aboutsummaryrefslogtreecommitdiffstats
path: root/lib/CodeGen
diff options
context:
space:
mode:
Diffstat (limited to 'lib/CodeGen')
-rw-r--r--lib/CodeGen/AggressiveAntiDepBreaker.cpp8
-rw-r--r--lib/CodeGen/Analysis.cpp67
-rw-r--r--lib/CodeGen/AsmPrinter/AsmPrinter.cpp32
-rw-r--r--lib/CodeGen/AsmPrinter/CMakeLists.txt10
-rw-r--r--lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp16
-rw-r--r--lib/CodeGen/AsmPrinter/DwarfDebug.cpp6
-rw-r--r--lib/CodeGen/AsmPrinter/DwarfException.cpp4
-rw-r--r--lib/CodeGen/AsmPrinter/LLVMBuild.txt1
-rw-r--r--lib/CodeGen/BranchFolding.cpp20
-rw-r--r--lib/CodeGen/CMakeLists.txt15
-rw-r--r--lib/CodeGen/CodeGen.cpp1
-rw-r--r--lib/CodeGen/CriticalAntiDepBreaker.cpp8
-rw-r--r--lib/CodeGen/DFAPacketizer.cpp98
-rw-r--r--lib/CodeGen/DeadMachineInstructionElim.cpp2
-rw-r--r--lib/CodeGen/ExecutionDepsFix.cpp2
-rw-r--r--lib/CodeGen/ExpandISelPseudos.cpp3
-rw-r--r--lib/CodeGen/ExpandPostRAPseudos.cpp2
-rw-r--r--lib/CodeGen/GCStrategy.cpp2
-rw-r--r--lib/CodeGen/IfConversion.cpp12
-rw-r--r--lib/CodeGen/InlineSpiller.cpp8
-rw-r--r--lib/CodeGen/LLVMBuild.txt4
-rw-r--r--lib/CodeGen/LLVMTargetMachine.cpp20
-rw-r--r--lib/CodeGen/LiveDebugVariables.cpp4
-rw-r--r--lib/CodeGen/LiveIntervalAnalysis.cpp2
-rw-r--r--lib/CodeGen/LiveRangeEdit.cpp22
-rw-r--r--lib/CodeGen/LiveRangeEdit.h7
-rw-r--r--lib/CodeGen/LiveVariables.cpp6
-rw-r--r--lib/CodeGen/MachineBasicBlock.cpp147
-rw-r--r--lib/CodeGen/MachineBlockPlacement.cpp156
-rw-r--r--lib/CodeGen/MachineCSE.cpp9
-rw-r--r--lib/CodeGen/MachineInstr.cpp116
-rw-r--r--lib/CodeGen/MachineInstrBundle.cpp180
-rw-r--r--lib/CodeGen/MachineLICM.cpp13
-rw-r--r--lib/CodeGen/MachineSSAUpdater.cpp6
-rw-r--r--lib/CodeGen/MachineSink.cpp164
-rw-r--r--lib/CodeGen/MachineVerifier.cpp36
-rw-r--r--lib/CodeGen/PHIElimination.cpp2
-rw-r--r--lib/CodeGen/PeepholeOptimizer.cpp10
-rw-r--r--lib/CodeGen/PostRASchedulerList.cpp7
-rw-r--r--lib/CodeGen/PrologEpilogInserter.cpp10
-rw-r--r--lib/CodeGen/RegAllocFast.cpp6
-rw-r--r--lib/CodeGen/RegAllocPBQP.cpp16
-rw-r--r--lib/CodeGen/RegisterCoalescer.cpp10
-rw-r--r--lib/CodeGen/ScheduleDAGInstrs.cpp58
-rw-r--r--lib/CodeGen/SelectionDAG/CMakeLists.txt10
-rw-r--r--lib/CodeGen/SelectionDAG/DAGCombiner.cpp135
-rw-r--r--lib/CodeGen/SelectionDAG/FastISel.cpp33
-rw-r--r--lib/CodeGen/SelectionDAG/InstrEmitter.cpp2
-rw-r--r--lib/CodeGen/SelectionDAG/LLVMBuild.txt1
-rw-r--r--lib/CodeGen/SelectionDAG/LegalizeDAG.cpp20
-rw-r--r--lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp32
-rw-r--r--lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp4
-rw-r--r--lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp4
-rw-r--r--lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp92
-rw-r--r--lib/CodeGen/SelectionDAG/SelectionDAG.cpp76
-rw-r--r--lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp137
-rw-r--r--lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h5
-rw-r--r--lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp177
-rw-r--r--lib/CodeGen/SelectionDAG/TargetLowering.cpp40
-rw-r--r--lib/CodeGen/ShrinkWrapping.cpp2
-rw-r--r--lib/CodeGen/SjLjEHPrepare.cpp41
-rw-r--r--lib/CodeGen/SplitKit.cpp2
-rw-r--r--lib/CodeGen/Splitter.cpp827
-rw-r--r--lib/CodeGen/Splitter.h101
-rw-r--r--lib/CodeGen/TailDuplication.cpp13
-rw-r--r--lib/CodeGen/TargetFrameLoweringImpl.cpp45
-rw-r--r--lib/CodeGen/TargetInstrInfoImpl.cpp68
-rw-r--r--lib/CodeGen/TargetLoweringObjectFileImpl.cpp4
-rw-r--r--lib/CodeGen/TargetOptionsImpl.cpp52
-rw-r--r--lib/CodeGen/TwoAddressInstructionPass.cpp34
70 files changed, 1759 insertions, 1526 deletions
diff --git a/lib/CodeGen/AggressiveAntiDepBreaker.cpp b/lib/CodeGen/AggressiveAntiDepBreaker.cpp
index 25842a7..6cf4571 100644
--- a/lib/CodeGen/AggressiveAntiDepBreaker.cpp
+++ b/lib/CodeGen/AggressiveAntiDepBreaker.cpp
@@ -148,7 +148,7 @@ void AggressiveAntiDepBreaker::StartBlock(MachineBasicBlock *BB) {
assert(State == NULL);
State = new AggressiveAntiDepState(TRI->getNumRegs(), BB);
- bool IsReturnBlock = (!BB->empty() && BB->back().getDesc().isReturn());
+ bool IsReturnBlock = (!BB->empty() && BB->back().isReturn());
std::vector<unsigned> &KillIndices = State->GetKillIndices();
std::vector<unsigned> &DefIndices = State->GetDefIndices();
@@ -384,7 +384,7 @@ void AggressiveAntiDepBreaker::PrescanInstruction(MachineInstr *MI,
// If MI's defs have a special allocation requirement, don't allow
// any def registers to be changed. Also assume all registers
// defined in a call must not be changed (ABI).
- if (MI->getDesc().isCall() || MI->getDesc().hasExtraDefRegAllocReq() ||
+ if (MI->isCall() || MI->hasExtraDefRegAllocReq() ||
TII->isPredicated(MI)) {
DEBUG(if (State->GetGroup(Reg) != 0) dbgs() << "->g0(alloc-req)");
State->UnionGroups(Reg, 0);
@@ -451,8 +451,8 @@ void AggressiveAntiDepBreaker::ScanInstruction(MachineInstr *MI,
// instruction which may not be executed. The second R6 def may or may not
// re-define R6 so it's not safe to change it since the last R6 use cannot be
// changed.
- bool Special = MI->getDesc().isCall() ||
- MI->getDesc().hasExtraSrcRegAllocReq() ||
+ bool Special = MI->isCall() ||
+ MI->hasExtraSrcRegAllocReq() ||
TII->isPredicated(MI);
// Scan the register uses for this instruction and update
diff --git a/lib/CodeGen/Analysis.cpp b/lib/CodeGen/Analysis.cpp
index fafc010..0c84be5 100644
--- a/lib/CodeGen/Analysis.cpp
+++ b/lib/CodeGen/Analysis.cpp
@@ -1,4 +1,4 @@
-//===-- Analysis.cpp - CodeGen LLVM IR Analysis Utilities --*- C++ ------*-===//
+//===-- Analysis.cpp - CodeGen LLVM IR Analysis Utilities -----------------===//
//
// The LLVM Compiler Infrastructure
//
@@ -12,6 +12,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/Analysis.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/Instructions.h"
@@ -149,33 +150,40 @@ llvm::hasInlineAsmMemConstraint(InlineAsm::ConstraintInfoVector &CInfos,
/// consideration of global floating-point math flags.
///
ISD::CondCode llvm::getFCmpCondCode(FCmpInst::Predicate Pred) {
- ISD::CondCode FPC, FOC;
switch (Pred) {
- case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
- case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
- case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
- case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
- case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
- case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
- case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
- case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
- case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
- case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
- case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
- case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
- case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
- case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
- case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
- case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
- default:
- llvm_unreachable("Invalid FCmp predicate opcode!");
- FOC = FPC = ISD::SETFALSE;
- break;
+ case FCmpInst::FCMP_FALSE: return ISD::SETFALSE;
+ case FCmpInst::FCMP_OEQ: return ISD::SETOEQ;
+ case FCmpInst::FCMP_OGT: return ISD::SETOGT;
+ case FCmpInst::FCMP_OGE: return ISD::SETOGE;
+ case FCmpInst::FCMP_OLT: return ISD::SETOLT;
+ case FCmpInst::FCMP_OLE: return ISD::SETOLE;
+ case FCmpInst::FCMP_ONE: return ISD::SETONE;
+ case FCmpInst::FCMP_ORD: return ISD::SETO;
+ case FCmpInst::FCMP_UNO: return ISD::SETUO;
+ case FCmpInst::FCMP_UEQ: return ISD::SETUEQ;
+ case FCmpInst::FCMP_UGT: return ISD::SETUGT;
+ case FCmpInst::FCMP_UGE: return ISD::SETUGE;
+ case FCmpInst::FCMP_ULT: return ISD::SETULT;
+ case FCmpInst::FCMP_ULE: return ISD::SETULE;
+ case FCmpInst::FCMP_UNE: return ISD::SETUNE;
+ case FCmpInst::FCMP_TRUE: return ISD::SETTRUE;
+ default: break;
+ }
+ llvm_unreachable("Invalid FCmp predicate opcode!");
+ return ISD::SETFALSE;
+}
+
+ISD::CondCode llvm::getFCmpCodeWithoutNaN(ISD::CondCode CC) {
+ switch (CC) {
+ case ISD::SETOEQ: case ISD::SETUEQ: return ISD::SETEQ;
+ case ISD::SETONE: case ISD::SETUNE: return ISD::SETNE;
+ case ISD::SETOLT: case ISD::SETULT: return ISD::SETLT;
+ case ISD::SETOLE: case ISD::SETULE: return ISD::SETLE;
+ case ISD::SETOGT: case ISD::SETUGT: return ISD::SETGT;
+ case ISD::SETOGE: case ISD::SETUGE: return ISD::SETGE;
+ default: break;
}
- if (NoNaNsFPMath)
- return FOC;
- else
- return FPC;
+ return CC;
}
/// getICmpCondCode - Return the ISD condition code corresponding to
@@ -221,12 +229,13 @@ bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
- (!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
+ (!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
+ !isa<UnreachableInst>(Term))) return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
- !I->isSafeToSpeculativelyExecute())
+ !isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
@@ -235,7 +244,7 @@ bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
- !BBI->isSafeToSpeculativelyExecute())
+ !isSafeToSpeculativelyExecute(BBI))
return false;
}
diff --git a/lib/CodeGen/AsmPrinter/AsmPrinter.cpp b/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
index 711b796..0c4d0d5 100644
--- a/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ b/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -1308,7 +1308,7 @@ void AsmPrinter::EmitXXStructorList(const Constant *List) {
}
// Emit the function pointers in reverse priority order.
- switch (MAI->getStructorOutputOrder()) {
+ switch (getObjFileLowering().getStructorOutputOrder()) {
case Structors::None:
break;
case Structors::PriorityOrder:
@@ -1659,6 +1659,28 @@ static void EmitGlobalConstantVector(const ConstantVector *CV,
AP.OutStreamer.EmitZeros(Padding, AddrSpace);
}
+static void LowerVectorConstant(const Constant *CV, unsigned AddrSpace,
+ AsmPrinter &AP) {
+ // Look through bitcasts
+ if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV))
+ if (CE->getOpcode() == Instruction::BitCast)
+ CV = CE->getOperand(0);
+
+ if (const ConstantVector *V = dyn_cast<ConstantVector>(CV))
+ return EmitGlobalConstantVector(V, AddrSpace, AP);
+
+ // If we get here, we're stuck; report the problem to the user.
+ // FIXME: Are there any other useful tricks for vectors?
+ {
+ std::string S;
+ raw_string_ostream OS(S);
+ OS << "Unsupported vector expression in static initializer: ";
+ WriteAsOperand(OS, CV, /*PrintType=*/false,
+ !AP.MF ? 0 : AP.MF->getFunction()->getParent());
+ report_fatal_error(OS.str());
+ }
+}
+
static void EmitGlobalConstantStruct(const ConstantStruct *CS,
unsigned AddrSpace, AsmPrinter &AP) {
// Print the fields in successive locations. Pad to align if needed!
@@ -1813,8 +1835,8 @@ static void EmitGlobalConstantImpl(const Constant *CV, unsigned AddrSpace,
return;
}
- if (const ConstantVector *V = dyn_cast<ConstantVector>(CV))
- return EmitGlobalConstantVector(V, AddrSpace, AP);
+ if (CV->getType()->isVectorTy())
+ return LowerVectorConstant(CV, AddrSpace, AP);
// Otherwise, it must be a ConstantExpr. Lower it to an MCExpr, then emit it
// thread the streamer with EmitValue.
@@ -1987,7 +2009,7 @@ static void EmitBasicBlockLoopComments(const MachineBasicBlock &MBB,
void AsmPrinter::EmitBasicBlockStart(const MachineBasicBlock *MBB) const {
// Emit an alignment directive for this block, if needed.
if (unsigned Align = MBB->getAlignment())
- EmitAlignment(Log2_32(Align));
+ EmitAlignment(Align);
// If the block has its address taken, emit any labels that were used to
// reference the block. It is possible that there is more than one label
@@ -2082,7 +2104,7 @@ isBlockOnlyReachableByFallthrough(const MachineBasicBlock *MBB) const {
MachineInstr &MI = *II;
// If it is not a simple branch, we are in a table somewhere.
- if (!MI.getDesc().isBranch() || MI.getDesc().isIndirectBranch())
+ if (!MI.isBranch() || MI.isIndirectBranch())
return false;
// If we are the operands of one of the branches, this is not
diff --git a/lib/CodeGen/AsmPrinter/CMakeLists.txt b/lib/CodeGen/AsmPrinter/CMakeLists.txt
index f6ce17d..58fe2ed 100644
--- a/lib/CodeGen/AsmPrinter/CMakeLists.txt
+++ b/lib/CodeGen/AsmPrinter/CMakeLists.txt
@@ -12,13 +12,3 @@ add_llvm_library(LLVMAsmPrinter
OcamlGCPrinter.cpp
Win64Exception.cpp
)
-
-add_llvm_library_dependencies(LLVMAsmPrinter
- LLVMAnalysis
- LLVMCodeGen
- LLVMCore
- LLVMMC
- LLVMMCParser
- LLVMSupport
- LLVMTarget
- )
diff --git a/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp b/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
index 237998a..8cb5156 100644
--- a/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
+++ b/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
@@ -791,13 +791,13 @@ void CompileUnit::constructTypeDIE(DIE &Buffer, DICompositeType CTy) {
DISubprogram SP(Element);
ElemDie = getOrCreateSubprogramDIE(DISubprogram(Element));
if (SP.isProtected())
- addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_protected);
else if (SP.isPrivate())
- addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_private);
else
- addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(ElemDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_public);
if (SP.isExplicit())
addUInt(ElemDie, dwarf::DW_AT_explicit, dwarf::DW_FORM_flag, 1);
@@ -988,7 +988,7 @@ DIE *CompileUnit::getOrCreateSubprogramDIE(DISubprogram SP) {
unsigned VK = SP.getVirtuality();
if (VK) {
- addUInt(SPDie, dwarf::DW_AT_virtuality, dwarf::DW_FORM_flag, VK);
+ addUInt(SPDie, dwarf::DW_AT_virtuality, dwarf::DW_FORM_data1, VK);
DIEBlock *Block = getDIEBlock();
addUInt(Block, 0, dwarf::DW_FORM_data1, dwarf::DW_OP_constu);
addUInt(Block, 0, dwarf::DW_FORM_udata, SP.getVirtualIndex());
@@ -1398,17 +1398,17 @@ DIE *CompileUnit::createMemberDIE(DIDerivedType DT) {
addBlock(MemberDie, dwarf::DW_AT_data_member_location, 0, MemLocationDie);
if (DT.isProtected())
- addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_protected);
else if (DT.isPrivate())
- addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_private);
// Otherwise C++ member and base classes are considered public.
else
- addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_flag,
+ addUInt(MemberDie, dwarf::DW_AT_accessibility, dwarf::DW_FORM_data1,
dwarf::DW_ACCESS_public);
if (DT.isVirtual())
- addUInt(MemberDie, dwarf::DW_AT_virtuality, dwarf::DW_FORM_flag,
+ addUInt(MemberDie, dwarf::DW_AT_virtuality, dwarf::DW_FORM_data1,
dwarf::DW_VIRTUALITY_virtual);
// Objective-C properties.
diff --git a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp
index dc46a58..a3db96a 100644
--- a/lib/CodeGen/AsmPrinter/DwarfDebug.cpp
+++ b/lib/CodeGen/AsmPrinter/DwarfDebug.cpp
@@ -442,6 +442,10 @@ DIE *DwarfDebug::constructInlinedScopeDIE(CompileUnit *TheCU,
TheCU->addUInt(ScopeDIE, dwarf::DW_AT_call_file, 0, TheCU->getID());
TheCU->addUInt(ScopeDIE, dwarf::DW_AT_call_line, 0, DL.getLineNumber());
+ // Add name to the name table, we do this here because we're guaranteed
+ // to have concrete versions of our DW_TAG_inlined_subprogram nodes.
+ addSubprogramNames(TheCU, InlinedSP, ScopeDIE);
+
return ScopeDIE;
}
@@ -1414,7 +1418,7 @@ void DwarfDebug::endFunction(const MachineFunction *MF) {
DIE *CurFnDIE = constructScopeDIE(TheCU, FnScope);
- if (!DisableFramePointerElim(*MF))
+ if (!MF->getTarget().Options.DisableFramePointerElim(*MF))
TheCU->addUInt(CurFnDIE, dwarf::DW_AT_APPLE_omit_frame_ptr,
dwarf::DW_FORM_flag, 1);
diff --git a/lib/CodeGen/AsmPrinter/DwarfException.cpp b/lib/CodeGen/AsmPrinter/DwarfException.cpp
index e0f2e85..bf7f7ee 100644
--- a/lib/CodeGen/AsmPrinter/DwarfException.cpp
+++ b/lib/CodeGen/AsmPrinter/DwarfException.cpp
@@ -184,7 +184,7 @@ ComputeActionsTable(const SmallVectorImpl<const LandingPadInfo*> &LandingPads,
/// CallToNoUnwindFunction - Return `true' if this is a call to a function
/// marked `nounwind'. Return `false' otherwise.
bool DwarfException::CallToNoUnwindFunction(const MachineInstr *MI) {
- assert(MI->getDesc().isCall() && "This should be a call instruction!");
+ assert(MI->isCall() && "This should be a call instruction!");
bool MarkedNoUnwind = false;
bool SawFunc = false;
@@ -243,7 +243,7 @@ ComputeCallSiteTable(SmallVectorImpl<CallSiteEntry> &CallSites,
for (MachineBasicBlock::const_iterator MI = I->begin(), E = I->end();
MI != E; ++MI) {
if (!MI->isLabel()) {
- if (MI->getDesc().isCall())
+ if (MI->isCall())
SawPotentiallyThrowing |= !CallToNoUnwindFunction(MI);
continue;
}
diff --git a/lib/CodeGen/AsmPrinter/LLVMBuild.txt b/lib/CodeGen/AsmPrinter/LLVMBuild.txt
index 0f2059f..20b1f7b 100644
--- a/lib/CodeGen/AsmPrinter/LLVMBuild.txt
+++ b/lib/CodeGen/AsmPrinter/LLVMBuild.txt
@@ -20,4 +20,3 @@ type = Library
name = AsmPrinter
parent = Libraries
required_libraries = Analysis CodeGen Core MC MCParser Support Target
-
diff --git a/lib/CodeGen/BranchFolding.cpp b/lib/CodeGen/BranchFolding.cpp
index 5dec368..89894c3 100644
--- a/lib/CodeGen/BranchFolding.cpp
+++ b/lib/CodeGen/BranchFolding.cpp
@@ -432,10 +432,9 @@ static unsigned EstimateRuntime(MachineBasicBlock::iterator I,
for (; I != E; ++I) {
if (I->isDebugValue())
continue;
- const MCInstrDesc &MCID = I->getDesc();
- if (MCID.isCall())
+ if (I->isCall())
Time += 10;
- else if (MCID.mayLoad() || MCID.mayStore())
+ else if (I->mayLoad() || I->mayStore())
Time += 2;
else
++Time;
@@ -502,7 +501,7 @@ static unsigned CountTerminators(MachineBasicBlock *MBB,
break;
}
--I;
- if (!I->getDesc().isTerminator()) break;
+ if (!I->isTerminator()) break;
++NumTerms;
}
return NumTerms;
@@ -550,8 +549,8 @@ static bool ProfitableToMerge(MachineBasicBlock *MBB1,
// heuristics.
unsigned EffectiveTailLen = CommonTailLen;
if (SuccBB && MBB1 != PredBB && MBB2 != PredBB &&
- !MBB1->back().getDesc().isBarrier() &&
- !MBB2->back().getDesc().isBarrier())
+ !MBB1->back().isBarrier() &&
+ !MBB2->back().isBarrier())
++EffectiveTailLen;
// Check if the common tail is long enough to be worthwhile.
@@ -927,8 +926,9 @@ bool BranchFolder::TailMergeBlocks(MachineFunction &MF) {
if (MergePotentials.size() >= 2)
MadeChange |= TryTailMergeBlocks(IBB, PredBB);
// Reinsert an unconditional branch if needed.
- // The 1 below can occur as a result of removing blocks in TryTailMergeBlocks.
- PredBB = prior(I); // this may have been changed in TryTailMergeBlocks
+ // The 1 below can occur as a result of removing blocks in
+ // TryTailMergeBlocks.
