_asmoperand : AsmOperandClass { let Name = "T2SOImm"; } def t2_so_imm : Operand<i32>, ImmLeaf<i32, [{ return ARM_AM::getT2SOImmVal(Imm) != -1; }]> { + let ParserMatchClass = t2_so_imm_asmoperand; let EncoderMethod = "getT2SOImmOpValue"; } @@ -463,7 +465,8 @@ multiclass T2I_un_irs<bits<4> opcod, string opc, /// changed to modify CPSR. multiclass T2I_bin_irs<bits<4> opcod, string opc, InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, - PatFrag opnode, bit Commutable = 0, string wide = ""> { + PatFrag opnode, string baseOpc, bit Commutable = 0, + string wide = ""> { // shifted imm def ri : T2sTwoRegImm< (outs rGPR:$Rd), (ins rGPR:$Rn, t2_so_imm:$imm), iii, @@ -495,14 +498,31 @@ multiclass T2I_bin_irs<bits<4> opcod, string opc, let Inst{26-25} = 0b01; let Inst{24-21} = opcod; } + // Assembly aliases for optional destination operand when it's the same + // as the source operand. + def : InstAlias<!strconcat(opc, "${s}${p} $Rdn, $imm"), + (!cast<Instruction>(!strconcat(baseOpc, "ri")) rGPR:$Rdn, rGPR:$Rdn, + t2_so_imm:$imm, pred:$p, + cc_out:$s)>, + Requires<[IsThumb2]>; + def : InstAlias<!strconcat(opc, "${s}${p}", wide, " $Rdn, $Rm"), + (!cast<Instruction>(!strconcat(baseOpc, "rr")) rGPR:$Rdn, rGPR:$Rdn, + rGPR:$Rm, pred:$p, + cc_out:$s)>, + Requires<[IsThumb2]>; + def : InstAlias<!strconcat(opc, "${s}${p}", wide, " $Rdn, $shift"), + (!cast<Instruction>(!strconcat(baseOpc, "rs")) rGPR:$Rdn, rGPR:$Rdn, + t2_so_reg:$shift, pred:$p, + cc_out:$s)>, + Requires<[IsThumb2]>; } /// T2I_bin_w_irs - Same as T2I_bin_irs except these operations need -// the ".w" prefix to indicate that they are wide. +// the ".w" suffix to indicate that they are wide. multiclass T2I_bin_w_irs<bits<4> opcod, string opc, InstrItinClass iii, InstrItinClass iir, InstrItinClass iis, - PatFrag opnode, bit Commutable = 0> : - T2I_bin_irs<opcod, opc, iii, iir, iis, opnode, Commutable, ".w">; + PatFrag opnode, string baseOpc, bit Commutable = 0> : + T2I_bin_irs<opcod, opc, iii, iir, iis, opnode, baseOpc, Commutable, ".w">; /// T2I_rbin_is - Same as T2I_bin_irs except the order of operands are /// reversed. The 'rr' form is only defined for the disassembler; for codegen @@ -696,18 +716,18 @@ let usesCustomInserter = 1 in { multiclass T2I_adde_sube_s_irs<PatFrag opnode, bit Commutable = 0> { // shifted imm def ri : t2PseudoInst<(outs rGPR:$Rd), (ins rGPR:$Rn, t2_so_imm:$imm), - Size4Bytes, IIC_iALUi, + 4, IIC_iALUi, [(set rGPR:$Rd, (opnode rGPR:$Rn, t2_so_imm:$imm))]>; // register def rr : t2PseudoInst<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - Size4Bytes, IIC_iALUr, + 4, IIC_iALUr, [(set rGPR:$Rd, (opnode rGPR:$Rn, rGPR:$Rm))]> { let isCommutable = Commutable; } // shifted register def rs : t2PseudoInst< (outs rGPR:$Rd), (ins rGPR:$Rn, t2_so_reg:$ShiftedRm), - Size4Bytes, IIC_iALUsi, + 4, IIC_iALUsi, [(set rGPR:$Rd, (opnode rGPR:$Rn, t2_so_reg:$ShiftedRm))]>; } } @@ -1018,7 +1038,8 @@ multiclass T2I_ext_rrot_uxtb16<bits<3> opcod, string opc, PatFrag opnode> { // supported yet. multiclass T2I_ext_rrot_sxtb16<bits<3> opcod, string opc> { def r : T2TwoReg<(outs rGPR:$Rd), (ins rGPR:$Rm), IIC_iEXTr, - opc, "\t$Rd, $Rm", []> { + opc, "\t$Rd, $Rm", []>, + Requires<[IsThumb2, HasT2ExtractPack]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0100; let Inst{22-20} = opcod; @@ -1028,7 +1049,8 @@ multiclass T2I_ext_rrot_sxtb16<bits<3> opcod, string opc> { let Inst{5-4} = 0b00; // rotate } def r_rot : T2TwoReg<(outs rGPR:$Rd), (ins rGPR:$Rm, i32imm:$rot), IIC_iEXTr, - opc, "\t$Rd, $Rm, ror $rot", []> { + opc, "\t$Rd, $Rm, ror $rot", []>, + Requires<[IsThumb2, HasT2ExtractPack]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0100; let Inst{22-20} = opcod; @@ -1084,7 +1106,7 @@ multiclass T2I_exta_rrot_DO<bits<3> opcod, string opc> { let Inst{7} = 1; let Inst{5-4} = 0b00; // rotate } - def rr_rot : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, i32imm:$rot), + def rr_rot :T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, i32imm:$rot), IIC_iEXTAsr, opc, "\t$Rd, $Rn, $Rm, ror $rot", []> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0100; @@ -1142,93 +1164,13 @@ def t2ADR : T2PCOneRegImm<(outs rGPR:$Rd), let neverHasSideEffects = 1, isReMaterializable = 1 in def t2LEApcrel : t2PseudoInst<(outs rGPR:$Rd), (ins i32imm:$label, pred:$p), - Size4Bytes, IIC_iALUi, []>; + 4, IIC_iALUi, []>; def t2LEApcrelJT : t2PseudoInst<(outs rGPR:$Rd), (ins i32imm:$label, nohash_imm:$id, pred:$p), - Size4Bytes, IIC_iALUi, + 4, IIC_iALUi, []>; -// FIXME: None of these add/sub SP special instructions should be necessary -// at all for thumb2 since they use the same encodings as the generic -// add/sub instructions. In thumb1 we need them since they have dedicated -// encodings. At the least, they should be pseudo instructions. -// ADD r, sp, {so_imm|i12} -let isCodeGenOnly = 1 in { -def t2ADDrSPi : T2sTwoRegImm<(outs GPR:$Rd), (ins GPR:$Rn, t2_so_imm:$imm), - IIC_iALUi, "add", ".w\t$Rd, $Rn, $imm", []> { - let Inst{31-27} = 0b11110; - let Inst{25} = 0; - let Inst{24-21} = 0b1000; - let Inst{15} = 0; -} -def t2ADDrSPi12 : T2TwoRegImm<(outs GPR:$Rd), (ins GPR:$Rn, imm0_4095:$imm), - IIC_iALUi, "addw", "\t$Rd, $Rn, $imm", []> { - let Inst{31-27} = 0b11110; - let Inst{25-20} = 0b100000; - let Inst{15} = 0; -} - -// ADD r, sp, so_reg -def t2ADDrSPs : T2sTwoRegShiftedReg< - (outs GPR:$Rd), (ins GPR:$Rn, t2_so_reg:$ShiftedRm), - IIC_iALUsi, "add", ".w\t$Rd, $Rn, $ShiftedRm", []> { - let Inst{31-27} = 0b11101; - let Inst{26-25} = 0b01; - let Inst{24-21} = 0b1000; - let Inst{15} = 0; -} - -// SUB r, sp, {so_imm|i12} -def t2SUBrSPi : T2sTwoRegImm<(outs GPR:$Rd), (ins GPR:$Rn, t2_so_imm:$imm), - IIC_iALUi, "sub", ".w\t$Rd, $Rn, $imm", []> { - let Inst{31-27} = 0b11110; - let Inst{25} = 0; - let Inst{24-21} = 0b1101; - let Inst{15} = 0; -} -def t2SUBrSPi12 : T2TwoRegImm<(outs GPR:$Rd), (ins GPR:$Rn, imm0_4095:$imm), - IIC_iALUi, "subw", "\t$Rd, $Rn, $imm", []> { - let Inst{31-27} = 0b11110; - let Inst{25-20} = 0b101010; - let Inst{15} = 0; -} - -// SUB r, sp, so_reg -def t2SUBrSPs : T2sTwoRegImm<(outs GPR:$Rd), (ins GPR:$Rn, t2_so_reg:$imm), - IIC_iALUsi, - "sub", "\t$Rd, $Rn, $imm", []> { - let Inst{31-27} = 0b11101; - let Inst{26-25} = 0b01; - let Inst{24-21} = 0b1101; - let Inst{19-16} = 0b1101; // Rn = sp - let Inst{15} = 0; -} -} // end isCodeGenOnly = 1 - -// Signed and unsigned division on v7-M -def t2SDIV : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iALUi, - "sdiv", "\t$Rd, $Rn, $Rm", - [(set rGPR:$Rd, (sdiv rGPR:$Rn, rGPR:$Rm))]>, - Requires<[HasDivide, IsThumb2]> { - let Inst{31-27} = 0b11111; - let Inst{26-21} = 0b011100; - let Inst{20} = 0b1; - let Inst{15-12} = 0b1111; - let Inst{7-4} = 0b1111; -} - -def t2UDIV : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iALUi, - "udiv", "\t$Rd, $Rn, $Rm", - [(set rGPR:$Rd, (udiv rGPR:$Rn, rGPR:$Rm))]>, - Requires<[HasDivide, IsThumb2]> { - let Inst{31-27} = 0b11111; - let Inst{26-21} = 0b011101; - let Inst{20} = 0b1; - let Inst{15-12} = 0b1111; - let Inst{7-4} = 0b1111; -} - //===----------------------------------------------------------------------===// // Load / store Instructions. // @@ -1668,6 +1610,10 @@ def t2MOVi : T2sOneRegImm<(outs rGPR:$Rd), (ins t2_so_imm:$imm), IIC_iMOVi, let Inst{15} = 0; } +def : InstAlias<"mov${s}${p} $Rd, $imm", (t2MOVi rGPR:$Rd, t2_so_imm:$imm, + pred:$p, cc_out:$s)>, + Requires<[IsThumb2]>; + let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in def t2MOVi16 : T2I<(outs rGPR:$Rd), (ins i32imm_hilo16:$imm), IIC_iMOVi, "movw", "\t$Rd, $imm", @@ -1788,8 +1734,10 @@ defm t2ADC : T2I_adde_sube_irs<0b1010, "adc", BinOpFrag<(adde_dead_carry node:$LHS, node:$RHS)>, 1>; defm t2SBC : T2I_adde_sube_irs<0b1011, "sbc", BinOpFrag<(sube_dead_carry node:$LHS, node:$RHS)>>; -defm t2ADCS : T2I_adde_sube_s_irs<BinOpFrag<(adde_live_carry node:$LHS, node:$RHS)>, 1>; -defm t2SBCS : T2I_adde_sube_s_irs<BinOpFrag<(sube_live_carry node:$LHS, node:$RHS)>>; +defm t2ADCS : T2I_adde_sube_s_irs<BinOpFrag<(adde_live_carry node:$LHS, + node:$RHS)>, 1>; +defm t2SBCS : T2I_adde_sube_s_irs<BinOpFrag<(sube_live_carry node:$LHS, + node:$RHS)>>; // RSB defm t2RSB : T2I_rbin_irs <0b1110, "rsb", @@ -1833,7 +1781,8 @@ def : T2Pat<(adde_live_carry rGPR:$src, t2_so_imm_not:$imm), // Select Bytes -- for disassembly only def t2SEL : T2ThreeReg<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), - NoItinerary, "sel", "\t$Rd, $Rn, $Rm", []> { + NoItinerary, "sel", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-24} = 0b010; let Inst{23} = 0b1; @@ -1849,7 +1798,8 @@ class T2I_pam<bits<3> op22_20, bits<4> op7_4, string opc, list<dag> pat = [/* For disassembly only; pattern left blank */], dag iops = (ins rGPR:$Rn, rGPR:$Rm), string asm = "\t$Rd, $Rn, $Rm"> - : T2I<(outs rGPR:$Rd), iops, NoItinerary, opc, asm, pat> { + : T2I<(outs rGPR:$Rd), iops, NoItinerary, opc, asm, pat>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0101; let Inst{22-20} = op22_20; @@ -1947,12 +1897,14 @@ class T2FourReg_mac<bit long, bits<3> op22_20, bits<4> op7_4, dag oops, def t2USAD8 : T2ThreeReg_mac<0, 0b111, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - NoItinerary, "usad8", "\t$Rd, $Rn, $Rm", []> { + NoItinerary, "usad8", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{15-12} = 0b1111; } def t2USADA8 : T2FourReg_mac<0, 0b111, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), NoItinerary, - "usada8", "\t$Rd, $Rn, $Rm, $Ra", []>; + "usada8", "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]>; // Signed/Unsigned saturate -- for disassembly only @@ -1985,7 +1937,8 @@ def t2SSAT: T2SatI< def t2SSAT16: T2SatI< (outs rGPR:$Rd), (ins ssat_imm:$sat_imm, rGPR:$Rn), NoItinerary, "ssat16", "\t$Rd, $sat_imm, $Rn", - [/* For disassembly only; pattern left blank */]> { + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11110; let Inst{25-22} = 0b1100; let Inst{20} = 0; @@ -2005,10 +1958,11 @@ def t2USAT: T2SatI< let Inst{15} = 0; } -def t2USAT16: T2SatI< - (outs rGPR:$dst), (ins i32imm:$sat_imm, rGPR:$Rn), NoItinerary, - "usat16", "\t$dst, $sat_imm, $Rn", - [/* For disassembly only; pattern left blank */]> { +def t2USAT16: T2SatI<(outs rGPR:$dst), (ins i32imm:$sat_imm, rGPR:$Rn), + NoItinerary, + "usat16", "\t$dst, $sat_imm, $Rn", + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11110; let Inst{25-22} = 0b1110; let Inst{20} = 0; @@ -2084,17 +2038,18 @@ def t2MOVsra_flag : T2TwoRegShiftImm< defm t2AND : T2I_bin_w_irs<0b0000, "and", IIC_iBITi, IIC_iBITr, IIC_iBITsi, - BinOpFrag<(and node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(and node:$LHS, node:$RHS)>, "t2AND", 1>; defm t2ORR : T2I_bin_w_irs<0b0010, "orr", IIC_iBITi, IIC_iBITr, IIC_iBITsi, - BinOpFrag<(or node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(or node:$LHS, node:$RHS)>, "t2ORR", 1>; defm t2EOR : T2I_bin_w_irs<0b0100, "eor", IIC_iBITi, IIC_iBITr, IIC_iBITsi, - BinOpFrag<(xor node:$LHS, node:$RHS)>, 1>; + BinOpFrag<(xor node:$LHS, node:$RHS)>, "t2EOR", 1>; defm t2BIC : T2I_bin_w_irs<0b0001, "bic", IIC_iBITi, IIC_iBITr, IIC_iBITsi, - BinOpFrag<(and node:$LHS, (not node:$RHS))>>; + BinOpFrag<(and node:$LHS, (not node:$RHS))>, + "t2BIC">; class T2BitFI<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> @@ -2194,7 +2149,8 @@ let Constraints = "$src = $Rd" in { defm t2ORN : T2I_bin_irs<0b0011, "orn", IIC_iBITi, IIC_iBITr, IIC_iBITsi, - BinOpFrag<(or node:$LHS, (not node:$RHS))>, 0, "">; + BinOpFrag<(or node:$LHS, (not node:$RHS))>, + "t2ORN", 0, "">; // Prefer over of t2EORri ra, rb, -1 because mvn has 16-bit version let AddedComplexity = 1 in @@ -2277,7 +2233,8 @@ def t2UMLAL : T2MulLong<0b110, 0b0000, def t2UMAAL : T2MulLong<0b110, 0b0110, (outs rGPR:$RdLo, rGPR:$RdHi), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMAC64, - "umaal", "\t$RdLo, $RdHi, $Rn, $Rm", []>; + "umaal", "\t$RdLo, $RdHi, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]>; } // neverHasSideEffects // Rounding variants of the below included for disassembly only @@ -2285,7 +2242,8 @@ def t2UMAAL : T2MulLong<0b110, 0b0110, // Most significant word multiply def t2SMMUL : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL32, "smmul", "\t$Rd, $Rn, $Rm", - [(set rGPR:$Rd, (mulhs rGPR:$Rn, rGPR:$Rm))]> { + [(set rGPR:$Rd, (mulhs rGPR:$Rn, rGPR:$Rm))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b101; @@ -2294,7 +2252,8 @@ def t2SMMUL : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL32, } def t2SMMULR : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL32, - "smmulr", "\t$Rd, $Rn, $Rm", []> { + "smmulr", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b101; @@ -2305,7 +2264,8 @@ def t2SMMULR : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL32, def t2SMMLA : T2FourReg< (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smmla", "\t$Rd, $Rn, $Rm, $Ra", - [(set rGPR:$Rd, (add (mulhs rGPR:$Rm, rGPR:$Rn), rGPR:$Ra))]> { + [(set rGPR:$Rd, (add (mulhs rGPR:$Rm, rGPR:$Rn), rGPR:$Ra))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b101; @@ -2314,7 +2274,8 @@ def t2SMMLA : T2FourReg< def t2SMMLAR: T2FourReg< (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, - "smmlar", "\t$Rd, $Rn, $Rm, $Ra", []> { + "smmlar", "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b101; @@ -2324,7 +2285,8 @@ def t2SMMLAR: T2FourReg< def t2SMMLS: T2FourReg< (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smmls", "\t$Rd, $Rn, $Rm, $Ra", - [(set rGPR:$Rd, (sub rGPR:$Ra, (mulhs rGPR:$Rn, rGPR:$Rm)))]> { + [(set rGPR:$Rd, (sub rGPR:$Ra, (mulhs rGPR:$Rn, rGPR:$Rm)))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b110; @@ -2333,7 +2295,8 @@ def t2SMMLS: T2FourReg< def t2SMMLSR:T2FourReg< (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, - "smmlsr", "\t$Rd, $Rn, $Rm, $Ra", []> { + "smmlsr", "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b110; @@ -2344,7 +2307,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def BB : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "bb"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (opnode (sext_inreg rGPR:$Rn, i16), - (sext_inreg rGPR:$Rm, i16)))]> { + (sext_inreg rGPR:$Rm, i16)))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2356,7 +2320,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def BT : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "bt"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (opnode (sext_inreg rGPR:$Rn, i16), - (sra rGPR:$Rm, (i32 16))))]> { + (sra rGPR:$Rm, (i32 16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2368,7 +2333,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def TB : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "tb"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (opnode (sra rGPR:$Rn, (i32 16)), - (sext_inreg rGPR:$Rm, i16)))]> { + (sext_inreg rGPR:$Rm, i16)))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2380,7 +2346,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def TT : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "tt"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (opnode (sra rGPR:$Rn, (i32 16)), - (sra rGPR:$Rm, (i32 16))))]> { + (sra rGPR:$Rm, (i32 16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2392,7 +2359,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def WB : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "wb"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (sra (opnode rGPR:$Rn, - (sext_inreg rGPR:$Rm, i16)), (i32 16)))]> { + (sext_inreg rGPR:$Rm, i16)), (i32 16)))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b011; @@ -2404,7 +2372,8 @@ multiclass T2I_smul<string opc, PatFrag opnode> { def WT : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iMUL16, !strconcat(opc, "wt"), "\t$Rd, $Rn, $Rm", [(set rGPR:$Rd, (sra (opnode rGPR:$Rn, - (sra rGPR:$Rm, (i32 16))), (i32 16)))]> { + (sra rGPR:$Rm, (i32 16))), (i32 16)))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b011; @@ -2421,7 +2390,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { !strconcat(opc, "bb"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (opnode (sext_inreg rGPR:$Rn, i16), - (sext_inreg rGPR:$Rm, i16))))]> { + (sext_inreg rGPR:$Rm, i16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2433,7 +2403,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC16, !strconcat(opc, "bt"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (opnode (sext_inreg rGPR:$Rn, i16), - (sra rGPR:$Rm, (i32 16)))))]> { + (sra rGPR:$Rm, (i32 16)))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2445,7 +2416,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC16, !strconcat(opc, "tb"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (opnode (sra rGPR:$Rn, (i32 16)), - (sext_inreg rGPR:$Rm, i16))))]> { + (sext_inreg rGPR:$Rm, i16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2457,7 +2429,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC16, !strconcat(opc, "tt"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (opnode (sra rGPR:$Rn, (i32 16)), - (sra rGPR:$Rm, (i32 16)))))]> { + (sra rGPR:$Rm, (i32 16)))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b001; @@ -2469,7 +2442,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC16, !strconcat(opc, "wb"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (sra (opnode rGPR:$Rn, - (sext_inreg rGPR:$Rm, i16)), (i32 16))))]> { + (sext_inreg rGPR:$Rm, i16)), (i32 16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b011; @@ -2481,7 +2455,8 @@ multiclass T2I_smla<string opc, PatFrag opnode> { (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC16, !strconcat(opc, "wt"), "\t$Rd, $Rn, $Rm, $Ra", [(set rGPR:$Rd, (add rGPR:$Ra, (sra (opnode rGPR:$Rn, - (sra rGPR:$Rm, (i32 16))), (i32 16))))]> { + (sra rGPR:$Rm, (i32 16))), (i32 16))))]>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{31-27} = 0b11111; let Inst{26-23} = 0b0110; let Inst{22-20} = 0b011; @@ -2496,66 +2471,108 @@ defm t2SMLA : T2I_smla<"smla", BinOpFrag<(mul node:$LHS, node:$RHS)>>; // Halfword multiple accumulate long: SMLAL<x><y> -- for disassembly only def t2SMLALBB : T2FourReg_mac<1, 0b100, 0b1000, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rn,rGPR:$Rm), IIC_iMAC64, "smlalbb", "\t$Ra, $Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>; + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLALBT : T2FourReg_mac<1, 0b100, 0b1001, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rn,rGPR:$Rm), IIC_iMAC64, "smlalbt", "\t$Ra, $Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>; + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLALTB : T2FourReg_mac<1, 0b100, 0b1010, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rn,rGPR:$Rm), IIC_iMAC64, "smlaltb", "\t$Ra, $Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>; + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLALTT : T2FourReg_mac<1, 0b100, 0b1011, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rn,rGPR:$Rm), IIC_iMAC64, "smlaltt", "\t$Ra, $Rd, $Rn, $Rm", - [/* For disassembly only; pattern left blank */]>; + [/* For disassembly only; pattern left blank */]>, + Requires<[IsThumb2, HasThumb2DSP]>; // Dual halfword multiple: SMUAD, SMUSD, SMLAD, SMLSD, SMLALD, SMLSLD // These are for disassembly only. def t2SMUAD: T2ThreeReg_mac< 0, 0b010, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - IIC_iMAC32, "smuad", "\t$Rd, $Rn, $Rm", []> { + IIC_iMAC32, "smuad", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{15-12} = 0b1111; } def t2SMUADX:T2ThreeReg_mac< 0, 0b010, 0b0001, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - IIC_iMAC32, "smuadx", "\t$Rd, $Rn, $Rm", []> { + IIC_iMAC32, "smuadx", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{15-12} = 0b1111; } def t2SMUSD: T2ThreeReg_mac< 0, 0b100, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - IIC_iMAC32, "smusd", "\t$Rd, $Rn, $Rm", []> { + IIC_iMAC32, "smusd", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{15-12} = 0b1111; } def t2SMUSDX:T2ThreeReg_mac< 0, 0b100, 0b0001, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), - IIC_iMAC32, "smusdx", "\t$Rd, $Rn, $Rm", []> { + IIC_iMAC32, "smusdx", "\t$Rd, $Rn, $Rm", []>, + Requires<[IsThumb2, HasThumb2DSP]> { let Inst{15-12} = 0b1111; } def t2SMLAD : T2ThreeReg_mac< 0, 0b010, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smlad", - "\t$Rd, $Rn, $Rm, $Ra", []>; + "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLADX : T2FourReg_mac< 0, 0b010, 0b0001, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smladx", - "\t$Rd, $Rn, $Rm, $Ra", []>; + "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLSD : T2FourReg_mac<0, 0b100, 0b0000, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smlsd", - "\t$Rd, $Rn, $Rm, $Ra", []>; + "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLSDX : T2FourReg_mac<0, 0b100, 0b0001, (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMAC32, "smlsdx", - "\t$Rd, $Rn, $Rm, $Ra", []>; + "\t$Rd, $Rn, $Rm, $Ra", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLALD : T2FourReg_mac<1, 0b100, 0b1100, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rm, rGPR:$Rn), IIC_iMAC64, "smlald", - "\t$Ra, $Rd, $Rm, $Rn", []>; + "\t$Ra, $Rd, $Rm, $Rn", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLALDX : T2FourReg_mac<1, 0b100, 0b1101, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rm,rGPR:$Rn), IIC_iMAC64, "smlaldx", - "\t$Ra, $Rd, $Rm, $Rn", []>; + "\t$Ra, $Rd, $Rm, $Rn", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLSLD : T2FourReg_mac<1, 0b101, 0b1100, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rm,rGPR:$Rn), IIC_iMAC64, "smlsld", - "\t$Ra, $Rd, $Rm, $Rn", []>; + "\t$Ra, $Rd, $Rm, $Rn", []>, + Requires<[IsThumb2, HasThumb2DSP]>; def t2SMLSLDX : T2FourReg_mac<1, 0b101, 0b1101, (outs rGPR:$Ra,rGPR:$Rd), (ins rGPR:$Rm,rGPR:$Rn), IIC_iMAC64, "smlsldx", - "\t$Ra, $Rd, $Rm, $Rn", []>; + "\t$Ra, $Rd, $Rm, $Rn", []>, + Requires<[IsThumb2, HasThumb2DSP]>; + +//===----------------------------------------------------------------------===// +// Division Instructions. +// Signed and unsigned division on v7-M +// +def t2SDIV : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iALUi, + "sdiv", "\t$Rd, $Rn, $Rm", + [(set rGPR:$Rd, (sdiv rGPR:$Rn, rGPR:$Rm))]>, + Requires<[HasDivide, IsThumb2]> { + let Inst{31-27} = 0b11111; + let Inst{26-21} = 0b011100; + let Inst{20} = 0b1; + let Inst{15-12} = 0b1111; + let Inst{7-4} = 0b1111; +} + +def t2UDIV : T2ThreeReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm), IIC_iALUi, + "udiv", "\t$Rd, $Rn, $Rm", + [(set rGPR:$Rd, (udiv rGPR:$Rn, rGPR:$Rm))]>, + Requires<[HasDivide, IsThumb2]> { + let Inst{31-27} = 0b11111; + let Inst{26-21} = 0b011101; + let Inst{20} = 0b1; + let Inst{15-12} = 0b1111; + let Inst{7-4} = 0b1111; +} //===----------------------------------------------------------------------===// // Misc. Arithmetic Instructions. @@ -2585,25 +2602,16 @@ def t2REV : T2I_misc<0b01, 0b00, (outs rGPR:$Rd), (ins rGPR:$Rm), IIC_iUNAr, def t2REV16 : T2I_misc<0b01, 0b01, (outs rGPR:$Rd), (ins rGPR:$Rm), IIC_iUNAr, "rev16", ".w\t$Rd, $Rm", - [(set rGPR:$Rd, - (or (and (srl rGPR:$Rm, (i32 8)), 0xFF), - (or (and (shl rGPR:$Rm, (i32 8)), 0xFF00), - (or (and (srl rGPR:$Rm, (i32 8)), 0xFF0000), - (and (shl rGPR:$Rm, (i32 8)), 0xFF000000)))))]>; + [(set rGPR:$Rd, (rotr (bswap rGPR:$Rm), (i32 16)))]>; def t2REVSH : T2I_misc<0b01, 0b11, (outs rGPR:$Rd), (ins rGPR:$Rm), IIC_iUNAr, "revsh", ".w\t$Rd, $Rm", - [(set rGPR:$Rd, - (sext_inreg - (or (srl rGPR:$Rm, (i32 8)), - (shl rGPR:$Rm, (i32 8))), i16))]>; + [(set rGPR:$Rd, (sra (bswap rGPR:$Rm), (i32 16)))]>; -def : T2Pat<(sext_inreg (or (srl (and rGPR:$Rm, 0xFF00), (i32 8)), - (shl