From d1a87a68064e0b9af3b71b681286954f861bb1b3 Mon Sep 17 00:00:00 2001 From: Sirish Pande Date: Thu, 12 Apr 2012 21:06:38 +0000 Subject: [PATCH] HexagonPacketizer patch. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@154616 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/CodeGen/DFAPacketizer.h | 44 +- lib/CodeGen/DFAPacketizer.cpp | 87 +- lib/Target/Hexagon/CMakeLists.txt | 2 +- lib/Target/Hexagon/Hexagon.h | 1 + lib/Target/Hexagon/HexagonAsmPrinter.cpp | 50 +- lib/Target/Hexagon/HexagonInstrFormats.td | 96 +- lib/Target/Hexagon/HexagonInstrFormatsV4.td | 27 +- lib/Target/Hexagon/HexagonInstrInfo.cpp | 1013 ++++- lib/Target/Hexagon/HexagonInstrInfo.h | 10 + lib/Target/Hexagon/HexagonInstrInfo.td | 4 + lib/Target/Hexagon/HexagonInstrInfoV3.td | 51 +- lib/Target/Hexagon/HexagonInstrInfoV4.td | 2913 +++++++++++-- lib/Target/Hexagon/HexagonMCInst.h | 41 + lib/Target/Hexagon/HexagonSchedule.td | 31 +- lib/Target/Hexagon/HexagonScheduleV4.td | 35 +- lib/Target/Hexagon/HexagonTargetMachine.cpp | 3 + lib/Target/Hexagon/HexagonVLIWPacketizer.cpp | 3654 +++++++++++++++++ .../InstPrinter/HexagonInstPrinter.cpp | 65 +- .../Hexagon/InstPrinter/HexagonInstPrinter.h | 2 + .../Hexagon/MCTargetDesc/HexagonBaseInfo.h | 31 +- 20 files changed, 7610 insertions(+), 550 deletions(-) create mode 100644 lib/Target/Hexagon/HexagonMCInst.h create mode 100644 lib/Target/Hexagon/HexagonVLIWPacketizer.cpp diff --git a/include/llvm/CodeGen/DFAPacketizer.h b/include/llvm/CodeGen/DFAPacketizer.h index ee1ed0779be..2d2db78144a 100644 --- a/include/llvm/CodeGen/DFAPacketizer.h +++ b/include/llvm/CodeGen/DFAPacketizer.h @@ -28,6 +28,7 @@ #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/ADT/DenseMap.h" +#include namespace llvm { @@ -36,7 +37,7 @@ class MachineInstr; class MachineLoopInfo; class MachineDominatorTree; class InstrItineraryData; -class ScheduleDAGInstrs; +class DefaultVLIWScheduler; class SUnit; class DFAPacketizer { @@ -77,6 +78,8 @@ public: // reserveResources - Reserve the resources occupied by a machine // instruction and change the current state to reflect that change. void reserveResources(llvm::MachineInstr *MI); + + const InstrItineraryData *getInstrItins() const { return InstrItins; } }; // VLIWPacketizerList - Implements a simple VLIW packetizer using DFA. The @@ -87,20 +90,21 @@ public: // and machine resource is marked as taken. If any dependency is found, a target // API call is made to prune the dependence. class VLIWPacketizerList { +protected: const TargetMachine &TM; const MachineFunction &MF; const TargetInstrInfo *TII; - // Encapsulate data types not exposed to the target interface. - ScheduleDAGInstrs *SchedulerImpl; + // The VLIW Scheduler. + DefaultVLIWScheduler *VLIWScheduler; -protected: // Vector of instructions assigned to the current packet. std::vector CurrentPacketMIs; // DFA resource tracker. DFAPacketizer *ResourceTracker; - // Scheduling units. - std::vector SUnits; + + // Generate MI -> SU map. + std::map MIToSUnit; public: VLIWPacketizerList( @@ -118,17 +122,32 @@ public: DFAPacketizer *getResourceTracker() {return ResourceTracker;} // addToPacket - Add MI to the current packet. - void addToPacket(MachineInstr *MI); + virtual MachineBasicBlock::iterator addToPacket(MachineInstr *MI) { + MachineBasicBlock::iterator MII = MI; + CurrentPacketMIs.push_back(MI); + ResourceTracker->reserveResources(MI); + return MII; + } // endPacket - End the current packet. - void endPacket(MachineBasicBlock *MBB, MachineInstr *I); + void endPacket(MachineBasicBlock *MBB, MachineInstr *MI); + + // initPacketizerState - perform initialization before packetizing + // an instruction. This function is supposed to be overrided by + // the target dependent packetizer. + virtual void initPacketizerState(void) { return; } // ignorePseudoInstruction - Ignore bundling of pseudo instructions. - bool ignorePseudoInstruction(MachineInstr *I, MachineBasicBlock *MBB); + virtual bool ignorePseudoInstruction(MachineInstr *I, + MachineBasicBlock *MBB) { + return false; + } - // isSoloInstruction - return true if instruction I must end previous - // packet. - bool isSoloInstruction(MachineInstr *I); + // isSoloInstruction - return true if instruction MI can not be packetized + // with any other instruction, which means that MI itself is a packet. + virtual bool isSoloInstruction(MachineInstr *MI) { + return true; + } // isLegalToPacketizeTogether - Is it legal to packetize SUI and SUJ // together. @@ -141,6 +160,7 @@ public: virtual bool isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ) { return false; } + }; } diff --git a/lib/CodeGen/DFAPacketizer.cpp b/lib/CodeGen/DFAPacketizer.cpp index 5ff641c7c84..bfbe7790998 100644 --- a/lib/CodeGen/DFAPacketizer.cpp +++ b/lib/CodeGen/DFAPacketizer.cpp @@ -23,10 +23,10 @@ // //===----------------------------------------------------------------------===// +#include "llvm/CodeGen/ScheduleDAGInstrs.h" #include "llvm/CodeGen/DFAPacketizer.h" #include "llvm/CodeGen/MachineInstr.h" #include "llvm/CodeGen/MachineInstrBundle.h" -#include "llvm/CodeGen/ScheduleDAGInstrs.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/MC/MCInstrItineraries.h" using namespace llvm; @@ -100,17 +100,17 @@ void DFAPacketizer::reserveResources(llvm::MachineInstr *MI) { reserveResources(&MID); } -namespace { +namespace llvm { // DefaultVLIWScheduler - This class extends ScheduleDAGInstrs and overrides // Schedule method to build the dependence graph. class DefaultVLIWScheduler : public ScheduleDAGInstrs { public: DefaultVLIWScheduler(MachineFunction &MF, MachineLoopInfo &MLI, - MachineDominatorTree &MDT, bool IsPostRA); + MachineDominatorTree &MDT, bool IsPostRA); // Schedule - Actual scheduling work. void schedule(); }; -} // end anonymous namespace +} DefaultVLIWScheduler::DefaultVLIWScheduler( MachineFunction &MF, MachineLoopInfo &MLI, MachineDominatorTree &MDT, @@ -129,49 +129,25 @@ VLIWPacketizerList::VLIWPacketizerList( bool IsPostRA) : TM(MF.getTarget()), MF(MF) { TII = TM.getInstrInfo(); ResourceTracker = TII->CreateTargetScheduleState(&TM, 0); - SchedulerImpl = new DefaultVLIWScheduler(MF, MLI, MDT, IsPostRA); + VLIWScheduler = new DefaultVLIWScheduler(MF, MLI, MDT, IsPostRA); } // VLIWPacketizerList Dtor VLIWPacketizerList::~VLIWPacketizerList() { - delete SchedulerImpl; - delete ResourceTracker; -} + if (VLIWScheduler) + delete VLIWScheduler; -// ignorePseudoInstruction - ignore pseudo instructions. -bool VLIWPacketizerList::ignorePseudoInstruction(MachineInstr *MI, - MachineBasicBlock *MBB) { - if (MI->isDebugValue()) - return true; - - if (TII->isSchedulingBoundary(MI, MBB, MF)) - return true; - - return false; -} - -// isSoloInstruction - return true if instruction I must end previous -// packet. -bool VLIWPacketizerList::isSoloInstruction(MachineInstr *I) { - if (I->isInlineAsm()) - return true; - - return false; -} - -// addToPacket - Add I to the current packet and reserve resource. -void VLIWPacketizerList::addToPacket(MachineInstr *MI) { - CurrentPacketMIs.push_back(MI); - ResourceTracker->reserveResources(MI); + if (ResourceTracker) + delete ResourceTracker; } // endPacket - End the current packet, bundle packet instructions and reset // DFA state. void VLIWPacketizerList::endPacket(MachineBasicBlock *MBB, - MachineInstr *I) { + MachineInstr *MI) { if (CurrentPacketMIs.size() > 1) { MachineInstr *MIFirst = CurrentPacketMIs.front(); - finalizeBundle(*MBB, MIFirst, I); + finalizeBundle(*MBB, MIFirst, MI); } CurrentPacketMIs.clear(); ResourceTracker->clearResources(); @@ -181,31 +157,36 @@ void VLIWPacketizerList::endPacket(MachineBasicBlock *MBB, void VLIWPacketizerList::PacketizeMIs(MachineBasicBlock *MBB, MachineBasicBlock::iterator BeginItr, MachineBasicBlock::iterator EndItr) { - assert(MBB->end() == EndItr && "Bad EndIndex"); + assert(VLIWScheduler && "VLIW Scheduler is not initialized!"); + VLIWScheduler->enterRegion(MBB, BeginItr, EndItr, MBB->size()); + VLIWScheduler->schedule(); + VLIWScheduler->exitRegion(); - SchedulerImpl->enterRegion(MBB, BeginItr, EndItr, MBB->size()); - - // Build the DAG without reordering instructions. - SchedulerImpl->schedule(); - - // Remember scheduling units. - SUnits = SchedulerImpl->SUnits; + // Generate MI -> SU map. + //std::map MIToSUnit; + MIToSUnit.clear(); + for (unsigned i = 0, e = VLIWScheduler->SUnits.size(); i != e; ++i) { + SUnit *SU = &VLIWScheduler->SUnits[i]; + MIToSUnit[SU->getInstr()] = SU; + } // The main packetizer loop. for (; BeginItr != EndItr; ++BeginItr) { MachineInstr *MI = BeginItr; - // Ignore pseudo instructions. - if (ignorePseudoInstruction(MI, MBB)) - continue; + this->initPacketizerState(); // End the current packet if needed. - if (isSoloInstruction(MI)) { + if (this->isSoloInstruction(MI)) { endPacket(MBB, MI); continue; } - SUnit *SUI = SchedulerImpl->getSUnit(MI); + // Ignore pseudo instructions. + if (this->ignorePseudoInstruction(MI, MBB)) + continue; + + SUnit *SUI = MIToSUnit[MI]; assert(SUI && "Missing SUnit Info!"); // Ask DFA if machine resource is available for MI. @@ -215,13 +196,13 @@ void VLIWPacketizerList::PacketizeMIs(MachineBasicBlock *MBB, for (std::vector::iterator VI = CurrentPacketMIs.begin(), VE = CurrentPacketMIs.end(); VI != VE; ++VI) { MachineInstr *MJ = *VI; - SUnit *SUJ = SchedulerImpl->getSUnit(MJ); + SUnit *SUJ = MIToSUnit[MJ]; assert(SUJ && "Missing SUnit Info!"); // Is it legal to packetize SUI and SUJ together. - if (!isLegalToPacketizeTogether(SUI, SUJ)) { + if (!this->isLegalToPacketizeTogether(SUI, SUJ)) { // Allow packetization if dependency can be pruned. - if (!isLegalToPruneDependencies(SUI, SUJ)) { + if (!this->isLegalToPruneDependencies(SUI, SUJ)) { // End the packet if dependency cannot be pruned. endPacket(MBB, MI); break; @@ -234,11 +215,9 @@ void VLIWPacketizerList::PacketizeMIs(MachineBasicBlock *MBB, } // Add MI to the current packet. - addToPacket(MI); + BeginItr = this->addToPacket(MI); } // For all instructions in BB. // End any packet left behind. endPacket(MBB, EndItr); - - SchedulerImpl->exitRegion(); } diff --git a/lib/Target/Hexagon/CMakeLists.txt b/lib/Target/Hexagon/CMakeLists.txt index af9e8136bf1..5f6bbf584f3 100644 --- a/lib/Target/Hexagon/CMakeLists.txt +++ b/lib/Target/Hexagon/CMakeLists.txt @@ -27,7 +27,7 @@ add_llvm_target(HexagonCodeGen HexagonSplitTFRCondSets.cpp HexagonSubtarget.cpp HexagonTargetMachine.cpp - HexagonTargetObjectFile.cpp + HexagonVLIWPacketizer.cpp ) add_subdirectory(TargetInfo) diff --git a/lib/Target/Hexagon/Hexagon.h b/lib/Target/Hexagon/Hexagon.h index 08083233367..43858b9624f 100644 --- a/lib/Target/Hexagon/Hexagon.h +++ b/lib/Target/Hexagon/Hexagon.h @@ -40,6 +40,7 @@ namespace llvm { FunctionPass *createHexagonHardwareLoops(); FunctionPass *createHexagonPeephole(); FunctionPass *createHexagonFixupHwLoops(); + FunctionPass *createHexagonPacketizer(); /* TODO: object output. MCCodeEmitter *createHexagonMCCodeEmitter(const Target &, diff --git a/lib/Target/Hexagon/HexagonAsmPrinter.cpp b/lib/Target/Hexagon/HexagonAsmPrinter.cpp index 39bf45d2d77..06e57ecccac 100644 --- a/lib/Target/Hexagon/HexagonAsmPrinter.cpp +++ b/lib/Target/Hexagon/HexagonAsmPrinter.cpp @@ -13,11 +13,11 @@ // //===----------------------------------------------------------------------===// - #define DEBUG_TYPE "asm-printer" #include "Hexagon.h" #include "HexagonAsmPrinter.h" #include "HexagonMachineFunctionInfo.h" +#include "HexagonMCInst.h" #include "HexagonTargetMachine.h" #include "HexagonSubtarget.h" #include "InstPrinter/HexagonInstPrinter.h" @@ -54,6 +54,7 @@ #include "llvm/ADT/SmallString.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringExtras.h" +#include using namespace llvm; @@ -196,10 +197,45 @@ void HexagonAsmPrinter::printPredicateOperand(const MachineInstr *MI, /// the current output stream. /// void HexagonAsmPrinter::EmitInstruction(const MachineInstr *MI) { - MCInst MCI; + if (MI->isBundle()) { + std::vector BundleMIs; - HexagonLowerToMC(MI, MCI, *this); - OutStreamer.EmitInstruction(MCI); + const MachineBasicBlock *MBB = MI->getParent(); + MachineBasicBlock::const_instr_iterator MII = MI; + ++MII; + unsigned int IgnoreCount = 0; + while (MII != MBB->end() && MII->isInsideBundle()) { + const MachineInstr *MInst = MII; + if (MInst->getOpcode() == TargetOpcode::DBG_VALUE || + MInst->getOpcode() == TargetOpcode::IMPLICIT_DEF) { + IgnoreCount++; + ++MII; + continue; + } + //BundleMIs.push_back(&*MII); + BundleMIs.push_back(MInst); + ++MII; + } + unsigned Size = BundleMIs.size(); + assert((Size+IgnoreCount) == MI->getBundleSize() && "Corrupt Bundle!"); + for (unsigned Index = 0; Index < Size; Index++) { + HexagonMCInst MCI; + MCI.setStartPacket(Index == 0); + MCI.setEndPacket(Index == (Size-1)); + + HexagonLowerToMC(BundleMIs[Index], MCI, *this); + OutStreamer.EmitInstruction(MCI); + } + } + else { + HexagonMCInst MCI; + if (MI->getOpcode() == Hexagon::ENDLOOP0) { + MCI.setStartPacket(true); + MCI.setEndPacket(true); + } + HexagonLowerToMC(MI, MCI, *this); + OutStreamer.EmitInstruction(MCI); + } return; } @@ -242,17 +278,17 @@ void HexagonAsmPrinter::printJumpTable(const MachineInstr *MI, int OpNo, raw_ostream &O) { const MachineOperand &MO = MI->getOperand(OpNo); assert( (MO.getType() == MachineOperand::MO_JumpTableIndex) && - "Expecting jump table index"); + "Expecting jump table index"); // Hexagon_TODO: Do we need name mangling? O << *GetJTISymbol(MO.getIndex()); } void HexagonAsmPrinter::printConstantPool(const MachineInstr *MI, int OpNo, - raw_ostream &O) { + raw_ostream &O) { const MachineOperand &MO = MI->getOperand(OpNo); assert( (MO.getType() == MachineOperand::MO_ConstantPoolIndex) && - "Expecting constant pool index"); + "Expecting constant pool index"); // Hexagon_TODO: Do we need name mangling? O << *GetCPISymbol(MO.getIndex()); diff --git a/lib/Target/Hexagon/HexagonInstrFormats.td b/lib/Target/Hexagon/HexagonInstrFormats.td index c9f16fb538a..48f0f01bb4c 100644 --- a/lib/Target/Hexagon/HexagonInstrFormats.td +++ b/lib/Target/Hexagon/HexagonInstrFormats.td @@ -13,13 +13,26 @@ // *** Must match HexagonBaseInfo.h *** //===----------------------------------------------------------------------===// +class Type t> { + bits<5> Value = t; +} +def TypePSEUDO : Type<0>; +def TypeALU32 : Type<1>; +def TypeCR : Type<2>; +def TypeJR : Type<3>; +def TypeJ : Type<4>; +def TypeLD : Type<5>; +def TypeST : Type<6>; +def TypeSYSTEM : Type<7>; +def TypeXTYPE : Type<8>; +def TypeMARKER : Type<31>; //===----------------------------------------------------------------------===// // Intruction Class Declaration + //===----------------------------------------------------------------------===// class InstHexagon pattern, - string cstr, InstrItinClass itin> : Instruction { + string cstr, InstrItinClass itin, Type type> : Instruction { field bits<32> Inst; let Namespace = "Hexagon"; @@ -31,11 +44,15 @@ class InstHexagon pattern, let Constraints = cstr; let Itinerary = itin; - // *** The code below must match HexagonBaseInfo.h *** + // *** Must match HexagonBaseInfo.h *** + Type HexagonType = type; + let TSFlags{4-0} = HexagonType.Value; + bits<1> isHexagonSolo = 0; + let TSFlags{5} = isHexagonSolo; // Predicated instructions. bits<1> isPredicated = 0; - let TSFlags{1} = isPredicated; + let TSFlags{6} = isPredicated; // *** The code above must match HexagonBaseInfo.h *** } @@ -47,28 +64,40 @@ class InstHexagon pattern, // LD Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class LDInst pattern> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<13> imm13; + let mayLoad = 1; } // LD Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class LDInstPost pattern, string cstr> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; bits<13> imm13; + let mayLoad = 1; } // ST Instruction Class in V2/V3 can take SLOT0 only. // ST Instruction Class in V4 can take SLOT0 & SLOT1. // Definition of the instruction class CHANGED from V2/V3 to V4. class STInst pattern> - : InstHexagon { + : InstHexagon { + bits<5> rd; + bits<5> rs; + bits<13> imm13; + let mayStore = 1; +} + +// SYSTEM Instruction Class in V4 can take SLOT0 only +// In V2/V3 we used ST for this but in v4 ST can take SLOT0 or SLOT1. +class SYSInst pattern> + : InstHexagon { bits<5> rd; bits<5> rs; bits<13> imm13; @@ -79,17 +108,18 @@ class STInst pattern> // Definition of the instruction class CHANGED from V2/V3 to V4. class STInstPost pattern, string cstr> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; bits<13> imm13; + let mayStore = 1; } // ALU32 Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class ALU32Type pattern> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -102,7 +132,17 @@ class ALU32Type pattern> // Definition of the instruction class NOT CHANGED. // Name of the Instruction Class changed from ALU64 to XTYPE from V2/V3 to V4. class ALU64Type pattern> - : InstHexagon { + : InstHexagon { + bits<5> rd; + bits<5> rs; + bits<5> rt; + bits<16> imm16; + bits<16> imm16_2; +} + +class ALU64_acc pattern, + string cstr> + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -115,7 +155,7 @@ class ALU64Type pattern> // Definition of the instruction class NOT CHANGED. // Name of the Instruction Class changed from M to XTYPE from V2/V3 to V4. class MInst pattern> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -126,8 +166,8 @@ class MInst pattern> // Definition of the instruction class NOT CHANGED. // Name of the Instruction Class changed from M to XTYPE from V2/V3 to V4. class MInst_acc pattern, - string cstr> - : InstHexagon { + string cstr> + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -138,9 +178,7 @@ class MInst_acc pattern, // Definition of the instruction class NOT CHANGED. // Name of the Instruction Class changed from S to XTYPE from V2/V3 to V4. class SInst pattern> -//: InstHexagon { - : InstHexagon { -// : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -151,8 +189,8 @@ class SInst pattern> // Definition of the instruction class NOT CHANGED. // Name of the Instruction Class changed from S to XTYPE from V2/V3 to V4. class SInst_acc pattern, - string cstr> - : InstHexagon { + string cstr> + : InstHexagon { // : InstHexagon { // : InstHexagon { bits<5> rd; @@ -163,14 +201,14 @@ class SInst_acc pattern, // J Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class JType pattern> - : InstHexagon { + : InstHexagon { bits<16> imm16; } // JR Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class JRType pattern> - : InstHexagon { + : InstHexagon { bits<5> rs; bits<5> pu; // Predicate register } @@ -178,15 +216,22 @@ class JRType pattern> // CR Instruction Class in V2/V3/V4. // Definition of the instruction class NOT CHANGED. class CRInst pattern> - : InstHexagon { + : InstHexagon { bits<5> rs; bits<10> imm10; } +class Marker pattern> + : InstHexagon { + let isCodeGenOnly = 1; + let isPseudo = 1; +} class Pseudo pattern> - : InstHexagon; - + : InstHexagon { + let isCodeGenOnly = 1; + let isPseudo = 1; +} //===----------------------------------------------------------------------===// // Intruction Classes Definitions - @@ -222,6 +267,11 @@ class ALU64_rr pattern> : ALU64Type { } +class ALU64_ri pattern> + : ALU64Type { + let rt{0-4} = 0; +} + // J Type Instructions. class JInst pattern> : JType { @@ -237,12 +287,14 @@ class JRInst pattern> class STInstPI pattern, string cstr> : STInstPost { let rt{0-4} = 0; + let mayStore = 1; } // Post increment LD Instruction. class LDInstPI pattern, string cstr> : LDInstPost { let rt{0-4} = 0; + let mayLoad = 1; } //===----------------------------------------------------------------------===// diff --git a/lib/Target/Hexagon/HexagonInstrFormatsV4.td b/lib/Target/Hexagon/HexagonInstrFormatsV4.td index bd5e4493d7c..49741a3d1b2 100644 --- a/lib/Target/Hexagon/HexagonInstrFormatsV4.td +++ b/lib/Target/Hexagon/HexagonInstrFormatsV4.td @@ -11,11 +11,25 @@ // //===----------------------------------------------------------------------===// +//----------------------------------------------------------------------------// +// Hexagon Intruction Flags + +// +// *** Must match BaseInfo.h *** +//----------------------------------------------------------------------------// + +def TypeMEMOP : Type<9>; +def TypeNV : Type<10>; +def TypePREFIX : Type<30>; + +//----------------------------------------------------------------------------// +// Intruction Classes Definitions + +//----------------------------------------------------------------------------// + // // NV type instructions. // class NVInst_V4 pattern> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<13> imm13; @@ -24,7 +38,7 @@ class NVInst_V4 pattern> // Definition of Post increment new value store. class NVInstPost_V4 pattern, string cstr> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<5> rt; @@ -39,8 +53,15 @@ class NVInstPI_V4 pattern, } class MEMInst_V4 pattern> - : InstHexagon { + : InstHexagon { bits<5> rd; bits<5> rs; bits<6> imm6; } + +class Immext pattern> + : InstHexagon { + let isCodeGenOnly = 1; + + bits<26> imm26; +} diff --git a/lib/Target/Hexagon/HexagonInstrInfo.cpp b/lib/Target/Hexagon/HexagonInstrInfo.cpp index 77b366372cf..6d3267ec877 100644 --- a/lib/Target/Hexagon/HexagonInstrInfo.cpp +++ b/lib/Target/Hexagon/HexagonInstrInfo.cpp @@ -11,10 +11,10 @@ // //===----------------------------------------------------------------------===// -#include "Hexagon.h" #include "HexagonInstrInfo.h" #include "HexagonRegisterInfo.h" #include "HexagonSubtarget.h" +#include "Hexagon.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallVector.h" #include "llvm/CodeGen/DFAPacketizer.h" @@ -466,7 +466,869 @@ unsigned HexagonInstrInfo::createVR(MachineFunction* MF, MVT VT) const { return NewReg; } +bool HexagonInstrInfo::isExtendable(const MachineInstr *MI) const { + switch(MI->getOpcode()) { + // JMP_EQri + case Hexagon::JMP_EQriPt_nv_V4: + case Hexagon::JMP_EQriPnt_nv_V4: + case Hexagon::JMP_EQriNotPt_nv_V4: + case Hexagon::JMP_EQriNotPnt_nv_V4: + // JMP_EQri - with -1 + case Hexagon::JMP_EQriPtneg_nv_V4: + case Hexagon::JMP_EQriPntneg_nv_V4: + case Hexagon::JMP_EQriNotPtneg_nv_V4: + case Hexagon::JMP_EQriNotPntneg_nv_V4: + + // JMP_EQrr + case Hexagon::JMP_EQrrPt_nv_V4: + case Hexagon::JMP_EQrrPnt_nv_V4: + case Hexagon::JMP_EQrrNotPt_nv_V4: + case Hexagon::JMP_EQrrNotPnt_nv_V4: + + // JMP_GTri + case Hexagon::JMP_GTriPt_nv_V4: + case Hexagon::JMP_GTriPnt_nv_V4: + case Hexagon::JMP_GTriNotPt_nv_V4: + case Hexagon::JMP_GTriNotPnt_nv_V4: + + // JMP_GTri - with -1 + case Hexagon::JMP_GTriPtneg_nv_V4: + case Hexagon::JMP_GTriPntneg_nv_V4: + case Hexagon::JMP_GTriNotPtneg_nv_V4: + case Hexagon::JMP_GTriNotPntneg_nv_V4: + + // JMP_GTrr + case Hexagon::JMP_GTrrPt_nv_V4: + case Hexagon::JMP_GTrrPnt_nv_V4: + case Hexagon::JMP_GTrrNotPt_nv_V4: + case Hexagon::JMP_GTrrNotPnt_nv_V4: + + // JMP_GTrrdn + case Hexagon::JMP_GTrrdnPt_nv_V4: + case Hexagon::JMP_GTrrdnPnt_nv_V4: + case Hexagon::JMP_GTrrdnNotPt_nv_V4: + case Hexagon::JMP_GTrrdnNotPnt_nv_V4: + + // JMP_GTUri + case Hexagon::JMP_GTUriPt_nv_V4: + case Hexagon::JMP_GTUriPnt_nv_V4: + case Hexagon::JMP_GTUriNotPt_nv_V4: + case Hexagon::JMP_GTUriNotPnt_nv_V4: + + // JMP_GTUrr + case Hexagon::JMP_GTUrrPt_nv_V4: + case Hexagon::JMP_GTUrrPnt_nv_V4: + case Hexagon::JMP_GTUrrNotPt_nv_V4: + case Hexagon::JMP_GTUrrNotPnt_nv_V4: + + // JMP_GTUrrdn + case Hexagon::JMP_GTUrrdnPt_nv_V4: + case Hexagon::JMP_GTUrrdnPnt_nv_V4: + case Hexagon::JMP_GTUrrdnNotPt_nv_V4: + case Hexagon::JMP_GTUrrdnNotPnt_nv_V4: + return true; + + // TFR_FI + case Hexagon::TFR_FI: + return true; + + + default: + return false; + } + return false; +} + +bool HexagonInstrInfo::isExtended(const MachineInstr *MI) const { + switch(MI->getOpcode()) { + // JMP_EQri + case Hexagon::JMP_EQriPt_ie_nv_V4: + case Hexagon::JMP_EQriPnt_ie_nv_V4: + case Hexagon::JMP_EQriNotPt_ie_nv_V4: + case Hexagon::JMP_EQriNotPnt_ie_nv_V4: + + // JMP_EQri - with -1 + case Hexagon::JMP_EQriPtneg_ie_nv_V4: + case Hexagon::JMP_EQriPntneg_ie_nv_V4: + case Hexagon::JMP_EQriNotPtneg_ie_nv_V4: + case Hexagon::JMP_EQriNotPntneg_ie_nv_V4: + + // JMP_EQrr + case Hexagon::JMP_EQrrPt_ie_nv_V4: + case Hexagon::JMP_EQrrPnt_ie_nv_V4: + case Hexagon::JMP_EQrrNotPt_ie_nv_V4: + case Hexagon::JMP_EQrrNotPnt_ie_nv_V4: + + // JMP_GTri + case Hexagon::JMP_GTriPt_ie_nv_V4: + case Hexagon::JMP_GTriPnt_ie_nv_V4: + case Hexagon::JMP_GTriNotPt_ie_nv_V4: + case Hexagon::JMP_GTriNotPnt_ie_nv_V4: + + // JMP_GTri - with -1 + case Hexagon::JMP_GTriPtneg_ie_nv_V4: + case Hexagon::JMP_GTriPntneg_ie_nv_V4: + case Hexagon::JMP_GTriNotPtneg_ie_nv_V4: + case Hexagon::JMP_GTriNotPntneg_ie_nv_V4: + + // JMP_GTrr + case Hexagon::JMP_GTrrPt_ie_nv_V4: + case Hexagon::JMP_GTrrPnt_ie_nv_V4: + case Hexagon::JMP_GTrrNotPt_ie_nv_V4: + case Hexagon::JMP_GTrrNotPnt_ie_nv_V4: + + // JMP_GTrrdn + case Hexagon::JMP_GTrrdnPt_ie_nv_V4: + case Hexagon::JMP_GTrrdnPnt_ie_nv_V4: + case Hexagon::JMP_GTrrdnNotPt_ie_nv_V4: + case Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4: + + // JMP_GTUri + case Hexagon::JMP_GTUriPt_ie_nv_V4: + case Hexagon::JMP_GTUriPnt_ie_nv_V4: + case Hexagon::JMP_GTUriNotPt_ie_nv_V4: + case Hexagon::JMP_GTUriNotPnt_ie_nv_V4: + + // JMP_GTUrr + case Hexagon::JMP_GTUrrPt_ie_nv_V4: + case Hexagon::JMP_GTUrrPnt_ie_nv_V4: + case Hexagon::JMP_GTUrrNotPt_ie_nv_V4: + case Hexagon::JMP_GTUrrNotPnt_ie_nv_V4: + + // JMP_GTUrrdn + case Hexagon::JMP_GTUrrdnPt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnPnt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4: + + // V4 absolute set addressing. + case Hexagon::LDrid_abs_setimm_V4: + case Hexagon::LDriw_abs_setimm_V4: + case Hexagon::LDrih_abs_setimm_V4: + case Hexagon::LDrib_abs_setimm_V4: + case Hexagon::LDriuh_abs_setimm_V4: + case Hexagon::LDriub_abs_setimm_V4: + + case Hexagon::STrid_abs_setimm_V4: + case Hexagon::STrib_abs_setimm_V4: + case Hexagon::STrih_abs_setimm_V4: + case Hexagon::STriw_abs_setimm_V4: + + // V4 global address load. + case Hexagon::LDrid_GP_cPt_V4 : + case Hexagon::LDrid_GP_cNotPt_V4 : + case Hexagon::LDrid_GP_cdnPt_V4 : + case Hexagon::LDrid_GP_cdnNotPt_V4 : + case Hexagon::LDrib_GP_cPt_V4 : + case Hexagon::LDrib_GP_cNotPt_V4 : + case Hexagon::LDrib_GP_cdnPt_V4 : + case Hexagon::LDrib_GP_cdnNotPt_V4 : + case Hexagon::LDriub_GP_cPt_V4 : + case Hexagon::LDriub_GP_cNotPt_V4 : + case Hexagon::LDriub_GP_cdnPt_V4 : + case Hexagon::LDriub_GP_cdnNotPt_V4 : + case Hexagon::LDrih_GP_cPt_V4 : + case Hexagon::LDrih_GP_cNotPt_V4 : + case Hexagon::LDrih_GP_cdnPt_V4 : + case Hexagon::LDrih_GP_cdnNotPt_V4 : + case Hexagon::LDriuh_GP_cPt_V4 : + case Hexagon::LDriuh_GP_cNotPt_V4 : + case Hexagon::LDriuh_GP_cdnPt_V4 : + case Hexagon::LDriuh_GP_cdnNotPt_V4 : + case Hexagon::LDriw_GP_cPt_V4 : + case Hexagon::LDriw_GP_cNotPt_V4 : + case Hexagon::LDriw_GP_cdnPt_V4 : + case Hexagon::LDriw_GP_cdnNotPt_V4 : + case Hexagon::LDd_GP_cPt_V4 : + case Hexagon::LDd_GP_cNotPt_V4 : + case Hexagon::LDd_GP_cdnPt_V4 : + case Hexagon::LDd_GP_cdnNotPt_V4 : + case Hexagon::LDb_GP_cPt_V4 : + case Hexagon::LDb_GP_cNotPt_V4 : + case Hexagon::LDb_GP_cdnPt_V4 : + case Hexagon::LDb_GP_cdnNotPt_V4 : + case Hexagon::LDub_GP_cPt_V4 : + case Hexagon::LDub_GP_cNotPt_V4 : + case Hexagon::LDub_GP_cdnPt_V4 : + case Hexagon::LDub_GP_cdnNotPt_V4 : + case Hexagon::LDh_GP_cPt_V4 : + case Hexagon::LDh_GP_cNotPt_V4 : + case Hexagon::LDh_GP_cdnPt_V4 : + case Hexagon::LDh_GP_cdnNotPt_V4 : + case Hexagon::LDuh_GP_cPt_V4 : + case Hexagon::LDuh_GP_cNotPt_V4 : + case Hexagon::LDuh_GP_cdnPt_V4 : + case Hexagon::LDuh_GP_cdnNotPt_V4 : + case Hexagon::LDw_GP_cPt_V4 : + case Hexagon::LDw_GP_cNotPt_V4 : + case Hexagon::LDw_GP_cdnPt_V4 : + case Hexagon::LDw_GP_cdnNotPt_V4 : + + // V4 global address store. + case Hexagon::STrid_GP_cPt_V4 : + case Hexagon::STrid_GP_cNotPt_V4 : + case Hexagon::STrid_GP_cdnPt_V4 : + case Hexagon::STrid_GP_cdnNotPt_V4 : + case Hexagon::STrib_GP_cPt_V4 : + case Hexagon::STrib_GP_cNotPt_V4 : + case Hexagon::STrib_GP_cdnPt_V4 : + case Hexagon::STrib_GP_cdnNotPt_V4 : + case Hexagon::STrih_GP_cPt_V4 : + case Hexagon::STrih_GP_cNotPt_V4 : + case Hexagon::STrih_GP_cdnPt_V4 : + case Hexagon::STrih_GP_cdnNotPt_V4 : + case Hexagon::STriw_GP_cPt_V4 : + case Hexagon::STriw_GP_cNotPt_V4 : + case Hexagon::STriw_GP_cdnPt_V4 : + case Hexagon::STriw_GP_cdnNotPt_V4 : + case Hexagon::STd_GP_cPt_V4 : + case Hexagon::STd_GP_cNotPt_V4 : + case Hexagon::STd_GP_cdnPt_V4 : + case Hexagon::STd_GP_cdnNotPt_V4 : + case Hexagon::STb_GP_cPt_V4 : + case Hexagon::STb_GP_cNotPt_V4 : + case Hexagon::STb_GP_cdnPt_V4 : + case Hexagon::STb_GP_cdnNotPt_V4 : + case Hexagon::STh_GP_cPt_V4 : + case Hexagon::STh_GP_cNotPt_V4 : + case Hexagon::STh_GP_cdnPt_V4 : + case Hexagon::STh_GP_cdnNotPt_V4 : + case Hexagon::STw_GP_cPt_V4 : + case Hexagon::STw_GP_cNotPt_V4 : + case Hexagon::STw_GP_cdnPt_V4 : + case Hexagon::STw_GP_cdnNotPt_V4 : + + // V4 predicated global address new value store. + case Hexagon::STrib_GP_cPt_nv_V4 : + case Hexagon::STrib_GP_cNotPt_nv_V4 : + case Hexagon::STrib_GP_cdnPt_nv_V4 : + case Hexagon::STrib_GP_cdnNotPt_nv_V4 : + case Hexagon::STrih_GP_cPt_nv_V4 : + case Hexagon::STrih_GP_cNotPt_nv_V4 : + case Hexagon::STrih_GP_cdnPt_nv_V4 : + case Hexagon::STrih_GP_cdnNotPt_nv_V4 : + case Hexagon::STriw_GP_cPt_nv_V4 : + case Hexagon::STriw_GP_cNotPt_nv_V4 : + case Hexagon::STriw_GP_cdnPt_nv_V4 : + case Hexagon::STriw_GP_cdnNotPt_nv_V4 : + case Hexagon::STb_GP_cPt_nv_V4 : + case Hexagon::STb_GP_cNotPt_nv_V4 : + case Hexagon::STb_GP_cdnPt_nv_V4 : + case Hexagon::STb_GP_cdnNotPt_nv_V4 : + case Hexagon::STh_GP_cPt_nv_V4 : + case Hexagon::STh_GP_cNotPt_nv_V4 : + case Hexagon::STh_GP_cdnPt_nv_V4 : + case Hexagon::STh_GP_cdnNotPt_nv_V4 : + case Hexagon::STw_GP_cPt_nv_V4 : + case Hexagon::STw_GP_cNotPt_nv_V4 : + case Hexagon::STw_GP_cdnPt_nv_V4 : + case Hexagon::STw_GP_cdnNotPt_nv_V4 : + + // TFR_FI + case Hexagon::TFR_FI_immext_V4: + return true; + + default: + return false; + } + return false; +} + +bool HexagonInstrInfo::isNewValueJump(const MachineInstr *MI) const { + switch (MI->getOpcode()) { + // JMP_EQri + case Hexagon::JMP_EQriPt_nv_V4: + case Hexagon::JMP_EQriPnt_nv_V4: + case Hexagon::JMP_EQriNotPt_nv_V4: + case Hexagon::JMP_EQriNotPnt_nv_V4: + case Hexagon::JMP_EQriPt_ie_nv_V4: + case Hexagon::JMP_EQriPnt_ie_nv_V4: + case Hexagon::JMP_EQriNotPt_ie_nv_V4: + case Hexagon::JMP_EQriNotPnt_ie_nv_V4: + + // JMP_EQri - with -1 + case Hexagon::JMP_EQriPtneg_nv_V4: + case Hexagon::JMP_EQriPntneg_nv_V4: + case Hexagon::JMP_EQriNotPtneg_nv_V4: + case Hexagon::JMP_EQriNotPntneg_nv_V4: + case Hexagon::JMP_EQriPtneg_ie_nv_V4: + case Hexagon::JMP_EQriPntneg_ie_nv_V4: + case Hexagon::JMP_EQriNotPtneg_ie_nv_V4: + case Hexagon::JMP_EQriNotPntneg_ie_nv_V4: + + // JMP_EQrr + case Hexagon::JMP_EQrrPt_nv_V4: + case Hexagon::JMP_EQrrPnt_nv_V4: + case Hexagon::JMP_EQrrNotPt_nv_V4: + case Hexagon::JMP_EQrrNotPnt_nv_V4: + case Hexagon::JMP_EQrrPt_ie_nv_V4: + case Hexagon::JMP_EQrrPnt_ie_nv_V4: + case Hexagon::JMP_EQrrNotPt_ie_nv_V4: + case Hexagon::JMP_EQrrNotPnt_ie_nv_V4: + + // JMP_GTri + case Hexagon::JMP_GTriPt_nv_V4: + case Hexagon::JMP_GTriPnt_nv_V4: + case Hexagon::JMP_GTriNotPt_nv_V4: + case Hexagon::JMP_GTriNotPnt_nv_V4: + case Hexagon::JMP_GTriPt_ie_nv_V4: + case Hexagon::JMP_GTriPnt_ie_nv_V4: + case Hexagon::JMP_GTriNotPt_ie_nv_V4: + case Hexagon::JMP_GTriNotPnt_ie_nv_V4: + + // JMP_GTri - with -1 + case Hexagon::JMP_GTriPtneg_nv_V4: + case Hexagon::JMP_GTriPntneg_nv_V4: + case Hexagon::JMP_GTriNotPtneg_nv_V4: + case Hexagon::JMP_GTriNotPntneg_nv_V4: + case Hexagon::JMP_GTriPtneg_ie_nv_V4: + case Hexagon::JMP_GTriPntneg_ie_nv_V4: + case Hexagon::JMP_GTriNotPtneg_ie_nv_V4: + case Hexagon::JMP_GTriNotPntneg_ie_nv_V4: + + // JMP_GTrr + case Hexagon::JMP_GTrrPt_nv_V4: + case Hexagon::JMP_GTrrPnt_nv_V4: + case Hexagon::JMP_GTrrNotPt_nv_V4: + case Hexagon::JMP_GTrrNotPnt_nv_V4: + case Hexagon::JMP_GTrrPt_ie_nv_V4: + case Hexagon::JMP_GTrrPnt_ie_nv_V4: + case Hexagon::JMP_GTrrNotPt_ie_nv_V4: + case Hexagon::JMP_GTrrNotPnt_ie_nv_V4: + + // JMP_GTrrdn + case Hexagon::JMP_GTrrdnPt_nv_V4: + case Hexagon::JMP_GTrrdnPnt_nv_V4: + case Hexagon::JMP_GTrrdnNotPt_nv_V4: + case Hexagon::JMP_GTrrdnNotPnt_nv_V4: + case Hexagon::JMP_GTrrdnPt_ie_nv_V4: + case Hexagon::JMP_GTrrdnPnt_ie_nv_V4: + case Hexagon::JMP_GTrrdnNotPt_ie_nv_V4: + case Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4: + + // JMP_GTUri + case Hexagon::JMP_GTUriPt_nv_V4: + case Hexagon::JMP_GTUriPnt_nv_V4: + case Hexagon::JMP_GTUriNotPt_nv_V4: + case Hexagon::JMP_GTUriNotPnt_nv_V4: + case Hexagon::JMP_GTUriPt_ie_nv_V4: + case Hexagon::JMP_GTUriPnt_ie_nv_V4: + case Hexagon::JMP_GTUriNotPt_ie_nv_V4: + case Hexagon::JMP_GTUriNotPnt_ie_nv_V4: + + // JMP_GTUrr + case Hexagon::JMP_GTUrrPt_nv_V4: + case Hexagon::JMP_GTUrrPnt_nv_V4: + case Hexagon::JMP_GTUrrNotPt_nv_V4: + case Hexagon::JMP_GTUrrNotPnt_nv_V4: + case Hexagon::JMP_GTUrrPt_ie_nv_V4: + case Hexagon::JMP_GTUrrPnt_ie_nv_V4: + case Hexagon::JMP_GTUrrNotPt_ie_nv_V4: + case Hexagon::JMP_GTUrrNotPnt_ie_nv_V4: + + // JMP_GTUrrdn + case Hexagon::JMP_GTUrrdnPt_nv_V4: + case Hexagon::JMP_GTUrrdnPnt_nv_V4: + case Hexagon::JMP_GTUrrdnNotPt_nv_V4: + case Hexagon::JMP_GTUrrdnNotPnt_nv_V4: + case Hexagon::JMP_GTUrrdnPt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnPnt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4: + case Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4: + return true; + + default: + return false; + } + return false; +} + +unsigned HexagonInstrInfo::getImmExtForm(const MachineInstr* MI) const { + switch(MI->getOpcode()) { + // JMP_EQri + case Hexagon::JMP_EQriPt_nv_V4: + return Hexagon::JMP_EQriPt_ie_nv_V4; + case Hexagon::JMP_EQriNotPt_nv_V4: + return Hexagon::JMP_EQriNotPt_ie_nv_V4; + case Hexagon::JMP_EQriPnt_nv_V4: + return Hexagon::JMP_EQriPnt_ie_nv_V4; + case Hexagon::JMP_EQriNotPnt_nv_V4: + return Hexagon::JMP_EQriNotPnt_ie_nv_V4; + + // JMP_EQri -- with -1 + case Hexagon::JMP_EQriPtneg_nv_V4: + return Hexagon::JMP_EQriPtneg_ie_nv_V4; + case Hexagon::JMP_EQriNotPtneg_nv_V4: + return Hexagon::JMP_EQriNotPtneg_ie_nv_V4; + case Hexagon::JMP_EQriPntneg_nv_V4: + return Hexagon::JMP_EQriPntneg_ie_nv_V4; + case Hexagon::JMP_EQriNotPntneg_nv_V4: + return Hexagon::JMP_EQriNotPntneg_ie_nv_V4; + + // JMP_EQrr + case Hexagon::JMP_EQrrPt_nv_V4: + return Hexagon::JMP_EQrrPt_ie_nv_V4; + case Hexagon::JMP_EQrrNotPt_nv_V4: + return Hexagon::JMP_EQrrNotPt_ie_nv_V4; + case Hexagon::JMP_EQrrPnt_nv_V4: + return Hexagon::JMP_EQrrPnt_ie_nv_V4; + case Hexagon::JMP_EQrrNotPnt_nv_V4: + return Hexagon::JMP_EQrrNotPnt_ie_nv_V4; + + // JMP_GTri + case Hexagon::JMP_GTriPt_nv_V4: + return Hexagon::JMP_GTriPt_ie_nv_V4; + case Hexagon::JMP_GTriNotPt_nv_V4: + return Hexagon::JMP_GTriNotPt_ie_nv_V4; + case Hexagon::JMP_GTriPnt_nv_V4: + return Hexagon::JMP_GTriPnt_ie_nv_V4; + case Hexagon::JMP_GTriNotPnt_nv_V4: + return Hexagon::JMP_GTriNotPnt_ie_nv_V4; + + // JMP_GTri -- with -1 + case Hexagon::JMP_GTriPtneg_nv_V4: + return Hexagon::JMP_GTriPtneg_ie_nv_V4; + case Hexagon::JMP_GTriNotPtneg_nv_V4: + return Hexagon::JMP_GTriNotPtneg_ie_nv_V4; + case Hexagon::JMP_GTriPntneg_nv_V4: + return Hexagon::JMP_GTriPntneg_ie_nv_V4; + case Hexagon::JMP_GTriNotPntneg_nv_V4: + return Hexagon::JMP_GTriNotPntneg_ie_nv_V4; + + // JMP_GTrr + case Hexagon::JMP_GTrrPt_nv_V4: + return Hexagon::JMP_GTrrPt_ie_nv_V4; + case Hexagon::JMP_GTrrNotPt_nv_V4: + return Hexagon::JMP_GTrrNotPt_ie_nv_V4; + case Hexagon::JMP_GTrrPnt_nv_V4: + return Hexagon::JMP_GTrrPnt_ie_nv_V4; + case Hexagon::JMP_GTrrNotPnt_nv_V4: + return Hexagon::JMP_GTrrNotPnt_ie_nv_V4; + + // JMP_GTrrdn + case Hexagon::JMP_GTrrdnPt_nv_V4: + return Hexagon::JMP_GTrrdnPt_ie_nv_V4; + case Hexagon::JMP_GTrrdnNotPt_nv_V4: + return Hexagon::JMP_GTrrdnNotPt_ie_nv_V4; + case Hexagon::JMP_GTrrdnPnt_nv_V4: + return Hexagon::JMP_GTrrdnPnt_ie_nv_V4; + case Hexagon::JMP_GTrrdnNotPnt_nv_V4: + return Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4; + + // JMP_GTUri + case Hexagon::JMP_GTUriPt_nv_V4: + return Hexagon::JMP_GTUriPt_ie_nv_V4; + case Hexagon::JMP_GTUriNotPt_nv_V4: + return Hexagon::JMP_GTUriNotPt_ie_nv_V4; + case Hexagon::JMP_GTUriPnt_nv_V4: + return Hexagon::JMP_GTUriPnt_ie_nv_V4; + case Hexagon::JMP_GTUriNotPnt_nv_V4: + return Hexagon::JMP_GTUriNotPnt_ie_nv_V4; + + // JMP_GTUrr + case Hexagon::JMP_GTUrrPt_nv_V4: + return Hexagon::JMP_GTUrrPt_ie_nv_V4; + case Hexagon::JMP_GTUrrNotPt_nv_V4: + return Hexagon::JMP_GTUrrNotPt_ie_nv_V4; + case Hexagon::JMP_GTUrrPnt_nv_V4: + return Hexagon::JMP_GTUrrPnt_ie_nv_V4; + case Hexagon::JMP_GTUrrNotPnt_nv_V4: + return