+ PredBB = prior(I); // this may have been changed in TryTailMergeBlocks
if (MergePotentials.size() == 1 &&
MergePotentials.begin()->getBlock() != PredBB)
FixTail(MergePotentials.begin()->getBlock(), IBB, TII);
@@ -983,7 +983,7 @@ static bool IsBranchOnlyBlock(MachineBasicBlock *MBB) {
if (!MBBI->isDebugValue())
break;
}
- return (MBBI->getDesc().isBranch());
+ return (MBBI->isBranch());
}
/// IsBetterFallthrough - Return true if it would be clearly better to
@@ -1011,7 +1011,7 @@ static bool IsBetterFallthrough(MachineBasicBlock *MBB1,
MachineBasicBlock::iterator MBB2I = --MBB2->end();
while (MBB2I->isDebugValue())
--MBB2I;
- return MBB2I->getDesc().isCall() && !MBB1I->getDesc().isCall();
+ return MBB2I->isCall() && !MBB1I->isCall();
}
/// OptimizeBlock - Analyze and optimize control flow related to the specified
diff --git a/lib/CodeGen/CMakeLists.txt b/lib/CodeGen/CMakeLists.txt
index c8d4dcf..7aee3bb 100644
--- a/lib/CodeGen/CMakeLists.txt
+++ b/lib/CodeGen/CMakeLists.txt
@@ -9,6 +9,7 @@ add_llvm_library(LLVMCodeGen
CodePlacementOpt.cpp
CriticalAntiDepBreaker.cpp
DeadMachineInstructionElim.cpp
+ DFAPacketizer.cpp
DwarfEHPrepare.cpp
EdgeBundles.cpp
ELFCodeEmitter.cpp
@@ -46,6 +47,7 @@ add_llvm_library(LLVMCodeGen
MachineFunctionPass.cpp
MachineFunctionPrinterPass.cpp
MachineInstr.cpp
+ MachineInstrBundle.cpp
MachineLICM.cpp
MachineLoopInfo.cpp
MachineLoopRanges.cpp
@@ -87,27 +89,18 @@ add_llvm_library(LLVMCodeGen
Spiller.cpp
SpillPlacement.cpp
SplitKit.cpp
- Splitter.cpp
StackProtector.cpp
StackSlotColoring.cpp
StrongPHIElimination.cpp
TailDuplication.cpp
+ TargetFrameLoweringImpl.cpp
TargetInstrInfoImpl.cpp
TargetLoweringObjectFileImpl.cpp
+ TargetOptionsImpl.cpp
TwoAddressInstructionPass.cpp
UnreachableBlockElim.cpp
VirtRegMap.cpp
)
-add_llvm_library_dependencies(LLVMCodeGen
- LLVMAnalysis
- LLVMCore
- LLVMMC
- LLVMScalarOpts
- LLVMSupport
- LLVMTarget
- LLVMTransformUtils
- )
-
add_subdirectory(SelectionDAG)
add_subdirectory(AsmPrinter)
diff --git a/lib/CodeGen/CodeGen.cpp b/lib/CodeGen/CodeGen.cpp
index 3112c22..48b71d9 100644
--- a/lib/CodeGen/CodeGen.cpp
+++ b/lib/CodeGen/CodeGen.cpp
@@ -45,7 +45,6 @@ void llvm::initializeCodeGen(PassRegistry &Registry) {
initializeRegisterCoalescerPass(Registry);
initializeRenderMachineFunctionPass(Registry);
initializeSlotIndexesPass(Registry);
- initializeLoopSplitterPass(Registry);
initializeStackProtectorPass(Registry);
initializeStackSlotColoringPass(Registry);
initializeStrongPHIEliminationPass(Registry);
diff --git a/lib/CodeGen/CriticalAntiDepBreaker.cpp b/lib/CodeGen/CriticalAntiDepBreaker.cpp
index 84c4d59..128143e 100644
--- a/lib/CodeGen/CriticalAntiDepBreaker.cpp
+++ b/lib/CodeGen/CriticalAntiDepBreaker.cpp
@@ -54,7 +54,7 @@ void CriticalAntiDepBreaker::StartBlock(MachineBasicBlock *BB) {
// Clear "do not change" set.
KeepRegs.clear();
- bool IsReturnBlock = (!BB->empty() && BB->back().getDesc().isReturn());
+ bool IsReturnBlock = (!BB->empty() && BB->back().isReturn());
// Determine the live-out physregs for this block.
if (IsReturnBlock) {
@@ -193,8 +193,8 @@ void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) {
// instruction which may not be executed. The second R6 def may or may not
// re-define R6 so it's not safe to change it since the last R6 use cannot be
// changed.
- bool Special = MI->getDesc().isCall() ||
- MI->getDesc().hasExtraSrcRegAllocReq() ||
+ bool Special = MI->isCall() ||
+ MI->hasExtraSrcRegAllocReq() ||
TII->isPredicated(MI);
// Scan the register operands for this instruction and update
@@ -572,7 +572,7 @@ BreakAntiDependencies(const std::vector<SUnit>& SUnits,
// If MI's defs have a special allocation requirement, don't allow
// any def registers to be changed. Also assume all registers
// defined in a call must not be changed (ABI).
- if (MI->getDesc().isCall() || MI->getDesc().hasExtraDefRegAllocReq() ||
+ if (MI->isCall() || MI->hasExtraDefRegAllocReq() ||
TII->isPredicated(MI))
// If this instruction's defs have special allocation requirement, don't
// break this anti-dependency.
diff --git a/lib/CodeGen/DFAPacketizer.cpp b/lib/CodeGen/DFAPacketizer.cpp
new file mode 100644
index 0000000..16276bd
--- /dev/null
+++ b/lib/CodeGen/DFAPacketizer.cpp
@@ -0,0 +1,98 @@
+//=- llvm/CodeGen/DFAPacketizer.cpp - DFA Packetizer for VLIW -*- C++ -*-=====//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// This class implements a deterministic finite automaton (DFA) based
+// packetizing mechanism for VLIW architectures. It provides APIs to
+// determine whether there exists a legal mapping of instructions to
+// functional unit assignments in a packet. The DFA is auto-generated from
+// the target's Schedule.td file.
+//
+// A DFA consists of 3 major elements: states, inputs, and transitions. For
+// the packetizing mechanism, the input is the set of instruction classes for
+// a target. The state models all possible combinations of functional unit
+// consumption for a given set of instructions in a packet. A transition
+// models the addition of an instruction to a packet. In the DFA constructed
+// by this class, if an instruction can be added to a packet, then a valid
+// transition exists from the corresponding state. Invalid transitions
+// indicate that the instruction cannot be added to the current packet.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/DFAPacketizer.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/MC/MCInstrItineraries.h"
+using namespace llvm;
+
+DFAPacketizer::DFAPacketizer(const InstrItineraryData *I, const int (*SIT)[2],
+ const unsigned *SET):
+ InstrItins(I), CurrentState(0), DFAStateInputTable(SIT),
+ DFAStateEntryTable(SET) {}
+
+
+//
+// ReadTable - Read the DFA transition table and update CachedTable.
+//
+// Format of the transition tables:
+// DFAStateInputTable[][2] = pairs of <Input, Transition> for all valid
+// transitions
+// DFAStateEntryTable[i] = Index of the first entry in DFAStateInputTable
+// for the ith state
+//
+void DFAPacketizer::ReadTable(unsigned int state) {
+ unsigned ThisState = DFAStateEntryTable[state];
+ unsigned NextStateInTable = DFAStateEntryTable[state+1];
+ // Early exit in case CachedTable has already contains this
+ // state's transitions.
+ if (CachedTable.count(UnsignPair(state,
+ DFAStateInputTable[ThisState][0])))
+ return;
+
+ for (unsigned i = ThisState; i < NextStateInTable; i++)
+ CachedTable[UnsignPair(state, DFAStateInputTable[i][0])] =
+ DFAStateInputTable[i][1];
+}
+
+
+// canReserveResources - Check if the resources occupied by a MCInstrDesc
+// are available in the current state.
+bool DFAPacketizer::canReserveResources(const llvm::MCInstrDesc *MID) {
+ unsigned InsnClass = MID->getSchedClass();
+ const llvm::InstrStage *IS = InstrItins->beginStage(InsnClass);
+ unsigned FuncUnits = IS->getUnits();
+ UnsignPair StateTrans = UnsignPair(CurrentState, FuncUnits);
+ ReadTable(CurrentState);
+ return (CachedTable.count(StateTrans) != 0);
+}
+
+
+// reserveResources - Reserve the resources occupied by a MCInstrDesc and
+// change the current state to reflect that change.
+void DFAPacketizer::reserveResources(const llvm::MCInstrDesc *MID) {
+ unsigned InsnClass = MID->getSchedClass();
+ const llvm::InstrStage *IS = InstrItins->beginStage(InsnClass);
+ unsigned FuncUnits = IS->getUnits();
+ UnsignPair StateTrans = UnsignPair(CurrentState, FuncUnits);
+ ReadTable(CurrentState);
+ assert(CachedTable.count(StateTrans) != 0);
+ CurrentState = CachedTable[StateTrans];
+}
+
+
+// canReserveResources - Check if the resources occupied by a machine
+// instruction are available in the current state.
+bool DFAPacketizer::canReserveResources(llvm::MachineInstr *MI) {
+ const llvm::MCInstrDesc &MID = MI->getDesc();
+ return canReserveResources(&MID);
+}
+
+// reserveResources - Reserve the resources occupied by a machine
+// instruction and change the current state to reflect that change.
+void DFAPacketizer::reserveResources(llvm::MachineInstr *MI) {
+ const llvm::MCInstrDesc &MID = MI->getDesc();
+ reserveResources(&MID);
+}
diff --git a/lib/CodeGen/DeadMachineInstructionElim.cpp b/lib/CodeGen/DeadMachineInstructionElim.cpp
index 6de6c0c..ba135e1 100644
--- a/lib/CodeGen/DeadMachineInstructionElim.cpp
+++ b/lib/CodeGen/DeadMachineInstructionElim.cpp
@@ -102,7 +102,7 @@ bool DeadMachineInstructionElim::runOnMachineFunction(MachineFunction &MF) {
LivePhysRegs = ReservedRegs;
// Also add any explicit live-out physregs for this block.
- if (!MBB->empty() && MBB->back().getDesc().isReturn())
+ if (!MBB->empty() && MBB->back().isReturn())
for (MachineRegisterInfo::liveout_iterator LOI = MRI->liveout_begin(),
LOE = MRI->liveout_end(); LOI != LOE; ++LOI) {
unsigned Reg = *LOI;
diff --git a/lib/CodeGen/ExecutionDepsFix.cpp b/lib/CodeGen/ExecutionDepsFix.cpp
index 300f037..4ec75cd 100644
--- a/lib/CodeGen/ExecutionDepsFix.cpp
+++ b/lib/CodeGen/ExecutionDepsFix.cpp
@@ -454,7 +454,7 @@ void ExeDepsFix::processDefs(MachineInstr *MI, bool Kill) {
assert(!MI->isDebugValue() && "Won't process debug values");
const MCInstrDesc &MCID = MI->getDesc();
for (unsigned i = 0,
- e = MCID.isVariadic() ? MI->getNumOperands() : MCID.getNumDefs();
+ e = MI->isVariadic() ? MI->getNumOperands() : MCID.getNumDefs();
i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg())
diff --git a/lib/CodeGen/ExpandISelPseudos.cpp b/lib/CodeGen/ExpandISelPseudos.cpp
index a67140e..b5f107d 100644
--- a/lib/CodeGen/ExpandISelPseudos.cpp
+++ b/lib/CodeGen/ExpandISelPseudos.cpp
@@ -62,8 +62,7 @@ bool ExpandISelPseudos::runOnMachineFunction(MachineFunction &MF) {
MachineInstr *MI = MBBI++;
// If MI is a pseudo, expand it.
- const MCInstrDesc &MCID = MI->getDesc();
- if (MCID.usesCustomInsertionHook()) {
+ if (MI->usesCustomInsertionHook()) {
Changed = true;
MachineBasicBlock *NewMBB =
TLI->EmitInstrWithCustomInserter(MI, MBB);
diff --git a/lib/CodeGen/ExpandPostRAPseudos.cpp b/lib/CodeGen/ExpandPostRAPseudos.cpp
index e2a14a8..3d23db0 100644
--- a/lib/CodeGen/ExpandPostRAPseudos.cpp
+++ b/lib/CodeGen/ExpandPostRAPseudos.cpp
@@ -207,7 +207,7 @@ bool ExpandPostRA::runOnMachineFunction(MachineFunction &MF) {
++mi;
// Only expand pseudos.
- if (!MI->getDesc().isPseudo())
+ if (!MI->isPseudo())
continue;
// Give targets a chance to expand even standard pseudos.
diff --git a/lib/CodeGen/GCStrategy.cpp b/lib/CodeGen/GCStrategy.cpp
index 9349797..e2c7132 100644
--- a/lib/CodeGen/GCStrategy.cpp
+++ b/lib/CodeGen/GCStrategy.cpp
@@ -386,7 +386,7 @@ void MachineCodeAnalysis::FindSafePoints(MachineFunction &MF) {
BBE = MF.end(); BBI != BBE; ++BBI)
for (MachineBasicBlock::iterator MI = BBI->begin(),
ME = BBI->end(); MI != ME; ++MI)
- if (MI->getDesc().isCall())
+ if (MI->isCall())
VisitCallPoint(MI);
}
diff --git a/lib/CodeGen/IfConversion.cpp b/lib/CodeGen/IfConversion.cpp
index d888939..bd31fdf 100644
--- a/lib/CodeGen/IfConversion.cpp
+++ b/lib/CodeGen/IfConversion.cpp
@@ -573,12 +573,12 @@ bool IfConverter::ValidDiamond(BBInfo &TrueBBI, BBInfo &FalseBBI,
// blocks, move the end iterators up past any branch instructions.
while (TIE != TIB) {
--TIE;
- if (!TIE->getDesc().isBranch())
+ if (!TIE->isBranch())
break;
}
while (FIE != FIB) {
--FIE;
- if (!FIE->getDesc().isBranch())
+ if (!FIE->isBranch())
break;
}
@@ -651,12 +651,11 @@ void IfConverter::ScanInstructions(BBInfo &BBI) {
if (I->isDebugValue())
continue;
- const MCInstrDesc &MCID = I->getDesc();
- if (MCID.isNotDuplicable())
+ if (I->isNotDuplicable())
BBI.CannotBeCopied = true;
bool isPredicated = TII->isPredicated(I);
- bool isCondBr = BBI.IsBrAnalyzable && MCID.isConditionalBranch();
+ bool isCondBr = BBI.IsBrAnalyzable && I->isConditionalBranch();
if (!isCondBr) {
if (!isPredicated) {
@@ -1395,9 +1394,8 @@ void IfConverter::CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI,
for (MachineBasicBlock::iterator I = FromBBI.BB->begin(),
E = FromBBI.BB->end(); I != E; ++I) {
- const MCInstrDesc &MCID = I->getDesc();
// Do not copy the end of the block branches.
- if (IgnoreBr && MCID.isBranch())
+ if (IgnoreBr && I->isBranch())
break;
MachineInstr *MI = MF.CloneMachineInstr(I);
diff --git a/lib/CodeGen/InlineSpiller.cpp b/lib/CodeGen/InlineSpiller.cpp
index 59907d9..9bf810e 100644
--- a/lib/CodeGen/InlineSpiller.cpp
+++ b/lib/CodeGen/InlineSpiller.cpp
@@ -759,7 +759,7 @@ void InlineSpiller::eliminateRedundantSpills(LiveInterval &SLI, VNInfo *VNI) {
// Find all spills and copies of VNI.
for (MachineRegisterInfo::use_nodbg_iterator UI = MRI.use_nodbg_begin(Reg);
MachineInstr *MI = UI.skipInstruction();) {
- if (!MI->isCopy() && !MI->getDesc().mayStore())
+ if (!MI->isCopy() && !MI->mayStore())
continue;
SlotIndex Idx = LIS.getInstructionIndex(MI);
if (LI->getVNInfoAt(Idx) != VNI)
@@ -878,7 +878,7 @@ bool InlineSpiller::reMaterializeFor(LiveInterval &VirtReg,
// Before rematerializing into a register for a single instruction, try to
// fold a load into the instruction. That avoids allocating a new register.
- if (RM.OrigMI->getDesc().canFoldAsLoad() &&
+ if (RM.OrigMI->canFoldAsLoad() &&
foldMemoryOperand(MI, Ops, RM.OrigMI)) {
Edit->markRematerialized(RM.ParentVNI);
++NumFoldedLoads;
@@ -957,7 +957,7 @@ void InlineSpiller::reMaterializeAll() {
if (DeadDefs.empty())
return;
DEBUG(dbgs() << "Remat created " << DeadDefs.size() << " dead defs.\n");
- Edit->eliminateDeadDefs(DeadDefs, LIS, VRM, TII);
+ Edit->eliminateDeadDefs(DeadDefs, LIS, VRM, TII, RegsToSpill);
// Get rid of deleted and empty intervals.
for (unsigned i = RegsToSpill.size(); i != 0; --i) {
@@ -1240,7 +1240,7 @@ void InlineSpiller::spillAll() {
// Hoisted spills may cause dead code.
if (!DeadDefs.empty()) {
DEBUG(dbgs() << "Eliminating " << DeadDefs.size() << " dead defs\n");
- Edit->eliminateDeadDefs(DeadDefs, LIS, VRM, TII);
+ Edit->eliminateDeadDefs(DeadDefs, LIS, VRM, TII, RegsToSpill);
}
// Finally delete the SnippetCopies.
diff --git a/lib/CodeGen/LLVMBuild.txt b/lib/CodeGen/LLVMBuild.txt
index 2eebb08..fee0347 100644
--- a/lib/CodeGen/LLVMBuild.txt
+++ b/lib/CodeGen/LLVMBuild.txt
@@ -15,9 +15,11 @@
;
;===------------------------------------------------------------------------===;
+[common]
+subdirectories = AsmPrinter SelectionDAG
+
[component_0]
type = Library
name = CodeGen
parent = Libraries
required_libraries = Analysis Core MC Scalar Support Target TransformUtils
-
diff --git a/lib/CodeGen/LLVMTargetMachine.cpp b/lib/CodeGen/LLVMTargetMachine.cpp
index 03b5693..62227fd 100644
--- a/lib/CodeGen/LLVMTargetMachine.cpp
+++ b/lib/CodeGen/LLVMTargetMachine.cpp
@@ -41,10 +41,6 @@
#include "llvm/Support/TargetRegistry.h"
using namespace llvm;
-namespace llvm {
- bool EnableFastISel;
-}
-
static cl::opt<bool> DisablePostRA("disable-post-ra", cl::Hidden,
cl::desc("Disable Post Regalloc"));
static cl::opt<bool> DisableBranchFold("disable-branch-fold", cl::Hidden,
@@ -114,9 +110,10 @@ EnableFastISelOption("fast-isel", cl::Hidden,
LLVMTargetMachine::LLVMTargetMachine(const Target &T, StringRef Triple,
StringRef CPU, StringRef FS,
+ TargetOptions Options,
Reloc::Model RM, CodeModel::Model CM,
CodeGenOpt::Level OL)
- : TargetMachine(T, Triple, CPU, FS) {
+ : TargetMachine(T, Triple, CPU, FS, Options) {
CodeGenInfo = T.createMCCodeGenInfo(Triple, RM, CM, OL);
AsmInfo = T.createMCAsmInfo(Triple);
// TargetSelect.h moved to a different directory between LLVM 2.9 and 3.0,
@@ -275,14 +272,15 @@ bool LLVMTargetMachine::addPassesToEmitMC(PassManagerBase &PM,
return false; // success!
}
-static void printNoVerify(PassManagerBase &PM, const char *Banner) {
- if (PrintMachineCode)
+void LLVMTargetMachine::printNoVerify(PassManagerBase &PM,
+ const char *Banner) const {
+ if (Options.PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(dbgs(), Banner));
}
-static void printAndVerify(PassManagerBase &PM,
- const char *Banner) {
- if (PrintMachineCode)
+void LLVMTargetMachine::printAndVerify(PassManagerBase &PM,
+ const char *Banner) const {
+ if (Options.PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(dbgs(), Banner));
if (VerifyMachineCode)
@@ -380,7 +378,7 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM,
if (EnableFastISelOption == cl::BOU_TRUE ||
(getOptLevel() == CodeGenOpt::None &&
EnableFastISelOption != cl::BOU_FALSE))
- EnableFastISel = true;
+ Options.EnableFastISel = true;
// Ask the target for an isel.
if (addInstSelector(PM))
diff --git a/lib/CodeGen/LiveDebugVariables.cpp b/lib/CodeGen/LiveDebugVariables.cpp
index eb54baa7..c35302a 100644
--- a/lib/CodeGen/LiveDebugVariables.cpp
+++ b/lib/CodeGen/LiveDebugVariables.cpp
@@ -920,8 +920,8 @@ findInsertLocation(MachineBasicBlock *MBB, SlotIndex Idx,
}
// Don't insert anything after the first terminator, though.
- return MI->getDesc().isTerminator() ? MBB->getFirstTerminator() :
- llvm::next(MachineBasicBlock::iterator(MI));
+ return MI->isTerminator() ? MBB->getFirstTerminator() :
+ llvm::next(MachineBasicBlock::iterator(MI));
}
DebugLoc UserValue::findDebugLoc() {
diff --git a/lib/CodeGen/LiveIntervalAnalysis.cpp b/lib/CodeGen/LiveIntervalAnalysis.cpp
index edcfebe..1e58173 100644
--- a/lib/CodeGen/LiveIntervalAnalysis.cpp
+++ b/lib/CodeGen/LiveIntervalAnalysis.cpp
@@ -794,7 +794,7 @@ LiveIntervals::getLastSplitPoint(const LiveInterval &li,
MachineBasicBlock::iterator I = mbb->end(), B = mbb->begin();
while (I != B) {
--I;
- if (I->getDesc().isCall())
+ if (I->isCall())
return I;
}
// The block contains no calls that can throw, so use the first terminator.
diff --git a/lib/CodeGen/LiveRangeEdit.cpp b/lib/CodeGen/LiveRangeEdit.cpp
index 2f283b2..a470877 100644
--- a/lib/CodeGen/LiveRangeEdit.cpp
+++ b/lib/CodeGen/LiveRangeEdit.cpp
@@ -129,7 +129,7 @@ bool LiveRangeEdit::canRematerializeAt(Remat &RM,
}
// If only cheap remats were requested, bail out early.
- if (cheapAsAMove && !RM.OrigMI->getDesc().isAsCheapAsAMove())
+ if (cheapAsAMove && !RM.OrigMI->isAsCheapAsAMove())
return false;
// Verify that all used registers are available with the same values.
@@ -174,7 +174,7 @@ bool LiveRangeEdit::foldAsLoad(LiveInterval *LI,
if (MO.isDef()) {
if (DefMI && DefMI != MI)
return false;
- if (!MI->getDesc().canFoldAsLoad())
+ if (!MI->canFoldAsLoad())
return false;
DefMI = MI;
} else if (!MO.isUndef()) {
@@ -210,7 +210,8 @@ bool LiveRangeEdit::foldAsLoad(LiveInterval *LI,
void LiveRangeEdit::eliminateDeadDefs(SmallVectorImpl<MachineInstr*> &Dead,
LiveIntervals &LIS, VirtRegMap &VRM,
- const TargetInstrInfo &TII) {
+ const TargetInstrInfo &TII,
+ ArrayRef<unsigned> RegsBeingSpilled) {
SetVector<LiveInterval*,
SmallVector<LiveInterval*, 8>,
SmallPtrSet<LiveInterval*, 8> > ToShrink;
@@ -290,6 +291,21 @@ void LiveRangeEdit::eliminateDeadDefs(SmallVectorImpl<MachineInstr*> &Dead,
delegate_->LRE_WillShrinkVirtReg(LI->reg);
if (!LIS.shrinkToUses(LI, &Dead))
continue;
+
+ // Don't create new intervals for a register being spilled.