rGPR:$Rm, (i32 8))), i16), +def : T2Pat<(or (sra (shl rGPR:$Rm, (i32 24)), (i32 16)), + (and (srl rGPR:$Rm, (i32 8)), 0xFF)), (t2REVSH rGPR:$Rm)>; -def : T2Pat<(sra (bswap rGPR:$Rm), (i32 16)), (t2REVSH rGPR:$Rm)>; - def t2PKHBT : T2ThreeReg< (outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, shift_imm:$sh), IIC_iBITsi, "pkhbt", "\t$Rd, $Rn, $Rm$sh", @@ -2699,33 +2707,21 @@ defm t2TEQ : T2I_cmp_irs<0b0100, "teq", // FIXME: should be able to write a pattern for ARMcmov, but can't use // a two-value operand where a dag node expects two operands. :( let neverHasSideEffects = 1 in { -def t2MOVCCr : T2TwoReg< - (outs rGPR:$Rd), (ins rGPR:$false, rGPR:$Rm), IIC_iCMOVr, - "mov", ".w\t$Rd, $Rm", +def t2MOVCCr : t2PseudoInst<(outs rGPR:$Rd), + (ins rGPR:$false, rGPR:$Rm, pred:$p), + 4, IIC_iCMOVr, [/*(set rGPR:$Rd, (ARMcmov rGPR:$false, rGPR:$Rm, imm:$cc, CCR:$ccr))*/]>, - RegConstraint<"$false = $Rd"> { - let Inst{31-27} = 0b11101; - let Inst{26-25} = 0b01; - let Inst{24-21} = 0b0010; - let Inst{20} = 0; // The S bit. - let Inst{19-16} = 0b1111; // Rn - let Inst{14-12} = 0b000; - let Inst{7-4} = 0b0000; -} + RegConstraint<"$false = $Rd">; let isMoveImm = 1 in -def t2MOVCCi : T2OneRegImm<(outs rGPR:$Rd), (ins rGPR:$false, t2_so_imm:$imm), - IIC_iCMOVi, "mov", ".w\t$Rd, $imm", +def t2MOVCCi : t2PseudoInst<(outs rGPR:$Rd), + (ins rGPR:$false, t2_so_imm:$imm, pred:$p), + 4, IIC_iCMOVi, [/*(set rGPR:$Rd,(ARMcmov rGPR:$false,t2_so_imm:$imm, imm:$cc, CCR:$ccr))*/]>, - RegConstraint<"$false = $Rd"> { - let Inst{31-27} = 0b11110; - let Inst{25} = 0; - let Inst{24-21} = 0b0010; - let Inst{20} = 0; // The S bit. - let Inst{19-16} = 0b1111; // Rn - let Inst{15} = 0; -} + RegConstraint<"$false = $Rd">; +// FIXME: Pseudo-ize these. For now, just mark codegen only. +let isCodeGenOnly = 1 in { let isMoveImm = 1 in def t2MOVCCi16 : T2I<(outs rGPR:$Rd), (ins rGPR:$false, i32imm_hilo16:$imm), IIC_iCMOVi, @@ -2792,6 +2788,7 @@ def t2MOVCCror : T2I_movcc_sh<0b11, (outs rGPR:$Rd), (ins rGPR:$false, rGPR:$Rm, i32imm:$imm), IIC_iCMOVsi, "ror", ".w\t$Rd, $Rm, $imm", []>, RegConstraint<"$false = $Rd">; +} // isCodeGenOnly = 1 } // neverHasSideEffects //===----------------------------------------------------------------------===// @@ -2826,7 +2823,7 @@ def t2ISB : AInoP<(outs), (ins), ThumbFrm, NoItinerary, "isb", "", let Inst{3-0} = 0b1111; } -class T2I_ldrex<bits<2> opcod, dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class T2I_ldrex<bits<2> opcod, dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern, bits<4> rt2 = 0b1111> : Thumb2I<oops, iops, am, sz, itin, opc, asm, cstr, pattern> { @@ -2842,7 +2839,7 @@ class T2I_ldrex<bits<2> opcod, dag oops, dag iops, AddrMode am, SizeFlagVal sz, let Inst{19-16} = addr; let Inst{15-12} = Rt; } -class T2I_strex<bits<2> opcod, dag oops, dag iops, AddrMode am, SizeFlagVal sz, +class T2I_strex<bits<2> opcod, dag oops, dag iops, AddrMode am, int sz, InstrItinClass itin, string opc, string asm, string cstr, list<dag> pattern, bits<4> rt2 = 0b1111> : Thumb2I<oops, iops, am, sz, itin, opc, asm, cstr, pattern> { @@ -2861,16 +2858,15 @@ class T2I_strex<bits<2> opcod, dag oops, dag iops, AddrMode am, SizeFlagVal sz, } let mayLoad = 1 in { -def t2LDREXB : T2I_ldrex<0b00, (outs rGPR:$Rt), (ins t2addrmode_reg:$addr), AddrModeNone, - Size4Bytes, NoItinerary, "ldrexb", "\t$Rt, $addr", - "", []>; -def t2LDREXH : T2I_ldrex<0b01, (outs rGPR:$Rt), (ins t2addrmode_reg:$addr), AddrModeNone, - Size4Bytes, NoItinerary, "ldrexh", "\t$Rt, $addr", - "", []>; -def t2LDREX : Thumb2I<(outs rGPR:$Rt), (ins t2addrmode_reg:$addr), AddrModeNone, - Size4Bytes, NoItinerary, - "ldrex", "\t$Rt, $addr", "", - []> { +def t2LDREXB : T2I_ldrex<0b00, (outs rGPR:$Rt), (ins t2addrmode_reg:$addr), + AddrModeNone, 4, NoItinerary, + "ldrexb", "\t$Rt, $addr", "", []>; +def t2LDREXH : T2I_ldrex<0b01, (outs rGPR:$Rt), (ins t2addrmode_reg:$addr), + AddrModeNone, 4, NoItinerary, + "ldrexh", "\t$Rt, $addr", "", []>; +def t2LDREX : Thumb2I<(outs rGPR:$Rt), (ins t2addrmode_reg:$addr), + AddrModeNone, 4, NoItinerary, + "ldrex", "\t$Rt, $addr", "", []> { let Inst{31-27} = 0b11101; let Inst{26-20} = 0b0000101; let Inst{11-8} = 0b1111; @@ -2884,7 +2880,7 @@ def t2LDREX : Thumb2I<(outs rGPR:$Rt), (ins t2addrmode_reg:$addr), AddrModeNone let hasExtraDefRegAllocReq = 1 in def t2LDREXD : T2I_ldrex<0b11, (outs rGPR:$Rt, rGPR:$Rt2), (ins t2addrmode_reg:$addr), - AddrModeNone, Size4Bytes, NoItinerary, + AddrModeNone, 4, NoItinerary, "ldrexd", "\t$Rt, $Rt2, $addr", "", [], {?, ?, ?, ?}> { bits<4> Rt2; @@ -2893,14 +2889,16 @@ def t2LDREXD : T2I_ldrex<0b11, (outs rGPR:$Rt, rGPR:$Rt2), } let mayStore = 1, Constraints = "@earlyclobber $Rd" in { -def t2STREXB : T2I_strex<0b00, (outs rGPR:$Rd), (ins rGPR:$Rt, t2addrmode_reg:$addr), - AddrModeNone, Size4Bytes, NoItinerary, - "strexb", "\t$Rd, $Rt, $addr", "", []>; -def t2STREXH : T2I_strex<0b01, (outs rGPR:$Rd), (ins rGPR:$Rt, t2addrmode_reg:$addr), - AddrModeNone, Size4Bytes, NoItinerary, - "strexh", "\t$Rd, $Rt, $addr", "", []>; +def t2STREXB : T2I_strex<0b00, (outs rGPR:$Rd), + (ins rGPR:$Rt, t2addrmode_reg:$addr), + AddrModeNone, 4, NoItinerary, + "strexb", "\t$Rd, $Rt, $addr", "", []>; +def t2STREXH : T2I_strex<0b01, (outs rGPR:$Rd), + (ins rGPR:$Rt, t2addrmode_reg:$addr), + AddrModeNone, 4, NoItinerary, + "strexh", "\t$Rd, $Rt, $addr", "", []>; def t2STREX : Thumb2I<(outs rGPR:$Rd), (ins rGPR:$Rt, t2addrmode_reg:$addr), - AddrModeNone, Size4Bytes, NoItinerary, + AddrModeNone, 4, NoItinerary, "strex", "\t$Rd, $Rt, $addr", "", []> { let Inst{31-27} = 0b11101; @@ -2919,7 +2917,7 @@ def t2STREX : Thumb2I<(outs rGPR:$Rd), (ins rGPR:$Rt, t2addrmode_reg:$addr), let hasExtraSrcRegAllocReq = 1, Constraints = "@earlyclobber $Rd" in def t2STREXD : T2I_strex<0b11, (outs rGPR:$Rd), (ins rGPR:$Rt, rGPR:$Rt2, t2addrmode_reg:$addr), - AddrModeNone, Size4Bytes, NoItinerary, + AddrModeNone, 4, NoItinerary, "strexd", "\t$Rd, $Rt, $Rt2, $addr", "", [], {?, ?, ?, ?}> { bits<4> Rt2; @@ -2940,22 +2938,6 @@ def t2CLREX : T2XI<(outs), (ins), NoItinerary, "clrex", } //===----------------------------------------------------------------------===// -// TLS Instructions -// - -// __aeabi_read_tp preserves the registers r1-r3. -let isCall = 1, - Defs = [R0, R12, LR, CPSR], Uses = [SP] in { - def t2TPsoft : T2XI<(outs), (ins), IIC_Br, - "bl\t__aeabi_read_tp", - [(set R0, ARMthread_pointer)]> { - let Inst{31-27} = 0b11110; - let Inst{15-14} = 0b11; - let Inst{12} = 1; - } -} - -//===----------------------------------------------------------------------===// // SJLJ Exception handling intrinsics // eh_sjlj_setjmp() is an instruction sequence to store the return // address and save #0 in R0 for the non-longjmp case. @@ -2973,7 +2955,7 @@ let Defs = QQQQ0, QQQQ1, QQQQ2, QQQQ3 ], hasSideEffects = 1, isBarrier = 1, isCodeGenOnly = 1 in { def t2Int_eh_sjlj_setjmp : Thumb2XI<(outs), (ins tGPR:$src, tGPR:$val), - AddrModeNone, SizeSpecial, NoItinerary, "", "", + AddrModeNone, 0, NoItinerary, "", "", [(set R0, (ARMeh_sjlj_setjmp tGPR:$src, tGPR:$val))]>, Requires<[IsThumb2, HasVFP2]>; } @@ -2982,7 +2964,7 @@ let Defs = [ R0, R1, R2, R3, R4, R5, R6, R7, R8, R9, R10, R11, R12, LR, CPSR ], hasSideEffects = 1, isBarrier = 1, isCodeGenOnly = 1 in { def t2Int_eh_sjlj_setjmp_nofp : Thumb2XI<(outs), (ins tGPR:$src, tGPR:$val), - AddrModeNone, SizeSpecial, NoItinerary, "", "", + AddrModeNone, 0, NoItinerary, "", "", [(set R0, (ARMeh_sjlj_setjmp tGPR:$src, tGPR:$val))]>, Requires<[IsThumb2, NoVFP]>; } @@ -2993,28 +2975,14 @@ let Defs = // // FIXME: remove when we have a way to marking a MI with these properties. -// FIXME: $dst1 should be a def. But the extra ops must be in the end of the -// operand list. // FIXME: Should pc be an implicit operand like PICADD, etc? let isReturn = 1, isTerminator = 1, isBarrier = 1, mayLoad = 1, hasExtraDefRegAllocReq = 1, isCodeGenOnly = 1 in -def t2LDMIA_RET: T2XIt<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, - reglist:$regs, variable_ops), - IIC_iLoad_mBr, - "ldmia${p}.w\t$Rn!, $regs", - "$Rn = $wb", []> { - bits<4> Rn; - bits<16> regs; - - let Inst{31-27} = 0b11101; - let Inst{26-25} = 0b00; - let Inst{24-23} = 0b01; // Increment After - let Inst{22} = 0; - let Inst{21} = 1; // Writeback - let Inst{20} = 1; - let Inst{19-16} = Rn; - let Inst{15-0} = regs; -} +def t2LDMIA_RET: t2PseudoExpand<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, + reglist:$regs, variable_ops), + 4, IIC_iLoad_mBr, [], + (t2LDMIA_UPD GPR:$wb, GPR:$Rn, pred:$p, reglist:$regs)>, + RegConstraint<"$Rn = $wb">; let isBranch = 1, isTerminator = 1, isBarrier = 1 in { let isPredicable = 1 in @@ -3036,17 +3004,17 @@ def t2B : T2XI<(outs), (ins uncondbrtarget:$target), IIC_Br, let isNotDuplicable = 1, isIndirectBranch = 1 in { def t2BR_JT : t2PseudoInst<(outs), (ins GPR:$target, GPR:$index, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, + 0, IIC_Br, [(ARMbr2jt GPR:$target, GPR:$index, tjumptable:$jt, imm:$id)]>; // FIXME: Add a non-pc based case that can be predicated. def t2TBB_JT : t2PseudoInst<(outs), (ins GPR:$index, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, []>; + 0, IIC_Br, []>; def t2TBH_JT : t2PseudoInst<(outs), (ins GPR:$index, i32imm:$jt, i32imm:$id), - SizeSpecial, IIC_Br, []>; + 0, IIC_Br, []>; def t2TBB : T2I<(outs), (ins GPR:$Rn, GPR:$Rm), IIC_Br, "tbb", "\t[$Rn, $Rm]", []> { @@ -3094,11 +3062,22 @@ def t2Bcc : T2I<(outs), (ins brtarget:$target), IIC_Br, let Inst{10-0} = target{11-1}; } +// Tail calls. The Darwin version of thumb tail calls uses a t2 branch, so +// it goes here. +let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in { + // Darwin version. + let Defs = [R0, R1, R2, R3, R9, R12, QQQQ0, QQQQ2, QQQQ3, PC], + Uses = [SP] in + def tTAILJMPd: tPseudoExpand<(outs), (ins uncondbrtarget:$dst, variable_ops), + 4, IIC_Br, [], + (t2B uncondbrtarget:$dst)>, + Requires<[IsThumb2, IsDarwin]>; +} // IT block let Defs = [ITSTATE] in def t2IT : Thumb2XI<(outs), (ins it_pred:$cc, it_mask:$mask), - AddrModeNone, Size2Bytes, IIC_iALUx, + AddrModeNone, 2, IIC_iALUx, "it$mask\t$cc", "", []> { // 16-bit instruction. let Inst{31-16} = 0x0000; @@ -3178,8 +3157,7 @@ def t2WFE : T2I_hint<0b00000010, "wfe", ".w">; def t2WFI : T2I_hint<0b00000011, "wfi", ".w">; def t2SEV : T2I_hint<0b00000100, "sev", ".w">; -def t2DBG : T2I<(outs),(ins i32imm:$opt), NoItinerary, "dbg", "\t$opt", - [/* For disassembly only; pattern left blank */]> { +def t2DBG : T2I<(outs), (ins imm0_15:$opt), NoItinerary, "dbg", "\t$opt", []> { let Inst{31-20} = 0xf3a; let Inst{15-14} = 0b10; let Inst{12} = 0; @@ -3347,12 +3325,13 @@ def t2MSR : T2SpecialReg<0b111100111000 /* op31-20 */, 0b10 /* op15-14 */, } //===----------------------------------------------------------------------===// -// Move between coprocessor and ARM core register -- for disassembly only +// Move between coprocessor and ARM core register // -class t2MovRCopro<string opc, bit direction, dag oops, dag iops, +class t2MovRCopro<bits<4> Op, string opc, bit direction, dag oops, dag iops, list<dag> pattern> - : T2Cop<oops, iops, !strconcat(opc, "\t$cop, $opc1, $Rt, $CRn, $CRm, $opc2"), + : T2Cop<Op, oops, iops, + !strconcat(opc, "\t$cop, $opc1, $Rt, $CRn, $CRm, $opc2"), pattern> { let Inst{27-24} = 0b1110; let Inst{20} = direction; @@ -3373,22 +3352,10 @@ class t2MovRCopro<string opc, bit direction, dag oops, dag iops, let Inst{19-16} = CRn; } -def t2MCR2 : t2MovRCopro<"mcr2", 0 /* from ARM core register to coprocessor */, - (outs), (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, c_imm:$CRn, - c_imm:$CRm, i32imm:$opc2), - [(int_arm_mcr2 imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, - imm:$CRm, imm:$opc2)]>; -def t2MRC2 : t2MovRCopro<"mrc2", 1 /* from coprocessor to ARM core register */, - (outs GPR:$Rt), (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, - c_imm:$CRm, i32imm:$opc2), []>; - -def : T2v6Pat<(int_arm_mrc2 imm:$cop, imm:$opc1, imm:$CRn, - imm:$CRm, imm:$opc2), - (t2MRC2 imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>; - -class t2MovRRCopro<string opc, bit direction, - list<dag> pattern = [/* For disassembly only */]> - : T2Cop<(outs), (ins p_imm:$cop, i32imm:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), +class t2MovRRCopro<bits<4> Op, string opc, bit direction, + list<dag> pattern = []> + : T2Cop<Op, (outs), + (ins p_imm:$cop, imm0_15:$opc1, GPR:$Rt, GPR:$Rt2, c_imm:$CRm), !strconcat(opc, "\t$cop, $opc1, $Rt, $Rt2, $CRm"), pattern> { let Inst{27-24} = 0b1100; let Inst{23-21} = 0b010; @@ -3407,19 +3374,77 @@ class t2MovRRCopro<string opc, bit direction, let Inst{3-0} = CRm; } -def t2MCRR2 : t2MovRRCopro<"mcrr2", - 0 /* from ARM core register to coprocessor */, +/* from ARM core register to coprocessor */ +def t2MCR : t2MovRCopro<0b1110, "mcr", 0, + (outs), + (ins p_imm:$cop, imm0_7:$opc1, GPR:$Rt, c_imm:$CRn, + c_imm:$CRm, imm0_7:$opc2), + [(int_arm_mcr imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, + imm:$CRm, imm:$opc2)]>; +def t2MCR2 : t2MovRCopro<0b1111, "mcr2", 0, + (outs), (ins p_imm:$cop, imm0_7:$opc1, GPR:$Rt, c_imm:$CRn, + c_imm:$CRm, imm0_7:$opc2), + [(int_arm_mcr2 imm:$cop, imm:$opc1, GPR:$Rt, imm:$CRn, + imm:$CRm, imm:$opc2)]>; + +/* from coprocessor to ARM core register */ +def t2MRC : t2MovRCopro<0b1110, "mrc", 1, + (outs GPR:$Rt), + (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), + []>; + +def t2MRC2 : t2MovRCopro<0b1111, "mrc2", 1, + (outs GPR:$Rt), (ins p_imm:$cop, i32imm:$opc1, c_imm:$CRn, + c_imm:$CRm, i32imm:$opc2), []>; + +def : T2v6Pat<(int_arm_mrc imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2), + (t2MRC imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>; + +def : T2v6Pat<(int_arm_mrc2 imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2), + (t2MRC2 imm:$cop, imm:$opc1, imm:$CRn, imm:$CRm, imm:$opc2)>; + + +/* from ARM core register to coprocessor */ +def t2MCRR : t2MovRRCopro<0b1110, "mcrr", 0, + [(int_arm_mcrr imm:$cop, imm:$opc1, GPR:$Rt, GPR:$Rt2, + imm:$CRm)]>; +def t2MCRR2 : t2MovRRCopro<0b1111, "mcrr2", 0, [(int_arm_mcrr2 imm:$cop, imm:$opc1, GPR:$Rt, GPR:$Rt2, imm:$CRm)]>; -def t2MRRC2 : t2MovRRCopro<"mrrc2", - 1 /* from coprocessor to ARM core register */>; +/* from coprocessor to ARM core register */ +def t2MRRC : t2MovRRCopro<0b1110, "mrrc", 1>; + +def t2MRRC2 : t2MovRRCopro<0b1111, "mrrc2", 1>; //===----------------------------------------------------------------------===// -// Other Coprocessor Instructions. For disassembly only. +// Other Coprocessor Instructions. // -def t2CDP2 : T2Cop<(outs), (ins p_imm:$cop, i32imm:$opc1, - c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, i32imm:$opc2), +def tCDP : T2Cop<0b1110, (outs), (ins p_imm:$cop, imm0_15:$opc1, + c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, imm0_7:$opc2), + "cdp\t$cop, $opc1, $CRd, $CRn, $CRm, $opc2", + [(int_arm_cdp imm:$cop, imm:$opc1, imm:$CRd, imm:$CRn, + imm:$CRm, imm:$opc2)]> { + let Inst{27-24} = 0b1110; + + bits<4> opc1; + bits<4> CRn; + bits<4> CRd; + bits<4> cop; + bits<3> opc2; + bits<4> CRm; + + let Inst{3-0} = CRm; + let Inst{4} = 0; + let Inst{7-5} = opc2; + let Inst{11-8} = cop; + let Inst{15-12} = CRd; + let Inst{19-16} = CRn; + let Inst{23-20} = opc1; +} + +def t2CDP2 : T2Cop<0b1111, (outs), (ins p_imm:$cop, imm0_15:$opc1, + c_imm:$CRd, c_imm:$CRn, c_imm:$CRm, imm0_7:$opc2), "cdp2\t$cop, $opc1, $CRd, $CRn, $CRm, $opc2", [(int_arm_cdp2 imm:$cop, imm:$opc1, imm:$CRd, imm:$CRn, imm:$CRm, imm:$opc2)]> { diff --git a/lib/Target/ARM/ARMInstrVFP.td b/lib/Target/ARM/ARMInstrVFP.td index 376bd96..f1f3cb9 100644 --- a/lib/Target/ARM/ARMInstrVFP.td +++ b/lib/Target/ARM/ARMInstrVFP.td @@ -94,7 +94,8 @@ multiclass vfp_ldst_mult<string asm, bit L_bit, let Inst{20} = L_bit; } def DIA_UPD : - AXDI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops), + AXDI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, + variable_ops), IndexModeUpd, itin_upd, !strconcat(asm, "ia${p}\t$Rn!, $regs"), "$Rn = $wb", []> { let Inst{24-23} = 0b01; // Increment After @@ -102,7 +103,8 @@ multiclass vfp_ldst_mult<string asm, bit L_bit, let Inst{20} = L_bit; } def DDB_UPD : - AXDI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, variable_ops), + AXDI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, dpr_reglist:$regs, + variable_ops), IndexModeUpd, itin_upd, !strconcat(asm, "db${p}\t$Rn!, $regs"), "$Rn = $wb", []> { let Inst{24-23} = 0b10; // Decrement Before @@ -124,7 +126,8 @@ multiclass vfp_ldst_mult<string asm, bit L_bit, let D = VFPNeonDomain; } def SIA_UPD : - AXSI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, spr_reglist:$regs, variable_ops), + AXSI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, spr_reglist:$regs, + variable_ops), IndexModeUpd, itin_upd, !strconcat(asm, "ia${p}\t$Rn!, $regs"), "$Rn = $wb", []> { let Inst{24-23} = 0b01; // Increment After @@ -136,7 +139,8 @@ multiclass vfp_ldst_mult<string asm, bit L_bit, let D = VFPNeonDomain; } def SDB_UPD : - AXSI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, spr_reglist:$regs, variable_ops), + AXSI4<(outs GPR:$wb), (ins GPR:$Rn, pred:$p, spr_reglist:$regs, + variable_ops), IndexModeUpd, itin_upd, !strconcat(asm, "db${p}\t$Rn!, $regs"), "$Rn = $wb", []> { let Inst{24-23} = 0b10; // Decrement Before @@ -162,6 +166,15 @@ defm VSTM : vfp_ldst_mult<"vstm", 0, IIC_fpLoad_m, IIC_fpLoad_mu>; def : MnemonicAlias<"vldm", "vldmia">; def : MnemonicAlias<"vstm", "vstmia">; +def : InstAlias<"vpush${p} $r", (VSTMDDB_UPD SP, pred:$p, dpr_reglist:$r)>, + Requires<[HasVFP2]>; +def : InstAlias<"vpush${p} $r", (VSTMSDB_UPD SP, pred:$p, spr_reglist:$r)>, + Requires<[HasVFP2]>; +def : InstAlias<"vpop${p} $r", (VLDMDIA_UPD SP, pred:$p, dpr_reglist:$r)>, + Requires<[HasVFP2]>; +def : InstAlias<"vpop${p} $r", (VLDMSIA_UPD SP, pred:$p, spr_reglist:$r)>, + Requires<[HasVFP2]>; + // FLDMX, FSTMX - mixing S/D registers for pre-armv6 cores //===----------------------------------------------------------------------===// @@ -860,7 +873,7 @@ def VULTOD : AVConv1XI<0b11101, 0b11, 0b1011, 0b1011, 1, } // End of 'let Constraints = "$a = $dst", isCodeGenOnly = 1 in' //===----------------------------------------------------------------------===// -// FP FMA Operations. +// FP Multiply-Accumulate Operations. // def VMLAD : ADbI<0b11100, 0b00, 0, 0, @@ -977,12 +990,12 @@ def : Pat<(fsub_mlx (fmul_su SPR:$a, SPR:$b), SPR:$dstin), let neverHasSideEffects = 1 in { def VMOVDcc : ARMPseudoInst<(outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm, pred:$p), - Size4Bytes, IIC_fpUNA64, + 4, IIC_fpUNA64, [/*(set DPR:$Dd, (ARMcmov DPR:$Dn, DPR:$Dm, imm:$cc))*/]>, RegConstraint<"$Dn = $Dd">; def VMOVScc : ARMPseudoInst<(outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm, pred:$p), - Size4Bytes, IIC_fpUNA32, + 4, IIC_fpUNA32, [/*(set SPR:$Sd, (ARMcmov SPR:$Sn, SPR:$Sm, imm:$cc))*/]>, RegConstraint<"$Sn = $Sd">; } // neverHasSideEffects diff --git a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp index f4645f1..c6efea1 100644 --- a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp +++ b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp @@ -329,13 +329,9 @@ ARMLoadStoreOpt::MergeOps(MachineBasicBlock &MBB, if (NewBase == 0) return false; } - int BaseOpc = !isThumb2 - ? ARM::ADDri - : ((Base == ARM::SP) ? ARM::t2ADDrSPi : ARM::t2ADDri); + int BaseOpc = !isThumb2 ? ARM::ADDri : ARM::t2ADDri; if (Offset < 0) { - BaseOpc = !isThumb2 - ? ARM::SUBri - : ((Base == ARM::SP) ? ARM::t2SUBrSPi : ARM::t2SUBri); + BaseOpc = !isThumb2 ? ARM::SUBri : ARM::t2SUBri; Offset = - Offset; } int ImmedOffset = isThumb2 @@ -516,8 +512,6 @@ static inline bool isMatchingDecrement(MachineInstr *MI, unsigned Base, if (!MI) return false; if (MI->getOpcode() != ARM::t2SUBri && - MI->getOpcode() != ARM::t2SUBrSPi && - MI->getOpcode() != ARM::t2SUBrSPi12 && MI->getOpcode() != ARM::tSUBspi && MI->getOpcode() != ARM::SUBri) return false; @@ -541,8 +535,6 @@ static inline bool isMatchingIncrement(MachineInstr *MI, unsigned Base, if (!MI) return false; if (MI->getOpcode() != ARM::t2ADDri && - MI->getOpcode() != ARM::t2ADDrSPi && - MI->getOpcode() != ARM::t2ADDrSPi12 && MI->getOpcode() != ARM::tADDspi && MI->getOpcode() != ARM::ADDri) return false; @@ -1461,19 +1453,19 @@ static bool IsSafeAndProfitableToMove(bool isLd, unsigned Base, while (++I != E) { if (I->isDebugValue() || MemOps.count(&*I)) continue; - const TargetInstrDesc &TID = I->getDesc(); - if (TID.isCall() || TID.isTerminator() || I->hasUnmodeledSideEffects()) + const MCInstrDesc &MCID = I->getDesc(); + if (MCID.isCall() || MCID.isTerminator() || I->hasUnmodeledSideEffects()) return false; - if (isLd && TID.mayStore()) + if (isLd && MCID.mayStore()) return false; if (!isLd) { - if (TID.mayLoad()) + if (MCID.mayLoad()) return false; // It's not safe to move the first 'str' down. // str r1, [r0] // strh r5, [r0] // str r4, [r0, #+4] - if (TID.mayStore()) + if (MCID.mayStore()) return false; } for (unsigned j = 0, NumOps = I->getNumOperands(); j != NumOps; ++j) { @@ -1672,14 +1664,14 @@ bool ARMPreAllocLoadStoreOpt::RescheduleOps(MachineBasicBlock *MBB, Ops.pop_back(); Ops.pop_back(); - const TargetInstrDesc &TID = TII->get(NewOpc); - const TargetRegisterClass *TRC = TID.OpInfo[0].getRegClass(TRI); + const MCInstrDesc &MCID = TII->get(NewOpc); + const TargetRegisterClass *TRC = TII->getRegClass(MCID, 0, TRI); MRI->constrainRegClass(EvenReg, TRC); MRI->constrainRegClass(OddReg, TRC); // Form the pair instruction. if (isLd) { - MachineInstrBuilder MIB = BuildMI(*MBB, InsertPos, dl, TID) + MachineInstrBuilder MIB = BuildMI(*MBB, InsertPos, dl, MCID) .addReg(EvenReg, RegState::Define) .addReg(OddReg, RegState::Define) .addReg(BaseReg); @@ -1691,7 +1683,7 @@ bool ARMPreAllocLoadStoreOpt::RescheduleOps(MachineBasicBlock *MBB, MIB.addImm(Offset).addImm(Pred).addReg(PredReg); ++NumLDRDFormed; } else { - MachineInstrBuilder MIB = BuildMI(*MBB, InsertPos, dl, TID) + MachineInstrBuilder MIB = BuildMI(*MBB, InsertPos, dl, MCID) .addReg(EvenReg) .addReg(OddReg) .addReg(BaseReg); @@ -1742,8 +1734,8 @@ ARMPreAllocLoadStoreOpt::RescheduleLoadStoreInstrs(MachineBasicBlock *MBB) { while (MBBI != E) { for (; MBBI != E; ++MBBI) { MachineInstr *MI = MBBI; - const TargetInstrDesc &TID = MI->getDesc(); - if (TID.isCall() || TID.isTerminator()) { + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.isCall() || MCID.isTerminator()) { // Stop at barriers. ++MBBI; break; diff --git a/lib/Target/ARM/ARMMCCodeEmitter.cpp b/lib/Target/ARM/ARMMCCodeEmitter.cpp index c5f727d..39be3f0 100644 --- a/lib/Target/ARM/ARMMCCodeEmitter.cpp +++ b/lib/Target/ARM/ARMMCCodeEmitter.cpp @@ -21,8 +21,11 @@ #include "llvm/MC/MCCodeEmitter.h" #include "llvm/MC/MCExpr.h" #include "llvm/MC/MCInst.h" +#include "llvm/MC/MCInstrInfo.h" +#include "llvm/MC/MCSubtargetInfo.h" #include "llvm/ADT/Statistic.h" #include "llvm/Support/raw_ostream.h" + using namespace llvm; STATISTIC(MCNumEmitted, "Number of MC instructions emitted."); @@ -32,19 +35,30 @@ namespace { class ARMMCCodeEmitter : public MCCodeEmitter { ARMMCCodeEmitter(const ARMMCCodeEmitter &); // DO NOT IMPLEMENT void operator=(const ARMMCCodeEmitter &); // DO NOT IMPLEMENT - const TargetMachine &TM; - const TargetInstrInfo &TII; - const ARMSubtarget *Subtarget; - MCContext &Ctx; + const MCInstrInfo &MCII; + const MCSubtargetInfo &STI; public: - ARMMCCodeEmitter(TargetMachine &tm, MCContext &ctx) - : TM(tm), TII(*TM.getInstrInfo()), - Subtarget(&TM.getSubtarget<ARMSubtarget>()), Ctx(ctx) { + ARMMCCodeEmitter(const MCInstrInfo &mcii, const MCSubtargetInfo &sti, + MCContext &ctx) + : MCII(mcii), STI(sti) { } ~ARMMCCodeEmitter() {} + bool isThumb() const { + // FIXME: Can tablegen auto-generate this? + return (STI.getFeatureBits() & ARM::ModeThumb) != 0; + } + bool isThumb2() const { + return isThumb() && (STI.getFeatureBits() & ARM::FeatureThumb2) != 0; + } + bool isTargetDarwin() const { + Triple TT(STI.getTargetTriple()); + Triple::OSType OS = TT.getOS(); + return OS == Triple::Darwin || OS == Triple::MacOSX || OS == Triple::IOS; + } + unsigned getMachineSoImmOpValue(unsigned SoImm) const; // getBinaryCodeForInstr - TableGen'erated function for getting the @@ -320,9 +334,10 @@ public: } // end anonymous namespace -MCCodeEmitter *llvm::createARMMCCodeEmitter(const Target &, TargetMachine &TM, +MCCodeEmitter *llvm::createARMMCCodeEmitter(const MCInstrInfo &MCII, + const MCSubtargetInfo &STI, MCContext &Ctx) { - return new ARMMCCodeEmitter(TM, Ctx); + return new ARMMCCodeEmitter(MCII, STI, Ctx); } /// NEONThumb2DataIPostEncoder - Post-process encoded NEON data-processing @@ -330,7 +345,7 @@ MCCodeEmitter *llvm::createARMMCCodeEmitter(const Target &, TargetMachine &TM, /// Thumb2 mode. unsigned ARMMCCodeEmitter::NEONThumb2DataIPostEncoder(const MCInst &MI, unsigned EncodedValue) const { - if (Subtarget->isThumb2()) { + if (isThumb2()) { // NEON Thumb2 data-processsing encodings are very simple: bit 24 is moved // to bit 12 of the high half-word (i.e. bit 28), and bits 27-24 are // set to 1111. @@ -349,7 +364,7 @@ unsigned ARMMCCodeEmitter::NEONThumb2DataIPostEncoder(const MCInst &MI, /// Thumb2 mode. unsigned ARMMCCodeEmitter::NEONThumb2LoadStorePostEncoder(const MCInst &MI, unsigned EncodedValue) const { - if (Subtarget->isThumb2()) { + if (isThumb2()) { EncodedValue &= 0xF0FFFFFF; EncodedValue |= 0x09000000; } @@ -362,7 +377,7 @@ unsigned ARMMCCodeEmitter::NEONThumb2LoadStorePostEncoder(const MCInst &MI, /// Thumb2 mode. unsigned ARMMCCodeEmitter::NEONThumb2DupPostEncoder(const MCInst &MI, unsigned EncodedValue) const { - if (Subtarget->isThumb2()) { + if (isThumb2()) { EncodedValue &= 0x00FFFFFF; EncodedValue |= 0xEE000000; } @@ -374,7 +389,7 @@ unsigned ARMMCCodeEmitter::NEONThumb2DupPostEncoder(const MCInst &MI, /// them to their Thumb2 form if we are currently in Thumb2 mode. unsigned ARMMCCodeEmitter:: VFPThumb2PostEncoder(const MCInst &MI, unsigned EncodedValue) const { - if (Subtarget->isThumb2()) { + if (isThumb2()) { EncodedValue &= 0x0FFFFFFF; EncodedValue |= 0xE0000000; } @@ -515,7 +530,7 @@ getBranchTargetOpValue(const MCInst &MI, unsigned OpIdx, SmallVectorImpl<MCFixup> &Fixups) const { // FIXME: This really, really shouldn't use TargetMachine. We don't want // coupling between MC and TM anywhere we can help it. - if (Subtarget->isThumb2()) + if (isThumb2()) return ::getBranchTargetOpValue(MI, OpIdx, ARM::fixup_t2_condbranch, Fixups); return getARMBranchTargetOpValue(MI, OpIdx, Fixups); @@ -624,7 +639,7 @@ getAddrModeImm12OpValue(const MCInst &MI, unsigned OpIdx, const MCExpr *Expr = MO.getExpr(); MCFixupKind Kind; - if (Subtarget->isThumb2()) + if (isThumb2()) Kind = MCFixupKind(ARM::fixup_t2_ldst_pcrel_12); else Kind = MCFixupKind(ARM::fixup_arm_ldst_pcrel_12); @@ -709,22 +724,22 @@ ARMMCCodeEmitter::getHiLo16ImmOpValue(const MCInst &MI, unsigned OpIdx, switch (ARM16Expr->getKind()) { default: assert(0 && "Unsupported ARMFixup"); case ARMMCExpr::VK_ARM_HI16: - if (!Subtarget->isTargetDarwin() && EvaluateAsPCRel(E)) - Kind = MCFixupKind(Subtarget->isThumb2() + if (!isTargetDarwin() && EvaluateAsPCRel(E)) + Kind = MCFixupKind(isThumb2() ? ARM::fixup_t2_movt_hi16_pcrel : ARM::fixup_arm_movt_hi16_pcrel); else - Kind = MCFixupKind(Subtarget->isThumb2() + Kind = MCFixupKind(isThumb2() ? ARM::fixup_t2_movt_hi16 : ARM::fixup_arm_movt_hi16); break; case ARMMCExpr::VK_ARM_LO16: - if (!Subtarget->isTargetDarwin() && EvaluateAsPCRel(E)) - Kind = MCFixupKind(Subtarget->isThumb2() + if (!isTargetDarwin() && EvaluateAsPCRel(E)) + Kind = MCFixupKind(isThumb2() ? ARM::fixup_t2_movw_lo16_pcrel : ARM::fixup_arm_movw_lo16_pcrel); else - Kind = MCFixupKind(Subtarget->isThumb2() + Kind = MCFixupKind(isThumb2() ? ARM::fixup_t2_movw_lo16 : ARM::fixup_arm_movw_lo16); break; @@ -898,7 +913,7 @@ getAddrMode5OpValue(const MCInst &MI, unsigned OpIdx, assert(MO.isExpr() && "Unexpected machine operand type!"); const MCExpr *Expr = MO.getExpr(); MCFixupKind Kind; - if (Subtarget->isThumb2()) + if (isThumb2()) Kind = MCFixupKind(ARM::fixup_t2_pcrel_10); else Kind = MCFixupKind(ARM::fixup_arm_pcrel_10); @@ -1274,21 +1289,21 @@ void ARMMCCodeEmitter:: EncodeInstruction(const MCInst &MI, raw_ostream &OS, SmallVectorImpl<MCFixup> &Fixups) const { // Pseudo instructions don't get encoded. - const TargetInstrDesc &Desc = TII.get(MI.getOpcode()); + const MCInstrDesc &Desc = MCII.get(MI.getOpcode()); uint64_t TSFlags = Desc.TSFlags; if ((TSFlags & ARMII::FormMask) == ARMII::Pseudo) return; + int Size; - // Basic size info comes from the TSFlags field. - switch ((TSFlags & ARMII::SizeMask) >> ARMII::SizeShift) { - default: llvm_unreachable("Unexpected instruction size!"); - case ARMII::Size2Bytes: Size = 2; break; - case ARMII::Size4Bytes: Size = 4; break; - } + if (Desc.getSize() == 2 || Desc.getSize() == 4) + Size = Desc.getSize(); + else + llvm_unreachable("Unexpected instruction size!"); + uint32_t Binary = getBinaryCodeForInstr(MI, Fixups); // Thumb 32-bit wide instructions need to emit the high order halfword // first. - if (Subtarget->isThumb() && Size == 4) { + if (isThumb() && Size == 4) { EmitConstant(Binary >> 16, 2, OS); EmitConstant(Binary & 0xffff, 2, OS); } else diff --git a/lib/Target/ARM/ARMMCInstLower.cpp b/lib/Target/ARM/ARMMCInstLower.cpp index 59d6050..7411b59 100644 --- a/lib/Target/ARM/ARMMCInstLower.cpp +++ b/lib/Target/ARM/ARMMCInstLower.cpp @@ -23,43 +23,94 @@ using namespace llvm; -static MCOperand GetSymbolRef(const MachineOperand &MO, const MCSymbol *Symbol, - ARMAsmPrinter &Printer) { - MCContext &Ctx = Printer.OutContext; +MCOperand ARMAsmPrinter::GetSymbolRef(const MachineOperand &MO, + const MCSymbol *Symbol) { const MCExpr *Expr; switch (MO.getTargetFlags()) { default: { - Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, Ctx); + Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, + OutContext); switch (MO.getTargetFlags()) { default: assert(0 && "Unknown target flag on symbol operand"); case 0: break; case ARMII::MO_LO16: - Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, Ctx); - Expr = ARMMCExpr::CreateLower16(Expr, Ctx); + Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, + OutContext); + Expr = ARMMCExpr::CreateLower16(Expr, OutContext); break; case ARMII::MO_HI16: - Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, Ctx); - Expr = ARMMCExpr::CreateUpper16(Expr, Ctx); + Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None, + OutContext); + Expr = ARMMCExpr::CreateUpper16(Expr, OutContext); break; } break; } case ARMII::MO_PLT: - Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_ARM_PLT, Ctx); + Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_ARM_PLT, + OutContext); break; } if (!MO.isJTI() && MO.getOffset()) Expr = MCBinaryExpr::CreateAdd(Expr, - MCConstantExpr::Create(MO.getOffset(), Ctx), - Ctx); + MCConstantExpr::Create(MO.getOffset(), + OutContext), + OutContext); return MCOperand::CreateExpr(Expr); } +bool ARMAsmPrinter::lowerOperand(const MachineOperand &MO, + MCOperand &MCOp) { + switch (MO.getType()) { + default: + assert(0 && "unknown operand type"); + return false; + case MachineOperand::MO_Register: + // Ignore all non-CPSR implicit register operands. + if (MO.isImplicit() && MO.getReg() != ARM::CPSR) + return false; + assert(!MO.getSubReg() && "Subregs should be eliminated!"); + MCOp = MCOperand::CreateReg(MO.getReg()); + break; + case MachineOperand::MO_Immediate: + MCOp = MCOperand::CreateImm(MO.getImm()); + break; + case MachineOperand::MO_MachineBasicBlock: + MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create( + MO.getMBB()->getSymbol(), OutContext)); + break; + case MachineOperand::MO_GlobalAddress: + MCOp = GetSymbolRef(MO, Mang->getSymbol(MO.getGlobal())); + break; + case MachineOperand::MO_ExternalSymbol: + MCOp = GetSymbolRef(MO, + GetExternalSymbolSymbol(MO.getSymbolName())); + break; + case MachineOperand::MO_JumpTableIndex: + MCOp = GetSymbolRef(MO, GetJTISymbol(MO.getIndex())); + break; + case MachineOperand::MO_ConstantPoolIndex: + MCOp = GetSymbolRef(MO, GetCPISymbol(MO.getIndex())); + break; + case MachineOperand::MO_BlockAddress: + MCOp = GetSymbolRef(MO, GetBlockAddressSymbol(MO.getBlockAddress())); + break; + case MachineOperand::MO_FPImmediate: { + APFloat Val = MO.getFPImm()->getValueAPF(); + bool ignored; + Val.convert(APFloat::IEEEdouble, APFloat::rmTowardZero, &ignored); + MCOp = MCOperand::CreateFPImm(Val.convertToDouble()); + break; + } + } + return true; +} + void llvm::LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI, ARMAsmPrinter &AP) { OutMI.setOpcode(MI->getOpcode()); @@ -68,48 +119,7 @@ void llvm::LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI, const MachineOperand &MO = MI->getOperand(i); MCOperand MCOp; - switch (MO.getType()) { - default: - MI->dump(); - assert(0 && "unknown operand type"); - case MachineOperand::MO_Register: - // Ignore all non-CPSR implicit register operands. - if (MO.isImplicit() && MO.getReg() != ARM::CPSR) continue; - assert(!MO.getSubReg() && "Subregs should be eliminated!"); - MCOp = MCOperand::CreateReg(MO.getReg()); - break; - case MachineOperand::MO_Immediate: - MCOp = MCOperand::CreateImm(MO.getImm()); - break; - case MachineOperand::MO_MachineBasicBlock: - MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create( - MO.getMBB()->getSymbol(), AP.OutContext)); - break; - case MachineOperand::MO_GlobalAddress: - MCOp = GetSymbolRef(MO, AP.Mang->getSymbol(MO.getGlobal()), AP); - break; - case MachineOperand::MO_ExternalSymbol: - MCOp = GetSymbolRef(MO, - AP.GetExternalSymbolSymbol(MO.getSymbolName()), AP); - break; - case MachineOperand::MO_JumpTableIndex: - MCOp = GetSymbolRef(MO, AP.GetJTISymbol(MO.getIndex()), AP); - break; - case MachineOperand::MO_ConstantPoolIndex: - MCOp = GetSymbolRef(MO, AP.GetCPISymbol(MO.getIndex()), AP); - break; - case MachineOperand::MO_BlockAddress: - MCOp = GetSymbolRef(MO,AP.GetBlockAddressSymbol(MO.getBlockAddress()),AP); - break; - case MachineOperand::MO_FPImmediate: { - APFloat Val = MO.getFPImm()->getValueAPF(); - bool ignored; - Val.convert(APFloat::IEEEdouble, APFloat::rmTowardZero, &ignored); - MCOp = MCOperand::CreateFPImm(Val.convertToDouble()); - break; - } - } - - OutMI.addOperand(MCOp); + if (AP.lowerOperand(MO, MCOp)) + OutMI.addOperand(MCOp); } } diff --git a/lib/Target/ARM/ARMMachObjectWriter.cpp b/lib/Target/ARM/ARMMachObjectWriter.cpp new file mode 100644 index 0000000..a36e47d --- /dev/null +++ b/lib/Target/ARM/ARMMachObjectWriter.cpp @@ -0,0 +1,389 @@ +//===-- ARMMachObjectWriter.cpp - ARM Mach Object Writer ------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "ARM.h" +#include "ARMFixupKinds.h" +#include "llvm/ADT/Twine.h" +#include "llvm/MC/MCAssembler.h" +#include "llvm/MC/MCAsmLayout.h" +#include "llvm/MC/MCMachObjectWriter.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCFixup.h" +#include "llvm/MC/MCFixupKindInfo.h" +#include "llvm/MC/MCValue.h" +#include "llvm/Object/MachOFormat.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Target/TargetAsmBackend.h" +using namespace llvm; +using namespace llvm::object; + +namespace { +class ARMMachObjectWriter : public MCMachObjectTargetWriter { + void RecordARMScatteredRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, + const MCAsmLayout &Layout, + const MCFragment *Fragment, + const MCFixup &Fixup, + MCValue Target, + unsigned Log2Size, + uint64_t &FixedValue); + void RecordARMMovwMovtRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, + const MCAsmLayout &Layout, + const MCFragment *Fragment, + const MCFixup &Fixup, MCValue Target, + uint64_t &FixedValue); + +public: + ARMMachObjectWriter(bool Is64Bit, uint32_t CPUType, + uint32_t CPUSubtype) + : MCMachObjectTargetWriter(Is64Bit, CPUType, CPUSubtype, + /*UseAggressiveSymbolFolding=*/true) {} + + void RecordRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, const MCAsmLayout &Layout, + const MCFragment *Fragment, const MCFixup &Fixup, + MCValue Target, uint64_t &FixedValue); +}; +} + +static bool getARMFixupKindMachOInfo(unsigned Kind, unsigned &RelocType, + unsigned &Log2Size) { + RelocType = unsigned(macho::RIT_Vanilla); + Log2Size = ~0U; + + switch (Kind) { + default: + return false; + + case FK_Data_1: + Log2Size = llvm::Log2_32(1); + return true; + case FK_Data_2: + Log2Size = llvm::Log2_32(2); + return true; + case FK_Data_4: + Log2Size = llvm::Log2_32(4); + return true; + case FK_Data_8: + Log2Size = llvm::Log2_32(8); + return true; + + // Handle 24-bit branch kinds. + case ARM::fixup_arm_ldst_pcrel_12: + case ARM::fixup_arm_pcrel_10: + case ARM::fixup_arm_adr_pcrel_12: + case ARM::fixup_arm_condbranch: + case ARM::fixup_arm_uncondbranch: + RelocType = unsigned(macho::RIT_ARM_Branch24Bit); + // Report as 'long', even though that is not quite accurate. + Log2Size = llvm::Log2_32(4); + return true; + + // Handle Thumb branches. + case ARM::fixup_arm_thumb_br: + RelocType = unsigned(macho::RIT_ARM_ThumbBranch22Bit); + Log2Size = llvm::Log2_32(2); + return true; + + case ARM::fixup_t2_uncondbranch: + case ARM::fixup_arm_thumb_bl: + case ARM::fixup_arm_thumb_blx: + RelocType = unsigned(macho::RIT_ARM_ThumbBranch22Bit); + Log2Size = llvm::Log2_32(4); + return true; + + case ARM::fixup_arm_movt_hi16: + case ARM::fixup_arm_movt_hi16_pcrel: + case ARM::fixup_t2_movt_hi16: + case ARM::fixup_t2_movt_hi16_pcrel: + RelocType = unsigned(macho::RIT_ARM_HalfDifference); + // Report as 'long', even though that is not quite accurate. + Log2Size = llvm::Log2_32(4); + return true; + + case ARM::fixup_arm_movw_lo16: + case ARM::fixup_arm_movw_lo16_pcrel: + case ARM::fixup_t2_movw_lo16: + case ARM::fixup_t2_movw_lo16_pcrel: + RelocType = unsigned(macho::RIT_ARM_Half); + // Report as 'long', even though that is not quite accurate. + Log2Size = llvm::Log2_32(4); + return true; + } +} + +void ARMMachObjectWriter:: +RecordARMMovwMovtRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, + const MCAsmLayout &Layout, + const MCFragment *Fragment, + const MCFixup &Fixup, + MCValue Target, + uint64_t &FixedValue) { + uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); + unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind()); + unsigned Type = macho::RIT_ARM_Half; + + // See <reloc.h>. + const MCSymbol *A = &Target.getSymA()->getSymbol(); + MCSymbolData *A_SD = &Asm.getSymbolData(*A); + + if (!A_SD->getFragment()) + report_fatal_error("symbol '" + A->getName() + + "' can not be undefined in a subtraction expression"); + + uint32_t Value = Writer->getSymbolAddress(A_SD, Layout); + uint32_t Value2 = 0; + uint64_t SecAddr = + Writer->getSectionAddress(A_SD->getFragment()->getParent()); + FixedValue += SecAddr; + + if (const MCSymbolRefExpr *B = Target.getSymB()) { + MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol()); + + if (!B_SD->getFragment()) + report_fatal_error("symbol '" + B->getSymbol().getName() + + "' can not be undefined in a subtraction expression"); + + // Select the appropriate difference relocation type. + Type = macho::RIT_ARM_HalfDifference; + Value2 = Writer->getSymbolAddress(B_SD, Layout); + FixedValue -= Writer->getSectionAddress(B_SD->getFragment()->getParent()); + } + + // Relocations are written out in reverse order, so the PAIR comes first. + // ARM_RELOC_HALF and ARM_RELOC_HALF_SECTDIFF abuse the r_length field: + // + // For these two r_type relocations they always have a pair following them and + // the r_length bits are used differently. The encoding of the r_length is as + // follows: + // low bit of r_length: + // 0 - :lower16: for movw instructions + // 1 - :upper16: for movt instructions + // high bit of r_length: + // 0 - arm instructions + // 1 - thumb instructions + // the other half of the relocated expression is in the following pair + // relocation entry in the the low 16 bits of r_address field. + unsigned ThumbBit = 0; + unsigned MovtBit = 0; + switch ((unsigned)Fixup.getKind()) { + default: break; + case ARM::fixup_arm_movt_hi16: + case ARM::fixup_arm_movt_hi16_pcrel: + MovtBit = 1; + break; + case ARM::fixup_t2_movt_hi16: + case ARM::fixup_t2_movt_hi16_pcrel: + MovtBit = 1; + // Fallthrough + case ARM::fixup_t2_movw_lo16: + case ARM::fixup_t2_movw_lo16_pcrel: + ThumbBit = 1; + break; + } + + + if (Type == macho::RIT_ARM_HalfDifference) { + uint32_t OtherHalf = MovtBit + ? (FixedValue & 0xffff) : ((FixedValue & 0xffff0000) >> 16); + + macho::RelocationEntry MRE; + MRE.Word0 = ((OtherHalf << 0) | + (macho::RIT_Pair << 24) | + (MovtBit << 28) | + (ThumbBit << 29) | + (IsPCRel << 30) | + macho::RF_Scattered); + MRE.Word1 = Value2; + Writer->addRelocation(Fragment->getParent(), MRE); + } + + macho::RelocationEntry MRE; + MRE.Word0 = ((FixupOffset << 0) | + (Type << 24) | + (MovtBit << 28) | + (ThumbBit << 29) | + (IsPCRel << 30) | + macho::RF_Scattered); + MRE.Word1 = Value; + Writer->addRelocation(Fragment->getParent(), MRE); +} + +void ARMMachObjectWriter::RecordARMScatteredRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, + const MCAsmLayout &Layout, + const MCFragment *Fragment, + const MCFixup &Fixup, + MCValue Target, + unsigned Log2Size, + uint64_t &FixedValue) { + uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); + unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind()); + unsigned Type = macho::RIT_Vanilla; + + // See <reloc.h>. + const MCSymbol *A = &Target.getSymA()->getSymbol(); + MCSymbolData *A_SD = &Asm.getSymbolData(*A); + + if (!A_SD->getFragment()) + report_fatal_error("symbol '" + A->getName() + + "' can not be undefined in a subtraction expression"); + + uint32_t Value = Writer->getSymbolAddress(A_SD, Layout); + uint64_t SecAddr = Writer->getSectionAddress(A_SD->getFragment()->getParent()); + FixedValue += SecAddr; + uint32_t Value2 = 0; + + if (const MCSymbolRefExpr *B = Target.getSymB()) { + MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol()); + + if (!B_SD->getFragment()) + report_fatal_error("symbol '" + B->getSymbol().getName() + + "' can not be undefined in a subtraction expression"); + + // Select the appropriate difference relocation type. + Type = macho::RIT_Difference; + Value2 = Writer->getSymbolAddress(B_SD, Layout); + FixedValue -= Writer->getSectionAddress(B_SD->getFragment()->getParent()); + } + + // Relocations are written out in reverse order, so the PAIR comes first. + if (Type == macho::RIT_Difference || + Type == macho::RIT_Generic_LocalDifference) { + macho::RelocationEntry MRE; + MRE.Word0 = ((0 << 0) | + (macho::RIT_Pair << 24) | + (Log2Size << 28) | + (IsPCRel << 30) | + macho::RF_Scattered); + MRE.Word1 = Value2; + Writer->addRelocation(Fragment->getParent(), MRE); + } + + macho::RelocationEntry MRE; + MRE.Word0 = ((FixupOffset << 0) | + (Type << 24) | + (Log2Size << 28) | + (IsPCRel << 30) | + macho::RF_Scattered); + MRE.Word1 = Value; + Writer->addRelocation(Fragment->getParent(), MRE); +} + +void ARMMachObjectWriter::RecordRelocation(MachObjectWriter *Writer, + const MCAssembler &Asm, + const MCAsmLayout &Layout, + const MCFragment *Fragment, + const MCFixup &Fixup, + MCValue Target, + uint64_t &FixedValue) { + unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind()); + unsigned Log2Size; + unsigned RelocType = macho::RIT_Vanilla; + if (!getARMFixupKindMachOInfo(Fixup.getKind(), RelocType, Log2Size)) { + report_fatal_error("unknown ARM fixup kind!"); + return; + } + + // If this is a difference or a defined symbol plus an offset, then we need a + // scattered relocation entry. Differences always require scattered + // relocations. + if (Target.getSymB()) { + if (RelocType == macho::RIT_ARM_Half || + RelocType == macho::RIT_ARM_HalfDifference) + return RecordARMMovwMovtRelocation(Writer, Asm, Layout, Fragment, Fixup, + Target, FixedValue); + return RecordARMScatteredRelocation(Writer, Asm, Layout, Fragment, Fixup, + Target, Log2Size, FixedValue); + } + + // Get the symbol data, if any. + MCSymbolData *SD = 0; + if (Target.getSymA()) + SD = &Asm.getSymbolData(Target.getSymA()->getSymbol()); + + // FIXME: For other platforms, we need to use scattered relocations for + // internal relocations with offsets. If this is an internal relocation with + // an offset, it also needs a scattered relocation entry. + // + // Is this right for ARM? + uint32_t Offset = Target.getConstant(); + if (IsPCRel && RelocType == macho::RIT_Vanilla) + Offset += 1 << Log2Size; + if (Offset && SD && !Writer->doesSymbolRequireExternRelocation(SD)) + return RecordARMScatteredRelocation(Writer, Asm, Layout, Fragment, Fixup, + Target, Log2Size, FixedValue); + + // See <reloc.h>. + uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); + unsigned Index = 0; + unsigned IsExtern = 0; + unsigned Type = 0; + + if (Target.isAbsolute()) { // constant + // FIXME! + report_fatal_error("FIXME: relocations to absolute targets " + "not yet implemented"); + } else { + // Resolve constant variables. + if (SD->getSymbol().isVariable()) { + int64_t Res; + if (SD->getSymbol().getVariableValue()->EvaluateAsAbsolute( + Res, Layout, Writer->getSectionAddressMap())) { + FixedValue = Res; + return; + } + } + + // Check whether we need an external or internal relocation. + if (Writer->doesSymbolRequireExternRelocation(SD)) { + IsExtern = 1; + Index = SD->getIndex(); + + // For external relocations, make sure to offset the fixup value to + // compensate for the addend of the symbol address, if it was + // undefined. This occurs with weak definitions, for example. + if (!SD->Symbol->isUndefined()) + FixedValue -= Layout.getSymbolOffset(SD); + } else { + // The index is the section ordinal (1-based). + const MCSectionData &SymSD = Asm.getSectionData( + SD->getSymbol().getSection()); + Index = SymSD.getOrdinal() + 1; + FixedValue += Writer->getSectionAddress(&SymSD); + } + if (IsPCRel) + FixedValue -= Writer->getSectionAddress(Fragment->getParent()); + + // The type is determined by the fixup kind. + Type = RelocType; + } + + // struct relocation_info (8 bytes) + macho::RelocationEntry MRE; + MRE.Word0 = FixupOffset; + MRE.Word1 = ((Index << 0) | + (IsPCRel << 24) | + (Log2Size << 25) | + (IsExtern << 27) | + (Type << 28)); + Writer->addRelocation(Fragment->getParent(), MRE); +} + +MCObjectWriter *llvm::createARMMachObjectWriter(raw_ostream &OS, + bool Is64Bit, + uint32_t CPUType, + uint32_t CPUSubtype) { + return createMachObjectWriter(new ARMMachObjectWriter(Is64Bit, + CPUType, + CPUSubtype), + OS, /*IsLittleEndian=*/true); +} diff --git a/lib/Target/ARM/ARMRegisterInfo.td b/lib/Target/ARM/ARMRegisterInfo.td index 99418733..76eb496 100644 --- a/lib/Target/ARM/ARMRegisterInfo.td +++ b/lib/Target/ARM/ARMRegisterInfo.td @@ -200,45 +200,16 @@ def FPEXC : ARMReg<8, "fpexc">; // r11 == Frame Pointer (arm-style backtraces) // r10 == Stack Limit // -def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, - R7, R8, R9, R10, R11, R12, - SP, LR, PC]> { - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - static const unsigned ARM_GPR_AO[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R12,ARM::LR, - ARM::R4, ARM::R5, ARM::R6, ARM::R7, - ARM::R8, ARM::R9, ARM::R10, ARM::R11 }; - - // For Thumb1 mode, we don't want to allocate hi regs at all, as we - // don't know how to spill them. If we