Hexagon::JMP_GTUrrNotPnt_ie_nv_V4; + + // JMP_GTUrrdn + case Hexagon::JMP_GTUrrdnPt_nv_V4: + return Hexagon::JMP_GTUrrdnPt_ie_nv_V4; + case Hexagon::JMP_GTUrrdnNotPt_nv_V4: + return Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4; + case Hexagon::JMP_GTUrrdnPnt_nv_V4: + return Hexagon::JMP_GTUrrdnPnt_ie_nv_V4; + case Hexagon::JMP_GTUrrdnNotPnt_nv_V4: + return Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4; + + case Hexagon::TFR_FI: + return Hexagon::TFR_FI_immext_V4; + + case Hexagon::MEMw_ADDSUBi_indexed_MEM_V4 : + case Hexagon::MEMw_ADDi_indexed_MEM_V4 : + case Hexagon::MEMw_SUBi_indexed_MEM_V4 : + case Hexagon::MEMw_ADDr_indexed_MEM_V4 : + case Hexagon::MEMw_SUBr_indexed_MEM_V4 : + case Hexagon::MEMw_ANDr_indexed_MEM_V4 : + case Hexagon::MEMw_ORr_indexed_MEM_V4 : + case Hexagon::MEMw_ADDSUBi_MEM_V4 : + case Hexagon::MEMw_ADDi_MEM_V4 : + case Hexagon::MEMw_SUBi_MEM_V4 : + case Hexagon::MEMw_ADDr_MEM_V4 : + case Hexagon::MEMw_SUBr_MEM_V4 : + case Hexagon::MEMw_ANDr_MEM_V4 : + case Hexagon::MEMw_ORr_MEM_V4 : + case Hexagon::MEMh_ADDSUBi_indexed_MEM_V4 : + case Hexagon::MEMh_ADDi_indexed_MEM_V4 : + case Hexagon::MEMh_SUBi_indexed_MEM_V4 : + case Hexagon::MEMh_ADDr_indexed_MEM_V4 : + case Hexagon::MEMh_SUBr_indexed_MEM_V4 : + case Hexagon::MEMh_ANDr_indexed_MEM_V4 : + case Hexagon::MEMh_ORr_indexed_MEM_V4 : + case Hexagon::MEMh_ADDSUBi_MEM_V4 : + case Hexagon::MEMh_ADDi_MEM_V4 : + case Hexagon::MEMh_SUBi_MEM_V4 : + case Hexagon::MEMh_ADDr_MEM_V4 : + case Hexagon::MEMh_SUBr_MEM_V4 : + case Hexagon::MEMh_ANDr_MEM_V4 : + case Hexagon::MEMh_ORr_MEM_V4 : + case Hexagon::MEMb_ADDSUBi_indexed_MEM_V4 : + case Hexagon::MEMb_ADDi_indexed_MEM_V4 : + case Hexagon::MEMb_SUBi_indexed_MEM_V4 : + case Hexagon::MEMb_ADDr_indexed_MEM_V4 : + case Hexagon::MEMb_SUBr_indexed_MEM_V4 : + case Hexagon::MEMb_ANDr_indexed_MEM_V4 : + case Hexagon::MEMb_ORr_indexed_MEM_V4 : + case Hexagon::MEMb_ADDSUBi_MEM_V4 : + case Hexagon::MEMb_ADDi_MEM_V4 : + case Hexagon::MEMb_SUBi_MEM_V4 : + case Hexagon::MEMb_ADDr_MEM_V4 : + case Hexagon::MEMb_SUBr_MEM_V4 : + case Hexagon::MEMb_ANDr_MEM_V4 : + case Hexagon::MEMb_ORr_MEM_V4 : + assert(0 && "Needs implementing"); + + default: + assert(0 && "Unknown type of instruction"); + } + assert(0 && "Unknown type of instruction"); +} + +unsigned HexagonInstrInfo::getNormalBranchForm(const MachineInstr* MI) const { + switch(MI->getOpcode()) { + // JMP_EQri + case Hexagon::JMP_EQriPt_ie_nv_V4: + return Hexagon::JMP_EQriPt_nv_V4; + case Hexagon::JMP_EQriNotPt_ie_nv_V4: + return Hexagon::JMP_EQriNotPt_nv_V4; + case Hexagon::JMP_EQriPnt_ie_nv_V4: + return Hexagon::JMP_EQriPnt_nv_V4; + case Hexagon::JMP_EQriNotPnt_ie_nv_V4: + return Hexagon::JMP_EQriNotPnt_nv_V4; + + // JMP_EQri -- with -1 + case Hexagon::JMP_EQriPtneg_ie_nv_V4: + return Hexagon::JMP_EQriPtneg_nv_V4; + case Hexagon::JMP_EQriNotPtneg_ie_nv_V4: + return Hexagon::JMP_EQriNotPtneg_nv_V4; + case Hexagon::JMP_EQriPntneg_ie_nv_V4: + return Hexagon::JMP_EQriPntneg_nv_V4; + case Hexagon::JMP_EQriNotPntneg_ie_nv_V4: + return Hexagon::JMP_EQriNotPntneg_nv_V4; + + // JMP_EQrr + case Hexagon::JMP_EQrrPt_ie_nv_V4: + return Hexagon::JMP_EQrrPt_nv_V4; + case Hexagon::JMP_EQrrNotPt_ie_nv_V4: + return Hexagon::JMP_EQrrNotPt_nv_V4; + case Hexagon::JMP_EQrrPnt_ie_nv_V4: + return Hexagon::JMP_EQrrPnt_nv_V4; + case Hexagon::JMP_EQrrNotPnt_ie_nv_V4: + return Hexagon::JMP_EQrrNotPnt_nv_V4; + + // JMP_GTri + case Hexagon::JMP_GTriPt_ie_nv_V4: + return Hexagon::JMP_GTriPt_nv_V4; + case Hexagon::JMP_GTriNotPt_ie_nv_V4: + return Hexagon::JMP_GTriNotPt_nv_V4; + case Hexagon::JMP_GTriPnt_ie_nv_V4: + return Hexagon::JMP_GTriPnt_nv_V4; + case Hexagon::JMP_GTriNotPnt_ie_nv_V4: + return Hexagon::JMP_GTriNotPnt_nv_V4; + + // JMP_GTri -- with -1 + case Hexagon::JMP_GTriPtneg_ie_nv_V4: + return Hexagon::JMP_GTriPtneg_nv_V4; + case Hexagon::JMP_GTriNotPtneg_ie_nv_V4: + return Hexagon::JMP_GTriNotPtneg_nv_V4; + case Hexagon::JMP_GTriPntneg_ie_nv_V4: + return Hexagon::JMP_GTriPntneg_nv_V4; + case Hexagon::JMP_GTriNotPntneg_ie_nv_V4: + return Hexagon::JMP_GTriNotPntneg_nv_V4; + + // JMP_GTrr + case Hexagon::JMP_GTrrPt_ie_nv_V4: + return Hexagon::JMP_GTrrPt_nv_V4; + case Hexagon::JMP_GTrrNotPt_ie_nv_V4: + return Hexagon::JMP_GTrrNotPt_nv_V4; + case Hexagon::JMP_GTrrPnt_ie_nv_V4: + return Hexagon::JMP_GTrrPnt_nv_V4; + case Hexagon::JMP_GTrrNotPnt_ie_nv_V4: + return Hexagon::JMP_GTrrNotPnt_nv_V4; + + // JMP_GTrrdn + case Hexagon::JMP_GTrrdnPt_ie_nv_V4: + return Hexagon::JMP_GTrrdnPt_nv_V4; + case Hexagon::JMP_GTrrdnNotPt_ie_nv_V4: + return Hexagon::JMP_GTrrdnNotPt_nv_V4; + case Hexagon::JMP_GTrrdnPnt_ie_nv_V4: + return Hexagon::JMP_GTrrdnPnt_nv_V4; + case Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4: + return Hexagon::JMP_GTrrdnNotPnt_nv_V4; + + // JMP_GTUri + case Hexagon::JMP_GTUriPt_ie_nv_V4: + return Hexagon::JMP_GTUriPt_nv_V4; + case Hexagon::JMP_GTUriNotPt_ie_nv_V4: + return Hexagon::JMP_GTUriNotPt_nv_V4; + case Hexagon::JMP_GTUriPnt_ie_nv_V4: + return Hexagon::JMP_GTUriPnt_nv_V4; + case Hexagon::JMP_GTUriNotPnt_ie_nv_V4: + return Hexagon::JMP_GTUriNotPnt_nv_V4; + + // JMP_GTUrr + case Hexagon::JMP_GTUrrPt_ie_nv_V4: + return Hexagon::JMP_GTUrrPt_nv_V4; + case Hexagon::JMP_GTUrrNotPt_ie_nv_V4: + return Hexagon::JMP_GTUrrNotPt_nv_V4; + case Hexagon::JMP_GTUrrPnt_ie_nv_V4: + return Hexagon::JMP_GTUrrPnt_nv_V4; + case Hexagon::JMP_GTUrrNotPnt_ie_nv_V4: + return Hexagon::JMP_GTUrrNotPnt_nv_V4; + + // JMP_GTUrrdn + case Hexagon::JMP_GTUrrdnPt_ie_nv_V4: + return Hexagon::JMP_GTUrrdnPt_nv_V4; + case Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4: + return Hexagon::JMP_GTUrrdnNotPt_nv_V4; + case Hexagon::JMP_GTUrrdnPnt_ie_nv_V4: + return Hexagon::JMP_GTUrrdnPnt_nv_V4; + case Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4: + return Hexagon::JMP_GTUrrdnNotPnt_nv_V4; + + default: + assert(0 && "Unknown type of jump instruction"); + } + assert(0 && "Unknown type of jump instruction"); +} + + +bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const { + switch (MI->getOpcode()) { + + // Store Byte + case Hexagon::STrib_nv_V4: + case Hexagon::STrib_indexed_nv_V4: + case Hexagon::STrib_indexed_shl_nv_V4: + case Hexagon::STrib_shl_nv_V4: + case Hexagon::STrib_GP_nv_V4: + case Hexagon::STb_GP_nv_V4: + case Hexagon::POST_STbri_nv_V4: + case Hexagon::STrib_cPt_nv_V4: + case Hexagon::STrib_cdnPt_nv_V4: + case Hexagon::STrib_cNotPt_nv_V4: + case Hexagon::STrib_cdnNotPt_nv_V4: + case Hexagon::STrib_indexed_cPt_nv_V4: + case Hexagon::STrib_indexed_cdnPt_nv_V4: + case Hexagon::STrib_indexed_cNotPt_nv_V4: + case Hexagon::STrib_indexed_cdnNotPt_nv_V4: + case Hexagon::STrib_indexed_shl_cPt_nv_V4: + case Hexagon::STrib_indexed_shl_cdnPt_nv_V4: + case Hexagon::STrib_indexed_shl_cNotPt_nv_V4: + case Hexagon::STrib_indexed_shl_cdnNotPt_nv_V4: + case Hexagon::POST_STbri_cPt_nv_V4: + case Hexagon::POST_STbri_cdnPt_nv_V4: + case Hexagon::POST_STbri_cNotPt_nv_V4: + case Hexagon::POST_STbri_cdnNotPt_nv_V4: + case Hexagon::STb_GP_cPt_nv_V4: + case Hexagon::STb_GP_cNotPt_nv_V4: + case Hexagon::STb_GP_cdnPt_nv_V4: + case Hexagon::STb_GP_cdnNotPt_nv_V4: + case Hexagon::STrib_GP_cPt_nv_V4: + case Hexagon::STrib_GP_cNotPt_nv_V4: + case Hexagon::STrib_GP_cdnPt_nv_V4: + case Hexagon::STrib_GP_cdnNotPt_nv_V4: + case Hexagon::STrib_abs_nv_V4: + case Hexagon::STrib_abs_cPt_nv_V4: + case Hexagon::STrib_abs_cdnPt_nv_V4: + case Hexagon::STrib_abs_cNotPt_nv_V4: + case Hexagon::STrib_abs_cdnNotPt_nv_V4: + case Hexagon::STrib_imm_abs_nv_V4: + case Hexagon::STrib_imm_abs_cPt_nv_V4: + case Hexagon::STrib_imm_abs_cdnPt_nv_V4: + case Hexagon::STrib_imm_abs_cNotPt_nv_V4: + case Hexagon::STrib_imm_abs_cdnNotPt_nv_V4: + + // Store Halfword + case Hexagon::STrih_nv_V4: + case Hexagon::STrih_indexed_nv_V4: + case Hexagon::STrih_indexed_shl_nv_V4: + case Hexagon::STrih_shl_nv_V4: + case Hexagon::STrih_GP_nv_V4: + case Hexagon::STh_GP_nv_V4: + case Hexagon::POST_SThri_nv_V4: + case Hexagon::STrih_cPt_nv_V4: + case Hexagon::STrih_cdnPt_nv_V4: + case Hexagon::STrih_cNotPt_nv_V4: + case Hexagon::STrih_cdnNotPt_nv_V4: + case Hexagon::STrih_indexed_cPt_nv_V4: + case Hexagon::STrih_indexed_cdnPt_nv_V4: + case Hexagon::STrih_indexed_cNotPt_nv_V4: + case Hexagon::STrih_indexed_cdnNotPt_nv_V4: + case Hexagon::STrih_indexed_shl_cPt_nv_V4: + case Hexagon::STrih_indexed_shl_cdnPt_nv_V4: + case Hexagon::STrih_indexed_shl_cNotPt_nv_V4: + case Hexagon::STrih_indexed_shl_cdnNotPt_nv_V4: + case Hexagon::POST_SThri_cPt_nv_V4: + case Hexagon::POST_SThri_cdnPt_nv_V4: + case Hexagon::POST_SThri_cNotPt_nv_V4: + case Hexagon::POST_SThri_cdnNotPt_nv_V4: + case Hexagon::STh_GP_cPt_nv_V4: + case Hexagon::STh_GP_cNotPt_nv_V4: + case Hexagon::STh_GP_cdnPt_nv_V4: + case Hexagon::STh_GP_cdnNotPt_nv_V4: + case Hexagon::STrih_GP_cPt_nv_V4: + case Hexagon::STrih_GP_cNotPt_nv_V4: + case Hexagon::STrih_GP_cdnPt_nv_V4: + case Hexagon::STrih_GP_cdnNotPt_nv_V4: + case Hexagon::STrih_abs_nv_V4: + case Hexagon::STrih_abs_cPt_nv_V4: + case Hexagon::STrih_abs_cdnPt_nv_V4: + case Hexagon::STrih_abs_cNotPt_nv_V4: + case Hexagon::STrih_abs_cdnNotPt_nv_V4: + case Hexagon::STrih_imm_abs_nv_V4: + case Hexagon::STrih_imm_abs_cPt_nv_V4: + case Hexagon::STrih_imm_abs_cdnPt_nv_V4: + case Hexagon::STrih_imm_abs_cNotPt_nv_V4: + case Hexagon::STrih_imm_abs_cdnNotPt_nv_V4: + + // Store Word + case Hexagon::STriw_nv_V4: + case Hexagon::STriw_indexed_nv_V4: + case Hexagon::STriw_indexed_shl_nv_V4: + case Hexagon::STriw_shl_nv_V4: + case Hexagon::STriw_GP_nv_V4: + case Hexagon::STw_GP_nv_V4: + case Hexagon::POST_STwri_nv_V4: + case Hexagon::STriw_cPt_nv_V4: + case Hexagon::STriw_cdnPt_nv_V4: + case Hexagon::STriw_cNotPt_nv_V4: + case Hexagon::STriw_cdnNotPt_nv_V4: + case Hexagon::STriw_indexed_cPt_nv_V4: + case Hexagon::STriw_indexed_cdnPt_nv_V4: + case Hexagon::STriw_indexed_cNotPt_nv_V4: + case Hexagon::STriw_indexed_cdnNotPt_nv_V4: + case Hexagon::STriw_indexed_shl_cPt_nv_V4: + case Hexagon::STriw_indexed_shl_cdnPt_nv_V4: + case Hexagon::STriw_indexed_shl_cNotPt_nv_V4: + case Hexagon::STriw_indexed_shl_cdnNotPt_nv_V4: + case Hexagon::POST_STwri_cPt_nv_V4: + case Hexagon::POST_STwri_cdnPt_nv_V4: + case Hexagon::POST_STwri_cNotPt_nv_V4: + case Hexagon::POST_STwri_cdnNotPt_nv_V4: + case Hexagon::STw_GP_cPt_nv_V4: + case Hexagon::STw_GP_cNotPt_nv_V4: + case Hexagon::STw_GP_cdnPt_nv_V4: + case Hexagon::STw_GP_cdnNotPt_nv_V4: + case Hexagon::STriw_GP_cPt_nv_V4: + case Hexagon::STriw_GP_cNotPt_nv_V4: + case Hexagon::STriw_GP_cdnPt_nv_V4: + case Hexagon::STriw_GP_cdnNotPt_nv_V4: + case Hexagon::STriw_abs_nv_V4: + case Hexagon::STriw_abs_cPt_nv_V4: + case Hexagon::STriw_abs_cdnPt_nv_V4: + case Hexagon::STriw_abs_cNotPt_nv_V4: + case Hexagon::STriw_abs_cdnNotPt_nv_V4: + case Hexagon::STriw_imm_abs_nv_V4: + case Hexagon::STriw_imm_abs_cPt_nv_V4: + case Hexagon::STriw_imm_abs_cdnPt_nv_V4: + case Hexagon::STriw_imm_abs_cNotPt_nv_V4: + case Hexagon::STriw_imm_abs_cdnNotPt_nv_V4: + return true; + + default: + return false; + } + return false; +} + +bool HexagonInstrInfo::isPostIncrement (const MachineInstr* MI) const { + switch (MI->getOpcode()) + { + // Load Byte + case Hexagon::POST_LDrib: + case Hexagon::POST_LDrib_cPt: + case Hexagon::POST_LDrib_cNotPt: + case Hexagon::POST_LDrib_cdnPt_V4: + case Hexagon::POST_LDrib_cdnNotPt_V4: + + // Load unsigned byte + case Hexagon::POST_LDriub: + case Hexagon::POST_LDriub_cPt: + case Hexagon::POST_LDriub_cNotPt: + case Hexagon::POST_LDriub_cdnPt_V4: + case Hexagon::POST_LDriub_cdnNotPt_V4: + + // Load halfword + case Hexagon::POST_LDrih: + case Hexagon::POST_LDrih_cPt: + case Hexagon::POST_LDrih_cNotPt: + case Hexagon::POST_LDrih_cdnPt_V4: + case Hexagon::POST_LDrih_cdnNotPt_V4: + + // Load unsigned halfword + case Hexagon::POST_LDriuh: + case Hexagon::POST_LDriuh_cPt: + case Hexagon::POST_LDriuh_cNotPt: + case Hexagon::POST_LDriuh_cdnPt_V4: + case Hexagon::POST_LDriuh_cdnNotPt_V4: + + // Load word + case Hexagon::POST_LDriw: + case Hexagon::POST_LDriw_cPt: + case Hexagon::POST_LDriw_cNotPt: + case Hexagon::POST_LDriw_cdnPt_V4: + case Hexagon::POST_LDriw_cdnNotPt_V4: + + // Load double word + case Hexagon::POST_LDrid: + case Hexagon::POST_LDrid_cPt: + case Hexagon::POST_LDrid_cNotPt: + case Hexagon::POST_LDrid_cdnPt_V4: + case Hexagon::POST_LDrid_cdnNotPt_V4: + + // Store byte + case Hexagon::POST_STbri: + case Hexagon::POST_STbri_cPt: + case Hexagon::POST_STbri_cNotPt: + case Hexagon::POST_STbri_cdnPt_V4: + case Hexagon::POST_STbri_cdnNotPt_V4: + + // Store halfword + case Hexagon::POST_SThri: + case Hexagon::POST_SThri_cPt: + case Hexagon::POST_SThri_cNotPt: + case Hexagon::POST_SThri_cdnPt_V4: + case Hexagon::POST_SThri_cdnNotPt_V4: + + // Store word + case Hexagon::POST_STwri: + case Hexagon::POST_STwri_cPt: + case Hexagon::POST_STwri_cNotPt: + case Hexagon::POST_STwri_cdnPt_V4: + case Hexagon::POST_STwri_cdnNotPt_V4: + + // Store double word + case Hexagon::POST_STdri: + case Hexagon::POST_STdri_cPt: + case Hexagon::POST_STdri_cNotPt: + case Hexagon::POST_STdri_cdnPt_V4: + case Hexagon::POST_STdri_cdnNotPt_V4: + return true; + + default: + return false; + } +} + +bool HexagonInstrInfo::isSaveCalleeSavedRegsCall(const MachineInstr *MI) const { + return MI->getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4; +} bool HexagonInstrInfo::isPredicable(MachineInstr *MI) const { bool isPred = MI->getDesc().isPredicable(); @@ -1587,6 +2449,24 @@ isSpillPredRegOp(const MachineInstr *MI) const { return false; } +bool HexagonInstrInfo:: +isConditionalTransfer (const MachineInstr *MI) const { + switch (MI->getOpcode()) { + case Hexagon::TFR_cPt: + case Hexagon::TFR_cNotPt: + case Hexagon::TFRI_cPt: + case Hexagon::TFRI_cNotPt: + case Hexagon::TFR_cdnPt: + case Hexagon::TFR_cdnNotPt: + case Hexagon::TFRI_cdnPt: + case Hexagon::TFRI_cdnNotPt: + return true; + + default: + return false; + } + return false; +} bool HexagonInstrInfo::isConditionalALU32 (const MachineInstr* MI) const { const HexagonRegisterInfo& QRI = getRegisterInfo(); @@ -1626,7 +2506,6 @@ bool HexagonInstrInfo::isConditionalALU32 (const MachineInstr* MI) const { } } - bool HexagonInstrInfo:: isConditionalLoad (const MachineInstr* MI) const { const HexagonRegisterInfo& QRI = getRegisterInfo(); @@ -1700,6 +2579,136 @@ isConditionalLoad (const MachineInstr* MI) const { } } +// Returns true if an instruction is a conditional store. +// +// Note: It doesn't include conditional new-value stores as they can't be +// converted to .new predicate. +// +// p.new NV store [ if(p0.new)memw(R0+#0)=R2.new ] +// ^ ^ +// / \ (not OK. it will cause new-value store to be +// / X conditional on p0.new while R2 producer is +// / \ on p0) +// / \. +// p.new store p.old NV store +// [if(p0.new)memw(R0+#0)=R2] [if(p0)memw(R0+#0)=R2.new] +// ^ ^ +// \ / +// \ / +// \ / +// p.old store +// [if (p0)memw(R0+#0)=R2] +// +// The above diagram shows the steps involoved in the conversion of a predicated +// store instruction to its .new predicated new-value form. +// +// The following set of instructions further explains the scenario where +// conditional new-value store becomes invalid when promoted to .new predicate +// form. +// +// { 1) if (p0) r0 = add(r1, r2) +// 2) p0 = cmp.eq(r3, #0) } +// +// 3) if (p0) memb(r1+#0) = r0 --> this instruction can't be grouped with +// the first two instructions because in instr 1, r0 is conditional on old value +// of p0 but its use in instr 3 is conditional on p0 modified by instr 2 which +// is not valid for new-value stores. +bool HexagonInstrInfo:: +isConditionalStore (const MachineInstr* MI) const { + const HexagonRegisterInfo& QRI = getRegisterInfo(); + switch (MI->getOpcode()) + { + case Hexagon::STrib_imm_cPt_V4 : + case Hexagon::STrib_imm_cNotPt_V4 : + case Hexagon::STrib_indexed_shl_cPt_V4 : + case Hexagon::STrib_indexed_shl_cNotPt_V4 : + case Hexagon::STrib_cPt : + case Hexagon::STrib_cNotPt : + case Hexagon::POST_STbri_cPt : + case Hexagon::POST_STbri_cNotPt : + case Hexagon::STrid_indexed_cPt : + case Hexagon::STrid_indexed_cNotPt : + case Hexagon::STrid_indexed_shl_cPt_V4 : + case Hexagon::POST_STdri_cPt : + case Hexagon::POST_STdri_cNotPt : + case Hexagon::STrih_cPt : + case Hexagon::STrih_cNotPt : + case Hexagon::STrih_indexed_cPt : + case Hexagon::STrih_indexed_cNotPt : + case Hexagon::STrih_imm_cPt_V4 : + case Hexagon::STrih_imm_cNotPt_V4 : + case Hexagon::STrih_indexed_shl_cPt_V4 : + case Hexagon::STrih_indexed_shl_cNotPt_V4 : + case Hexagon::POST_SThri_cPt : + case Hexagon::POST_SThri_cNotPt : + case Hexagon::STriw_cPt : + case Hexagon::STriw_cNotPt : + case Hexagon::STriw_indexed_cPt : + case Hexagon::STriw_indexed_cNotPt : + case Hexagon::STriw_imm_cPt_V4 : + case Hexagon::STriw_imm_cNotPt_V4 : + case Hexagon::STriw_indexed_shl_cPt_V4 : + case Hexagon::STriw_indexed_shl_cNotPt_V4 : + case Hexagon::POST_STwri_cPt : + case Hexagon::POST_STwri_cNotPt : + return QRI.Subtarget.hasV4TOps(); + + // V4 global address store before promoting to dot new. + case Hexagon::STrid_GP_cPt_V4 : + case Hexagon::STrid_GP_cNotPt_V4 : + case Hexagon::STrib_GP_cPt_V4 : + case Hexagon::STrib_GP_cNotPt_V4 : + case Hexagon::STrih_GP_cPt_V4 : + case Hexagon::STrih_GP_cNotPt_V4 : + case Hexagon::STriw_GP_cPt_V4 : + case Hexagon::STriw_GP_cNotPt_V4 : + case Hexagon::STd_GP_cPt_V4 : + case Hexagon::STd_GP_cNotPt_V4 : + case Hexagon::STb_GP_cPt_V4 : + case Hexagon::STb_GP_cNotPt_V4 : + case Hexagon::STh_GP_cPt_V4 : + case Hexagon::STh_GP_cNotPt_V4 : + case Hexagon::STw_GP_cPt_V4 : + case Hexagon::STw_GP_cNotPt_V4 : + return QRI.Subtarget.hasV4TOps(); + + // Predicated new value stores (i.e. if (p0) memw(..)=r0.new) are excluded + // from the "Conditional Store" list. Because a predicated new value store + // would NOT be promoted to a double dot new store. See diagram below: + // This function returns yes for those stores that are predicated but not + // yet promoted to predicate dot new instructions. + // + // +---------------------+ + // /-----| if (p0) memw(..)=r0 |---------\~ + // || +---------------------+ || + // promote || /\ /\ || promote + // || /||\ /||\ || + // \||/ demote || \||/ + // \/ || || \/ + // +-------------------------+ || +-------------------------+ + // | if (p0.new) memw(..)=r0 | || | if (p0) memw(..)=r0.new | + // +-------------------------+ || +-------------------------+ + // || || || + // || demote \||/ + // promote || \/ NOT possible + // || || /\~ + // \||/ || /||\~ + // \/ || || + // +-----------------------------+ + // | if (p0.new) memw(..)=r0.new | + // +-----------------------------+ + // Double Dot New Store + // + + default: + return false; + + } + return false; +} + + + DFAPacketizer *HexagonInstrInfo:: CreateTargetScheduleState(const TargetMachine *TM, const ScheduleDAG *DAG) const { diff --git a/lib/Target/Hexagon/HexagonInstrInfo.h b/lib/Target/Hexagon/HexagonInstrInfo.h index 730687036c8..6a45871b67e 100644 --- a/lib/Target/Hexagon/HexagonInstrInfo.h +++ b/lib/Target/Hexagon/HexagonInstrInfo.h @@ -160,10 +160,20 @@ public: bool isS8_Immediate(const int value) const; bool isS6_Immediate(const int value) const; + bool isSaveCalleeSavedRegsCall(const MachineInstr* MI) const; + bool isConditionalTransfer(const MachineInstr* MI) const; bool isConditionalALU32 (const MachineInstr* MI) const; bool isConditionalLoad (const MachineInstr* MI) const; + bool isConditionalStore(const MachineInstr* MI) const; bool isDeallocRet(const MachineInstr *MI) const; unsigned getInvertedPredicatedOpcode(const int Opc) const; + bool isExtendable(const MachineInstr* MI) const; + bool isExtended(const MachineInstr* MI) const; + bool isPostIncrement(const MachineInstr* MI) const; + bool isNewValueStore(const MachineInstr* MI) const; + bool isNewValueJump(const MachineInstr* MI) const; + unsigned getImmExtForm(const MachineInstr* MI) const; + unsigned getNormalBranchForm(const MachineInstr* MI) const; private: int getMatchingCondBranchOpcode(int Opc, bool sense) const; diff --git a/lib/Target/Hexagon/HexagonInstrInfo.td b/lib/Target/Hexagon/HexagonInstrInfo.td index b563ac3c613..fd5adef0f63 100644 --- a/lib/Target/Hexagon/HexagonInstrInfo.td +++ b/lib/Target/Hexagon/HexagonInstrInfo.td @@ -3046,3 +3046,7 @@ include "HexagonInstrInfoV3.td" //===----------------------------------------------------------------------===// include "HexagonInstrInfoV4.td" + +//===----------------------------------------------------------------------===// +// V4 Instructions - +//===----------------------------------------------------------------------===// diff --git a/lib/Target/Hexagon/HexagonInstrInfoV3.td b/lib/Target/Hexagon/HexagonInstrInfoV3.td index a73897ee345..2bd6770efd7 100644 --- a/lib/Target/Hexagon/HexagonInstrInfoV3.td +++ b/lib/Target/Hexagon/HexagonInstrInfoV3.td @@ -41,10 +41,11 @@ let isCall = 1, neverHasSideEffects = 1, } +// Jump to address from register // if(p?.new) jumpr:t r? let isReturn = 1, isTerminator = 1, isBarrier = 1, Defs = [PC], Uses = [R31] in { - def JMPR_cPnewt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), + def JMPR_cdnPt_V3: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), "if ($src1.new) jumpr:t $src2", []>, Requires<[HasV3T]>; } @@ -52,7 +53,7 @@ let isReturn = 1, isTerminator = 1, isBarrier = 1, // if (!p?.new) jumpr:t r? let isReturn = 1, isTerminator = 1, isBarrier = 1, Defs = [PC], Uses = [R31] in { - def JMPR_cNotPnewt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), + def JMPR_cdnNotPt_V3: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), "if (!$src1.new) jumpr:t $src2", []>, Requires<[HasV3T]>; } @@ -61,7 +62,7 @@ let isReturn = 1, isTerminator = 1, isBarrier = 1, // if(p?.new) jumpr:nt r? let isReturn = 1, isTerminator = 1, isBarrier = 1, Defs = [PC], Uses = [R31] in { - def JMPR_cPnewNt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), + def JMPR_cdnPnt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), "if ($src1.new) jumpr:nt $src2", []>, Requires<[HasV3T]>; } @@ -69,7 +70,7 @@ let isReturn = 1, isTerminator = 1, isBarrier = 1, // if (!p?.new) jumpr:nt r? let isReturn = 1, isTerminator = 1, isBarrier = 1, Defs = [PC], Uses = [R31] in { - def JMPR_cNotPnewNt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), + def JMPR_cdnNotPnt: JRInst<(outs), (ins PredRegs:$src1, IntRegs:$src2), "if (!$src1.new) jumpr:nt $src2", []>, Requires<[HasV3T]>; } @@ -86,20 +87,22 @@ let AddedComplexity = 200 in def MAXw_dd : ALU64_rr<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2), "$dst = max($src2, $src1)", - [(set DoubleRegs:$dst, (select (i1 (setlt DoubleRegs:$src2, - DoubleRegs:$src1)), - DoubleRegs:$src1, - DoubleRegs:$src2))]>, + [(set (i64 DoubleRegs:$dst), + (i64 (select (i1 (setlt (i64 DoubleRegs:$src2), + (i64 DoubleRegs:$src1))), + (i64 DoubleRegs:$src1), + (i64 DoubleRegs:$src2))))]>, Requires<[HasV3T]>; let AddedComplexity = 200 in def MINw_dd : ALU64_rr<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2), "$dst = min($src2, $src1)", - [(set DoubleRegs:$dst, (select (i1 (setgt DoubleRegs:$src2, - DoubleRegs:$src1)), - DoubleRegs:$src1, - DoubleRegs:$src2))]>, + [(set (i64 DoubleRegs:$dst), + (i64 (select (i1 (setgt (i64 DoubleRegs:$src2), + (i64 DoubleRegs:$src1))), + (i64 DoubleRegs:$src1), + (i64 DoubleRegs:$src2))))]>, Requires<[HasV3T]>; //===----------------------------------------------------------------------===// @@ -109,25 +112,25 @@ Requires<[HasV3T]>; -//def : Pat <(brcond (i1 (seteq IntRegs:$src1, 0)), bb:$offset), -// (JMP_RegEzt IntRegs:$src1, bb:$offset)>, Requires<[HasV3T]>; +//def : Pat <(brcond (i1 (seteq (i32 IntRegs:$src1), 0)), bb:$offset), +// (JMP_RegEzt (i32 IntRegs:$src1), bb:$offset)>, Requires<[HasV3T]>; -//def : Pat <(brcond (i1 (setne IntRegs:$src1, 0)), bb:$offset), -// (JMP_RegNzt IntRegs:$src1, bb:$offset)>, Requires<[HasV3T]>; +//def : Pat <(brcond (i1 (setne (i32 IntRegs:$src1), 0)), bb:$offset), +// (JMP_RegNzt (i32 IntRegs:$src1), bb:$offset)>, Requires<[HasV3T]>; -//def : Pat <(brcond (i1 (setle IntRegs:$src1, 0)), bb:$offset), -// (JMP_RegLezt IntRegs:$src1, bb:$offset)>, Requires<[HasV3T]>; +//def : Pat <(brcond (i1 (setle (i32 IntRegs:$src1), 0)), bb:$offset), +// (JMP_RegLezt (i32 IntRegs:$src1), bb:$offset)>, Requires<[HasV3T]>; -//def : Pat <(brcond (i1 (setge IntRegs:$src1, 0)), bb:$offset), -// (JMP_RegGezt IntRegs:$src1, bb:$offset)>, Requires<[HasV3T]>; +//def : Pat <(brcond (i1 (setge (i32 IntRegs:$src1), 0)), bb:$offset), +// (JMP_RegGezt (i32 IntRegs:$src1), bb:$offset)>, Requires<[HasV3T]>; -//def : Pat <(brcond (i1 (setgt IntRegs:$src1, -1)), bb:$offset), -// (JMP_RegGezt IntRegs:$src1, bb:$offset)>, Requires<[HasV3T]>; +//def : Pat <(brcond (i1 (setgt (i32 IntRegs:$src1), -1)), bb:$offset), +// (JMP_RegGezt (i32 IntRegs:$src1), bb:$offset)>, Requires<[HasV3T]>; // Map call instruction -def : Pat<(call IntRegs:$dst), - (CALLRv3 IntRegs:$dst)>, Requires<[HasV3T]>; +def : Pat<(call (i32 IntRegs:$dst)), + (CALLRv3 (i32 IntRegs:$dst))>, Requires<[HasV3T]>; def : Pat<(call tglobaladdr:$dst), (CALLv3 tglobaladdr:$dst)>, Requires<[HasV3T]>; def : Pat<(call texternalsym:$dst), diff --git a/lib/Target/Hexagon/HexagonInstrInfoV4.td b/lib/Target/Hexagon/HexagonInstrInfoV4.td index 9e60cf26d08..f507e4f37c1 100644 --- a/lib/Target/Hexagon/HexagonInstrInfoV4.td +++ b/lib/Target/Hexagon/HexagonInstrInfoV4.td @@ -11,6 +11,11 @@ // //===----------------------------------------------------------------------===// +def IMMEXT : Immext<(outs), (ins), + "##immext //should never emit this", + []>, + Requires<[HasV4T]>; + // Hexagon V4 Architecture spec defines 8 instruction classes: // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the // compiler) @@ -250,23 +255,151 @@ def ZXTH_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst), []>, Requires<[HasV4T]>; +// Generate frame index addresses. +let neverHasSideEffects = 1, isReMaterializable = 1 in +def TFR_FI_immext_V4 : ALU32_ri<(outs IntRegs:$dst), + (ins IntRegs:$src1, s32Imm:$offset), + "$dst = add($src1, ##$offset)", + []>, + Requires<[HasV4T]>; + //===----------------------------------------------------------------------===// // ALU32 - //===----------------------------------------------------------------------===// +//===----------------------------------------------------------------------===// +// ALU32/PERM + +//===----------------------------------------------------------------------===// + +// Combine +// Rdd=combine(Rs, #s8) +let neverHasSideEffects = 1 in +def COMBINE_ri_V4 : ALU32_ri<(outs DoubleRegs:$dst), + (ins IntRegs:$src1, s8Imm:$src2), + "$dst = combine($src1, #$src2)", + []>, + Requires<[HasV4T]>; +// Rdd=combine(#s8, Rs) +let neverHasSideEffects = 1 in +def COMBINE_ir_V4 : ALU32_ir<(outs DoubleRegs:$dst), + (ins s8Imm:$src1, IntRegs:$src2), + "$dst = combine(#$src1, $src2)", + []>, + Requires<[HasV4T]>; +//===----------------------------------------------------------------------===// +// ALU32/PERM + +//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===// // LD + //===----------------------------------------------------------------------===// -/// -/// Make sure that in post increment load, the first operand is always the post -/// increment operand. -/// -//// Load doubleword. -// Rdd=memd(Re=#U6) +// +// These absolute set addressing mode instructions accept immediate as +// an operand. We have duplicated these patterns to take global address. +let neverHasSideEffects = 1 in +def LDrid_abs_setimm_V4 : LDInst<(outs DoubleRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memd($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memb(Re=#U6) +let neverHasSideEffects = 1 in +def LDrib_abs_setimm_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memb($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memh(Re=#U6) +let neverHasSideEffects = 1 in +def LDrih_abs_setimm_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memh($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memub(Re=#U6) +let neverHasSideEffects = 1 in +def LDriub_abs_setimm_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memub($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memuh(Re=#U6) +let neverHasSideEffects = 1 in +def LDriuh_abs_setimm_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memuh($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memw(Re=#U6) +let neverHasSideEffects = 1 in +def LDriw_abs_setimm_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins u6Imm:$addr), + "$dst1 = memw($dst2=#$addr)", + []>, + Requires<[HasV4T]>; + +// Following patterns are defined for absolute set addressing mode +// instruction which take global address as operand. +let mayLoad = 1, neverHasSideEffects = 1 in +def LDrid_abs_set_V4 : LDInst<(outs DoubleRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memd($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memb(Re=#U6) +let mayLoad = 1, neverHasSideEffects = 1 in +def LDrib_abs_set_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memb($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memh(Re=#U6) +let mayLoad = 1, neverHasSideEffects = 1 in +def LDrih_abs_set_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memh($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memub(Re=#U6) +let mayLoad = 1, neverHasSideEffects = 1 in +def LDriub_abs_set_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memub($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memuh(Re=#U6) +let mayLoad = 1, neverHasSideEffects = 1 in +def LDriuh_abs_set_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memuh($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Rd=memw(Re=#U6) +let mayLoad = 1, neverHasSideEffects = 1 in +def LDriw_abs_set_V4 : LDInst<(outs IntRegs:$dst1, IntRegs:$dst2), + (ins globaladdress:$addr), + "$dst1 = memw($dst2=##$addr)", + []>, + Requires<[HasV4T]>; + +// Load doubleword. +// +// Make sure that in post increment load, the first operand is always the post +// increment operand. +// // Rdd=memd(Rs+Rt<<#u2) // Special case pattern for indexed load without offset which is easier to // match. AddedComplexity of this pattern should be lower than base+offset load @@ -276,17 +409,19 @@ let AddedComplexity = 10, isPredicable = 1 in def LDrid_indexed_V4 : LDInst<(outs DoubleRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memd($src1+$src2<<#0)", - [(set DoubleRegs:$dst, (load (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i64 DoubleRegs:$dst), + (i64 (load (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDrid_indexed_shl_V4 : LDInst<(outs DoubleRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memd($src1+$src2<<#$offset)", - [(set DoubleRegs:$dst, (load (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i64 DoubleRegs:$dst), + (i64 (load (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; //// Load doubleword conditionally. @@ -362,60 +497,62 @@ def LDrid_indexed_shl_cdnNotPt_V4 : LDInst<(outs DoubleRegs:$dst), // Rdd=memd(Rt<<#u2+#U6) //// Load byte. -// Rd=memb(Re=#U6) - // Rd=memb(Rs+Rt<<#u2) let AddedComplexity = 10, isPredicable = 1 in def LDrib_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memb($src1+$src2<<#0)", - [(set IntRegs:$dst, (sextloadi8 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (sextloadi8 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 10, isPredicable = 1 in def LDriub_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memub($src1+$src2<<#0)", - [(set IntRegs:$dst, (zextloadi8 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (zextloadi8 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 10, isPredicable = 1 in def LDriub_ae_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memub($src1+$src2<<#0)", - [(set IntRegs:$dst, (extloadi8 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (extloadi8 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDrib_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memb($src1+$src2<<#$offset)", - [(set IntRegs:$dst, - (sextloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (sextloadi8 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDriub_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memub($src1+$src2<<#$offset)", - [(set IntRegs:$dst, - (zextloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (zextloadi8 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDriub_ae_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memub($src1+$src2<<#$offset)", - [(set IntRegs:$dst, (extloadi8 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (extloadi8 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; //// Load byte conditionally. @@ -561,31 +698,32 @@ def LDriub_indexed_shl_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), // Rd=memb(Rt<<#u2+#U6) //// Load halfword -// Rd=memh(Re=#U6) - // Rd=memh(Rs+Rt<<#u2) let AddedComplexity = 10, isPredicable = 1 in def LDrih_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memh($src1+$src2<<#0)", - [(set IntRegs:$dst, (sextloadi16 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (sextloadi16 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 10, isPredicable = 1 in def LDriuh_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memuh($src1+$src2<<#0)", - [(set IntRegs:$dst, (zextloadi16 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (zextloadi16 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; let AddedComplexity = 10, isPredicable = 1 in def LDriuh_ae_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memuh($src1+$src2<<#0)", - [(set IntRegs:$dst, (extloadi16 (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (extloadi16 (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; // Rd=memh(Rs+Rt<<#u2) @@ -593,30 +731,30 @@ let AddedComplexity = 40, isPredicable = 1 in def LDrih_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memh($src1+$src2<<#$offset)", - [(set IntRegs:$dst, - (sextloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (sextloadi16 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDriuh_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memuh($src1+$src2<<#$offset)", - [(set IntRegs:$dst, - (zextloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (zextloadi16 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; let AddedComplexity = 40, isPredicable = 1 in def LDriuh_ae_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memuh($src1+$src2<<#$offset)", - [(set IntRegs:$dst, - (extloadi16 (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (extloadi16 (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; //// Load halfword conditionally. @@ -762,6 +900,14 @@ def LDriuh_indexed_shl_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), // Rd=memh(Rt<<#u2+#U6) //// Load word. +// Load predicate: Fix for bug 5279. +let mayLoad = 1, neverHasSideEffects = 1 in +def LDriw_pred_V4 : LDInst<(outs PredRegs:$dst), + (ins MEMri:$addr), + "Error; should not emit", + []>, + Requires<[HasV4T]>; + // Rd=memw(Re=#U6) // Rd=memw(Rs+Rt<<#u2) @@ -769,8 +915,9 @@ let AddedComplexity = 10, isPredicable = 1 in def LDriw_indexed_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst=memw($src1+$src2<<#0)", - [(set IntRegs:$dst, (load (add IntRegs:$src1, - IntRegs:$src2)))]>, + [(set (i32 IntRegs:$dst), + (i32 (load (add (i32 IntRegs:$src1), + (i32 IntRegs:$src2)))))]>, Requires<[HasV4T]>; // Rd=memw(Rs+Rt<<#u2) @@ -778,9 +925,10 @@ let AddedComplexity = 40, isPredicable = 1 in def LDriw_indexed_shl_V4 : LDInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset), "$dst=memw($src1+$src2<<#$offset)", - [(set IntRegs:$dst, (load (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$offset))))]>, + [(set (i32 IntRegs:$dst), + (i32 (load (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$offset)))))]>, Requires<[HasV4T]>; //// Load word conditionally. @@ -955,6 +1103,633 @@ def POST_LDriw_cdnNotPt_V4 : LDInstPI<(outs IntRegs:$dst1, IntRegs:$dst2), "$src2 = $dst2">, Requires<[HasV4T]>; +/// Load from global offset + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDrid_GP_V4 : LDInst<(outs DoubleRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memd(#$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrid_GP_cPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memd(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrid_GP_cNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memd(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrid_GP_cdnPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memd(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrid_GP_cdnNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memd(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDrib_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memb(#$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrib_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memb(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrib_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memb(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrib_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memb(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrib_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memb(##$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDriub_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memub(#$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriub_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memub(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriub_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memub(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriub_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memub(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriub_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memub(##$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDrih_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memh(#$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrih_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrih_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrih_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDrih_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDriuh_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memuh(#$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriuh_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memuh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriuh_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memuh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriuh_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memuh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriuh_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memuh(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDriw_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global, u16Imm:$offset), + "$dst=memw(#$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriw_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1) $dst=memw(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriw_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1) $dst=memw(##$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriw_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if ($src1.new) $dst=memw(##$global+$offset)", + []>, + Requires<[HasV4T]>; + +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDriw_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset), + "if (!