+ // The new intervals would have to be spilled anyway so its not worth it.
+ // Also they currently aren't spilled so creating them and not spilling
+ // them results in incorrect code.
+ bool BeingSpilled = false;
+ for (unsigned i = 0, e = RegsBeingSpilled.size(); i != e; ++i) {
+ if (LI->reg == RegsBeingSpilled[i]) {
+ BeingSpilled = true;
+ break;
+ }
+ }
+
+ if (BeingSpilled) continue;
+
// LI may have been separated, create new intervals.
LI->RenumberValues(LIS);
diff --git a/lib/CodeGen/LiveRangeEdit.h b/lib/CodeGen/LiveRangeEdit.h
index 9b0a671..057d9bb 100644
--- a/lib/CodeGen/LiveRangeEdit.h
+++ b/lib/CodeGen/LiveRangeEdit.h
@@ -191,9 +191,14 @@ public:
/// eliminateDeadDefs - Try to delete machine instructions that are now dead
/// (allDefsAreDead returns true). This may cause live intervals to be trimmed
/// and further dead efs to be eliminated.
+ /// RegsBeingSpilled lists registers currently being spilled by the register
+ /// allocator. These registers should not be split into new intervals
+ /// as currently those new intervals are not guaranteed to spill.
void eliminateDeadDefs(SmallVectorImpl<MachineInstr*> &Dead,
LiveIntervals&, VirtRegMap&,
- const TargetInstrInfo&);
+ const TargetInstrInfo&,
+ ArrayRef<unsigned> RegsBeingSpilled
+ = ArrayRef<unsigned>());
/// calculateRegClassAndHint - Recompute register class and hint for each new
/// register.
diff --git a/lib/CodeGen/LiveVariables.cpp b/lib/CodeGen/LiveVariables.cpp
index 2ca90f9..7477d91 100644
--- a/lib/CodeGen/LiveVariables.cpp
+++ b/lib/CodeGen/LiveVariables.cpp
@@ -590,8 +590,8 @@ bool LiveVariables::runOnMachineFunction(MachineFunction &mf) {
// them. The tail callee need not take the same registers as input
// that it produces as output, and there are dependencies for its input
// registers elsewhere.
- if (!MBB->empty() && MBB->back().getDesc().isReturn()
- && !MBB->back().getDesc().isCall()) {
+ if (!MBB->empty() && MBB->back().isReturn()
+ && !MBB->back().isCall()) {
MachineInstr *Ret = &MBB->back();
for (MachineRegisterInfo::liveout_iterator
@@ -754,7 +754,7 @@ void LiveVariables::addNewBlock(MachineBasicBlock *BB,
const unsigned NumNew = BB->getNumber();
// All registers used by PHI nodes in SuccBB must be live through BB.
- for (MachineBasicBlock::const_iterator BBI = SuccBB->begin(),
+ for (MachineBasicBlock::iterator BBI = SuccBB->begin(),
BBE = SuccBB->end(); BBI != BBE && BBI->isPHI(); ++BBI)
for (unsigned i = 1, e = BBI->getNumOperands(); i != e; i += 2)
if (BBI->getOperand(i+1).getMBB() == BB)
diff --git a/lib/CodeGen/MachineBasicBlock.cpp b/lib/CodeGen/MachineBasicBlock.cpp
index b9d1ef7..6734916 100644
--- a/lib/CodeGen/MachineBasicBlock.cpp
+++ b/lib/CodeGen/MachineBasicBlock.cpp
@@ -73,7 +73,8 @@ void ilist_traits<MachineBasicBlock>::addNodeToList(MachineBasicBlock *N) {
// Make sure the instructions have their operands in the reginfo lists.
MachineRegisterInfo &RegInfo = MF.getRegInfo();
- for (MachineBasicBlock::iterator I = N->begin(), E = N->end(); I != E; ++I)
+ for (MachineBasicBlock::instr_iterator
+ I = N->instr_begin(), E = N->instr_end(); I != E; ++I)
I->AddRegOperandsToUseLists(RegInfo);
LeakDetector::removeGarbageObject(N);
@@ -120,8 +121,8 @@ void ilist_traits<MachineInstr>::removeNodeFromList(MachineInstr *N) {
/// lists.
void ilist_traits<MachineInstr>::
transferNodesFromList(ilist_traits<MachineInstr> &fromList,
- MachineBasicBlock::iterator first,
- MachineBasicBlock::iterator last) {
+ ilist_iterator<MachineInstr> first,
+ ilist_iterator<MachineInstr> last) {
assert(Parent->getParent() == fromList.Parent->getParent() &&
"MachineInstr parent mismatch!");
@@ -140,9 +141,10 @@ void ilist_traits<MachineInstr>::deleteNode(MachineInstr* MI) {
}
MachineBasicBlock::iterator MachineBasicBlock::getFirstNonPHI() {
- iterator I = begin();
+ instr_iterator I = instr_begin();
while (I != end() && I->isPHI())
++I;
+ assert(!I->isInsideBundle() && "First non-phi MI cannot be inside a bundle!");
return I;
}
@@ -150,23 +152,63 @@ MachineBasicBlock::iterator
MachineBasicBlock::SkipPHIsAndLabels(MachineBasicBlock::iterator I) {
while (I != end() && (I->isPHI() || I->isLabel() || I->isDebugValue()))
++I;
+ // FIXME: This needs to change if we wish to bundle labels / dbg_values
+ // inside the bundle.
+ assert(!I->isInsideBundle() &&
+ "First non-phi / non-label instruction is inside a bundle!");
return I;
}
MachineBasicBlock::iterator MachineBasicBlock::getFirstTerminator() {
iterator I = end();
- while (I != begin() && ((--I)->getDesc().isTerminator() || I->isDebugValue()))
+ while (I != begin() && ((--I)->isTerminator() || I->isDebugValue()))
; /*noop */
- while (I != end() && !I->getDesc().isTerminator())
+ while (I != end() && !I->isTerminator())
+ ++I;
+ return I;
+}
+
+MachineBasicBlock::const_iterator
+MachineBasicBlock::getFirstTerminator() const {
+ const_iterator I = end();
+ while (I != begin() && ((--I)->isTerminator() || I->isDebugValue()))
+ ; /*noop */
+ while (I != end() && !I->isTerminator())
+ ++I;
+ return I;
+}
+
+MachineBasicBlock::instr_iterator MachineBasicBlock::getFirstInstrTerminator() {
+ instr_iterator I = instr_end();
+ while (I != instr_begin() && ((--I)->isTerminator() || I->isDebugValue()))
+ ; /*noop */
+ while (I != instr_end() && !I->isTerminator())
++I;
return I;
}
MachineBasicBlock::iterator MachineBasicBlock::getLastNonDebugInstr() {
- iterator B = begin(), I = end();
+ // Skip over end-of-block dbg_value instructions.
+ instr_iterator B = instr_begin(), I = instr_end();
while (I != B) {
--I;
- if (I->isDebugValue())
+ // Return instruction that starts a bundle.
+ if (I->isDebugValue() || I->isInsideBundle())
+ continue;
+ return I;
+ }
+ // The block is all debug values.
+ return end();
+}
+
+MachineBasicBlock::const_iterator
+MachineBasicBlock::getLastNonDebugInstr() const {
+ // Skip over end-of-block dbg_value instructions.
+ const_instr_iterator B = instr_begin(), I = instr_end();
+ while (I != B) {
+ --I;
+ // Return instruction that starts a bundle.
+ if (I->isDebugValue() || I->isInsideBundle())
continue;
return I;
}
@@ -203,8 +245,6 @@ void MachineBasicBlock::print(raw_ostream &OS, SlotIndexes *Indexes) const {
return;
}
- if (Alignment) { OS << "Alignment " << Alignment << "\n"; }
-
if (Indexes)
OS << Indexes->getMBBStartIdx(this) << '\t';
@@ -218,6 +258,12 @@ void MachineBasicBlock::print(raw_ostream &OS, SlotIndexes *Indexes) const {
}
if (isLandingPad()) { OS << Comma << "EH LANDING PAD"; Comma = ", "; }
if (hasAddressTaken()) { OS << Comma << "ADDRESS TAKEN"; Comma = ", "; }
+ if (Alignment) {
+ OS << Comma << "Align " << Alignment << " (" << (1u << Alignment)
+ << " bytes)";
+ Comma = ", ";
+ }
+
OS << '\n';
const TargetRegisterInfo *TRI = MF->getTarget().getRegisterInfo();
@@ -237,13 +283,15 @@ void MachineBasicBlock::print(raw_ostream &OS, SlotIndexes *Indexes) const {
OS << '\n';
}
- for (const_iterator I = begin(); I != end(); ++I) {
+ for (const_instr_iterator I = instr_begin(); I != instr_end(); ++I) {
if (Indexes) {
if (Indexes->hasIndex(I))
OS << Indexes->getInstructionIndex(I);
OS << '\t';
}
OS << '\t';
+ if (I->isInsideBundle())
+ OS << " * ";
I->print(OS, &getParent()->getTarget());
}
@@ -449,8 +497,8 @@ MachineBasicBlock::transferSuccessorsAndUpdatePHIs(MachineBasicBlock *fromMBB) {
fromMBB->removeSuccessor(Succ);
// Fix up any PHI nodes in the successor.
- for (MachineBasicBlock::iterator MI = Succ->begin(), ME = Succ->end();
- MI != ME && MI->isPHI(); ++MI)
+ for (MachineBasicBlock::instr_iterator MI = Succ->instr_begin(),
+ ME = Succ->instr_end(); MI != ME && MI->isPHI(); ++MI)
for (unsigned i = 2, e = MI->getNumOperands()+1; i != e; i += 2) {
MachineOperand &MO = MI->getOperand(i);
if (MO.getMBB() == fromMBB)
@@ -492,8 +540,8 @@ bool MachineBasicBlock::canFallThrough() {
// Barrier is predicated and thus no longer an actual control barrier. This
// is over-conservative though, because if an instruction isn't actually
// predicated we could still treat it like a barrier.
- return empty() || !back().getDesc().isBarrier() ||
- back().getDesc().isPredicable();
+ return empty() || !back().isBarrier() ||
+ back().isPredicable();
}
// If there is no branch, control always falls through.
@@ -552,7 +600,8 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
// Collect a list of virtual registers killed by the terminators.
SmallVector<unsigned, 4> KilledRegs;
if (LV)
- for (iterator I = getFirstTerminator(), E = end(); I != E; ++I) {
+ for (instr_iterator I = getFirstInstrTerminator(), E = instr_end();
+ I != E; ++I) {
MachineInstr *MI = I;
for (MachineInstr::mop_iterator OI = MI->operands_begin(),
OE = MI->operands_end(); OI != OE; ++OI) {
@@ -579,7 +628,8 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
}
// Fix PHI nodes in Succ so they refer to NMBB instead of this
- for (MachineBasicBlock::iterator i = Succ->begin(), e = Succ->end();
+ for (MachineBasicBlock::instr_iterator
+ i = Succ->instr_begin(),e = Succ->instr_end();
i != e && i->isPHI(); ++i)
for (unsigned ni = 1, ne = i->getNumOperands(); ni != ne; ni += 2)
if (i->getOperand(ni+1).getMBB() == this)
@@ -595,7 +645,7 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
// Restore kills of virtual registers that were killed by the terminators.
while (!KilledRegs.empty()) {
unsigned Reg = KilledRegs.pop_back_val();
- for (iterator I = end(), E = begin(); I != E;) {
+ for (instr_iterator I = instr_end(), E = instr_begin(); I != E;) {
if (!(--I)->addRegisterKilled(Reg, NULL, /* addIfNotFound= */ false))
continue;
LV->getVarInfo(Reg).Kills.push_back(I);
@@ -664,6 +714,41 @@ MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
return NMBB;
}
+MachineBasicBlock::iterator
+MachineBasicBlock::erase(MachineBasicBlock::iterator I) {
+ if (I->isBundle()) {
+ MachineBasicBlock::iterator E = llvm::next(I);
+ return Insts.erase(I.getInstrIterator(), E.getInstrIterator());
+ }
+
+ return Insts.erase(I.getInstrIterator());
+}
+
+MachineInstr *MachineBasicBlock::remove(MachineInstr *I) {
+ if (I->isBundle()) {
+ MachineBasicBlock::instr_iterator MII = I; ++MII;
+ while (MII != end() && MII->isInsideBundle()) {
+ MachineInstr *MI = &*MII++;
+ Insts.remove(MI);
+ }
+ }
+
+ return Insts.remove(I);
+}
+
+void MachineBasicBlock::splice(MachineBasicBlock::iterator where,
+ MachineBasicBlock *Other,
+ MachineBasicBlock::iterator From) {
+ if (From->isBundle()) {
+ MachineBasicBlock::iterator To = llvm::next(From);
+ Insts.splice(where.getInstrIterator(), Other->Insts,
+ From.getInstrIterator(), To.getInstrIterator());
+ return;
+ }
+
+ Insts.splice(where.getInstrIterator(), Other->Insts, From.getInstrIterator());
+}
+
/// removeFromParent - This method unlinks 'this' from the containing function,
/// and returns it, but does not delete it.
MachineBasicBlock *MachineBasicBlock::removeFromParent() {
@@ -687,10 +772,10 @@ void MachineBasicBlock::ReplaceUsesOfBlockWith(MachineBasicBlock *Old,
MachineBasicBlock *New) {
assert(Old != New && "Cannot replace self with self!");
- MachineBasicBlock::iterator I = end();
- while (I != begin()) {
+ MachineBasicBlock::instr_iterator I = instr_end();
+ while (I != instr_begin()) {
--I;
- if (!I->getDesc().isTerminator()) break;
+ if (!I->isTerminator()) break;
// Scan the operands of this machine instruction, replacing any uses of Old
// with New.
@@ -769,17 +854,17 @@ bool MachineBasicBlock::CorrectExtraCFGEdges(MachineBasicBlock *DestA,
/// findDebugLoc - find the next valid DebugLoc starting at MBBI, skipping
/// any DBG_VALUE instructions. Return UnknownLoc if there is none.
DebugLoc
-MachineBasicBlock::findDebugLoc(MachineBasicBlock::iterator &MBBI) {
+MachineBasicBlock::findDebugLoc(instr_iterator MBBI) {
DebugLoc DL;
- MachineBasicBlock::iterator E = end();
- if (MBBI != E) {
- // Skip debug declarations, we don't want a DebugLoc from them.
- MachineBasicBlock::iterator MBBI2 = MBBI;
- while (MBBI2 != E && MBBI2->isDebugValue())
- MBBI2++;
- if (MBBI2 != E)
- DL = MBBI2->getDebugLoc();
- }
+ instr_iterator E = instr_end();
+ if (MBBI == E)
+ return DL;
+
+ // Skip debug declarations, we don't want a DebugLoc from them.
+ while (MBBI != E && MBBI->isDebugValue())
+ MBBI++;
+ if (MBBI != E)
+ DL = MBBI->getDebugLoc();
return DL;
}
diff --git a/lib/CodeGen/MachineBlockPlacement.cpp b/lib/CodeGen/MachineBlockPlacement.cpp
index 55d804b..638d895 100644
--- a/lib/CodeGen/MachineBlockPlacement.cpp
+++ b/lib/CodeGen/MachineBlockPlacement.cpp
@@ -36,10 +36,7 @@
#include "llvm/CodeGen/Passes.h"
#include "llvm/Support/Allocator.h"
#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/PostOrderIterator.h"
-#include "llvm/ADT/SCCIterator.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
@@ -56,22 +53,6 @@ STATISTIC(UncondBranchTakenFreq,
"Potential frequency of taking unconditional branches");
namespace {
-/// \brief A structure for storing a weighted edge.
-///
-/// This stores an edge and its weight, computed as the product of the
-/// frequency that the starting block is entered with the probability of
-/// a particular exit block.
-struct WeightedEdge {
- BlockFrequency EdgeFrequency;
- MachineBasicBlock *From, *To;
-
- bool operator<(const WeightedEdge &RHS) const {
- return EdgeFrequency < RHS.EdgeFrequency;
- }
-};
-}
-
-namespace {
class BlockChain;
/// \brief Type for our function-wide basic block -> block chain mapping.
typedef DenseMap<MachineBasicBlock *, BlockChain *> BlockToChainMapType;
@@ -222,6 +203,9 @@ class MachineBlockPlacement : public MachineFunctionPass {
void buildChain(MachineBasicBlock *BB, BlockChain &Chain,
SmallVectorImpl<MachineBasicBlock *> &BlockWorkList,
const BlockFilterSet *BlockFilter = 0);
+ MachineBasicBlock *findBestLoopTop(MachineFunction &F,
+ MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet);
void buildLoopChains(MachineFunction &F, MachineLoop &L);
void buildCFGChains(MachineFunction &F);
void AlignLoops(MachineFunction &F);
@@ -546,12 +530,134 @@ void MachineBlockPlacement::buildChain(
markChainSuccessors(SuccChain, LoopHeaderBB, BlockWorkList, BlockFilter);
Chain.merge(BestSucc, &SuccChain);
BB = *llvm::prior(Chain.end());
- };
+ }
DEBUG(dbgs() << "Finished forming chain for header block "
<< getBlockNum(*Chain.begin()) << "\n");
}
+/// \brief Find the best loop top block for layout.
+///
+/// This routine implements the logic to analyze the loop looking for the best
+/// block to layout at the top of the loop. Typically this is done to maximize
+/// fallthrough opportunities.
+MachineBasicBlock *
+MachineBlockPlacement::findBestLoopTop(MachineFunction &F,
+ MachineLoop &L,
+ const BlockFilterSet &LoopBlockSet) {
+ BlockFrequency BestExitEdgeFreq;
+ MachineBasicBlock *ExitingBB = 0;
+ MachineBasicBlock *LoopingBB = 0;
+ // If there are exits to outer loops, loop rotation can severely limit
+ // fallthrough opportunites unless it selects such an exit. Keep a set of
+ // blocks where rotating to exit with that block will reach an outer loop.
+ SmallPtrSet<MachineBasicBlock *, 4> BlocksExitingToOuterLoop;
+
+ DEBUG(dbgs() << "Finding best loop exit for: "
+ << getBlockName(L.getHeader()) << "\n");
+ for (MachineLoop::block_iterator I = L.block_begin(),
+ E = L.block_end();
+ I != E; ++I) {
+ BlockChain &Chain = *BlockToChain[*I];
+ // Ensure that this block is at the end of a chain; otherwise it could be
+ // mid-way through an inner loop or a successor of an analyzable branch.
+ if (*I != *llvm::prior(Chain.end()))
+ continue;
+
+ // Now walk the successors. We need to establish whether this has a viable
+ // exiting successor and whether it has a viable non-exiting successor.
+ // We store the old exiting state and restore it if a viable looping
+ // successor isn't found.
+ MachineBasicBlock *OldExitingBB = ExitingBB;
+ BlockFrequency OldBestExitEdgeFreq = BestExitEdgeFreq;
+ // We also compute and store the best looping successor for use in layout.
+ MachineBasicBlock *BestLoopSucc = 0;
+ // FIXME: Due to the performance of the probability and weight routines in
+ // the MBPI analysis, we use the internal weights. This is only valid
+ // because it is purely a ranking function, we don't care about anything
+ // but the relative values.
+ uint32_t BestLoopSuccWeight = 0;
+ // FIXME: We also manually compute the probabilities to avoid quadratic
+ // behavior.
+ uint32_t WeightScale = 0;
+ uint32_t SumWeight = MBPI->getSumForBlock(*I, WeightScale);
+ for (MachineBasicBlock::succ_iterator SI = (*I)->succ_begin(),
+ SE = (*I)->succ_end();
+ SI != SE; ++SI) {
+ if ((*SI)->isLandingPad())
+ continue;
+ if (*SI == *I)
+ continue;
+ BlockChain &SuccChain = *BlockToChain[*SI];
+ // Don't split chains, either this chain or the successor's chain.
+ if (&Chain == &SuccChain || *SI != *SuccChain.begin()) {
+ DEBUG(dbgs() << " " << (LoopBlockSet.count(*SI) ? "looping: "
+ : "exiting: ")
+ << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " (chain conflict)\n");
+ continue;
+ }
+
+ uint32_t SuccWeight = MBPI->getEdgeWeight(*I, *SI);
+ if (LoopBlockSet.count(*SI)) {
+ DEBUG(dbgs() << " looping: " << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " (" << SuccWeight << ")\n");
+ if (BestLoopSucc && BestLoopSuccWeight >= SuccWeight)
+ continue;
+
+ BestLoopSucc = *SI;
+ BestLoopSuccWeight = SuccWeight;
+ continue;
+ }
+
+ BranchProbability SuccProb(SuccWeight / WeightScale, SumWeight);
+ BlockFrequency ExitEdgeFreq = MBFI->getBlockFreq(*I) * SuccProb;
+ DEBUG(dbgs() << " exiting: " << getBlockName(*I) << " -> "
+ << getBlockName(*SI) << " (" << ExitEdgeFreq << ")\n");
+ // Note that we slightly bias this toward an existing layout successor to
+ // retain incoming order in the absence of better information.
+ // FIXME: Should we bias this more strongly? It's pretty weak.
+ if (!ExitingBB || ExitEdgeFreq > BestExitEdgeFreq ||
+ ((*I)->isLayoutSuccessor(*SI) &&
+ !(ExitEdgeFreq < BestExitEdgeFreq))) {
+ BestExitEdgeFreq = ExitEdgeFreq;
+ ExitingBB = *I;
+ }
+
+ if (MachineLoop *ExitLoop = MLI->getLoopFor(*SI))
+ if (ExitLoop->contains(&L))
+ BlocksExitingToOuterLoop.insert(*I);
+ }
+
+ // Restore the old exiting state, no viable looping successor was found.
+ if (!BestLoopSucc) {
+ ExitingBB = OldExitingBB;
+ BestExitEdgeFreq = OldBestExitEdgeFreq;
+ continue;
+ }
+
+ // If this was best exiting block thus far, also record the looping block.
+ if (ExitingBB == *I)
+ LoopingBB = BestLoopSucc;
+ }
+ // Without a candidate exitting block or with only a single block in the
+ // loop, just use the loop header to layout the loop.
+ if (!ExitingBB || L.getNumBlocks() == 1)
+ return L.getHeader();
+
+ // Also, if we have exit blocks which lead to outer loops but didn't select
+ // one of them as the exiting block we are rotating toward, disable loop
+ // rotation altogether.