make our prologue/epilogue code - // smarter at some point, we can go back to using the above allocation - // orders for the Thumb1 instructions that know how to use hi regs. - static const unsigned THUMB_GPR_AO[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R4, ARM::R5, ARM::R6, ARM::R7 }; - - GPRClass::iterator - GPRClass::allocation_order_begin(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.isThumb1Only()) - return THUMB_GPR_AO; - return ARM_GPR_AO; - } - - GPRClass::iterator - GPRClass::allocation_order_end(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.isThumb1Only()) - return THUMB_GPR_AO + (sizeof(THUMB_GPR_AO)/sizeof(unsigned)); - return ARM_GPR_AO + (sizeof(ARM_GPR_AO)/sizeof(unsigned)); - } +def GPR : RegisterClass<"ARM", [i32], 32, (add (sequence "R%u", 0, 12), + SP, LR, PC)> { + // Allocate LR as the first CSR since it is always saved anyway. + // For Thumb1 mode, we don't want to allocate hi regs at all, as we don't + // know how to spill them. If we make our prologue/epilogue code smarter at + // some point, we can go back to using the above allocation orders for the + // Thumb1 instructions that know how to use hi regs. + let AltOrders = [(add LR, GPR), (trunc GPR, 8)]; + let AltOrderSelect = [{ + return 1 + MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only(); }]; } @@ -246,263 +217,98 @@ def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, // register range for operands, but have undefined behaviours when PC // or SP (R13 or R15) are used. The ARM ISA refers to these operands // via the BadReg() pseudo-code description. -def rGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, - R7, R8, R9, R10, R11, R12, LR]> { - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - static const unsigned ARM_rGPR_AO[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R12,ARM::LR, - ARM::R4, ARM::R5, ARM::R6, ARM::R7, - ARM::R8, ARM::R9, ARM::R10, - ARM::R11 }; - - // For Thumb1 mode, we don't want to allocate hi regs at all, as we - // don't know how to spill them. If we make our prologue/epilogue code - // smarter at some point, we can go back to using the above allocation - // orders for the Thumb1 instructions that know how to use hi regs. - static const unsigned THUMB_rGPR_AO[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R4, ARM::R5, ARM::R6, ARM::R7 }; - - rGPRClass::iterator - rGPRClass::allocation_order_begin(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.isThumb1Only()) - return THUMB_rGPR_AO; - return ARM_rGPR_AO; - } - - rGPRClass::iterator - rGPRClass::allocation_order_end(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - - if (Subtarget.isThumb1Only()) - return THUMB_rGPR_AO + (sizeof(THUMB_rGPR_AO)/sizeof(unsigned)); - return ARM_rGPR_AO + (sizeof(ARM_rGPR_AO)/sizeof(unsigned)); - } +def rGPR : RegisterClass<"ARM", [i32], 32, (sub GPR, SP, PC)> { + let AltOrders = [(add LR, rGPR), (trunc rGPR, 8)]; + let AltOrderSelect = [{ + return 1 + MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only(); }]; } // Thumb registers are R0-R7 normally. Some instructions can still use // the general GPR register class above (MOV, e.g.) -def tGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, R7]> {} +def tGPR : RegisterClass<"ARM", [i32], 32, (trunc GPR, 8)>; + +// The high registers in thumb mode, R8-R15. +def hGPR : RegisterClass<"ARM", [i32], 32, (sub GPR, tGPR)>; // For tail calls, we can't use callee-saved registers, as they are restored // to the saved value before the tail call, which would clobber a call address. // Note, getMinimalPhysRegClass(R0) returns tGPR because of the names of // this class and the preceding one(!) This is what we want. -def tcGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R9, R12]> { - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - // R9 is available. - static const unsigned ARM_GPR_R9_TC[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R9, ARM::R12 }; - // R9 is not available. - static const unsigned ARM_GPR_NOR9_TC[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3, - ARM::R12 }; - - // For Thumb1 mode, we don't want to allocate hi regs at all, as we - // don't know how to spill them. If we make our prologue/epilogue code - // smarter at some point, we can go back to using the above allocation - // orders for the Thumb1 instructions that know how to use hi regs. - static const unsigned THUMB_GPR_AO_TC[] = { - ARM::R0, ARM::R1, ARM::R2, ARM::R3 }; - - tcGPRClass::iterator - tcGPRClass::allocation_order_begin(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.isThumb1Only()) - return THUMB_GPR_AO_TC; - return Subtarget.isTargetDarwin() ? ARM_GPR_R9_TC : ARM_GPR_NOR9_TC; - } - - tcGPRClass::iterator - tcGPRClass::allocation_order_end(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - - if (Subtarget.isThumb1Only()) - return THUMB_GPR_AO_TC + (sizeof(THUMB_GPR_AO_TC)/sizeof(unsigned)); - - return Subtarget.isTargetDarwin() ? - ARM_GPR_R9_TC + (sizeof(ARM_GPR_R9_TC)/sizeof(unsigned)) : - ARM_GPR_NOR9_TC + (sizeof(ARM_GPR_NOR9_TC)/sizeof(unsigned)); - } +def tcGPR : RegisterClass<"ARM", [i32], 32, (add R0, R1, R2, R3, R9, R12)> { + let AltOrders = [(and tcGPR, tGPR)]; + let AltOrderSelect = [{ + return MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only(); }]; } - // Scalar single precision floating point register class.. -def SPR : RegisterClass<"ARM", [f32], 32, [S0, S1, S2, S3, S4, S5, S6, S7, S8, - S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22, - S23, S24, S25, S26, S27, S28, S29, S30, S31]>; +def SPR : RegisterClass<"ARM", [f32], 32, (sequence "S%u", 0, 31)>; // Subset of SPR which can be used as a source of NEON scalars for 16-bit // operations -def SPR_8 : RegisterClass<"ARM", [f32], 32, - [S0, S1, S2, S3, S4, S5, S6, S7, - S8, S9, S10, S11, S12, S13, S14, S15]>; +def SPR_8 : RegisterClass<"ARM", [f32], 32, (trunc SPR, 16)>; // Scalar double precision floating point / generic 64-bit vector register // class. // ARM requires only word alignment for double. It's more performant if it // is double-word alignment though. def DPR : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64, - [D0, D1, D2, D3, D4, D5, D6, D7, - D8, D9, D10, D11, D12, D13, D14, D15, - D16, D17, D18, D19, D20, D21, D22, D23, - D24, D25, D26, D27, D28, D29, D30, D31]> { - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - // VFP2 / VFPv3-D16 - static const unsigned ARM_DPR_VFP2[] = { - ARM::D0, ARM::D1, ARM::D2, ARM::D3, - ARM::D4, ARM::D5, ARM::D6, ARM::D7, - ARM::D8, ARM::D9, ARM::D10, ARM::D11, - ARM::D12, ARM::D13, ARM::D14, ARM::D15 }; - // VFP3: D8-D15 are callee saved and should be allocated last. - // Save other low registers for use as DPR_VFP2 and DPR_8 classes. - static const unsigned ARM_DPR_VFP3[] = { - ARM::D16, ARM::D17, ARM::D18, ARM::D19, - ARM::D20, ARM::D21, ARM::D22, ARM::D23, - ARM::D24, ARM::D25, ARM::D26, ARM::D27, - ARM::D28, ARM::D29, ARM::D30, ARM::D31, - ARM::D0, ARM::D1, ARM::D2, ARM::D3, - ARM::D4, ARM::D5, ARM::D6, ARM::D7, - ARM::D8, ARM::D9, ARM::D10, ARM::D11, - ARM::D12, ARM::D13, ARM::D14, ARM::D15 }; - - DPRClass::iterator - DPRClass::allocation_order_begin(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.hasVFP3() && !Subtarget.hasD16()) - return ARM_DPR_VFP3; - return ARM_DPR_VFP2; - } - - DPRClass::iterator - DPRClass::allocation_order_end(const MachineFunction &MF) const { - const TargetMachine &TM = MF.getTarget(); - const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>(); - if (Subtarget.hasVFP3() && !Subtarget.hasD16()) - return ARM_DPR_VFP3 + (sizeof(ARM_DPR_VFP3)/sizeof(unsigned)); - else - return ARM_DPR_VFP2 + (sizeof(ARM_DPR_VFP2)/sizeof(unsigned)); - } - }]; + (sequence "D%u", 0, 31)> { + // Allocate non-VFP2 registers D16-D31 first. + let AltOrders = [(rotl DPR, 16)]; + let AltOrderSelect = [{ return 1; }]; } // Subset of DPR that are accessible with VFP2 (and so that also have // 32-bit SPR subregs). def DPR_VFP2 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64, - [D0, D1, D2, D3, D4, D5, D6, D7, - D8, D9, D10, D11, D12, D13, D14, D15]> { + (trunc DPR, 16)> { let SubRegClasses = [(SPR ssub_0, ssub_1)]; } // Subset of DPR which can be used as a source of NEON scalars for 16-bit // operations def DPR_8 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64, - [D0, D1, D2, D3, D4, D5, D6, D7]> { + (trunc DPR, 8)> { let SubRegClasses = [(SPR_8 ssub_0, ssub_1)]; } // Generic 128-bit vector register class. def QPR : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 128, - [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7, - Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15]> { + (sequence "Q%u", 0, 15)> { let SubRegClasses = [(DPR dsub_0, dsub_1)]; - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - // Q4-Q7 are callee saved and should be allocated last. - // Save other low registers for use as QPR_VFP2 and QPR_8 classes. - static const unsigned ARM_QPR[] = { - ARM::Q8, ARM::Q9, ARM::Q10, ARM::Q11, - ARM::Q12, ARM::Q13, ARM::Q14, ARM::Q15, - ARM::Q0, ARM::Q1, ARM::Q2, ARM::Q3, - ARM::Q4, ARM::Q5, ARM::Q6, ARM::Q7 }; - - QPRClass::iterator - QPRClass::allocation_order_begin(const MachineFunction &MF) const { - return ARM_QPR; - } - - QPRClass::iterator - QPRClass::allocation_order_end(const MachineFunction &MF) const { - return ARM_QPR + (sizeof(ARM_QPR)/sizeof(unsigned)); - } - }]; + // Allocate non-VFP2 aliases Q8-Q15 first. + let AltOrders = [(rotl QPR, 8)]; + let AltOrderSelect = [{ return 1; }]; } // Subset of QPR that have 32-bit SPR subregs. def QPR_VFP2 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], - 128, - [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]> { + 128, (trunc QPR, 8)> { let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3), (DPR_VFP2 dsub_0, dsub_1)]; } // Subset of QPR that have DPR_8 and SPR_8 subregs. def QPR_8 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], - 128, - [Q0, Q1, Q2, Q3]> { + 128, (trunc QPR, 4)> { let SubRegClasses = [(SPR_8 ssub_0, ssub_1, ssub_2, ssub_3), (DPR_8 dsub_0, dsub_1)]; } // Pseudo 256-bit vector register class to model pairs of Q registers // (4 consecutive D registers). -def QQPR : RegisterClass<"ARM", [v4i64], - 256, - [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7]> { +def QQPR : RegisterClass<"ARM", [v4i64], 256, (sequence "QQ%u", 0, 7)> { let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3), (QPR qsub_0, qsub_1)]; - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - // QQ2-QQ3 are callee saved and should be allocated last. - // Save other low registers for use as QPR_VFP2 and QPR_8 classes. - static const unsigned ARM_QQPR[] = { - ARM::QQ4, ARM::QQ5, ARM::QQ6, ARM::QQ7, - ARM::QQ0, ARM::QQ1, ARM::QQ2, ARM::QQ3 }; - - QQPRClass::iterator - QQPRClass::allocation_order_begin(const MachineFunction &MF) const { - return ARM_QQPR; - } - - QQPRClass::iterator - QQPRClass::allocation_order_end(const MachineFunction &MF) const { - return ARM_QQPR + (sizeof(ARM_QQPR)/sizeof(unsigned)); - } - }]; + // Allocate non-VFP2 aliases first. + let AltOrders = [(rotl QQPR, 4)]; + let AltOrderSelect = [{ return 1; }]; } // Subset of QQPR that have 32-bit SPR subregs. -def QQPR_VFP2 : RegisterClass<"ARM", [v4i64], - 256, - [QQ0, QQ1, QQ2, QQ3]> { +def QQPR_VFP2 : RegisterClass<"ARM", [v4i64], 256, (trunc QQPR, 4)> { let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3), (DPR_VFP2 dsub_0, dsub_1, dsub_2, dsub_3), (QPR_VFP2 qsub_0, qsub_1)]; @@ -511,35 +317,16 @@ def QQPR_VFP2 : RegisterClass<"ARM", [v4i64], // Pseudo 512-bit vector register class to model 4 consecutive Q registers // (8 consecutive D registers). -def QQQQPR : RegisterClass<"ARM", [v8i64], - 256, - [QQQQ0, QQQQ1, QQQQ2, QQQQ3]> { +def QQQQPR : RegisterClass<"ARM", [v8i64], 256, (sequence "QQQQ%u", 0, 3)> { let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3, dsub_4, dsub_5, dsub_6, dsub_7), (QPR qsub_0, qsub_1, qsub_2, qsub_3)]; - let MethodProtos = [{ - iterator allocation_order_begin(const MachineFunction &MF) const; - iterator allocation_order_end(const MachineFunction &MF) const; - }]; - let MethodBodies = [{ - // QQQQ1 is callee saved and should be allocated last. - // Save QQQQ0 for use as QPR_VFP2 and QPR_8 classes. - static const unsigned ARM_QQQQPR[] = { - ARM::QQQQ2, ARM::QQQQ3, ARM::QQQQ0, ARM::QQQQ1 }; - - QQQQPRClass::iterator - QQQQPRClass::allocation_order_begin(const MachineFunction &MF) const { - return ARM_QQQQPR; - } - - QQQQPRClass::iterator - QQQQPRClass::allocation_order_end(const MachineFunction &MF) const { - return ARM_QQQQPR + (sizeof(ARM_QQQQPR)/sizeof(unsigned)); - } - }]; + // Allocate non-VFP2 aliases first. + let AltOrders = [(rotl QQQQPR, 2)]; + let AltOrderSelect = [{ return 1; }]; } // Condition code registers. -def CCR : RegisterClass<"ARM", [i32], 32, [CPSR]> { +def CCR : RegisterClass<"ARM", [i32], 32, (add CPSR)> { let isAllocatable = 0; } diff --git a/lib/Target/ARM/ARMSubtarget.cpp b/lib/Target/ARM/ARMSubtarget.cpp index c6f266b..1cab9e4 100644 --- a/lib/Target/ARM/ARMSubtarget.cpp +++ b/lib/Target/ARM/ARMSubtarget.cpp @@ -7,17 +7,21 @@ // //===----------------------------------------------------------------------===// // -// This file implements the ARM specific subclass of TargetSubtarget. +// This file implements the ARM specific subclass of TargetSubtargetInfo. // //===----------------------------------------------------------------------===// #include "ARMSubtarget.h" -#include "ARMGenSubtarget.inc" #include "ARMBaseRegisterInfo.h" #include "llvm/GlobalValue.h" -#include "llvm/Target/TargetOptions.h" +#include "llvm/Target/TargetSubtargetInfo.h" #include "llvm/Support/CommandLine.h" #include "llvm/ADT/SmallVector.h" + +#define GET_SUBTARGETINFO_TARGET_DESC +#define GET_SUBTARGETINFO_CTOR +#include "ARMGenSubtargetInfo.inc" + using namespace llvm; static cl::opt<bool> @@ -31,17 +35,25 @@ static cl::opt<bool> StrictAlign("arm-strict-align", cl::Hidden, cl::desc("Disallow all unaligned memory accesses")); -ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS, - bool isT) - : ARMArchVersion(V4) +ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &CPU, + const std::string &FS) + : ARMGenSubtargetInfo(TT, CPU, FS) , ARMProcFamily(Others) - , ARMFPUType(None) + , HasV4TOps(false) + , HasV5TOps(false) + , HasV5TEOps(false) + , HasV6Ops(false) + , HasV6T2Ops(false) + , HasV7Ops(false) + , HasVFPv2(false) + , HasVFPv3(false) + , HasNEON(false) , UseNEONForSinglePrecisionFP(false) , SlowFPVMLx(false) , HasVMLxForwarding(false) , SlowFPBrcc(false) - , IsThumb(isT) - , ThumbMode(Thumb1) + , InThumbMode(false) + , HasThumb2(false) , NoARM(false) , PostRAScheduler(false) , IsR9Reserved(ReserveR9) @@ -56,94 +68,40 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS, , HasMPExtension(false) , FPOnlySP(false) , AllowsUnalignedMem(false) + , Thumb2DSP(false) , stackAlignment(4) - , CPUString("generic") + , CPUString(CPU) , TargetTriple(TT) , TargetABI(ARM_ABI_APCS) { - // Default to soft float ABI - if (FloatABIType == FloatABI::Default) - FloatABIType = FloatABI::Soft; - // Determine default and user specified characteristics - - // When no arch is specified either by CPU or by attributes, make the default - // ARMv4T. - const char *ARMArchFeature = ""; - if (CPUString == "generic" && (FS.empty() || FS == "generic")) { - ARMArchVersion = V4T; - ARMArchFeature = ",+v4t"; + if (CPUString.empty()) + CPUString = "generic"; + + // Insert the architecture feature derived from the target triple into the + // feature string. This is important for setting features that are implied + // based on the architecture version. + std::string ArchFS = ARM_MC::ParseARMTriple(TT); + if (!FS.empty()) { + if (!ArchFS.empty()) + ArchFS = ArchFS + "," + FS; + else + ArchFS = FS; } + ParseSubtargetFeatures(CPUString, ArchFS); - // Set the boolean corresponding to the current target triple, or the default - // if one cannot be determined, to true. - unsigned Len = TT.length(); - unsigned Idx = 0; + // Thumb2 implies at least V6T2. FIXME: Fix tests to explicitly specify a + // ARM version or CPU and then remove this. + if (!HasV6T2Ops && hasThumb2()) + HasV4TOps = HasV5TOps = HasV5TEOps = HasV6Ops = HasV6T2Ops = true; - if (Len >= 5 && TT.substr(0, 4) == "armv") - Idx = 4; - else if (Len >= 6 && TT.substr(0, 5) == "thumb") { - IsThumb = true; - if (Len >= 7 && TT[5] == 'v') - Idx = 6; - } - if (Idx) { - unsigned SubVer = TT[Idx]; - if (SubVer >= '7' && SubVer <= '9') { - ARMArchVersion = V7A; - ARMArchFeature = ",+v7a"; - if (Len >= Idx+2 && TT[Idx+1] == 'm') { - ARMArchVersion = V7M; - ARMArchFeature = ",+v7m"; - } - } else if (SubVer == '6') { - ARMArchVersion = V6; - ARMArchFeature = ",+v6"; - if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2') { - ARMArchVersion = V6T2; - ARMArchFeature = ",+v6t2"; - } - } else if (SubVer == '5') { - ARMArchVersion = V5T; - ARMArchFeature = ",+v5t"; - if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e') { - ARMArchVersion = V5TE; - ARMArchFeature = ",+v5te"; - } - } else if (SubVer == '4') { - if (Len >= Idx+2 && TT[Idx+1] == 't') { - ARMArchVersion = V4T; - ARMArchFeature = ",+v4t"; - } else { - ARMArchVersion = V4; - ARMArchFeature = ""; - } - } - } - - if (TT.find("eabi") != std::string::npos) - TargetABI = ARM_ABI_AAPCS; - - // Parse features string. If the first entry in FS (the CPU) is missing, - // insert the architecture feature derived from the target triple. This is - // important for setting features that are implied based on the architecture - // version. - std::string FSWithArch; - if (FS.empty()) - FSWithArch = std::string(ARMArchFeature); - else if (FS.find(',') == 0) - FSWithArch = std::string(ARMArchFeature) + FS; - else - FSWithArch = FS; - CPUString = ParseSubtargetFeatures(FSWithArch, CPUString); + // Initialize scheduling itinerary for the specified CPU. + InstrItins = getInstrItineraryForCPU(CPUString); // After parsing Itineraries, set ItinData.IssueWidth. computeIssueWidth(); - // Thumb2 implies at least V6T2. - if (ARMArchVersion >= V6T2) - ThumbMode = Thumb2; - else if (ThumbMode >= Thumb2) - ARMArchVersion = V6T2; + if (TT.find("eabi") != std::string::npos) + TargetABI = ARM_ABI_AAPCS; if (isAAPCS_ABI()) stackAlignment = 8; @@ -151,7 +109,7 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS, if (!isTargetDarwin()) UseMovt = hasV6T2Ops(); else { - IsR9Reserved = ReserveR9 | (ARMArchVersion < V6); + IsR9Reserved = ReserveR9 | !HasV6Ops; UseMovt = DarwinUseMOVT && hasV6T2Ops(); } @@ -247,9 +205,9 @@ void ARMSubtarget::computeIssueWidth() { bool ARMSubtarget::enablePostRAScheduler( CodeGenOpt::Level OptLevel, - TargetSubtarget::AntiDepBreakMode& Mode, + TargetSubtargetInfo::AntiDepBreakMode& Mode, RegClassVector& CriticalPathRCs) const { - Mode = TargetSubtarget::ANTIDEP_CRITICAL; + Mode = TargetSubtargetInfo::ANTIDEP_CRITICAL; CriticalPathRCs.clear(); CriticalPathRCs.push_back(&ARM::GPRRegClass); return PostRAScheduler && OptLevel >= CodeGenOpt::Default; diff --git a/lib/Target/ARM/ARMSubtarget.h b/lib/Target/ARM/ARMSubtarget.h index 0271c87..c650872 100644 --- a/lib/Target/ARM/ARMSubtarget.h +++ b/lib/Target/ARM/ARMSubtarget.h @@ -7,50 +7,49 @@ // //===----------------------------------------------------------------------===// // -// This file declares the ARM specific subclass of TargetSubtarget. +// This file declares the ARM specific subclass of TargetSubtargetInfo. // //===----------------------------------------------------------------------===// #ifndef ARMSUBTARGET_H #define ARMSUBTARGET_H -#include "llvm/Target/TargetInstrItineraries.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetSubtarget.h" +#include "MCTargetDesc/ARMMCTargetDesc.h" +#include "llvm/Target/TargetSubtargetInfo.h" +#include "llvm/MC/MCInstrItineraries.h" #include "llvm/ADT/Triple.h" #include <string> +#define GET_SUBTARGETINFO_HEADER +#include "ARMGenSubtargetInfo.inc" + namespace llvm { class GlobalValue; +class StringRef; -class ARMSubtarget : public TargetSubtarget { +class ARMSubtarget : public ARMGenSubtargetInfo { protected: - enum ARMArchEnum { - V4, V4T, V5T, V5TE, V6, V6M, V6T2, V7A, V7M - }; - enum ARMProcFamilyEnum { Others, CortexA8, CortexA9 }; - enum ARMFPEnum { - None, VFPv2, VFPv3, NEON - }; - - enum ThumbTypeEnum { - Thumb1, - Thumb2 - }; - - /// ARMArchVersion - ARM architecture version: V4, V4T (base), V5T, V5TE, - /// V6, V6T2, V7A, V7M. - ARMArchEnum ARMArchVersion; - /// ARMProcFamily - ARM processor family: Cortex-A8, Cortex-A9, and others. ARMProcFamilyEnum ARMProcFamily; - /// ARMFPUType - Floating Point Unit type. - ARMFPEnum ARMFPUType; + /// HasV4TOps, HasV5TOps, HasV5TEOps, HasV6Ops, HasV6T2Ops, HasV7Ops - + /// Specify whether target support specific ARM ISA variants. + bool HasV4TOps; + bool HasV5TOps; + bool HasV5TEOps; + bool HasV6Ops; + bool HasV6T2Ops; + bool HasV7Ops; + + /// HasVFPv2, HasVFPv3, HasNEON - Specify what floating point ISAs are + /// supported. + bool HasVFPv2; + bool HasVFPv3; + bool HasNEON; /// UseNEONForSinglePrecisionFP - if the NEONFP attribute has been /// specified. Use the method useNEONForSinglePrecisionFP() to @@ -68,11 +67,11 @@ protected: /// SlowFPBrcc - True if floating point compare + branch is slow. bool SlowFPBrcc; - /// IsThumb - True if we are in thumb mode, false if in ARM mode. - bool IsThumb; + /// InThumbMode - True if compiling for Thumb, false for ARM. + bool InThumbMode; - /// ThumbMode - Indicates supported Thumb version. - ThumbTypeEnum ThumbMode; + /// HasThumb2 - True if Thumb2 instructions are supported. + bool HasThumb2; /// NoARM - True if subtarget does not support ARM mode execution. bool NoARM; @@ -128,6 +127,10 @@ protected: /// ARMTargetLowering::allowsUnalignedMemoryAccesses(). bool AllowsUnalignedMem; + /// Thumb2DSP - If true, the subtarget supports the v7 DSP (saturating arith + /// and such) instructions in Thumb2 code. + bool Thumb2DSP; + /// stackAlignment - The minimum alignment known to hold of the stack frame on /// entry to the function and which must be maintained by every function. unsigned stackAlignment; @@ -154,7 +157,8 @@ protected: /// This constructor initializes the data members to match that /// of the specified triple. /// - ARMSubtarget(const std::string &TT, const std::string &FS, bool isThumb); + ARMSubtarget(const std::string &TT, const std::string &CPU, + const std::string &FS); /// getMaxInlineSizeThreshold - Returns the maximum memset / memcpy size /// that still makes it profitable to inline the call. @@ -165,28 +169,28 @@ protected: } /// ParseSubtargetFeatures - Parses features string setting specified /// subtarget options. Definition of function is auto generated by tblgen. - std::string ParseSubtargetFeatures(const std::string &FS, - const std::string &CPU); + void ParseSubtargetFeatures(StringRef CPU, StringRef FS); void computeIssueWidth(); - bool hasV4TOps() const { return ARMArchVersion >= V4T; } - bool hasV5TOps() const { return ARMArchVersion >= V5T; } - bool hasV5TEOps() const { return ARMArchVersion >= V5TE; } - bool hasV6Ops() const { return ARMArchVersion >= V6; } - bool hasV6T2Ops() const { return ARMArchVersion >= V6T2; } - bool hasV7Ops() const { return ARMArchVersion >= V7A; } + bool hasV4TOps() const { return HasV4TOps; } + bool hasV5TOps() const { return HasV5TOps; } + bool hasV5TEOps() const { return HasV5TEOps; } + bool hasV6Ops() const { return HasV6Ops; } + bool hasV6T2Ops() const { return HasV6T2Ops; } + bool hasV7Ops() const { return HasV7Ops; } bool isCortexA8() const { return ARMProcFamily == CortexA8; } bool isCortexA9() const { return ARMProcFamily == CortexA9; } bool hasARMOps() const { return !NoARM; } - bool hasVFP2() const { return ARMFPUType >= VFPv2; } - bool hasVFP3() const { return ARMFPUType >= VFPv3; } - bool hasNEON() const { return ARMFPUType >= NEON; } + bool hasVFP2() const { return HasVFPv2; } + bool hasVFP3() const { return HasVFPv3; } + bool hasNEON() const { return HasNEON; } bool useNEONForSinglePrecisionFP() const { return hasNEON() && UseNEONForSinglePrecisionFP; } + bool hasDivide() const { return HasHardwareDivide; } bool hasT2ExtractPack() const { return HasT2ExtractPack; } bool hasDataBarrier() const { return HasDataBarrier; } @@ -197,6 +201,7 @@ protected: bool prefers32BitThumb() const { return Pref32BitThumb; } bool avoidCPSRPartialUpdate() const { return AvoidCPSRPartialUpdate; } bool hasMPExtension() const { return HasMPExtension; } + bool hasThumb2DSP() const { return Thumb2DSP; } bool hasFP16() const { return HasFP16; } bool hasD16() const { return HasD16; } @@ -209,10 +214,10 @@ protected: bool isAPCS_ABI() const { return TargetABI == ARM_ABI_APCS; } bool isAAPCS_ABI() const { return TargetABI == ARM_ABI_AAPCS; } - bool isThumb() const { return IsThumb; } - bool isThumb1Only() const { return IsThumb && (ThumbMode == Thumb1); } - bool isThumb2() const { return IsThumb && (ThumbMode == Thumb2); } - bool hasThumb2() const { return ThumbMode >= Thumb2; } + bool isThumb() const { return InThumbMode; } + bool isThumb1Only() const { return InThumbMode && !HasThumb2; } + bool isThumb2() const { return InThumbMode && HasThumb2; } + bool hasThumb2() const { return HasThumb2; } bool isR9Reserved() const { return IsR9Reserved; } @@ -226,7 +231,7 @@ protected: /// enablePostRAScheduler - True at 'More' optimization. bool enablePostRAScheduler(CodeGenOpt::Level OptLevel, - TargetSubtarget::AntiDepBreakMode& Mode, + TargetSubtargetInfo::AntiDepBreakMode& Mode, RegClassVector& CriticalPathRCs) const; /// getInstrItins - Return the instruction itineraies based on subtarget diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp index 29aa4f7..f0b176a 100644 --- a/lib/Target/ARM/ARMTargetMachine.cpp +++ b/lib/Target/ARM/ARMTargetMachine.cpp @@ -11,7 +11,6 @@ //===----------------------------------------------------------------------===// #include "ARMTargetMachine.h" -#include "ARMMCAsmInfo.h" #include "ARMFrameLowering.h" #include "ARM.h" #include "llvm/PassManager.h" @@ -22,15 +21,6 @@ #include "llvm/Target/TargetRegistry.h" using namespace llvm; -static MCAsmInfo *createMCAsmInfo(const Target &T, StringRef TT) { - Triple TheTriple(TT); - - if (TheTriple.isOSDarwin()) - return new ARMMCAsmInfoDarwin(); - - return new ARMELFMCAsmInfo(); -} - // This is duplicated code. Refactor this. static MCStreamer *createMCStreamer(const Target &T, const std::string &TT, MCContext &Ctx, TargetAsmBackend &TAB, @@ -56,10 +46,6 @@ extern "C" void LLVMInitializeARMTarget() { RegisterTargetMachine<ARMTargetMachine> X(TheARMTarget); RegisterTargetMachine<ThumbTargetMachine> Y(TheThumbTarget); - // Register the target asm info. - RegisterAsmInfoFn A(TheARMTarget, createMCAsmInfo); - RegisterAsmInfoFn B(TheThumbTarget, createMCAsmInfo); - // Register the MC Code Emitter TargetRegistry::RegisterCodeEmitter(TheARMTarget, createARMMCCodeEmitter); TargetRegistry::RegisterCodeEmitter(TheThumbTarget, createARMMCCodeEmitter); @@ -78,18 +64,23 @@ extern "C" void LLVMInitializeARMTarget() { /// ARMBaseTargetMachine::ARMBaseTargetMachine(const Target &T, const std::string &TT, - const std::string &FS, - bool isThumb) - : LLVMTargetMachine(T, TT), - Subtarget(TT, FS, isThumb), + const std::string &CPU, + const std::string &FS) + : LLVMTargetMachine(T, TT, CPU, FS), + Subtarget(TT, CPU, FS), JITInfo(), InstrItins(Subtarget.getInstrItineraryData()) { DefRelocModel = getRelocationModel(); + + // Default to soft float ABI + if (FloatABIType == FloatABI::Default) + FloatABIType = FloatABI::Soft; } ARMTargetMachine::ARMTargetMachine(const Target &T, const std::string &TT, + const std::string &CPU, const std::string &FS) - : ARMBaseTargetMachine(T, TT, FS, false), InstrInfo(Subtarget), + : ARMBaseTargetMachine(T, TT, CPU, FS), InstrInfo(Subtarget), DataLayout(Subtarget.isAPCS_ABI() ? std::string("e-p:32:32-f64:32:64-i64:32:64-" "v128:32:128-v64:32:64-n32") : @@ -105,8 +96,9 @@ ARMTargetMachine::ARMTargetMachine(const Target &T, const std::string &TT, } ThumbTargetMachine::ThumbTargetMachine(const Target &T, const std::string &TT, + const std::string &CPU, const std::string &FS) - : ARMBaseTargetMachine(T, TT, FS, true), + : ARMBaseTargetMachine(T, TT, CPU, FS), InstrInfo(Subtarget.hasThumb2() ? ((ARMBaseInstrInfo*)new Thumb2InstrInfo(Subtarget)) : ((ARMBaseInstrInfo*)new Thumb1InstrInfo(Subtarget))), diff --git a/lib/Target/ARM/ARMTargetMachine.h b/lib/Target/ARM/ARMTargetMachine.h index e0aa149..bc3d46a 100644 --- a/lib/Target/ARM/ARMTargetMachine.h +++ b/lib/Target/ARM/ARMTargetMachine.h @@ -41,7 +41,7 @@ private: public: ARMBaseTargetMachine(const Target &T, const std::string &TT, - const std::string &FS, bool isThumb); + const std::string &CPU, const std::string &FS); virtual ARMJITInfo *getJITInfo() { return &JITInfo; } virtual const ARMSubtarget *getSubtargetImpl() const { return &Subtarget; } @@ -70,7 +70,7 @@ class ARMTargetMachine : public ARMBaseTargetMachine { ARMFrameLowering FrameLowering; public: ARMTargetMachine(const Target &T, const std::string &TT, - const std::string &FS); + const std::string &CPU, const std::string &FS); virtual const ARMRegisterInfo *getRegisterInfo() const { return &InstrInfo.getRegisterInfo(); @@ -109,7 +109,7 @@ class ThumbTargetMachine : public ARMBaseTargetMachine { OwningPtr<ARMFrameLowering> FrameLowering; public: ThumbTargetMachine(const Target &T, const std::string &TT, - const std::string &FS); + const std::string &CPU, const std::string &FS); /// returns either Thumb1RegisterInfo or Thumb2RegisterInfo virtual const ARMBaseRegisterInfo *getRegisterInfo() const { diff --git a/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp b/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp index 2428ce1..d9a5fa2 100644 --- a/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp +++ b/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp @@ -87,8 +87,9 @@ public: : ARMBaseAsmLexer(T, MAI) { std::string tripleString("arm-unknown-unknown"); std::string featureString; + std::string CPU; OwningPtr<const TargetMachine> - targetMachine(T.createTargetMachine(tripleString, featureString)); + targetMachine(T.createTargetMachine(tripleString, CPU, featureString)); InitRegisterMap(targetMachine->getRegisterInfo()); } }; @@ -99,8 +100,9 @@ public: : ARMBaseAsmLexer(T, MAI) { std::string tripleString("thumb-unknown-unknown"); std::string featureString; + std::string CPU; OwningPtr<const TargetMachine> - targetMachine(T.createTargetMachine(tripleString, featureString)); + targetMachine(T.createTargetMachine(tripleString, CPU, featureString)); InitRegisterMap(targetMachine->getRegisterInfo()); } }; diff --git a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp index 4bc12c9..a474127 100644 --- a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp +++ b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp @@ -20,14 +20,17 @@ #include "llvm/MC/MCStreamer.h" #include "llvm/MC/MCExpr.h" #include "llvm/MC/MCInst.h" +#include "llvm/MC/MCSubtargetInfo.h" #include "llvm/Target/TargetRegistry.h" #include "llvm/Target/TargetAsmParser.h" #include "llvm/Support/SourceMgr.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringExtras.h" #include "llvm/ADT/StringSwitch.h" #include "llvm/ADT/Twine.h" + using namespace llvm; namespace { @@ -35,8 +38,8 @@ namespace { class ARMOperand; class ARMAsmParser : public TargetAsmParser { + MCSubtargetInfo &STI; MCAsmParser &Parser; - TargetMachine &TM; MCAsmParser &getParser() const { return Parser; } MCAsmLexer &getLexer() const { return Parser.getLexer(); } @@ -47,7 +50,7 @@ class ARMAsmParser : public TargetAsmParser { int TryParseRegister(); virtual bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc); bool TryParseRegisterWithWriteBack(SmallVectorImpl<MCParsedAsmOperand*> &); - bool TryParseShiftRegister(SmallVectorImpl<MCParsedAsmOperand*> &); + int TryParseShiftRegister(SmallVectorImpl<MCParsedAsmOperand*> &); bool ParseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &); bool ParseMemory(SmallVectorImpl<MCParsedAsmOperand*> &, ARMII::AddrMode AddrMode); @@ -79,6 +82,18 @@ class ARMAsmParser : public TargetAsmParser { void GetMnemonicAcceptInfo(StringRef Mnemonic, bool &CanAcceptCarrySet, bool &CanAcceptPredicationCode); + bool isThumb() const { + // FIXME: Can tablegen auto-generate this? + return (STI.getFeatureBits() & ARM::ModeThumb) != 0; + } + bool isThumbOne() const { + return isThumb() && (STI.getFeatureBits() & ARM::FeatureThumb2) == 0; + } + void SwitchMode() { + unsigned FB = ComputeAvailableFeatures(STI.ToggleFeature(ARM::ModeThumb)); + setAvailableFeatures(FB); + } + /// @name Auto-generated Match Functions /// { @@ -113,13 +128,13 @@ class ARMAsmParser : public TargetAsmParser { const SmallVectorImpl<MCParsedAsmOperand*> &); public: - ARMAsmParser(const Target &T, MCAsmParser &_Parser, TargetMachine &_TM) - : TargetAsmParser(T), Parser(_Parser), TM(_TM) { - MCAsmParserExtension::Initialize(_Parser); - // Initialize the set of available features. - setAvailableFeatures(ComputeAvailableFeatures( - &TM.getSubtarget<ARMSubtarget>())); - } + ARMAsmParser(MCSubtargetInfo &_STI, MCAsmParser &_Parser) + : TargetAsmParser(), STI(_STI), Parser(_Parser) { + MCAsmParserExtension::Initialize(_Parser); + + // Initialize the set of available features. + setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits())); + } virtual bool ParseInstruction(StringRef Name, SMLoc NameLoc, SmallVectorImpl<MCParsedAsmOperand*> &Operands); @@ -146,6 +161,7 @@ class ARMOperand : public MCParsedAsmOperand { RegisterList, DPRRegisterList, SPRRegisterList, + ShiftedRegister, Shifter, Token } Kind; @@ -207,8 +223,14 @@ class ARMOperand : public MCParsedAsmOperand { struct { ARM_AM::ShiftOpc ShiftTy; - unsigned RegNum; + unsigned Imm; } Shift; + struct { + ARM_AM::ShiftOpc ShiftTy; + unsigned SrcReg; + unsigned ShiftReg; + unsigned ShiftImm; + } ShiftedReg; }; ARMOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {} @@ -255,6 +277,9 @@ public: case Shifter: Shift = o.Shift; break; + case ShiftedRegister: + ShiftedReg = o.ShiftedReg; + break; } } @@ -350,6 +375,46 @@ public: bool isCondCode() const { return Kind == CondCode; } bool isCCOut() const { return Kind == CCOut; } bool isImm() const { return Kind == Immediate; } + bool isImm0_255() const { + if (Kind != Immediate) + return false; + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); + if (!CE) return false; + int64_t Value = CE->getValue(); + return Value >= 0 && Value < 256; + } + bool isImm0_7() const { + if (Kind != Immediate) + return false; + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); + if (!CE) return false; + int64_t Value = CE->getValue(); + return Value >= 0 && Value < 8; + } + bool isImm0_15() const { + if (Kind != Immediate) + return false; + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); + if (!CE) return false; + int64_t Value = CE->getValue(); + return Value >= 0 && Value < 16; + } + bool isImm0_65535() const { + if (Kind != Immediate) + return false; + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); + if (!CE) return false; + int64_t Value = CE->getValue(); + return Value >= 0 && Value < 65536; + } + bool isT2SOImm() const { + if (Kind != Immediate) + return false; + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); + if (!CE) return false; + int64_t Value = CE->getValue(); + return ARM_AM::getT2SOImmVal(Value) != -1; + } bool isReg() const { return Kind == Register; } bool isRegList() const { return Kind == RegisterList; } bool isDPRRegList() const { return Kind == DPRRegisterList; } @@ -358,6 +423,7 @@ public: bool isMemBarrierOpt() const { return Kind == MemBarrierOpt; } bool isMemory() const { return Kind == Memory; } bool isShifter() const { return Kind == Shifter; } + bool isShiftedReg() const { return Kind == ShiftedRegister; } bool isMemMode2() const { if (getMemAddrMode() != ARMII::AddrMode2) return false; @@ -488,6 +554,18 @@ public: Inst.addOperand(MCOperand::CreateReg(getReg())); } + void addShiftedRegOperands(MCInst &Inst, unsigned N) const { + assert(N == 3 && "Invalid number of operands!"); + assert(isShiftedReg() && "addShiftedRegOperands() on non ShiftedReg!"); + assert((ShiftedReg.ShiftReg == 0 || + ARM_AM::getSORegOffset(ShiftedReg.ShiftImm) == 0) && + "Invalid shifted register operand!"); + Inst.addOperand(MCOperand::CreateReg(ShiftedReg.SrcReg)); + Inst.addOperand(MCOperand::CreateReg(ShiftedReg.ShiftReg)); + Inst.addOperand(MCOperand::CreateImm( + ARM_AM::getSORegOpc(ShiftedReg.ShiftTy, ShiftedReg.ShiftImm))); + } + void addShifterOperands(MCInst &Inst, unsigned N) const { assert(N == 1 && "Invalid number of operands!"); Inst.addOperand(MCOperand::CreateImm( @@ -515,6 +593,31 @@ public: addExpr(Inst, getImm()); } + void addImm0_255Operands(MCInst &Inst, unsigned N) const { + assert(N == 1 && "Invalid number of operands!"); + addExpr(Inst, getImm()); + } + + void addImm0_7Operands(MCInst &Inst, unsigned N) const { + assert(N == 1 && "Invalid number of operands!"); + addExpr(Inst, getImm()); + } + + void addImm0_15Operands(MCInst &Inst, unsigned N) const { + assert(N == 1 && "Invalid number of operands!"); + addExpr(Inst, getImm()); + } + + void addImm0_65535Operands(MCInst &Inst, unsigned N) const { + assert(N == 1 && "Invalid number of operands!"); + addExpr(Inst, getImm()); + } + + void addT2SOImmOperands(MCInst &Inst, unsigned N) const { + assert(N == 1 && "Invalid number of operands!"); + addExpr(Inst, getImm()); + } + void addMemBarrierOptOperands(MCInst &Inst, unsigned N) const { assert(N == 1 && "Invalid number of operands!"); Inst.addOperand(MCOperand::CreateImm(unsigned(getMemBarrierOpt()))); @@ -648,7 +751,7 @@ public: Inst.addOperand(MCOperand::CreateImm(unsigned(getProcIFlags()))); } - virtual void dump(raw_ostream &OS) const; + virtual void print(raw_ostream &OS) const; static ARMOperand *CreateCondCode(ARMCC::CondCodes CC, SMLoc S) { ARMOperand *Op = new ARMOperand(CondCode); @@ -699,6 +802,21 @@ public: return Op; } + static ARMOperand *CreateShiftedRegister(ARM_AM::ShiftOpc ShTy, + unsigned SrcReg, + unsigned ShiftReg, + unsigned ShiftImm, + SMLoc S, SMLoc E) { + ARMOperand *Op = new ARMOperand(ShiftedRegister); + Op->ShiftedReg.ShiftTy = ShTy; + Op->ShiftedReg.SrcReg = SrcReg; + Op->ShiftedReg.ShiftReg = ShiftReg; + Op->ShiftedReg.ShiftImm = ShiftImm; + Op->StartLoc = S; + Op->EndLoc = E; + return Op; + } + static ARMOperand *CreateShifter(ARM_AM::ShiftOpc ShTy, SMLoc S, SMLoc E) { ARMOperand *Op = new ARMOperand(Shifter); @@ -802,7 +920,7 @@ public: } // end anonymous namespace. -void ARMOperand::dump(raw_ostream &OS) const { +void ARMOperand::print(raw_ostream &OS) const { switch (Kind) { case CondCode: OS << "<ARMCC::" << ARMCondCodeToString(getCondCode()) << ">"; @@ -863,7 +981,15 @@ void ARMOperand::dump(raw_ostream &OS) const { OS << "<register " << getReg() << ">"; break; case Shifter: - OS << "<shifter " << getShiftOpcStr(Shift.ShiftTy) << ">"; + OS << "<shifter " << ARM_AM::getShiftOpcStr(Shift.ShiftTy) << ">"; + break; + case ShiftedRegister: + OS << "<so_reg" + << ShiftedReg.SrcReg + << ARM_AM::getShiftOpcStr(ARM_AM::getSORegShOp(ShiftedReg.ShiftImm)) + << ", " << ShiftedReg.ShiftReg << ", " + << ARM_AM::getSORegOffset(ShiftedReg.ShiftImm) + << ">"; break; case RegisterList: case DPRRegisterList: @@ -927,11 +1053,12 @@ int ARMAsmParser::TryParseRegister() { return RegNum; } -/// Try to parse a register name. The token must be an Identifier when called, -/// and if it is a register name the token is eaten and the register number is -/// returned. Otherwise return -1. -/// -bool ARMAsmParser::TryParseShiftRegister( +// Try to parse a shifter (e.g., "lsl <amt>"). On success, return 0. +// If a recoverable error occurs, return 1. If an irrecoverable error +// occurs, return -1. An irrecoverable error is one where tokens have been +// consumed in the process of trying to parse the shifter (i.e., when it is +// indeed a shifter operand, but malformed). +int ARMAsmParser::TryParseShiftRegister( SmallVectorImpl<MCParsedAsmOperand*> &Operands) { SMLoc S = Parser.getTok().getLoc(); const AsmToken &Tok = Parser.getTok(); @@ -948,18 +1075,69 @@ bool ARMAsmParser::TryParseShiftRegister( .Default(ARM_AM::no_shift); if (ShiftTy == ARM_AM::no_shift) - return true; - - Parser.Lex(); // Eat shift-type operand; - int RegNum = TryParseRegister(); - if (RegNum == -1) - return Error(Parser.getTok().getLoc(), "register expected"); + return 1; + + Parser.Lex(); // Eat the operator. + + // The source register for the shift has already been added to the + // operand list, so we need to pop it off and combine it into the shifted + // register operand instead. + OwningPtr<ARMOperand> PrevOp((ARMOperand*)Operands.pop_back_val()); + if (!PrevOp->isReg()) + return Error(PrevOp->getStartLoc(), "shift must be of a register"); + int SrcReg = PrevOp->getReg(); + int64_t Imm = 0; + int ShiftReg = 0; + if (ShiftTy == ARM_AM::rrx) { + // RRX Doesn't have an explicit shift amount. The encoder expects + // the shift register to be the same as the source register. Seems odd, + // but OK. + ShiftReg = SrcReg; + } else { + // Figure out if this is shifted by a constant or a register (for non-RRX). + if (Parser.getTok().is(AsmToken::Hash)) { + Parser.Lex(); // Eat hash. + SMLoc ImmLoc = Parser.getTok().getLoc(); + const MCExpr *ShiftExpr = 0; + if (getParser().ParseExpression(ShiftExpr)) { + Error(ImmLoc, "invalid immediate shift value"); + return -1; + } + // The expression must be evaluatable as an immediate. + const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ShiftExpr); + if (!CE) { + Error(ImmLoc, "invalid immediate shift value"); + return -1; + } + // Range check the immediate. + // lsl, ror: 0 <= imm <= 31 + // lsr, asr: 0 <= imm <= 32 + Imm = CE->getValue(); + if (Imm < 0 || + ((ShiftTy == ARM_AM::lsl || ShiftTy == ARM_AM::ror) && Imm > 31) || + ((ShiftTy == ARM_AM::lsr || ShiftTy == ARM_AM::asr) && Imm > 32)) { + Error(ImmLoc, "immediate shift value out of range"); + return -1; + } + } else if (Parser.getTok().is(AsmToken::Identifier)) { + ShiftReg = TryParseRegister(); + SMLoc L = Parser.getTok().getLoc(); + if (ShiftReg == -1) { + Error (L, "expected immediate or register in shift operand"); + return -1; + } + } else { + Error (Parser.getTok().getLoc(), + "expected immediate or register in shift operand"); + return -1; + } + } - Operands.push_back(ARMOperand::CreateReg(RegNum,S, Parser.getTok().getLoc())); - Operands.push_back(ARMOperand::CreateShifter(ShiftTy, + Operands.push_back(ARMOperand::CreateShiftedRegister(ShiftTy, SrcReg, + ShiftReg, Imm, S, Parser.getTok().getLoc())); - return false; + return 0; } @@ -1162,10 +1340,14 @@ tryParseMemBarrierOptOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { unsigned Opt = StringSwitch<unsigned>(OptStr.slice(0, OptStr.size())) .Case("sy", ARM_MB::SY) .Case("st", ARM_MB::ST) + .Case("sh", ARM_MB::ISH) .Case("ish", ARM_MB::ISH) + .Case("shst", ARM_MB::ISHST) .Case("ishst", ARM_MB::ISHST) .Case("nsh", ARM_MB::NSH) + .Case("un", ARM_MB::NSH) .Case("nshst", ARM_MB::NSHST) + .Case("unst", ARM_MB::NSHST) .Case("osh", ARM_MB::OSH) .Case("oshst", ARM_MB::OSHST) .Default(~0U); @@ -1604,15 +1786,18 @@ bool ARMAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands, default: Error(Parser.getTok().getLoc(), "unexpected token in operand"); return true; - case AsmToken::Identifier: + case AsmToken::Identifier: { if (!TryParseRegisterWithWriteBack(Operands)) return false; - if (!TryParseShiftRegister(Operands)) + int Res = TryParseShiftRegister(Operands); + if (Res == 0) // success return false; - + else if (Res == -1) // irrecoverable error + return true; // Fall though for the Identifier case that is not a register or a // special name. + } case AsmToken::Integer: // things like 1f and 2b as a branch targets case AsmToken::Dot: { // . as a branch target // This was not a register so parse other operands that start with an @@ -1761,30 +1946,35 @@ static StringRef SplitMnemonic(StringRef Mnemonic, Mnemonic == "vcle" || (Mnemonic == "smlal" || Mnemonic == "umaal" || Mnemonic == "umlal" || Mnemonic == "vabal" || Mnemonic == "vmlal" || Mnemonic == "vpadal" || - Mnemonic == "vqdmlal")) + Mnemonic == "vqdmlal" || Mnemonic == "bics")) return Mnemonic; - // First, split out any predication code. - unsigned CC = StringSwitch<unsigned>(Mnemonic.substr(Mnemonic.size()-2)) - .Case("eq", ARMCC::EQ) - .Case("ne", ARMCC::NE) - .Case("hs", ARMCC::HS) - .Case("lo", ARMCC::LO) - .Case("mi", ARMCC::MI) - .Case("pl", ARMCC::PL) - .Case("vs", ARMCC::VS) - .Case("vc", ARMCC::VC) - .Case("hi", ARMCC::HI) - .Case("ls", ARMCC::LS) - .Case("ge", ARMCC::GE) - .Case("lt", ARMCC::LT) - .Case("gt", ARMCC::GT) - .Case("le", ARMCC::LE) - .Case("al", ARMCC::AL) - .Default(~0U); - if (CC != ~0U) { - Mnemonic = Mnemonic.slice(0, Mnemonic.size() - 2); - PredicationCode = CC; + // First, split out any predication code. Ignore mnemonics we know aren't + // predicated but do have a carry-set and so weren't caught above. + if (Mnemonic != "adcs") { + unsigned CC = StringSwitch<unsigned>(Mnemonic.substr(Mnemonic.size()-2)) + .Case("eq", ARMCC::EQ) + .Case("ne", ARMCC::NE) + .Case("hs", ARMCC::HS) + .Case("cs", ARMCC::HS) + .Case("lo", ARMCC::LO) + .Case("cc", ARMCC::LO) + .Case("mi", ARMCC::MI) + .Case("pl", ARMCC::PL) + .Case("vs", ARMCC::VS) + .Case("vc", ARMCC::VC) + .Case("hi", ARMCC::HI) + .Case("ls", ARMCC::LS) + .Case("ge", ARMCC::GE) + .Case("lt", ARMCC::LT) + .Case("gt", ARMCC::GT) + .Case("le", ARMCC::LE) + .Case("al", ARMCC::AL) + .Default(~0U); + if (CC != ~0U) { + Mnemonic = Mnemonic.slice(0, Mnemonic.size() - 2); + PredicationCode = CC; + } } // Next, determine if we have a carry setting bit. We explicitly ignore all @@ -1824,8 +2014,6 @@ static StringRef SplitMnemonic(StringRef Mnemonic, void ARMAsmParser:: GetMnemonicAcceptInfo(StringRef Mnemonic, bool &CanAcceptCarrySet, bool &CanAcceptPredicationCode) { - bool isThumb = TM.getSubtarget<ARMSubtarget>().isThumb(); - if (Mnemonic == "and" || Mnemonic == "lsl" || Mnemonic == "lsr" || Mnemonic == "rrx" || Mnemonic == "ror" || Mnemonic == "sub" || Mnemonic == "smull" || Mnemonic == "add" || Mnemonic == "adc" || @@ -1834,7 +2022,7 @@ GetMnemonicAcceptInfo(StringRef Mnemonic, bool &CanAcceptCarrySet, Mnemonic == "rsb" || Mnemonic == "rsc" || Mnemonic == "orn" || Mnemonic == "sbc" || Mnemonic == "mla" || Mnemonic == "umull" || Mnemonic == "eor" || Mnemonic == "smlal" || - (Mnemonic == "mov" && !isThumb)) { + (Mnemonic == "mov" && !isThumbOne())) { CanAcceptCarrySet = true; } else { CanAcceptCarrySet = false; @@ -1851,10 +2039,9 @@ GetMnemonicAcceptInfo(StringRef Mnemonic, bool &CanAcceptCarrySet, CanAcceptPredicationCode = true; } - if (isThumb) + if (isThumb()) if (Mnemonic == "bkpt" || Mnemonic == "mcr" || Mnemonic == "mcrr" || - Mnemonic == "mrc" || Mnemonic == "mrrc" || Mnemonic == "cdp" || - Mnemonic == "mov") + Mnemonic == "mrc" || Mnemonic == "mrrc" || Mnemonic == "cdp") CanAcceptPredicationCode = false; } @@ -1884,20 +2071,22 @@ bool ARMAsmParser::ParseInstruction(StringRef Name, SMLoc NameLoc, bool CanAcceptCarrySet, CanAcceptPredicationCode; GetMnemonicAcceptInfo(Head, CanAcceptCarrySet, CanAcceptPredicationCode); + // If