$src1.new) $dst=memw(##$global+$offset)", + []>, + Requires<[HasV4T]>; + + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDd_GP_V4 : LDInst<(outs DoubleRegs:$dst), + (ins globaladdress:$global), + "$dst=memd(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rtt=memd(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDd_GP_cPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memd(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rtt=memd(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDd_GP_cNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memd(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rtt=memd(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDd_GP_cdnPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memd(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rtt=memd(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDd_GP_cdnNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memd(##$global)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDb_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global), + "$dst=memb(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memb(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDb_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memb(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memb(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDb_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memb(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memb(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDb_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memb(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memb(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDb_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memb(##$global)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDub_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global), + "$dst=memub(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memub(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDub_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memub(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rt=memub(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDub_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memub(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memub(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDub_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memub(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rt=memub(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDub_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memub(##$global)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDh_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global), + "$dst=memh(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDh_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDh_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDh_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDh_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memh(##$global)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDuh_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global), + "$dst=memuh(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memuh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDuh_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memuh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memuh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDuh_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memuh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memuh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDuh_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memuh(##$global)", + []>, + Requires<[HasV4T]>; + +// if (!Pv) Rt=memuh(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDuh_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memuh(##$global)", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayLoad = 1, neverHasSideEffects = 1 in +def LDw_GP_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$global), + "$dst=memw(#$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memw(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDw_GP_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1) $dst=memw(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rt=memw(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDw_GP_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1) $dst=memw(##$global)", + []>, + Requires<[HasV4T]>; + +// if (Pv) Rt=memw(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDw_GP_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if ($src1.new) $dst=memw(##$global)", + []>, + Requires<[HasV4T]>; + + +// if (!Pv) Rt=memw(##global) +let mayLoad = 1, neverHasSideEffects = 1, isPredicated = 1 in +def LDw_GP_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$global), + "if (!$src1.new) $dst=memw(##$global)", + []>, + Requires<[HasV4T]>; + + + +def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)), + (i64 (LDd_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)), + (i32 (LDw_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)), + (i32 (LDuh_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)), + (i32 (LDub_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memw(#foo + 0) +let AddedComplexity = 100 in +def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))), + (i64 (LDd_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd +let AddedComplexity = 100 in +def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))), + (i1 (TFR_PdRs (i32 (LDb_GP_V4 tglobaladdr:$global))))>, + Requires<[HasV4T]>; + +// When the Interprocedural Global Variable optimizer realizes that a certain +// global variable takes only two constant values, it shrinks the global to +// a boolean. Catch those loads here in the following 3 patterns. +let AddedComplexity = 100 in +def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDb_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +let AddedComplexity = 100 in +def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDb_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memb(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDb_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memb(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDb_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +let AddedComplexity = 100 in +def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDub_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memub(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDub_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memh(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDh_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memh(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDh_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memuh(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDuh_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress) -> memw(#foo) +let AddedComplexity = 100 in +def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))), + (i32 (LDw_GP_V4 tglobaladdr:$global))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_64 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (i64 (LDrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_32 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (i32 (LDriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_16 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (i32 (LDriuh_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +def : Pat <(atomic_load_8 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (i32 (LDriub_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memd(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i64 (load (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i64 (LDrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memb(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (extloadi8 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memb(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (sextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memub(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (zextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDriub_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memuh(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (extloadi16 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memh(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (sextloadi16 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + + +// Map from load(globaladdress + x) -> memuh(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (zextloadi16 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDriuh_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + +// Map from load(globaladdress + x) -> memw(#foo + x) +let AddedComplexity = 100 in +def : Pat <(i32 (load (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset))), + (i32 (LDriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>, + Requires<[HasV4T]>; + //===----------------------------------------------------------------------===// // LD - @@ -971,18 +1746,70 @@ def POST_LDriw_cdnNotPt_V4 : LDInstPI<(outs IntRegs:$dst1, IntRegs:$dst2), /// last operand. /// -// Store doubleword. // memd(Re=#U6)=Rtt -// TODO: needs to be implemented +def STrid_abs_setimm_V4 : STInst<(outs IntRegs:$dst1), + (ins DoubleRegs:$src1, u6Imm:$src2), + "memd($dst1=#$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memb(Re=#U6)=Rs +def STrib_abs_setimm_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, u6Imm:$src2), + "memb($dst1=#$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memh(Re=#U6)=Rs +def STrih_abs_setimm_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, u6Imm:$src2), + "memh($dst1=#$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memw(Re=#U6)=Rs +def STriw_abs_setimm_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, u6Imm:$src2), + "memw($dst1=#$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memd(Re=#U6)=Rtt +def STrid_abs_set_V4 : STInst<(outs IntRegs:$dst1), + (ins DoubleRegs:$src1, globaladdress:$src2), + "memd($dst1=##$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memb(Re=#U6)=Rs +def STrib_abs_set_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, globaladdress:$src2), + "memb($dst1=##$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memh(Re=#U6)=Rs +def STrih_abs_set_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, globaladdress:$src2), + "memh($dst1=##$src2) = $src1", + []>, + Requires<[HasV4T]>; + +// memw(Re=#U6)=Rs +def STriw_abs_set_V4 : STInst<(outs IntRegs:$dst1), + (ins IntRegs:$src1, globaladdress:$src2), + "memw($dst1=##$src2) = $src1", + []>, + Requires<[HasV4T]>; -// memd(Rs+#s11:3)=Rtt // memd(Rs+Ru<<#u2)=Rtt let AddedComplexity = 10, isPredicable = 1 in def STrid_indexed_shl_V4 : STInst<(outs), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, DoubleRegs:$src4), "memd($src1+$src2<<#$src3) = $src4", - [(store DoubleRegs:$src4, (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$src3)))]>, + [(store (i64 DoubleRegs:$src4), + (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), u2ImmPred:$src3)))]>, Requires<[HasV4T]>; // memd(Ru<<#u2+#U6)=Rtt @@ -990,9 +1817,9 @@ let AddedComplexity = 10 in def STrid_shl_V4 : STInst<(outs), (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, DoubleRegs:$src4), "memd($src1<<#$src2+#$src3) = $src4", - [(store DoubleRegs:$src4, (shl IntRegs:$src1, - (add u2ImmPred:$src2, - u6ImmPred:$src3)))]>, + [(store (i64 DoubleRegs:$src4), + (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), + u6ImmPred:$src3))]>, Requires<[HasV4T]>; // memd(Rx++#s4:3)=Rtt @@ -1009,7 +1836,8 @@ def STrid_shl_V4 : STInst<(outs), // if ([!]Pv[.new]) memd(Rs+#u6:3)=Rtt // if (Pv) memd(Rs+#u6:3)=Rtt // if (Pv.new) memd(Rs+#u6:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2), "if ($src1.new) memd($addr) = $src2", @@ -1018,7 +1846,8 @@ def STrid_cdnPt_V4 : STInst<(outs), // if (!Pv) memd(Rs+#u6:3)=Rtt // if (!Pv.new) memd(Rs+#u6:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2), "if (!$src1.new) memd($addr) = $src2", @@ -1027,7 +1856,8 @@ def STrid_cdnNotPt_V4 : STInst<(outs), // if (Pv) memd(Rs+#u6:3)=Rtt // if (Pv.new) memd(Rs+#u6:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3, DoubleRegs:$src4), @@ -1037,7 +1867,8 @@ def STrid_indexed_cdnPt_V4 : STInst<(outs), // if (!Pv) memd(Rs+#u6:3)=Rtt // if (!Pv.new) memd(Rs+#u6:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3, DoubleRegs:$src4), @@ -1047,7 +1878,8 @@ def STrid_indexed_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memd(Rs+Ru<<#u2)=Rtt // if (Pv) memd(Rs+Ru<<#u2)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_shl_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, DoubleRegs:$src5), @@ -1056,15 +1888,17 @@ def STrid_indexed_shl_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memd(Rs+Ru<<#u2)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_shl_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, DoubleRegs:$src5), - "if ($src1) memd($src2+$src3<<#$src4) = $src5", + "if ($src1.new) memd($src2+$src3<<#$src4) = $src5", []>, Requires<[HasV4T]>; // if (!Pv) memd(Rs+Ru<<#u2)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_shl_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, DoubleRegs:$src5), @@ -1072,7 +1906,8 @@ def STrid_indexed_shl_cNotPt_V4 : STInst<(outs), []>, Requires<[HasV4T]>; // if (!Pv.new) memd(Rs+Ru<<#u2)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrid_indexed_shl_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, DoubleRegs:$src5), @@ -1083,7 +1918,8 @@ def STrid_indexed_shl_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memd(Rx++#s4:3)=Rtt // if (Pv) memd(Rx++#s4:3)=Rtt // if (Pv.new) memd(Rx++#s4:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def POST_STdri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3, s4_3Imm:$offset), @@ -1094,7 +1930,8 @@ def POST_STdri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), // if (!Pv) memd(Rx++#s4:3)=Rtt // if (!Pv.new) memd(Rx++#s4:3)=Rtt -let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1 in +let AddedComplexity = 10, mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def POST_STdri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3, s4_3Imm:$offset), @@ -1105,15 +1942,12 @@ def POST_STdri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), // Store byte. -// memb(Re=#U6)=Rt -// TODO: needs to be implemented. -// memb(Rs+#s11:0)=Rt // memb(Rs+#u6:0)=#S8 let AddedComplexity = 10, isPredicable = 1 in def STrib_imm_V4 : STInst<(outs), (ins IntRegs:$src1, u6_0Imm:$src2, s8Imm:$src3), "memb($src1+#$src2) = #$src3", - [(truncstorei8 s8ImmPred:$src3, (add IntRegs:$src1, + [(truncstorei8 s8ImmPred:$src3, (add (i32 IntRegs:$src1), u6_0ImmPred:$src2))]>, Requires<[HasV4T]>; @@ -1122,9 +1956,10 @@ let AddedComplexity = 10, isPredicable = 1 in def STrib_indexed_shl_V4 : STInst<(outs), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4), "memb($src1+$src2<<#$src3) = $src4", - [(truncstorei8 IntRegs:$src4, (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$src3)))]>, + [(truncstorei8 (i32 IntRegs:$src4), + (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$src3)))]>, Requires<[HasV4T]>; // memb(Ru<<#u2+#U6)=Rt @@ -1132,9 +1967,9 @@ let AddedComplexity = 10 in def STrib_shl_V4 : STInst<(outs), (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4), "memb($src1<<#$src2+#$src3) = $src4", - [(truncstorei8 IntRegs:$src4, (shl IntRegs:$src1, - (add u2ImmPred:$src2, - u6ImmPred:$src3)))]>, + [(truncstorei8 (i32 IntRegs:$src4), + (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), + u6ImmPred:$src3))]>, Requires<[HasV4T]>; // memb(Rx++#s4:0:circ(Mu))=Rt @@ -1148,7 +1983,8 @@ def STrib_shl_V4 : STInst<(outs), // if ([!]Pv[.new]) memb(#u6)=Rt // if ([!]Pv[.new]) memb(Rs+#u6:0)=#S6 // if (Pv) memb(Rs+#u6:0)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_imm_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4), "if ($src1) memb($src2+#$src3) = #$src4", @@ -1156,7 +1992,8 @@ def STrib_imm_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memb(Rs+#u6:0)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_imm_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4), "if ($src1.new) memb($src2+#$src3) = #$src4", @@ -1164,7 +2001,8 @@ def STrib_imm_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memb(Rs+#u6:0)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_imm_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4), "if (!$src1) memb($src2+#$src3) = #$src4", @@ -1172,7 +2010,8 @@ def STrib_imm_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+#u6:0)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_imm_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4), "if (!$src1.new) memb($src2+#$src3) = #$src4", @@ -1182,7 +2021,8 @@ def STrib_imm_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memb(Rs+#u6:0)=Rt // if (Pv) memb(Rs+#u6:0)=Rt // if (Pv.new) memb(Rs+#u6:0)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memb($addr) = $src2", @@ -1191,7 +2031,8 @@ def STrib_cdnPt_V4 : STInst<(outs), // if (!Pv) memb(Rs+#u6:0)=Rt // if (!Pv.new) memb(Rs+#u6:0)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memb($addr) = $src2", @@ -1201,7 +2042,8 @@ def STrib_cdnNotPt_V4 : STInst<(outs), // if (Pv) memb(Rs+#u6:0)=Rt // if (!Pv) memb(Rs+#u6:0)=Rt // if (Pv.new) memb(Rs+#u6:0)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if ($src1.new) memb($src2+#$src3) = $src4", @@ -1209,7 +2051,8 @@ def STrib_indexed_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+#u6:0)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if (!$src1.new) memb($src2+#$src3) = $src4", @@ -1218,7 +2061,8 @@ def STrib_indexed_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memb(Rs+Ru<<#u2)=Rt // if (Pv) memb(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1227,7 +2071,8 @@ def STrib_indexed_shl_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memb(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1236,7 +2081,8 @@ def STrib_indexed_shl_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memb(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1245,7 +2091,8 @@ def STrib_indexed_shl_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1256,7 +2103,8 @@ def STrib_indexed_shl_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memb(Rx++#s4:0)=Rt // if (Pv) memb(Rx++#s4:0)=Rt // if (Pv.new) memb(Rx++#s4:0)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if ($src1.new) memb($src3++#$offset) = $src2", @@ -1265,7 +2113,8 @@ def POST_STbri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), // if (!Pv) memb(Rx++#s4:0)=Rt // if (!Pv.new) memb(Rx++#s4:0)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if (!$src1.new) memb($src3++#$offset) = $src2", @@ -1274,20 +2123,15 @@ def POST_STbri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), // Store halfword. +// TODO: needs to be implemented // memh(Re=#U6)=Rt.H -// TODO: needs to be implemented - -// memh(Re=#U6)=Rt -// TODO: needs to be implemented - // memh(Rs+#s11:1)=Rt.H -// memh(Rs+#s11:1)=Rt // memh(Rs+#u6:1)=#S8 let AddedComplexity = 10, isPredicable = 1 in def STrih_imm_V4 : STInst<(outs), (ins IntRegs:$src1, u6_1Imm:$src2, s8Imm:$src3), "memh($src1+#$src2) = #$src3", - [(truncstorei16 s8ImmPred:$src3, (add IntRegs:$src1, + [(truncstorei16 s8ImmPred:$src3, (add (i32 IntRegs:$src1), u6_1ImmPred:$src2))]>, Requires<[HasV4T]>; @@ -1299,9 +2143,10 @@ let AddedComplexity = 10, isPredicable = 1 in def STrih_indexed_shl_V4 : STInst<(outs), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4), "memh($src1+$src2<<#$src3) = $src4", - [(truncstorei16 IntRegs:$src4, (add IntRegs:$src1, - (shl IntRegs:$src2, - u2ImmPred:$src3)))]>, + [(truncstorei16 (i32 IntRegs:$src4), + (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$src3)))]>, Requires<[HasV4T]>; // memh(Ru<<#u2+#U6)=Rt.H @@ -1310,9 +2155,9 @@ let AddedComplexity = 10 in def STrih_shl_V4 : STInst<(outs), (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4), "memh($src1<<#$src2+#$src3) = $src4", - [(truncstorei16 IntRegs:$src4, (shl IntRegs:$src1, - (add u2ImmPred:$src2, - u6ImmPred:$src3)))]>, + [(truncstorei16 (i32 IntRegs:$src4), + (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), + u6ImmPred:$src3))]>, Requires<[HasV4T]>; // memh(Rx++#s4:1:circ(Mu))=Rt.H @@ -1323,17 +2168,14 @@ def STrih_shl_V4 : STInst<(outs), // memh(Rx++Mu)=Rt // memh(Rx++Mu:brev)=Rt.H // memh(Rx++Mu:brev)=Rt -// memh(gp+#u16:1)=Rt.H // memh(gp+#u16:1)=Rt - - -// Store halfword conditionally. // if ([!]Pv[.new]) memh(#u6)=Rt.H // if ([!]Pv[.new]) memh(#u6)=Rt // if ([!]Pv[.new]) memh(Rs+#u6:1)=#S6 // if (Pv) memh(Rs+#u6:1)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_imm_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4), "if ($src1) memh($src2+#$src3) = #$src4", @@ -1341,7 +2183,8 @@ def STrih_imm_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+#u6:1)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_imm_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4), "if ($src1.new) memh($src2+#$src3) = #$src4", @@ -1349,7 +2192,8 @@ def STrih_imm_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memh(Rs+#u6:1)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_imm_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4), "if (!$src1) memh($src2+#$src3) = #$src4", @@ -1357,7 +2201,8 @@ def STrih_imm_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+#u6:1)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_imm_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4), "if (!$src1.new) memh($src2+#$src3) = #$src4", @@ -1370,7 +2215,8 @@ def STrih_imm_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt // if (Pv) memh(Rs+#u6:1)=Rt // if (Pv.new) memh(Rs+#u6:1)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memh($addr) = $src2", @@ -1379,7 +2225,8 @@ def STrih_cdnPt_V4 : STInst<(outs), // if (!Pv) memh(Rs+#u6:1)=Rt // if (!Pv.new) memh(Rs+#u6:1)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memh($addr) = $src2", @@ -1387,7 +2234,8 @@ def STrih_cdnNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+#u6:1)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if ($src1.new) memh($src2+#$src3) = $src4", @@ -1395,7 +2243,8 @@ def STrih_indexed_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+#u6:1)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if (!$src1.new) memh($src2+#$src3) = $src4", @@ -1405,7 +2254,8 @@ def STrih_indexed_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Rt.H // if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Rt // if (Pv) memh(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1414,6 +2264,8 @@ def STrih_indexed_shl_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+Ru<<#u2)=Rt +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1422,7 +2274,8 @@ def STrih_indexed_shl_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memh(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1431,7 +2284,8 @@ def STrih_indexed_shl_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1445,7 +2299,8 @@ def STrih_indexed_shl_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt // if (Pv) memh(Rx++#s4:1)=Rt // if (Pv.new) memh(Rx++#s4:1)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if ($src1.new) memh($src3++#$offset) = $src2", @@ -1454,7 +2309,8 @@ def POST_SThri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), // if (!Pv) memh(Rx++#s4:1)=Rt // if (!Pv.new) memh(Rx++#s4:1)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if (!$src1.new) memh($src3++#$offset) = $src2", @@ -1466,13 +2322,22 @@ def POST_SThri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), // memw(Re=#U6)=Rt // TODO: Needs to be implemented. -// memw(Rs+#s11:2)=Rt +// Store predicate: +let mayStore = 1, neverHasSideEffects = 1 in +def STriw_pred_V4 : STInst<(outs), + (ins MEMri:$addr, PredRegs:$src1), + "Error; should not emit", + []>, + Requires<[HasV4T]>; + + // memw(Rs+#u6:2)=#S8 let AddedComplexity = 10, isPredicable = 1 in def STriw_imm_V4 : STInst<(outs), (ins IntRegs:$src1, u6_2Imm:$src2, s8Imm:$src3), "memw($src1+#$src2) = #$src3", - [(store s8ImmPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2))]>, + [(store s8ImmPred:$src3, (add (i32 IntRegs:$src1), + u6_2ImmPred:$src2))]>, Requires<[HasV4T]>; // memw(Rs+Ru<<#u2)=Rt @@ -1480,8 +2345,9 @@ let AddedComplexity = 10, isPredicable = 1 in def STriw_indexed_shl_V4 : STInst<(outs), (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4), "memw($src1+$src2<<#$src3) = $src4", - [(store IntRegs:$src4, (add IntRegs:$src1, - (shl IntRegs:$src2, u2ImmPred:$src3)))]>, + [(store (i32 IntRegs:$src4), (add (i32 IntRegs:$src1), + (shl (i32 IntRegs:$src2), + u2ImmPred:$src3)))]>, Requires<[HasV4T]>; // memw(Ru<<#u2+#U6)=Rt @@ -1489,8 +2355,9 @@ let AddedComplexity = 10 in def STriw_shl_V4 : STInst<(outs), (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4), "memw($src1<<#$src2+#$src3) = $src4", - [(store IntRegs:$src4, (shl IntRegs:$src1, - (add u2ImmPred:$src2, u6ImmPred:$src3)))]>, + [(store (i32 IntRegs:$src4), + (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2), + u6ImmPred:$src3))]>, Requires<[HasV4T]>; // memw(Rx++#s4:2)=Rt @@ -1502,12 +2369,11 @@ def STriw_shl_V4 : STInst<(outs), // Store word conditionally. -// if ([!]Pv[.new]) memw(#u6)=Rt -// TODO: Needs to be implemented. // if ([!]Pv[.new]) memw(Rs+#u6:2)=#S6 // if (Pv) memw(Rs+#u6:2)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_imm_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4), "if ($src1) memw($src2+#$src3) = #$src4", @@ -1515,7 +2381,8 @@ def STriw_imm_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memw(Rs+#u6:2)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_imm_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4), "if ($src1.new) memw($src2+#$src3) = #$src4", @@ -1523,7 +2390,8 @@ def STriw_imm_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memw(Rs+#u6:2)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_imm_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4), "if (!$src1) memw($src2+#$src3) = #$src4", @@ -1531,7 +2399,8 @@ def STriw_imm_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+#u6:2)=#S6 -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_imm_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4), "if (!$src1.new) memw($src2+#$src3) = #$src4", @@ -1541,7 +2410,8 @@ def STriw_imm_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memw(Rs+#u6:2)=Rt // if (Pv) memw(Rs+#u6:2)=Rt // if (Pv.new) memw(Rs+#u6:2)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memw($addr) = $src2", @@ -1550,7 +2420,8 @@ def STriw_cdnPt_V4 : STInst<(outs), // if (!Pv) memw(Rs+#u6:2)=Rt // if (!Pv.new) memw(Rs+#u6:2)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memw($addr) = $src2", @@ -1560,7 +2431,8 @@ def STriw_cdnNotPt_V4 : STInst<(outs), // if (Pv) memw(Rs+#u6:2)=Rt // if (!Pv) memw(Rs+#u6:2)=Rt // if (Pv.new) memw(Rs+#u6:2)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if ($src1.new) memw($src2+#$src3) = $src4", @@ -1568,7 +2440,8 @@ def STriw_indexed_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+#u6:2)=Rt -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if (!$src1.new) memw($src2+#$src3) = $src4", @@ -1577,7 +2450,8 @@ def STriw_indexed_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memw(Rs+Ru<<#u2)=Rt // if (Pv) memw(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1586,7 +2460,8 @@ def STriw_indexed_shl_cPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (Pv.new) memw(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cdnPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1595,7 +2470,8 @@ def STriw_indexed_shl_cdnPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv) memw(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1604,7 +2480,8 @@ def STriw_indexed_shl_cNotPt_V4 : STInst<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+Ru<<#u2)=Rt -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cdnNotPt_V4 : STInst<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1615,7 +2492,8 @@ def STriw_indexed_shl_cdnNotPt_V4 : STInst<(outs), // if ([!]Pv[.new]) memw(Rx++#s4:2)=Rt // if (Pv) memw(Rx++#s4:2)=Rt // if (Pv.new) memw(Rx++#s4:2)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if ($src1.new) memw($src3++#$offset) = $src2", @@ -1624,7 +2502,8 @@ def POST_STwri_cdnPt_V4 : STInstPI<(outs IntRegs:$dst), // if (!Pv) memw(Rx++#s4:2)=Rt // if (!Pv.new) memw(Rx++#s4:2)=Rt -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if (!$src1.new) memw($src3++#$offset) = $src2", @@ -1632,6 +2511,439 @@ def POST_STwri_cdnNotPt_V4 : STInstPI<(outs IntRegs:$dst), Requires<[HasV4T]>; +/// store to global address + +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STrid_GP_V4 : STInst<(outs), + (ins globaladdress:$global, u16Imm:$offset, DoubleRegs:$src), + "memd(#$global+$offset) = $src", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrid_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + DoubleRegs:$src2), + "if ($src1) memd(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrid_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + DoubleRegs:$src2), + "if (!$src1) memd(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrid_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + DoubleRegs:$src2), + "if ($src1.new) memd(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrid_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + DoubleRegs:$src2), + "if (!$src1.new) memd(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STrib_GP_V4 : STInst<(outs), + (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src), + "memb(#$global+$offset) = $src", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrib_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memb(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrib_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memb(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrib_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memb(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrib_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memb(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STrih_GP_V4 : STInst<(outs), + (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src), + "memh(#$global+$offset) = $src", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrih_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memh(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrih_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memh(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrih_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memh(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STrih_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memh(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STriw_GP_V4 : STInst<(outs), + (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src), + "memw(#$global+$offset) = $src", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STriw_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memw(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STriw_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memw(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STriw_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memw(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STriw_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memw(##$global+$offset) = $src2", + []>, + Requires<[HasV4T]>; + +// memd(#global)=Rtt +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STd_GP_V4 : STInst<(outs), + (ins globaladdress:$global, DoubleRegs:$src), + "memd(#$global) = $src", + []>, + Requires<[HasV4T]>; + +// if (Pv) memd(##global) = Rtt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STd_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2), + "if ($src1) memd(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memd(##global) = Rtt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STd_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2), + "if (!$src1) memd(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (Pv) memd(##global) = Rtt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STd_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2), + "if ($src1.new) memd(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memd(##global) = Rtt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STd_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2), + "if (!