+ if (!BlocksExitingToOuterLoop.empty() &&
+ !BlocksExitingToOuterLoop.count(ExitingBB))
+ return L.getHeader();
+
+ assert(LoopingBB && "All successors of a loop block are exit blocks!");
+ DEBUG(dbgs() << " Best exiting block: " << getBlockName(ExitingBB) << "\n");
+ DEBUG(dbgs() << " Best top block: " << getBlockName(LoopingBB) << "\n");
+ return LoopingBB;
+}
+
/// \brief Forms basic block chains from the natural loop structures.
///
/// These chains are designed to preserve the existing *structure* of the code
@@ -567,17 +673,21 @@ void MachineBlockPlacement::buildLoopChains(MachineFunction &F,
SmallVector<MachineBasicBlock *, 16> BlockWorkList;
BlockFilterSet LoopBlockSet(L.block_begin(), L.block_end());
- BlockChain &LoopChain = *BlockToChain[L.getHeader()];
+
+ MachineBasicBlock *LayoutTop = findBestLoopTop(F, L, LoopBlockSet);
+ BlockChain &LoopChain = *BlockToChain[LayoutTop];
// FIXME: This is a really lame way of walking the chains in the loop: we
// walk the blocks, and use a set to prevent visiting a particular chain
// twice.
SmallPtrSet<BlockChain *, 4> UpdatedPreds;
+ assert(LoopChain.LoopPredecessors == 0);
+ UpdatedPreds.insert(&LoopChain);
for (MachineLoop::block_iterator BI = L.block_begin(),
BE = L.block_end();
BI != BE; ++BI) {
BlockChain &Chain = *BlockToChain[*BI];
- if (!UpdatedPreds.insert(&Chain) || BI == L.block_begin())
+ if (!UpdatedPreds.insert(&Chain))
continue;
assert(Chain.LoopPredecessors == 0);
@@ -597,7 +707,7 @@ void MachineBlockPlacement::buildLoopChains(MachineFunction &F,
BlockWorkList.push_back(*Chain.begin());
}
- buildChain(*L.block_begin(), LoopChain, BlockWorkList, &LoopBlockSet);
+ buildChain(LayoutTop, LoopChain, BlockWorkList, &LoopBlockSet);
DEBUG({
// Crash at the end so we get all of the debugging output first.
diff --git a/lib/CodeGen/MachineCSE.cpp b/lib/CodeGen/MachineCSE.cpp
index 7eda8c1..8c02cd7 100644
--- a/lib/CodeGen/MachineCSE.cpp
+++ b/lib/CodeGen/MachineCSE.cpp
@@ -260,12 +260,11 @@ bool MachineCSE::isCSECandidate(MachineInstr *MI) {
return false;
// Ignore stuff that we obviously can't move.
- const MCInstrDesc &MCID = MI->getDesc();
- if (MCID.mayStore() || MCID.isCall() || MCID.isTerminator() ||
+ if (MI->mayStore() || MI->isCall() || MI->isTerminator() ||
MI->hasUnmodeledSideEffects())
return false;
- if (MCID.mayLoad()) {
+ if (MI->mayLoad()) {
// Okay, this instruction does a load. As a refinement, we allow the target
// to decide whether the loaded value is actually a constant. If so, we can
// actually use it as a load.
@@ -287,7 +286,7 @@ bool MachineCSE::isProfitableToCSE(unsigned CSReg, unsigned Reg,
// Heuristics #1: Don't CSE "cheap" computation if the def is not local or in
// an immediate predecessor. We don't want to increase register pressure and
// end up causing other computation to be spilled.
- if (MI->getDesc().isAsCheapAsAMove()) {
+ if (MI->isAsCheapAsAMove()) {
MachineBasicBlock *CSBB = CSMI->getParent();
MachineBasicBlock *BB = MI->getParent();
if (CSBB != BB && !CSBB->isSuccessor(BB))
@@ -376,7 +375,7 @@ bool MachineCSE::ProcessBlock(MachineBasicBlock *MBB) {
// Commute commutable instructions.
bool Commuted = false;
- if (!FoundCSE && MI->getDesc().isCommutable()) {
+ if (!FoundCSE && MI->isCommutable()) {
MachineInstr *NewMI = TII->commuteInstruction(MI);
if (NewMI) {
Commuted = true;
diff --git a/lib/CodeGen/MachineInstr.cpp b/lib/CodeGen/MachineInstr.cpp
index b0ef9d4..ec5a1cd 100644
--- a/lib/CodeGen/MachineInstr.cpp
+++ b/lib/CodeGen/MachineInstr.cpp
@@ -178,6 +178,7 @@ void MachineOperand::ChangeToRegister(unsigned Reg, bool isDef, bool isImp,
IsKill = isKill;
IsDead = isDead;
IsUndef = isUndef;
+ IsInternalRead = false;
IsEarlyClobber = false;
IsDebug = isDebug;
SubReg = 0;
@@ -240,7 +241,7 @@ void MachineOperand::print(raw_ostream &OS, const TargetMachine *TM) const {
OS << PrintReg(getReg(), TRI, getSubReg());
if (isDef() || isKill() || isDead() || isImplicit() || isUndef() ||
- isEarlyClobber()) {
+ isInternalRead() || isEarlyClobber()) {
OS << '<';
bool NeedComma = false;
if (isDef()) {
@@ -256,14 +257,26 @@ void MachineOperand::print(raw_ostream &OS, const TargetMachine *TM) const {
NeedComma = true;
}
- if (isKill() || isDead() || isUndef()) {
+ if (isKill() || isDead() || isUndef() || isInternalRead()) {
if (NeedComma) OS << ',';
- if (isKill()) OS << "kill";
- if (isDead()) OS << "dead";
+ NeedComma = false;
+ if (isKill()) {
+ OS << "kill";
+ NeedComma = true;
+ }
+ if (isDead()) {
+ OS << "dead";
+ NeedComma = true;
+ }
if (isUndef()) {
- if (isKill() || isDead())
- OS << ',';
+ if (NeedComma) OS << ',';
OS << "undef";
+ NeedComma = true;
+ }
+ if (isInternalRead()) {
+ if (NeedComma) OS << ',';
+ OS << "internal";
+ NeedComma = true;
}
}
OS << '>';
@@ -735,6 +748,27 @@ void MachineInstr::addMemOperand(MachineFunction &MF,
MemRefsEnd = NewMemRefsEnd;
}
+bool
+MachineInstr::hasProperty(unsigned MCFlag, QueryType Type) const {
+ if (Type == IgnoreBundle || !isBundle())
+ return getDesc().getFlags() & (1 << MCFlag);
+
+ const MachineBasicBlock *MBB = getParent();
+ MachineBasicBlock::const_instr_iterator MII = *this; ++MII;
+ while (MII != MBB->end() && MII->isInsideBundle()) {
+ if (MII->getDesc().getFlags() & (1 << MCFlag)) {
+ if (Type == AnyInBundle)
+ return true;
+ } else {
+ if (Type == AllInBundle)
+ return false;
+ }
+ ++MII;
+ }
+
+ return Type == AllInBundle;
+}
+
bool MachineInstr::isIdenticalTo(const MachineInstr *Other,
MICheckType Check) const {
// If opcodes or number of operands are not the same then the two
@@ -743,6 +777,19 @@ bool MachineInstr::isIdenticalTo(const MachineInstr *Other,
Other->getNumOperands() != getNumOperands())
return false;
+ if (isBundle()) {
+ // Both instructions are bundles, compare MIs inside the bundle.
+ MachineBasicBlock::const_instr_iterator I1 = *this;
+ MachineBasicBlock::const_instr_iterator E1 = getParent()->instr_end();
+ MachineBasicBlock::const_instr_iterator I2 = *Other;
+ MachineBasicBlock::const_instr_iterator E2= Other->getParent()->instr_end();
+ while (++I1 != E1 && I1->isInsideBundle()) {
+ ++I2;
+ if (I2 == E2 || !I2->isInsideBundle() || !I1->isIdenticalTo(I2, Check))
+ return false;
+ }
+ }
+
// Check operands to make sure they match.
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
const MachineOperand &MO = getOperand(i);
@@ -789,6 +836,18 @@ bool MachineInstr::isIdenticalTo(const MachineInstr *Other,
/// block, and returns it, but does not delete it.
MachineInstr *MachineInstr::removeFromParent() {
assert(getParent() && "Not embedded in a basic block!");
+
+ // If it's a bundle then remove the MIs inside the bundle as well.
+ if (isBundle()) {
+ MachineBasicBlock *MBB = getParent();
+ MachineBasicBlock::instr_iterator MII = *this; ++MII;
+ MachineBasicBlock::instr_iterator E = MBB->instr_end();
+ while (MII != E && MII->isInsideBundle()) {
+ MachineInstr *MI = &*MII;
+ ++MII;
+ MBB->remove(MI);
+ }
+ }
getParent()->remove(this);
return this;
}
@@ -798,6 +857,17 @@ MachineInstr *MachineInstr::removeFromParent() {
/// block, and deletes it.
void MachineInstr::eraseFromParent() {
assert(getParent() && "Not embedded in a basic block!");
+ // If it's a bundle then remove the MIs inside the bundle as well.
+ if (isBundle()) {
+ MachineBasicBlock *MBB = getParent();
+ MachineBasicBlock::instr_iterator MII = *this; ++MII;
+ MachineBasicBlock::instr_iterator E = MBB->instr_end();
+ while (MII != E && MII->isInsideBundle()) {
+ MachineInstr *MI = &*MII;
+ ++MII;
+ MBB->erase(MI);
+ }
+ }
getParent()->erase(this);
}
@@ -887,6 +957,20 @@ MachineInstr::getRegClassConstraint(unsigned OpIdx,
return NULL;
}
+/// getBundleSize - Return the number of instructions inside the MI bundle.
+unsigned MachineInstr::getBundleSize() const {
+ assert(isBundle() && "Expecting a bundle");
+
+ MachineBasicBlock::const_instr_iterator I = *this;
+ unsigned Size = 0;
+ while ((++I)->isInsideBundle()) {
+ ++Size;
+ }
+ assert(Size > 1 && "Malformed bundle");
+
+ return Size;
+}
+
/// findRegisterUseOperandIdx() - Returns the MachineOperand that is a use of
/// the specific register or -1 if it is not found. It further tightens
/// the search criteria to a use that kills the register if isKill is true.
@@ -1118,6 +1202,8 @@ void MachineInstr::copyKillDeadInfo(const MachineInstr *MI) {
/// copyPredicates - Copies predicate operand(s) from MI.
void MachineInstr::copyPredicates(const MachineInstr *MI) {
+ assert(!isBundle() && "MachineInstr::copyPredicates() can't handle bundles");
+
const MCInstrDesc &MCID = MI->getDesc();
if (!MCID.isPredicable())
return;
@@ -1159,13 +1245,13 @@ bool MachineInstr::isSafeToMove(const TargetInstrInfo *TII,
AliasAnalysis *AA,
bool &SawStore) const {
// Ignore stuff that we obviously can't move.
- if (MCID->mayStore() || MCID->isCall()) {
+ if (mayStore() || isCall()) {
SawStore = true;
return false;
}
if (isLabel() || isDebugValue() ||
- MCID->isTerminator() || hasUnmodeledSideEffects())
+ isTerminator() || hasUnmodeledSideEffects())
return false;
// See if this instruction does a load. If so, we have to guarantee that the
@@ -1173,7 +1259,7 @@ bool MachineInstr::isSafeToMove(const TargetInstrInfo *TII,
// destination. The check for isInvariantLoad gives the targe the chance to
// classify the load as always returning a constant, e.g. a constant pool
// load.
- if (MCID->mayLoad() && !isInvariantLoad(AA))
+ if (mayLoad() && !isInvariantLoad(AA))
// Otherwise, this is a real load. If there is a store between the load and
// end of block, or if the load is volatile, we can't move it.
return !SawStore && !hasVolatileMemoryRef();
@@ -1213,9 +1299,9 @@ bool MachineInstr::isSafeToReMat(const TargetInstrInfo *TII,
/// have no volatile memory references.
bool MachineInstr::hasVolatileMemoryRef() const {
// An instruction known never to access memory won't have a volatile access.
- if (!MCID->mayStore() &&
- !MCID->mayLoad() &&
- !MCID->isCall() &&
+ if (!mayStore() &&
+ !mayLoad() &&
+ !isCall() &&
!hasUnmodeledSideEffects())
return false;
@@ -1239,7 +1325,7 @@ bool MachineInstr::hasVolatileMemoryRef() const {
/// *all* loads the instruction does are invariant (if it does multiple loads).
bool MachineInstr::isInvariantLoad(AliasAnalysis *AA) const {
// If the instruction doesn't load at all, it isn't an invariant load.
- if (!MCID->mayLoad())
+ if (!mayLoad())
return false;
// If the instruction has lost its memoperands, conservatively assume that
@@ -1292,7 +1378,7 @@ unsigned MachineInstr::isConstantValuePHI() const {
}
bool MachineInstr::hasUnmodeledSideEffects() const {
- if (getDesc().hasUnmodeledSideEffects())
+ if (hasProperty(MCID::UnmodeledSideEffects))
return true;
if (isInlineAsm()) {
unsigned ExtraInfo = getOperand(InlineAsm::MIOp_ExtraInfo).getImm();
@@ -1420,7 +1506,7 @@ void MachineInstr::print(raw_ostream &OS, const TargetMachine *TM) const {
// call instructions much less noisy on targets where calls clobber lots
// of registers. Don't rely on MO.isDead() because we may be called before
// LiveVariables is run, or we may be looking at a non-allocatable reg.
- if (MF && getDesc().isCall() &&
+ if (MF && isCall() &&
MO.isReg() && MO.isImplicit() && MO.isDef()) {
unsigned Reg = MO.getReg();
if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
diff --git a/lib/CodeGen/MachineInstrBundle.cpp b/lib/CodeGen/MachineInstrBundle.cpp
new file mode 100644
index 0000000..b766d08
--- /dev/null
+++ b/lib/CodeGen/MachineInstrBundle.cpp
@@ -0,0 +1,180 @@
+//===-- lib/CodeGen/MachineInstrBundle.cpp --------------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineInstrBundle.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
+using namespace llvm;
+
+namespace {
+ class UnpackMachineBundles : public MachineFunctionPass {
+ public:
+ static char ID; // Pass identification
+ UnpackMachineBundles() : MachineFunctionPass(ID) {
+ initializeUnpackMachineBundlesPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnMachineFunction(MachineFunction &MF);
+ };
+} // end anonymous namespace
+
+char UnpackMachineBundles::ID = 0;
+INITIALIZE_PASS(UnpackMachineBundles, "unpack-mi-bundle",
+ "Unpack machine instruction bundles", false, false)
+
+FunctionPass *llvm::createUnpackMachineBundlesPass() {
+ return new UnpackMachineBundles();
+}
+
+bool UnpackMachineBundles::runOnMachineFunction(MachineFunction &MF) {
+ bool Changed = false;
+ for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
+ MachineBasicBlock *MBB = &*I;
+
+ for (MachineBasicBlock::instr_iterator MII = MBB->instr_begin(),
+ MIE = MBB->instr_end(); MII != MIE; ) {
+ MachineInstr *MI = &*MII;
+
+ // Remove BUNDLE instruction and the InsideBundle flags from bundled
+ // instructions.
+ if (MI->isBundle()) {
+ while (++MII != MIE && MII->isInsideBundle()) {
+ MII->setIsInsideBundle(false);
+ for (unsigned i = 0, e = MII->getNumOperands(); i != e; ++i) {
+ MachineOperand &MO = MII->getOperand(i);
+ if (MO.isReg() && MO.isInternalRead())
+ MO.setIsInternalRead(false);
+ }
+ }
+ MI->eraseFromParent();
+
+ Changed = true;
+ continue;
+ }
+
+ ++MII;
+ }
+ }
+
+ return Changed;
+}
+
+/// FinalizeBundle - Finalize a machine instruction bundle which includes
+/// a sequence of instructions starting from FirstMI to LastMI (inclusive).
+/// This routine adds a BUNDLE instruction to represent the bundle, it adds
+/// IsInternalRead markers to MachineOperands which are defined inside the
+/// bundle, and it copies externally visible defs and uses to the BUNDLE
+/// instruction.
+void llvm::FinalizeBundle(MachineBasicBlock &MBB,
+ MachineBasicBlock::instr_iterator FirstMI,
+ MachineBasicBlock::instr_iterator LastMI) {
+ const TargetMachine &TM = MBB.getParent()->getTarget();
+ const TargetInstrInfo *TII = TM.getInstrInfo();
+ const TargetRegisterInfo *TRI = TM.getRegisterInfo();
+
+ MachineInstrBuilder MIB = BuildMI(MBB, FirstMI, FirstMI->getDebugLoc(),
+ TII->get(TargetOpcode::BUNDLE));
+
+ SmallVector<unsigned, 8> LocalDefs;
+ SmallSet<unsigned, 8> LocalDefSet;
+ SmallSet<unsigned, 8> DeadDefSet;
+ SmallSet<unsigned, 8> KilledDefSet;
+ SmallVector<unsigned, 8> ExternUses;
+ SmallSet<unsigned, 8> ExternUseSet;
+ SmallSet<unsigned, 8> KilledUseSet;
+ SmallSet<unsigned, 8> UndefUseSet;
+ SmallVector<MachineOperand*, 4> Defs;
+ do {
+ for (unsigned i = 0, e = FirstMI->getNumOperands(); i != e; ++i) {
+ MachineOperand &MO = FirstMI->getOperand(i);
+ if (!MO.isReg())
+ continue;
+ if (MO.isDef()) {
+ Defs.push_back(&MO);
+ continue;
+ }
+
+ unsigned Reg = MO.getReg();
+ if (!Reg)
+ continue;
+ assert(TargetRegisterInfo::isPhysicalRegister(Reg));
+ if (LocalDefSet.count(Reg)) {
+ MO.setIsInternalRead();
+ if (MO.isKill())
+ // Internal def is now killed.
+ KilledDefSet.insert(Reg);
+ } else {
+ if (ExternUseSet.insert(Reg)) {
+ ExternUses.push_back(Reg);
+ if (MO.isUndef())
+ UndefUseSet.insert(Reg);
+ }
+ if (MO.isKill())
+ // External def is now killed.
+ KilledUseSet.insert(Reg);
+ }
+ }
+
+ for (unsigned i = 0, e = Defs.size(); i != e; ++i) {
+ MachineOperand &MO = *Defs[i];
+ unsigned Reg = MO.getReg();
+ if (!Reg)
+ continue;
+
+ if (LocalDefSet.insert(Reg)) {
+ LocalDefs.push_back(Reg);
+ if (MO.isDead()) {
+ DeadDefSet.insert(Reg);
+ }
+ } else {
+ // Re-defined inside the bundle, it's no longer killed.
+ KilledDefSet.erase(Reg);
+ if (!MO.isDead())
+ // Previously defined but dead.
+ DeadDefSet.erase(Reg);
+ }
+
+ if (!MO.isDead()) {
+ for (const unsigned *SubRegs = TRI->getSubRegisters(Reg);
+ unsigned SubReg = *SubRegs; ++SubRegs) {
+ if (LocalDefSet.insert(SubReg))
+ LocalDefs.push_back(SubReg);
+ }
+ }
+ }
+
+ FirstMI->setIsInsideBundle();
+ Defs.clear();
+ } while (FirstMI++ != LastMI);
+
+ SmallSet<unsigned, 8> Added;
+ for (unsigned i = 0, e = LocalDefs.size(); i != e; ++i) {
+ unsigned Reg = LocalDefs[i];
+ if (Added.insert(Reg)) {
+ // If it's not live beyond end of the bundle, mark it dead.
+ bool isDead = DeadDefSet.count(Reg) || KilledDefSet.count(Reg);
+ MIB.addReg(Reg, getDefRegState(true) | getDeadRegState(isDead) |
+ getImplRegState(true));
+ }
+ }
+
+ for (unsigned i = 0, e = ExternUses.size(); i != e; ++i) {
+ unsigned Reg = ExternUses[i];
+ bool isKill = KilledUseSet.count(Reg);
+ bool isUndef = UndefUseSet.count(Reg);
+ MIB.addReg(Reg, getKillRegState(isKill) | getUndefRegState(isUndef) |
+ getImplRegState(true));
+ }
+}
diff --git a/lib/CodeGen/MachineLICM.cpp b/lib/CodeGen/MachineLICM.cpp
index e5e8c51..764429d 100644
--- a/lib/CodeGen/MachineLICM.cpp
+++ b/lib/CodeGen/MachineLICM.cpp
@@ -765,7 +765,7 @@ void MachineLICM::UpdateRegPressure(const MachineInstr *MI) {
/// isLoadFromGOTOrConstantPool - Return true if this machine instruction
/// loads from global offset table or constant pool.
static bool isLoadFromGOTOrConstantPool(MachineInstr &MI) {
- assert (MI.getDesc().mayLoad() && "Expected MI that loads!");
+ assert (MI.mayLoad() && "Expected MI that loads!");
for (MachineInstr::mmo_iterator I = MI.memoperands_begin(),
E = MI.memoperands_end(); I != E; ++I) {
if (const Value *V = (*I)->getValue()) {
@@ -792,7 +792,7 @@ bool MachineLICM::IsLICMCandidate(MachineInstr &I) {
// from constant memory are not safe to speculate all the time, for example
// indexed load from a jump table.
// Stores and side effects are already checked by isSafeToMove.
- if (I.getDesc().mayLoad() && !isLoadFromGOTOrConstantPool(I) &&
+ if (I.mayLoad() && !isLoadFromGOTOrConstantPool(I) &&
!IsGuaranteedToExecute(I.getParent()))
return false;
@@ -921,7 +921,7 @@ bool MachineLICM::HasHighOperandLatency(MachineInstr &MI,
/// IsCheapInstruction - Return true if the instruction is marked "cheap" or
/// the operand latency between its def and a use is one or less.
bool MachineLICM::IsCheapInstruction(MachineInstr &MI) const {
- if (MI.getDesc().isAsCheapAsAMove() || MI.isCopyLike())
+ if (MI.isAsCheapAsAMove() || MI.isCopyLike())
return true;
if (!InstrItins || InstrItins->isEmpty())
return false;
@@ -1105,7 +1105,7 @@ bool MachineLICM::IsProfitableToHoist(MachineInstr &MI) {
MachineInstr *MachineLICM::ExtractHoistableLoad(MachineInstr *MI) {
// Don't unfold simple loads.
- if (MI->getDesc().canFoldAsLoad())
+ if (MI->canFoldAsLoad())
return 0;
// If not, we may be able to unfold a load and hoist that.