we had a carry-set on an instruction that can't do that, issue an + // error. + if (!CanAcceptCarrySet && CarrySetting) { + Parser.EatToEndOfStatement(); + return Error(NameLoc, "instruction '" + Head + + "' can not set flags, but 's' suffix specified"); + } + // Add the carry setting operand, if necessary. // // FIXME: It would be awesome if we could somehow invent a location such that // match errors on this operand would print a nice diagnostic about how the // 's' character in the mnemonic resulted in a CCOut operand. - if (CanAcceptCarrySet) { + if (CanAcceptCarrySet) Operands.push_back(ARMOperand::CreateCCOut(CarrySetting ? ARM::CPSR : 0, NameLoc)); - } else { - // This mnemonic can't ever accept a carry set, but the user wrote one (or - // misspelled another mnemonic). - - // FIXME: Issue a nice error. - } // Add the predication code operand, if necessary. if (CanAcceptPredicationCode) { @@ -1988,7 +2177,7 @@ MatchAndEmitInstruction(SMLoc IDLoc, // that updates the condition codes if it ends in 's'. So see if the // mnemonic ends in 's' and if so try removing the 's' and adding a CCOut // operand with a value of CPSR. - else if(MatchResult == Match_MnemonicFail) { + else if (MatchResult == Match_MnemonicFail) { // Get the instruction mnemonic, which is the first token. StringRef Mnemonic = ((ARMOperand*)Operands[0])->getToken(); if (Mnemonic.substr(Mnemonic.size()-1) == "s") { @@ -2174,20 +2363,15 @@ bool ARMAsmParser::ParseDirectiveCode(SMLoc L) { return Error(Parser.getTok().getLoc(), "unexpected token in directive"); Parser.Lex(); - // FIXME: We need to be able switch subtargets at this point so that - // MatchInstructionImpl() will work when it gets the AvailableFeatures which - // includes Feature_IsThumb or not to match the right instructions. This is - // blocked on the FIXME in llvm-mc.cpp when creating the TargetMachine. - if (Val == 16){ - assert(TM.getSubtarget<ARMSubtarget>().isThumb() && - "switching between arm/thumb not yet suppported via .code 16)"); + if (Val == 16) { + if (!isThumb()) + SwitchMode(); getParser().getStreamer().EmitAssemblerFlag(MCAF_Code16); - } - else{ - assert(!TM.getSubtarget<ARMSubtarget>().isThumb() && - "switching between thumb/arm not yet suppported via .code 32)"); + } else { + if (isThumb()) + SwitchMode(); getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32); - } + } return false; } diff --git a/lib/Target/ARM/CMakeLists.txt b/lib/Target/ARM/CMakeLists.txt index d3b8b54..21608d0 100644 --- a/lib/Target/ARM/CMakeLists.txt +++ b/lib/Target/ARM/CMakeLists.txt @@ -1,18 +1,16 @@ set(LLVM_TARGET_DEFINITIONS ARM.td) -tablegen(ARMGenRegisterInfo.h.inc -gen-register-desc-header) -tablegen(ARMGenRegisterNames.inc -gen-register-enums) -tablegen(ARMGenRegisterInfo.inc -gen-register-desc) -tablegen(ARMGenInstrNames.inc -gen-instr-enums) -tablegen(ARMGenInstrInfo.inc -gen-instr-desc) +tablegen(ARMGenRegisterInfo.inc -gen-register-info) +tablegen(ARMGenInstrInfo.inc -gen-instr-info) tablegen(ARMGenCodeEmitter.inc -gen-emitter) tablegen(ARMGenMCCodeEmitter.inc -gen-emitter -mc-emitter) +tablegen(ARMGenMCPseudoLowering.inc -gen-pseudo-lowering) tablegen(ARMGenAsmWriter.inc -gen-asm-writer) tablegen(ARMGenAsmMatcher.inc -gen-asm-matcher) tablegen(ARMGenDAGISel.inc -gen-dag-isel) tablegen(ARMGenFastISel.inc -gen-fast-isel) tablegen(ARMGenCallingConv.inc -gen-callingconv) -tablegen(ARMGenSubtarget.inc -gen-subtarget) +tablegen(ARMGenSubtargetInfo.inc -gen-subtarget) tablegen(ARMGenEDInfo.inc -gen-enhanced-disassembly-info) tablegen(ARMGenDecoderTables.inc -gen-arm-decoder) @@ -34,10 +32,10 @@ add_llvm_target(ARMCodeGen ARMISelLowering.cpp ARMInstrInfo.cpp ARMJITInfo.cpp + ARMMachObjectWriter.cpp ARMMCCodeEmitter.cpp ARMMCExpr.cpp ARMLoadStoreOptimizer.cpp - ARMMCAsmInfo.cpp ARMMCInstLower.cpp ARMRegisterInfo.cpp ARMSelectionDAGInfo.cpp @@ -67,3 +65,4 @@ add_subdirectory(TargetInfo) add_subdirectory(AsmParser) add_subdirectory(Disassembler) add_subdirectory(InstPrinter) +add_subdirectory(MCTargetDesc) diff --git a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp index 271ca8c..d89c80a 100644 --- a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp +++ b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp @@ -24,8 +24,8 @@ //#define DEBUG(X) do { X; } while (0) /// ARMGenInstrInfo.inc - ARMGenInstrInfo.inc contains the static const -/// TargetInstrDesc ARMInsts[] definition and the TargetOperandInfo[]'s -/// describing the operand info for each ARMInsts[i]. +/// MCInstrDesc ARMInsts[] definition and the MCOperandInfo[]'s describing the +/// operand info for each ARMInsts[i]. /// /// Together with an instruction's encoding format, we can take advantage of the /// NumOperands and the OpInfo fields of the target instruction description in @@ -46,10 +46,10 @@ /// dag DefaultOps = (ops (i32 14), (i32 zero_reg)); /// } /// -/// which is manifested by the TargetOperandInfo[] of: +/// which is manifested by the MCOperandInfo[] of: /// -/// { 0, 0|(1<<TOI::Predicate), 0 }, -/// { ARM::CCRRegClassID, 0|(1<<TOI::Predicate), 0 } +/// { 0, 0|(1<<MCOI::Predicate), 0 }, +/// { ARM::CCRRegClassID, 0|(1<<MCOI::Predicate), 0 } /// /// So the first predicate MCOperand corresponds to the immediate part of the /// ARM condition field (Inst{31-28}), and the second predicate MCOperand @@ -66,12 +66,14 @@ /// dag DefaultOps = (ops (i32 zero_reg)); /// } /// -/// which is manifested by the one TargetOperandInfo of: +/// which is manifested by the one MCOperandInfo of: /// -/// { ARM::CCRRegClassID, 0|(1<<TOI::OptionalDef), 0 } +/// { ARM::CCRRegClassID, 0|(1<<MCOI::OptionalDef), 0 } /// -/// And this maps to one MCOperand with the regsiter kind of ARM::CPSR. -#include "ARMGenInstrInfo.inc" + +namespace llvm { +extern MCInstrDesc ARMInsts[]; +} using namespace llvm; @@ -588,9 +590,9 @@ static bool BadRegsMulFrm(unsigned Opcode, uint32_t insn) { static bool DisassembleMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - unsigned short NumDefs = TID.getNumDefs(); - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + unsigned short NumDefs = MCID.getNumDefs(); + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -739,9 +741,9 @@ static bool DisassembleCoprocessor(MCInst &MI, unsigned Opcode, uint32_t insn, if (PW) { MI.addOperand(MCOperand::CreateReg(0)); ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; - const TargetInstrDesc &TID = ARMInsts[Opcode]; + const MCInstrDesc &MCID = ARMInsts[Opcode]; unsigned IndexMode = - (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; + (MCID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; unsigned Offset = ARM_AM::getAM2Opc(AddrOpcode, slice(insn, 7, 0) << 2, ARM_AM::no_shift, IndexMode); MI.addOperand(MCOperand::CreateImm(Offset)); @@ -802,7 +804,7 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, if (CoprocessorOpcode(Opcode)) return DisassembleCoprocessor(MI, Opcode, insn, NumOps, NumOpsAdded, B); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; // MRS and MRSsys take one GPR reg Rd. @@ -901,7 +903,7 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleBrMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; unsigned &OpIdx = NumOpsAdded; @@ -976,10 +978,10 @@ static bool BadRegsDPFrm(unsigned Opcode, uint32_t insn) { static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - unsigned short NumDefs = TID.getNumDefs(); - bool isUnary = isUnaryDP(TID.TSFlags); - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + unsigned short NumDefs = MCID.getNumDefs(); + bool isUnary = isUnaryDP(MCID.TSFlags); + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1041,7 +1043,7 @@ static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, } // If this is a two-address operand, skip it, e.g., MOVCCr operand 1. - if (isUnary && (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1)) { + if (isUnary && (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1)) { MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; } @@ -1089,10 +1091,10 @@ static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleDPSoRegFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - unsigned short NumDefs = TID.getNumDefs(); - bool isUnary = isUnaryDP(TID.TSFlags); - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + unsigned short NumDefs = MCID.getNumDefs(); + bool isUnary = isUnaryDP(MCID.TSFlags); + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1118,7 +1120,7 @@ static bool DisassembleDPSoRegFrm(MCInst &MI, unsigned Opcode, uint32_t insn, } // If this is a two-address operand, skip it, e.g., MOVCCs operand 1. - if (isUnary && (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1)) { + if (isUnary && (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1)) { MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; } @@ -1244,17 +1246,17 @@ static bool BadRegsLdStFrm(unsigned Opcode, uint32_t insn, bool Store, bool WBac static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool isStore, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - bool isPrePost = isPrePostLdSt(TID.TSFlags); - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + bool isPrePost = isPrePostLdSt(MCID.TSFlags); + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; - assert(((!isStore && TID.getNumDefs() > 0) || - (isStore && (TID.getNumDefs() == 0 || isPrePost))) + assert(((!isStore && MCID.getNumDefs() > 0) || + (isStore && (MCID.getNumDefs() == 0 || isPrePost))) && "Invalid arguments"); // Operand 0 of a pre- and post-indexed store is the address base writeback. @@ -1291,7 +1293,7 @@ static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && "Reg operand expected"); - assert((!isPrePost || (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1)) + assert((!isPrePost || (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1)) && "Index mode or tied_to operand expected"); MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)))); @@ -1308,7 +1310,7 @@ static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; unsigned IndexMode = - (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; + (MCID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; if (getIBit(insn) == 0) { // For pre- and post-indexed case, add a reg0 operand (Addressing Mode #2). // Otherwise, skip the reg operand since for addrmode_imm12, Rn has already @@ -1379,17 +1381,17 @@ static bool HasDualReg(unsigned Opcode) { static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool isStore, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - bool isPrePost = isPrePostLdSt(TID.TSFlags); - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + bool isPrePost = isPrePostLdSt(MCID.TSFlags); + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; - assert(((!isStore && TID.getNumDefs() > 0) || - (isStore && (TID.getNumDefs() == 0 || isPrePost))) + assert(((!isStore && MCID.getNumDefs() > 0) || + (isStore && (MCID.getNumDefs() == 0 || isPrePost))) && "Invalid arguments"); // Operand 0 of a pre- and post-indexed store is the address base writeback. @@ -1433,7 +1435,7 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && "Reg operand expected"); - assert((!isPrePost || (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1)) + assert((!isPrePost || (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1)) && "Offset mode or tied_to operand expected"); MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)))); @@ -1451,7 +1453,7 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; unsigned IndexMode = - (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; + (MCID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; if (getAM3IBit(insn) == 1) { MI.addOperand(MCOperand::CreateReg(0)); @@ -1539,7 +1541,7 @@ static bool DisassembleLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleLdStExFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; unsigned &OpIdx = NumOpsAdded; @@ -1591,7 +1593,7 @@ static bool DisassembleLdStExFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleArithMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1653,8 +1655,8 @@ static bool DisassembleSatFrm(MCInst &MI, unsigned Opcode, uint32_t insn, if (decodeRd(insn) == 15 || decodeRm(insn) == 15) return false; - const TargetInstrDesc &TID = ARMInsts[Opcode]; - NumOpsAdded = TID.getNumOperands() - 2; // ignore predicate operands + const MCInstrDesc &MCID = ARMInsts[Opcode]; + NumOpsAdded = MCID.getNumOperands() - 2; // ignore predicate operands // Disassemble register def. MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -1696,7 +1698,7 @@ static bool DisassembleExtFrm(MCInst &MI, unsigned Opcode, uint32_t insn, if (decodeRd(insn) == 15 || decodeRm(insn) == 15) return false; - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1802,7 +1804,7 @@ static bool DisassembleVFPUnaryFrm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(NumOps >= 1 && "VFPUnaryFrm expects NumOps >= 1"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1842,8 +1844,8 @@ static bool DisassembleVFPBinaryFrm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(NumOps >= 3 && "VFPBinaryFrm expects NumOps >= 3"); - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1858,7 +1860,7 @@ static bool DisassembleVFPBinaryFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ++OpIdx; // Skip tied_to operand constraint. - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) { + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) { assert(NumOps >= 4 && "Expect >=4 operands"); MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; @@ -1886,8 +1888,8 @@ static bool DisassembleVFPConv1Frm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(NumOps >= 2 && "VFPConv1Frm expects NumOps >= 2"); - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; bool SP = slice(insn, 8, 8) == 0; // A8.6.295 & A8.6.297 @@ -1903,7 +1905,7 @@ static bool DisassembleVFPConv1Frm(MCInst &MI, unsigned Opcode, uint32_t insn, getRegisterEnum(B, RegClassID, decodeVFPRd(insn, SP)))); - assert(TID.getOperandConstraint(1, TOI::TIED_TO) != -1 && + assert(MCID.getOperandConstraint(1, MCOI::TIED_TO) != -1 && "Tied to operand expected"); MI.addOperand(MI.getOperand(0)); @@ -1961,7 +1963,7 @@ static bool DisassembleVFPConv3Frm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(NumOps >= 3 && "VFPConv3Frm expects NumOps >= 3"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -2011,7 +2013,7 @@ static bool DisassembleVFPConv5Frm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(NumOps >= 3 && "VFPConv5Frm expects NumOps >= 3"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -2136,7 +2138,7 @@ static bool DisassembleVFPLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleVFPMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -2402,8 +2404,8 @@ static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool Store, bool DblSpaced, unsigned alignment, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; // At least one DPR register plus addressing mode #6. assert(NumOps >= 3 && "Expect >= 3 operands"); @@ -2507,7 +2509,7 @@ static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, } while (OpIdx < NumOps && (unsigned)OpInfo[OpIdx].RegClass == RegClass) { - assert(TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1 && + assert(MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1 && "Tied to operand expected"); MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; @@ -2757,8 +2759,8 @@ static bool DisassembleNLdSt(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleN1RegModImmFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; assert(NumOps >= 2 && (OpInfo[0].RegClass == ARM::DPRRegClassID || @@ -2848,8 +2850,8 @@ enum N2VFlag { static bool DisassembleNVdVmOptImm(MCInst &MI, unsigned Opc, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, N2VFlag Flag, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opc]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opc]; + const MCOperandInfo *OpInfo = MCID.OpInfo; assert(NumOps >= 2 && (OpInfo[0].RegClass == ARM::DPRRegClassID || @@ -2878,7 +2880,7 @@ static bool DisassembleNVdVmOptImm(MCInst &MI, unsigned Opc, uint32_t insn, ++OpIdx; // VPADAL... - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) { + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) { // TIED_TO operand. MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; @@ -2892,7 +2894,7 @@ static bool DisassembleNVdVmOptImm(MCInst &MI, unsigned Opc, uint32_t insn, // VZIP and others have two TIED_TO reg operands. int Idx; while (OpIdx < NumOps && - (Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + (Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // Add TIED_TO operand. MI.addOperand(MI.getOperand(Idx)); ++OpIdx; @@ -2945,8 +2947,8 @@ static bool DisassembleNVecDupLnFrm(MCInst &MI, unsigned Opc, uint32_t insn, static bool DisassembleNVectorShift(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool LeftShift, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; assert(NumOps >= 3 && (OpInfo[0].RegClass == ARM::DPRRegClassID || @@ -2964,7 +2966,7 @@ static bool DisassembleNVectorShift(MCInst &MI, unsigned Opcode, uint32_t insn, decodeNEONRd(insn)))); ++OpIdx; - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) { + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) { // TIED_TO operand. MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; @@ -3044,8 +3046,8 @@ enum N3VFlag { static bool DisassembleNVdVnVmOptImm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, N3VFlag Flag, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; // No checking for OpInfo[2] because of MOVDneon/MOVQ with only two regs. assert(NumOps >= 3 && @@ -3076,7 +3078,7 @@ static bool DisassembleNVdVnVmOptImm(MCInst &MI, unsigned Opcode, uint32_t insn, ++OpIdx; // VABA, VABAL, VBSLd, VBSLq, ... - if (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1) { + if (MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO) != -1) { // TIED_TO operand. MI.addOperand(MCOperand::CreateReg(0)); ++OpIdx; @@ -3091,11 +3093,6 @@ static bool DisassembleNVdVnVmOptImm(MCInst &MI, unsigned Opcode, uint32_t insn, : decodeNEONRm(insn)))); ++OpIdx; - // Special case handling for VMOVDneon and VMOVQ because they are marked as - // N3RegFrm. - if (Opcode == ARM::VMOVDneon || Opcode == ARM::VMOVQ) - return true; - // Dm = Inst{5:3-0} => NEON Rm // or // Dm is restricted to D0-D7 if size is 16, D0-D15 otherwise @@ -3163,8 +3160,8 @@ static bool DisassembleNVecMulScalarFrm(MCInst &MI, unsigned Opcode, static bool DisassembleNVTBLFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; assert(NumOps >= 3 && @@ -3192,7 +3189,7 @@ static bool DisassembleNVTBLFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // Process tied_to operand constraint. int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { MI.addOperand(MI.getOperand(Idx)); ++OpIdx; } @@ -3221,11 +3218,11 @@ static bool DisassembleNVTBLFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleNGetLnFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; - assert(TID.getNumDefs() == 1 && NumOps >= 3 && + assert(MCID.getNumDefs() == 1 && NumOps >= 3 && OpInfo[0].RegClass == ARM::GPRRegClassID && OpInfo[1].RegClass == ARM::DPRRegClassID && OpInfo[2].RegClass < 0 && @@ -3255,14 +3252,14 @@ static bool DisassembleNGetLnFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleNSetLnFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; if (!OpInfo) return false; - assert(TID.getNumDefs() == 1 && NumOps >= 3 && + assert(MCID.getNumDefs() == 1 && NumOps >= 3 && OpInfo[0].RegClass == ARM::DPRRegClassID && OpInfo[1].RegClass == ARM::DPRRegClassID && - TID.getOperandConstraint(1, TOI::TIED_TO) != -1 && + MCID.getOperandConstraint(1, MCOI::TIED_TO) != -1 && OpInfo[2].RegClass == ARM::GPRRegClassID && OpInfo[3].RegClass < 0 && "Expect >= 3 operands with one dst operand"); @@ -3294,7 +3291,7 @@ static bool DisassembleNSetLnFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleNDupFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; assert(NumOps >= 2 && (OpInfo[0].RegClass == ARM::DPRRegClassID || @@ -3379,7 +3376,7 @@ static bool DisassemblePreLoadFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - if (Opcode == ARM::DMB || Opcode == ARM::DSB) { + if (Opcode == ARM::DMB || Opcode == ARM::DSB || Opcode == ARM::ISB) { // Inst{3-0} encodes the memory barrier option for the variants. unsigned opt = slice(insn, 3, 0); switch (opt) { @@ -3604,11 +3601,11 @@ bool ARMBasicMCBuilder::DoPredicateOperands(MCInst& MI, unsigned Opcode, assert(NumOpsRemaining > 0 && "Invalid argument"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned Idx = MI.getNumOperands(); // First, we check whether this instr specifies the PredicateOperand through - // a pair of TargetOperandInfos with isPredicate() property. + // a pair of MCOperandInfos with isPredicate() property. if (NumOpsRemaining >= 2 && OpInfo[Idx].isPredicate() && OpInfo[Idx+1].isPredicate() && OpInfo[Idx].RegClass < 0 && @@ -3636,13 +3633,13 @@ bool ARMBasicMCBuilder::TryPredicateAndSBitModifier(MCInst& MI, unsigned Opcode, assert(NumOpsRemaining > 0 && "Invalid argument"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; const std::string &Name = ARMInsts[Opcode].Name; unsigned Idx = MI.getNumOperands(); uint64_t TSFlags = ARMInsts[Opcode].TSFlags; // First, we check whether this instr specifies the PredicateOperand through - // a pair of TargetOperandInfos with isPredicate() property. + // a pair of MCOperandInfos with isPredicate() property. if (NumOpsRemaining >= 2 && OpInfo[Idx].isPredicate() && OpInfo[Idx+1].isPredicate() && OpInfo[Idx].RegClass < 0 && diff --git a/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h b/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h index 9639c8a..834c6f6 100644 --- a/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h +++ b/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h @@ -350,7 +350,7 @@ static inline unsigned decodeRotate(uint32_t insn) { static bool DisassembleThumb1General(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -425,8 +425,8 @@ static bool DisassembleThumb1General(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb1DP(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -454,7 +454,7 @@ static bool DisassembleThumb1DP(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpIdx < NumOps && OpInfo[OpIdx].RegClass == ARM::tGPRRegClassID && "Thumb reg operand expected"); int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // The reg operand is tied to the first reg operand. MI.addOperand(MI.getOperand(Idx)); ++OpIdx; @@ -511,8 +511,8 @@ static bool DisassembleThumb1Special(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -530,7 +530,7 @@ static bool DisassembleThumb1Special(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpIdx < NumOps && "More operands expected"); int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // The reg operand is tied to the first reg operand. MI.addOperand(MI.getOperand(Idx)); ++OpIdx; @@ -554,7 +554,7 @@ static bool DisassembleThumb1Special(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb1LdPC(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 2 && OpInfo[0].RegClass == ARM::tGPRRegClassID && @@ -602,7 +602,7 @@ static bool DisassembleThumb1LdPC(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2Ldpci(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 2 && @@ -630,8 +630,8 @@ static bool DisassembleThumb2Ldpci(MCInst &MI, unsigned Opcode, static bool DisassembleThumb1LdSt(unsigned opA, MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; assert(NumOps >= 2 @@ -680,7 +680,7 @@ static bool DisassembleThumb1LdStSP(MCInst &MI, unsigned Opcode, uint32_t insn, assert((Opcode == ARM::tLDRspi || Opcode == ARM::tSTRspi) && "Unexpected opcode"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 3 && @@ -708,7 +708,7 @@ static bool DisassembleThumb1AddPCi(MCInst &MI, unsigned Opcode, uint32_t insn, assert(Opcode == ARM::tADDrPCi && "Unexpected opcode"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 2 && OpInfo[0].RegClass == ARM::tGPRRegClassID && @@ -733,7 +733,7 @@ static bool DisassembleThumb1AddSPi(MCInst &MI, unsigned Opcode, uint32_t insn, assert(Opcode == ARM::tADDrSPi && "Unexpected opcode"); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 3 && @@ -810,7 +810,7 @@ static bool DisassembleThumb1Misc(MCInst &MI, unsigned Opcode, uint32_t insn, if (Opcode == ARM::tPUSH || Opcode == ARM::tPOP) return DisassembleThumb1PushPop(MI, Opcode, insn, NumOps, NumOpsAdded, B); - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; // Predicate operands are handled elsewhere. if (NumOps == 2 && @@ -958,7 +958,7 @@ static bool DisassembleThumb1CondBr(MCInst &MI, unsigned Opcode, uint32_t insn, if (Opcode == ARM::tTRAP) return true; - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps == 3 && OpInfo[0].RegClass < 0 && @@ -989,7 +989,7 @@ static bool DisassembleThumb1CondBr(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb1Br(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps == 1 && OpInfo[0].RegClass < 0 && "1 imm operand expected"); @@ -1226,7 +1226,7 @@ static bool DisassembleThumb2LdStMul(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2LdStEx(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; unsigned &OpIdx = NumOpsAdded; @@ -1316,7 +1316,7 @@ static bool DisassembleThumb2LdStEx(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2LdStDual(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; if (!OpInfo) return false; assert(NumOps >= 4 @@ -1423,8 +1423,8 @@ static inline bool Thumb2ShiftOpcode(unsigned Opcode) { static bool DisassembleThumb2DPSoReg(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; // Special case handling. @@ -1467,7 +1467,7 @@ static bool DisassembleThumb2DPSoReg(MCInst &MI, unsigned Opcode, uint32_t insn, if (ThreeReg) { int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // Process tied_to operand constraint. MI.addOperand(MI.getOperand(Idx)); ++OpIdx; @@ -1521,8 +1521,8 @@ static bool DisassembleThumb2DPSoReg(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2DPModImm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1550,7 +1550,7 @@ static bool DisassembleThumb2DPModImm(MCInst &MI, unsigned Opcode, return false; } int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // The reg operand is tied to the first reg operand. MI.addOperand(MI.getOperand(Idx)); } else { @@ -1590,8 +1590,8 @@ static inline bool Thumb2SaturateOpcode(unsigned Opcode) { /// o t2SSAT16, t2USAT16: Rs sat_pos Rn static bool DisassembleThumb2Sat(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - NumOpsAdded = TID.getNumOperands() - 2; // ignore predicate operands + const MCInstrDesc &MCID = ARMInsts[Opcode]; + NumOpsAdded = MCID.getNumOperands() - 2; // ignore predicate operands // Disassemble the register def. MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::rGPRRegClassID, @@ -1635,8 +1635,8 @@ static bool DisassembleThumb2Sat(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2DPBinImm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1659,7 +1659,7 @@ static bool DisassembleThumb2DPBinImm(MCInst &MI, unsigned Opcode, if (TwoReg) { assert(NumOps >= 3 && "Expect >= 3 operands"); int Idx; - if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + if ((Idx = MCID.getOperandConstraint(OpIdx, MCOI::TIED_TO)) != -1) { // Process tied_to operand constraint. MI.addOperand(MI.getOperand(Idx)); } else { @@ -1907,8 +1907,8 @@ static bool DisassembleThumb2PreLoad(MCInst &MI, unsigned Opcode, uint32_t insn, // t2PLDs: Rn Rm imm2=Inst{5-4} // Same pattern applies for t2PLDW* and t2PLI*. - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -2073,8 +2073,8 @@ static bool DisassembleThumb2LdSt(bool Load, MCInst &MI, unsigned Opcode, // See, for example, A6.3.7 Load word: Table A6-18 Load word. if (Load && Rn == 15) return DisassembleThumb2Ldpci(MI, Opcode, insn, NumOps, NumOpsAdded, B); - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -2085,7 +2085,7 @@ static bool DisassembleThumb2LdSt(bool Load, MCInst &MI, unsigned Opcode, "Expect >= 3 operands and first two as reg operands"); bool ThreeReg = (OpInfo[2].RegClass > 0); - bool TIED_TO = ThreeReg && TID.getOperandConstraint(2, TOI::TIED_TO) != -1; + bool TIED_TO = ThreeReg && MCID.getOperandConstraint(2, MCOI::TIED_TO) != -1; bool Imm12 = !ThreeReg && slice(insn, 23, 23) == 1; // ARMInstrThumb2.td // Build the register operands, followed by the immediate. @@ -2160,8 +2160,8 @@ static bool DisassembleThumb2LdSt(bool Load, MCInst &MI, unsigned Opcode, static bool DisassembleThumb2DPReg(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetInstrDesc &TID = ARMInsts[Opcode]; - const TargetOperandInfo *OpInfo = TID.OpInfo; + const MCInstrDesc &MCID = ARMInsts[Opcode]; + const MCOperandInfo *OpInfo = MCID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -2214,7 +2214,7 @@ static bool DisassembleThumb2DPReg(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2Mul(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; assert(NumOps >= 3 && OpInfo[0].RegClass == ARM::rGPRRegClassID && @@ -2259,7 +2259,7 @@ static bool DisassembleThumb2Mul(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2LongMul(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const MCOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; assert(NumOps >= 3 && OpInfo[0].RegClass == ARM::rGPRRegClassID && diff --git a/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp b/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp index 8ae87f8..78d3e47 100644 --- a/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp +++ b/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp @@ -126,38 +126,6 @@ void ARMInstPrinter::printOperand(const MCInst *MI, unsigned OpNo, } } -static void printSOImm(raw_ostream &O, int64_t V, raw_ostream *CommentStream, - const MCAsmInfo *MAI) { - // Break it up into two parts that make up a shifter immediate. - V = ARM_AM::getSOImmVal(V); - assert(V != -1 && "Not a valid so_imm value!"); - - unsigned Imm = ARM_AM::getSOImmValImm(V); - unsigned Rot = ARM_AM::getSOImmValRot(V); - - // Print low-level immediate formation info, per - // A5.2.3: Data-processing (immediate), and - // A5.2.4: Modified immediate constants in ARM instructions - if (Rot) { - O << "#" << Imm << ", #" << Rot; - // Pretty printed version. - if (CommentStream) - *CommentStream << (int)ARM_AM::rotr32(Imm, Rot) << "\n"; - } else { - O << "#" << Imm; - } -} - - -/// printSOImmOperand - SOImm is 4-bit rotate amount in bits 8-11 with 8-bit -/// immediate in bits 0-7. -void ARMInstPrinter::printSOImmOperand(const MCInst *MI, unsigned OpNum, - raw_ostream &O) { - const MCOperand &MO = MI->getOperand(OpNum); - assert(MO.isImm() && "Not a valid so_imm value!"); - printSOImm(O, MO.getImm(), CommentStream, &MAI); -} - // so_reg is a 4-operand unit corresponding to register forms of the A5.1 // "Addressing Mode 1 - Data-processing operands" forms. This includes: // REG 0 0 - e.g. R5 @@ -174,6 +142,8 @@ void ARMInstPrinter::printSORegOperand(const MCInst *MI, unsigned OpNum, // Print the shift opc. ARM_AM::ShiftOpc ShOpc = ARM_AM::getSORegShOp(MO3.getImm()); O << ", " << ARM_AM::getShiftOpcStr(ShOpc); + if (ShOpc == ARM_AM::rrx) + return; if (MO2.getReg()) { O << ' ' << getRegisterName(MO2.getReg()); assert(ARM_AM::getSORegOffset(MO3.getImm()) == 0); diff --git a/lib/Target/ARM/InstPrinter/ARMInstPrinter.h b/lib/Target/ARM/InstPrinter/ARMInstPrinter.h index bde0eb9..d5f238b 100644 --- a/lib/Target/ARM/InstPrinter/ARMInstPrinter.h +++ b/lib/Target/ARM/InstPrinter/ARMInstPrinter.h @@ -19,11 +19,10 @@ namespace llvm { class MCOperand; -class TargetMachine; class ARMInstPrinter : public MCInstPrinter { public: - ARMInstPrinter(TargetMachine &TM, const MCAsmInfo &MAI) + ARMInstPrinter(const MCAsmInfo &MAI) : MCInstPrinter(MAI) {} virtual void printInst(const MCInst *MI, raw_ostream &O); @@ -39,8 +38,6 @@ public: void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O); - void printSOImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O); - void printSORegOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O); void printAddrMode2Operand(const MCInst *MI, unsigned OpNum, raw_ostream &O); diff --git a/lib/Target/ARM/ARMMCAsmInfo.cpp b/lib/Target/ARM/MCTargetDesc/ARMMCAsmInfo.cpp index 53b4c95..53b4c95 100644 --- a/lib/Target/ARM/ARMMCAsmInfo.cpp +++ b/lib/Target/ARM/MCTargetDesc/ARMMCAsmInfo.cpp diff --git a/lib/Target/ARM/ARMMCAsmInfo.h b/lib/Target/ARM/MCTargetDesc/ARMMCAsmInfo.h index 90f7822..90f7822 100644 --- a/lib/Target/ARM/ARMMCAsmInfo.h +++ b/lib/Target/ARM/MCTargetDesc/ARMMCAsmInfo.h diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp new file mode 100644 index 0000000..f8fcf2b --- /dev/null +++ b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp @@ -0,0 +1,144 @@ +//===-- ARMMCTargetDesc.cpp - ARM Target Descriptions -----------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file provides ARM specific target descriptions. +// +//===----------------------------------------------------------------------===// + +#include "ARMMCTargetDesc.h" +#include "ARMMCAsmInfo.h" +#include "llvm/MC/MCInstrInfo.h" +#include "llvm/MC/MCRegisterInfo.h" +#include "llvm/MC/MCSubtargetInfo.h" +#include "llvm/Target/TargetRegistry.h" + +#define GET_REGINFO_MC_DESC +#include "ARMGenRegisterInfo.inc" + +#define GET_INSTRINFO_MC_DESC +#include "ARMGenInstrInfo.inc" + +#define GET_SUBTARGETINFO_MC_DESC +#include "ARMGenSubtargetInfo.inc" + +using namespace llvm; + +std::string ARM_MC::ParseARMTriple(StringRef TT) { + // Set the boolean corresponding to the current target triple, or the default + // if one cannot be determined, to true. + unsigned Len = TT.size(); + unsigned Idx = 0; + + // FIXME: Enahnce Triple helper class to extract ARM version. + bool isThumb = false; + if (Len >= 5 && TT.substr(0, 4) == "armv") + Idx = 4; + else if (Len >= 6 && TT.substr(0, 5) == "thumb") { + isThumb = true; + if (Len >= 7 && TT[5] == 'v') + Idx = 6; + } + + std::string ARMArchFeature; + if (Idx) { + unsigned SubVer = TT[Idx]; + if (SubVer >= '7' && SubVer <= '9') { + if (Len >= Idx+2 && TT[Idx+1] == 'm') { + // v7m: FeatureNoARM, FeatureDB, FeatureHWDiv + ARMArchFeature = "+v7,+noarm,+db,+hwdiv"; + } else if (Len >= Idx+3 && TT[Idx+1] == 'e'&& TT[Idx+2] == 'm') { + // v7em: FeatureNoARM, FeatureDB, FeatureHWDiv, FeatureDSPThumb2, + // FeatureT2XtPk + ARMArchFeature = "+v7,+noarm,+db,+hwdiv,+t2dsp,t2xtpk"; + } else + // v7a: FeatureNEON, FeatureDB, FeatureDSPThumb2 + ARMArchFeature = "+v7,+neon,+db,+t2dsp"; + } else if (SubVer == '6') { + if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2') + ARMArchFeature = "+v6t2"; + else + ARMArchFeature = "+v6"; + } else if (SubVer == '5') { + if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e') + ARMArchFeature = "+v5te"; + else + ARMArchFeature = "+v5t"; + } else if (SubVer == '4' && Len >= Idx+2 && TT[Idx+1] == 't') + ARMArchFeature = "+v4t"; + } + + if (isThumb) { + if (ARMArchFeature.empty()) + ARMArchFeature = "+thumb-mode"; + else + ARMArchFeature += ",+thumb-mode"; + } + + return ARMArchFeature; +} + +MCSubtargetInfo *ARM_MC::createARMMCSubtargetInfo(StringRef TT, StringRef CPU, + StringRef FS) { + std::string ArchFS = ARM_MC::ParseARMTriple(TT); + if (!FS.empty()) { + if (!ArchFS.empty()) + ArchFS = ArchFS + "," + FS.str(); + else + ArchFS = FS; + } + + MCSubtargetInfo *X = new MCSubtargetInfo(); + InitARMMCSubtargetInfo(X, TT, CPU, ArchFS); + return X; +} + +// Force static initialization. +extern "C" void LLVMInitializeARMMCSubtargetInfo() { + TargetRegistry::RegisterMCSubtargetInfo(TheARMTarget, + ARM_MC::createARMMCSubtargetInfo); + TargetRegistry::RegisterMCSubtargetInfo(TheThumbTarget, + ARM_MC::createARMMCSubtargetInfo); +} + +static MCInstrInfo *createARMMCInstrInfo() { + MCInstrInfo *X = new MCInstrInfo(); + InitARMMCInstrInfo(X); + return X; +} + +extern "C" void LLVMInitializeARMMCInstrInfo() { + TargetRegistry::RegisterMCInstrInfo(TheARMTarget, createARMMCInstrInfo); + TargetRegistry::RegisterMCInstrInfo(TheThumbTarget, createARMMCInstrInfo); +} + +static MCRegisterInfo *createARMMCRegisterInfo() { + MCRegisterInfo *X = new MCRegisterInfo(); + InitARMMCRegisterInfo(X); + return X; +} + +extern "C" void LLVMInitializeARMMCRegInfo() { + TargetRegistry::RegisterMCRegInfo(TheARMTarget, createARMMCRegisterInfo); + TargetRegistry::RegisterMCRegInfo(TheThumbTarget, createARMMCRegisterInfo); +} + +static MCAsmInfo *createARMMCAsmInfo(const Target &T, StringRef TT) { + Triple TheTriple(TT); + + if (TheTriple.isOSDarwin()) + return new ARMMCAsmInfoDarwin(); + + return new ARMELFMCAsmInfo(); +} + +extern "C" void LLVMInitializeARMMCAsmInfo() { + // Register the target asm info. + RegisterMCAsmInfoFn A(TheARMTarget, createARMMCAsmInfo); + RegisterMCAsmInfoFn B(TheThumbTarget, createARMMCAsmInfo); +} diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.h b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.h new file mode 100644 index 0000000..74701e3 --- /dev/null +++ b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.h @@ -0,0 +1,52 @@ +//===-- ARMMCTargetDesc.h - ARM Target Descriptions -------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file provides ARM specific target descriptions. +// +//===----------------------------------------------------------------------===// + +#ifndef ARMMCTARGETDESC_H +#define ARMMCTARGETDESC_H + +#include <string> + +namespace llvm { +class MCSubtargetInfo; +class Target; +class StringRef; + +extern Target TheARMTarget, TheThumbTarget; + +namespace ARM_MC { + std::string ParseARMTriple(StringRef TT); + + /// createARMMCSubtargetInfo - Create a ARM MCSubtargetInfo instance. + /// This is exposed so Asm parser, etc. do not need to go through + /// TargetRegistry. + MCSubtargetInfo *createARMMCSubtargetInfo(StringRef TT, StringRef CPU, + StringRef FS); +} + +} // End llvm namespace + +// Defines symbolic names for ARM registers. This defines a mapping from +// register name to register number. +// +#define GET_REGINFO_ENUM +#include "ARMGenRegisterInfo.inc" + +// Defines symbolic names for the ARM instructions. +// +#define GET_INSTRINFO_ENUM +#include "ARMGenInstrInfo.inc" + +#define GET_SUBTARGETINFO_ENUM +#include "ARMGenSubtargetInfo.inc" + +#endif diff --git a/lib/Target/ARM/MCTargetDesc/CMakeLists.txt b/lib/Target/ARM/MCTargetDesc/CMakeLists.txt new file mode 100644 index 0000000..68daf42 --- /dev/null +++ b/lib/Target/ARM/MCTargetDesc/CMakeLists.txt @@ -0,0 +1,7 @@ +add_llvm_library(LLVMARMDesc + ARMMCTargetDesc.cpp + ARMMCAsmInfo.cpp + ) + +# Hack: we need to include 'main' target directory to grab private headers +include_directories(${CMAKE_CURRENT_SOURCE_DIR}/.. ${CMAKE_CURRENT_BINARY_DIR}/..) diff --git a/lib/Target/ARM/MCTargetDesc/Makefile b/lib/Target/ARM/MCTargetDesc/Makefile new file mode 100644 index 0000000..448ed9d --- /dev/null +++ b/lib/Target/ARM/MCTargetDesc/Makefile @@ -0,0 +1,16 @@ +##===- lib/Target/ARM/TargetDesc/Makefile ------------------*- Makefile -*-===## +# +# The LLVM Compiler Infrastructure +# +# This file is distributed under the University of Illinois Open Source +# License. See LICENSE.TXT for details. +# +##===----------------------------------------------------------------------===## + +LEVEL = ../../../.. +LIBRARYNAME = LLVMARMDesc + +# Hack: we need to include 'main' target directory to grab private headers +CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/.. + +include $(LEVEL)/Makefile.common diff --git a/lib/Target/ARM/MLxExpansionPass.cpp b/lib/Target/ARM/MLxExpansionPass.cpp index f6d0242..2df0053 100644 --- a/lib/Target/ARM/MLxExpansionPass.cpp +++ b/lib/Target/ARM/MLxExpansionPass.cpp @@ -137,11 +137,11 @@ unsigned MLxExpansion::getDefReg(MachineInstr *MI) const { bool MLxExpansion::hasRAWHazard(unsigned Reg, MachineInstr *MI) const { // FIXME: Detect integer instructions properly. - const TargetInstrDesc &TID = MI->getDesc(); - unsigned Domain = TID.TSFlags & ARMII::DomainMask; - if (TID.mayStore()) + const MCInstrDesc &MCID = MI->getDesc(); + unsigned Domain = MCID.TSFlags & ARMII::DomainMask; + if (MCID.mayStore()) return false; - unsigned Opcode = TID.getOpcode(); + unsigned Opcode = MCID.getOpcode(); if (Opcode == ARM::VMOVRS || Opcode == ARM::VMOVRRD) return false; if ((Domain & ARMII::DomainVFP) || (Domain & ARMII::DomainNEON)) @@ -218,18 +218,18 @@ MLxExpansion::ExpandFPMLxInstruction(MachineBasicBlock &MBB, MachineInstr *MI, ARMCC::CondCodes Pred = (ARMCC::CondCodes)MI->getOperand(NextOp).getImm(); unsigned PredReg = MI->getOperand(++NextOp).getReg(); - const TargetInstrDesc &TID1 = TII->get(MulOpc); - const TargetInstrDesc &TID2 = TII->get(AddSubOpc); - unsigned TmpReg = MRI->createVirtualRegister(TID1.getRegClass(0, TRI)); + const MCInstrDesc &MCID1 = TII->get(MulOpc); + const MCInstrDesc &MCID2 = TII->get(AddSubOpc); + unsigned TmpReg = MRI->createVirtualRegister(TII->getRegClass(MCID1, 0, TRI)); - MachineInstrBuilder MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), TID1, TmpReg) + MachineInstrBuilder MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), MCID1, TmpReg) .addReg(Src1Reg, getKillRegState(Src1Kill)) .addReg(Src2Reg, getKillRegState(Src2Kill)); if (HasLane) MIB.addImm(LaneImm); MIB.addImm(Pred).addReg(PredReg); - MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), TID2) + MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), MCID2) .addReg(DstReg, getDefRegState(true) | getDeadRegState(DstDead)); if (NegAcc) { @@ -273,15 +273,15 @@ bool MLxExpansion::ExpandFPMLxInstructions(MachineBasicBlock &MBB) { continue; } - const TargetInstrDesc &TID = MI->getDesc(); - if (TID.isBarrier()) { + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.isBarrier()) { clearStack(); Skip = 0; ++MII; continue; } - unsigned Domain = TID.TSFlags & ARMII::DomainMask; + unsigned Domain = MCID.TSFlags & ARMII::DomainMask; if (Domain == ARMII::DomainGeneral) { if (++Skip == 2) // Assume dual issues of non-VFP / NEON instructions. @@ -291,7 +291,7 @@ bool MLxExpansion::ExpandFPMLxInstructions(MachineBasicBlock &MBB) { unsigned MulOpc, AddSubOpc; bool NegAcc, HasLane; - if (!TII->isFpMLxInstruction(TID.getOpcode(), + if (!TII->isFpMLxInstruction(MCID.getOpcode(), MulOpc, AddSubOpc, NegAcc, HasLane) || !FindMLxHazard(MI)) pushStack(MI); diff --git a/lib/Target/ARM/Makefile b/lib/Target/ARM/Makefile index 65a6494..eb8c603 100644 --- a/lib/Target/ARM/Makefile +++ b/lib/Target/ARM/Makefile @@ -12,14 +12,14 @@ LIBRARYNAME = LLVMARMCodeGen TARGET = ARM # Make sure that tblgen is run, first thing. -BUILT_SOURCES = ARMGenRegisterInfo.h.inc ARMGenRegisterNames.inc \ - ARMGenRegisterInfo.inc ARMGenInstrNames.inc \ - ARMGenInstrInfo.inc ARMGenAsmWriter.inc ARMGenAsmMatcher.inc \ - ARMGenDAGISel.inc ARMGenSubtarget.inc \ +BUILT_SOURCES = ARMGenRegisterInfo.inc ARMGenInstrInfo.inc \ + ARMGenAsmWriter.inc ARMGenAsmMatcher.inc \ + ARMGenDAGISel.inc ARMGenSubtargetInfo.inc \ ARMGenCodeEmitter.inc ARMGenCallingConv.inc \ ARMGenDecoderTables.inc ARMGenEDInfo.inc \ - ARMGenFastISel.inc ARMGenMCCodeEmitter.inc + ARMGenFastISel.inc ARMGenMCCodeEmitter.inc \ + ARMGenMCPseudoLowering.inc -DIRS = InstPrinter AsmParser Disassembler TargetInfo +DIRS = InstPrinter AsmParser Disassembler TargetInfo MCTargetDesc include $(LEVEL)/Makefile.common diff --git a/lib/Target/ARM/NEONMoveFix.cpp b/lib/Target/ARM/NEONMoveFix.cpp index 965665c..c85d1e9 100644 --- a/lib/Target/ARM/NEONMoveFix.cpp +++ b/lib/Target/ARM/NEONMoveFix.cpp @@ -77,7 +77,7 @@ bool NEONMoveFixPass::InsertMoves(MachineBasicBlock &MBB) { } if (inNEONDomain(Domain, isA8)) { - // Convert VMOVD to VMOVDneon + // Convert VMOVD to VORRd unsigned DestReg = MI->getOperand(0).getReg(); DEBUG({errs() << "vmov convert: "; MI->dump();}); @@ -88,7 +88,8 @@ bool NEONMoveFixPass::InsertMoves(MachineBasicBlock &MBB) { // - The imp-defs / imp-uses are superregs only, we don't care about // them. AddDefaultPred(BuildMI(MBB, *MI, MI->getDebugLoc(), - TII->get(ARM::VMOVDneon), DestReg).addReg(SrcReg)); + TII->get(ARM::VORRd), DestReg) + .addReg(SrcReg).addReg(SrcReg)); MBB.erase(MI); MachineBasicBlock::iterator I = prior(NextMII); MI = &*I; diff --git a/lib/Target/ARM/README.txt b/lib/Target/ARM/README.txt index 8ba9a27..2f6842e 100644 --- a/lib/Target/ARM/README.txt +++ b/lib/Target/ARM/README.txt @@ -681,3 +681,21 @@ is compiled and optimized to: str r1, [r0] //===---------------------------------------------------------------------===// + +Improve codegen for select's: +if (x != 0) x = 1 +if (x == 1) x = 1 + +ARM codegen used to look like this: + mov r1, r0 + cmp r1, #1 + mov r0, #0 + moveq r0, #1 + +The naive lowering select between two different values. It should recognize the +test is equality test so it's more a conditional move rather than a select: + cmp r0, #1 + movne r0, #0 + +Currently this is a ARM specific dag combine. We probably should make it into a +target-neutral one. diff --git a/lib/Target/ARM/Thumb1FrameLowering.cpp b/lib/Target/ARM/Thumb1FrameLowering.cpp index dee3d27..c258870 100644 --- a/lib/Target/ARM/Thumb1FrameLowering.cpp +++ b/lib/Target/ARM/Thumb1FrameLowering.cpp @@ -136,8 +136,8 @@ void Thumb1FrameLowering::emitPrologue(MachineFunction &MF) const { BuildMI(MBB, MBBI, dl, TII.get(ARM::tADDrSPi), FramePtr) .addFrameIndex(FramePtrSpillFI).addImm(0) .setMIFlags(MachineInstr::FrameSetup); - if (NumBytes > 7) - // If offset is > 7 then sp cannot be adjusted in a single instruction, + if (NumBytes > 508) + // If offset is > 508 then sp cannot be adjusted in a single instruction, // try restoring from fp instead. AFI->setShouldRestoreSPFromFP(true); } @@ -160,7 +160,8 @@ void Thumb1FrameLowering::emitPrologue(MachineFunction &MF) const { // will be allocated after this, so we can still use the base pointer // to reference locals. if (RegInfo->hasBasePointer(MF)) - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), BasePtr).addReg(ARM::SP); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), BasePtr) + .addReg(ARM::SP)); // If the frame has variable sized objects then the epilogue must restore // the sp from fp. We can assume there's an FP here since hasFP already @@ -177,7 +178,7 @@ static bool isCalleeSavedRegister(unsigned Reg, const unsigned *CSRegs) { } static bool isCSRestore(MachineInstr *MI, const unsigned *CSRegs) { - if (MI->getOpcode() == ARM::tRestore && + if (MI->getOpcode() == ARM::tLDRspi && MI->getOperand(1).isFI() && isCalleeSavedRegister(MI->getOperand(0).getReg(), CSRegs)) return true; @@ -239,11 +240,13 @@ void Thumb1FrameLowering::emitEpilogue(MachineFunction &MF, "No scratch register to restore SP from FP!"); emitThumbRegPlusImmediate(MBB, MBBI, dl, ARM::R4, FramePtr, -NumBytes, TII, *RegInfo); - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP) - .addReg(ARM::R4); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), + ARM::SP) + .addReg(ARM::R4)); } else - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP) - .addReg(FramePtr); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), + ARM::SP) + .addReg(FramePtr)); } else { if (MBBI->getOpcode() == ARM::tBX_RET && &MBB.front() != MBBI && @@ -270,8 +273,8 @@ void Thumb1FrameLowering::emitEpilogue(MachineFunction &MF, emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, VARegSaveSize); - BuildMI(MBB, MBBI, dl, TII.get(ARM::tBX_RET_vararg)) - .addReg(ARM::R3, RegState::Kill); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tBX_RET_vararg)) + .addReg(ARM::R3, RegState::Kill)); // erase the old tBX_RET instruction MBB.erase(MBBI); } diff --git a/lib/Target/ARM/Thumb1InstrInfo.cpp b/lib/Target/ARM/Thumb1InstrInfo.cpp index 3fbb433..218311d 100644 --- a/lib/Target/ARM/Thumb1InstrInfo.cpp +++ b/lib/Target/ARM/Thumb1InstrInfo.cpp @@ -13,7 +13,6 @@ #include "Thumb1InstrInfo.h" #include "ARM.h" -#include "ARMGenInstrInfo.inc" #include "ARMMachineFunctionInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineInstrBuilder.h" @@ -37,18 +36,8 @@ void Thumb1InstrInfo::copyPhysReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL, unsigned DestReg, unsigned SrcReg, bool KillSrc) const { - bool tDest = ARM::tGPRRegClass.contains(DestReg); - bool tSrc = ARM::tGPRRegClass.contains(SrcReg); - unsigned Opc = ARM::tMOVgpr2gpr; - if (tDest && tSrc) - Opc = ARM::tMOVr; - else if (tSrc) - Opc = ARM::tMOVtgpr2gpr; - else if (tDest) - Opc = ARM::tMOVgpr2tgpr; - - BuildMI(MBB, I, DL, get(Opc), DestReg) - .addReg(SrcReg, getKillRegState(KillSrc)); + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tMOVr), DestReg) + .addReg(SrcReg, getKillRegState(KillSrc))); assert(ARM::GPRRegClass.contains(DestReg, SrcReg) && "Thumb1 can only copy GPR registers"); } @@ -76,7 +65,7 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, MachineMemOperand::MOStore, MFI.getObjectSize(FI), MFI.getObjectAlignment(FI)); - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tSpill)) + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tSTRspi)) .addReg(SrcReg, getKillRegState(isKill)) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); } @@ -105,7 +94,7 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, MachineMemOperand::MOLoad, MFI.getObjectSize(FI), MFI.getObjectAlignment(FI)); - AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tRestore), DestReg) + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tLDRspi), DestReg) .addFrameIndex(FI).addImm(0).addMemOperand(MMO)); } } diff --git a/lib/Target/ARM/Thumb1RegisterInfo.cpp b/lib/Target/ARM/Thumb1RegisterInfo.cpp index 6bf5650..4eb0b6c 100644 --- a/lib/Target/ARM/Thumb1RegisterInfo.cpp +++ b/lib/Target/ARM/Thumb1RegisterInfo.cpp @@ -239,13 +239,13 @@ void llvm::emitThumbRegPlusImmediate(MachineBasicBlock &MBB, unsigned Chunk = (1 << 3) - 1; unsigned ThisVal = (Bytes > Chunk) ? Chunk : Bytes; Bytes -= ThisVal; - const TargetInstrDesc &TID = TII.get(isSub ? ARM::tSUBi3 : ARM::tADDi3); + const MCInstrDesc &MCID = TII.get(isSub ? ARM::tSUBi3 : ARM::tADDi3); const MachineInstrBuilder MIB = - AddDefaultT1CC(BuildMI(MBB, MBBI, dl, TID, DestReg).setMIFlags(MIFlags)); + AddDefaultT1CC(BuildMI(MBB, MBBI, dl, MCID, DestReg).setMIFlags(MIFlags)); AddDefaultPred(MIB.addReg(BaseReg, RegState::Kill).addImm(ThisVal)); } else { - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), DestReg) - .addReg(BaseReg, RegState::Kill) + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), DestReg) + .addReg(BaseReg, RegState::Kill)) .setMIFlags(MIFlags); } BaseReg = DestReg; @@ -291,8 +291,8 @@ void llvm::emitThumbRegPlusImmediate(MachineBasicBlock &MBB, } if (ExtraOpc) { - const TargetInstrDesc &TID = TII.get(ExtraOpc); - AddDefaultPred(AddDefaultT1CC(BuildMI(MBB, MBBI, dl, TID, DestReg)) + const MCInstrDesc &MCID = TII.get(ExtraOpc); + AddDefaultPred(AddDefaultT1CC(BuildMI(MBB, MBBI, dl, MCID, DestReg)) .addReg(DestReg, RegState::Kill) .addImm(((unsigned)NumBytes) & 3) .setMIFlags(MIFlags)); @@ -360,8 +360,8 @@ static void emitThumbConstant(MachineBasicBlock &MBB, if (Imm > 0) emitThumbRegPlusImmediate(MBB, MBBI, dl, DestReg, DestReg, Imm, TII, MRI); if (isSub) { - const TargetInstrDesc &TID = TII.get(ARM::tRSB); - AddDefaultPred(AddDefaultT1CC(BuildMI(MBB, MBBI, dl, TID, DestReg)) + const MCInstrDesc &MCID = TII.get(ARM::tRSB); + AddDefaultPred(AddDefaultT1CC(BuildMI(MBB, MBBI, dl, MCID, DestReg)) .addReg(DestReg, RegState::Kill)); } } @@ -377,11 +377,9 @@ static void removeOperands(MachineInstr &MI, unsigned i) { static unsigned convertToNonSPOpcode(unsigned Opcode) { switch (Opcode) { case ARM::tLDRspi: - case ARM::tRestore: // FIXME: Should this opcode be here? return ARM::tLDRi; case ARM::tSTRspi: - case ARM::tSpill: // FIXME: Should this opcode be here? return ARM::tSTRi; } @@ -396,7 +394,7 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx, MachineBasicBlock &MBB = *MI.getParent(); DebugLoc dl = MI.getDebugLoc(); unsigned Opcode = MI.getOpcode(); - const TargetInstrDesc &Desc = MI.getDesc(); + const MCInstrDesc &Desc = MI.getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); if (Opcode == ARM::tADDrSPi) { @@ -419,13 +417,12 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx, unsigned PredReg; if (Offset == 0 && getInstrPredicate(&MI, PredReg) == ARMCC::AL) { // Turn it into a move. - MI.setDesc(TII.get(ARM::tMOVgpr2tgpr)); + MI.setDesc(TII.get(ARM::tMOVr)); MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false); - // Remove offset and remaining explicit predicate operands. - do MI.RemoveOperand(FrameRegIdx+1); - while (MI.getNumOperands() > FrameRegIdx+1 && - (!MI.getOperand(FrameRegIdx+1).isReg() || - !MI.getOperand(FrameRegIdx+1).isImm())); + // Remove offset and add predicate operands. + MI.RemoveOperand(FrameRegIdx+1); + MachineInstrBuilder MIB(&MI); + AddDefaultPred(MIB); return true; } @@ -524,7 +521,7 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx, // If this is a thumb spill / restore, we will be using a constpool load to // materialize the offset. - if (Opcode == ARM::tRestore || Opcode == ARM::tSpill) { + if (Opcode == ARM::tLDRspi || Opcode == ARM::tSTRspi) { ImmOp.ChangeToImmediate(0); } else { // Otherwise, it didn't fit. Pull in what we can to simplify the immed. @@ -567,8 +564,9 @@ Thumb1RegisterInfo::saveScavengerRegister(MachineBasicBlock &MBB, // the function, the offset will be negative. Use R12 instead since that's // a call clobbered register that we know won't be used in Thumb1 mode. DebugLoc DL; - BuildMI(MBB, I, DL, TII.get(ARM::tMOVtgpr2gpr)). - addReg(ARM::R12, RegState::Define).addReg(Reg, RegState::Kill); + AddDefaultPred(BuildMI(MBB, I, DL, TII.get(ARM::tMOVr)) + .addReg(ARM::R12, RegState::Define) + .addReg(Reg, RegState::Kill)); // The UseMI is where we would like to restore the register. If there's // interference with R12 before then, however, we'll need to restore it @@ -591,8 +589,8 @@ Thumb1RegisterInfo::saveScavengerRegister(MachineBasicBlock &MBB, } } // Restore the register from R12 - BuildMI(MBB, UseMI, DL, TII.get(ARM::tMOVgpr2tgpr)). - addReg(Reg, RegState::Define).addReg(ARM::R12, RegState::Kill); + AddDefaultPred(BuildMI(MBB, UseMI, DL, TII.get(ARM::tMOVr)). + addReg(Reg, RegState::Define).addReg(ARM::R12, RegState::Kill)); return true; } @@ -653,7 +651,7 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, assert(Offset && "This code isn't needed if offset already handled!"); unsigned Opcode = MI.getOpcode(); - const TargetInstrDesc &Desc = MI.getDesc(); + const MCInstrDesc &Desc = MI.getDesc(); // Remove predicate first. int PIdx = MI.findFirstPredOperandIdx(); @@ -664,7 +662,7 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, // Use the destination register to materialize sp + offset. unsigned TmpReg = MI.getOperand(0).getReg(); bool UseRR = false; - if (Opcode == ARM::tRestore) { + if (Opcode == ARM::tLDRspi) { if (FrameReg == ARM::SP) emitThumbRegPlusImmInReg(MBB, II, dl, TmpReg, FrameReg, Offset, false, TII, *this); @@ -687,7 +685,7 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, VReg = MF.getRegInfo().createVirtualRegister(ARM::tGPRRegisterClass); bool UseRR = false; - if (Opcode == ARM::tSpill) { + if (Opcode == ARM::tSTRspi) { if (FrameReg == ARM::SP) emitThumbRegPlusImmInReg(MBB, II, dl, VReg, FrameReg, Offset, false, TII, *this); diff --git a/lib/Target/ARM/Thumb2ITBlockPass.cpp b/lib/Target/ARM/Thumb2ITBlockPass.cpp index 45e6937..360ec00 100644 --- a/lib/Target/ARM/Thumb2ITBlockPass.cpp +++ b/lib/Target/ARM/Thumb2ITBlockPass.cpp @@ -98,9 +98,6 @@ static bool isCopy(MachineInstr *MI) { case ARM::MOVr: case ARM::MOVr_TC: case ARM::tMOVr: - case ARM::tMOVgpr2tgpr: - case ARM::tMOVtgpr2gpr: - case ARM::tMOVgpr2gpr: case ARM::t2MOVr: return true; } diff --git a/lib/Target/ARM/Thumb2InstrInfo.cpp b/lib/Target/ARM/Thumb2InstrInfo.cpp index d169dbb..51b56aa 100644 --- a/lib/Target/ARM/Thumb2InstrInfo.cpp +++ b/lib/Target/ARM/Thumb2InstrInfo.cpp @@ -15,7 +15,6 @@ #include "ARM.h" #include "ARMConstantPoolValue.h" #include "ARMAddressingModes.h" -#include "ARMGenInstrInfo.inc" #include "ARMMachineFunctionInfo.h" #include "Thumb2InstrInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" @@ -113,18 +112,8 @@ void Thumb2InstrInfo::copyPhysReg(MachineBasicBlock &MBB, if (!ARM::GPRRegClass.contains(DestReg, SrcReg)) return ARMBaseInstrInfo::copyPhysReg(MBB, I, DL, DestReg, SrcReg, KillSrc); - bool tDest = ARM::tGPRRegClass.contains(DestReg); - bool tSrc = ARM::tGPRRegClass.contains(SrcReg); - unsigned Opc = ARM::tMOVgpr2gpr; - if (tDest && tSrc) - Opc = ARM::tMOVr; - else if (tSrc) - Opc = ARM::tMOVtgpr2gpr; - else if (tDest) - Opc = ARM::tMOVgpr2tgpr; - - BuildMI(MBB, I, DL, get(Opc), DestReg) - .addReg(SrcReg, getKillRegState(KillSrc)); + AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::tMOVr), DestReg) + .addReg(SrcReg, getKillRegState(KillSrc))); } void Thumb2InstrInfo:: @@ -232,8 +221,8 @@ void llvm::emitT2RegPlusImmediate(MachineBasicBlock &MBB, unsigned Opc = 0; if (DestReg == ARM::SP && BaseReg != ARM::SP) { // mov sp, rn. Note t2MOVr cannot be used. - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr),DestReg) - .addReg(BaseReg).setMIFlags(MIFlags); + AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr),DestReg) + .addReg(BaseReg).setMIFlags(MIFlags)); BaseReg = ARM::SP; continue; } @@ -252,7 +241,7 @@ void llvm::emitT2RegPlusImmediate(MachineBasicBlock &MBB, } // sub rd, sp, so_imm - Opc = isSub ? ARM::t2SUBrSPi : ARM::t2ADDrSPi; + Opc = isSub ? ARM::t2SUBri : ARM::t2ADDri; if (ARM_AM::getT2SOImmVal(NumBytes) != -1) { NumBytes = 0; } else { @@ -396,7 +385,7 @@ bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx, unsigned FrameReg, int &Offset, const ARMBaseInstrInfo &TII) { unsigned Opcode = MI.getOpcode(); - const TargetInstrDesc &Desc = MI.getDesc(); + const MCInstrDesc &Desc = MI.getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); bool isSub = false; @@ -410,25 +399,24 @@ bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx, unsigned PredReg; if (Offset == 0 && getInstrPredicate(&MI, PredReg) == ARMCC::AL) { // Turn it into a move. - MI.setDesc(TII.get(ARM::tMOVgpr2gpr)); + MI.setDesc(TII.get(ARM::tMOVr)); MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false); // Remove offset and remaining explicit predicate operands. do MI.RemoveOperand(FrameRegIdx+1); - while (MI.getNumOperands() > FrameRegIdx+1 && - (!MI.getOperand(FrameRegIdx+1).isReg() || - !MI.getOperand(FrameRegIdx+1).isImm())); + while (MI.getNumOperands() > FrameRegIdx+1); + MachineInstrBuilder MIB(&MI); + AddDefaultPred(MIB); return true; } - bool isSP = FrameReg == ARM::SP; bool HasCCOut = Opcode != ARM::t2ADDri12; if (Offset < 0) { Offset = -Offset; isSub = true; - MI.setDesc(TII.get(isSP ? ARM::t2SUBrSPi : ARM::t2SUBri)); + MI.setDesc(TII.get(ARM::t2SUBri)); } else { - MI.setDesc(TII.get(isSP ? ARM::t2ADDrSPi : ARM::t2ADDri)); + MI.setDesc(TII.get(ARM::t2ADDri)); } // Common case: small offset, fits into instruction. @@ -444,9 +432,7 @@ bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx, // Another common case: imm12. if (Offset < 4096 && (!HasCCOut || MI.getOperand(MI.getNumOperands()-1).getReg() == 0)) { - unsigned NewOpc = isSP - ? (isSub ? ARM::t2SUBrSPi12 : ARM::t2ADDrSPi12) - : (isSub ? ARM::t2SUBri12 : ARM::t2ADDri12); + unsigned NewOpc = isSub ? ARM::t2SUBri12 : ARM::t2ADDri12; MI.setDesc(TII.get(NewOpc)); MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false); MI.getOperand(FrameRegIdx+1).ChangeToImmediate(Offset); @@ -579,8 +565,7 @@ void Thumb2InstrInfo::scheduleTwoAddrSource(MachineInstr *SrcMI, MachineInstr *UseMI, const TargetRegisterInfo &TRI) const { - if (SrcMI->getOpcode() != ARM::tMOVgpr2gpr || - SrcMI->getOperand(1).isKill()) + if (SrcMI->getOpcode() != ARM::tMOVr || SrcMI->getOperand(1).isKill()) return; unsigned PredReg = 0; diff --git a/lib/Target/ARM/Thumb2SizeReduction.cpp b/lib/Target/ARM/Thumb2SizeReduction.cpp index ce2e966..c741a6e 100644 --- a/lib/Target/ARM/Thumb2SizeReduction.cpp +++ b/lib/Target/ARM/Thumb2SizeReduction.cpp @@ -57,10 +57,8 @@ namespace { static const ReduceEntry ReduceTable[] = { // Wide, Narrow1, Narrow2, imm1,imm2, lo1, lo2, P/C, PF, S { ARM::t2ADCrr, 0, ARM::tADC, 0, 0, 0, 1, 0,0, 0,0 }, - { ARM::t2ADDri, ARM::tADDi3, ARM::tADDi8, 3, 8, 1, 1, 0,0, 0,0 }, + { ARM::t2ADDri, ARM::tADDi3, ARM::tADDi8, 3, 8, 1, 1, 0,0, 0,1 }, { ARM::t2ADDrr, ARM::tADDrr, ARM::tADDhirr, 0, 0, 1, 0, 0,1, 0,0 }, - // Note: immediate scale is 4. - { ARM::t2ADDrSPi,ARM::tADDrSPi,0, 8, 0, 1, 0, 1,0, 0,1 }, { ARM::t2ADDSri,ARM::tADDi3, ARM::tADDi8, 3, 8, 1, 1, 2,2, 0,1 }, { ARM::t2ADDSrr,ARM::tADDrr, 0, 0, 0, 1, 0, 2,0, 0,1 }, { ARM::t2ANDrr, 0, ARM::tAND, 0, 0, 0, 1, 0,0, 1,0 }, @@ -84,9 +82,7 @@ namespace { { ARM::t2MOVi, ARM::tMOVi8, 0, 8, 0, 1, 0, 0,0, 0,0 }, { ARM::t2MOVi16,ARM::tMOVi8, 0, 8, 0, 1, 0, 0,0, 0,1 }, // FIXME: Do we need the 16-bit 'S' variant? - { ARM::t2MOVr,ARM::tMOVgpr2gpr,0, 0, 0, 0, 0, 1,0, 0,0 }, - { ARM::t2MOVCCr,0, ARM::tMOVCCr, 0, 0, 0, 0, 0,1, 0,0 }, - { ARM::t2MOVCCi,0, ARM::tMOVCCi, 0, 8, 0, 1, 0,1, 0,0 }, + { ARM::t2MOVr,ARM::tMOVr, 0, 0, 0, 0, 0, 1,0, 0,0 }, { ARM::t2MUL, 0, ARM::tMUL, 0, 0, 0, 1, 0,0, 1,0 }, { ARM::t2MVNr, ARM::tMVN, 0, 0, 0, 1, 0, 0,0, 0,0 }, { ARM::t2ORRrr, 0, ARM::tORR, 0, 0, 0, 1, 0,0, 1,0 }, @@ -189,8 +185,8 @@ Thumb2SizeReduce::Thumb2SizeReduce() : MachineFunctionPass(ID) { } } -static bool HasImplicitCPSRDef(const TargetInstrDesc &TID) { - for (const unsigned *Regs = TID.ImplicitDefs; *Regs; ++Regs) +static bool HasImplicitCPSRDef(const MCInstrDesc &MCID) { + for (const unsigned *Regs = MCID.ImplicitDefs; *Regs; ++Regs) if (*Regs == ARM::CPSR) return true; return false; @@ -291,7 +287,7 @@ static bool VerifyLowRegs(MachineInstr *MI) { Opc == ARM::t2LDMDB || Opc == ARM::t2LDMIA_UPD || Opc == ARM::t2LDMDB_UPD); bool isLROk = (Opc == ARM::t2STMIA_UPD || Opc == ARM::t2STMDB_UPD); - bool isSPOk = isPCOk || isLROk || (Opc == ARM::t2ADDrSPi); + bool isSPOk = isPCOk || isLROk; for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { const MachineOperand &MO = MI->getOperand(i); if (!MO.isReg() || MO.isImplicit()) @@ -481,14 +477,54 @@ bool Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI, const ReduceEntry &Entry, bool LiveCPSR, MachineInstr *CPSRDef) { + unsigned Opc = MI->getOpcode(); + if (Opc == ARM::t2ADDri) { + // If the source register is SP, try to reduce to tADDrSPi, otherwise + // it's a normal reduce. + if (MI->getOperand(1).getReg() != ARM::SP) { + if (ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, CPSRDef)) + return true; + return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef); + } + // Try to reduce to tADDrSPi. + unsigned Imm = MI->getOperand(2).getImm(); + // The immediate must be in range, the destination register must be a low + // reg, the predicate must be "always" and the condition flags must not + // be being set. + if (Imm & 3 || Imm > 1020) + return false; + if (!isARMLowRegister(MI->getOperand(0).getReg())) + return false; + if (MI->getOperand(3).getImm() != ARMCC::AL) + return false; + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.hasOptionalDef() && + MI->getOperand(MCID.getNumOperands()-1).getReg() == ARM::CPSR) + return false; + + MachineInstrBuilder MIB = BuildMI(MBB, *MI, MI->getDebugLoc(), + TII->get(ARM::tADDrSPi)) + .addOperand(MI->getOperand(0)) + .addOperand(MI->getOperand(1)) + .addImm(Imm / 4); // The tADDrSPi has an implied scale by four. + + // Transfer MI flags. + MIB.setMIFlags(MI->getFlags()); + + DEBUG(errs() << "Converted 32-bit: " << *MI << " to 16-bit: " <<*MIB); + + MBB.erase(MI); + ++NumNarrows; + return true; + } + if (Entry.LowRegs1 && !VerifyLowRegs(MI)) return false; - const TargetInstrDesc &TID = MI->getDesc(); - if (TID.mayLoad() || TID.mayStore()) + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.mayLoad() || MCID.mayStore()) return ReduceLoadStore(MBB, MI, Entry); - unsigned Opc = MI->getOpcode(); switch (Opc) { default: break; case ARM::t2ADDSri: @@ -531,13 +567,6 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI, return true; return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef); } - case ARM::t2ADDrSPi: { - static const ReduceEntry NarrowEntry = - { ARM::t2ADDrSPi,ARM::tADDspi, 0, 7, 0, 1, 0, 1, 0, 0,1 }; - if (MI->getOperand(0).getReg() == ARM::SP) - return ReduceToNarrow(MBB, MI, NarrowEntry, LiveCPSR, CPSRDef); - return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef); - } } return false; } @@ -576,23 +605,23 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI, } // Check if it's possible / necessary to transfer the predicate. - const TargetInstrDesc &NewTID = TII->get(Entry.NarrowOpc2); + const MCInstrDesc &NewMCID = TII->get(Entry.NarrowOpc2); unsigned PredReg = 0; ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg); bool SkipPred = false; if (Pred != ARMCC::AL) { - if (!NewTID.isPredicable()) + if (!NewMCID.isPredicable()) // Can't transfer predicate, fail. return false; } else { - SkipPred = !NewTID.isPredicable(); + SkipPred = !NewMCID.isPredicable(); } bool HasCC = false; bool CCDead = false; - const TargetInstrDesc &TID = MI->getDesc(); - if (TID.hasOptionalDef()) { - unsigned NumOps = TID.getNumOperands(); + const MCInstrDesc &MCID = MI->getDesc(); + if (MCID.hasOptionalDef()) { + unsigned NumOps = MCID.getNumOperands(); HasCC = (MI->getOperand(NumOps-1).getReg() == ARM::CPSR); if (HasCC && MI->getOperand(NumOps-1).isDead()) CCDead = true; @@ -602,15 +631,15 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI, // Avoid adding a false dependency on partial flag update by some 16-bit // instructions which has the 's' bit set. - if (Entry.PartFlag && NewTID.hasOptionalDef() && HasCC && + if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC && canAddPseudoFlagDep(CPSRDef, MI)) return false; // Add the 16-bit instruction. DebugLoc dl = MI->getDebugLoc(); - MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewTID); + MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewMCID); MIB.addOperand(MI->getOperand(0)); - if (NewTID.hasOptionalDef()) { + if (NewMCID.hasOptionalDef()) { if (HasCC) AddDefaultT1CC(MIB, CCDead); else @@ -618,11 +647,11 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI, } // Transfer the rest of operands. - unsigned NumOps = TID.getNumOperands(); + unsigned NumOps = MCID.getNumOperands(); for (unsigned i = 1, e = MI->getNumOperands(); i != e; ++i) { - if (i < NumOps && TID.OpInfo[i].isOptionalDef()) + if (i < NumOps && MCID.OpInfo[i].isOptionalDef()) continue; - if (SkipPred && TID.OpInfo[i].isPredicate()) + if (SkipPred && MCID.OpInfo[i].isPredicate()) continue; MIB.addOperand(MI->getOperand(i)); } @@ -645,47 +674,44 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI, return false; unsigned Limit = ~0U; - unsigned Scale = (Entry.WideOpc == ARM::t2ADDrSPi) ? 4 : 1; if (Entry.Imm1Limit) - Limit = ((1 << Entry.Imm1Limit) - 1) * Scale; + Limit = (1 << Entry.Imm1Limit) - 1; - const TargetInstrDesc &TID = MI->getDesc(); - for (unsigned i = 0, e = TID.getNumOperands(); i != e; ++i) { - if (TID.OpInfo[i].isPredicate()) + const MCInstrDesc &MCID = MI->getDesc(); + for (unsigned i = 0, e = MCID.getNumOperands(); i != e; ++i) { + if (MCID.OpInfo[i].isPredicate()) continue; const MachineOperand &MO = MI->getOperand(i); if (MO.isReg()) { unsigned Reg = MO.getReg(); if (!Reg || Reg == ARM::CPSR) continue; - if (Entry.WideOpc == ARM::t2ADDrSPi && Reg == ARM::SP) - continue; if (Entry.LowRegs1 && !isARMLowRegister(Reg)) return false; } else if (MO.isImm() && - !TID.OpInfo[i].isPredicate()) { - if (((unsigned)MO.getImm()) > Limit || (MO.getImm() & (Scale-1)) != 0) + !MCID.OpInfo[i].isPredicate()) { + if (((unsigned)MO.getImm()) > Limit) return false; } } // Check if it's possible / necessary to transfer the predicate. - const TargetInstrDesc &NewTID = TII->get(Entry.NarrowOpc1); + const MCInstrDesc &NewMCID = TII->get(Entry.NarrowOpc1); unsigned PredReg = 0; ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg); bool SkipPred = false; if (Pred != ARMCC::AL) { - if (!NewTID.isPredicable()) + if (!NewMCID.isPredicable()) // Can't transfer predicate, fail. return false; } else { - SkipPred = !NewTID.isPredicable(); + SkipPred = !NewMCID.isPredicable(); } bool HasCC = false; bool CCDead = false; - if (TID.hasOptionalDef()) { - unsigned NumOps = TID.getNumOperands(); + if (MCID.hasOptionalDef()) { + unsigned NumOps = MCID.getNumOperands(); HasCC = (MI->getOperand(NumOps-1).getReg() == ARM::CPSR); if (HasCC && MI->getOperand(NumOps-1).isDead()) CCDead = true; @@ -695,15 +721,15 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI, // Avoid adding a false dependency on partial flag update by some 16-bit // instructions which has the 's' bit set. - if (Entry.PartFlag && NewTID.hasOptionalDef() && HasCC && + if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC && canAddPseudoFlagDep(CPSRDef, MI)) return false; // Add the 16-bit instruction. DebugLoc dl = MI->getDebugLoc(); - MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewTID); + MachineInstrBuilder MIB = BuildMI(MBB, *MI, dl, NewMCID); MIB.addOperand(MI->getOperand(0)); - if (NewTID.hasOptionalDef()) { + if (NewMCID.hasOptionalDef()) { if (HasCC) AddDefaultT1CC(MIB, CCDead); else @@ -711,29 +737,25 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI, } // Transfer the rest of operands. - unsigned NumOps = TID.getNumOperands(); + unsigned NumOps = MCID.getNumOperands(); for (unsigned i = 1, e = MI->getNumOperands(); i != e; ++i) { - if (i < NumOps && TID.OpInfo[i].isOptionalDef()) + if (i < NumOps && MCID.OpInfo[i].isOptionalDef()) continue; - if ((TID.getOpcode() == ARM::t2RSBSri || - TID.getOpcode() == ARM::t2RSBri) && i == 2) + if ((MCID.getOpcode() == ARM::t2RSBSri || + MCID.getOpcode() == ARM::t2RSBri) && i == 2) // Skip the zero immediate operand, it's now implicit. continue; - bool isPred = (i < NumOps && TID.OpInfo[i].isPredicate()); + bool isPred = (i < NumOps && MCID.OpInfo[i].isPredicate()); if (SkipPred && isPred) continue; const MachineOperand &MO = MI->getOperand(i); - if (Scale > 1 && !isPred && MO.isImm()) - MIB.addImm(MO.getImm() / Scale); - else { - if (MO.isReg() && MO.isImplicit() && MO.getReg() == ARM::CPSR) - // Skip implicit def of CPSR. Either it's modeled as an optional - // def now or it's already an implicit def on the new instruction. - continue; - MIB.addOperand(MO); - } + if (MO.isReg() && MO.isImplicit() && MO.getReg() == ARM::CPSR) + // Skip implicit def of CPSR. Either it's modeled as an optional + // def now or it's already an implicit def on the new instruction. + continue; + MIB.addOperand(MO); } - if (!TID.isPredicable() && NewTID.isPredicable()) + if (!MCID.isPredicable() && NewMCID.isPredicable()) AddDefaultPred(MIB); // Transfer MI flags. |