$src1.new) memd(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// memb(#global)=Rt +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_V4 : STInst<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memb(#$global) = $src", + []>, + Requires<[HasV4T]>; + +// if (Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STb_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memb(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STb_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memb(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STb_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memb(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STb_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memb(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// memh(#global)=Rt +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_V4 : STInst<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memh(#$global) = $src", + []>, + Requires<[HasV4T]>; + +// if (Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STh_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memh(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STh_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memh(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STh_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memh(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STh_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memh(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// memw(#global)=Rt +let isPredicable = 1, mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_V4 : STInst<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memw(#$global) = $src", + []>, + Requires<[HasV4T]>; + +// if (Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STw_GP_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memw(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STw_GP_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memw(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STw_GP_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memw(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1, isPredicated = 1 in +def STw_GP_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memw(##$global) = $src2", + []>, + Requires<[HasV4T]>; + +// 64 bit atomic store +def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global), + (i64 DoubleRegs:$src1)), + (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress) -> memd(#foo) +let AddedComplexity = 100 in +def : Pat <(store (i64 DoubleRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)), + (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>, + Requires<[HasV4T]>; + +// 8 bit atomic store +def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global), + (i32 IntRegs:$src1)), + (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress) -> memb(#foo) +let AddedComplexity = 100 in +def : Pat<(truncstorei8 (i32 IntRegs:$src1), + (HexagonCONST32_GP tglobaladdr:$global)), + (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1" +// to "r0 = 1; memw(#foo) = r0" +let AddedComplexity = 100 in +def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)), + (STb_GP_V4 tglobaladdr:$global, (TFRI 1))>, + Requires<[HasV4T]>; + +def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global), + (i32 IntRegs:$src1)), + (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress) -> memh(#foo) +let AddedComplexity = 100 in +def : Pat<(truncstorei16 (i32 IntRegs:$src1), + (HexagonCONST32_GP tglobaladdr:$global)), + (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// 32 bit atomic store +def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global), + (i32 IntRegs:$src1)), + (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress) -> memw(#foo) +let AddedComplexity = 100 in +def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)), + (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +def : Pat<(atomic_store_64 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset), + (i64 DoubleRegs:$src1)), + (STrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, + (i64 DoubleRegs:$src1))>, + Requires<[HasV4T]>; + +def : Pat<(atomic_store_32 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset), + (i32 IntRegs:$src1)), + (STriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +def : Pat<(atomic_store_16 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset), + (i32 IntRegs:$src1)), + (STrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +def : Pat<(atomic_store_8 (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset), + (i32 IntRegs:$src1)), + (STrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress + x) -> memd(#foo + x) +let AddedComplexity = 100 in +def : Pat<(store (i64 DoubleRegs:$src1), (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (STrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, + (i64 DoubleRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress + x) -> memb(#foo + x) +let AddedComplexity = 100 in +def : Pat<(truncstorei8 (i32 IntRegs:$src1), + (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (STrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress + x) -> memh(#foo + x) +let AddedComplexity = 100 in +def : Pat<(truncstorei16 (i32 IntRegs:$src1), + (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (STrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + +// Map from store(globaladdress + x) -> memw(#foo + x) +let AddedComplexity = 100 in +def : Pat<(store (i32 IntRegs:$src1), + (add (HexagonCONST32_GP tglobaladdr:$global), + u16ImmPred:$offset)), + (STriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>, + Requires<[HasV4T]>; + + + //===----------------------------------------------------------------------=== // ST - //===----------------------------------------------------------------------=== @@ -1696,11 +3008,19 @@ def STrib_GP_nv_V4 : NVInst_V4<(outs), []>, Requires<[HasV4T]>; +// memb(#global)=Nt.new +let mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_nv_V4 : NVInst_V4<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memb(#$global) = $src.new", + []>, + Requires<[HasV4T]>; // Store new-value byte conditionally. // if ([!]Pv[.new]) memb(#u6)=Nt.new // if (Pv) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1) memb($addr) = $src2.new", @@ -1708,7 +3028,8 @@ def STrib_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memb($addr) = $src2.new", @@ -1716,7 +3037,8 @@ def STrib_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1) memb($addr) = $src2.new", @@ -1724,7 +3046,8 @@ def STrib_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memb($addr) = $src2.new", @@ -1732,7 +3055,8 @@ def STrib_cdnNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if ($src1) memb($src2+#$src3) = $src4.new", @@ -1740,7 +3064,8 @@ def STrib_indexed_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if ($src1.new) memb($src2+#$src3) = $src4.new", @@ -1748,7 +3073,8 @@ def STrib_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if (!$src1) memb($src2+#$src3) = $src4.new", @@ -1756,7 +3082,8 @@ def STrib_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+#u6:0)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrib_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4), "if (!$src1.new) memb($src2+#$src3) = $src4.new", @@ -1766,7 +3093,8 @@ def STrib_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memb(Rs+Ru<<#u2)=Nt.new // if (Pv) memb(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1775,7 +3103,8 @@ def STrib_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memb(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1784,7 +3113,8 @@ def STrib_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memb(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1793,7 +3123,8 @@ def STrib_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memb(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrib_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1803,7 +3134,8 @@ def STrib_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memb(Rx++#s4:0)=Nt.new // if (Pv) memb(Rx++#s4:0)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if ($src1) memb($src3++#$offset) = $src2.new", @@ -1811,7 +3143,8 @@ def POST_STbri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (Pv.new) memb(Rx++#s4:0)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if ($src1.new) memb($src3++#$offset) = $src2.new", @@ -1819,7 +3152,8 @@ def POST_STbri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv) memb(Rx++#s4:0)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if (!$src1) memb($src3++#$offset) = $src2.new", @@ -1827,7 +3161,8 @@ def POST_STbri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv.new) memb(Rx++#s4:0)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STbri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset), "if (!$src1.new) memb($src3++#$offset) = $src2.new", @@ -1889,6 +3224,14 @@ def STrih_GP_nv_V4 : NVInst_V4<(outs), []>, Requires<[HasV4T]>; +// memh(#global)=Nt.new +let mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_nv_V4 : NVInst_V4<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memh(#$global) = $src.new", + []>, + Requires<[HasV4T]>; + // Store new-value halfword conditionally. @@ -1896,7 +3239,8 @@ def STrih_GP_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memh(Rs+#u6:1)=Nt.new // if (Pv) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1) memh($addr) = $src2.new", @@ -1904,7 +3248,8 @@ def STrih_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memh($addr) = $src2.new", @@ -1912,7 +3257,8 @@ def STrih_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1) memh($addr) = $src2.new", @@ -1920,7 +3266,8 @@ def STrih_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memh($addr) = $src2.new", @@ -1928,7 +3275,8 @@ def STrih_cdnNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if ($src1) memh($src2+#$src3) = $src4.new", @@ -1936,7 +3284,8 @@ def STrih_indexed_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if ($src1.new) memh($src2+#$src3) = $src4.new", @@ -1944,7 +3293,8 @@ def STrih_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if (!$src1) memh($src2+#$src3) = $src4.new", @@ -1952,7 +3302,8 @@ def STrih_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+#u6:1)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STrih_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4), "if (!$src1.new) memh($src2+#$src3) = $src4.new", @@ -1961,7 +3312,8 @@ def STrih_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Nt.new // if (Pv) memh(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1970,7 +3322,8 @@ def STrih_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memh(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1979,7 +3332,8 @@ def STrih_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memh(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1988,7 +3342,8 @@ def STrih_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memh(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STrih_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -1998,7 +3353,8 @@ def STrih_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[]) memh(Rx++#s4:1)=Nt.new // if (Pv) memh(Rx++#s4:1)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if ($src1) memh($src3++#$offset) = $src2.new", @@ -2006,7 +3362,8 @@ def POST_SThri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (Pv.new) memh(Rx++#s4:1)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if ($src1.new) memh($src3++#$offset) = $src2.new", @@ -2014,7 +3371,8 @@ def POST_SThri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv) memh(Rx++#s4:1)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if (!$src1) memh($src3++#$offset) = $src2.new", @@ -2022,7 +3380,8 @@ def POST_SThri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv.new) memh(Rx++#s4:1)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_SThri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset), "if (!$src1.new) memh($src3++#$offset) = $src2.new", @@ -2085,6 +3444,12 @@ def STriw_GP_nv_V4 : NVInst_V4<(outs), []>, Requires<[HasV4T]>; +let mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_nv_V4 : NVInst_V4<(outs), + (ins globaladdress:$global, IntRegs:$src), + "memw(#$global) = $src.new", + []>, + Requires<[HasV4T]>; // Store new-value word conditionally. @@ -2092,7 +3457,8 @@ def STriw_GP_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memw(Rs+#u6:2)=Nt.new // if (Pv) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1) memw($addr) = $src2.new", @@ -2100,7 +3466,8 @@ def STriw_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if ($src1.new) memw($addr) = $src2.new", @@ -2108,7 +3475,8 @@ def STriw_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1) memw($addr) = $src2.new", @@ -2116,7 +3484,8 @@ def STriw_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2), "if (!$src1.new) memw($addr) = $src2.new", @@ -2124,7 +3493,8 @@ def STriw_cdnNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if ($src1) memw($src2+#$src3) = $src4.new", @@ -2132,7 +3502,8 @@ def STriw_indexed_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if ($src1.new) memw($src2+#$src3) = $src4.new", @@ -2140,7 +3511,8 @@ def STriw_indexed_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if (!$src1) memw($src2+#$src3) = $src4.new", @@ -2148,7 +3520,8 @@ def STriw_indexed_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+#u6:2)=Nt.new -let mayStore = 1, neverHasSideEffects = 1 in +let mayStore = 1, neverHasSideEffects = 1, + isPredicated = 1 in def STriw_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4), "if (!$src1.new) memw($src2+#$src3) = $src4.new", @@ -2158,7 +3531,8 @@ def STriw_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memw(Rs+Ru<<#u2)=Nt.new // if (Pv) memw(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -2167,7 +3541,8 @@ def STriw_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (Pv.new) memw(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -2176,7 +3551,8 @@ def STriw_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv) memw(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -2185,7 +3561,8 @@ def STriw_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs), Requires<[HasV4T]>; // if (!Pv.new) memw(Rs+Ru<<#u2)=Nt.new -let mayStore = 1, AddedComplexity = 10 in +let mayStore = 1, AddedComplexity = 10, + isPredicated = 1 in def STriw_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4, IntRegs:$src5), @@ -2195,7 +3572,8 @@ def STriw_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs), // if ([!]Pv[.new]) memw(Rx++#s4:2)=Nt.new // if (Pv) memw(Rx++#s4:2)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if ($src1) memw($src3++#$offset) = $src2.new", @@ -2203,7 +3581,8 @@ def POST_STwri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (Pv.new) memw(Rx++#s4:2)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if ($src1.new) memw($src3++#$offset) = $src2.new", @@ -2211,7 +3590,8 @@ def POST_STwri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv) memw(Rx++#s4:2)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if (!$src1) memw($src3++#$offset) = $src2.new", @@ -2219,7 +3599,8 @@ def POST_STwri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; // if (!Pv.new) memw(Rx++#s4:2)=Nt.new -let mayStore = 1, hasCtrlDep = 1 in +let mayStore = 1, hasCtrlDep = 1, + isPredicated = 1 in def POST_STwri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset), "if (!$src1.new) memw($src3++#$offset) = $src2.new", @@ -2227,6 +3608,199 @@ def POST_STwri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst), Requires<[HasV4T]>; + +// if (Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memb(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memb(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memb(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memb(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STb_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memb(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memh(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memh(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memh(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memh(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STh_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memh(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1) memw(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1) memw(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if ($src1.new) memw(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +// if (!Pv) memw(##global) = Rt +let mayStore = 1, neverHasSideEffects = 1 in +def STw_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2), + "if (!$src1.new) memw(##$global) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrib_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memb(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrib_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memb(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrib_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memb(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrib_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memb(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrih_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memh(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrih_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memh(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrih_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memh(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STrih_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memh(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STriw_GP_cPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1) memw(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STriw_GP_cNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1) memw(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STriw_GP_cdnPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if ($src1.new) memw(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + +let mayStore = 1, neverHasSideEffects = 1 in +def STriw_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs), + (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset, + IntRegs:$src2), + "if (!$src1.new) memw(##$global+$offset) = $src2.new", + []>, + Requires<[HasV4T]>; + //===----------------------------------------------------------------------===// // NV/ST - //===----------------------------------------------------------------------===// @@ -2416,16 +3990,18 @@ let isBranch = 1, isTerminator=1, neverHasSideEffects = 1, Defs = [PC] in { def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, s6Imm:$src3), "$dst = add($src1, add($src2, #$src3))", - [(set IntRegs:$dst, - (add IntRegs:$src1, (add IntRegs:$src2, s6ImmPred:$src3)))]>, + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2), + s6ImmPred:$src3)))]>, Requires<[HasV4T]>; // Rd=add(Rs,sub(#s6,Ru)) def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s6Imm:$src2, IntRegs:$src3), "$dst = add($src1, sub(#$src2, $src3))", - [(set IntRegs:$dst, - (add IntRegs:$src1, (sub s6ImmPred:$src2, IntRegs:$src3)))]>, + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (sub s6ImmPred:$src2, + (i32 IntRegs:$src3))))]>, Requires<[HasV4T]>; // Generates the same instruction as ADDr_SUBri_V4 but matches different @@ -2434,8 +4010,9 @@ def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst), def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s6Imm:$src2, IntRegs:$src3), "$dst = add($src1, sub(#$src2, $src3))", - [(set IntRegs:$dst, - (sub (add IntRegs:$src1, s6ImmPred:$src2), IntRegs:$src3))]>, + [(set (i32 IntRegs:$dst), + (sub (add (i32 IntRegs:$src1), s6ImmPred:$src2), + (i32 IntRegs:$src3)))]>, Requires<[HasV4T]>; @@ -2451,16 +4028,16 @@ def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst), def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2), "$dst = and($src1, ~$src2)", - [(set DoubleRegs:$dst, (and DoubleRegs:$src1, - (not DoubleRegs:$src2)))]>, + [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1), + (not (i64 DoubleRegs:$src2))))]>, Requires<[HasV4T]>; // Rdd=or(Rtt,~Rss) def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2), "$dst = or($src1, ~$src2)", - [(set DoubleRegs:$dst, - (or DoubleRegs:$src1, (not DoubleRegs:$src2)))]>, + [(set (i64 DoubleRegs:$dst), + (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>, Requires<[HasV4T]>; @@ -2469,8 +4046,9 @@ def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst), def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3), "$dst ^= xor($src2, $src3)", - [(set DoubleRegs:$dst, - (xor DoubleRegs:$src1, (xor DoubleRegs:$src2, DoubleRegs:$src3)))], + [(set (i64 DoubleRegs:$dst), + (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2), + (i64 DoubleRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2480,8 +4058,9 @@ def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst), def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3), "$dst = or($src1, and($src2, #$src3))", - [(set IntRegs:$dst, - (or IntRegs:$src1, (and IntRegs:$src2, s10ImmPred:$src3)))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + s10ImmPred:$src3)))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2490,8 +4069,9 @@ def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst), def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst &= and($src2, $src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (and IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2499,8 +4079,9 @@ def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst |= and($src2, $src3)", - [(set IntRegs:$dst, - (or IntRegs:$src1, (and IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2508,8 +4089,9 @@ def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst ^= and($src2, $src3)", - [(set IntRegs:$dst, - (xor IntRegs:$src1, (and IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2518,8 +4100,9 @@ def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst), def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst &= and($src2, ~$src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (and IntRegs:$src2, (not IntRegs:$src3))))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (not (i32 IntRegs:$src3)))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2527,8 +4110,9 @@ def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst |= and($src2, ~$src3)", - [(set IntRegs:$dst, - (or IntRegs:$src1, (and IntRegs:$src2, (not IntRegs:$src3))))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (not (i32 IntRegs:$src3)))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2536,8 +4120,9 @@ def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst ^= and($src2, ~$src3)", - [(set IntRegs:$dst, - (xor IntRegs:$src1, (and IntRegs:$src2, (not IntRegs:$src3))))], + [(set (i32 IntRegs:$dst), + (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + (not (i32 IntRegs:$src3)))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2546,8 +4131,9 @@ def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst), def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst &= or($src2, $src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (or IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2555,8 +4141,9 @@ def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst |= or($src2, $src3)", - [(set IntRegs:$dst, - (or IntRegs:$src1, (or IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2564,8 +4151,9 @@ def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst ^= or($src2, $src3)", - [(set IntRegs:$dst, - (xor IntRegs:$src1, (or IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2574,8 +4162,9 @@ def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst), def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst &= xor($src2, $src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (xor IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2583,8 +4172,9 @@ def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst |= xor($src2, $src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (xor IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2592,8 +4182,9 @@ def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3), "$dst ^= xor($src2, $src3)", - [(set IntRegs:$dst, - (and IntRegs:$src1, (xor IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2601,8 +4192,9 @@ def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3), "$dst |= and($src2, #$src3)", - [(set IntRegs:$dst, - (or IntRegs:$src1, (and IntRegs:$src2, s10ImmPred:$src3)))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + s10ImmPred:$src3)))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2610,8 +4202,9 @@ def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst), def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3), "$dst |= or($src2, #$src3)", - [(set IntRegs:$dst, - (or IntRegs:$src1, (and IntRegs:$src2, s10ImmPred:$src3)))], + [(set (i32 IntRegs:$dst), + (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2), + s10ImmPred:$src3)))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2663,8 +4256,9 @@ def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst), def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst), (ins u6Imm:$src1, IntRegs:$src2, u6Imm:$src3), "$dst = add(#$src1, mpyi($src2, #$src3))", - [(set IntRegs:$dst, - (add (mul IntRegs:$src2, u6ImmPred:$src3), u6ImmPred:$src1))]>, + [(set (i32 IntRegs:$dst), + (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3), + u6ImmPred:$src1))]>, Requires<[HasV4T]>; // Rd=add(#u6,mpyi(Rs,Rt)) @@ -2672,32 +4266,36 @@ def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst), def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst), (ins u6Imm:$src1, IntRegs:$src2, IntRegs:$src3), "$dst = add(#$src1, mpyi($src2, $src3))", - [(set IntRegs:$dst, - (add (mul IntRegs:$src2, IntRegs:$src3), u6ImmPred:$src1))]>, + [(set (i32 IntRegs:$dst), + (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)), + u6ImmPred:$src1))]>, Requires<[HasV4T]>; // Rd=add(Ru,mpyi(#u6:2,Rs)) def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3), "$dst = add($src1, mpyi(#$src2, $src3))", - [(set IntRegs:$dst, - (add IntRegs:$src1, (mul IntRegs:$src3, u6_2ImmPred:$src2)))]>, + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3), + u6_2ImmPred:$src2)))]>, Requires<[HasV4T]>; // Rd=add(Ru,mpyi(Rs,#u6)) def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, u6Imm:$src3), "$dst = add($src1, mpyi($src2, #$src3))", - [(set IntRegs:$dst, - (add IntRegs:$src1, (mul IntRegs:$src2, u6ImmPred:$src3)))]>, + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2), + u6ImmPred:$src3)))]>, Requires<[HasV4T]>; // Rx=add(Ru,mpyi(Rx,Rs)) def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3), "$dst = add($src1, mpyi($src2, $src3))", - [(set IntRegs:$dst, - (add IntRegs:$src1, (mul IntRegs:$src2, IntRegs:$src3)))], + [(set (i32 IntRegs:$dst), + (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2), + (i32 IntRegs:$src3))))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2745,8 +4343,9 @@ def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst), def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = add(#$src1, asl($src2, #$src3))", - [(set IntRegs:$dst, - (add (shl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2754,8 +4353,9 @@ def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = add(#$src1, lsr($src2, #$src3))", - [(set IntRegs:$dst, - (add (srl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2763,8 +4363,9 @@ def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = sub(#$src1, asl($src2, #$src3))", - [(set IntRegs:$dst, - (sub (shl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2772,8 +4373,9 @@ def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = sub(#$src1, lsr($src2, #$src3))", - [(set IntRegs:$dst, - (sub (srl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2783,8 +4385,9 @@ def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = and(#$src1, asl($src2, #$src3))", - [(set IntRegs:$dst, - (and (shl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2792,26 +4395,31 @@ def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = and(#$src1, lsr($src2, #$src3))", - [(set IntRegs:$dst, - (and (srl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; //Rx=or(#u8,asl(Rx,#U5)) +let AddedComplexity = 30 in def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = or(#$src1, asl($src2, #$src3))", - [(set IntRegs:$dst, - (or (shl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; //Rx=or(#u8,lsr(Rx,#U5)) +let AddedComplexity = 30 in def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3), "$dst = or(#$src1, lsr($src2, #$src3))", - [(set IntRegs:$dst, - (or (srl IntRegs:$src2, u5ImmPred:$src3), u8ImmPred:$src1))], + [(set (i32 IntRegs:$dst), + (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3), + u8ImmPred:$src1))], "$src2 = $dst">, Requires<[HasV4T]>; @@ -2820,7 +4428,8 @@ def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst), //Rd=lsl(#s6,Rt) def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2), "$dst = lsl(#$src1, $src2)", - [(set IntRegs:$dst, (shl s6ImmPred:$src1, IntRegs:$src2))]>, + [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1, + (i32 IntRegs:$src2)))]>, Requires<[HasV4T]>; @@ -2829,8 +4438,9 @@ def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2), def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), "$dst ^= asl($src2, $src3)", - [(set DoubleRegs:$dst, - (xor DoubleRegs:$src1, (shl DoubleRegs:$src2, IntRegs:$src3)))], + [(set (i64 DoubleRegs:$dst), + (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2838,8 +4448,9 @@ def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), "$dst ^= asr($src2, $src3)", - [(set DoubleRegs:$dst, - (xor DoubleRegs:$src1, (sra DoubleRegs:$src2, IntRegs:$src3)))], + [(set (i64 DoubleRegs:$dst), + (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2847,8 +4458,9 @@ def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), "$dst ^= lsl($src2, $src3)", - [(set DoubleRegs:$dst, - (xor DoubleRegs:$src1, (shl DoubleRegs:$src2, IntRegs:$src3)))], + [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1), + (shl (i64 DoubleRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2856,8 +4468,9 @@ def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3), "$dst ^= lsr($src2, $src3)", - [(set DoubleRegs:$dst, - (xor DoubleRegs:$src1, (srl DoubleRegs:$src2, IntRegs:$src3)))], + [(set (i64 DoubleRegs:$dst), + (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2), + (i32 IntRegs:$src3))))], "$src1 = $dst">, Requires<[HasV4T]>; @@ -2903,16 +4516,16 @@ let AddedComplexity = 30 in def MEMw_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, m6Imm:$addend), "Error; should not emit", - [(store (add (load (add IntRegs:$base, u6_2ImmPred:$offset)), -m6ImmPred:$addend), - (add IntRegs:$base, u6_2ImmPred:$offset))]>, + [(store (add (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)), + m6ImmPred:$addend), + (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memw(Rs+#u6:2) += #U5 let AddedComplexity = 30 in def MEMw_ADDi_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, u5Imm:$addend), - "memw($base+#$offset) += $addend", + "memw($base+#$offset) += #$addend", []>, Requires<[HasV4T, UseMEMOP]>; @@ -2920,7 +4533,7 @@ def MEMw_ADDi_indexed_MEM_V4 : MEMInst_V4<(outs), let AddedComplexity = 30 in def MEMw_SUBi_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, u5Imm:$subend), - "memw($base+#$offset) -= $subend", + "memw($base+#$offset) -= #$subend", []>, Requires<[HasV4T, UseMEMOP]>; @@ -2929,9 +4542,9 @@ let AddedComplexity = 30 in def MEMw_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$addend), "memw($base+#$offset) += $addend", - [(store (add (load (add IntRegs:$base, u6_2ImmPred:$offset)), -IntRegs:$addend), - (add IntRegs:$base, u6_2ImmPred:$offset))]>, + [(store (add (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)), + (i32 IntRegs:$addend)), + (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memw(Rs+#u6:2) -= Rt @@ -2939,19 +4552,19 @@ let AddedComplexity = 30 in def MEMw_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$subend), "memw($base+#$offset) -= $subend", - [(store (sub (load (add IntRegs:$base, u6_2ImmPred:$offset)), -IntRegs:$subend), - (add IntRegs:$base, u6_2ImmPred:$offset))]>, + [(store (sub (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)), + (i32 IntRegs:$subend)), + (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memw(Rs+#u6:2) &= Rt let AddedComplexity = 30 in def MEMw_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$andend), - "memw($base+#$offset) += $andend", - [(store (and (load (add IntRegs:$base, u6_2ImmPred:$offset)), -IntRegs:$andend), - (add IntRegs:$base, u6_2ImmPred:$offset))]>, + "memw($base+#$offset) &= $andend", + [(store (and (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)), + (i32 IntRegs:$andend)), + (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memw(Rs+#u6:2) |= Rt @@ -2959,9 +4572,9 @@ let AddedComplexity = 30 in def MEMw_ORr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$orend), "memw($base+#$offset) |= $orend", - [(store (or (load (add IntRegs:$base, u6_2ImmPred:$offset)), - IntRegs:$orend), - (add IntRegs:$base, u6_2ImmPred:$offset))]>, + [(store (or (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)), + (i32 IntRegs:$orend)), + (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // MEMw_ADDSUBi_V4: @@ -2996,7 +4609,7 @@ let AddedComplexity = 30 in def MEMw_ADDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$addend), "memw($addr) += $addend", - [(store (add (load ADDRriU6_2:$addr), IntRegs:$addend), + [(store (add (load ADDRriU6_2:$addr), (i32 IntRegs:$addend)), ADDRriU6_2:$addr)]>, Requires<[HasV4T, UseMEMOP]>; @@ -3005,7 +4618,7 @@ let AddedComplexity = 30 in def MEMw_SUBr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$subend), "memw($addr) -= $subend", - [(store (sub (load ADDRriU6_2:$addr), IntRegs:$subend), + [(store (sub (load ADDRriU6_2:$addr), (i32 IntRegs:$subend)), ADDRriU6_2:$addr)]>, Requires<[HasV4T, UseMEMOP]>; @@ -3014,7 +4627,7 @@ let AddedComplexity = 30 in def MEMw_ANDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$andend), "memw($addr) &= $andend", - [(store (and (load ADDRriU6_2:$addr), IntRegs:$andend), + [(store (and (load ADDRriU6_2:$addr), (i32 IntRegs:$andend)), ADDRriU6_2:$addr)]>, Requires<[HasV4T, UseMEMOP]>; @@ -3023,8 +4636,8 @@ let AddedComplexity = 30 in def MEMw_ORr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$orend), "memw($addr) |= $orend", - [(store (or (load ADDRriU6_2:$addr), IntRegs:$orend), -ADDRriU6_2:$addr)]>, + [(store (or (load ADDRriU6_2:$addr), (i32 IntRegs:$orend)), + ADDRriU6_2:$addr)]>, Requires<[HasV4T, UseMEMOP]>; //===----------------------------------------------------------------------===// @@ -3060,10 +4673,10 @@ let AddedComplexity = 30 in def MEMh_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_1Imm:$offset, m6Imm:$addend), "Error; should not emit", - [(truncstorei16 (add (sextloadi16 (add IntRegs:$base, + [(truncstorei16 (add (sextloadi16 (add (i32 IntRegs:$base), u6_1ImmPred:$offset)), m6ImmPred:$addend), - (add IntRegs:$base, u6_1ImmPred:$offset))]>, + (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) += #U5 @@ -3087,10 +4700,10 @@ let AddedComplexity = 30 in def MEMh_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$addend), "memh($base+#$offset) += $addend", - [(truncstorei16 (add (sextloadi16 (add IntRegs:$base, + [(truncstorei16 (add (sextloadi16 (add (i32 IntRegs:$base), u6_1ImmPred:$offset)), - IntRegs:$addend), - (add IntRegs:$base, u6_1ImmPred:$offset))]>, + (i32 IntRegs:$addend)), + (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) -= Rt @@ -3098,10 +4711,10 @@ let AddedComplexity = 30 in def MEMh_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$subend), "memh($base+#$offset) -= $subend", - [(truncstorei16 (sub (sextloadi16 (add IntRegs:$base, + [(truncstorei16 (sub (sextloadi16 (add (i32 IntRegs:$base), u6_1ImmPred:$offset)), - IntRegs:$subend), - (add IntRegs:$base, u6_1ImmPred:$offset))]>, + (i32 IntRegs:$subend)), + (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) &= Rt @@ -3109,10 +4722,10 @@ let AddedComplexity = 30 in def MEMh_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$andend), "memh($base+#$offset) += $andend", - [(truncstorei16 (and (sextloadi16 (add IntRegs:$base, + [(truncstorei16 (and (sextloadi16 (add (i32 IntRegs:$base), u6_1ImmPred:$offset)), - IntRegs:$andend), - (add IntRegs:$base, u6_1ImmPred:$offset))]>, + (i32 IntRegs:$andend)), + (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) |= Rt @@ -3120,10 +4733,10 @@ let AddedComplexity = 30 in def MEMh_ORr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$orend), "memh($base+#$offset) |= $orend", - [(truncstorei16 (or (sextloadi16 (add IntRegs:$base, + [(truncstorei16 (or (sextloadi16 (add (i32 IntRegs:$base), u6_1ImmPred:$offset)), - IntRegs:$orend), - (add IntRegs:$base, u6_1ImmPred:$offset))]>, + (i32 IntRegs:$orend)), + (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // MEMh_ADDSUBi_V4: @@ -3159,7 +4772,7 @@ def MEMh_ADDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$addend), "memh($addr) += $addend", [(truncstorei16 (add (sextloadi16 ADDRriU6_1:$addr), - IntRegs:$addend), ADDRriU6_1:$addr)]>, + (i32 IntRegs:$addend)), ADDRriU6_1:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) -= Rt @@ -3168,7 +4781,7 @@ def MEMh_SUBr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$subend), "memh($addr) -= $subend", [(truncstorei16 (sub (sextloadi16 ADDRriU6_1:$addr), - IntRegs:$subend), ADDRriU6_1:$addr)]>, + (i32 IntRegs:$subend)), ADDRriU6_1:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) &= Rt @@ -3177,7 +4790,7 @@ def MEMh_ANDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$andend), "memh($addr) &= $andend", [(truncstorei16 (and (sextloadi16 ADDRriU6_1:$addr), - IntRegs:$andend), ADDRriU6_1:$addr)]>, + (i32 IntRegs:$andend)), ADDRriU6_1:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memh(Rs+#u6:1) |= Rt @@ -3186,7 +4799,7 @@ def MEMh_ORr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$orend), "memh($addr) |= $orend", [(truncstorei16 (or (sextloadi16 ADDRriU6_1:$addr), - IntRegs:$orend), ADDRriU6_1:$addr)]>, + (i32 IntRegs:$orend)), ADDRriU6_1:$addr)]>, Requires<[HasV4T, UseMEMOP]>; @@ -3223,10 +4836,10 @@ let AddedComplexity = 30 in def MEMb_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_0Imm:$offset, m6Imm:$addend), "Error; should not emit", - [(truncstorei8 (add (sextloadi8 (add IntRegs:$base, + [(truncstorei8 (add (sextloadi8 (add (i32 IntRegs:$base), u6_0ImmPred:$offset)), m6ImmPred:$addend), - (add IntRegs:$base, u6_0ImmPred:$offset))]>, + (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) += #U5 @@ -3250,10 +4863,10 @@ let AddedComplexity = 30 in def MEMb_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$addend), "memb($base+#$offset) += $addend", - [(truncstorei8 (add (sextloadi8 (add IntRegs:$base, + [(truncstorei8 (add (sextloadi8 (add (i32 IntRegs:$base), u6_0ImmPred:$offset)), - IntRegs:$addend), - (add IntRegs:$base, u6_0ImmPred:$offset))]>, + (i32 IntRegs:$addend)), + (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) -= Rt @@ -3261,10 +4874,10 @@ let AddedComplexity = 30 in def MEMb_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$subend), "memb($base+#$offset) -= $subend", - [(truncstorei8 (sub (sextloadi8 (add IntRegs:$base, + [(truncstorei8 (sub (sextloadi8 (add (i32 IntRegs:$base), u6_0ImmPred:$offset)), - IntRegs:$subend), - (add IntRegs:$base, u6_0ImmPred:$offset))]>, + (i32 IntRegs:$subend)), + (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) &= Rt @@ -3272,10 +4885,10 @@ let AddedComplexity = 30 in def MEMb_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$andend), "memb($base+#$offset) += $andend", - [(truncstorei8 (and (sextloadi8 (add IntRegs:$base, + [(truncstorei8 (and (sextloadi8 (add (i32 IntRegs:$base), u6_0ImmPred:$offset)), - IntRegs:$andend), - (add IntRegs:$base, u6_0ImmPred:$offset))]>, + (i32 IntRegs:$andend)), + (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) |= Rt @@ -3283,10 +4896,10 @@ let AddedComplexity = 30 in def MEMb_ORr_indexed_MEM_V4 : MEMInst_V4<(outs), (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$orend), "memb($base+#$offset) |= $orend", - [(truncstorei8 (or (sextloadi8 (add IntRegs:$base, + [(truncstorei8 (or (sextloadi8 (add (i32 IntRegs:$base), u6_0ImmPred:$offset)), - IntRegs:$orend), - (add IntRegs:$base, u6_0ImmPred:$offset))]>, + (i32 IntRegs:$orend)), + (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>, Requires<[HasV4T, UseMEMOP]>; // MEMb_ADDSUBi_V4: @@ -3322,7 +4935,7 @@ def MEMb_ADDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$addend), "memb($addr) += $addend", [(truncstorei8 (add (sextloadi8 ADDRriU6_0:$addr), - IntRegs:$addend), ADDRriU6_0:$addr)]>, + (i32 IntRegs:$addend)), ADDRriU6_0:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) -= Rt @@ -3331,7 +4944,7 @@ def MEMb_SUBr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$subend), "memb($addr) -= $subend", [(truncstorei8 (sub (sextloadi8 ADDRriU6_0:$addr), - IntRegs:$subend), ADDRriU6_0:$addr)]>, + (i32 IntRegs:$subend)), ADDRriU6_0:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) &= Rt @@ -3340,7 +4953,7 @@ def MEMb_ANDr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$andend), "memb($addr) &= $andend", [(truncstorei8 (and (sextloadi8 ADDRriU6_0:$addr), - IntRegs:$andend), ADDRriU6_0:$addr)]>, + (i32 IntRegs:$andend)), ADDRriU6_0:$addr)]>, Requires<[HasV4T, UseMEMOP]>; // memb(Rs+#u6:0) |= Rt @@ -3349,7 +4962,7 @@ def MEMb_ORr_MEM_V4 : MEMInst_V4<(outs), (ins MEMri:$addr, IntRegs:$orend), "memb($addr) |= $orend", [(truncstorei8 (or (sextloadi8 ADDRriU6_0:$addr), - IntRegs:$orend), ADDRriU6_0:$addr)]>, + (i32 IntRegs:$orend)), ADDRriU6_0:$addr)]>, Requires<[HasV4T, UseMEMOP]>; @@ -3364,13 +4977,16 @@ def MEMb_ORr_MEM_V4 : MEMInst_V4<(outs), // The implemented patterns are: EQ/GT/GTU. // Missing patterns are: GE/GEU/LT/LTU/LE/LEU. +// Following instruction is not being extended as it results into the +// incorrect code for negative numbers. // Pd=cmpb.eq(Rs,#u8) + let isCompare = 1 in def CMPbEQri_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u8Imm:$src2), "$dst = cmpb.eq($src1, #$src2)", - [(set PredRegs:$dst, (seteq (and IntRegs:$src1, 255), - u8ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), + (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>, Requires<[HasV4T]>; // Pd=cmpb.eq(Rs,Rt) @@ -3378,10 +4994,9 @@ let isCompare = 1 in def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmpb.eq($src1, $src2)", - [(set PredRegs:$dst, (seteq (and (xor IntRegs:$src1, - IntRegs:$src2), - 255), - 0))]>, + [(set (i1 PredRegs:$dst), + (seteq (and (xor (i32 IntRegs:$src1), + (i32 IntRegs:$src2)), 255), 0))]>, Requires<[HasV4T]>; // Pd=cmpb.eq(Rs,Rt) @@ -3389,26 +5004,31 @@ let isCompare = 1 in def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmpb.eq($src1, $src2)", - [(set PredRegs:$dst, (seteq (shl IntRegs:$src1, (i32 24)), - (shl IntRegs:$src2, (i32 24))))]>, + [(set (i1 PredRegs:$dst), + (seteq (shl (i32 IntRegs:$src1), (i32 24)), + (shl (i32 IntRegs:$src2), (i32 24))))]>, Requires<[HasV4T]>; +/* Incorrect Pattern -- immediate should be right shifted before being +used in the cmpb.gt instruction. // Pd=cmpb.gt(Rs,#s8) let isCompare = 1 in def CMPbGTri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, s32Imm:$src2), + (ins IntRegs:$src1, s8Imm:$src2), "$dst = cmpb.gt($src1, #$src2)", - [(set PredRegs:$dst, (setgt (shl IntRegs:$src1, (i32 24)), - s32_24ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), (setgt (shl (i32 IntRegs:$src1), (i32 24)), + s8ImmPred:$src2))]>, Requires<[HasV4T]>; +*/ // Pd=cmpb.gt(Rs,Rt) let isCompare = 1 in def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmpb.gt($src1, $src2)", - [(set PredRegs:$dst, (setgt (shl IntRegs:$src1, (i32 24)), - (shl IntRegs:$src2, (i32 24))))]>, + [(set (i1 PredRegs:$dst), + (setgt (shl (i32 IntRegs:$src1), (i32 24)), + (shl (i32 IntRegs:$src2), (i32 24))))]>, Requires<[HasV4T]>; // Pd=cmpb.gtu(Rs,#u7) @@ -3416,8 +5036,8 @@ let isCompare = 1 in def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u7Imm:$src2), "$dst = cmpb.gtu($src1, #$src2)", - [(set PredRegs:$dst, (setugt (and IntRegs:$src1, 255), - u7ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255), + u7ImmPred:$src2))]>, Requires<[HasV4T]>; // Pd=cmpb.gtu(Rs,Rt) @@ -3425,18 +5045,21 @@ let isCompare = 1 in def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmpb.gtu($src1, $src2)", - [(set PredRegs:$dst, (setugt (and IntRegs:$src1, 255), - (and IntRegs:$src2, 255)))]>, + [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255), + (and (i32 IntRegs:$src2), 255)))]>, Requires<[HasV4T]>; +// Following instruction is not being extended as it results into the incorrect +// code for negative numbers. + // Signed half compare(.eq) ri. // Pd=cmph.eq(Rs,#s8) let isCompare = 1 in def CMPhEQri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, u16Imm:$src2), + (ins IntRegs:$src1, s8Imm:$src2), "$dst = cmph.eq($src1, #$src2)", - [(set PredRegs:$dst, (seteq (and IntRegs:$src1, 65535), - u16_s8ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), (seteq (and (i32 IntRegs:$src1), 65535), + s8ImmPred:$src2))]>, Requires<[HasV4T]>; // Signed half compare(.eq) rr. @@ -3449,10 +5072,9 @@ let isCompare = 1 in def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmph.eq($src1, $src2)", - [(set PredRegs:$dst, (seteq (and (xor IntRegs:$src1, - IntRegs:$src2), - 65535), - 0))]>, + [(set (i1 PredRegs:$dst), (seteq (and (xor (i32 IntRegs:$src1), + (i32 IntRegs:$src2)), + 65535), 0))]>, Requires<[HasV4T]>; // Signed half compare(.eq) rr. @@ -3465,19 +5087,25 @@ let isCompare = 1 in def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmph.eq($src1, $src2)", - [(set PredRegs:$dst, (seteq (shl IntRegs:$src1, (i32 16)), - (shl IntRegs:$src2, (i32 16))))]>, + [(set (i1 PredRegs:$dst), + (seteq (shl (i32 IntRegs:$src1), (i32 16)), + (shl (i32 IntRegs:$src2), (i32 16))))]>, Requires<[HasV4T]>; +/* Incorrect Pattern -- immediate should be right shifted before being +used in the cmph.gt instruction. // Signed half compare(.gt) ri. // Pd=cmph.gt(Rs,#s8) + let isCompare = 1 in def CMPhGTri_V4 : MInst<(outs PredRegs:$dst), - (ins IntRegs:$src1, s32Imm:$src2), + (ins IntRegs:$src1, s8Imm:$src2), "$dst = cmph.gt($src1, #$src2)", - [(set PredRegs:$dst, (setgt (shl IntRegs:$src1, (i32 16)), - s32_16s8ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), + (setgt (shl (i32 IntRegs:$src1), (i32 16)), + s8ImmPred:$src2))]>, Requires<[HasV4T]>; +*/ // Signed half compare(.gt) rr. // Pd=cmph.gt(Rs,Rt) @@ -3485,8 +5113,9 @@ let isCompare = 1 in def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmph.gt($src1, $src2)", - [(set PredRegs:$dst, (setgt (shl IntRegs:$src1, (i32 16)), - (shl IntRegs:$src2, (i32 16))))]>, + [(set (i1 PredRegs:$dst), + (setgt (shl (i32 IntRegs:$src1), (i32 16)), + (shl (i32 IntRegs:$src2), (i32 16))))]>, Requires<[HasV4T]>; // Unsigned half compare rr (.gtu). @@ -3495,8 +5124,9 @@ let isCompare = 1 in def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2), "$dst = cmph.gtu($src1, $src2)", - [(set PredRegs:$dst, (setugt (and IntRegs:$src1, 65535), - (and IntRegs:$src2, 65535)))]>, + [(set (i1 PredRegs:$dst), + (setugt (and (i32 IntRegs:$src1), 65535), + (and (i32 IntRegs:$src2), 65535)))]>, Requires<[HasV4T]>; // Unsigned half compare ri (.gtu). @@ -3505,8 +5135,8 @@ let isCompare = 1 in def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u7Imm:$src2), "$dst = cmph.gtu($src1, #$src2)", - [(set PredRegs:$dst, (setugt (and IntRegs:$src1, 65535), - u7ImmPred:$src2))]>, + [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535), + u7ImmPred:$src2))]>, Requires<[HasV4T]>; //===----------------------------------------------------------------------===// @@ -3523,9 +5153,37 @@ let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1, Requires<[HasV4T]>; } +// Restore registers and dealloc return function call. +let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1, + Defs = [R29, R30, R31, PC] in { + def RESTORE_DEALLOC_RET_JMP_V4 : JInst<(outs), (ins calltarget:$dst, variable_ops), + "jump $dst // Restore_and_dealloc_return", + []>, + Requires<[HasV4T]>; +} + +// Restore registers and dealloc frame before a tail call. +let isCall = 1, isBarrier = 1, + Defs = [R29, R30, R31, PC] in { + def RESTORE_DEALLOC_BEFORE_TAILCALL_V4 : JInst<(outs), (ins calltarget:$dst, variable_ops), + "call $dst // Restore_and_dealloc_before_tailcall", + []>, + Requires<[HasV4T]>; +} + +// Save registers function call. +let isCall = 1, isBarrier = 1, + Uses = [R29, R31] in { + def SAVE_REGISTERS_CALL_V4 : JInst<(outs), (ins calltarget:$dst, variable_ops), + "call $dst // Save_calle_saved_registers", + []>, + Requires<[HasV4T]>; +} + // if (Ps) dealloc_return let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cPt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if ($src1) dealloc_return", []>, @@ -3534,7 +5192,8 @@ let isReturn = 1, isTerminator = 1, // if (!Ps) dealloc_return let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cNotPt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if (!$src1) dealloc_return", @@ -3544,7 +5203,8 @@ let isReturn = 1, isTerminator = 1, // if (Ps.new) dealloc_return:nt let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cdnPnt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if ($src1.new) dealloc_return:nt", @@ -3554,7 +5214,8 @@ let isReturn = 1, isTerminator = 1, // if (!Ps.new) dealloc_return:nt let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cNotdnPnt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if (!$src1.new) dealloc_return:nt", @@ -3564,7 +5225,8 @@ let isReturn = 1, isTerminator = 1, // if (Ps.new) dealloc_return:t let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cdnPt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if ($src1.new) dealloc_return:t", @@ -3574,10 +5236,511 @@ let isReturn = 1, isTerminator = 1, // if (!Ps.new) dealloc_return:nt let isReturn = 1, isTerminator = 1, - Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1 in { + Defs = [R29, R30, R31, PC], Uses = [R29, R31], neverHasSideEffects = 1, + isPredicated = 1 in { def DEALLOC_RET_cNotdnPt_V4 : NVInst_V4<(outs), (ins PredRegs:$src1, i32imm:$amt1), "if (!$src1.new) dealloc_return:t", []>, Requires<[HasV4T]>; } + + +// Load/Store with absolute addressing mode +// memw(#u6)=Rt + +multiclass ST_abs { + let isPredicable = 1 in + def _abs_V4 : STInst<(outs), + (ins globaladdress:$absaddr, IntRegs:$src), + !strconcat(OpcStr, "(##$absaddr) = $src"), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if ($src1)", !strconcat(OpcStr, "(##$absaddr) = $src2")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if (!$src1)", !strconcat(OpcStr, "(##$absaddr) = $src2")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if ($src1.new)", !strconcat(OpcStr, "(##$absaddr) = $src2")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if (!$src1.new)", !strconcat(OpcStr, "(##$absaddr) = $src2")), + []>, + Requires<[HasV4T]>; + + def _abs_nv_V4 : STInst<(outs), + (ins globaladdress:$absaddr, IntRegs:$src), + !strconcat(OpcStr, "(##$absaddr) = $src.new"), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if ($src1)", !strconcat(OpcStr, "(##$absaddr) = $src2.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if (!$src1)", !strconcat(OpcStr, "(##$absaddr) = $src2.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if ($src1.new)", !strconcat(OpcStr, "(##$absaddr) = $src2.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2), + !strconcat("if (!$src1.new)", !strconcat(OpcStr, "(##$absaddr) = $src2.new")), + []>, + Requires<[HasV4T]>; +} + +let AddedComplexity = 30, isPredicable = 1 in +def STrid_abs_V4 : STInst<(outs), + (ins globaladdress:$absaddr, DoubleRegs:$src), + "memd(##$absaddr) = $src", + [(store (i64 DoubleRegs:$src), (HexagonCONST32 tglobaladdr:$absaddr))]>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def STrid_abs_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2), + "if ($src1) memd(##$absaddr) = $src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def STrid_abs_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2), + "if (!$src1) memd(##$absaddr) = $src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def STrid_abs_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2), + "if ($src1.new) memd(##$absaddr) = $src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def STrid_abs_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2), + "if (!$src1.new) memd(##$absaddr) = $src2", + []>, + Requires<[HasV4T]>; + +defm STrib : ST_abs<"memb">; +defm STrih : ST_abs<"memh">; +defm STriw : ST_abs<"memw">; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(truncstorei8 (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)), + (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(truncstorei16 (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)), + (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)), + (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>; + + +multiclass LD_abs { + let isPredicable = 1 in + def _abs_V4 : LDInst<(outs IntRegs:$dst), + (ins globaladdress:$absaddr), + !strconcat("$dst = ", !strconcat(OpcStr, "(##$absaddr)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + !strconcat("if ($src1) $dst = ", !strconcat(OpcStr, "(##$absaddr)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + !strconcat("if (!$src1) $dst = ", !strconcat(OpcStr, "(##$absaddr)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + !strconcat("if ($src1.new) $dst = ", !strconcat(OpcStr, "(##$absaddr)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + !strconcat("if (!$src1.new) $dst = ", !strconcat(OpcStr, "(##$absaddr)")), + []>, + Requires<[HasV4T]>; +} + +let AddedComplexity = 30 in +def LDrid_abs_V4 : LDInst<(outs DoubleRegs:$dst), + (ins globaladdress:$absaddr), + "$dst = memd(##$absaddr)", + [(set (i64 DoubleRegs:$dst), (load (HexagonCONST32 tglobaladdr:$absaddr)))]>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def LDrid_abs_cPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + "if ($src1) $dst = memd(##$absaddr)", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def LDrid_abs_cNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + "if (!$src1) $dst = memd(##$absaddr)", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def LDrid_abs_cdnPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + "if ($src1.new) $dst = memd(##$absaddr)", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 30, isPredicated = 1 in +def LDrid_abs_cdnNotPt_V4 : LDInst<(outs DoubleRegs:$dst), + (ins PredRegs:$src1, globaladdress:$absaddr), + "if (!$src1.new) $dst = memd(##$absaddr)", + []>, + Requires<[HasV4T]>; + +defm LDrib : LD_abs<"memb">; +defm LDriub : LD_abs<"memub">; +defm LDrih : LD_abs<"memh">; +defm LDriuh : LD_abs<"memuh">; +defm LDriw : LD_abs<"memw">; + + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))), + (LDriw_abs_V4 tglobaladdr: $absaddr)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))), + (LDrib_abs_V4 tglobaladdr:$absaddr)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))), + (LDriub_abs_V4 tglobaladdr:$absaddr)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))), + (LDrih_abs_V4 tglobaladdr:$absaddr)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))), + (LDriuh_abs_V4 tglobaladdr:$absaddr)>; + +// Transfer global address into a register +let AddedComplexity=50, isMoveImm = 1, isReMaterializable = 1 in +def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins globaladdress:$src1), + "$dst = ##$src1", + [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>, + Requires<[HasV4T]>; + +let AddedComplexity=50, neverHasSideEffects = 1, isPredicated = 1 in +def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$src2), + "if($src1) $dst = ##$src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity=50, neverHasSideEffects = 1, isPredicated = 1 in +def TFRI_cNotPt_V4 : ALU32_ri<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$src2), + "if(!$src1) $dst = ##$src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity=50, neverHasSideEffects = 1, isPredicated = 1 in +def TFRI_cdnPt_V4 : ALU32_ri<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$src2), + "if($src1.new) $dst = ##$src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity=50, neverHasSideEffects = 1, isPredicated = 1 in +def TFRI_cdnNotPt_V4 : ALU32_ri<(outs IntRegs:$dst), + (ins PredRegs:$src1, globaladdress:$src2), + "if(!$src1.new) $dst = ##$src2", + []>, + Requires<[HasV4T]>; + +let AddedComplexity = 50, Predicates = [HasV4T] in +def : Pat<(HexagonCONST32_GP tglobaladdr:$src1), + (TFRI_V4 tglobaladdr:$src1)>; + + +// Load - Indirect with long offset: These instructions take global address +// as an operand +let AddedComplexity = 10 in +def LDrid_ind_lo_V4 : LDInst<(outs DoubleRegs:$dst), + (ins IntRegs:$src1, u2Imm:$src2, globaladdress:$offset), + "$dst=memd($src1<<#$src2+##$offset)", + [(set (i64 DoubleRegs:$dst), + (load (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$offset))))]>, + Requires<[HasV4T]>; + +let AddedComplexity = 10 in +multiclass LD_indirect_lo { + def _lo_V4 : LDInst<(outs IntRegs:$dst), + (ins IntRegs:$src1, u2Imm:$src2, globaladdress:$offset), + !strconcat("$dst = ", !strconcat(OpcStr, "($src1<<#$src2+##$offset)")), + [(set IntRegs:$dst, + (i32 (OpNode (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$offset)))))]>, + Requires<[HasV4T]>; +} + +defm LDrib_ind : LD_indirect_lo<"memb", sextloadi8>; +defm LDriub_ind : LD_indirect_lo<"memub", zextloadi8>; +defm LDrih_ind : LD_indirect_lo<"memh", sextloadi16>; +defm LDriuh_ind : LD_indirect_lo<"memuh", zextloadi16>; +defm LDriw_ind : LD_indirect_lo<"memw", load>; + +// Store - Indirect with long offset: These instructions take global address +// as an operand +let AddedComplexity = 10 in +def STrid_ind_lo_V4 : STInst<(outs), + (ins IntRegs:$src1, u2Imm:$src2, globaladdress:$src3, + DoubleRegs:$src4), + "memd($src1<<#$src2+#$src3) = $src4", + [(store (i64 DoubleRegs:$src4), + (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$src3)))]>, + Requires<[HasV4T]>; + +let AddedComplexity = 10 in +multiclass ST_indirect_lo { + def _lo_V4 : STInst<(outs), + (ins IntRegs:$src1, u2Imm:$src2, globaladdress:$src3, + IntRegs:$src4), + !strconcat(OpcStr, "($src1<<#$src2+##$src3) = $src4"), + [(OpNode (i32 IntRegs:$src4), + (add (shl IntRegs:$src1, u2ImmPred:$src2), + (HexagonCONST32 tglobaladdr:$src3)))]>, + Requires<[HasV4T]>; +} + +defm STrib_ind : ST_indirect_lo<"memb", truncstorei8>; +defm STrih_ind : ST_indirect_lo<"memh", truncstorei16>; +defm STriw_ind : ST_indirect_lo<"memw", store>; + +// Store - absolute addressing mode: These instruction take constant +// value as the extended operand +multiclass ST_absimm { + let isPredicable = 1 in + def _abs_V4 : STInst<(outs), + (ins u6Imm:$src1, IntRegs:$src2), + !strconcat(OpcStr, "(#$src1) = $src2"), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if ($src1)", !strconcat(OpcStr, "(#$src2) = $src3")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if (!$src1)", !strconcat(OpcStr, "(#$src2) = $src3")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if ($src1.new)", !strconcat(OpcStr, "(#$src2) = $src3")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if (!$src1.new)", !strconcat(OpcStr, "(#$src2) = $src3")), + []>, + Requires<[HasV4T]>; + + def _abs_nv_V4 : STInst<(outs), + (ins u6Imm:$src1, IntRegs:$src2), + !strconcat(OpcStr, "(#$src1) = $src2.new"), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if ($src1)", !strconcat(OpcStr, "(#$src2) = $src3.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if (!$src1)", !strconcat(OpcStr, "(#$src2) = $src3.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if ($src1.new)", !strconcat(OpcStr, "(#$src2) = $src3.new")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_nv_V4 : STInst<(outs), + (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3), + !strconcat("if (!$src1.new)", !strconcat(OpcStr, "(#$src2) = $src3.new")), + []>, + Requires<[HasV4T]>; +} + +defm STrib_imm : ST_absimm<"memb">; +defm STrih_imm : ST_absimm<"memh">; +defm STriw_imm : ST_absimm<"memw">; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(truncstorei8 (i32 IntRegs:$src1), u6ImmPred:$src2), + (STrib_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(truncstorei16 (i32 IntRegs:$src1), u6ImmPred:$src2), + (STrih_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(store (i32 IntRegs:$src1), u6ImmPred:$src2), + (STriw_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>; + + +// Load - absolute addressing mode: These instruction take constant +// value as the extended operand + +multiclass LD_absimm { + let isPredicable = 1 in + def _abs_V4 : LDInst<(outs IntRegs:$dst), + (ins u6Imm:$src), + !strconcat("$dst = ", !strconcat(OpcStr, "(#$src)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, u6Imm:$src2), + !strconcat("if ($src1) $dst = ", !strconcat(OpcStr, "(#$src2)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, u6Imm:$src2), + !strconcat("if (!$src1) $dst = ", !strconcat(OpcStr, "(#$src2)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, u6Imm:$src2), + !strconcat("if ($src1.new) $dst = ", !strconcat(OpcStr, "(#$src2)")), + []>, + Requires<[HasV4T]>; + + let isPredicated = 1 in + def _abs_cdnNotPt_V4 : LDInst<(outs IntRegs:$dst), + (ins PredRegs:$src1, u6Imm:$src2), + !strconcat("if (!$src1.new) $dst = ", !strconcat(OpcStr, "(#$src2)")), + []>, + Requires<[HasV4T]>; +} + +defm LDrib_imm : LD_absimm<"memb">; +defm LDriub_imm : LD_absimm<"memub">; +defm LDrih_imm : LD_absimm<"memh">; +defm LDriuh_imm : LD_absimm<"memuh">; +defm LDriw_imm : LD_absimm<"memw">; + +let Predicates = [HasV4T], AddedComplexity = 30 in +def : Pat<(i32 (load u6ImmPred:$src)), + (LDriw_imm_abs_V4 u6ImmPred:$src)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (sextloadi8 u6ImmPred:$src)), + (LDrib_imm_abs_V4 u6ImmPred:$src)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (zextloadi8 u6ImmPred:$src)), + (LDriub_imm_abs_V4 u6ImmPred:$src)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (sextloadi16 u6ImmPred:$src)), + (LDrih_imm_abs_V4 u6ImmPred:$src)>; + +let Predicates = [HasV4T], AddedComplexity=30 in +def : Pat<(i32 (zextloadi16 u6ImmPred:$src)), + (LDriuh_imm_abs_V4 u6ImmPred:$src)>; + + +// Indexed store double word - global address. +// memw(Rs+#u6:2)=#S8 +let AddedComplexity = 10 in +def STriw_offset_ext_V4 : STInst<(outs), + (ins IntRegs:$src1, u6_2Imm:$src2, globaladdress:$src3), + "memw($src1+#$src2) = ##$src3", + [(store (HexagonCONST32 tglobaladdr:$src3), + (add IntRegs:$src1, u6_2ImmPred:$src2))]>, + Requires<[HasV4T]>; + + +// Indexed store double word - global address. +// memw(Rs+#u6:2)=#S8 +let AddedComplexity = 10 in +def STrih_offset_ext_V4 : STInst<(outs), + (ins IntRegs:$src1, u6_1Imm:$src2, globaladdress:$src3), + "memh($src1+#$src2) = ##$src3", + [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3), + (add IntRegs:$src1, u6_1ImmPred:$src2))]>, + Requires<[HasV4T]>; diff --git a/lib/Target/Hexagon/HexagonMCInst.h b/lib/Target/Hexagon/HexagonMCInst.h new file mode 100644 index 00000000000..16ea7cf6ed7 --- /dev/null +++ b/lib/Target/Hexagon/HexagonMCInst.h @@ -0,0 +1,41 @@ +//===- HexagonMCInst.h - Hexagon sub-class of MCInst ----------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This class extends MCInst to allow some VLIW annotation. +// +//===----------------------------------------------------------------------===// + +#ifndef HEXAGONMCINST_H +#define HEXAGONMCINST_H + +#include "llvm/MC/MCInst.h" +#include "llvm/CodeGen/MachineInstr.h" + +namespace llvm { + class HexagonMCInst: public MCInst { + // Packet start and end markers + unsigned startPacket: 1, endPacket: 1; + const MachineInstr *MachineI; + public: + explicit HexagonMCInst(): MCInst(), + startPacket(0), endPacket(0) {} + + const MachineInstr* getMI() const { return MachineI; }; + + void setMI(const MachineInstr *MI) { MachineI = MI; }; + + bool isStartPacket() const { return (startPacket); }; + bool isEndPacket() const { return (endPacket); }; + + void setStartPacket(bool yes) { startPacket = yes; }; + void setEndPacket(bool yes) { endPacket = yes; }; + }; +} + +#endif diff --git a/lib/Target/Hexagon/HexagonSchedule.td b/lib/Target/Hexagon/HexagonSchedule.td index fbea4452ec6..c4887963895 100644 --- a/lib/Target/Hexagon/HexagonSchedule.td +++ b/lib/Target/Hexagon/HexagonSchedule.td @@ -13,7 +13,6 @@ def LSUNIT : FuncUnit; def MUNIT : FuncUnit; def SUNIT : FuncUnit; - // Itinerary classes def ALU32 : InstrItinClass; def ALU64 : InstrItinClass; @@ -24,23 +23,25 @@ def LD : InstrItinClass; def M : InstrItinClass; def ST : InstrItinClass; def S : InstrItinClass; +def SYS : InstrItinClass; +def MARKER : InstrItinClass; def PSEUDO : InstrItinClass; - def HexagonItineraries : - ProcessorItineraries<[LUNIT, LSUNIT, MUNIT, SUNIT], [], [ - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]> -]>; - + ProcessorItineraries<[LUNIT, LSUNIT, MUNIT, SUNIT], [], [ + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]> + ]>; //===----------------------------------------------------------------------===// // V4 Machine Info + diff --git a/lib/Target/Hexagon/HexagonScheduleV4.td b/lib/Target/Hexagon/HexagonScheduleV4.td index 4cf66fe7434..1d82dbb90e9 100644 --- a/lib/Target/Hexagon/HexagonScheduleV4.td +++ b/lib/Target/Hexagon/HexagonScheduleV4.td @@ -23,7 +23,6 @@ // | SLOT3 | XTYPE ALU32 J CR | // |===========|==================================================| - // Functional Units. def SLOT0 : FuncUnit; def SLOT1 : FuncUnit; @@ -34,22 +33,26 @@ def SLOT3 : FuncUnit; def NV_V4 : InstrItinClass; def MEM_V4 : InstrItinClass; // ALU64/M/S Instruction classes of V2 are collectively knownn as XTYPE in V4. +def PREFIX : InstrItinClass; -def HexagonItinerariesV4 : ProcessorItineraries< - [SLOT0, SLOT1, SLOT2, SLOT3], [], [ - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]>, - InstrItinData]> -]>; +def HexagonItinerariesV4 : + ProcessorItineraries<[SLOT0, SLOT1, SLOT2, SLOT3], [], [ + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]>, + InstrItinData]> + ]>; //===----------------------------------------------------------------------===// // Hexagon V4 Resource Definitions - diff --git a/lib/Target/Hexagon/HexagonTargetMachine.cpp b/lib/Target/Hexagon/HexagonTargetMachine.cpp index b9e68949366..411325bf963 100644 --- a/lib/Target/Hexagon/HexagonTargetMachine.cpp +++ b/lib/Target/Hexagon/HexagonTargetMachine.cpp @@ -138,5 +138,8 @@ bool HexagonPassConfig::addPreEmitPass() { // Split up TFRcondsets into conditional transfers. PM.add(createHexagonSplitTFRCondSets(getHexagonTargetMachine())); + // Create Packets. + PM.add(createHexagonPacketizer()); + return false; } diff --git a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp new file mode 100644 index 00000000000..0fe97ef5f70 --- /dev/null +++ b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp @@ -0,0 +1,3654 @@ +//===----- HexagonPacketizer.cpp - vliw packetizer ---------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This implements a simple VLIW packetizer using DFA. The packetizer works on +// machine basic blocks. For each instruction I in BB, the packetizer consults +// the DFA to see if machine resources are available to execute I. If so, the +// packetizer checks if I depends on any instruction J in the current packet. +// If no dependency is found, I is added to current packet and machine resource +// is marked as taken. If any dependency is found, a target API call is made to +// prune the dependence. +// +//===----------------------------------------------------------------------===// +#define DEBUG_TYPE "packets" +#include "llvm/CodeGen/DFAPacketizer.h" +#include "llvm/CodeGen/Passes.h" +#include "llvm/CodeGen/MachineDominators.h" +#include "llvm/CodeGen/MachineFunctionPass.h" +#include "llvm/CodeGen/MachineLoopInfo.h" +#include "llvm/CodeGen/ScheduleDAG.h" +#include "llvm/CodeGen/ScheduleDAGInstrs.h" +#include "llvm/CodeGen/LatencyPriorityQueue.h" +#include "llvm/CodeGen/SchedulerRegistry.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/MachineFunctionAnalysis.h" +#include "llvm/CodeGen/ScheduleHazardRecognizer.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Support/MathExtras.h" +#include "llvm/MC/MCInstrItineraries.h" +#include "llvm/Support/Compiler.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" +#include "Hexagon.h" +#include "HexagonTargetMachine.h" +#include "HexagonRegisterInfo.h" +#include "HexagonSubtarget.h" +#include "HexagonMachineFunctionInfo.h" + +#include + +using namespace llvm; + +namespace { + class HexagonPacketizer : public MachineFunctionPass { + + public: + static char ID; + HexagonPacketizer() : MachineFunctionPass(ID) {} + + void getAnalysisUsage(AnalysisUsage &AU) const { + AU.setPreservesCFG(); + AU.addRequired(); + AU.addPreserved(); + AU.addRequired(); + AU.addPreserved(); + MachineFunctionPass::getAnalysisUsage(AU); + } + + const char *getPassName() const { + return "Hexagon Packetizer"; + } + + bool runOnMachineFunction(MachineFunction &Fn); + }; + char HexagonPacketizer::ID = 0; + + class HexagonPacketizerList : public VLIWPacketizerList { + + private: + + // Has the instruction been promoted to a dot-new instruction. + bool PromotedToDotNew; + + // Has the instruction been glued to allocframe. + bool GlueAllocframeStore; + + // Has the feeder instruction been glued to new value jump. + bool GlueToNewValueJump; + + // Check if there is a dependence between some instruction already in this + // packet and this instruction. + bool Dependence; + + // Only check for dependence if there are resources available to + // schedule this instruction. + bool FoundSequentialDependence; + + public: + // Ctor. + HexagonPacketizerList(MachineFunction &MF, MachineLoopInfo &MLI, + MachineDominatorTree &MDT); + + // initPacketizerState - initialize some internal flags. + void initPacketizerState(void); + + // ignorePseudoInstruction - Ignore bundling of pseudo instructions. + bool ignorePseudoInstruction(MachineInstr *MI, MachineBasicBlock *MBB); + + // isSoloInstruction - return true if instruction MI can not be packetized + // with any other instruction, which means that MI itself is a packet. + bool isSoloInstruction(MachineInstr *MI); + + // isLegalToPacketizeTogether - Is it legal to packetize SUI and SUJ + // together. + bool isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ); + + // isLegalToPruneDependencies - Is it legal to prune dependece between SUI + // and SUJ. + bool isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ); + + MachineBasicBlock::iterator addToPacket(MachineInstr *MI); + private: + bool IsCallDependent(MachineInstr* MI, SDep::Kind DepType, unsigned DepReg); + bool PromoteToDotNew(MachineInstr* MI, SDep::Kind DepType, + MachineBasicBlock::iterator &MII, + const TargetRegisterClass* RC); + bool CanPromoteToDotNew(MachineInstr* MI, SUnit* PacketSU, + unsigned DepReg, + std::map MIToSUnit, + MachineBasicBlock::iterator &MII, + const TargetRegisterClass* RC); + bool CanPromoteToNewValue(MachineInstr* MI, SUnit* PacketSU, + unsigned DepReg, + std::map MIToSUnit, + MachineBasicBlock::iterator &MII); + bool CanPromoteToNewValueStore(MachineInstr* MI, MachineInstr* PacketMI, + unsigned DepReg, + std::map MIToSUnit); + bool DemoteToDotOld(MachineInstr* MI); + bool ArePredicatesComplements(MachineInstr* MI1, MachineInstr* MI2, + std::map MIToSUnit); + bool RestrictingDepExistInPacket(MachineInstr*, + unsigned, std::map ); + bool isNewifiable(MachineInstr* MI); + bool isCondInst(MachineInstr* MI); + bool IsNewifyStore (MachineInstr* MI); + bool tryAllocateResourcesForConstExt(MachineInstr* MI); + bool canReserveResourcesForConstExt(MachineInstr *MI); + void reserveResourcesForConstExt(MachineInstr* MI); + bool isNewValueInst(MachineInstr* MI); + bool isDotNewInst(MachineInstr* MI); + }; +} + +// HexagonPacketizerList Ctor. +HexagonPacketizerList::HexagonPacketizerList( + MachineFunction &MF, MachineLoopInfo &MLI,MachineDominatorTree &MDT) + : VLIWPacketizerList(MF, MLI, MDT, true){ +} + +bool HexagonPacketizer::runOnMachineFunction(MachineFunction &Fn) { + const TargetInstrInfo *TII = Fn.getTarget().getInstrInfo(); + MachineLoopInfo &MLI = getAnalysis(); + MachineDominatorTree &MDT = getAnalysis(); + + // Instantiate the packetizer. + HexagonPacketizerList Packetizer(Fn, MLI, MDT); + + // DFA state table should not be empty. + assert(Packetizer.getResourceTracker() && "Empty DFA table!"); + + // + // Loop over all basic blocks and remove KILL pseudo-instructions + // These instructions confuse the dependence analysis. Consider: + // D0 = ... (Insn 0) + // R0 = KILL R0, D0 (Insn 1) + // R0 = ... (Insn 2) + // Here, Insn 1 will result in the dependence graph not emitting an output + // dependence between Insn 0 and Insn 2. This can lead to incorrect + // packetization + // + for (MachineFunction::iterator MBB = Fn.begin(), MBBe = Fn.end(); + MBB != MBBe; ++MBB) { + MachineBasicBlock::iterator End = MBB->end(); + MachineBasicBlock::iterator MI = MBB->begin(); + while (MI != End) { + if (MI->isKill()) { + MachineBasicBlock::iterator DeleteMI = MI; + ++MI; + MBB->erase(DeleteMI); + End = MBB->end(); + continue; + } + ++MI; + } + } + + // Loop over all of the basic blocks. + for (MachineFunction::iterator MBB = Fn.begin(), MBBe = Fn.end(); + MBB != MBBe; ++MBB) { + // Find scheduling regions and schedule / packetize each region. + unsigned RemainingCount = MBB->size(); + for(MachineBasicBlock::iterator RegionEnd = MBB->end(); + RegionEnd != MBB->begin();) { + // The next region starts above the previous region. Look backward in the + // instruction stream until we find the nearest boundary. + MachineBasicBlock::iterator I = RegionEnd; + for(;I != MBB->begin(); --I, --RemainingCount) { + if (TII->isSchedulingBoundary(llvm::prior(I), MBB, Fn)) + break; + } + I = MBB->begin(); + + // Skip empty scheduling regions. + if (I == RegionEnd) { + RegionEnd = llvm::prior(RegionEnd); + --RemainingCount; + continue; + } + // Skip