@@ -1141,8 +1141,9 @@ MachineInstr *MachineLICM::ExtractHoistableLoad(MachineInstr *MI) {
assert(NewMIs.size() == 2 &&
"Unfolded a load into multiple instructions!");
MachineBasicBlock *MBB = MI->getParent();
- MBB->insert(MI, NewMIs[0]);
- MBB->insert(MI, NewMIs[1]);
+ MachineBasicBlock::iterator Pos = MI;
+ MBB->insert(Pos, NewMIs[0]);
+ MBB->insert(Pos, NewMIs[1]);
// If unfolding produced a load that wasn't loop-invariant or profitable to
// hoist, discard the new instructions and bail.
if (!IsLoopInvariantInst(*NewMIs[0]) || !IsProfitableToHoist(*NewMIs[0])) {
diff --git a/lib/CodeGen/MachineSSAUpdater.cpp b/lib/CodeGen/MachineSSAUpdater.cpp
index 84d6df2..8cb6112 100644
--- a/lib/CodeGen/MachineSSAUpdater.cpp
+++ b/lib/CodeGen/MachineSSAUpdater.cpp
@@ -81,7 +81,7 @@ unsigned LookForIdenticalPHI(MachineBasicBlock *BB,
if (BB->empty())
return 0;
- MachineBasicBlock::iterator I = BB->front();
+ MachineBasicBlock::iterator I = BB->begin();
if (!I->isPHI())
return 0;
@@ -182,7 +182,7 @@ unsigned MachineSSAUpdater::GetValueInMiddleOfBlock(MachineBasicBlock *BB) {
return DupPHI;
// Otherwise, we do need a PHI: insert one now.
- MachineBasicBlock::iterator Loc = BB->empty() ? BB->end() : BB->front();
+ MachineBasicBlock::iterator Loc = BB->empty() ? BB->end() : BB->begin();
MachineInstr *InsertedPHI = InsertNewDef(TargetOpcode::PHI, BB,
Loc, VRC, MRI, TII);
@@ -311,7 +311,7 @@ public:
/// Add it into the specified block and return the register.
static unsigned CreateEmptyPHI(MachineBasicBlock *BB, unsigned NumPreds,
MachineSSAUpdater *Updater) {
- MachineBasicBlock::iterator Loc = BB->empty() ? BB->end() : BB->front();
+ MachineBasicBlock::iterator Loc = BB->empty() ? BB->end() : BB->begin();
MachineInstr *PHI = InsertNewDef(TargetOpcode::PHI, BB, Loc,
Updater->VRC, Updater->MRI,
Updater->TII);
diff --git a/lib/CodeGen/MachineSink.cpp b/lib/CodeGen/MachineSink.cpp
index 29cfb49..e47360d 100644
--- a/lib/CodeGen/MachineSink.cpp
+++ b/lib/CodeGen/MachineSink.cpp
@@ -90,6 +90,12 @@ namespace {
bool AllUsesDominatedByBlock(unsigned Reg, MachineBasicBlock *MBB,
MachineBasicBlock *DefMBB,
bool &BreakPHIEdge, bool &LocalUse) const;
+ MachineBasicBlock *FindSuccToSinkTo(MachineInstr *MI, MachineBasicBlock *MBB,
+ bool &BreakPHIEdge);
+ bool isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
+ MachineBasicBlock *MBB,
+ MachineBasicBlock *SuccToSinkTo);
+
bool PerformTrivialForwardCoalescing(MachineInstr *MI,
MachineBasicBlock *MBB);
};
@@ -147,14 +153,10 @@ MachineSinking::AllUsesDominatedByBlock(unsigned Reg,
assert(TargetRegisterInfo::isVirtualRegister(Reg) &&
"Only makes sense for vregs");
+ // Ignore debug uses because debug info doesn't affect the code.
if (MRI->use_nodbg_empty(Reg))
return true;
- // Ignoring debug uses is necessary so debug info doesn't affect the code.
- // This may leave a referencing dbg_value in the original block, before
- // the definition of the vreg. Dwarf generator handles this although the
- // user might not get the right info at runtime.
-
// BreakPHIEdge is true if all the uses are in the successor MBB being sunken
// into and they are all PHI nodes. In this case, machine-sink must break
// the critical edge first. e.g.
@@ -291,7 +293,7 @@ bool MachineSinking::isWorthBreakingCriticalEdge(MachineInstr *MI,
if (!CEBCandidates.insert(std::make_pair(From, To)))
return true;
- if (!MI->isCopy() && !MI->getDesc().isAsCheapAsAMove())
+ if (!MI->isCopy() && !MI->isAsCheapAsAMove())
return true;
// MI is cheap, we probably don't want to break the critical edge for it.
@@ -401,35 +403,76 @@ static void collectDebugValues(MachineInstr *MI,
}
}
-/// SinkInstruction - Determine whether it is safe to sink the specified machine
-/// instruction out of its current block into a successor.
-bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
- // Don't sink insert_subreg, subreg_to_reg, reg_sequence. These are meant to
- // be close to the source to make it easier to coalesce.
- if (AvoidsSinking(MI, MRI))
+/// isPostDominatedBy - Return true if A is post dominated by B.
+static bool isPostDominatedBy(MachineBasicBlock *A, MachineBasicBlock *B) {
+
+ // FIXME - Use real post dominator.
+ if (A->succ_size() != 2)
+ return false;
+ MachineBasicBlock::succ_iterator I = A->succ_begin();
+ if (B == *I)
+ ++I;
+ MachineBasicBlock *OtherSuccBlock = *I;
+ if (OtherSuccBlock->succ_size() != 1 ||
+ *(OtherSuccBlock->succ_begin()) != B)
return false;
- // Check if it's safe to move the instruction.
- if (!MI->isSafeToMove(TII, AA, SawStore))
+ return true;
+}
+
+/// isProfitableToSinkTo - Return true if it is profitable to sink MI.
+bool MachineSinking::isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
+ MachineBasicBlock *MBB,
+ MachineBasicBlock *SuccToSinkTo) {
+ assert (MI && "Invalid MachineInstr!");
+ assert (SuccToSinkTo && "Invalid SinkTo Candidate BB");
+
+ if (MBB == SuccToSinkTo)
return false;
- // FIXME: This should include support for sinking instructions within the
- // block they are currently in to shorten the live ranges. We often get
- // instructions sunk into the top of a large block, but it would be better to
- // also sink them down before their first use in the block. This xform has to
- // be careful not to *increase* register pressure though, e.g. sinking
- // "x = y + z" down if it kills y and z would increase the live ranges of y
- // and z and only shrink the live range of x.
+ // It is profitable if SuccToSinkTo does not post dominate current block.
+ if (!isPostDominatedBy(MBB, SuccToSinkTo))
+ return true;
+
+ // Check if only use in post dominated block is PHI instruction.
+ bool NonPHIUse = false;
+ for (MachineRegisterInfo::use_nodbg_iterator
+ I = MRI->use_nodbg_begin(Reg), E = MRI->use_nodbg_end();
+ I != E; ++I) {
+ MachineInstr *UseInst = &*I;
+ MachineBasicBlock *UseBlock = UseInst->getParent();
+ if (UseBlock == SuccToSinkTo && !UseInst->isPHI())
+ NonPHIUse = true;
+ }
+ if (!NonPHIUse)
+ return true;
+
+ // If SuccToSinkTo post dominates then also it may be profitable if MI
+ // can further profitably sinked into another block in next round.
+ bool BreakPHIEdge = false;
+ // FIXME - If finding successor is compile time expensive then catch results.
+ if (MachineBasicBlock *MBB2 = FindSuccToSinkTo(MI, SuccToSinkTo, BreakPHIEdge))
+ return isProfitableToSinkTo(Reg, MI, SuccToSinkTo, MBB2);
+
+ // If SuccToSinkTo is final destination and it is a post dominator of current
+ // block then it is not profitable to sink MI into SuccToSinkTo block.
+ return false;
+}
+
+/// FindSuccToSinkTo - Find a successor to sink this instruction to.
+MachineBasicBlock *MachineSinking::FindSuccToSinkTo(MachineInstr *MI,
+ MachineBasicBlock *MBB,
+ bool &BreakPHIEdge) {
+
+ assert (MI && "Invalid MachineInstr!");
+ assert (MBB && "Invalid MachineBasicBlock!");
// Loop over all the operands of the specified instruction. If there is
// anything we can't handle, bail out.
- MachineBasicBlock *ParentBlock = MI->getParent();
// SuccToSinkTo - This is the successor to sink this instruction to, once we
// decide.
MachineBasicBlock *SuccToSinkTo = 0;
-
- bool BreakPHIEdge = false;
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg()) continue; // Ignore non-register operands.
@@ -443,23 +486,23 @@ bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
// and we can freely move its uses. Alternatively, if it's allocatable,
// it could get allocated to something with a def during allocation.
if (!MRI->def_empty(Reg))
- return false;
+ return NULL;
if (AllocatableSet.test(Reg))
- return false;
+ return NULL;
// Check for a def among the register's aliases too.
for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
unsigned AliasReg = *Alias;
if (!MRI->def_empty(AliasReg))
- return false;
+ return NULL;
if (AllocatableSet.test(AliasReg))
- return false;
+ return NULL;
}
} else if (!MO.isDead()) {
// A def that isn't dead. We can't move it.
- return false;
+ return NULL;
}
} else {
// Virtual register uses are always safe to sink.
@@ -467,7 +510,7 @@ bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
// If it's not safe to move defs of the register class, then abort.
if (!TII->isSafeToMoveRegClassDefs(MRI->getRegClass(Reg)))
- return false;
+ return NULL;
// FIXME: This picks a successor to sink into based on having one
// successor that dominates all the uses. However, there are cases where
@@ -488,48 +531,79 @@ bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
// If a previous operand picked a block to sink to, then this operand
// must be sinkable to the same block.
bool LocalUse = false;
- if (!AllUsesDominatedByBlock(Reg, SuccToSinkTo, ParentBlock,
+ if (!AllUsesDominatedByBlock(Reg, SuccToSinkTo, MBB,
BreakPHIEdge, LocalUse))
- return false;
+ return NULL;
continue;
}
// Otherwise, we should look at all the successors and decide which one
// we should sink to.
- for (MachineBasicBlock::succ_iterator SI = ParentBlock->succ_begin(),
- E = ParentBlock->succ_end(); SI != E; ++SI) {
+ for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(),
+ E = MBB->succ_end(); SI != E; ++SI) {
+ MachineBasicBlock *SuccBlock = *SI;
bool LocalUse = false;
- if (AllUsesDominatedByBlock(Reg, *SI, ParentBlock,
+ if (AllUsesDominatedByBlock(Reg, SuccBlock, MBB,
BreakPHIEdge, LocalUse)) {
- SuccToSinkTo = *SI;
+ SuccToSinkTo = SuccBlock;
break;
}
if (LocalUse)
// Def is used locally, it's never safe to move this def.
- return false;
+ return NULL;
}
// If we couldn't find a block to sink to, ignore this instruction.
if (SuccToSinkTo == 0)
- return false;
+ return NULL;
+ else if (!isProfitableToSinkTo(Reg, MI, MBB, SuccToSinkTo))
+ return NULL;
}
}
- // If there are no outputs, it must have side-effects.
- if (SuccToSinkTo == 0)
- return false;
+ // It is not possible to sink an instruction into its own block. This can
+ // happen with loops.
+ if (MBB == SuccToSinkTo)
+ return NULL;
// It's not safe to sink instructions to EH landing pad. Control flow into
// landing pad is implicitly defined.
- if (SuccToSinkTo->isLandingPad())
+ if (SuccToSinkTo && SuccToSinkTo->isLandingPad())
+ return NULL;
+
+ return SuccToSinkTo;
+}
+
+/// SinkInstruction - Determine whether it is safe to sink the specified machine
+/// instruction out of its current block into a successor.
+bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
+ // Don't sink insert_subreg, subreg_to_reg, reg_sequence. These are meant to
+ // be close to the source to make it easier to coalesce.
+ if (AvoidsSinking(MI, MRI))
return false;
- // It is not possible to sink an instruction into its own block. This can
- // happen with loops.
- if (MI->getParent() == SuccToSinkTo)
+ // Check if it's safe to move the instruction.
+ if (!MI->isSafeToMove(TII, AA, SawStore))
return false;
+ // FIXME: This should include support for sinking instructions within the
+ // block they are currently in to shorten the live ranges. We often get
+ // instructions sunk into the top of a large block, but it would be better to
+ // also sink them down before their first use in the block. This xform has to
+ // be careful not to *increase* register pressure though, e.g. sinking
+ // "x = y + z" down if it kills y and z would increase the live ranges of y
+ // and z and only shrink the live range of x.
+
+ bool BreakPHIEdge = false;
+ MachineBasicBlock *ParentBlock = MI->getParent();
+ MachineBasicBlock *SuccToSinkTo = FindSuccToSinkTo(MI, ParentBlock, BreakPHIEdge);
+
+ // If there are no outputs, it must have side-effects.
+ if (SuccToSinkTo == 0)
+ return false;
+
+
// If the instruction to move defines a dead physical register which is live
// when leaving the basic block, don't move it because it could turn into a
// "zombie" define of that preg. E.g., EFLAGS. (<rdar://problem/8030636>)
diff --git a/lib/CodeGen/MachineVerifier.cpp b/lib/CodeGen/MachineVerifier.cpp
index f231e3c..0a2c2f8 100644
--- a/lib/CodeGen/MachineVerifier.cpp
+++ b/lib/CodeGen/MachineVerifier.cpp
@@ -279,13 +279,17 @@ bool MachineVerifier::runOnMachineFunction(MachineFunction &MF) {
for (MachineFunction::const_iterator MFI = MF.begin(), MFE = MF.end();
MFI!=MFE; ++MFI) {
visitMachineBasicBlockBefore(MFI);
- for (MachineBasicBlock::const_iterator MBBI = MFI->begin(),
- MBBE = MFI->end(); MBBI != MBBE; ++MBBI) {
+ for (MachineBasicBlock::const_instr_iterator MBBI = MFI->instr_begin(),
+ MBBE = MFI->instr_end(); MBBI != MBBE; ++MBBI) {
if (MBBI->getParent() != MFI) {
report("Bad instruction parent pointer", MFI);
*OS << "Instruction: " << *MBBI;
continue;
}
+ // Skip BUNDLE instruction for now. FIXME: We should add code to verify
+ // the BUNDLE's specifically.
+ if (MBBI->isBundle())
+ continue;
visitMachineInstrBefore(MBBI);
for (unsigned I = 0, E = MBBI->getNumOperands(); I != E; ++I)
visitMachineOperand(&MBBI->getOperand(I), I);
@@ -435,7 +439,7 @@ MachineVerifier::visitMachineBasicBlockBefore(const MachineBasicBlock *MBB) {
report("MBB exits via unconditional fall-through but its successor "
"differs from its CFG successor!", MBB);
}
- if (!MBB->empty() && MBB->back().getDesc().isBarrier() &&
+ if (!MBB->empty() && MBB->back().isBarrier() &&
!TII->isPredicated(&MBB->back())) {
report("MBB exits via unconditional fall-through but ends with a "
"barrier instruction!", MBB);
@@ -456,10 +460,10 @@ MachineVerifier::visitMachineBasicBlockBefore(const MachineBasicBlock *MBB) {
if (MBB->empty()) {
report("MBB exits via unconditional branch but doesn't contain "
"any instructions!", MBB);
- } else if (!MBB->back().getDesc().isBarrier()) {
+ } else if (!MBB->back().isBarrier()) {
report("MBB exits via unconditional branch but doesn't end with a "
"barrier instruction!", MBB);
- } else if (!MBB->back().getDesc().isTerminator()) {
+ } else if (!MBB->back().isTerminator()) {
report("MBB exits via unconditional branch but the branch isn't a "
"terminator instruction!", MBB);
}
@@ -479,10 +483,10 @@ MachineVerifier::visitMachineBasicBlockBefore(const MachineBasicBlock *MBB) {
if (MBB->empty()) {
report("MBB exits via conditional branch/fall-through but doesn't "
"contain any instructions!", MBB);
- } else if (MBB->back().getDesc().isBarrier()) {
+ } else if (MBB->back().isBarrier()) {
report("MBB exits via conditional branch/fall-through but ends with a "
"barrier instruction!", MBB);
- } else if (!MBB->back().getDesc().isTerminator()) {
+ } else if (!MBB->back().isTerminator()) {
report("MBB exits via conditional branch/fall-through but the branch "
"isn't a terminator instruction!", MBB);
}
@@ -499,10 +503,10 @@ MachineVerifier::visitMachineBasicBlockBefore(const MachineBasicBlock *MBB) {
if (MBB->empty()) {
report("MBB exits via conditional branch/branch but doesn't "
"contain any instructions!", MBB);
- } else if (!MBB->back().getDesc().isBarrier()) {
+ } else if (!MBB->back().isBarrier()) {
report("MBB exits via conditional branch/branch but doesn't end with a "
"barrier instruction!", MBB);
- } else if (!MBB->back().getDesc().isTerminator()) {
+ } else if (!MBB->back().isTerminator()) {
report("MBB exits via conditional branch/branch but the branch "
"isn't a terminator instruction!", MBB);
}
@@ -555,9 +559,9 @@ void MachineVerifier::visitMachineInstrBefore(const MachineInstr *MI) {
// Check the MachineMemOperands for basic consistency.
for (MachineInstr::mmo_iterator I = MI->memoperands_begin(),
E = MI->memoperands_end(); I != E; ++I) {
- if ((*I)->isLoad() && !MCID.mayLoad())
+ if ((*I)->isLoad() && !MI->mayLoad())
report("Missing mayLoad flag", MI);
- if ((*I)->isStore() && !MCID.mayStore())
+ if ((*I)->isStore() && !MI->mayStore())
report("Missing mayStore flag", MI);
}
@@ -575,7 +579,7 @@ void MachineVerifier::visitMachineInstrBefore(const MachineInstr *MI) {
}
// Ensure non-terminators don't follow terminators.
- if (MCID.isTerminator()) {
+ if (MI->isTerminator()) {
if (!FirstTerminator)
FirstTerminator = MI;
} else if (FirstTerminator) {
@@ -606,7 +610,7 @@ MachineVerifier::visitMachineOperand(const MachineOperand *MO, unsigned MONum) {
// Don't check if it's the last operand in a variadic instruction. See,
// e.g., LDM_RET in the arm back end.
if (MO->isReg() &&
- !(MCID.isVariadic() && MONum == MCID.getNumOperands()-1)) {
+ !(MI->isVariadic() && MONum == MCID.getNumOperands()-1)) {
if (MO->isDef() && !MCOI.isOptionalDef())
report("Explicit operand marked as def", MO, MONum);
if (MO->isImplicit())
@@ -614,7 +618,7 @@ MachineVerifier::visitMachineOperand(const MachineOperand *MO, unsigned MONum) {
}
} else {
// ARM adds %reg0 operands to indicate predicates. We'll allow that.
- if (MO->isReg() && !MO->isImplicit() && !MCID.isVariadic() && MO->getReg())
+ if (MO->isReg() && !MO->isImplicit() && !MI->isVariadic() && MO->getReg())
report("Extra explicit operand on non-variadic instruction", MO, MONum);
}
@@ -800,11 +804,11 @@ MachineVerifier::visitMachineOperand(const MachineOperand *MO, unsigned MONum) {
LiveInts && !LiveInts->isNotInMIMap(MI)) {
LiveInterval &LI = LiveStks->getInterval(MO->getIndex());
SlotIndex Idx = LiveInts->getInstructionIndex(MI);
- if (MCID.mayLoad() && !LI.liveAt(Idx.getRegSlot(true))) {
+ if (MI->mayLoad() && !LI.liveAt(Idx.getRegSlot(true))) {
report("Instruction loads from dead spill slot", MO, MONum);
*OS << "Live stack: " << LI << '\n';
}
- if (MCID.mayStore() && !LI.liveAt(Idx.getRegSlot())) {
+ if (MI->mayStore() && !LI.liveAt(Idx.getRegSlot())) {
report("Instruction stores to dead spill slot", MO, MONum);
*OS << "Live stack: " << LI << '\n';
}
diff --git a/lib/CodeGen/PHIElimination.cpp b/lib/CodeGen/PHIElimination.cpp
index 6994aa5..0e52496 100644
--- a/lib/CodeGen/PHIElimination.cpp
+++ b/lib/CodeGen/PHIElimination.cpp
@@ -410,7 +410,7 @@ bool PHIElimination::SplitPHIEdges(MachineFunction &MF,
return false; // Quick exit for basic blocks without PHIs.
bool Changed = false;
- for (MachineBasicBlock::const_iterator BBI = MBB.begin(), BBE = MBB.end();
+ for (MachineBasicBlock::iterator BBI = MBB.begin(), BBE = MBB.end();
BBI != BBE && BBI->isPHI(); ++BBI) {
for (unsigned i = 1, e = BBI->getNumOperands(); i != e; i += 2) {
unsigned Reg = BBI->getOperand(i).getReg();
diff --git a/lib/CodeGen/PeepholeOptimizer.cpp b/lib/CodeGen/PeepholeOptimizer.cpp
index bbc7ce2..2a5652a 100644
--- a/lib/CodeGen/PeepholeOptimizer.cpp
+++ b/lib/CodeGen/PeepholeOptimizer.cpp
@@ -292,7 +292,7 @@ bool PeepholeOptimizer::OptimizeBitcastInstr(MachineInstr *MI,
assert(Def && Src && "Malformed bitcast instruction!");
MachineInstr *DefMI = MRI->getVRegDef(Src);
- if (!DefMI || !DefMI->getDesc().isBitcast())
+ if (!DefMI || !DefMI->isBitcast())
return false;
unsigned SrcSrc = 0;
@@ -353,7 +353,7 @@ bool PeepholeOptimizer::isMoveImmediate(MachineInstr *MI,
SmallSet<unsigned, 4> &ImmDefRegs,
DenseMap<unsigned, MachineInstr*> &ImmDefMIs) {
const MCInstrDesc &MCID = MI->getDesc();
- if (!MCID.isMoveImmediate())
+ if (!MI->isMoveImmediate())
return false;
if (MCID.getNumDefs() != 1)
return false;
@@ -428,9 +428,7 @@ bool PeepholeOptimizer::runOnMachineFunction(MachineFunction &MF) {
continue;
}
- const MCInstrDesc &MCID = MI->getDesc();
-
- if (MCID.isBitcast()) {
+ if (MI->isBitcast()) {
if (OptimizeBitcastInstr(MI, MBB)) {
// MI is deleted.
LocalMIs.erase(MI);
@@ -438,7 +436,7 @@ bool PeepholeOptimizer::runOnMachineFunction(MachineFunction &MF) {
MII = First ? I->begin() : llvm::next(PMII);
continue;
}
- } else if (MCID.isCompare()) {
+ } else if (MI->isCompare()) {
if (OptimizeCmpInstr(MI, MBB)) {
// MI is deleted.
LocalMIs.erase(MI);
diff --git a/lib/CodeGen/PostRASchedulerList.cpp b/lib/CodeGen/PostRASchedulerList.cpp
index 7205ed6..fa832c8 100644
--- a/lib/CodeGen/PostRASchedulerList.cpp
+++ b/lib/CodeGen/PostRASchedulerList.cpp
@@ -212,7 +212,8 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
RegClassInfo.runOnMachineFunction(Fn);
// Check for explicit enable/disable of post-ra scheduling.