regions with one instruction. + if (I == llvm::prior(RegionEnd)) { + RegionEnd = llvm::prior(RegionEnd); + continue; + } + + Packetizer.PacketizeMIs(MBB, I, RegionEnd); + RegionEnd = I; + } + } + + return true; +} + + +static bool IsIndirectCall(MachineInstr* MI) { + return ((MI->getOpcode() == Hexagon::CALLR) || + (MI->getOpcode() == Hexagon::CALLRv3)); +} + +// Reserve resources for constant extender. Trigure an assertion if +// reservation fail. +void HexagonPacketizerList::reserveResourcesForConstExt(MachineInstr* MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + MachineInstr *PseudoMI = MI->getParent()->getParent()->CreateMachineInstr( + QII->get(Hexagon::IMMEXT), MI->getDebugLoc()); + + if (ResourceTracker->canReserveResources(PseudoMI)) { + ResourceTracker->reserveResources(PseudoMI); + MI->getParent()->getParent()->DeleteMachineInstr(PseudoMI); + } else { + MI->getParent()->getParent()->DeleteMachineInstr(PseudoMI); + assert(0 && "can not reserve resources for constant extender."); + } + return; +} + +bool HexagonPacketizerList::canReserveResourcesForConstExt(MachineInstr *MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + assert(QII->isExtended(MI) && + "Should only be called for constant extended instructions"); + MachineFunction *MF = MI->getParent()->getParent(); + MachineInstr *PseudoMI = MF->CreateMachineInstr(QII->get(Hexagon::IMMEXT), + MI->getDebugLoc()); + bool CanReserve = ResourceTracker->canReserveResources(PseudoMI); + MF->DeleteMachineInstr(PseudoMI); + return CanReserve; +} + +// Allocate resources (i.e. 4 bytes) for constant extender. If succeed, return +// true, otherwise, return false. +bool HexagonPacketizerList::tryAllocateResourcesForConstExt(MachineInstr* MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + MachineInstr *PseudoMI = MI->getParent()->getParent()->CreateMachineInstr( + QII->get(Hexagon::IMMEXT), MI->getDebugLoc()); + + if (ResourceTracker->canReserveResources(PseudoMI)) { + ResourceTracker->reserveResources(PseudoMI); + MI->getParent()->getParent()->DeleteMachineInstr(PseudoMI); + return true; + } else { + MI->getParent()->getParent()->DeleteMachineInstr(PseudoMI); + return false; + } +} + + +bool HexagonPacketizerList::IsCallDependent(MachineInstr* MI, + SDep::Kind DepType, + unsigned DepReg) { + + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + const HexagonRegisterInfo* QRI = (const HexagonRegisterInfo *) TM.getRegisterInfo(); + + // Check for lr dependence + if (DepReg == QRI->getRARegister()) { + return true; + } + + if (QII->isDeallocRet(MI)) { + if (DepReg == QRI->getFrameRegister() || + DepReg == QRI->getStackRegister()) + return true; + } + + // Check if this is a predicate dependence + const TargetRegisterClass* RC = QRI->getMinimalPhysRegClass(DepReg); + if (RC == Hexagon::PredRegsRegisterClass) { + return true; + } + + // + // Lastly check for an operand used in an indirect call + // If we had an attribute for checking if an instruction is an indirect call, + // then we could have avoided this relatively brittle implementation of + // IsIndirectCall() + // + // Assumes that the first operand of the CALLr is the function address + // + if (IsIndirectCall(MI) && (DepType == SDep::Data)) { + MachineOperand MO = MI->getOperand(0); + if (MO.isReg() && MO.isUse() && (MO.getReg() == DepReg)) { + return true; + } + } + + return false; +} + +static bool IsRegDependence(const SDep::Kind DepType) { + return (DepType == SDep::Data || DepType == SDep::Anti || + DepType == SDep::Output); +} + +static bool IsDirectJump(MachineInstr* MI) { + return (MI->getOpcode() == Hexagon::JMP); +} + +static bool IsSchedBarrier(MachineInstr* MI) { + switch (MI->getOpcode()) { + case Hexagon::BARRIER: + return true; + } + return false; +} + +static bool IsControlFlow(MachineInstr* MI) { + return (MI->getDesc().isTerminator() || MI->getDesc().isCall()); +} + +bool HexagonPacketizerList::isNewValueInst(MachineInstr* MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + if (QII->isNewValueJump(MI)) + return true; + + if (QII->isNewValueStore(MI)) + return true; + + return false; +} + +// Function returns true if an instruction can be promoted to the new-value +// store. It will always return false for v2 and v3. +// It lists all the conditional and unconditional stores that can be promoted +// to the new-value stores. + +bool HexagonPacketizerList::IsNewifyStore (MachineInstr* MI) { + const HexagonRegisterInfo* QRI = (const HexagonRegisterInfo *) TM.getRegisterInfo(); + switch (MI->getOpcode()) + { + // store byte + case Hexagon::STrib: + case Hexagon::STrib_indexed: + case Hexagon::STrib_indexed_shl_V4: + case Hexagon::STrib_shl_V4: + case Hexagon::STrib_GP_V4: + case Hexagon::STb_GP_V4: + case Hexagon::POST_STbri: + case Hexagon::STrib_cPt: + case Hexagon::STrib_cdnPt_V4: + case Hexagon::STrib_cNotPt: + case Hexagon::STrib_cdnNotPt_V4: + case Hexagon::STrib_indexed_cPt: + case Hexagon::STrib_indexed_cdnPt_V4: + case Hexagon::STrib_indexed_cNotPt: + case Hexagon::STrib_indexed_cdnNotPt_V4: + case Hexagon::STrib_indexed_shl_cPt_V4: + case Hexagon::STrib_indexed_shl_cdnPt_V4: + case Hexagon::STrib_indexed_shl_cNotPt_V4: + case Hexagon::STrib_indexed_shl_cdnNotPt_V4: + case Hexagon::POST_STbri_cPt: + case Hexagon::POST_STbri_cdnPt_V4: + case Hexagon::POST_STbri_cNotPt: + case Hexagon::POST_STbri_cdnNotPt_V4: + case Hexagon::STb_GP_cPt_V4: + case Hexagon::STb_GP_cNotPt_V4: + case Hexagon::STb_GP_cdnPt_V4: + case Hexagon::STb_GP_cdnNotPt_V4: + case Hexagon::STrib_GP_cPt_V4: + case Hexagon::STrib_GP_cNotPt_V4: + case Hexagon::STrib_GP_cdnPt_V4: + case Hexagon::STrib_GP_cdnNotPt_V4: + + // store halfword + case Hexagon::STrih: + case Hexagon::STrih_indexed: + case Hexagon::STrih_indexed_shl_V4: + case Hexagon::STrih_shl_V4: + case Hexagon::STrih_GP_V4: + case Hexagon::STh_GP_V4: + case Hexagon::POST_SThri: + case Hexagon::STrih_cPt: + case Hexagon::STrih_cdnPt_V4: + case Hexagon::STrih_cNotPt: + case Hexagon::STrih_cdnNotPt_V4: + case Hexagon::STrih_indexed_cPt: + case Hexagon::STrih_indexed_cdnPt_V4: + case Hexagon::STrih_indexed_cNotPt: + case Hexagon::STrih_indexed_cdnNotPt_V4: + case Hexagon::STrih_indexed_shl_cPt_V4: + case Hexagon::STrih_indexed_shl_cdnPt_V4: + case Hexagon::STrih_indexed_shl_cNotPt_V4: + case Hexagon::STrih_indexed_shl_cdnNotPt_V4: + case Hexagon::POST_SThri_cPt: + case Hexagon::POST_SThri_cdnPt_V4: + case Hexagon::POST_SThri_cNotPt: + case Hexagon::POST_SThri_cdnNotPt_V4: + case Hexagon::STh_GP_cPt_V4: + case Hexagon::STh_GP_cNotPt_V4: + case Hexagon::STh_GP_cdnPt_V4: + case Hexagon::STh_GP_cdnNotPt_V4: + case Hexagon::STrih_GP_cPt_V4: + case Hexagon::STrih_GP_cNotPt_V4: + case Hexagon::STrih_GP_cdnPt_V4: + case Hexagon::STrih_GP_cdnNotPt_V4: + + // store word + case Hexagon::STriw: + case Hexagon::STriw_indexed: + case Hexagon::STriw_indexed_shl_V4: + case Hexagon::STriw_shl_V4: + case Hexagon::STriw_GP_V4: + case Hexagon::STw_GP_V4: + case Hexagon::POST_STwri: + case Hexagon::STriw_cPt: + case Hexagon::STriw_cdnPt_V4: + case Hexagon::STriw_cNotPt: + case Hexagon::STriw_cdnNotPt_V4: + case Hexagon::STriw_indexed_cPt: + case Hexagon::STriw_indexed_cdnPt_V4: + case Hexagon::STriw_indexed_cNotPt: + case Hexagon::STriw_indexed_cdnNotPt_V4: + case Hexagon::STriw_indexed_shl_cPt_V4: + case Hexagon::STriw_indexed_shl_cdnPt_V4: + case Hexagon::STriw_indexed_shl_cNotPt_V4: + case Hexagon::STriw_indexed_shl_cdnNotPt_V4: + case Hexagon::POST_STwri_cPt: + case Hexagon::POST_STwri_cdnPt_V4: + case Hexagon::POST_STwri_cNotPt: + case Hexagon::POST_STwri_cdnNotPt_V4: + case Hexagon::STw_GP_cPt_V4: + case Hexagon::STw_GP_cNotPt_V4: + case Hexagon::STw_GP_cdnPt_V4: + case Hexagon::STw_GP_cdnNotPt_V4: + case Hexagon::STriw_GP_cPt_V4: + case Hexagon::STriw_GP_cNotPt_V4: + case Hexagon::STriw_GP_cdnPt_V4: + case Hexagon::STriw_GP_cdnNotPt_V4: + return QRI->Subtarget.hasV4TOps(); + } + return false; +} + +static bool IsLoopN(MachineInstr *MI) { + return (MI->getOpcode() == Hexagon::LOOP0_i || + MI->getOpcode() == Hexagon::LOOP0_r); +} + +/// DoesModifyCalleeSavedReg - Returns true if the instruction modifies a +/// callee-saved register. +static bool DoesModifyCalleeSavedReg(MachineInstr *MI, + const TargetRegisterInfo *TRI) { + for (const uint16_t *CSR = TRI->getCalleeSavedRegs(); *CSR; ++CSR) { + unsigned CalleeSavedReg = *CSR; + if (MI->modifiesRegister(CalleeSavedReg, TRI)) + return true; + } + return false; +} + +// Return the new value instruction for a given store. +static int GetDotNewOp(const int opc) { + switch (opc) { + + // store new value byte + case Hexagon::STrib: + return Hexagon::STrib_nv_V4; + + case Hexagon::STrib_indexed: + return Hexagon::STrib_indexed_nv_V4; + + case Hexagon::STrib_indexed_shl_V4: + return Hexagon::STrib_indexed_shl_nv_V4; + + case Hexagon::STrib_shl_V4: + return Hexagon::STrib_shl_nv_V4; + + case Hexagon::STrib_GP_V4: + return Hexagon::STrib_GP_nv_V4; + + case Hexagon::STb_GP_V4: + return Hexagon::STb_GP_nv_V4; + + case Hexagon::POST_STbri: + return Hexagon::POST_STbri_nv_V4; + + case Hexagon::STrib_cPt: + return Hexagon::STrib_cPt_nv_V4; + + case Hexagon::STrib_cdnPt_V4: + return Hexagon::STrib_cdnPt_nv_V4; + + case Hexagon::STrib_cNotPt: + return Hexagon::STrib_cNotPt_nv_V4; + + case Hexagon::STrib_cdnNotPt_V4: + return Hexagon::STrib_cdnNotPt_nv_V4; + + case Hexagon::STrib_indexed_cPt: + return Hexagon::STrib_indexed_cPt_nv_V4; + + case Hexagon::STrib_indexed_cdnPt_V4: + return Hexagon::STrib_indexed_cdnPt_nv_V4; + + case Hexagon::STrib_indexed_cNotPt: + return Hexagon::STrib_indexed_cNotPt_nv_V4; + + case Hexagon::STrib_indexed_cdnNotPt_V4: + return Hexagon::STrib_indexed_cdnNotPt_nv_V4; + + case Hexagon::STrib_indexed_shl_cPt_V4: + return Hexagon::STrib_indexed_shl_cPt_nv_V4; + + case Hexagon::STrib_indexed_shl_cdnPt_V4: + return Hexagon::STrib_indexed_shl_cdnPt_nv_V4; + + case Hexagon::STrib_indexed_shl_cNotPt_V4: + return Hexagon::STrib_indexed_shl_cNotPt_nv_V4; + + case Hexagon::STrib_indexed_shl_cdnNotPt_V4: + return Hexagon::STrib_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_STbri_cPt: + return Hexagon::POST_STbri_cPt_nv_V4; + + case Hexagon::POST_STbri_cdnPt_V4: + return Hexagon::POST_STbri_cdnPt_nv_V4; + + case Hexagon::POST_STbri_cNotPt: + return Hexagon::POST_STbri_cNotPt_nv_V4; + + case Hexagon::POST_STbri_cdnNotPt_V4: + return Hexagon::POST_STbri_cdnNotPt_nv_V4; + + case Hexagon::STb_GP_cPt_V4: + return Hexagon::STb_GP_cPt_nv_V4; + + case Hexagon::STb_GP_cNotPt_V4: + return Hexagon::STb_GP_cNotPt_nv_V4; + + case Hexagon::STb_GP_cdnPt_V4: + return Hexagon::STb_GP_cdnPt_nv_V4; + + case Hexagon::STb_GP_cdnNotPt_V4: + return Hexagon::STb_GP_cdnNotPt_nv_V4; + + case Hexagon::STrib_GP_cPt_V4: + return Hexagon::STrib_GP_cPt_nv_V4; + + case Hexagon::STrib_GP_cNotPt_V4: + return Hexagon::STrib_GP_cNotPt_nv_V4; + + case Hexagon::STrib_GP_cdnPt_V4: + return Hexagon::STrib_GP_cdnPt_nv_V4; + + case Hexagon::STrib_GP_cdnNotPt_V4: + return Hexagon::STrib_GP_cdnNotPt_nv_V4; + + // store new value halfword + case Hexagon::STrih: + return Hexagon::STrih_nv_V4; + + case Hexagon::STrih_indexed: + return Hexagon::STrih_indexed_nv_V4; + + case Hexagon::STrih_indexed_shl_V4: + return Hexagon::STrih_indexed_shl_nv_V4; + + case Hexagon::STrih_shl_V4: + return Hexagon::STrih_shl_nv_V4; + + case Hexagon::STrih_GP_V4: + return Hexagon::STrih_GP_nv_V4; + + case Hexagon::STh_GP_V4: + return Hexagon::STh_GP_nv_V4; + + case Hexagon::POST_SThri: + return Hexagon::POST_SThri_nv_V4; + + case Hexagon::STrih_cPt: + return Hexagon::STrih_cPt_nv_V4; + + case Hexagon::STrih_cdnPt_V4: + return Hexagon::STrih_cdnPt_nv_V4; + + case Hexagon::STrih_cNotPt: + return Hexagon::STrih_cNotPt_nv_V4; + + case Hexagon::STrih_cdnNotPt_V4: + return Hexagon::STrih_cdnNotPt_nv_V4; + + case Hexagon::STrih_indexed_cPt: + return Hexagon::STrih_indexed_cPt_nv_V4; + + case Hexagon::STrih_indexed_cdnPt_V4: + return Hexagon::STrih_indexed_cdnPt_nv_V4; + + case Hexagon::STrih_indexed_cNotPt: + return Hexagon::STrih_indexed_cNotPt_nv_V4; + + case Hexagon::STrih_indexed_cdnNotPt_V4: + return Hexagon::STrih_indexed_cdnNotPt_nv_V4; + + case Hexagon::STrih_indexed_shl_cPt_V4: + return Hexagon::STrih_indexed_shl_cPt_nv_V4; + + case Hexagon::STrih_indexed_shl_cdnPt_V4: + return Hexagon::STrih_indexed_shl_cdnPt_nv_V4; + + case Hexagon::STrih_indexed_shl_cNotPt_V4: + return Hexagon::STrih_indexed_shl_cNotPt_nv_V4; + + case Hexagon::STrih_indexed_shl_cdnNotPt_V4: + return Hexagon::STrih_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_SThri_cPt: + return Hexagon::POST_SThri_cPt_nv_V4; + + case Hexagon::POST_SThri_cdnPt_V4: + return Hexagon::POST_SThri_cdnPt_nv_V4; + + case Hexagon::POST_SThri_cNotPt: + return Hexagon::POST_SThri_cNotPt_nv_V4; + + case Hexagon::POST_SThri_cdnNotPt_V4: + return Hexagon::POST_SThri_cdnNotPt_nv_V4; + + case Hexagon::STh_GP_cPt_V4: + return Hexagon::STh_GP_cPt_nv_V4; + + case Hexagon::STh_GP_cNotPt_V4: + return Hexagon::STh_GP_cNotPt_nv_V4; + + case Hexagon::STh_GP_cdnPt_V4: + return Hexagon::STh_GP_cdnPt_nv_V4; + + case Hexagon::STh_GP_cdnNotPt_V4: + return Hexagon::STh_GP_cdnNotPt_nv_V4; + + case Hexagon::STrih_GP_cPt_V4: + return Hexagon::STrih_GP_cPt_nv_V4; + + case Hexagon::STrih_GP_cNotPt_V4: + return Hexagon::STrih_GP_cNotPt_nv_V4; + + case Hexagon::STrih_GP_cdnPt_V4: + return Hexagon::STrih_GP_cdnPt_nv_V4; + + case Hexagon::STrih_GP_cdnNotPt_V4: + return Hexagon::STrih_GP_cdnNotPt_nv_V4; + + // store new value word + case Hexagon::STriw: + return Hexagon::STriw_nv_V4; + + case Hexagon::STriw_indexed: + return Hexagon::STriw_indexed_nv_V4; + + case Hexagon::STriw_indexed_shl_V4: + return Hexagon::STriw_indexed_shl_nv_V4; + + case Hexagon::STriw_shl_V4: + return Hexagon::STriw_shl_nv_V4; + + case Hexagon::STriw_GP_V4: + return Hexagon::STriw_GP_nv_V4; + + case Hexagon::STw_GP_V4: + return Hexagon::STw_GP_nv_V4; + + case Hexagon::POST_STwri: + return Hexagon::POST_STwri_nv_V4; + + case Hexagon::STriw_cPt: + return Hexagon::STriw_cPt_nv_V4; + + case Hexagon::STriw_cdnPt_V4: + return Hexagon::STriw_cdnPt_nv_V4; + + case Hexagon::STriw_cNotPt: + return Hexagon::STriw_cNotPt_nv_V4; + + case Hexagon::STriw_cdnNotPt_V4: + return Hexagon::STriw_cdnNotPt_nv_V4; + + case Hexagon::STriw_indexed_cPt: + return Hexagon::STriw_indexed_cPt_nv_V4; + + case Hexagon::STriw_indexed_cdnPt_V4: + return Hexagon::STriw_indexed_cdnPt_nv_V4; + + case Hexagon::STriw_indexed_cNotPt: + return Hexagon::STriw_indexed_cNotPt_nv_V4; + + case Hexagon::STriw_indexed_cdnNotPt_V4: + return Hexagon::STriw_indexed_cdnNotPt_nv_V4; + + case Hexagon::STriw_indexed_shl_cPt_V4: + return Hexagon::STriw_indexed_shl_cPt_nv_V4; + + case Hexagon::STriw_indexed_shl_cdnPt_V4: + return Hexagon::STriw_indexed_shl_cdnPt_nv_V4; + + case Hexagon::STriw_indexed_shl_cNotPt_V4: + return Hexagon::STriw_indexed_shl_cNotPt_nv_V4; + + case Hexagon::STriw_indexed_shl_cdnNotPt_V4: + return Hexagon::STriw_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_STwri_cPt: + return Hexagon::POST_STwri_cPt_nv_V4; + + case Hexagon::POST_STwri_cdnPt_V4: + return Hexagon::POST_STwri_cdnPt_nv_V4; + + case Hexagon::POST_STwri_cNotPt: + return Hexagon::POST_STwri_cNotPt_nv_V4; + + case Hexagon::POST_STwri_cdnNotPt_V4: + return Hexagon::POST_STwri_cdnNotPt_nv_V4; + + case Hexagon::STw_GP_cPt_V4: + return Hexagon::STw_GP_cPt_nv_V4; + + case Hexagon::STw_GP_cNotPt_V4: + return Hexagon::STw_GP_cNotPt_nv_V4; + + case Hexagon::STw_GP_cdnPt_V4: + return Hexagon::STw_GP_cdnPt_nv_V4; + + case Hexagon::STw_GP_cdnNotPt_V4: + return Hexagon::STw_GP_cdnNotPt_nv_V4; + + case Hexagon::STriw_GP_cPt_V4: + return Hexagon::STriw_GP_cPt_nv_V4; + + case Hexagon::STriw_GP_cNotPt_V4: + return Hexagon::STriw_GP_cNotPt_nv_V4; + + case Hexagon::STriw_GP_cdnPt_V4: + return Hexagon::STriw_GP_cdnPt_nv_V4; + + case Hexagon::STriw_GP_cdnNotPt_V4: + return Hexagon::STriw_GP_cdnNotPt_nv_V4; + + default: + assert(0 && "Unknown .new type"); + } + return 0; +} + +// Return .new predicate version for an instruction +static int GetDotNewPredOp(const int opc) { + switch (opc) { + // Conditional stores + // Store byte conditionally + case Hexagon::STrib_cPt : + return Hexagon::STrib_cdnPt_V4; + + case Hexagon::STrib_cNotPt : + return Hexagon::STrib_cdnNotPt_V4; + + case Hexagon::STrib_indexed_cPt : + return Hexagon::STrib_indexed_cdnPt_V4; + + case Hexagon::STrib_indexed_cNotPt : + return Hexagon::STrib_indexed_cdnNotPt_V4; + + case Hexagon::STrib_imm_cPt_V4 : + return Hexagon::STrib_imm_cdnPt_V4; + + case Hexagon::STrib_imm_cNotPt_V4 : + return Hexagon::STrib_imm_cdnNotPt_V4; + + case Hexagon::POST_STbri_cPt : + return Hexagon::POST_STbri_cdnPt_V4; + + case Hexagon::POST_STbri_cNotPt : + return Hexagon::POST_STbri_cdnNotPt_V4; + + case Hexagon::STrib_indexed_shl_cPt_V4 : + return Hexagon::STrib_indexed_shl_cdnPt_V4; + + case Hexagon::STrib_indexed_shl_cNotPt_V4 : + return Hexagon::STrib_indexed_shl_cdnNotPt_V4; + + case Hexagon::STb_GP_cPt_V4 : + return Hexagon::STb_GP_cdnPt_V4; + + case Hexagon::STb_GP_cNotPt_V4 : + return Hexagon::STb_GP_cdnNotPt_V4; + + case Hexagon::STrib_GP_cPt_V4 : + return Hexagon::STrib_GP_cdnPt_V4; + + case Hexagon::STrib_GP_cNotPt_V4 : + return Hexagon::STrib_GP_cdnNotPt_V4; + + // Store doubleword conditionally + case Hexagon::STrid_cPt : + return Hexagon::STrid_cdnPt_V4; + + case Hexagon::STrid_cNotPt : + return Hexagon::STrid_cdnNotPt_V4; + + case Hexagon::STrid_indexed_cPt : + return Hexagon::STrid_indexed_cdnPt_V4; + + case Hexagon::STrid_indexed_cNotPt : + return Hexagon::STrid_indexed_cdnNotPt_V4; + + case Hexagon::STrid_indexed_shl_cPt_V4 : + return Hexagon::STrid_indexed_shl_cdnPt_V4; + + case Hexagon::STrid_indexed_shl_cNotPt_V4 : + return Hexagon::STrid_indexed_shl_cdnNotPt_V4; + + case Hexagon::POST_STdri_cPt : + return Hexagon::POST_STdri_cdnPt_V4; + + case Hexagon::POST_STdri_cNotPt : + return Hexagon::POST_STdri_cdnNotPt_V4; + + case Hexagon::STd_GP_cPt_V4 : + return Hexagon::STd_GP_cdnPt_V4; + + case Hexagon::STd_GP_cNotPt_V4 : + return Hexagon::STd_GP_cdnNotPt_V4; + + case Hexagon::STrid_GP_cPt_V4 : + return Hexagon::STrid_GP_cdnPt_V4; + + case Hexagon::STrid_GP_cNotPt_V4 : + return Hexagon::STrid_GP_cdnNotPt_V4; + + // Store halfword conditionally + case Hexagon::STrih_cPt : + return Hexagon::STrih_cdnPt_V4; + + case Hexagon::STrih_cNotPt : + return Hexagon::STrih_cdnNotPt_V4; + + case Hexagon::STrih_indexed_cPt : + return Hexagon::STrih_indexed_cdnPt_V4; + + case Hexagon::STrih_indexed_cNotPt : + return Hexagon::STrih_indexed_cdnNotPt_V4; + + case Hexagon::STrih_imm_cPt_V4 : + return Hexagon::STrih_imm_cdnPt_V4; + + case Hexagon::STrih_imm_cNotPt_V4 : + return Hexagon::STrih_imm_cdnNotPt_V4; + + case Hexagon::STrih_indexed_shl_cPt_V4 : + return Hexagon::STrih_indexed_shl_cdnPt_V4; + + case Hexagon::STrih_indexed_shl_cNotPt_V4 : + return Hexagon::STrih_indexed_shl_cdnNotPt_V4; + + case Hexagon::POST_SThri_cPt : + return Hexagon::POST_SThri_cdnPt_V4; + + case Hexagon::POST_SThri_cNotPt : + return Hexagon::POST_SThri_cdnNotPt_V4; + + case Hexagon::STh_GP_cPt_V4 : + return Hexagon::STh_GP_cdnPt_V4; + + case Hexagon::STh_GP_cNotPt_V4 : + return Hexagon::STh_GP_cdnNotPt_V4; + + case Hexagon::STrih_GP_cPt_V4 : + return Hexagon::STrih_GP_cdnPt_V4; + + case Hexagon::STrih_GP_cNotPt_V4 : + return Hexagon::STrih_GP_cdnNotPt_V4; + + // Store word conditionally + case Hexagon::STriw_cPt : + return Hexagon::STriw_cdnPt_V4; + + case Hexagon::STriw_cNotPt : + return Hexagon::STriw_cdnNotPt_V4; + + case Hexagon::STriw_indexed_cPt : + return Hexagon::STriw_indexed_cdnPt_V4; + + case Hexagon::STriw_indexed_cNotPt : + return Hexagon::STriw_indexed_cdnNotPt_V4; + + case Hexagon::STriw_imm_cPt_V4 : + return Hexagon::STriw_imm_cdnPt_V4; + + case Hexagon::STriw_imm_cNotPt_V4 : + return Hexagon::STriw_imm_cdnNotPt_V4; + + case Hexagon::STriw_indexed_shl_cPt_V4 : + return Hexagon::STriw_indexed_shl_cdnPt_V4; + + case Hexagon::STriw_indexed_shl_cNotPt_V4 : + return Hexagon::STriw_indexed_shl_cdnNotPt_V4; + + case Hexagon::POST_STwri_cPt : + return Hexagon::POST_STwri_cdnPt_V4; + + case Hexagon::POST_STwri_cNotPt : + return Hexagon::POST_STwri_cdnNotPt_V4; + + case Hexagon::STw_GP_cPt_V4 : + return Hexagon::STw_GP_cdnPt_V4; + + case Hexagon::STw_GP_cNotPt_V4 : + return Hexagon::STw_GP_cdnNotPt_V4; + + case Hexagon::STriw_GP_cPt_V4 : + return Hexagon::STriw_GP_cdnPt_V4; + + case Hexagon::STriw_GP_cNotPt_V4 : + return Hexagon::STriw_GP_cdnNotPt_V4; + + // Condtional Jumps + case Hexagon::JMP_c: + return Hexagon::JMP_cdnPt; + + case Hexagon::JMP_cNot: + return Hexagon::JMP_cdnNotPt; + + case Hexagon::JMPR_cPt: + return Hexagon::JMPR_cdnPt_V3; + + case Hexagon::JMPR_cNotPt: + return Hexagon::JMPR_cdnNotPt_V3; + + // Conditional Transfers + case Hexagon::TFR_cPt: + return Hexagon::TFR_cdnPt; + + case Hexagon::TFR_cNotPt: + return Hexagon::TFR_cdnNotPt; + + case Hexagon::TFRI_cPt: + return Hexagon::TFRI_cdnPt; + + case Hexagon::TFRI_cNotPt: + return Hexagon::TFRI_cdnNotPt; + + // Load double word + case Hexagon::LDrid_cPt : + return Hexagon::LDrid_cdnPt; + + case Hexagon::LDrid_cNotPt : + return Hexagon::LDrid_cdnNotPt; + + case Hexagon::LDrid_indexed_cPt : + return Hexagon::LDrid_indexed_cdnPt; + + case Hexagon::LDrid_indexed_cNotPt : + return Hexagon::LDrid_indexed_cdnNotPt; + + case Hexagon::POST_LDrid_cPt : + return Hexagon::POST_LDrid_cdnPt_V4; + + case Hexagon::POST_LDrid_cNotPt : + return Hexagon::POST_LDrid_cdnNotPt_V4; + + // Load word + case Hexagon::LDriw_cPt : + return Hexagon::LDriw_cdnPt; + + case Hexagon::LDriw_cNotPt : + return Hexagon::LDriw_cdnNotPt; + + case Hexagon::LDriw_indexed_cPt : + return Hexagon::LDriw_indexed_cdnPt; + + case Hexagon::LDriw_indexed_cNotPt : + return Hexagon::LDriw_indexed_cdnNotPt; + + case Hexagon::POST_LDriw_cPt : + return Hexagon::POST_LDriw_cdnPt_V4; + + case Hexagon::POST_LDriw_cNotPt : + return Hexagon::POST_LDriw_cdnNotPt_V4; + + // Load halfword + case Hexagon::LDrih_cPt : + return Hexagon::LDrih_cdnPt; + + case Hexagon::LDrih_cNotPt : + return Hexagon::LDrih_cdnNotPt; + + case Hexagon::LDrih_indexed_cPt : + return Hexagon::LDrih_indexed_cdnPt; + + case Hexagon::LDrih_indexed_cNotPt : + return Hexagon::LDrih_indexed_cdnNotPt; + + case Hexagon::POST_LDrih_cPt : + return Hexagon::POST_LDrih_cdnPt_V4; + + case Hexagon::POST_LDrih_cNotPt : + return Hexagon::POST_LDrih_cdnNotPt_V4; + + // Load byte + case Hexagon::LDrib_cPt : + return Hexagon::LDrib_cdnPt; + + case Hexagon::LDrib_cNotPt : + return Hexagon::LDrib_cdnNotPt; + + case Hexagon::LDrib_indexed_cPt : + return Hexagon::LDrib_indexed_cdnPt; + + case Hexagon::LDrib_indexed_cNotPt : + return Hexagon::LDrib_indexed_cdnNotPt; + + case Hexagon::POST_LDrib_cPt : + return Hexagon::POST_LDrib_cdnPt_V4; + + case Hexagon::POST_LDrib_cNotPt : + return Hexagon::POST_LDrib_cdnNotPt_V4; + + // Load unsigned halfword + case Hexagon::LDriuh_cPt : + return Hexagon::LDriuh_cdnPt; + + case Hexagon::LDriuh_cNotPt : + return Hexagon::LDriuh_cdnNotPt; + + case Hexagon::LDriuh_indexed_cPt : + return Hexagon::LDriuh_indexed_cdnPt; + + case Hexagon::LDriuh_indexed_cNotPt : + return Hexagon::LDriuh_indexed_cdnNotPt; + + case Hexagon::POST_LDriuh_cPt : + return Hexagon::POST_LDriuh_cdnPt_V4; + + case Hexagon::POST_LDriuh_cNotPt : + return Hexagon::POST_LDriuh_cdnNotPt_V4; + + // Load unsigned byte + case Hexagon::LDriub_cPt : + return Hexagon::LDriub_cdnPt; + + case Hexagon::LDriub_cNotPt : + return Hexagon::LDriub_cdnNotPt; + + case Hexagon::LDriub_indexed_cPt : + return Hexagon::LDriub_indexed_cdnPt; + + case Hexagon::LDriub_indexed_cNotPt : + return Hexagon::LDriub_indexed_cdnNotPt; + + case Hexagon::POST_LDriub_cPt : + return Hexagon::POST_LDriub_cdnPt_V4; + + case Hexagon::POST_LDriub_cNotPt : + return Hexagon::POST_LDriub_cdnNotPt_V4; + + // V4 indexed+scaled load + + case Hexagon::LDrid_indexed_cPt_V4 : + return Hexagon::LDrid_indexed_cdnPt_V4; + + case Hexagon::LDrid_indexed_cNotPt_V4 : + return Hexagon::LDrid_indexed_cdnNotPt_V4; + + case Hexagon::LDrid_indexed_shl_cPt_V4 : + return Hexagon::LDrid_indexed_shl_cdnPt_V4; + + case Hexagon::LDrid_indexed_shl_cNotPt_V4 : + return Hexagon::LDrid_indexed_shl_cdnNotPt_V4; + + case Hexagon::LDrib_indexed_cPt_V4 : + return Hexagon::LDrib_indexed_cdnPt_V4; + + case Hexagon::LDrib_indexed_cNotPt_V4 : + return Hexagon::LDrib_indexed_cdnNotPt_V4; + + case Hexagon::LDrib_indexed_shl_cPt_V4 : + return Hexagon::LDrib_indexed_shl_cdnPt_V4; + + case Hexagon::LDrib_indexed_shl_cNotPt_V4 : + return Hexagon::LDrib_indexed_shl_cdnNotPt_V4; + + case Hexagon::LDriub_indexed_cPt_V4 : + return Hexagon::LDriub_indexed_cdnPt_V4; + + case Hexagon::LDriub_indexed_cNotPt_V4 : + return Hexagon::LDriub_indexed_cdnNotPt_V4; + + case Hexagon::LDriub_indexed_shl_cPt_V4 : + return Hexagon::LDriub_indexed_shl_cdnPt_V4; + + case Hexagon::LDriub_indexed_shl_cNotPt_V4 : + return Hexagon::LDriub_indexed_shl_cdnNotPt_V4; + + case Hexagon::LDrih_indexed_cPt_V4 : + return Hexagon::LDrih_indexed_cdnPt_V4; + + case Hexagon::LDrih_indexed_cNotPt_V4 : + return Hexagon::LDrih_indexed_cdnNotPt_V4; + + case Hexagon::LDrih_indexed_shl_cPt_V4 : + return Hexagon::LDrih_indexed_shl_cdnPt_V4; + + case Hexagon::LDrih_indexed_shl_cNotPt_V4 : + return Hexagon::LDrih_indexed_shl_cdnNotPt_V4; + + case Hexagon::LDriuh_indexed_cPt_V4 : + return Hexagon::LDriuh_indexed_cdnPt_V4; + + case Hexagon::LDriuh_indexed_cNotPt_V4 : + return Hexagon::LDriuh_indexed_cdnNotPt_V4; + + case Hexagon::LDriuh_indexed_shl_cPt_V4 : + return Hexagon::LDriuh_indexed_shl_cdnPt_V4; + + case Hexagon::LDriuh_indexed_shl_cNotPt_V4 : + return Hexagon::LDriuh_indexed_shl_cdnNotPt_V4; + + case Hexagon::LDriw_indexed_cPt_V4 : + return Hexagon::LDriw_indexed_cdnPt_V4; + + case Hexagon::LDriw_indexed_cNotPt_V4 : + return Hexagon::LDriw_indexed_cdnNotPt_V4; + + case Hexagon::LDriw_indexed_shl_cPt_V4 : + return Hexagon::LDriw_indexed_shl_cdnPt_V4; + + case Hexagon::LDriw_indexed_shl_cNotPt_V4 : + return Hexagon::LDriw_indexed_shl_cdnNotPt_V4; + + // V4 global address load + + case Hexagon::LDd_GP_cPt_V4: + return Hexagon::LDd_GP_cdnPt_V4; + + case Hexagon::LDd_GP_cNotPt_V4: + return Hexagon::LDd_GP_cdnNotPt_V4; + + case Hexagon::LDb_GP_cPt_V4: + return Hexagon::LDb_GP_cdnPt_V4; + + case Hexagon::LDb_GP_cNotPt_V4: + return Hexagon::LDb_GP_cdnNotPt_V4; + + case Hexagon::LDub_GP_cPt_V4: + return Hexagon::LDub_GP_cdnPt_V4; + + case Hexagon::LDub_GP_cNotPt_V4: + return Hexagon::LDub_GP_cdnNotPt_V4; + + case Hexagon::LDh_GP_cPt_V4: + return Hexagon::LDh_GP_cdnPt_V4; + + case Hexagon::LDh_GP_cNotPt_V4: + return Hexagon::LDh_GP_cdnNotPt_V4; + + case Hexagon::LDuh_GP_cPt_V4: + return Hexagon::LDuh_GP_cdnPt_V4; + + case Hexagon::LDuh_GP_cNotPt_V4: + return Hexagon::LDuh_GP_cdnNotPt_V4; + + case Hexagon::LDw_GP_cPt_V4: + return Hexagon::LDw_GP_cdnPt_V4; + + case Hexagon::LDw_GP_cNotPt_V4: + return Hexagon::LDw_GP_cdnNotPt_V4; + + case Hexagon::LDrid_GP_cPt_V4: + return Hexagon::LDrid_GP_cdnPt_V4; + + case Hexagon::LDrid_GP_cNotPt_V4: + return Hexagon::LDrid_GP_cdnNotPt_V4; + + case Hexagon::LDrib_GP_cPt_V4: + return Hexagon::LDrib_GP_cdnPt_V4; + + case Hexagon::LDrib_GP_cNotPt_V4: + return Hexagon::LDrib_GP_cdnNotPt_V4; + + case Hexagon::LDriub_GP_cPt_V4: + return Hexagon::LDriub_GP_cdnPt_V4; + + case Hexagon::LDriub_GP_cNotPt_V4: + return Hexagon::LDriub_GP_cdnNotPt_V4; + + case Hexagon::LDrih_GP_cPt_V4: + return Hexagon::LDrih_GP_cdnPt_V4; + + case Hexagon::LDrih_GP_cNotPt_V4: + return Hexagon::LDrih_GP_cdnNotPt_V4; + + case Hexagon::LDriuh_GP_cPt_V4: + return Hexagon::LDriuh_GP_cdnPt_V4; + + case Hexagon::LDriuh_GP_cNotPt_V4: + return Hexagon::LDriuh_GP_cdnNotPt_V4; + + case Hexagon::LDriw_GP_cPt_V4: + return Hexagon::LDriw_GP_cdnPt_V4; + + case Hexagon::LDriw_GP_cNotPt_V4: + return Hexagon::LDriw_GP_cdnNotPt_V4; + + // Conditional store new-value byte + case Hexagon::STrib_cPt_nv_V4 : + return Hexagon::STrib_cdnPt_nv_V4; + case Hexagon::STrib_cNotPt_nv_V4 : + return Hexagon::STrib_cdnNotPt_nv_V4; + + case Hexagon::STrib_indexed_cPt_nv_V4 : + return Hexagon::STrib_indexed_cdnPt_nv_V4; + case Hexagon::STrib_indexed_cNotPt_nv_V4 : + return Hexagon::STrib_indexed_cdnNotPt_nv_V4; + + case Hexagon::STrib_indexed_shl_cPt_nv_V4 : + return Hexagon::STrib_indexed_shl_cdnPt_nv_V4; + case Hexagon::STrib_indexed_shl_cNotPt_nv_V4 : + return Hexagon::STrib_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_STbri_cPt_nv_V4 : + return Hexagon::POST_STbri_cdnPt_nv_V4; + case Hexagon::POST_STbri_cNotPt_nv_V4 : + return Hexagon::POST_STbri_cdnNotPt_nv_V4; + + case Hexagon::STb_GP_cPt_nv_V4 : + return Hexagon::STb_GP_cdnPt_nv_V4; + + case Hexagon::STb_GP_cNotPt_nv_V4 : + return Hexagon::STb_GP_cdnNotPt_nv_V4; + + case Hexagon::STrib_GP_cPt_nv_V4 : + return Hexagon::STrib_GP_cdnPt_nv_V4; + + case Hexagon::STrib_GP_cNotPt_nv_V4 : + return Hexagon::STrib_GP_cdnNotPt_nv_V4; + + // Conditional store new-value halfword + case Hexagon::STrih_cPt_nv_V4 : + return Hexagon::STrih_cdnPt_nv_V4; + case Hexagon::STrih_cNotPt_nv_V4 : + return Hexagon::STrih_cdnNotPt_nv_V4; + + case Hexagon::STrih_indexed_cPt_nv_V4 : + return Hexagon::STrih_indexed_cdnPt_nv_V4; + case Hexagon::STrih_indexed_cNotPt_nv_V4 : + return Hexagon::STrih_indexed_cdnNotPt_nv_V4; + + case Hexagon::STrih_indexed_shl_cPt_nv_V4 : + return Hexagon::STrih_indexed_shl_cdnPt_nv_V4; + case Hexagon::STrih_indexed_shl_cNotPt_nv_V4 : + return Hexagon::STrih_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_SThri_cPt_nv_V4 : + return Hexagon::POST_SThri_cdnPt_nv_V4; + case Hexagon::POST_SThri_cNotPt_nv_V4 : + return Hexagon::POST_SThri_cdnNotPt_nv_V4; + + case Hexagon::STh_GP_cPt_nv_V4 : + return Hexagon::STh_GP_cdnPt_nv_V4; + + case Hexagon::STh_GP_cNotPt_nv_V4 : + return Hexagon::STh_GP_cdnNotPt_nv_V4; + + case Hexagon::STrih_GP_cPt_nv_V4 : + return Hexagon::STrih_GP_cdnPt_nv_V4; + + case Hexagon::STrih_GP_cNotPt_nv_V4 : + return Hexagon::STrih_GP_cdnNotPt_nv_V4; + + // Conditional store new-value word + case Hexagon::STriw_cPt_nv_V4 : + return Hexagon::STriw_cdnPt_nv_V4; + case Hexagon::STriw_cNotPt_nv_V4 : + return Hexagon::STriw_cdnNotPt_nv_V4; + + case Hexagon::STriw_indexed_cPt_nv_V4 : + return Hexagon::STriw_indexed_cdnPt_nv_V4; + case Hexagon::STriw_indexed_cNotPt_nv_V4 : + return Hexagon::STriw_indexed_cdnNotPt_nv_V4; + + case Hexagon::STriw_indexed_shl_cPt_nv_V4 : + return Hexagon::STriw_indexed_shl_cdnPt_nv_V4; + case Hexagon::STriw_indexed_shl_cNotPt_nv_V4 : + return Hexagon::STriw_indexed_shl_cdnNotPt_nv_V4; + + case Hexagon::POST_STwri_cPt_nv_V4 : + return Hexagon::POST_STwri_cdnPt_nv_V4; + case Hexagon::POST_STwri_cNotPt_nv_V4: + return Hexagon::POST_STwri_cdnNotPt_nv_V4; + + case Hexagon::STw_GP_cPt_nv_V4 : + return Hexagon::STw_GP_cdnPt_nv_V4; + + case Hexagon::STw_GP_cNotPt_nv_V4 : + return Hexagon::STw_GP_cdnNotPt_nv_V4; + + case Hexagon::STriw_GP_cPt_nv_V4 : + return Hexagon::STriw_GP_cdnPt_nv_V4; + + case Hexagon::STriw_GP_cNotPt_nv_V4 : + return Hexagon::STriw_GP_cdnNotPt_nv_V4; + + // Conditional add + case Hexagon::ADD_ri_cPt : + return Hexagon::ADD_ri_cdnPt; + case Hexagon::ADD_ri_cNotPt : + return Hexagon::ADD_ri_cdnNotPt; + + case Hexagon::ADD_rr_cPt : + return Hexagon::ADD_rr_cdnPt; + case Hexagon::ADD_rr_cNotPt : + return Hexagon::ADD_rr_cdnNotPt; + + // Conditional logical Operations + case Hexagon::XOR_rr_cPt : + return Hexagon::XOR_rr_cdnPt; + case Hexagon::XOR_rr_cNotPt : + return Hexagon::XOR_rr_cdnNotPt; + + case Hexagon::AND_rr_cPt : + return Hexagon::AND_rr_cdnPt; + case Hexagon::AND_rr_cNotPt : + return Hexagon::AND_rr_cdnNotPt; + + case Hexagon::OR_rr_cPt : + return Hexagon::OR_rr_cdnPt; + case Hexagon::OR_rr_cNotPt : + return Hexagon::OR_rr_cdnNotPt; + + // Conditional Subtract + case Hexagon::SUB_rr_cPt : + return Hexagon::SUB_rr_cdnPt; + case Hexagon::SUB_rr_cNotPt : + return Hexagon::SUB_rr_cdnNotPt; + + // Conditional combine + case Hexagon::COMBINE_rr_cPt : + return Hexagon::COMBINE_rr_cdnPt; + case Hexagon::COMBINE_rr_cNotPt : + return Hexagon::COMBINE_rr_cdnNotPt; + + case Hexagon::ASLH_cPt_V4 : + return Hexagon::ASLH_cdnPt_V4; + case Hexagon::ASLH_cNotPt_V4 : + return Hexagon::ASLH_cdnNotPt_V4; + + case Hexagon::ASRH_cPt_V4 : + return Hexagon::ASRH_cdnPt_V4; + case Hexagon::ASRH_cNotPt_V4 : + return Hexagon::ASRH_cdnNotPt_V4; + + case Hexagon::SXTB_cPt_V4 : + return Hexagon::SXTB_cdnPt_V4; + case Hexagon::SXTB_cNotPt_V4 : + return Hexagon::SXTB_cdnNotPt_V4; + + case Hexagon::SXTH_cPt_V4 : + return Hexagon::SXTH_cdnPt_V4; + case Hexagon::SXTH_cNotPt_V4 : + return Hexagon::SXTH_cdnNotPt_V4; + + case Hexagon::ZXTB_cPt_V4 : + return Hexagon::ZXTB_cdnPt_V4; + case Hexagon::ZXTB_cNotPt_V4 : + return Hexagon::ZXTB_cdnNotPt_V4; + + case Hexagon::ZXTH_cPt_V4 : + return Hexagon::ZXTH_cdnPt_V4; + case Hexagon::ZXTH_cNotPt_V4 : + return Hexagon::ZXTH_cdnNotPt_V4; + + + default: + assert(0 && "Unknown .new type"); + } + return 0; +} + +// Returns true if an instruction can be promoted to .new predicate +// or new-value store. +bool HexagonPacketizerList::isNewifiable(MachineInstr* MI) { + if ( isCondInst(MI) || IsNewifyStore(MI)) + return true; + else + return false; +} + +bool HexagonPacketizerList::isCondInst (MachineInstr* MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + const MCInstrDesc& TID = MI->getDesc(); + // bug 5670: until that is fixed, + // this portion is disabled. + if ( TID.isConditionalBranch() // && !IsRegisterJump(MI)) || + || QII->isConditionalTransfer(MI) + || QII->isConditionalALU32(MI) + || QII->isConditionalLoad(MI) + || QII->isConditionalStore(MI)) { + return true; + } + return false; +} + + +// Promote an instructiont to its .new form. +// At this time, we have already made a call to CanPromoteToDotNew +// and made sure that it can *indeed* be promoted. +bool HexagonPacketizerList::PromoteToDotNew(MachineInstr* MI, + SDep::Kind DepType, MachineBasicBlock::iterator &MII, + const TargetRegisterClass* RC) { + + assert (DepType == SDep::Data); + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + + int NewOpcode; + if (RC == Hexagon::PredRegsRegisterClass) + NewOpcode = GetDotNewPredOp(MI->getOpcode()); + else + NewOpcode = GetDotNewOp(MI->getOpcode()); + MI->setDesc(QII->get(NewOpcode)); + + return true; +} + +// Returns the most basic instruction for the .new predicated instructions and +// new-value stores. +// For example, all of the following instructions will be converted back to the +// same instruction: +// 1) if (p0.new) memw(R0+#0) = R1.new ---> +// 2) if (p0) memw(R0+#0)= R1.new -------> if (p0) memw(R0+#0) = R1 +// 3) if (p0.new) memw(R0+#0) = R1 ---> +// +// To understand the translation of instruction 1 to its original form, consider +// a packet with 3 instructions. +// { p0 = cmp.eq(R0,R1) +// if (p0.new) R2 = add(R3, R4) +// R5 = add (R3, R1) +// } +// if (p0) memw(R5+#0) = R2 <--- trying to include it in the previous packet +// +// This instruction can be part of the previous packet only if both p0 and R2 +// are promoted to .new values. This promotion happens in steps, first +// predicate register is promoted to .new and in the next iteration R2 is +// promoted. Therefore, in case of dependence check failure (due to R5) during +// next iteration, it should be converted back to its most basic form. + +static int GetDotOldOp(const int opc) { + switch (opc) { + case Hexagon::TFR_cdnPt: + return Hexagon::TFR_cPt; + + case Hexagon::TFR_cdnNotPt: + return Hexagon::TFR_cNotPt; + + case Hexagon::TFRI_cdnPt: + return Hexagon::TFRI_cPt; + + case Hexagon::TFRI_cdnNotPt: + return Hexagon::TFRI_cNotPt; + + case Hexagon::JMP_cdnPt: + return Hexagon::JMP_c; + + case Hexagon::JMP_cdnNotPt: + return Hexagon::JMP_cNot; + + case Hexagon::JMPR_cdnPt_V3: + return Hexagon::JMPR_cPt; + + case Hexagon::JMPR_cdnNotPt_V3: + return Hexagon::JMPR_cNotPt; + + // Load double word + + case Hexagon::LDrid_cdnPt : + return Hexagon::LDrid_cPt; + + case Hexagon::LDrid_cdnNotPt : + return Hexagon::LDrid_cNotPt; + + case Hexagon::LDrid_indexed_cdnPt : + return Hexagon::LDrid_indexed_cPt; + + case Hexagon::LDrid_indexed_cdnNotPt : + return Hexagon::LDrid_indexed_cNotPt; + + case Hexagon::POST_LDrid_cdnPt_V4 : + return Hexagon::POST_LDrid_cPt; + + case Hexagon::POST_LDrid_cdnNotPt_V4 : + return Hexagon::POST_LDrid_cNotPt; + + // Load word + + case Hexagon::LDriw_cdnPt : + return Hexagon::LDriw_cPt; + + case Hexagon::LDriw_cdnNotPt : + return Hexagon::LDriw_cNotPt; + + case Hexagon::LDriw_indexed_cdnPt : + return Hexagon::LDriw_indexed_cPt; + + case Hexagon::LDriw_indexed_cdnNotPt : + return