- TargetSubtargetInfo::AntiDepBreakMode AntiDepMode = TargetSubtargetInfo::ANTIDEP_NONE;
+ TargetSubtargetInfo::AntiDepBreakMode AntiDepMode =
+ TargetSubtargetInfo::ANTIDEP_NONE;
SmallVector<TargetRegisterClass*, 4> CriticalPathRCs;
if (EnablePostRAScheduler.getPosition() > 0) {
if (!EnablePostRAScheduler)
@@ -271,6 +272,8 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
}
I = MI;
--Count;
+ if (MI->isBundle())
+ Count -= MI->getBundleSize();
}
assert(Count == 0 && "Instruction count mismatch!");
assert((MBB->begin() == Current || CurrentCount != 0) &&
@@ -364,7 +367,7 @@ void SchedulePostRATDList::StartBlockForKills(MachineBasicBlock *BB) {
KillIndices[i] = ~0u;
// Determine the live-out physregs for this block.
- if (!BB->empty() && BB->back().getDesc().isReturn()) {
+ if (!BB->empty() && BB->back().isReturn()) {
// In a return block, examine the function live-out regs.
for (MachineRegisterInfo::liveout_iterator I = MRI.liveout_begin(),
E = MRI.liveout_end(); I != E; ++I) {
diff --git a/lib/CodeGen/PrologEpilogInserter.cpp b/lib/CodeGen/PrologEpilogInserter.cpp
index 32c9325..b4fd1cb 100644
--- a/lib/CodeGen/PrologEpilogInserter.cpp
+++ b/lib/CodeGen/PrologEpilogInserter.cpp
@@ -332,7 +332,7 @@ void PEI::insertCSRSpillsAndRestores(MachineFunction &Fn) {
// Skip over all terminator instructions, which are part of the return
// sequence.
MachineBasicBlock::iterator I2 = I;
- while (I2 != MBB->begin() && (--I2)->getDesc().isTerminator())
+ while (I2 != MBB->begin() && (--I2)->isTerminator())
I = I2;
bool AtStart = I == MBB->begin();
@@ -426,11 +426,11 @@ void PEI::insertCSRSpillsAndRestores(MachineFunction &Fn) {
// Skip over all terminator instructions, which are part of the
// return sequence.
- if (! I->getDesc().isTerminator()) {
+ if (! I->isTerminator()) {
++I;
} else {
MachineBasicBlock::iterator I2 = I;
- while (I2 != MBB->begin() && (--I2)->getDesc().isTerminator())
+ while (I2 != MBB->begin() && (--I2)->isTerminator())
I = I2;
}
}
@@ -698,7 +698,7 @@ void PEI::insertPrologEpilogCode(MachineFunction &Fn) {
// Add epilogue to restore the callee-save registers in each exiting block
for (MachineFunction::iterator I = Fn.begin(), E = Fn.end(); I != E; ++I) {
// If last instruction is a return instruction, add an epilogue
- if (!I->empty() && I->back().getDesc().isReturn())
+ if (!I->empty() && I->back().isReturn())
TFI.emitEpilogue(Fn, *I);
}
@@ -706,7 +706,7 @@ void PEI::insertPrologEpilogCode(MachineFunction &Fn) {
// we've been asked for it. This, when linked with a runtime with support
// for segmented stacks (libgcc is one), will result in allocating stack
// space in small chunks instead of one large contiguous block.
- if (EnableSegmentedStacks)
+ if (Fn.getTarget().Options.EnableSegmentedStacks)
TFI.adjustForSegmentedStacks(Fn);
}
diff --git a/lib/CodeGen/RegAllocFast.cpp b/lib/CodeGen/RegAllocFast.cpp
index 4664a3c..c2656c5 100644
--- a/lib/CodeGen/RegAllocFast.cpp
+++ b/lib/CodeGen/RegAllocFast.cpp
@@ -748,8 +748,8 @@ void RAFast::AllocateBasicBlock() {
// and return are tail calls; do not do this for them. The tail callee need
// not take the same registers as input that it produces as output, and there
// are dependencies for its input registers elsewhere.
- if (!MBB->empty() && MBB->back().getDesc().isReturn() &&
- !MBB->back().getDesc().isCall()) {
+ if (!MBB->empty() && MBB->back().isReturn() &&
+ !MBB->back().isCall()) {
MachineInstr *Ret = &MBB->back();
for (MachineRegisterInfo::liveout_iterator
@@ -968,7 +968,7 @@ void RAFast::AllocateBasicBlock() {
}
unsigned DefOpEnd = MI->getNumOperands();
- if (MCID.isCall()) {
+ if (MI->isCall()) {
// Spill all virtregs before a call. This serves two purposes: 1. If an
// exception is thrown, the landing pad is going to expect to find
// registers in their spill slots, and 2. we don't have to wade through
diff --git a/lib/CodeGen/RegAllocPBQP.cpp b/lib/CodeGen/RegAllocPBQP.cpp
index 845ee12..a053ccc 100644
--- a/lib/CodeGen/RegAllocPBQP.cpp
+++ b/lib/CodeGen/RegAllocPBQP.cpp
@@ -34,13 +34,14 @@
#include "LiveRangeEdit.h"
#include "RenderMachineFunction.h"
#include "Spiller.h"
-#include "Splitter.h"
#include "VirtRegMap.h"
#include "RegisterCoalescer.h"
+#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/CalcSpillWeights.h"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "llvm/CodeGen/LiveStackAnalysis.h"
#include "llvm/CodeGen/RegAllocPBQP.h"
+#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
@@ -68,11 +69,6 @@ pbqpCoalescing("pbqp-coalescing",
cl::desc("Attempt coalescing during PBQP register allocation."),
cl::init(false), cl::Hidden);
-static cl::opt<bool>
-pbqpPreSplitting("pbqp-pre-splitting",
- cl::desc("Pre-split before PBQP register allocation."),
- cl::init(false), cl::Hidden);
-
namespace {
///
@@ -93,7 +89,6 @@ public:
initializeCalculateSpillWeightsPass(*PassRegistry::getPassRegistry());
initializeLiveStacksPass(*PassRegistry::getPassRegistry());
initializeMachineLoopInfoPass(*PassRegistry::getPassRegistry());
- initializeLoopSplitterPass(*PassRegistry::getPassRegistry());
initializeVirtRegMapPass(*PassRegistry::getPassRegistry());
initializeRenderMachineFunctionPass(*PassRegistry::getPassRegistry());
}
@@ -444,6 +439,9 @@ void PBQPBuilderWithCoalescing::addVirtRegCoalesce(
void RegAllocPBQP::getAnalysisUsage(AnalysisUsage &au) const {
+ au.setPreservesCFG();
+ au.addRequired<AliasAnalysis>();
+ au.addPreserved<AliasAnalysis>();
au.addRequired<SlotIndexes>();
au.addPreserved<SlotIndexes>();
au.addRequired<LiveIntervals>();
@@ -454,10 +452,10 @@ void RegAllocPBQP::getAnalysisUsage(AnalysisUsage &au) const {
au.addRequired<CalculateSpillWeights>();
au.addRequired<LiveStacks>();
au.addPreserved<LiveStacks>();
+ au.addRequired<MachineDominatorTree>();
+ au.addPreserved<MachineDominatorTree>();
au.addRequired<MachineLoopInfo>();
au.addPreserved<MachineLoopInfo>();
- if (pbqpPreSplitting)
- au.addRequired<LoopSplitter>();
au.addRequired<VirtRegMap>();
au.addRequired<RenderMachineFunction>();
MachineFunctionPass::getAnalysisUsage(au);
diff --git a/lib/CodeGen/RegisterCoalescer.cpp b/lib/CodeGen/RegisterCoalescer.cpp
index 22d6a3b..cd181cd 100644
--- a/lib/CodeGen/RegisterCoalescer.cpp
+++ b/lib/CodeGen/RegisterCoalescer.cpp
@@ -651,8 +651,7 @@ bool RegisterCoalescer::RemoveCopyByCommutingDef(const CoalescerPair &CP,
MachineInstr *DefMI = LIS->getInstructionFromIndex(AValNo->def);
if (!DefMI)
return false;
- const MCInstrDesc &MCID = DefMI->getDesc();
- if (!MCID.isCommutable())
+ if (!DefMI->isCommutable())
return false;
// If DefMI is a two-address instruction then commuting it will change the
// destination register.
@@ -718,7 +717,8 @@ bool RegisterCoalescer::RemoveCopyByCommutingDef(const CoalescerPair &CP,
return false;
if (NewMI != DefMI) {
LIS->ReplaceMachineInstrInMaps(DefMI, NewMI);
- MBB->insert(DefMI, NewMI);
+ MachineBasicBlock::iterator Pos = DefMI;
+ MBB->insert(Pos, NewMI);
MBB->erase(DefMI);
}
unsigned OpIdx = NewMI->findRegisterUseOperandIdx(IntA.reg, false);
@@ -809,14 +809,14 @@ bool RegisterCoalescer::ReMaterializeTrivialDef(LiveInterval &SrcInt,
if (!DefMI)
return false;
assert(DefMI && "Defining instruction disappeared");
- const MCInstrDesc &MCID = DefMI->getDesc();
- if (!MCID.isAsCheapAsAMove())
+ if (!DefMI->isAsCheapAsAMove())
return false;
if (!TII->isTriviallyReMaterializable(DefMI, AA))
return false;
bool SawStore = false;
if (!DefMI->isSafeToMove(TII, AA, SawStore))
return false;
+ const MCInstrDesc &MCID = DefMI->getDesc();
if (MCID.getNumDefs() != 1)
return false;
if (!DefMI->isImplicitDef()) {
diff --git a/lib/CodeGen/ScheduleDAGInstrs.cpp b/lib/CodeGen/ScheduleDAGInstrs.cpp
index 34b8ab0..4418f40 100644
--- a/lib/CodeGen/ScheduleDAGInstrs.cpp
+++ b/lib/CodeGen/ScheduleDAGInstrs.cpp
@@ -157,7 +157,7 @@ void ScheduleDAGInstrs::AddSchedBarrierDeps() {
MachineInstr *ExitMI = InsertPos != BB->end() ? &*InsertPos : 0;
ExitSU.setInstr(ExitMI);
bool AllDepKnown = ExitMI &&
- (ExitMI->getDesc().isCall() || ExitMI->getDesc().isBarrier());
+ (ExitMI->isCall() || ExitMI->isBarrier());
if (ExitMI && AllDepKnown) {
// If it's a call or a barrier, add dependencies on the defs and uses of
// instruction.
@@ -238,13 +238,12 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
continue;
}
- const MCInstrDesc &MCID = MI->getDesc();
- assert(!MCID.isTerminator() && !MI->isLabel() &&
+ assert(!MI->isTerminator() && !MI->isLabel() &&
"Cannot schedule terminators or labels!");
// Create the SUnit for this MI.
SUnit *SU = NewSUnit(MI);
- SU->isCall = MCID.isCall();
- SU->isCommutable = MCID.isCommutable();
+ SU->isCall = MI->isCall();
+ SU->isCommutable = MI->isCommutable();
// Assign the Latency field of SU using target-provided information.
if (UnitLatencies)
@@ -278,8 +277,15 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
continue;
if (DefSU != SU &&
(Kind != SDep::Output || !MO.isDead() ||
- !DefSU->getInstr()->registerDefIsDead(Reg)))
- DefSU->addPred(SDep(SU, Kind, AOLatency, /*Reg=*/Reg));
+ !DefSU->getInstr()->registerDefIsDead(Reg))) {
+ if (Kind == SDep::Anti)
+ DefSU->addPred(SDep(SU, Kind, 0, /*Reg=*/Reg));
+ else {
+ unsigned AOLat = TII->getOutputLatency(InstrItins, MI, j,
+ DefSU->getInstr());
+ DefSU->addPred(SDep(SU, Kind, AOLat, /*Reg=*/Reg));
+ }
+ }
}
for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
std::vector<SUnit *> &MemDefList = Defs[*Alias];
@@ -315,7 +321,7 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
int RegUseIndex = UseMI->findRegisterUseOperandIdx(Reg);
assert(RegUseIndex >= 0 && "UseMI doesn's use register!");
if (RegUseIndex >= 0 &&
- (UseMCID.mayLoad() || UseMCID.mayStore()) &&
+ (UseMI->mayLoad() || UseMI->mayStore()) &&
(unsigned)RegUseIndex < UseMCID.getNumOperands() &&
UseMCID.OpInfo[RegUseIndex].isLookupPtrRegClass())
LDataLatency += SpecialAddressLatency;
@@ -419,9 +425,9 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
// produce more precise dependence information.
#define STORE_LOAD_LATENCY 1
unsigned TrueMemOrderLatency = 0;
- if (MCID.isCall() || MI->hasUnmodeledSideEffects() ||
+ if (MI->isCall() || MI->hasUnmodeledSideEffects() ||
(MI->hasVolatileMemoryRef() &&
- (!MCID.mayLoad() || !MI->isInvariantLoad(AA)))) {
+ (!MI->mayLoad() || !MI->isInvariantLoad(AA)))) {
// Be conservative with these and add dependencies on all memory
// references, even those that are known to not alias.
for (std::map<const Value *, SUnit *>::iterator I =
@@ -460,7 +466,7 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
PendingLoads.clear();
AliasMemDefs.clear();
AliasMemUses.clear();
- } else if (MCID.mayStore()) {
+ } else if (MI->mayStore()) {
bool MayAlias = true;
TrueMemOrderLatency = STORE_LOAD_LATENCY;
if (const Value *V = getUnderlyingObjectForInstr(MI, MFI, MayAlias)) {
@@ -516,7 +522,7 @@ void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
/*Reg=*/0, /*isNormalMemory=*/false,
/*isMustAlias=*/false,
/*isArtificial=*/true));
- } else if (MCID.mayLoad()) {
+ } else if (MI->mayLoad()) {
bool MayAlias = true;
TrueMemOrderLatency = 0;
if (MI->isInvariantLoad(AA)) {
@@ -576,7 +582,7 @@ void ScheduleDAGInstrs::ComputeLatency(SUnit *SU) {
// Simplistic target-independent heuristic: assume that loads take
// extra time.
- if (SU->getInstr()->getDesc().mayLoad())
+ if (SU->getInstr()->mayLoad())
SU->Latency += 2;
} else {
SU->Latency = TII->getInstrLatency(InstrItins, SU->getInstr());
@@ -658,39 +664,33 @@ std::string ScheduleDAGInstrs::getGraphNodeLabel(const SUnit *SU) const {
// EmitSchedule - Emit the machine code in scheduled order.
MachineBasicBlock *ScheduleDAGInstrs::EmitSchedule() {
- // For MachineInstr-based scheduling, we're rescheduling the instructions in
- // the block, so start by removing them from the block.
- while (Begin != InsertPos) {
- MachineBasicBlock::iterator I = Begin;
- ++Begin;
- BB->remove(I);
- }
+ Begin = InsertPos;
// If first instruction was a DBG_VALUE then put it back.
if (FirstDbgValue)
- BB->insert(InsertPos, FirstDbgValue);
+ BB->splice(InsertPos, BB, FirstDbgValue);
// Then re-insert them according to the given schedule.
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
if (SUnit *SU = Sequence[i])
- BB->insert(InsertPos, SU->getInstr());
+ BB->splice(InsertPos, BB, SU->getInstr());
else
// Null SUnit* is a noop.
EmitNoop();
- }
- // Update the Begin iterator, as the first instruction in the block
- // may have been scheduled later.
- if (!Sequence.empty())
- Begin = Sequence[0]->getInstr();
+ // Update the Begin iterator, as the first instruction in the block
+ // may have been scheduled later.
+ if (i == 0)
+ Begin = prior(InsertPos);
+ }
// Reinsert any remaining debug_values.
for (std::vector<std::pair<MachineInstr *, MachineInstr *> >::iterator
DI = DbgValues.end(), DE = DbgValues.begin(); DI != DE; --DI) {
std::pair<MachineInstr *, MachineInstr *> P = *prior(DI);
MachineInstr *DbgValue = P.first;
- MachineInstr *OrigPrivMI = P.second;
- BB->insertAfter(OrigPrivMI, DbgValue);
+ MachineBasicBlock::iterator OrigPrivMI = P.second;
+ BB->splice(++OrigPrivMI, BB, DbgValue);
}
DbgValues.clear();
FirstDbgValue = NULL;
diff --git a/lib/CodeGen/SelectionDAG/CMakeLists.txt b/lib/CodeGen/SelectionDAG/CMakeLists.txt
index ff4184f..6023326 100644
--- a/lib/CodeGen/SelectionDAG/CMakeLists.txt
+++ b/lib/CodeGen/SelectionDAG/CMakeLists.txt
@@ -20,13 +20,3 @@ add_llvm_library(LLVMSelectionDAG
TargetLowering.cpp
TargetSelectionDAGInfo.cpp
)
-
-add_llvm_library_dependencies(LLVMSelectionDAG
- LLVMAnalysis
- LLVMCodeGen
- LLVMCore
- LLVMMC
- LLVMSupport
- LLVMTarget
- LLVMTransformUtils
- )
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index d8208a4..80cf0a8 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -180,7 +180,9 @@ namespace {
SDValue visitSRA(SDNode *N);
SDValue visitSRL(SDNode *N);
SDValue visitCTLZ(SDNode *N);
+ SDValue visitCTLZ_ZERO_UNDEF(SDNode *N);
SDValue visitCTTZ(SDNode *N);
+ SDValue visitCTTZ_ZERO_UNDEF(SDNode *N);
SDValue visitCTPOP(SDNode *N);
SDValue visitSELECT(SDNode *N);
SDValue visitSELECT_CC(SDNode *N);
@@ -361,6 +363,7 @@ CommitTargetLoweringOpt(const TargetLowering::TargetLoweringOpt &TLO) {
/// specified expression for the same cost as the expression itself, or 2 if we
/// can compute the negated form more cheaply than the expression itself.
static char isNegatibleForFree(SDValue Op, bool LegalOperations,
+ const TargetOptions *Options,
unsigned Depth = 0) {
// No compile time optimizations on this type.
if (Op.getValueType() == MVT::ppcf128)
@@ -383,34 +386,39 @@ static char isNegatibleForFree(SDValue Op, bool LegalOperations,
return LegalOperations ? 0 : 1;
case ISD::FADD:
// FIXME: determine better conditions for this xform.
- if (!UnsafeFPMath) return 0;
+ if (!Options->UnsafeFPMath) return 0;
// fold (fsub (fadd A, B)) -> (fsub (fneg A), B)
- if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, Depth+1))
+ if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, Options,
+ Depth + 1))
return V;
// fold (fneg (fadd A, B)) -> (fsub (fneg B), A)
- return isNegatibleForFree(Op.getOperand(1), LegalOperations, Depth+1);
+ return isNegatibleForFree(Op.getOperand(1), LegalOperations, Options,
+ Depth + 1);
case ISD::FSUB:
// We can't turn -(A-B) into B-A when we honor signed zeros.
- if (!UnsafeFPMath) return 0;
+ if (!Options->UnsafeFPMath) return 0;
// fold (fneg (fsub A, B)) -> (fsub B, A)
return 1;
case ISD::FMUL:
case ISD::FDIV:
- if (HonorSignDependentRoundingFPMath()) return 0;
+ if (Options->HonorSignDependentRoundingFPMath()) return 0;
// fold (fneg (fmul X, Y)) -> (fmul (fneg X), Y) or (fmul X, (fneg Y))
- if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, Depth+1))
+ if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, Options,
+ Depth + 1))
return V;
- return isNegatibleForFree(Op.getOperand(1), LegalOperations, Depth+1);
+ return isNegatibleForFree(Op.getOperand(1), LegalOperations, Options,
+ Depth + 1);
case ISD::FP_EXTEND:
case ISD::FP_ROUND:
case ISD::FSIN:
- return isNegatibleForFree(Op.getOperand(0), LegalOperations, Depth+1);
+ return isNegatibleForFree(Op.getOperand(0), LegalOperations, Options,
+ Depth + 1);
}
}
@@ -434,10 +442,11 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG,
}
case ISD::FADD:
// FIXME: determine better conditions for this xform.
- assert(UnsafeFPMath);
+ assert(DAG.getTarget().Options.UnsafeFPMath);
// fold (fneg (fadd A, B)) -> (fsub (fneg A), B)
- if (isNegatibleForFree(Op.getOperand(0), LegalOperations, Depth+1))
+ if (isNegatibleForFree(Op.getOperand(0), LegalOperations,
+ &DAG.getTarget().Options, Depth+1))
return DAG.getNode(ISD::FSUB, Op.getDebugLoc(), Op.getValueType(),
GetNegatedExpression(Op.getOperand(0), DAG,
LegalOperations, Depth+1),
@@ -449,7 +458,7 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG,
Op.getOperand(0));
case ISD::FSUB:
// We can't turn -(A-B) into B-A when we honor signed zeros.