Hexagon::LDriw_indexed_cNotPt; + + case Hexagon::POST_LDriw_cdnPt_V4 : + return Hexagon::POST_LDriw_cPt; + + case Hexagon::POST_LDriw_cdnNotPt_V4 : + return Hexagon::POST_LDriw_cNotPt; + + // Load half + + case Hexagon::LDrih_cdnPt : + return Hexagon::LDrih_cPt; + + case Hexagon::LDrih_cdnNotPt : + return Hexagon::LDrih_cNotPt; + + case Hexagon::LDrih_indexed_cdnPt : + return Hexagon::LDrih_indexed_cPt; + + case Hexagon::LDrih_indexed_cdnNotPt : + return Hexagon::LDrih_indexed_cNotPt; + + case Hexagon::POST_LDrih_cdnPt_V4 : + return Hexagon::POST_LDrih_cPt; + + case Hexagon::POST_LDrih_cdnNotPt_V4 : + return Hexagon::POST_LDrih_cNotPt; + + // Load byte + + case Hexagon::LDrib_cdnPt : + return Hexagon::LDrib_cPt; + + case Hexagon::LDrib_cdnNotPt : + return Hexagon::LDrib_cNotPt; + + case Hexagon::LDrib_indexed_cdnPt : + return Hexagon::LDrib_indexed_cPt; + + case Hexagon::LDrib_indexed_cdnNotPt : + return Hexagon::LDrib_indexed_cNotPt; + + case Hexagon::POST_LDrib_cdnPt_V4 : + return Hexagon::POST_LDrib_cPt; + + case Hexagon::POST_LDrib_cdnNotPt_V4 : + return Hexagon::POST_LDrib_cNotPt; + + // Load unsigned half + + case Hexagon::LDriuh_cdnPt : + return Hexagon::LDriuh_cPt; + + case Hexagon::LDriuh_cdnNotPt : + return Hexagon::LDriuh_cNotPt; + + case Hexagon::LDriuh_indexed_cdnPt : + return Hexagon::LDriuh_indexed_cPt; + + case Hexagon::LDriuh_indexed_cdnNotPt : + return Hexagon::LDriuh_indexed_cNotPt; + + case Hexagon::POST_LDriuh_cdnPt_V4 : + return Hexagon::POST_LDriuh_cPt; + + case Hexagon::POST_LDriuh_cdnNotPt_V4 : + return Hexagon::POST_LDriuh_cNotPt; + + // Load unsigned byte + case Hexagon::LDriub_cdnPt : + return Hexagon::LDriub_cPt; + + case Hexagon::LDriub_cdnNotPt : + return Hexagon::LDriub_cNotPt; + + case Hexagon::LDriub_indexed_cdnPt : + return Hexagon::LDriub_indexed_cPt; + + case Hexagon::LDriub_indexed_cdnNotPt : + return Hexagon::LDriub_indexed_cNotPt; + + case Hexagon::POST_LDriub_cdnPt_V4 : + return Hexagon::POST_LDriub_cPt; + + case Hexagon::POST_LDriub_cdnNotPt_V4 : + return Hexagon::POST_LDriub_cNotPt; + + // V4 indexed+scaled Load + + case Hexagon::LDrid_indexed_cdnPt_V4 : + return Hexagon::LDrid_indexed_cPt_V4; + + case Hexagon::LDrid_indexed_cdnNotPt_V4 : + return Hexagon::LDrid_indexed_cNotPt_V4; + + case Hexagon::LDrid_indexed_shl_cdnPt_V4 : + return Hexagon::LDrid_indexed_shl_cPt_V4; + + case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDrid_indexed_shl_cNotPt_V4; + + case Hexagon::LDrib_indexed_cdnPt_V4 : + return Hexagon::LDrib_indexed_cPt_V4; + + case Hexagon::LDrib_indexed_cdnNotPt_V4 : + return Hexagon::LDrib_indexed_cNotPt_V4; + + case Hexagon::LDrib_indexed_shl_cdnPt_V4 : + return Hexagon::LDrib_indexed_shl_cPt_V4; + + case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDrib_indexed_shl_cNotPt_V4; + + case Hexagon::LDriub_indexed_cdnPt_V4 : + return Hexagon::LDriub_indexed_cPt_V4; + + case Hexagon::LDriub_indexed_cdnNotPt_V4 : + return Hexagon::LDriub_indexed_cNotPt_V4; + + case Hexagon::LDriub_indexed_shl_cdnPt_V4 : + return Hexagon::LDriub_indexed_shl_cPt_V4; + + case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDriub_indexed_shl_cNotPt_V4; + + case Hexagon::LDrih_indexed_cdnPt_V4 : + return Hexagon::LDrih_indexed_cPt_V4; + + case Hexagon::LDrih_indexed_cdnNotPt_V4 : + return Hexagon::LDrih_indexed_cNotPt_V4; + + case Hexagon::LDrih_indexed_shl_cdnPt_V4 : + return Hexagon::LDrih_indexed_shl_cPt_V4; + + case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDrih_indexed_shl_cNotPt_V4; + + case Hexagon::LDriuh_indexed_cdnPt_V4 : + return Hexagon::LDriuh_indexed_cPt_V4; + + case Hexagon::LDriuh_indexed_cdnNotPt_V4 : + return Hexagon::LDriuh_indexed_cNotPt_V4; + + case Hexagon::LDriuh_indexed_shl_cdnPt_V4 : + return Hexagon::LDriuh_indexed_shl_cPt_V4; + + case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDriuh_indexed_shl_cNotPt_V4; + + case Hexagon::LDriw_indexed_cdnPt_V4 : + return Hexagon::LDriw_indexed_cPt_V4; + + case Hexagon::LDriw_indexed_cdnNotPt_V4 : + return Hexagon::LDriw_indexed_cNotPt_V4; + + case Hexagon::LDriw_indexed_shl_cdnPt_V4 : + return Hexagon::LDriw_indexed_shl_cPt_V4; + + case Hexagon::LDriw_indexed_shl_cdnNotPt_V4 : + return Hexagon::LDriw_indexed_shl_cNotPt_V4; + + // V4 global address load + + case Hexagon::LDd_GP_cdnPt_V4: + return Hexagon::LDd_GP_cPt_V4; + + case Hexagon::LDd_GP_cdnNotPt_V4: + return Hexagon::LDd_GP_cNotPt_V4; + + case Hexagon::LDb_GP_cdnPt_V4: + return Hexagon::LDb_GP_cPt_V4; + + case Hexagon::LDb_GP_cdnNotPt_V4: + return Hexagon::LDb_GP_cNotPt_V4; + + case Hexagon::LDub_GP_cdnPt_V4: + return Hexagon::LDub_GP_cPt_V4; + + case Hexagon::LDub_GP_cdnNotPt_V4: + return Hexagon::LDub_GP_cNotPt_V4; + + case Hexagon::LDh_GP_cdnPt_V4: + return Hexagon::LDh_GP_cPt_V4; + + case Hexagon::LDh_GP_cdnNotPt_V4: + return Hexagon::LDh_GP_cNotPt_V4; + + case Hexagon::LDuh_GP_cdnPt_V4: + return Hexagon::LDuh_GP_cPt_V4; + + case Hexagon::LDuh_GP_cdnNotPt_V4: + return Hexagon::LDuh_GP_cNotPt_V4; + + case Hexagon::LDw_GP_cdnPt_V4: + return Hexagon::LDw_GP_cPt_V4; + + case Hexagon::LDw_GP_cdnNotPt_V4: + return Hexagon::LDw_GP_cNotPt_V4; + + case Hexagon::LDrid_GP_cdnPt_V4: + return Hexagon::LDrid_GP_cPt_V4; + + case Hexagon::LDrid_GP_cdnNotPt_V4: + return Hexagon::LDrid_GP_cNotPt_V4; + + case Hexagon::LDrib_GP_cdnPt_V4: + return Hexagon::LDrib_GP_cPt_V4; + + case Hexagon::LDrib_GP_cdnNotPt_V4: + return Hexagon::LDrib_GP_cNotPt_V4; + + case Hexagon::LDriub_GP_cdnPt_V4: + return Hexagon::LDriub_GP_cPt_V4; + + case Hexagon::LDriub_GP_cdnNotPt_V4: + return Hexagon::LDriub_GP_cNotPt_V4; + + case Hexagon::LDrih_GP_cdnPt_V4: + return Hexagon::LDrih_GP_cPt_V4; + + case Hexagon::LDrih_GP_cdnNotPt_V4: + return Hexagon::LDrih_GP_cNotPt_V4; + + case Hexagon::LDriuh_GP_cdnPt_V4: + return Hexagon::LDriuh_GP_cPt_V4; + + case Hexagon::LDriuh_GP_cdnNotPt_V4: + return Hexagon::LDriuh_GP_cNotPt_V4; + + case Hexagon::LDriw_GP_cdnPt_V4: + return Hexagon::LDriw_GP_cPt_V4; + + case Hexagon::LDriw_GP_cdnNotPt_V4: + return Hexagon::LDriw_GP_cNotPt_V4; + + // Conditional add + + case Hexagon::ADD_ri_cdnPt : + return Hexagon::ADD_ri_cPt; + case Hexagon::ADD_ri_cdnNotPt : + return Hexagon::ADD_ri_cNotPt; + + case Hexagon::ADD_rr_cdnPt : + return Hexagon::ADD_rr_cPt; + case Hexagon::ADD_rr_cdnNotPt: + return Hexagon::ADD_rr_cNotPt; + + // Conditional logical Operations + + case Hexagon::XOR_rr_cdnPt : + return Hexagon::XOR_rr_cPt; + case Hexagon::XOR_rr_cdnNotPt : + return Hexagon::XOR_rr_cNotPt; + + case Hexagon::AND_rr_cdnPt : + return Hexagon::AND_rr_cPt; + case Hexagon::AND_rr_cdnNotPt : + return Hexagon::AND_rr_cNotPt; + + case Hexagon::OR_rr_cdnPt : + return Hexagon::OR_rr_cPt; + case Hexagon::OR_rr_cdnNotPt : + return Hexagon::OR_rr_cNotPt; + + // Conditional Subtract + + case Hexagon::SUB_rr_cdnPt : + return Hexagon::SUB_rr_cPt; + case Hexagon::SUB_rr_cdnNotPt : + return Hexagon::SUB_rr_cNotPt; + + // Conditional combine + + case Hexagon::COMBINE_rr_cdnPt : + return Hexagon::COMBINE_rr_cPt; + case Hexagon::COMBINE_rr_cdnNotPt : + return Hexagon::COMBINE_rr_cNotPt; + +// Conditional shift operations + + case Hexagon::ASLH_cdnPt_V4 : + return Hexagon::ASLH_cPt_V4; + case Hexagon::ASLH_cdnNotPt_V4 : + return Hexagon::ASLH_cNotPt_V4; + + case Hexagon::ASRH_cdnPt_V4 : + return Hexagon::ASRH_cPt_V4; + case Hexagon::ASRH_cdnNotPt_V4 : + return Hexagon::ASRH_cNotPt_V4; + + case Hexagon::SXTB_cdnPt_V4 : + return Hexagon::SXTB_cPt_V4; + case Hexagon::SXTB_cdnNotPt_V4 : + return Hexagon::SXTB_cNotPt_V4; + + case Hexagon::SXTH_cdnPt_V4 : + return Hexagon::SXTH_cPt_V4; + case Hexagon::SXTH_cdnNotPt_V4 : + return Hexagon::SXTH_cNotPt_V4; + + case Hexagon::ZXTB_cdnPt_V4 : + return Hexagon::ZXTB_cPt_V4; + case Hexagon::ZXTB_cdnNotPt_V4 : + return Hexagon::ZXTB_cNotPt_V4; + + case Hexagon::ZXTH_cdnPt_V4 : + return Hexagon::ZXTH_cPt_V4; + case Hexagon::ZXTH_cdnNotPt_V4 : + return Hexagon::ZXTH_cNotPt_V4; + + // Store byte + + case Hexagon::STrib_imm_cdnPt_V4 : + return Hexagon::STrib_imm_cPt_V4; + + case Hexagon::STrib_imm_cdnNotPt_V4 : + return Hexagon::STrib_imm_cNotPt_V4; + + case Hexagon::STrib_cdnPt_nv_V4 : + case Hexagon::STrib_cPt_nv_V4 : + case Hexagon::STrib_cdnPt_V4 : + return Hexagon::STrib_cPt; + + case Hexagon::STrib_cdnNotPt_nv_V4 : + case Hexagon::STrib_cNotPt_nv_V4 : + case Hexagon::STrib_cdnNotPt_V4 : + return Hexagon::STrib_cNotPt; + + case Hexagon::STrib_indexed_cdnPt_V4 : + case Hexagon::STrib_indexed_cPt_nv_V4 : + case Hexagon::STrib_indexed_cdnPt_nv_V4 : + return Hexagon::STrib_indexed_cPt; + + case Hexagon::STrib_indexed_cdnNotPt_V4 : + case Hexagon::STrib_indexed_cNotPt_nv_V4 : + case Hexagon::STrib_indexed_cdnNotPt_nv_V4 : + return Hexagon::STrib_indexed_cNotPt; + + case Hexagon::STrib_indexed_shl_cdnPt_nv_V4: + case Hexagon::STrib_indexed_shl_cPt_nv_V4 : + case Hexagon::STrib_indexed_shl_cdnPt_V4 : + return Hexagon::STrib_indexed_shl_cPt_V4; + + case Hexagon::STrib_indexed_shl_cdnNotPt_nv_V4: + case Hexagon::STrib_indexed_shl_cNotPt_nv_V4 : + case Hexagon::STrib_indexed_shl_cdnNotPt_V4 : + return Hexagon::STrib_indexed_shl_cNotPt_V4; + + case Hexagon::POST_STbri_cdnPt_nv_V4 : + case Hexagon::POST_STbri_cPt_nv_V4 : + case Hexagon::POST_STbri_cdnPt_V4 : + return Hexagon::POST_STbri_cPt; + + case Hexagon::POST_STbri_cdnNotPt_nv_V4 : + case Hexagon::POST_STbri_cNotPt_nv_V4: + case Hexagon::POST_STbri_cdnNotPt_V4 : + return Hexagon::POST_STbri_cNotPt; + + case Hexagon::STb_GP_cdnPt_nv_V4: + case Hexagon::STb_GP_cdnPt_V4: + case Hexagon::STb_GP_cPt_nv_V4: + return Hexagon::STb_GP_cPt_V4; + + case Hexagon::STb_GP_cdnNotPt_nv_V4: + case Hexagon::STb_GP_cdnNotPt_V4: + case Hexagon::STb_GP_cNotPt_nv_V4: + return Hexagon::STb_GP_cNotPt_V4; + + case Hexagon::STrib_GP_cdnPt_nv_V4: + case Hexagon::STrib_GP_cdnPt_V4: + case Hexagon::STrib_GP_cPt_nv_V4: + return Hexagon::STrib_GP_cPt_V4; + + case Hexagon::STrib_GP_cdnNotPt_nv_V4: + case Hexagon::STrib_GP_cdnNotPt_V4: + case Hexagon::STrib_GP_cNotPt_nv_V4: + return Hexagon::STrib_GP_cNotPt_V4; + + // Store new-value byte - unconditional + case Hexagon::STrib_nv_V4: + return Hexagon::STrib; + + case Hexagon::STrib_indexed_nv_V4: + return Hexagon::STrib_indexed; + + case Hexagon::STrib_indexed_shl_nv_V4: + return Hexagon::STrib_indexed_shl_V4; + + case Hexagon::STrib_shl_nv_V4: + return Hexagon::STrib_shl_V4; + + case Hexagon::STrib_GP_nv_V4: + return Hexagon::STrib_GP_V4; + + case Hexagon::STb_GP_nv_V4: + return Hexagon::STb_GP_V4; + + case Hexagon::POST_STbri_nv_V4: + return Hexagon::POST_STbri; + + // Store halfword + case Hexagon::STrih_imm_cdnPt_V4 : + return Hexagon::STrih_imm_cPt_V4; + + case Hexagon::STrih_imm_cdnNotPt_V4 : + return Hexagon::STrih_imm_cNotPt_V4; + + case Hexagon::STrih_cdnPt_nv_V4 : + case Hexagon::STrih_cPt_nv_V4 : + case Hexagon::STrih_cdnPt_V4 : + return Hexagon::STrih_cPt; + + case Hexagon::STrih_cdnNotPt_nv_V4 : + case Hexagon::STrih_cNotPt_nv_V4 : + case Hexagon::STrih_cdnNotPt_V4 : + return Hexagon::STrih_cNotPt; + + case Hexagon::STrih_indexed_cdnPt_nv_V4: + case Hexagon::STrih_indexed_cPt_nv_V4 : + case Hexagon::STrih_indexed_cdnPt_V4 : + return Hexagon::STrih_indexed_cPt; + + case Hexagon::STrih_indexed_cdnNotPt_nv_V4: + case Hexagon::STrih_indexed_cNotPt_nv_V4 : + case Hexagon::STrih_indexed_cdnNotPt_V4 : + return Hexagon::STrih_indexed_cNotPt; + + case Hexagon::STrih_indexed_shl_cdnPt_nv_V4 : + case Hexagon::STrih_indexed_shl_cPt_nv_V4 : + case Hexagon::STrih_indexed_shl_cdnPt_V4 : + return Hexagon::STrih_indexed_shl_cPt_V4; + + case Hexagon::STrih_indexed_shl_cdnNotPt_nv_V4 : + case Hexagon::STrih_indexed_shl_cNotPt_nv_V4 : + case Hexagon::STrih_indexed_shl_cdnNotPt_V4 : + return Hexagon::STrih_indexed_shl_cNotPt_V4; + + case Hexagon::POST_SThri_cdnPt_nv_V4 : + case Hexagon::POST_SThri_cPt_nv_V4 : + case Hexagon::POST_SThri_cdnPt_V4 : + return Hexagon::POST_SThri_cPt; + + case Hexagon::POST_SThri_cdnNotPt_nv_V4 : + case Hexagon::POST_SThri_cNotPt_nv_V4 : + case Hexagon::POST_SThri_cdnNotPt_V4 : + return Hexagon::POST_SThri_cNotPt; + + case Hexagon::STh_GP_cdnPt_nv_V4: + case Hexagon::STh_GP_cdnPt_V4: + case Hexagon::STh_GP_cPt_nv_V4: + return Hexagon::STh_GP_cPt_V4; + + case Hexagon::STh_GP_cdnNotPt_nv_V4: + case Hexagon::STh_GP_cdnNotPt_V4: + case Hexagon::STh_GP_cNotPt_nv_V4: + return Hexagon::STh_GP_cNotPt_V4; + + case Hexagon::STrih_GP_cdnPt_nv_V4: + case Hexagon::STrih_GP_cdnPt_V4: + case Hexagon::STrih_GP_cPt_nv_V4: + return Hexagon::STrih_GP_cPt_V4; + + case Hexagon::STrih_GP_cdnNotPt_nv_V4: + case Hexagon::STrih_GP_cdnNotPt_V4: + case Hexagon::STrih_GP_cNotPt_nv_V4: + return Hexagon::STrih_GP_cNotPt_V4; + + // Store new-value halfword - unconditional + + case Hexagon::STrih_nv_V4: + return Hexagon::STrih; + + case Hexagon::STrih_indexed_nv_V4: + return Hexagon::STrih_indexed; + + case Hexagon::STrih_indexed_shl_nv_V4: + return Hexagon::STrih_indexed_shl_V4; + + case Hexagon::STrih_shl_nv_V4: + return Hexagon::STrih_shl_V4; + + case Hexagon::STrih_GP_nv_V4: + return Hexagon::STrih_GP_V4; + + case Hexagon::STh_GP_nv_V4: + return Hexagon::STh_GP_V4; + + case Hexagon::POST_SThri_nv_V4: + return Hexagon::POST_SThri; + + // Store word + + case Hexagon::STriw_imm_cdnPt_V4 : + return Hexagon::STriw_imm_cPt_V4; + + case Hexagon::STriw_imm_cdnNotPt_V4 : + return Hexagon::STriw_imm_cNotPt_V4; + + case Hexagon::STriw_cdnPt_nv_V4 : + case Hexagon::STriw_cPt_nv_V4 : + case Hexagon::STriw_cdnPt_V4 : + return Hexagon::STriw_cPt; + + case Hexagon::STriw_cdnNotPt_nv_V4 : + case Hexagon::STriw_cNotPt_nv_V4 : + case Hexagon::STriw_cdnNotPt_V4 : + return Hexagon::STriw_cNotPt; + + case Hexagon::STriw_indexed_cdnPt_nv_V4 : + case Hexagon::STriw_indexed_cPt_nv_V4 : + case Hexagon::STriw_indexed_cdnPt_V4 : + return Hexagon::STriw_indexed_cPt; + + case Hexagon::STriw_indexed_cdnNotPt_nv_V4 : + case Hexagon::STriw_indexed_cNotPt_nv_V4 : + case Hexagon::STriw_indexed_cdnNotPt_V4 : + return Hexagon::STriw_indexed_cNotPt; + + case Hexagon::STriw_indexed_shl_cdnPt_nv_V4 : + case Hexagon::STriw_indexed_shl_cPt_nv_V4 : + case Hexagon::STriw_indexed_shl_cdnPt_V4 : + return Hexagon::STriw_indexed_shl_cPt_V4; + + case Hexagon::STriw_indexed_shl_cdnNotPt_nv_V4 : + case Hexagon::STriw_indexed_shl_cNotPt_nv_V4 : + case Hexagon::STriw_indexed_shl_cdnNotPt_V4 : + return Hexagon::STriw_indexed_shl_cNotPt_V4; + + case Hexagon::POST_STwri_cdnPt_nv_V4 : + case Hexagon::POST_STwri_cPt_nv_V4 : + case Hexagon::POST_STwri_cdnPt_V4 : + return Hexagon::POST_STwri_cPt; + + case Hexagon::POST_STwri_cdnNotPt_nv_V4 : + case Hexagon::POST_STwri_cNotPt_nv_V4 : + case Hexagon::POST_STwri_cdnNotPt_V4 : + return Hexagon::POST_STwri_cNotPt; + + case Hexagon::STw_GP_cdnPt_nv_V4: + case Hexagon::STw_GP_cdnPt_V4: + case Hexagon::STw_GP_cPt_nv_V4: + return Hexagon::STw_GP_cPt_V4; + + case Hexagon::STw_GP_cdnNotPt_nv_V4: + case Hexagon::STw_GP_cdnNotPt_V4: + case Hexagon::STw_GP_cNotPt_nv_V4: + return Hexagon::STw_GP_cNotPt_V4; + + case Hexagon::STriw_GP_cdnPt_nv_V4: + case Hexagon::STriw_GP_cdnPt_V4: + case Hexagon::STriw_GP_cPt_nv_V4: + return Hexagon::STriw_GP_cPt_V4; + + case Hexagon::STriw_GP_cdnNotPt_nv_V4: + case Hexagon::STriw_GP_cdnNotPt_V4: + case Hexagon::STriw_GP_cNotPt_nv_V4: + return Hexagon::STriw_GP_cNotPt_V4; + + // Store new-value word - unconditional + + case Hexagon::STriw_nv_V4: + return Hexagon::STriw; + + case Hexagon::STriw_indexed_nv_V4: + return Hexagon::STriw_indexed; + + case Hexagon::STriw_indexed_shl_nv_V4: + return Hexagon::STriw_indexed_shl_V4; + + case Hexagon::STriw_shl_nv_V4: + return Hexagon::STriw_shl_V4; + + case Hexagon::STriw_GP_nv_V4: + return Hexagon::STriw_GP_V4; + + case Hexagon::STw_GP_nv_V4: + return Hexagon::STw_GP_V4; + + case Hexagon::POST_STwri_nv_V4: + return Hexagon::POST_STwri; + + // Store doubleword + + case Hexagon::STrid_cdnPt_V4 : + return Hexagon::STrid_cPt; + + case Hexagon::STrid_cdnNotPt_V4 : + return Hexagon::STrid_cNotPt; + + case Hexagon::STrid_indexed_cdnPt_V4 : + return Hexagon::STrid_indexed_cPt; + + case Hexagon::STrid_indexed_cdnNotPt_V4 : + return Hexagon::STrid_indexed_cNotPt; + + case Hexagon::STrid_indexed_shl_cdnPt_V4 : + return Hexagon::STrid_indexed_shl_cPt_V4; + + case Hexagon::STrid_indexed_shl_cdnNotPt_V4 : + return Hexagon::STrid_indexed_shl_cNotPt_V4; + + case Hexagon::POST_STdri_cdnPt_V4 : + return Hexagon::POST_STdri_cPt; + + case Hexagon::POST_STdri_cdnNotPt_V4 : + return Hexagon::POST_STdri_cNotPt; + + case Hexagon::STd_GP_cdnPt_V4 : + return Hexagon::STd_GP_cPt_V4; + + case Hexagon::STd_GP_cdnNotPt_V4 : + return Hexagon::STd_GP_cNotPt_V4; + + case Hexagon::STrid_GP_cdnPt_V4 : + return Hexagon::STrid_GP_cPt_V4; + + case Hexagon::STrid_GP_cdnNotPt_V4 : + return Hexagon::STrid_GP_cNotPt_V4; + + default: + assert(0 && "Unknown .old type"); + } + return 0; +} + +bool HexagonPacketizerList::DemoteToDotOld(MachineInstr* MI) { + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + int NewOpcode = GetDotOldOp(MI->getOpcode()); + MI->setDesc(QII->get(NewOpcode)); + return true; +} + +// Returns true if an instruction is predicated on p0 and false if it's +// predicated on !p0. + +static bool GetPredicateSense(MachineInstr* MI, + const HexagonInstrInfo *QII) { + + switch (MI->getOpcode()) { + case Hexagon::TFR_cPt: + case Hexagon::TFR_cdnPt: + case Hexagon::TFRI_cPt: + case Hexagon::TFRI_cdnPt: + case Hexagon::STrib_cPt : + case Hexagon::STrib_cdnPt_V4 : + case Hexagon::STrib_indexed_cPt : + case Hexagon::STrib_indexed_cdnPt_V4 : + case Hexagon::STrib_indexed_shl_cPt_V4 : + case Hexagon::STrib_indexed_shl_cdnPt_V4 : + case Hexagon::POST_STbri_cPt : + case Hexagon::POST_STbri_cdnPt_V4 : + case Hexagon::STrih_cPt : + case Hexagon::STrih_cdnPt_V4 : + case Hexagon::STrih_indexed_cPt : + case Hexagon::STrih_indexed_cdnPt_V4 : + case Hexagon::STrih_indexed_shl_cPt_V4 : + case Hexagon::STrih_indexed_shl_cdnPt_V4 : + case Hexagon::POST_SThri_cPt : + case Hexagon::POST_SThri_cdnPt_V4 : + case Hexagon::STriw_cPt : + case Hexagon::STriw_cdnPt_V4 : + case Hexagon::STriw_indexed_cPt : + case Hexagon::STriw_indexed_cdnPt_V4 : + case Hexagon::STriw_indexed_shl_cPt_V4 : + case Hexagon::STriw_indexed_shl_cdnPt_V4 : + case Hexagon::POST_STwri_cPt : + case Hexagon::POST_STwri_cdnPt_V4 : + case Hexagon::STrib_imm_cPt_V4 : + case Hexagon::STrib_imm_cdnPt_V4 : + case Hexagon::STrid_cPt : + case Hexagon::STrid_cdnPt_V4 : + case Hexagon::STrid_indexed_cPt : + case Hexagon::STrid_indexed_cdnPt_V4 : + case Hexagon::STrid_indexed_shl_cPt_V4 : + case Hexagon::STrid_indexed_shl_cdnPt_V4 : + case Hexagon::POST_STdri_cPt : + case Hexagon::POST_STdri_cdnPt_V4 : + case Hexagon::STrih_imm_cPt_V4 : + case Hexagon::STrih_imm_cdnPt_V4 : + case Hexagon::STriw_imm_cPt_V4 : + case Hexagon::STriw_imm_cdnPt_V4 : + case Hexagon::JMP_cdnPt : + case Hexagon::LDrid_cPt : + case Hexagon::LDrid_cdnPt : + case Hexagon::LDrid_indexed_cPt : + case Hexagon::LDrid_indexed_cdnPt : + case Hexagon::POST_LDrid_cPt : + case Hexagon::POST_LDrid_cdnPt_V4 : + case Hexagon::LDriw_cPt : + case Hexagon::LDriw_cdnPt : + case Hexagon::LDriw_indexed_cPt : + case Hexagon::LDriw_indexed_cdnPt : + case Hexagon::POST_LDriw_cPt : + case Hexagon::POST_LDriw_cdnPt_V4 : + case Hexagon::LDrih_cPt : + case Hexagon::LDrih_cdnPt : + case Hexagon::LDrih_indexed_cPt : + case Hexagon::LDrih_indexed_cdnPt : + case Hexagon::POST_LDrih_cPt : + case Hexagon::POST_LDrih_cdnPt_V4 : + case Hexagon::LDrib_cPt : + case Hexagon::LDrib_cdnPt : + case Hexagon::LDrib_indexed_cPt : + case Hexagon::LDrib_indexed_cdnPt : + case Hexagon::POST_LDrib_cPt : + case Hexagon::POST_LDrib_cdnPt_V4 : + case Hexagon::LDriuh_cPt : + case Hexagon::LDriuh_cdnPt : + case Hexagon::LDriuh_indexed_cPt : + case Hexagon::LDriuh_indexed_cdnPt : + case Hexagon::POST_LDriuh_cPt : + case Hexagon::POST_LDriuh_cdnPt_V4 : + case Hexagon::LDriub_cPt : + case Hexagon::LDriub_cdnPt : + case Hexagon::LDriub_indexed_cPt : + case Hexagon::LDriub_indexed_cdnPt : + case Hexagon::POST_LDriub_cPt : + case Hexagon::POST_LDriub_cdnPt_V4 : + case Hexagon::LDrid_indexed_cPt_V4 : + case Hexagon::LDrid_indexed_cdnPt_V4 : + case Hexagon::LDrid_indexed_shl_cPt_V4 : + case Hexagon::LDrid_indexed_shl_cdnPt_V4 : + case Hexagon::LDrib_indexed_cPt_V4 : + case Hexagon::LDrib_indexed_cdnPt_V4 : + case Hexagon::LDrib_indexed_shl_cPt_V4 : + case Hexagon::LDrib_indexed_shl_cdnPt_V4 : + case Hexagon::LDriub_indexed_cPt_V4 : + case Hexagon::LDriub_indexed_cdnPt_V4 : + case Hexagon::LDriub_indexed_shl_cPt_V4 : + case Hexagon::LDriub_indexed_shl_cdnPt_V4 : + case Hexagon::LDrih_indexed_cPt_V4 : + case Hexagon::LDrih_indexed_cdnPt_V4 : + case Hexagon::LDrih_indexed_shl_cPt_V4 : + case Hexagon::LDrih_indexed_shl_cdnPt_V4 : + case Hexagon::LDriuh_indexed_cPt_V4 : + case Hexagon::LDriuh_indexed_cdnPt_V4 : + case Hexagon::LDriuh_indexed_shl_cPt_V4 : + case Hexagon::LDriuh_indexed_shl_cdnPt_V4 : + case Hexagon::LDriw_indexed_cPt_V4 : + case Hexagon::LDriw_indexed_cdnPt_V4 : + case Hexagon::LDriw_indexed_shl_cPt_V4 : + case Hexagon::LDriw_indexed_shl_cdnPt_V4 : + case Hexagon::ADD_ri_cPt : + case Hexagon::ADD_ri_cdnPt : + case Hexagon::ADD_rr_cPt : + case Hexagon::ADD_rr_cdnPt : + case Hexagon::XOR_rr_cPt : + case Hexagon::XOR_rr_cdnPt : + case Hexagon::AND_rr_cPt : + case Hexagon::AND_rr_cdnPt : + case Hexagon::OR_rr_cPt : + case Hexagon::OR_rr_cdnPt : + case Hexagon::SUB_rr_cPt : + case Hexagon::SUB_rr_cdnPt : + case Hexagon::COMBINE_rr_cPt : + case Hexagon::COMBINE_rr_cdnPt : + case Hexagon::ASLH_cPt_V4 : + case Hexagon::ASLH_cdnPt_V4 : + case Hexagon::ASRH_cPt_V4 : + case Hexagon::ASRH_cdnPt_V4 : + case Hexagon::SXTB_cPt_V4 : + case Hexagon::SXTB_cdnPt_V4 : + case Hexagon::SXTH_cPt_V4 : + case Hexagon::SXTH_cdnPt_V4 : + case Hexagon::ZXTB_cPt_V4 : + case Hexagon::ZXTB_cdnPt_V4 : + case Hexagon::ZXTH_cPt_V4 : + case Hexagon::ZXTH_cdnPt_V4 : + case Hexagon::LDrid_GP_cPt_V4 : + case Hexagon::LDrib_GP_cPt_V4 : + case Hexagon::LDriub_GP_cPt_V4 : + case Hexagon::LDrih_GP_cPt_V4 : + case Hexagon::LDriuh_GP_cPt_V4 : + case Hexagon::LDriw_GP_cPt_V4 : + case Hexagon::LDd_GP_cPt_V4 : + case Hexagon::LDb_GP_cPt_V4 : + case Hexagon::LDub_GP_cPt_V4 : + case Hexagon::LDh_GP_cPt_V4 : + case Hexagon::LDuh_GP_cPt_V4 : + case Hexagon::LDw_GP_cPt_V4 : + case Hexagon::STrid_GP_cPt_V4 : + case Hexagon::STrib_GP_cPt_V4 : + case Hexagon::STrih_GP_cPt_V4 : + case Hexagon::STriw_GP_cPt_V4 : + case Hexagon::STd_GP_cPt_V4 : + case Hexagon::STb_GP_cPt_V4 : + case Hexagon::STh_GP_cPt_V4 : + case Hexagon::STw_GP_cPt_V4 : + case Hexagon::LDrid_GP_cdnPt_V4 : + case Hexagon::LDrib_GP_cdnPt_V4 : + case Hexagon::LDriub_GP_cdnPt_V4 : + case Hexagon::LDrih_GP_cdnPt_V4 : + case Hexagon::LDriuh_GP_cdnPt_V4 : + case Hexagon::LDriw_GP_cdnPt_V4 : + case Hexagon::LDd_GP_cdnPt_V4 : + case Hexagon::LDb_GP_cdnPt_V4 : + case Hexagon::LDub_GP_cdnPt_V4 : + case Hexagon::LDh_GP_cdnPt_V4 : + case Hexagon::LDuh_GP_cdnPt_V4 : + case Hexagon::LDw_GP_cdnPt_V4 : + case Hexagon::STrid_GP_cdnPt_V4 : + case Hexagon::STrib_GP_cdnPt_V4 : + case Hexagon::STrih_GP_cdnPt_V4 : + case Hexagon::STriw_GP_cdnPt_V4 : + case Hexagon::STd_GP_cdnPt_V4 : + case Hexagon::STb_GP_cdnPt_V4 : + case Hexagon::STh_GP_cdnPt_V4 : + case Hexagon::STw_GP_cdnPt_V4 : + return true; + + case Hexagon::TFR_cNotPt: + case Hexagon::TFR_cdnNotPt: + case Hexagon::TFRI_cNotPt: + case Hexagon::TFRI_cdnNotPt: + case Hexagon::STrib_cNotPt : + case Hexagon::STrib_cdnNotPt_V4 : + case Hexagon::STrib_indexed_cNotPt : + case Hexagon::STrib_indexed_cdnNotPt_V4 : + case Hexagon::STrib_indexed_shl_cNotPt_V4 : + case Hexagon::STrib_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_STbri_cNotPt : + case Hexagon::POST_STbri_cdnNotPt_V4 : + case Hexagon::STrih_cNotPt : + case Hexagon::STrih_cdnNotPt_V4 : + case Hexagon::STrih_indexed_cNotPt : + case Hexagon::STrih_indexed_cdnNotPt_V4 : + case Hexagon::STrih_indexed_shl_cNotPt_V4 : + case Hexagon::STrih_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_SThri_cNotPt : + case Hexagon::POST_SThri_cdnNotPt_V4 : + case Hexagon::STriw_cNotPt : + case Hexagon::STriw_cdnNotPt_V4 : + case Hexagon::STriw_indexed_cNotPt : + case Hexagon::STriw_indexed_cdnNotPt_V4 : + case Hexagon::STriw_indexed_shl_cNotPt_V4 : + case Hexagon::STriw_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_STwri_cNotPt : + case Hexagon::POST_STwri_cdnNotPt_V4 : + case Hexagon::STrib_imm_cNotPt_V4 : + case Hexagon::STrib_imm_cdnNotPt_V4 : + case Hexagon::STrid_cNotPt : + case Hexagon::STrid_cdnNotPt_V4 : + case Hexagon::STrid_indexed_cdnNotPt_V4 : + case Hexagon::STrid_indexed_cNotPt : + case Hexagon::STrid_indexed_shl_cNotPt_V4 : + case Hexagon::STrid_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_STdri_cNotPt : + case Hexagon::POST_STdri_cdnNotPt_V4 : + case Hexagon::STrih_imm_cNotPt_V4 : + case Hexagon::STrih_imm_cdnNotPt_V4 : + case Hexagon::STriw_imm_cNotPt_V4 : + case Hexagon::STriw_imm_cdnNotPt_V4 : + case Hexagon::JMP_cdnNotPt : + case Hexagon::LDrid_cNotPt : + case Hexagon::LDrid_cdnNotPt : + case Hexagon::LDrid_indexed_cNotPt : + case Hexagon::LDrid_indexed_cdnNotPt : + case Hexagon::POST_LDrid_cNotPt : + case Hexagon::POST_LDrid_cdnNotPt_V4 : + case Hexagon::LDriw_cNotPt : + case Hexagon::LDriw_cdnNotPt : + case Hexagon::LDriw_indexed_cNotPt : + case Hexagon::LDriw_indexed_cdnNotPt : + case Hexagon::POST_LDriw_cNotPt : + case Hexagon::POST_LDriw_cdnNotPt_V4 : + case Hexagon::LDrih_cNotPt : + case Hexagon::LDrih_cdnNotPt : + case Hexagon::LDrih_indexed_cNotPt : + case Hexagon::LDrih_indexed_cdnNotPt : + case Hexagon::POST_LDrih_cNotPt : + case Hexagon::POST_LDrih_cdnNotPt_V4 : + case Hexagon::LDrib_cNotPt : + case Hexagon::LDrib_cdnNotPt : + case Hexagon::LDrib_indexed_cNotPt : + case Hexagon::LDrib_indexed_cdnNotPt : + case Hexagon::POST_LDrib_cNotPt : + case Hexagon::POST_LDrib_cdnNotPt_V4 : + case Hexagon::LDriuh_cNotPt : + case Hexagon::LDriuh_cdnNotPt : + case Hexagon::LDriuh_indexed_cNotPt : + case Hexagon::LDriuh_indexed_cdnNotPt : + case Hexagon::POST_LDriuh_cNotPt : + case Hexagon::POST_LDriuh_cdnNotPt_V4 : + case Hexagon::LDriub_cNotPt : + case Hexagon::LDriub_cdnNotPt : + case Hexagon::LDriub_indexed_cNotPt : + case Hexagon::LDriub_indexed_cdnNotPt : + case Hexagon::POST_LDriub_cNotPt : + case Hexagon::POST_LDriub_cdnNotPt_V4 : + case Hexagon::LDrid_indexed_cNotPt_V4 : + case Hexagon::LDrid_indexed_cdnNotPt_V4 : + case Hexagon::LDrid_indexed_shl_cNotPt_V4 : + case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDrib_indexed_cNotPt_V4 : + case Hexagon::LDrib_indexed_cdnNotPt_V4 : + case Hexagon::LDrib_indexed_shl_cNotPt_V4 : + case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriub_indexed_cNotPt_V4 : + case Hexagon::LDriub_indexed_cdnNotPt_V4 : + case Hexagon::LDriub_indexed_shl_cNotPt_V4 : + case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDrih_indexed_cNotPt_V4 : + case Hexagon::LDrih_indexed_cdnNotPt_V4 : + case Hexagon::LDrih_indexed_shl_cNotPt_V4 : + case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriuh_indexed_cNotPt_V4 : + case Hexagon::LDriuh_indexed_cdnNotPt_V4 : + case Hexagon::LDriuh_indexed_shl_cNotPt_V4 : + case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriw_indexed_cNotPt_V4 : + case Hexagon::LDriw_indexed_cdnNotPt_V4 : + case Hexagon::LDriw_indexed_shl_cNotPt_V4 : + case Hexagon::LDriw_indexed_shl_cdnNotPt_V4 : + case Hexagon::ADD_ri_cNotPt : + case Hexagon::ADD_ri_cdnNotPt : + case Hexagon::ADD_rr_cNotPt : + case Hexagon::ADD_rr_cdnNotPt : + case Hexagon::XOR_rr_cNotPt : + case Hexagon::XOR_rr_cdnNotPt : + case Hexagon::AND_rr_cNotPt : + case Hexagon::AND_rr_cdnNotPt : + case Hexagon::OR_rr_cNotPt : + case Hexagon::OR_rr_cdnNotPt : + case Hexagon::SUB_rr_cNotPt : + case Hexagon::SUB_rr_cdnNotPt : + case Hexagon::COMBINE_rr_cNotPt : + case Hexagon::COMBINE_rr_cdnNotPt : + case Hexagon::ASLH_cNotPt_V4 : + case Hexagon::ASLH_cdnNotPt_V4 : + case Hexagon::ASRH_cNotPt_V4 : + case Hexagon::ASRH_cdnNotPt_V4 : + case Hexagon::SXTB_cNotPt_V4 : + case Hexagon::SXTB_cdnNotPt_V4 : + case Hexagon::SXTH_cNotPt_V4 : + case Hexagon::SXTH_cdnNotPt_V4 : + case Hexagon::ZXTB_cNotPt_V4 : + case Hexagon::ZXTB_cdnNotPt_V4 : + case Hexagon::ZXTH_cNotPt_V4 : + case Hexagon::ZXTH_cdnNotPt_V4 : + + case Hexagon::LDrid_GP_cNotPt_V4 : + case Hexagon::LDrib_GP_cNotPt_V4 : + case Hexagon::LDriub_GP_cNotPt_V4 : + case Hexagon::LDrih_GP_cNotPt_V4 : + case Hexagon::LDriuh_GP_cNotPt_V4 : + case Hexagon::LDriw_GP_cNotPt_V4 : + case Hexagon::LDd_GP_cNotPt_V4 : + case Hexagon::LDb_GP_cNotPt_V4 : + case Hexagon::LDub_GP_cNotPt_V4 : + case Hexagon::LDh_GP_cNotPt_V4 : + case Hexagon::LDuh_GP_cNotPt_V4 : + case Hexagon::LDw_GP_cNotPt_V4 : + case Hexagon::STrid_GP_cNotPt_V4 : + case Hexagon::STrib_GP_cNotPt_V4 : + case Hexagon::STrih_GP_cNotPt_V4 : + case Hexagon::STriw_GP_cNotPt_V4 : + case Hexagon::STd_GP_cNotPt_V4 : + case Hexagon::STb_GP_cNotPt_V4 : + case Hexagon::STh_GP_cNotPt_V4 : + case Hexagon::STw_GP_cNotPt_V4 : + case Hexagon::LDrid_GP_cdnNotPt_V4 : + case Hexagon::LDrib_GP_cdnNotPt_V4 : + case Hexagon::LDriub_GP_cdnNotPt_V4 : + case Hexagon::LDrih_GP_cdnNotPt_V4 : + case Hexagon::LDriuh_GP_cdnNotPt_V4 : + case Hexagon::LDriw_GP_cdnNotPt_V4 : + case Hexagon::LDd_GP_cdnNotPt_V4 : + case Hexagon::LDb_GP_cdnNotPt_V4 : + case Hexagon::LDub_GP_cdnNotPt_V4 : + case Hexagon::LDh_GP_cdnNotPt_V4 : + case Hexagon::LDuh_GP_cdnNotPt_V4 : + case Hexagon::LDw_GP_cdnNotPt_V4 : + case Hexagon::STrid_GP_cdnNotPt_V4 : + case Hexagon::STrib_GP_cdnNotPt_V4 : + case Hexagon::STrih_GP_cdnNotPt_V4 : + case Hexagon::STriw_GP_cdnNotPt_V4 : + case Hexagon::STd_GP_cdnNotPt_V4 : + case Hexagon::STb_GP_cdnNotPt_V4 : + case Hexagon::STh_GP_cdnNotPt_V4 : + case Hexagon::STw_GP_cdnNotPt_V4 : + return false; + + default: + assert (false && "Unknown predicate sense of the instruction"); + } + // return *some value* to avoid compiler warning + return false; +} + +bool HexagonPacketizerList::isDotNewInst(MachineInstr* MI) { + if (isNewValueInst(MI)) + return true; + + switch (MI->getOpcode()) { + case Hexagon::TFR_cdnNotPt: + case Hexagon::TFR_cdnPt: + case Hexagon::TFRI_cdnNotPt: + case Hexagon::TFRI_cdnPt: + case Hexagon::LDrid_cdnPt : + case Hexagon::LDrid_cdnNotPt : + case Hexagon::LDrid_indexed_cdnPt : + case Hexagon::LDrid_indexed_cdnNotPt : + case Hexagon::POST_LDrid_cdnPt_V4 : + case Hexagon::POST_LDrid_cdnNotPt_V4 : + case Hexagon::LDriw_cdnPt : + case Hexagon::LDriw_cdnNotPt : + case Hexagon::LDriw_indexed_cdnPt : + case Hexagon::LDriw_indexed_cdnNotPt : + case Hexagon::POST_LDriw_cdnPt_V4 : + case Hexagon::POST_LDriw_cdnNotPt_V4 : + case Hexagon::LDrih_cdnPt : + case Hexagon::LDrih_cdnNotPt : + case Hexagon::LDrih_indexed_cdnPt : + case Hexagon::LDrih_indexed_cdnNotPt : + case Hexagon::POST_LDrih_cdnPt_V4 : + case Hexagon::POST_LDrih_cdnNotPt_V4 : + case Hexagon::LDrib_cdnPt : + case Hexagon::LDrib_cdnNotPt : + case Hexagon::LDrib_indexed_cdnPt : + case Hexagon::LDrib_indexed_cdnNotPt : + case Hexagon::POST_LDrib_cdnPt_V4 : + case Hexagon::POST_LDrib_cdnNotPt_V4 : + case Hexagon::LDriuh_cdnPt : + case Hexagon::LDriuh_cdnNotPt : + case Hexagon::LDriuh_indexed_cdnPt : + case Hexagon::LDriuh_indexed_cdnNotPt : + case Hexagon::POST_LDriuh_cdnPt_V4 : + case Hexagon::POST_LDriuh_cdnNotPt_V4 : + case Hexagon::LDriub_cdnPt : + case Hexagon::LDriub_cdnNotPt : + case Hexagon::LDriub_indexed_cdnPt : + case Hexagon::LDriub_indexed_cdnNotPt : + case Hexagon::POST_LDriub_cdnPt_V4 : + case Hexagon::POST_LDriub_cdnNotPt_V4 : + + case Hexagon::LDrid_indexed_cdnPt_V4 : + case Hexagon::LDrid_indexed_cdnNotPt_V4 : + case Hexagon::LDrid_indexed_shl_cdnPt_V4 : + case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDrib_indexed_cdnPt_V4 : + case Hexagon::LDrib_indexed_cdnNotPt_V4 : + case Hexagon::LDrib_indexed_shl_cdnPt_V4 : + case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriub_indexed_cdnPt_V4 : + case Hexagon::LDriub_indexed_cdnNotPt_V4 : + case Hexagon::LDriub_indexed_shl_cdnPt_V4 : + case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDrih_indexed_cdnPt_V4 : + case Hexagon::LDrih_indexed_cdnNotPt_V4 : + case Hexagon::LDrih_indexed_shl_cdnPt_V4 : + case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriuh_indexed_cdnPt_V4 : + case Hexagon::LDriuh_indexed_cdnNotPt_V4 : + case Hexagon::LDriuh_indexed_shl_cdnPt_V4 : + case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 : + case Hexagon::LDriw_indexed_cdnPt_V4 : + case Hexagon::LDriw_indexed_cdnNotPt_V4 : + case Hexagon::LDriw_indexed_shl_cdnPt_V4 : + case Hexagon::LDriw_indexed_shl_cdnNotPt_V4 : + +// Coditional add + case Hexagon::ADD_ri_cdnPt: + case Hexagon::ADD_ri_cdnNotPt: + case Hexagon::ADD_rr_cdnPt: + case Hexagon::ADD_rr_cdnNotPt: + + // Conditional logical operations + case Hexagon::XOR_rr_cdnPt : + case Hexagon::XOR_rr_cdnNotPt : + case Hexagon::AND_rr_cdnPt : + case Hexagon::AND_rr_cdnNotPt : + case Hexagon::OR_rr_cdnPt : + case Hexagon::OR_rr_cdnNotPt : + + // Conditonal subtract + case Hexagon::SUB_rr_cdnPt : + case Hexagon::SUB_rr_cdnNotPt : + + // Conditional combine + case Hexagon::COMBINE_rr_cdnPt : + case Hexagon::COMBINE_rr_cdnNotPt : + + // Conditional shift operations + case Hexagon::ASLH_cdnPt_V4: + case Hexagon::ASLH_cdnNotPt_V4: + case Hexagon::ASRH_cdnPt_V4: + case Hexagon::ASRH_cdnNotPt_V4: + case Hexagon::SXTB_cdnPt_V4: + case Hexagon::SXTB_cdnNotPt_V4: + case Hexagon::SXTH_cdnPt_V4: + case Hexagon::SXTH_cdnNotPt_V4: + case Hexagon::ZXTB_cdnPt_V4: + case Hexagon::ZXTB_cdnNotPt_V4: + case Hexagon::ZXTH_cdnPt_V4: + case Hexagon::ZXTH_cdnNotPt_V4: + + // Conditional stores + case Hexagon::STrib_imm_cdnPt_V4 : + case Hexagon::STrib_imm_cdnNotPt_V4 : + case Hexagon::STrib_cdnPt_V4 : + case Hexagon::STrib_cdnNotPt_V4 : + case Hexagon::STrib_indexed_cdnPt_V4 : + case Hexagon::STrib_indexed_cdnNotPt_V4 : + case Hexagon::POST_STbri_cdnPt_V4 : + case Hexagon::POST_STbri_cdnNotPt_V4 : + case Hexagon::STrib_indexed_shl_cdnPt_V4 : + case Hexagon::STrib_indexed_shl_cdnNotPt_V4 : + + // Store doubleword conditionally + case Hexagon::STrid_indexed_cdnPt_V4 : + case Hexagon::STrid_indexed_cdnNotPt_V4 : + case Hexagon::STrid_indexed_shl_cdnPt_V4 : + case Hexagon::STrid_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_STdri_cdnPt_V4 : + case Hexagon::POST_STdri_cdnNotPt_V4 : + + // Store halfword conditionally + case Hexagon::STrih_cdnPt_V4 : + case Hexagon::STrih_cdnNotPt_V4 : + case Hexagon::STrih_indexed_cdnPt_V4 : + case Hexagon::STrih_indexed_cdnNotPt_V4 : + case Hexagon::STrih_imm_cdnPt_V4 : + case Hexagon::STrih_imm_cdnNotPt_V4 : + case Hexagon::STrih_indexed_shl_cdnPt_V4 : + case Hexagon::STrih_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_SThri_cdnPt_V4 : + case Hexagon::POST_SThri_cdnNotPt_V4 : + + // Store word conditionally + case Hexagon::STriw_cdnPt_V4 : + case Hexagon::STriw_cdnNotPt_V4 : + case Hexagon::STriw_indexed_cdnPt_V4 : + case Hexagon::STriw_indexed_cdnNotPt_V4 : + case Hexagon::STriw_imm_cdnPt_V4 : + case Hexagon::STriw_imm_cdnNotPt_V4 : + case Hexagon::STriw_indexed_shl_cdnPt_V4 : + case Hexagon::STriw_indexed_shl_cdnNotPt_V4 : + case Hexagon::POST_STwri_cdnPt_V4 : + case Hexagon::POST_STwri_cdnNotPt_V4 : + + case Hexagon::LDd_GP_cdnPt_V4: + case Hexagon::LDd_GP_cdnNotPt_V4: + case Hexagon::LDb_GP_cdnPt_V4: + case Hexagon::LDb_GP_cdnNotPt_V4: + case Hexagon::LDub_GP_cdnPt_V4: + case Hexagon::LDub_GP_cdnNotPt_V4: + case Hexagon::LDh_GP_cdnPt_V4: + case Hexagon::LDh_GP_cdnNotPt_V4: + case Hexagon::LDuh_GP_cdnPt_V4: + case Hexagon::LDuh_GP_cdnNotPt_V4: + case Hexagon::LDw_GP_cdnPt_V4: + case Hexagon::LDw_GP_cdnNotPt_V4: + case Hexagon::LDrid_GP_cdnPt_V4: + case Hexagon::LDrid_GP_cdnNotPt_V4: + case Hexagon::LDrib_GP_cdnPt_V4: + case Hexagon::LDrib_GP_cdnNotPt_V4: + case Hexagon::LDriub_GP_cdnPt_V4: + case Hexagon::LDriub_GP_cdnNotPt_V4: + case Hexagon::LDrih_GP_cdnPt_V4: + case Hexagon::LDrih_GP_cdnNotPt_V4: + case Hexagon::LDriuh_GP_cdnPt_V4: + case Hexagon::LDriuh_GP_cdnNotPt_V4: + case Hexagon::LDriw_GP_cdnPt_V4: + case Hexagon::LDriw_GP_cdnNotPt_V4: + + case Hexagon::STrid_GP_cdnPt_V4: + case Hexagon::STrid_GP_cdnNotPt_V4: + case Hexagon::STrib_GP_cdnPt_V4: + case Hexagon::STrib_GP_cdnNotPt_V4: + case Hexagon::STrih_GP_cdnPt_V4: + case Hexagon::STrih_GP_cdnNotPt_V4: + case Hexagon::STriw_GP_cdnPt_V4: + case Hexagon::STriw_GP_cdnNotPt_V4: + case Hexagon::STd_GP_cdnPt_V4: + case Hexagon::STd_GP_cdnNotPt_V4: + case Hexagon::STb_GP_cdnPt_V4: + case Hexagon::STb_GP_cdnNotPt_V4: + case Hexagon::STh_GP_cdnPt_V4: + case Hexagon::STh_GP_cdnNotPt_V4: + case Hexagon::STw_GP_cdnPt_V4: + case Hexagon::STw_GP_cdnNotPt_V4: + + return true; + } + return false; +} + +static MachineOperand& GetPostIncrementOperand(MachineInstr *MI, + const HexagonInstrInfo *QII) { + assert(QII->isPostIncrement(MI) && "Not a post increment operation."); +#ifndef NDEBUG + // Post Increment means duplicates. Use dense map to find duplicates in the + // list. Caution: Densemap initializes with the minimum of 64 buckets, + // whereas there are at most 5 operands in the post increment. + DenseMap DefRegsSet; + for(unsigned opNum = 0; opNum < MI->getNumOperands(); opNum++) + if (MI->getOperand(opNum).isReg() && + MI->getOperand(opNum).isDef()) { + DefRegsSet[MI->getOperand(opNum).getReg()] = 1; + } + + for(unsigned opNum = 0; opNum < MI->getNumOperands(); opNum++) + if (MI->getOperand(opNum).isReg() && + MI->getOperand(opNum).isUse()) { + if (DefRegsSet[MI->getOperand(opNum).getReg()]) { + return MI->getOperand(opNum); + } + } +#else + if (MI->getDesc().mayLoad()) { + // The 2nd operand is always the post increment operand in load. + assert(MI->getOperand(1).isReg() && + "Post increment operand has be to a register."); + return (MI->getOperand(1)); + } + if (MI->getDesc().mayStore()) { + // The 1st operand is always the post increment operand in store. + assert(MI->getOperand(0).isReg() && + "Post increment operand has be to a register."); + return (MI->getOperand(0)); + } +#endif + // we should never come here. + assert(0 && "mayLoad or mayStore not set for Post Increment operation"); + + // return *some value* to avoid compiler warning + return MI->getOperand(0); +} + +// get the value being stored +static MachineOperand& GetStoreValueOperand(MachineInstr *MI) { + // value being stored is always the last operand. + return (MI->getOperand(MI->getNumOperands()-1)); +} + +// can be new value store? +// Following restrictions are to be respected in convert a store into +// a new value store. +// 1. If an instruction uses auto-increment, its address register cannot +// be a new-value register. Arch Spec 5.4.2.1 +// 2. If an instruction uses absolute-set addressing mode, +// its address register cannot be a new-value register. +// Arch Spec 5.4.2.1.TODO: This is not enabled as +// as absolute-set address mode patters are not implemented. +// 3. If an instruction produces a 64-bit result, its registers cannot be used +// as new-value registers. Arch Spec 5.4.2.2. +// 4. If the instruction that sets a new-value register is conditional, then +// the instruction that uses the new-value register must also be conditional, +// and both must always have their predicates evaluate identically. +// Arch Spec 5.4.2.3. +// 5. There is an implied restriction of a packet can not have another store, +// if there is a new value store in the packet. Corollary, if there is +// already a store in a packet, there can not be a new value store. +// Arch Spec: 3.4.4.2 +bool HexagonPacketizerList::CanPromoteToNewValueStore( MachineInstr *MI, + MachineInstr *PacketMI, unsigned DepReg, + std::map MIToSUnit) +{ + // Make sure we are looking at the store + if (!IsNewifyStore(MI)) + return false; + + // Make sure there is dependency and can be new value'ed + if (GetStoreValueOperand(MI).isReg() && + GetStoreValueOperand(MI).getReg() != DepReg) + return false; + + const HexagonRegisterInfo* QRI = (const HexagonRegisterInfo *) TM.getRegisterInfo(); + const MCInstrDesc& MCID = PacketMI->getDesc(); + // first operand is always the result + + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + const TargetRegisterClass* PacketRC = QII->getRegClass(MCID, 0, QRI); + + // if there is already an store in the packet, no can do new value store + // Arch Spec 3.4.4.2. + for (std::vector::iterator VI = CurrentPacketMIs.begin(), + VE = CurrentPacketMIs.end(); + (VI != VE); ++VI) { + SUnit* PacketSU = MIToSUnit[*VI]; + if (PacketSU->getInstr()->getDesc().mayStore() || + // if we have mayStore = 1 set on ALLOCFRAME and DEALLOCFRAME, + // then we don't need this + PacketSU->getInstr()->getOpcode() == Hexagon::ALLOCFRAME || + PacketSU->getInstr()->getOpcode() == Hexagon::DEALLOCFRAME) + return false; + } + + if (PacketRC == Hexagon::DoubleRegsRegisterClass) { + // new value store constraint: double regs can not feed into new value store + // arch spec section: 5.4.2.2 + return false; + } + + // Make sure it's NOT the post increment register that we are going to + // new value. + if (QII->isPostIncrement(MI) && + MI->getDesc().mayStore() && + GetPostIncrementOperand(MI, QII).getReg() == DepReg) { + return false; + } + + if (QII->isPostIncrement(PacketMI) && + PacketMI->getDesc().mayLoad() && + GetPostIncrementOperand(PacketMI, QII).getReg() == DepReg) { + // if source is post_inc, or absolute-set addressing, + // it can not feed into new value store + // r3 = memw(r2++#4) + // memw(r30 + #-1404) = r2.new -> can not be new value store + // arch spec section: 5.4.2.1 + return false; + } + + // If the source that feeds the store is predicated, new value store must also be + // also predicated. + if (QII->isPredicated(PacketMI)) { + if (!QII->isPredicated(MI)) + return false; + + // Check to make sure that they both will have their predicates + // evaluate identically + unsigned predRegNumSrc; + unsigned predRegNumDst; + const TargetRegisterClass* predRegClass; + + // Get predicate register used in the source instruction + for(unsigned opNum = 0; opNum < PacketMI->getNumOperands(); opNum++) { + if ( PacketMI->getOperand(opNum).isReg()) + predRegNumSrc = PacketMI->getOperand(opNum).getReg(); + predRegClass = QRI->getMinimalPhysRegClass(predRegNumSrc); + if (predRegClass == Hexagon::PredRegsRegisterClass) { + break; + } + } + assert ((predRegClass == Hexagon::PredRegsRegisterClass ) && + ("predicate register not found in a predicated PacketMI instruction")); + + // Get predicate register used in new-value store instruction + for(unsigned opNum = 0; opNum < MI->getNumOperands(); opNum++) { + if ( MI->getOperand(opNum).isReg()) + predRegNumDst = MI->getOperand(opNum).getReg(); + predRegClass = QRI->getMinimalPhysRegClass(predRegNumDst); + if (predRegClass == Hexagon::PredRegsRegisterClass) { + break; + } + } + assert ((predRegClass == Hexagon::PredRegsRegisterClass ) && + ("predicate register not found in a predicated MI instruction")); + + // New-value register producer and user (store) need to satisfy these + // constraints: + // 1) Both instructions should be predicated on the same register. + // 2) If producer of the new-value register is .new predicated then store + // should also be .new predicated and if producer is not .new predicated + // then store should not be .new predicated. + // 3) Both new-value register producer and user should have same predicate + // sense, i.e, either both should be negated or both should be none negated. + + if (( predRegNumDst != predRegNumSrc) || + isDotNewInst(PacketMI) != isDotNewInst(MI) || + GetPredicateSense(MI, QII) != GetPredicateSense(PacketMI, QII)) { + return false; + } + } + + // Make sure that other than the new-value register no other store instruction + // register has been modified in the same packet. Predicate registers can be + // modified by they should not be modified between the producer and the store + // instruction as it will make them both conditional on different values. + // We already know this to be true for all the instructions before and + // including PacketMI. Howerver, we need to perform the check for the + // remaining instructions in the packet. + + std::vector::iterator VI; + std::vector::iterator VE; + unsigned StartCheck = 0; + + for (VI=CurrentPacketMIs.begin(), VE = CurrentPacketMIs.end(); + (VI != VE); ++VI) { + SUnit* TempSU = MIToSUnit[*VI]; + MachineInstr* TempMI = TempSU->getInstr(); + + // Following condition is true for all the instructions until PacketMI is + // reached (StartCheck is set to 0 before the for loop). + // StartCheck flag is 1 for all the instructions after PacketMI. + if (TempMI != PacketMI && !StartCheck) // start processing only after + continue; // encountering PacketMI + + StartCheck = 1; + if (TempMI == PacketMI) // We don't want to check PacketMI for dependence + continue; + + for(unsigned opNum = 0; opNum < MI->getNumOperands(); opNum++) { + if (MI->getOperand(opNum).isReg() && + TempSU->getInstr()->modifiesRegister(MI->getOperand(opNum).getReg(), QRI)) + return false; + } + } + + // Make sure that for non POST_INC stores: + // 1. The only use of reg is DepReg and no other registers. + // This handles V4 base+index registers. + // The following store can not be dot new. + // Eg. r0 = add(r0, #3)a + // memw(r1+r0<<#2) = r0 + if (!QII->isPostIncrement(MI) && + GetStoreValueOperand(MI).isReg() && + GetStoreValueOperand(MI).getReg() == DepReg) { + for(unsigned opNum = 0; opNum < MI->getNumOperands()-1; opNum++) { + if (MI->getOperand(opNum).isReg() && + MI->getOperand(opNum).getReg() == DepReg) { + return false; + } + } + // 2. If data definition is because of implicit definition of the register, + // do not newify the store. Eg. + // %R9 = ZXTH %R12, %D6, %R12 + // STrih_indexed %R8, 2, %R12; mem:ST2[%scevgep343] + for(unsigned opNum = 0; opNum < PacketMI->getNumOperands(); opNum++) { + if (PacketMI->getOperand(opNum).isReg() && + PacketMI->getOperand(opNum).getReg() == DepReg && + PacketMI->getOperand(opNum).isDef() && + PacketMI->getOperand(opNum).isImplicit()) { + return false; + } + } + } + + // Can be dot new store. + return true; +} + +// can this MI to promoted to either +// new value store or new value jump +bool HexagonPacketizerList::CanPromoteToNewValue( MachineInstr *MI, + SUnit *PacketSU, unsigned DepReg, + std::map MIToSUnit, + MachineBasicBlock::iterator &MII) +{ + + const HexagonRegisterInfo* QRI = (const HexagonRegisterInfo *) TM.getRegisterInfo(); + if (!QRI->Subtarget.hasV4TOps() || + !IsNewifyStore(MI)) + return false; + + MachineInstr *PacketMI = PacketSU->getInstr(); + + // Check to see the store can be new value'ed. + if (CanPromoteToNewValueStore(MI, PacketMI, DepReg, MIToSUnit)) + return true; + + // Check to see the compare/jump can be new value'ed. + // This is done as a pass on its own. Don't need to check it here. + return false; +} + +// Check to see if an instruction can be dot new +// There are three kinds. +// 1. dot new on predicate - V2/V3/V4 +// 2. dot new on stores NV/ST - V4 +// 3. dot new on jump NV/J - V4 -- This is generated in a pass. +bool HexagonPacketizerList::CanPromoteToDotNew( MachineInstr *MI, + SUnit *PacketSU, unsigned DepReg, + std::map MIToSUnit, + MachineBasicBlock::iterator &MII, + const TargetRegisterClass* RC ) +{ + // already a dot new instruction + if (isDotNewInst(MI) && !IsNewifyStore(MI)) + return false; + + if (!isNewifiable(MI)) + return false; + + // predicate .new + if (RC == Hexagon::PredRegsRegisterClass && isCondInst(MI)) + return true; + else if (RC != Hexagon::PredRegsRegisterClass && + !IsNewifyStore(MI)) // MI is not a new-value store + return false; + else { + // Create a dot new machine instruction to see if resources can be + // allocated. If not, bail out now. + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + int NewOpcode = GetDotNewOp(MI->getOpcode()); + const MCInstrDesc &desc = QII->get(NewOpcode); + DebugLoc dl; + MachineInstr *NewMI = MI->getParent()->getParent()->CreateMachineInstr(desc, dl); + bool ResourcesAvailable = ResourceTracker->canReserveResources(NewMI); + MI->getParent()->getParent()->DeleteMachineInstr(NewMI); + + if (!ResourcesAvailable) + return false; + + // new value store only + // new new value jump generated as a passes + if (!CanPromoteToNewValue(MI, PacketSU, DepReg, MIToSUnit, MII)) { + return false; + } + } + return true; +} + +// Go through the packet instructions and search for anti dependency +// between them and DepReg from MI +// Consider this case: +// Trying to add +// a) %R1 = TFRI_cdNotPt %P3, 2 +// to this packet: +// { +// b) %P0 = OR_pp %P3, %P0 +// c) %P3 = TFR_PdRs %R23 +// d) %R1 = TFRI_cdnPt %P3, 4 +// } +// The P3 from a) and d) will be complements after +// a)'s P3 is converted to .new form +// Anti Dep between c) and b) is irrelevant for this case +bool HexagonPacketizerList::RestrictingDepExistInPacket (MachineInstr* MI, + unsigned DepReg, + std::map MIToSUnit) { + + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + SUnit* PacketSUDep = MIToSUnit[MI]; + + for (std::vector::iterator VIN = CurrentPacketMIs.begin(), + VEN = CurrentPacketMIs.end(); (VIN != VEN); ++VIN) { + + // We only care for dependencies to predicated instructions + if(!QII->isPredicated(*VIN)) continue; + + // Scheduling Unit for current insn in the packet + SUnit* PacketSU = MIToSUnit[*VIN]; + + // Look at dependencies between current members of the packet + // and predicate defining instruction MI. + // Make sure that dependency is on the exact register + // we care about. + if (PacketSU->isSucc(PacketSUDep)) { + for (unsigned i = 0; i < PacketSU->Succs.size(); ++i) { + if ((PacketSU->Succs[i].getSUnit() == PacketSUDep) && + (PacketSU->Succs[i].getKind() == SDep::Anti) && + (PacketSU->Succs[i].getReg() == DepReg)) { + return true; + } + } + } + } + + return false; +} + + +// Given two predicated instructions, this function detects whether +// the predicates are complements +bool HexagonPacketizerList::ArePredicatesComplements (MachineInstr* MI1, + MachineInstr* MI2, std::map MIToSUnit) { + + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + // Currently can only reason about conditional transfers + if (!QII->isConditionalTransfer(MI1) || !QII->isConditionalTransfer(MI2)) { + return false; + } + + // Scheduling unit for candidate + SUnit* SU = MIToSUnit[MI1]; + + // One corner case deals with the following scenario: + // Trying to add + // a) %R24 = TFR_cPt %P0, %R25 + // to this packet: + // + // { + // b) %R25 = TFR_cNotPt %P0, %R24 + // c) %P0 = CMPEQri %R26, 1 + // } + // + // On general check a) and b) are complements, but + // presence of c) will convert a) to .new form, and + // then it is not a complement + // We attempt to detect it by analyzing existing + // dependencies in the packet + + // Analyze relationships between all existing members of the packet. + // Look for Anti dependecy on the same predicate reg + // as used in the candidate + for (std::vector::iterator VIN = CurrentPacketMIs.begin(), + VEN = CurrentPacketMIs.end(); (VIN != VEN); ++VIN) { + + // Scheduling Unit for current insn in the packet + SUnit* PacketSU = MIToSUnit[*VIN]; + + // If this instruction in the packet is succeeded by the candidate... + if (PacketSU->isSucc(SU)) { + for (unsigned i = 0; i < PacketSU->Succs.size(); ++i) { + // The corner case exist when there is true data + // dependency between candidate and one of current + // packet members, this dep is on predicate reg, and + // there already exist anti dep on the same pred in + // the packet. + if (PacketSU->Succs[i].getSUnit() == SU && + Hexagon::PredRegsRegisterClass->contains( + PacketSU->Succs[i].getReg()) && + PacketSU->Succs[i].getKind() == SDep::Data && + // Here I know that *VIN is predicate setting instruction + // with true data dep to candidate on the register + // we care about - c) in the above example. + // Now I need to see if there is an anti dependency + // from c) to any other instruction in the + // same packet on the pred reg of interest + RestrictingDepExistInPacket(*VIN,PacketSU->Succs[i].getReg(), + MIToSUnit)) { + return false; + } + } + } + } + + // If the above case does not apply, check regular + // complement condition. + // Check that the predicate register is the same and + // that the predicate sense is different + // We also need to differentiate .old vs. .new: + // !p0 is not complimentary to p0.new + return ((MI1->getOperand(1).getReg() == MI2->getOperand(1).getReg()) && + (GetPredicateSense(MI1, QII) != GetPredicateSense(MI2, QII)) && + (isDotNewInst(MI1) == isDotNewInst(MI2))); +} + +// initPacketizerState - Initialize packetizer flags +void HexagonPacketizerList::initPacketizerState(void) { + + Dependence = false; + PromotedToDotNew = false; + GlueToNewValueJump = false; + GlueAllocframeStore = false; + FoundSequentialDependence = false; + + return; +} + +// ignorePseudoInstruction - Ignore bundling of pseudo instructions. +bool HexagonPacketizerList::ignorePseudoInstruction(MachineInstr *MI, + MachineBasicBlock *MBB) { + if (MI->isDebugValue()) + return true; + + // We must print out inline assembly + if (MI->isInlineAsm()) + return false; + + // We check if MI has any functional units mapped to it. + // If it doesn't, we ignore the instruction. + const MCInstrDesc& TID = MI->getDesc(); + unsigned SchedClass = TID.getSchedClass(); + const InstrStage* IS = ResourceTracker->getInstrItins()->beginStage(SchedClass); + unsigned FuncUnits = IS->getUnits(); + return !FuncUnits; +} + +// isSoloInstruction: - Returns true for instructions that must be +// scheduled in their own packet. +bool HexagonPacketizerList::isSoloInstruction(MachineInstr *MI) { + + if (MI->isInlineAsm()) + return true; + + if (MI->isEHLabel()) + return true; + + // From Hexagon V4 Programmer's Reference Manual 3.4.4 Grouping constraints: + // trap, pause, barrier, icinva, isync, and syncht are solo instructions. + // They must not be grouped with other instructions in a packet. + if (IsSchedBarrier(MI)) + return true; + + return false; +} + +// isLegalToPacketizeTogether: +// SUI is the current instruction that is out side of the current packet. +// SUJ is the current instruction inside the current packet against which that +// SUI will be packetized. +bool HexagonPacketizerList::isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) { + MachineInstr *I = SUI->getInstr(); + MachineInstr *J = SUJ->getInstr(); + assert(I && J && "Unable to packetize null instruction!"); + + const MCInstrDesc &MCIDI = I->getDesc(); + const MCInstrDesc &MCIDJ = J->getDesc(); + + MachineBasicBlock::iterator II = I; + + const unsigned FrameSize = MF.getFrameInfo()->getStackSize(); + const HexagonRegisterInfo* QRI = (const HexagonRegisterInfo *) TM.getRegisterInfo(); + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + + // Inline asm cannot go in the packet. + if (I->getOpcode() == Hexagon::INLINEASM) + assert(0 && "Should not meet inline asm here!"); + + if (isSoloInstruction(I)) + assert(0 && "Should not meet solo instr here!"); + + // A save callee-save register function call can only be in a packet + // with instructions that don't write to the callee-save registers. + if ((QII->isSaveCalleeSavedRegsCall(I) && + DoesModifyCalleeSavedReg(J, QRI)) || + (QII->isSaveCalleeSavedRegsCall(J) && + DoesModifyCalleeSavedReg(I, QRI))) { + Dependence = true; + return false; + } + + // Two control flow instructions cannot go in the same packet. + if (IsControlFlow(I) && IsControlFlow(J)) { + Dependence = true; + return false; + } + + // A LoopN instruction cannot appear in the same packet as a jump or call. + if (IsLoopN(I) && ( IsDirectJump(J) + || MCIDJ.isCall() + || QII->isDeallocRet(J))) { + Dependence = true; + return false; + } + if (IsLoopN(J) && ( IsDirectJump(I) + || MCIDI.isCall() + || QII->isDeallocRet(I))) { + Dependence = true; + return false; + } + + // dealloc_return cannot appear in the same packet as a conditional or + // unconditional jump. + if (QII->isDeallocRet(I) && ( MCIDJ.isBranch() + || MCIDJ.isCall() + || MCIDJ.isBarrier())) { + Dependence = true; + return false; + } + + + // V4 allows dual store. But does not allow second store, if the + // first store is not in SLOT0. New value store, new value jump, + // dealloc_return and memop always take SLOT0. + // Arch spec 3.4.4.2 + if (QRI->Subtarget.hasV4TOps()) { + + if (MCIDI.mayStore() && MCIDJ.mayStore() && isNewValueInst(J)) { + Dependence = true; + return false; + } + + if ( (QII->isMemOp(J) && MCIDI.mayStore()) + || (MCIDJ.mayStore() && QII->isMemOp(I)) + || (QII->isMemOp(J) && QII->isMemOp(I))) { + Dependence = true; + return false; + } + + //if dealloc_return + if (MCIDJ.mayStore() && QII->isDeallocRet(I)){ + Dependence = true; + return false; + } + + // If an instruction feeds new value jump, glue it. + MachineBasicBlock::iterator NextMII = I; + ++NextMII; + MachineInstr *NextMI = NextMII; + + if (QII->isNewValueJump(NextMI)) { + + bool secondRegMatch = false; + bool maintainNewValueJump = false; + + if (NextMI->getOperand(1).isReg() && + I->getOperand(0).getReg() == NextMI->getOperand(1).getReg()) { + secondRegMatch = true; + maintainNewValueJump = true; + } + + if (!secondRegMatch && + I->getOperand(0).getReg() == NextMI->getOperand(0).getReg()) { + maintainNewValueJump = true; + } + + for (std::vector::iterator + VI = CurrentPacketMIs.begin(), + VE = CurrentPacketMIs.end(); + (VI != VE && maintainNewValueJump); ++VI) { + SUnit* PacketSU = MIToSUnit[*VI]; + + // NVJ can not be part of the dual jump - Arch Spec: section 7.8 + if (PacketSU->getInstr()->getDesc().isCall()) { + Dependence = true; + break; + } + // Validate + // 1. Packet does not have a store in it. + // 2. If the first operand of the nvj is newified, and the second + // operand is also a reg, it (second reg) is not defined in + // the same packet. + // 3. If the second operand of the nvj is newified, (which means + // first operand is also a reg), first reg is not defined in + // the same packet. + if (PacketSU->getInstr()->getDesc().mayStore() || + PacketSU->getInstr()->getOpcode() == Hexagon::ALLOCFRAME || + // Check #2. + (!secondRegMatch && NextMI->getOperand(1).isReg() && + PacketSU->getInstr()->modifiesRegister( + NextMI->getOperand(1).getReg(), QRI)) || + // Check #3. + (secondRegMatch && + PacketSU->getInstr()->modifiesRegister( + NextMI->getOperand(0).getReg(), QRI))) { + Dependence = true; + break; + } + } + if (!Dependence) + GlueToNewValueJump = true; + else + return false; + } + } + + if (SUJ->isSucc(SUI)) { + for (unsigned i = 0; + (i < SUJ->Succs.size()) && !FoundSequentialDependence; + ++i) { + + if (SUJ->Succs[i].getSUnit() != SUI) { + continue; + } + + SDep::Kind DepType = SUJ->Succs[i].getKind(); + + // For direct calls: + // Ignore register dependences for call instructions for + // packetization purposes except for those due to r31 and + // predicate registers. + // + // For indirect calls: + // Same as direct calls + check for true dependences to the register + // used in the indirect call. + // + // We completely ignore Order dependences for call instructions + // + // For returns: + // Ignore register dependences for return instructions like jumpr, + // dealloc return unless we have dependencies on the explicit uses + // of the registers used by jumpr (like r31) or dealloc return + // (like r29 or r30). + // + // TODO: Currently, jumpr is handling only return of r31. So, the + // following logic (specificaly IsCallDependent) is working fine. + // We need to enable jumpr for register other than r31 and then, + // we need to rework the last part, where it handles indirect call + // of that (IsCallDependent) function. Bug 6216 is opened for this. + // + unsigned DepReg; + const TargetRegisterClass* RC; + if (DepType == SDep::Data) { + DepReg = SUJ->Succs[i].getReg(); + RC = QRI->getMinimalPhysRegClass(DepReg); + } + if ((MCIDI.isCall() || MCIDI.isReturn()) && + (!IsRegDependence(DepType) || + !IsCallDependent(I, DepType, SUJ->Succs[i].getReg()))) { + /* do nothing */ + } + + // For instructions that can be promoted to dot-new, try to promote. + else if ((DepType == SDep::Data) && + CanPromoteToDotNew(I, SUJ, DepReg, MIToSUnit, II, RC) && + PromoteToDotNew(I, DepType, II, RC)) { + PromotedToDotNew = true; + /* do nothing */ + } + + else if ((DepType == SDep::Data) && + (QII->isNewValueJump(I))) { + /* do nothing */ + } + + // For predicated instructions, if the predicates are complements + // then there can be no dependence. + else if (QII->isPredicated(I) && + QII->isPredicated(J) && + ArePredicatesComplements(I, J, MIToSUnit)) { + /* do nothing */ + + } + else if (IsDirectJump(I) && + !MCIDJ.isBranch() && + !MCIDJ.isCall() && + (DepType == SDep::Order)) { + // Ignore Order dependences between unconditional direct branches + // and non-control-flow instructions + /* do nothing */ + } + else if (MCIDI.isConditionalBranch() && (DepType != SDep::Data) && + (DepType != SDep::Output)) { + // Ignore all dependences for jumps except for true and output + // dependences + /* do nothing */ + } + + // Ignore output dependences due to superregs. We can + // write to two different subregisters of R1:0 for instance + // in the same cycle + // + + // + // Let the + // If neither I nor J defines DepReg, then this is a + // superfluous output dependence. The dependence must be of the + // form: + // R0 = ... + // R1 = ... + // and there is an output dependence between the two instructions + // with + // DepReg = D0 + // We want to ignore these dependences. + // Ideally, the dependence constructor should annotate such + // dependences. We can then avoid this relatively expensive check. + // + else if (DepType == SDep::Output) { + // DepReg is the register that's responsible for the dependence. + unsigned DepReg = SUJ->Succs[i].getReg(); + + // Check if I and J really defines DepReg. + if (I->definesRegister(DepReg) || + J->definesRegister(DepReg)) { + FoundSequentialDependence = true; + break; + } + } + + // We ignore Order dependences for + // 1. Two loads unless they are volatile. + // 2. Two stores in V4 unless they are volatile. + else if ((DepType == SDep::Order) && + !I->hasVolatileMemoryRef() && + !J->hasVolatileMemoryRef()) { + if (QRI->Subtarget.hasV4TOps() && + // hexagonv4 allows dual store. + MCIDI.mayStore() && MCIDJ.mayStore()) { + /* do nothing */ + } + // store followed by store-- not OK on V2 + // store followed by load -- not OK on all (OK if addresses + // are not aliased) + // load followed by store -- OK on all + // load followed by load -- OK on all + else if ( !MCIDJ.mayStore()) { + /* do nothing */ + } + else { + FoundSequentialDependence = true; + break; + } + } + + // For V4, special case ALLOCFRAME. Even though there is dependency + // between ALLOCAFRAME and subsequent store, allow it to be + // packetized in a same packet. This implies that the store is using + // caller's SP. Hense, offset needs to be updated accordingly. + else if (DepType == SDep::Data + && QRI->Subtarget.hasV4TOps() + && J->getOpcode() == Hexagon::ALLOCFRAME + && (I->getOpcode() == Hexagon::STrid + || I->getOpcode() == Hexagon::STriw + || I->getOpcode() == Hexagon::STrib) + && I->getOperand(0).getReg() == QRI->getStackRegister() + && QII->isValidOffset(I->getOpcode(), + I->getOperand(1).getImm() - + (FrameSize + HEXAGON_LRFP_SIZE))) + { + GlueAllocframeStore = true; + // Since this store is to be glued with allocframe in the same + // packet, it will use SP of the previous stack frame, i.e + // caller's SP. Therefore, we need to recalculate offset according + // to this change. + I->getOperand(1).setImm(I->getOperand(1).getImm() - + (FrameSize + HEXAGON_LRFP_SIZE)); + } + + // + // Skip over anti-dependences. Two instructions that are + // anti-dependent can share a packet + // + else if (DepType != SDep::Anti) { + FoundSequentialDependence = true; + break; + } + } + + if (FoundSequentialDependence) { + Dependence = true; + return false; + } + } + + return true; +} + +// isLegalToPruneDependencies +bool HexagonPacketizerList::isLegalToPruneDependencies(SUnit *SUI, SUnit *SUJ) { + MachineInstr *I = SUI->getInstr(); + MachineInstr *J = SUJ->getInstr(); + assert(I && J && "Unable to packetize null instruction!"); + + const unsigned FrameSize = MF.getFrameInfo()->getStackSize(); + + if (Dependence) { + + // Check if the instruction was promoted to a dot-new. If so, demote it + // back into a dot-old. + if (PromotedToDotNew) { + DemoteToDotOld(I); + } + + // Check if the instruction (must be a store) was glued with an Allocframe + // instruction. If so, restore its offset to its original value, i.e. use + // curent SP instead of caller's SP. + if (GlueAllocframeStore) { + I->getOperand(1).setImm(I->getOperand(1).getImm() + + FrameSize + HEXAGON_LRFP_SIZE); + } + + return false; + } + return true; +} + +MachineBasicBlock::iterator HexagonPacketizerList::addToPacket(MachineInstr *MI) { + + MachineBasicBlock::iterator MII = MI; + MachineBasicBlock *MBB = MI->getParent(); + + const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII; + + if (GlueToNewValueJump) { + + ++MII; + MachineInstr *nvjMI = MII; + assert(ResourceTracker->canReserveResources(MI)); + ResourceTracker->reserveResources(MI); + if (QII->isExtended(MI) && + !tryAllocateResourcesForConstExt(MI)) { + endPacket(MBB, MI); + ResourceTracker->reserveResources(MI); + assert(canReserveResourcesForConstExt(MI) && + "Ensure that there is a slot"); + reserveResourcesForConstExt(MI); + // Reserve resources for new value jump constant extender. + assert(canReserveResourcesForConstExt(MI) && + "Ensure that there is a slot"); + reserveResourcesForConstExt(nvjMI); + assert(ResourceTracker->canReserveResources(nvjMI) && + "Ensure that there is a slot"); + + } else if ( // Extended instruction takes two slots in the packet. + // Try reserve and allocate 4-byte in the current packet first. + (QII->isExtended(nvjMI) + && (!tryAllocateResourcesForConstExt(nvjMI) + || !ResourceTracker->canReserveResources(nvjMI))) + || // For non-extended instruction, no need to allocate extra 4 bytes. + (!QII->isExtended(nvjMI) && !ResourceTracker->canReserveResources(nvjMI))) + { + endPacket(MBB, MI); + // A new and empty packet starts. + // We are sure that the resources requirements can be satisfied. + // Therefore, do not need to call "canReserveResources" anymore. + ResourceTracker->reserveResources(MI); + if (QII->isExtended(nvjMI)) + reserveResourcesForConstExt(nvjMI); + } + // Here, we are sure that "reserveResources" would succeed. + ResourceTracker->reserveResources(nvjMI); + CurrentPacketMIs.push_back(MI); + CurrentPacketMIs.push_back(nvjMI); + } else { + if ( QII->isExtended(MI) + && ( !tryAllocateResourcesForConstExt(MI) + || !ResourceTracker->canReserveResources(MI))) + { + endPacket(MBB, MI); + // Check if the instruction was promoted to a dot-new. If so, demote it + // back into a dot-old + if (PromotedToDotNew) { + DemoteToDotOld(MI); + } + reserveResourcesForConstExt(MI); + } + // In case that "MI" is not an extended insn, + // the resource availability has already been checked. + ResourceTracker->reserveResources(MI); + CurrentPacketMIs.push_back(MI); + } + return MII; +} + +//===----------------------------------------------------------------------===// +// Public Constructor Functions +//===----------------------------------------------------------------------===// + +FunctionPass *llvm::createHexagonPacketizer() { + return new HexagonPacketizer(); +} + diff --git a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp index ef3688138c9..f78965ea39f 100644 --- a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp +++ b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp @@ -15,6 +15,7 @@ #include "Hexagon.h" #include "HexagonAsmPrinter.h" #include "HexagonInstPrinter.h" +#include "HexagonMCInst.h" #include "llvm/MC/MCInst.h" #include "llvm/MC/MCAsmInfo.h" #include "llvm/MC/MCExpr.h" @@ -37,20 +38,50 @@ StringRef HexagonInstPrinter::getRegName(unsigned RegNo) const { void HexagonInstPrinter::printInst(const MCInst *MI, raw_ostream &O, StringRef Annot) { + printInst((const HexagonMCInst*)(MI), O, Annot); +} + +void HexagonInstPrinter::printInst(const HexagonMCInst *MI, raw_ostream &O, + StringRef Annot) { const char packetPadding[] = " "; const char startPacket = '{', endPacket = '}'; // TODO: add outer HW loop when it's supported too. if (MI->getOpcode() == Hexagon::ENDLOOP0) { - MCInst Nop; + // Ending a harware loop is different from ending an regular packet. + assert(MI->isEndPacket() && "Loop end must also end the packet"); - O << packetPadding << startPacket << '\n'; - Nop.setOpcode(Hexagon::NOP); - printInstruction(&Nop, O); - O << packetPadding << endPacket; + if (MI->isStartPacket()) { + // There must be a packet to end a loop. + // FIXME: when shuffling is always run, this shouldn't be needed. + HexagonMCInst Nop; + StringRef NoAnnot; + + Nop.setOpcode (Hexagon::NOP); + Nop.setStartPacket (MI->isStartPacket()); + printInst (&Nop, O, NoAnnot); + } + + // Close the packet. + if (MI->isEndPacket()) + O << packetPadding << endPacket; + + printInstruction(MI, O); + } + else { + // Prefix the insn opening the packet. + if (MI->isStartPacket()) + O << packetPadding << startPacket << '\n'; + + printInstruction(MI, O); + + // Suffix the insn closing the packet. + if (MI->isEndPacket()) + // Suffix the packet in a new line always, since the GNU assembler has + // issues with a closing brace on the same line as CONST{32,64}. + O << '\n' << packetPadding << endPacket; } - printInstruction(MI, O); printAnnotation(O, Annot); } @@ -69,18 +100,18 @@ void HexagonInstPrinter::printOperand(const MCInst *MI, unsigned OpNo, } } -void HexagonInstPrinter::printImmOperand - (const MCInst *MI, unsigned OpNo, raw_ostream &O) const { +void HexagonInstPrinter::printImmOperand(const MCInst *MI, unsigned OpNo, + raw_ostream &O) const { O << MI->getOperand(OpNo).getImm(); } void HexagonInstPrinter::printExtOperand(const MCInst *MI, unsigned OpNo, - raw_ostream &O) const { + raw_ostream &O) const { O << MI->getOperand(OpNo).getImm(); } -void HexagonInstPrinter::printUnsignedImmOperand - (const MCInst *MI, unsigned OpNo, raw_ostream &O) const { +void HexagonInstPrinter::printUnsignedImmOperand(const MCInst *MI, unsigned OpNo, + raw_ostream &O) const { O << MI->getOperand(OpNo).getImm(); } @@ -89,13 +120,13 @@ void HexagonInstPrinter::printNegImmOperand(const MCInst *MI, unsigned OpNo, O << -MI->getOperand(OpNo).getImm(); } -void HexagonInstPrinter::printNOneImmOperand - (const MCInst *MI, unsigned OpNo, raw_ostream &O) const { +void HexagonInstPrinter::printNOneImmOperand(const MCInst *MI, unsigned OpNo, + raw_ostream &O) const { O << -1; } -void HexagonInstPrinter::printMEMriOperand - (const MCInst *MI, unsigned OpNo, raw_ostream &O) const { +void HexagonInstPrinter::printMEMriOperand(const MCInst *MI, unsigned OpNo, + raw_ostream &O) const { const MCOperand& MO0 = MI->getOperand(OpNo); const MCOperand& MO1 = MI->getOperand(OpNo + 1); @@ -103,8 +134,8 @@ void HexagonInstPrinter::printMEMriOperand O << " + #" << MO1.getImm(); } -void HexagonInstPrinter::printFrameIndexOperand - (const MCInst *MI, unsigned OpNo, raw_ostream &O) const { +void HexagonInstPrinter::printFrameIndexOperand(const MCInst *MI, unsigned OpNo, + raw_ostream &O) const { const MCOperand& MO0 = MI->getOperand(OpNo); const MCOperand& MO1 = MI->getOperand(OpNo + 1); diff --git a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h index dad4334c3eb..3ce7dfcbdbe 100644 --- a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h +++ b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h @@ -14,6 +14,7 @@ #ifndef HEXAGONINSTPRINTER_H #define HEXAGONINSTPRINTER_H +#include "HexagonMCInst.h" #include "llvm/MC/MCInstPrinter.h" namespace llvm { @@ -25,6 +26,7 @@ namespace llvm { : MCInstPrinter(MAI, MII, MRI) {} virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot); + void printInst(const HexagonMCInst *MI, raw_ostream &O, StringRef Annot); virtual StringRef getOpcodeName(unsigned Opcode) const; void printInstruction(const MCInst *MI, raw_ostream &O); StringRef getRegName(unsigned RegNo) const; diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h b/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h index ed55c3c1c15..7221e906342 100644 --- a/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h +++ b/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h @@ -23,14 +23,41 @@ namespace llvm { /// instruction info tracks. /// namespace HexagonII { - // *** The code below must match HexagonInstrFormat*.td *** // + // Insn types. + // *** Must match HexagonInstrFormat*.td *** + enum Type { + TypePSEUDO = 0, + TypeALU32 = 1, + TypeCR = 2, + TypeJR = 3, + TypeJ = 4, + TypeLD = 5, + TypeST = 6, + TypeSYSTEM = 7, + TypeXTYPE = 8, + TypeMEMOP = 9, + TypeNV = 10, + TypePREFIX = 30, // Such as extenders. + TypeMARKER = 31 // Such as end of a HW loop. + }; + + + // MCInstrDesc TSFlags + // *** Must match HexagonInstrFormat*.td *** enum { + // This 5-bit field describes the insn type. + TypePos = 0, + TypeMask = 0x1f, + + // Solo instructions. + SoloPos = 5, + SoloMask = 0x1, // Predicated instructions. - PredicatedPos = 1, + PredicatedPos = 6, PredicatedMask = 0x1 };