- assert(UnsafeFPMath);
+ assert(DAG.getTarget().Options.UnsafeFPMath);
// fold (fneg (fsub 0, B)) -> B
if (ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(Op.getOperand(0)))
@@ -462,10 +471,11 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG,
case ISD::FMUL:
case ISD::FDIV:
- assert(!HonorSignDependentRoundingFPMath());
+ assert(!DAG.getTarget().Options.HonorSignDependentRoundingFPMath());
// fold (fneg (fmul X, Y)) -> (fmul (fneg X), Y)
- if (isNegatibleForFree(Op.getOperand(0), LegalOperations, Depth+1))
+ if (isNegatibleForFree(Op.getOperand(0), LegalOperations,
+ &DAG.getTarget().Options, Depth+1))
return DAG.getNode(Op.getOpcode(), Op.getDebugLoc(), Op.getValueType(),
GetNegatedExpression(Op.getOperand(0), DAG,
LegalOperations, Depth+1),
@@ -1070,7 +1080,9 @@ SDValue DAGCombiner::visit(SDNode *N) {
case ISD::SRA: return visitSRA(N);
case ISD::SRL: return visitSRL(N);
case ISD::CTLZ: return visitCTLZ(N);
+ case ISD::CTLZ_ZERO_UNDEF: return visitCTLZ_ZERO_UNDEF(N);
case ISD::CTTZ: return visitCTTZ(N);
+ case ISD::CTTZ_ZERO_UNDEF: return visitCTTZ_ZERO_UNDEF(N);
case ISD::CTPOP: return visitCTPOP(N);
case ISD::SELECT: return visitSELECT(N);
case ISD::SELECT_CC: return visitSELECT_CC(N);
@@ -1769,7 +1781,7 @@ SDValue DAGCombiner::visitSDIV(SDNode *N) {
N0, N1);
}
// fold (sdiv X, pow2) -> simple ops after legalize
- if (N1C && !N1C->isNullValue() && !TLI.isIntDivCheap() &&
+ if (N1C && !N1C->isNullValue() &&
(N1C->getAPIntValue().isPowerOf2() ||
(-N1C->getAPIntValue()).isPowerOf2())) {
// If dividing by powers of two is cheap, then don't perform the following
@@ -3709,6 +3721,16 @@ SDValue DAGCombiner::visitCTLZ(SDNode *N) {
return SDValue();
}
+SDValue DAGCombiner::visitCTLZ_ZERO_UNDEF(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ EVT VT = N->getValueType(0);
+
+ // fold (ctlz_zero_undef c1) -> c2
+ if (isa<ConstantSDNode>(N0))
+ return DAG.getNode(ISD::CTLZ_ZERO_UNDEF, N->getDebugLoc(), VT, N0);
+ return SDValue();
+}
+
SDValue DAGCombiner::visitCTTZ(SDNode *N) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
@@ -3719,6 +3741,16 @@ SDValue DAGCombiner::visitCTTZ(SDNode *N) {
return SDValue();
}
+SDValue DAGCombiner::visitCTTZ_ZERO_UNDEF(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ EVT VT = N->getValueType(0);
+
+ // fold (cttz_zero_undef c1) -> c2
+ if (isa<ConstantSDNode>(N0))
+ return DAG.getNode(ISD::CTTZ_ZERO_UNDEF, N->getDebugLoc(), VT, N0);
+ return SDValue();
+}
+
SDValue DAGCombiner::visitCTPOP(SDNode *N) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
@@ -5254,20 +5286,22 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
if (N0CFP && !N1CFP)
return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N1, N0);
// fold (fadd A, 0) -> A
- if (UnsafeFPMath && N1CFP && N1CFP->getValueAPF().isZero())
+ if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+ N1CFP->getValueAPF().isZero())
return N0;
// fold (fadd A, (fneg B)) -> (fsub A, B)
- if (isNegatibleForFree(N1, LegalOperations) == 2)
+ if (isNegatibleForFree(N1, LegalOperations, &DAG.getTarget().Options) == 2)
return DAG.getNode(ISD::FSUB, N->getDebugLoc(), VT, N0,
GetNegatedExpression(N1, DAG, LegalOperations));
// fold (fadd (fneg A), B) -> (fsub B, A)
- if (isNegatibleForFree(N0, LegalOperations) == 2)
+ if (isNegatibleForFree(N0, LegalOperations, &DAG.getTarget().Options) == 2)
return DAG.getNode(ISD::FSUB, N->getDebugLoc(), VT, N1,
GetNegatedExpression(N0, DAG, LegalOperations));
// If allowed, fold (fadd (fadd x, c1), c2) -> (fadd x, (fadd c1, c2))
- if (UnsafeFPMath && N1CFP && N0.getOpcode() == ISD::FADD &&
- N0.getNode()->hasOneUse() && isa<ConstantFPSDNode>(N0.getOperand(1)))
+ if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+ N0.getOpcode() == ISD::FADD && N0.getNode()->hasOneUse() &&
+ isa<ConstantFPSDNode>(N0.getOperand(1)))
return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N0.getOperand(0),
DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
N0.getOperand(1), N1));
@@ -5292,17 +5326,19 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) {
if (N0CFP && N1CFP && VT != MVT::ppcf128)
return DAG.getNode(ISD::FSUB, N->getDebugLoc(), VT, N0, N1);
// fold (fsub A, 0) -> A
- if (UnsafeFPMath && N1CFP && N1CFP->getValueAPF().isZero())
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ N1CFP && N1CFP->getValueAPF().isZero())
return N0;
// fold (fsub 0, B) -> -B
- if (UnsafeFPMath && N0CFP && N0CFP->getValueAPF().isZero()) {
- if (isNegatibleForFree(N1, LegalOperations))
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ N0CFP && N0CFP->getValueAPF().isZero()) {
+ if (isNegatibleForFree(N1, LegalOperations, &DAG.getTarget().Options))
return GetNegatedExpression(N1, DAG, LegalOperations);
if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))
return DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT, N1);
}
// fold (fsub A, (fneg B)) -> (fadd A, B)
- if (isNegatibleForFree(N1, LegalOperations))
+ if (isNegatibleForFree(N1, LegalOperations, &DAG.getTarget().Options))
return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N0,
GetNegatedExpression(N1, DAG, LegalOperations));
@@ -5329,10 +5365,12 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) {
if (N0CFP && !N1CFP)
return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT, N1, N0);
// fold (fmul A, 0) -> 0
- if (UnsafeFPMath && N1CFP && N1CFP->getValueAPF().isZero())
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ N1CFP && N1CFP->getValueAPF().isZero())
return N1;
// fold (fmul A, 0) -> 0, vector edition.
- if (UnsafeFPMath && ISD::isBuildVectorAllZeros(N1.getNode()))
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ ISD::isBuildVectorAllZeros(N1.getNode()))
return N1;
// fold (fmul X, 2.0) -> (fadd X, X)
if (N1CFP && N1CFP->isExactlyValue(+2.0))
@@ -5343,8 +5381,10 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) {
return DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT, N0);
// fold (fmul (fneg X), (fneg Y)) -> (fmul X, Y)
- if (char LHSNeg = isNegatibleForFree(N0, LegalOperations)) {
- if (char RHSNeg = isNegatibleForFree(N1, LegalOperations)) {
+ if (char LHSNeg = isNegatibleForFree(N0, LegalOperations,
+ &DAG.getTarget().Options)) {
+ if (char RHSNeg = isNegatibleForFree(N1, LegalOperations,
+ &DAG.getTarget().Options)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
if (LHSNeg == 2 || RHSNeg == 2)
@@ -5355,7 +5395,8 @@ SDValue DAGCombiner::visitFMUL(SDNode *N) {
}
// If allowed, fold (fmul (fmul x, c1), c2) -> (fmul x, (fmul c1, c2))
- if (UnsafeFPMath && N1CFP && N0.getOpcode() == ISD::FMUL &&
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ N1CFP && N0.getOpcode() == ISD::FMUL &&
N0.getNode()->hasOneUse() && isa<ConstantFPSDNode>(N0.getOperand(1)))
return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT, N0.getOperand(0),
DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
@@ -5383,8 +5424,10 @@ SDValue DAGCombiner::visitFDIV(SDNode *N) {
// (fdiv (fneg X), (fneg Y)) -> (fdiv X, Y)
- if (char LHSNeg = isNegatibleForFree(N0, LegalOperations)) {
- if (char RHSNeg = isNegatibleForFree(N1, LegalOperations)) {
+ if (char LHSNeg = isNegatibleForFree(N0, LegalOperations,
+ &DAG.getTarget().Options)) {
+ if (char RHSNeg = isNegatibleForFree(N1, LegalOperations,
+ &DAG.getTarget().Options)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
if (LHSNeg == 2 || RHSNeg == 2)
@@ -5637,7 +5680,7 @@ SDValue DAGCombiner::visitFNEG(SDNode *N) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
- if (isNegatibleForFree(N0, LegalOperations))
+ if (isNegatibleForFree(N0, LegalOperations, &DAG.getTarget().Options))
return GetNegatedExpression(N0, DAG, LegalOperations);
// Transform fneg(bitconvert(x)) -> bitconvert(x^sign) to avoid loading
@@ -7162,19 +7205,23 @@ SDValue DAGCombiner::visitEXTRACT_SUBVECTOR(SDNode* N) {
if (NVT != SmallVT || NVT.getSizeInBits()*2 != BigVT.getSizeInBits())
return SDValue();
- // Combine:
- // (extract_subvec (insert_subvec V1, V2, InsIdx), ExtIdx)
- // Into:
- // indicies are equal => V1
- // otherwise => (extract_subvec V1, ExtIdx)
- //
- SDValue InsIdx = N->getOperand(1);
- SDValue ExtIdx = V->getOperand(2);
-
- if (InsIdx == ExtIdx)
- return V->getOperand(1);
- return DAG.getNode(ISD::EXTRACT_SUBVECTOR, N->getDebugLoc(), NVT,
- V->getOperand(0), N->getOperand(1));
+ // Only handle cases where both indexes are constants with the same type.
+ ConstantSDNode *InsIdx = dyn_cast<ConstantSDNode>(N->getOperand(1));
+ ConstantSDNode *ExtIdx = dyn_cast<ConstantSDNode>(V->getOperand(2));
+
+ if (InsIdx && ExtIdx &&
+ InsIdx->getValueType(0).getSizeInBits() <= 64 &&
+ ExtIdx->getValueType(0).getSizeInBits() <= 64) {
+ // Combine:
+ // (extract_subvec (insert_subvec V1, V2, InsIdx), ExtIdx)
+ // Into:
+ // indices are equal => V1
+ // otherwise => (extract_subvec V1, ExtIdx)
+ if (InsIdx->getZExtValue() == ExtIdx->getZExtValue())
+ return V->getOperand(1);
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, N->getDebugLoc(), NVT,
+ V->getOperand(0), N->getOperand(1));
+ }
}
return SDValue();
diff --git a/lib/CodeGen/SelectionDAG/FastISel.cpp b/lib/CodeGen/SelectionDAG/FastISel.cpp
index cff37c2..b4946ec 100644
--- a/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -62,8 +62,11 @@
#include "llvm/ADT/Statistic.h"
using namespace llvm;
-STATISTIC(NumFastIselSuccessIndependent, "Number of insts selected by target-independent selector");
-STATISTIC(NumFastIselSuccessTarget, "Number of insts selected by target-specific selector");
+STATISTIC(NumFastIselSuccessIndependent, "Number of insts selected by "
+ "target-independent selector");
+STATISTIC(NumFastIselSuccessTarget, "Number of insts selected by "
+ "target-specific selector");
+STATISTIC(NumFastIselDead, "Number of dead insts removed on failure");
/// startNewBlock - Set the current block to which generated machine
/// instructions will be appended, and clear the local CSE map.
@@ -307,6 +310,18 @@ void FastISel::recomputeInsertPt() {
++FuncInfo.InsertPt;
}
+void FastISel::removeDeadCode(MachineBasicBlock::iterator I,
+ MachineBasicBlock::iterator E) {
+ assert (I && E && std::distance(I, E) > 0 && "Invalid iterator!");
+ while (I != E) {
+ MachineInstr *Dead = &*I;
+ ++I;
+ Dead->eraseFromParent();
+ ++NumFastIselDead;
+ }
+ recomputeInsertPt();
+}
+
FastISel::SavePoint FastISel::enterLocalValueArea() {
MachineBasicBlock::iterator OldInsertPt = FuncInfo.InsertPt;
DebugLoc OldDL = DL;
@@ -792,19 +807,33 @@ FastISel::SelectInstruction(const Instruction *I) {
DL = I->getDebugLoc();
+ MachineBasicBlock::iterator SavedInsertPt = FuncInfo.InsertPt;
+
// First, try doing target-independent selection.
if (SelectOperator(I, I->getOpcode())) {
++NumFastIselSuccessIndependent;
DL = DebugLoc();
return true;
}
+ // Remove dead code. However, ignore call instructions since we've flushed
+ // the local value map and recomputed the insert point.
+ if (!isa<CallInst>(I)) {
+ recomputeInsertPt();
+ if (SavedInsertPt != FuncInfo.InsertPt)
+ removeDeadCode(FuncInfo.InsertPt, SavedInsertPt);
+ }
// Next, try calling the target to attempt to handle the instruction.
+ SavedInsertPt = FuncInfo.InsertPt;
if (TargetSelectInstruction(I)) {
++NumFastIselSuccessTarget;
DL = DebugLoc();
return true;
}
+ // Check for dead code and remove as necessary.
+ recomputeInsertPt();
+ if (SavedInsertPt != FuncInfo.InsertPt)
+ removeDeadCode(FuncInfo.InsertPt, SavedInsertPt);
DL = DebugLoc();
return false;
diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 2ff66f8..cb6fd53 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -294,7 +294,7 @@ InstrEmitter::AddRegisterOperand(MachineInstr *MI, SDValue Op,
const TargetRegisterClass *DstRC = 0;
if (IIOpNum < II->getNumOperands())
DstRC = TII->getRegClass(*II, IIOpNum, TRI);
- assert((DstRC || (MCID.isVariadic() && IIOpNum >= MCID.getNumOperands())) &&
+ assert((DstRC || (MI->isVariadic() && IIOpNum >= MCID.getNumOperands())) &&
"Don't have operand info for this instruction!");
if (DstRC && !MRI->constrainRegClass(VReg, DstRC, MinRCSize)) {
unsigned NewVReg = MRI->createVirtualRegister(DstRC);
diff --git a/lib/CodeGen/SelectionDAG/LLVMBuild.txt b/lib/CodeGen/SelectionDAG/LLVMBuild.txt
index 10a849f..81d2e00 100644
--- a/lib/CodeGen/SelectionDAG/LLVMBuild.txt
+++ b/lib/CodeGen/SelectionDAG/LLVMBuild.txt
@@ -20,4 +20,3 @@ type = Library
name = SelectionDAG
parent = CodeGen
required_libraries = Analysis CodeGen Core MC Support Target TransformUtils
-
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 156cc70..75f5761 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -785,7 +785,6 @@ void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
case ISD::FRAME_TO_ARGS_OFFSET:
case ISD::EH_SJLJ_SETJMP:
case ISD::EH_SJLJ_LONGJMP:
- case ISD::EH_SJLJ_DISPATCHSETUP:
// These operations lie about being legal: when they claim to be legal,
// they should actually be expanded.
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
@@ -2383,6 +2382,9 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
return Op;
}
+ case ISD::CTLZ_ZERO_UNDEF:
+ // This trivially expands to CTLZ.
+ return DAG.getNode(ISD::CTLZ, dl, Op.getValueType(), Op);
case ISD::CTLZ: {
// for now, we do this:
// x = x | (x >> 1);
@@ -2404,6 +2406,9 @@ SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op,
Op = DAG.getNOT(dl, Op, VT);
return DAG.getNode(ISD::CTPOP, dl, VT, Op);
}
+ case ISD::CTTZ_ZERO_UNDEF:
+ // This trivially expands to CTTZ.
+ return DAG.getNode(ISD::CTTZ, dl, Op.getValueType(), Op);
case ISD::CTTZ: {
// for now, we use: { return popcount(~x & (x - 1)); }
// unless the target has ctlz but not ctpop, in which case we use:
@@ -2518,7 +2523,9 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) {
switch (Node->getOpcode()) {
case ISD::CTPOP:
case ISD::CTLZ:
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTTZ:
+ case ISD::CTTZ_ZERO_UNDEF:
Tmp1 = ExpandBitCount(Node->getOpcode(), Node->getOperand(0), dl);
Results.push_back(Tmp1);
break;
@@ -2538,7 +2545,6 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) {
case ISD::PREFETCH:
case ISD::VAEND:
case ISD::EH_SJLJ_LONGJMP:
- case ISD::EH_SJLJ_DISPATCHSETUP:
// If the target didn't expand these, there's nothing to do, so just
// preserve the chain and be done.
Results.push_back(Node->getOperand(0));
@@ -3421,20 +3427,24 @@ void SelectionDAGLegalize::PromoteNode(SDNode *Node) {
SDValue Tmp1, Tmp2, Tmp3;
switch (Node->getOpcode()) {
case ISD::CTTZ:
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTLZ:
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTPOP:
// Zero extend the argument.
Tmp1 = DAG.getNode(ISD::ZERO_EXTEND, dl, NVT, Node->getOperand(0));
- // Perform the larger operation.
+ // Perform the larger operation. For CTPOP and CTTZ_ZERO_UNDEF, this is
+ // already the correct result.
Tmp1 = DAG.getNode(Node->getOpcode(), dl, NVT, Tmp1);
if (Node->getOpcode() == ISD::CTTZ) {
- //if Tmp1 == sizeinbits(NVT) then Tmp1 = sizeinbits(Old VT)
+ // FIXME: This should set a bit in the zero extended value instead.
Tmp2 = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT),
Tmp1, DAG.getConstant(NVT.getSizeInBits(), NVT),
ISD::SETEQ);
Tmp1 = DAG.getNode(ISD::SELECT, dl, NVT, Tmp2,
DAG.getConstant(OVT.getSizeInBits(), NVT), Tmp1);
- } else if (Node->getOpcode() == ISD::CTLZ) {
+ } else if (Node->getOpcode() == ISD::CTLZ ||
+ Node->getOpcode() == ISD::CTLZ_ZERO_UNDEF) {
// Tmp1 = Tmp1 - (sizeinbits(NVT) - sizeinbits(Old VT))
Tmp1 = DAG.getNode(ISD::SUB, dl, NVT, Tmp1,
DAG.getConstant(NVT.getSizeInBits() -
diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index fd24238..1c02c4f 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -56,8 +56,10 @@ void DAGTypeLegalizer::PromoteIntegerResult(SDNode *N, unsigned ResNo) {
case ISD::Constant: Res = PromoteIntRes_Constant(N); break;
case ISD::CONVERT_RNDSAT:
Res = PromoteIntRes_CONVERT_RNDSAT(N); break;
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTLZ: Res = PromoteIntRes_CTLZ(N); break;
case ISD::CTPOP: Res = PromoteIntRes_CTPOP(N); break;
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTTZ: Res = PromoteIntRes_CTTZ(N); break;
case ISD::EXTRACT_VECTOR_ELT:
Res = PromoteIntRes_EXTRACT_VECTOR_ELT(N); break;
@@ -216,7 +218,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_BITCAST(SDNode *N) {
case TargetLowering::TypeLegal:
break;
case TargetLowering::TypePromoteInteger:
- if (NOutVT.bitsEq(NInVT))
+ if (NOutVT.bitsEq(NInVT) && !NOutVT.isVector() && !NInVT.isVector())
// The input promotes to the same size. Convert the promoted value.
return DAG.getNode(ISD::BITCAST, dl, NOutVT, GetPromotedInteger(InOp));
break;
@@ -311,7 +313,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTLZ(SDNode *N) {
DebugLoc dl = N->getDebugLoc();
EVT OVT = N->getValueType(0);
EVT NVT = Op.getValueType();
- Op = DAG.getNode(ISD::CTLZ, dl, NVT, Op);
+ Op = DAG.getNode(N->getOpcode(), dl, NVT, Op);
// Subtract off the extra leading bits in the bigger type.
return DAG.getNode(ISD::SUB, dl, NVT, Op,
DAG.getConstant(NVT.getSizeInBits() -
@@ -329,13 +331,15 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTTZ(SDNode *N) {
EVT OVT = N->getValueType(0);
EVT NVT = Op.getValueType();
DebugLoc dl = N->getDebugLoc();
- // The count is the same in the promoted type except if the original
- // value was zero. This can be handled by setting the bit just off
- // the top of the original type.
- APInt TopBit(NVT.getSizeInBits(), 0);
- TopBit.setBit(OVT.getSizeInBits());
- Op = DAG.getNode(ISD::OR, dl, NVT, Op, DAG.getConstant(TopBit, NVT));
- return DAG.getNode(ISD::CTTZ, dl, NVT, Op);
+ if (N->getOpcode() == ISD::CTTZ) {
+ // The count is the same in the promoted type except if the original
+ // value was zero. This can be handled by setting the bit just off
+ // the top of the original type.
+ APInt TopBit(NVT.getSizeInBits(), 0);
+ TopBit.setBit(OVT.getSizeInBits());
+ Op = DAG.getNode(ISD::OR, dl, NVT, Op, DAG.getConstant(TopBit, NVT));
+ }
+ return DAG.getNode(N->getOpcode(), dl, NVT, Op);
}
SDValue DAGTypeLegalizer::PromoteIntRes_EXTRACT_VECTOR_ELT(SDNode *N) {
@@ -1097,8 +1101,10 @@ void DAGTypeLegalizer::ExpandIntegerResult(SDNode *N, unsigned ResNo) {
case ISD::AssertZext: ExpandIntRes_AssertZext(N, Lo, Hi); break;
case ISD::BSWAP: ExpandIntRes_BSWAP(N, Lo, Hi); break;
case ISD::Constant: ExpandIntRes_Constant(N, Lo, Hi); break;
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTLZ: ExpandIntRes_CTLZ(N, Lo, Hi); break;
case ISD::CTPOP: ExpandIntRes_CTPOP(N, Lo, Hi); break;
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTTZ: ExpandIntRes_CTTZ(N, Lo, Hi); break;
case ISD::FP_TO_SINT: ExpandIntRes_FP_TO_SINT(N, Lo, Hi); break;
case ISD::FP_TO_UINT: ExpandIntRes_FP_TO_UINT(N, Lo, Hi); break;
@@ -1701,8 +1707,8 @@ void DAGTypeLegalizer::ExpandIntRes_CTLZ(SDNode *N,
SDValue HiNotZero = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT), Hi,
DAG.getConstant(0, NVT), ISD::SETNE);
- SDValue LoLZ = DAG.getNode(ISD::CTLZ, dl, NVT, Lo);
- SDValue HiLZ = DAG.getNode(ISD::CTLZ, dl, NVT, Hi);
+ SDValue LoLZ = DAG.getNode(N->getOpcode(), dl, NVT, Lo);
+ SDValue HiLZ = DAG.getNode(ISD::CTLZ_ZERO_UNDEF, dl, NVT, Hi);
Lo = DAG.getNode(ISD::SELECT, dl, NVT, HiNotZero, HiLZ,
DAG.getNode(ISD::ADD, dl, NVT, LoLZ,
@@ -1731,8 +1737,8 @@ void DAGTypeLegalizer::ExpandIntRes_CTTZ(SDNode *N,
SDValue LoNotZero = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT), Lo,
DAG.getConstant(0, NVT), ISD::SETNE);
- SDValue LoLZ = DAG.getNode(ISD::CTTZ, dl, NVT, Lo);
- SDValue HiLZ = DAG.getNode(ISD::CTTZ, dl, NVT, Hi);
+ SDValue LoLZ = DAG.getNode(ISD::CTTZ_ZERO_UNDEF, dl, NVT, Lo);
+ SDValue HiLZ = DAG.getNode(N->getOpcode(), dl, NVT, Hi);
Lo = DAG.getNode(ISD::SELECT, dl, NVT, LoNotZero, LoLZ,
DAG.getNode(ISD::ADD, dl, NVT, HiLZ,
diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
index 4e02b90..4696c0d 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -185,8 +185,10 @@ SDValue VectorLegalizer::LegalizeOp(SDValue Op) {
case ISD::SRL:
case ISD::ROTL:
case ISD::ROTR:
- case ISD::CTTZ:
case ISD::CTLZ:
+ case ISD::CTTZ:
+ case ISD::CTLZ_ZERO_UNDEF:
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTPOP:
case ISD::SELECT:
case ISD::VSELECT:
diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index ad83565..7ca0d1e 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -441,8 +441,10 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
case ISD::ANY_EXTEND:
case ISD::CONVERT_RNDSAT:
case ISD::CTLZ:
- case ISD::CTPOP:
case ISD::CTTZ:
+ case ISD::CTLZ_ZERO_UNDEF:
+ case ISD::CTTZ_ZERO_UNDEF:
+ case ISD::CTPOP:
case ISD::FABS:
case ISD::FCEIL:
case ISD::FCOS:
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
index cd0da37..80162d7 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
@@ -102,17 +102,6 @@ static cl::opt<unsigned> AvgIPC(
"sched-avg-ipc", cl::Hidden, cl::init(1),
cl::desc("Average inst/cycle whan no target itinerary exists."));
-#ifndef NDEBUG
-namespace {
- // For sched=list-ilp, Count the number of times each factor comes into play.
- enum { FactPressureDiff, FactRegUses, FactStall, FactHeight, FactDepth,
- FactStatic, FactOther, NumFactors };
-}
-static const char *FactorName[NumFactors] =
-{"PressureDiff", "RegUses", "Stall", "Height", "Depth","Static", "Other"};
-static int FactorCount[NumFactors];
-#endif //!NDEBUG
-
namespace {
//===----------------------------------------------------------------------===//
/// ScheduleDAGRRList - The actual register reduction list scheduler
@@ -157,6 +146,10 @@ private:
/// and similar queries.
ScheduleDAGTopologicalSort Topo;
+ // Hack to keep track of the inverse of FindCallSeqStart without more crazy
+ // DAG crawling.
+ DenseMap<SUnit*, SUnit*> CallSeqEndForStart;
+
public:
ScheduleDAGRRList(MachineFunction &mf, bool needlatency,
SchedulingPriorityQueue *availqueue,
@@ -308,11 +301,6 @@ void ScheduleDAGRRList::Schedule() {
DEBUG(dbgs()
<< "********** List Scheduling BB#" << BB->getNumber()
<< " '" << BB->getName() << "' **********\n");
-#ifndef NDEBUG
- for (int i = 0; i < NumFactors; ++i) {
- FactorCount[i] = 0;
- }
-#endif //!NDEBUG
CurCycle = 0;
IssueCount = 0;
@@ -322,6 +310,7 @@ void ScheduleDAGRRList::Schedule() {
// to track the virtual resource of a calling sequence.
LiveRegDefs.resize(TRI->getNumRegs() + 1, NULL);
LiveRegGens.resize(TRI->getNumRegs() + 1, NULL);
+ CallSeqEndForStart.clear();
// Build the scheduling graph.
BuildSchedGraph(NULL);
@@ -337,11 +326,6 @@ void ScheduleDAGRRList::Schedule() {
// Execute the actual scheduling loop.
ListScheduleBottomUp();
-#ifndef NDEBUG
- for (int i = 0; i < NumFactors; ++i) {
- DEBUG(dbgs() << FactorName[i] << "\t" << FactorCount[i] << "\n");
- }
-#endif // !NDEBUG
AvailableQueue->releaseState();
}
@@ -545,6 +529,8 @@ void ScheduleDAGRRList::ReleasePredecessors(SUnit *SU) {
SDNode *N = FindCallSeqStart(Node, NestLevel, MaxNest, TII);
SUnit *Def = &SUnits[N->getNodeId()];
+ CallSeqEndForStart[Def] = SU;
+
++NumLiveRegs;
LiveRegDefs[CallResource] = Def;
LiveRegGens[CallResource] = SU;
@@ -811,7 +797,7 @@ void ScheduleDAGRRList::UnscheduleNodeBottomUp(SUnit *SU) {
SUNode->getMachineOpcode() == (unsigned)TII->getCallFrameSetupOpcode()) {
++NumLiveRegs;
LiveRegDefs[CallResource] = SU;
- LiveRegGens[CallResource] = NULL;
+ LiveRegGens[CallResource] = CallSeqEndForStart[SU];
}
}
@@ -832,12 +818,11 @@ void ScheduleDAGRRList::UnscheduleNodeBottomUp(SUnit *SU) {
for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
I != E; ++I) {
if (I->isAssignedRegDep()) {
+ if (!LiveRegDefs[I->getReg()])
+ ++NumLiveRegs;
// This becomes the nearest def. Note that an earlier def may still be
// pending if this is a two-address node.
LiveRegDefs[I->getReg()] = SU;
- if (!LiveRegDefs[I->getReg()]) {
- ++NumLiveRegs;
- }
if (LiveRegGens[I->getReg()] == NULL ||
I->getSUnit()->getHeight() < LiveRegGens[I->getReg()]->getHeight())
LiveRegGens[I->getReg()] = I->getSUnit();
@@ -2296,28 +2281,20 @@ static int BUCompareLatency(SUnit *left, SUnit *right, bool checkPref,
// If scheduling either one of the node will cause a pipeline stall, sort
// them according to their height.
if (LStall) {
- if (!RStall) {
- DEBUG(++FactorCount[FactStall]);
+ if (!RStall)
return 1;
- }
- if (LHeight != RHeight) {
- DEBUG(++FactorCount[FactStall]);
+ if (LHeight != RHeight)
return LHeight > RHeight ? 1 : -1;
- }
- } else if (RStall) {
- DEBUG(++FactorCount[FactStall]);
+ } else if (RStall)
return -1;
- }
// If either node is scheduling for latency, sort them by height/depth
// and latency.
if (!checkPref || (left->SchedulingPref == Sched::ILP ||
right->SchedulingPref == Sched::ILP)) {
if (DisableSchedCycles) {
- if (LHeight != RHeight) {
- DEBUG(++FactorCount[FactHeight]);
+ if (LHeight != RHeight)
return LHeight > RHeight ? 1 : -1;
- }
}
else {
// If neither instruction stalls (!LStall && !RStall) then
@@ -2326,17 +2303,14 @@ static int BUCompareLatency(SUnit *left, SUnit *right, bool checkPref,
int LDepth = left->getDepth() - LPenalty;
int RDepth = right->getDepth() - RPenalty;
if (LDepth != RDepth) {
- DEBUG(++FactorCount[FactDepth]);
DEBUG(dbgs() << " Comparing latency of SU (" << left->NodeNum
<< ") depth " << LDepth << " vs SU (" << right->NodeNum
<< ") depth " << RDepth << "\n");
return LDepth < RDepth ? 1 : -1;
}
}
- if (left->Latency != right->Latency) {
- DEBUG(++FactorCount[FactOther]);
+ if (left->Latency != right->Latency)
return left->Latency > right->Latency ? 1 : -1;
- }
}
return 0;
}
@@ -2350,7 +2324,6 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
bool LHasPhysReg = left->hasPhysRegDefs;
bool RHasPhysReg = right->hasPhysRegDefs;
if (LHasPhysReg != RHasPhysReg) {
- DEBUG(++FactorCount[FactRegUses]);
#ifndef NDEBUG
const char *PhysRegMsg[] = {" has no physreg", " defines a physreg"};
#endif
@@ -2376,10 +2349,8 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
LPriority = (LPriority > LNumVals) ? (LPriority - LNumVals) : 0;
}
- if (LPriority != RPriority) {
- DEBUG(++FactorCount[FactStatic]);
+ if (LPriority != RPriority)
return LPriority > RPriority;
- }
// One or both of the nodes are calls and their sethi-ullman numbers are the
// same, then keep source order.
@@ -2412,18 +2383,14 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
// This creates more short live intervals.
unsigned LDist = closestSucc(left);
unsigned RDist = closestSucc(right);
- if (LDist != RDist) {
- DEBUG(++FactorCount[FactOther]);
+ if (LDist != RDist)
return LDist < RDist;
- }
// How many registers becomes live when the node is scheduled.
unsigned LScratch = calcMaxScratches(left);
unsigned RScratch = calcMaxScratches(right);
- if (LScratch != RScratch) {
- DEBUG(++FactorCount[FactOther]);
+ if (LScratch != RScratch)
return LScratch > RScratch;
- }
// Comparing latency against a call makes little sense unless the node
// is register pressure-neutral.
@@ -2438,20 +2405,15 @@ static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) {
return result > 0;
}
else {
- if (left->getHeight() != right->getHeight()) {
- DEBUG(++FactorCount[FactHeight]);
+ if (left->getHeight() != right->getHeight())
return left->getHeight() > right->getHeight();
- }
- if (left->getDepth() != right->getDepth()) {
- DEBUG(++FactorCount[FactDepth]);
+ if (left->getDepth() != right->getDepth())
return left->getDepth() < right->getDepth();
- }
}
assert(left->NodeQueueId && right->NodeQueueId &&
"NodeQueueId cannot be zero");
- DEBUG(++FactorCount[FactOther]);
return (left->NodeQueueId > right->NodeQueueId);
}
@@ -2511,13 +2473,11 @@ bool hybrid_ls_rr_sort::operator()(SUnit *left, SUnit *right) const {
// Avoid causing spills. If register pressure is high, schedule for
// register pressure reduction.
if (LHigh && !RHigh) {
- DEBUG(++FactorCount[FactPressureDiff]);
DEBUG(dbgs() << " pressure SU(" << left->NodeNum << ") > SU("
<< right->NodeNum << ")\n");
return true;
}
else if (!LHigh && RHigh) {
- DEBUG(++FactorCount[FactPressureDiff]);
DEBUG(dbgs() << " pressure SU(" << right->NodeNum << ") > SU("
<< left->NodeNum << ")\n");
return false;
@@ -2581,7 +2541,6 @@ bool ilp_ls_rr_sort::operator()(SUnit *left, SUnit *right) const {
RPDiff = SPQ->RegPressureDiff(right, RLiveUses);
}
if (!DisableSchedRegPressure && LPDiff != RPDiff) {
- DEBUG(++FactorCount[FactPressureDiff]);
DEBUG(dbgs() << "RegPressureDiff SU(" << left->NodeNum << "): " << LPDiff
<< " != SU(" << right->NodeNum << "): " << RPDiff << "\n");
return LPDiff > RPDiff;
@@ -2590,7 +2549,6 @@ bool ilp_ls_rr_sort::operator()(SUnit *left, SUnit *right) const {
if (!DisableSchedRegPressure && (LPDiff > 0 || RPDiff > 0)) {
bool LReduce = canEnableCoalescing(left);
bool RReduce = canEnableCoalescing(right);
- DEBUG(if (LReduce != RReduce) ++FactorCount[FactPressureDiff]);
if (LReduce && !RReduce) return false;
if (RReduce && !LReduce) return true;
}
@@ -2598,17 +2556,14 @@ bool ilp_ls_rr_sort::operator()(SUnit *left, SUnit *right) const {
if (!DisableSchedLiveUses && (LLiveUses != RLiveUses)) {
DEBUG(dbgs() << "Live uses SU(" << left->NodeNum << "): " << LLiveUses
<< " != SU(" << right->NodeNum << "): " << RLiveUses << "\n");
- DEBUG(++FactorCount[FactRegUses]);
return LLiveUses < RLiveUses;
}
if (!DisableSchedStalls) {
bool LStall = BUHasStall(left, left->getHeight(), SPQ);
bool RStall = BUHasStall(right, right->getHeight(), SPQ);
- if (LStall != RStall) {
- DEBUG(++FactorCount[FactHeight]);
+ if (LStall != RStall)
return left->getHeight() > right->getHeight();
- }
}
if (!DisableSchedCriticalPath) {
@@ -2617,17 +2572,14 @@ bool ilp_ls_rr_sort::operator()(SUnit *left, SUnit *right) const {
DEBUG(dbgs() << "Depth of SU(" << left->NodeNum << "): "
<< left->getDepth() << " != SU(" << right->NodeNum << "): "
<< right->getDepth() << "\n");
- DEBUG(++FactorCount[FactDepth]);
return left->getDepth() < right->getDepth();
}
}
if (!DisableSchedHeight && left->getHeight() != right->getHeight()) {
int spread = (int)left->getHeight() - (int)right->getHeight();
- if (std::abs(spread) > MaxReorderWindow) {
- DEBUG(++FactorCount[FactHeight]);
+ if (std::abs(spread) > MaxReorderWindow)
return left->getHeight() > right->getHeight();
- }
}
return BURRSort(left, right, SPQ);
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 497c286..dd626e2 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -840,9 +840,9 @@ unsigned SelectionDAG::getEVTAlignment(EVT VT) const {
}
// EntryNode could meaningfully have debug info if we can find it...
-SelectionDAG::SelectionDAG(const TargetMachine &tm)
+SelectionDAG::SelectionDAG(const TargetMachine &tm, CodeGenOpt::Level OL)
: TM(tm), TLI(*tm.getTargetLowering()), TSI(*tm.getSelectionDAGInfo()),
- EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)),
+ OptLevel(OL), EntryNode(ISD::EntryToken, DebugLoc(), getVTList(MVT::Other)),
Root(getEntryNode()), Ordering(0) {
AllNodes.push_back(&EntryNode);
Ordering = new SDNodeOrdering();
@@ -1856,7 +1856,9 @@ void SelectionDAG::ComputeMaskedBits(SDValue Op, const APInt &Mask,
return;
}
case ISD::CTTZ:
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTLZ:
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTPOP: {
unsigned LowBits = Log2_32(BitWidth)+1;
KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
@@ -2334,7 +2336,7 @@ bool SelectionDAG::isBaseWithConstantOffset(SDValue Op) const {
bool SelectionDAG::isKnownNeverNaN(SDValue Op) const {
// If we're told that NaNs won't happen, assume they won't.
- if (NoNaNsFPMath)
+ if (getTarget().Options.NoNaNsFPMath)
return true;
// If the value is a constant, we can obviously see if it is a NaN or not.
@@ -2429,8 +2431,10 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL,
case ISD::CTPOP:
return getConstant(Val.countPopulation(), VT);
case ISD::CTLZ:
+ case ISD::CTLZ_ZERO_UNDEF:
return getConstant(Val.countLeadingZeros(), VT);
case ISD::CTTZ:
+ case ISD::CTTZ_ZERO_UNDEF:
return getConstant(Val.countTrailingZeros(), VT);
}
}
@@ -2607,7 +2611,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL,
break;
case ISD::FNEG:
// -(X-Y) -> (Y-X) is unsafe because when X==Y, -0.0 != +0.0
- if (UnsafeFPMath && OpOpcode == ISD::FSUB)
+ if (getTarget().Options.UnsafeFPMath && OpOpcode == ISD::FSUB)
return getNode(ISD::FSUB, DL, VT, Operand.getNode()->getOperand(1),
Operand.getNode()->getOperand(0));
if (OpOpcode == ISD::FNEG) // --X -> X
@@ -2742,7 +2746,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT,
case ISD::FMUL:
case ISD::FDIV:
case ISD::FREM:
- if (UnsafeFPMath) {
+ if (getTarget().Options.UnsafeFPMath) {
if (Opcode == ISD::FADD) {
// 0+x --> x
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N1))
@@ -3065,7 +3069,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, DebugLoc DL, EVT VT,
case ISD::FMUL:
case ISD::FDIV:
case ISD::FREM:
- if (UnsafeFPMath)
+ if (getTarget().Options.UnsafeFPMath)
return N2;
break;
case ISD::MUL:
@@ -4914,6 +4918,20 @@ SDNode *SelectionDAG::SelectNodeTo(SDNode *N, unsigned MachineOpc,
return N;
}
+/// UpdadeDebugLocOnMergedSDNode - If the opt level is -O0 then it throws away
+/// the line number information on the merged node since it is not possible to
+/// preserve the information that operation is associated with multiple lines.
+/// This will make the debugger working better at -O0, were there is a higher
+/// probability having other instructions associated with that line.
+///
+SDNode *SelectionDAG::UpdadeDebugLocOnMergedSDNode(SDNode *N, DebugLoc OLoc) {
+ DebugLoc NLoc = N->getDebugLoc();
+ if (!(NLoc.isUnknown()) && (OptLevel == CodeGenOpt::None) && (OLoc != NLoc)) {
+ N->setDebugLoc(DebugLoc());
+ }
+ return N;
+}
+
/// MorphNodeTo - This *mutates* the specified node to have the specified
/// return type, opcode, and operands.
///
@@ -4935,7 +4953,7 @@ SDNode *SelectionDAG::MorphNodeTo(SDNode *N, unsigned Opc,
FoldingSetNodeID ID;
AddNodeIDNode(ID, Opc, VTs, Ops, NumOps);
if (SDNode *ON = CSEMap.FindNodeOrInsertPos(ID, IP))
- return ON;
+ return UpdadeDebugLocOnMergedSDNode(ON, N->getDebugLoc());
}
if (!RemoveNodeFromCSEMaps(N))
@@ -5139,8 +5157,9 @@ SelectionDAG::getMachineNode(unsigned Opcode, DebugLoc DL, SDVTList VTs,
FoldingSetNodeID ID;
AddNodeIDNode(ID, ~Opcode, VTs, Ops, NumOps);
IP = 0;
- if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
- return cast<MachineSDNode>(E);
+ if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) {
+ return cast<MachineSDNode>(UpdadeDebugLocOnMergedSDNode(E, DL));
+ }
}
// Allocate a new MachineSDNode.
@@ -5943,7 +5962,6 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::EH_RETURN: return "EH_RETURN";
case ISD::EH_SJLJ_SETJMP: return "EH_SJLJ_SETJMP";
case ISD::EH_SJLJ_LONGJMP: return "EH_SJLJ_LONGJMP";
- case ISD::EH_SJLJ_DISPATCHSETUP: return "EH_SJLJ_DISPATCHSETUP";
case ISD::ConstantPool: return "ConstantPool";
case ISD::ExternalSymbol: return "ExternalSymbol";
case ISD::BlockAddress: return "BlockAddress";
@@ -6112,10 +6130,12 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::TRAP: return "trap";
// Bit manipulation
- case ISD::BSWAP: return "bswap";
- case ISD::CTPOP: return "ctpop";
- case ISD::CTTZ: return "cttz";
- case ISD::CTLZ: return "ctlz";
+ case ISD::BSWAP: return "bswap";
+ case ISD::CTPOP: return "ctpop";
+ case ISD::CTTZ: return "cttz";
+ case ISD::CTTZ_ZERO_UNDEF: return "cttz_zero_undef";
+ case ISD::CTLZ: return "ctlz";
+ case ISD::CTLZ_ZERO_UNDEF: return "ctlz_zero_undef";
// Trampolines
case ISD::INIT_TRAMPOLINE: return "init_trampoline";
@@ -6146,6 +6166,11 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::SETLT: return "setlt";
case ISD::SETLE: return "setle";
case ISD::SETNE: return "setne";
+
+ case ISD::SETTRUE: return "settrue";
+ case ISD::SETTRUE2: return "settrue2";
+ case ISD::SETFALSE: return "setfalse";
+ case ISD::SETFALSE2: return "setfalse2";
}
}
}
@@ -6554,20 +6579,15 @@ unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const {
const GlobalValue *GV;
int64_t GVOffset = 0;
if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset)) {
- // If GV has specified alignment, then use it. Otherwise, use the preferred
- // alignment.
- unsigned Align = GV->getAlignment();
- if (!Align) {
- if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV)) {
- if (GVar->hasInitializer()) {
- const TargetData *TD = TLI.getTargetData();
- Align = TD->getPreferredAlignment(GVar);
- }
- }
- if (!Align)
- Align = TLI.getTargetData()->getABITypeAlignment(GV->getType());
- }
- return MinAlign(Align, GVOffset);
+ unsigned PtrWidth = TLI.getPointerTy().getSizeInBits();
+ APInt AllOnes = APInt::getAllOnesValue(PtrWidth);
+ APInt KnownZero(PtrWidth, 0), KnownOne(PtrWidth, 0);
+ llvm::ComputeMaskedBits(const_cast<GlobalValue*>(GV), AllOnes,
+ KnownZero, KnownOne, TLI.getTargetData());
+ unsigned AlignBits = KnownZero.countTrailingOnes();
+ unsigned Align = AlignBits ? 1 << std::min(31U, AlignBits) : 0;
+ if (Align)
+ return MinAlign(Align, GVOffset);
}
// If this is a direct reference to a stack slot, use information about the
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 8d02350..68c9514 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -47,6 +47,7 @@
#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
@@ -812,9 +813,11 @@ void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
}
}
-void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa) {
+void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa,
+ const TargetLibraryInfo *li) {
AA = &aa;
GFI = gfi;
+ LibInfo = li;
TD = DAG.getTarget().getTargetData();
LPadToCallSiteMap.clear();
}
@@ -1335,6 +1338,8 @@ SelectionDAGBuilder::EmitBranchForMergedCondition(const Value *Cond,
Condition = getICmpCondCode(IC->getPredicate());
} else if (const FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
Condition = getFCmpCondCode(FC->getPredicate());
+ if (TM.Options.NoNaNsFPMath)
+ Condition = getFCmpCodeWithoutNaN(Condition);
} else {
Condition = ISD::SETEQ; // silence warning.
llvm_unreachable("Unknown compare instruction");
@@ -2002,7 +2007,7 @@ bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR,
}
static inline bool areJTsAllowed(const TargetLowering &TLI) {
- return !DisableJumpTables &&
+ return !TLI.getTargetMachine().Options.DisableJumpTables &&
(TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other));
}
@@ -2625,6 +2630,8 @@ void SelectionDAGBuilder::visitFCmp(const User &I) {
SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1));
ISD::CondCode Condition = getFCmpCondCode(predicate);
+ if (TM.Options.NoNaNsFPMath)
+ Condition = getFCmpCodeWithoutNaN(Condition);
EVT DestVT = TLI.getValueType(I.getType());
setValue(&I, DAG.getSetCC(getCurDebugLoc(), DestVT, Op1, Op2, Condition));
}
@@ -3095,7 +3102,7 @@ void SelectionDAGBuilder::visitGetElementPtr(const User &I) {
unsigned Amt = ElementSize.logBase2();
IdxN = DAG.getNode(ISD::SHL, getCurDebugLoc(),
N.getValueType(), IdxN,
- DAG.getConstant(Amt, TLI.getPointerTy()));
+ DAG.getConstant(Amt, IdxN.getValueType()));
} else {
SDValue Scale = DAG.getConstant(ElementSize, TLI.getPointerTy());
IdxN = DAG.getNode(ISD::MUL, getCurDebugLoc(),
@@ -4775,11 +4782,6 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
getRoot(), getValue(I.getArgOperand(0))));
return 0;
}
- case Intrinsic::eh_sjlj_dispatch_setup: {
- DAG.setRoot(DAG.getNode(ISD::EH_SJLJ_DISPATCHSETUP, dl, MVT::Other,
- getRoot(), getValue(I.getArgOperand(0))));
- return 0;
- }
case Intrinsic::x86_mmx_pslli_w:
case Intrinsic::x86_mmx_pslli_d:
@@ -4946,14 +4948,18 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
return 0;
case Intrinsic::cttz: {
SDValue Arg = getValue(I.getArgOperand(0));
+ ConstantInt *CI = cast<ConstantInt>(I.getArgOperand(1));
EVT Ty = Arg.getValueType();
- setValue(&I, DAG.getNode(ISD::CTTZ, dl, Ty, Arg));
+ setValue(&I, DAG.getNode(CI->isZero() ? ISD::CTTZ : ISD::CTTZ_ZERO_UNDEF,
+ dl, Ty, Arg));
return 0;
}
case Intrinsic::ctlz: {
SDValue Arg = getValue(I.getArgOperand(0));
+ ConstantInt *CI = cast<ConstantInt>(I.getArgOperand(1));
EVT Ty = Arg.getValueType();
- setValue(&I, DAG.getNode(ISD::CTLZ, dl, Ty, Arg));
+ setValue(&I, DAG.getNode(CI->isZero() ? ISD::CTLZ : ISD::CTLZ_ZERO_UNDEF,
+ dl, Ty, Arg));
return 0;
}
case Intrinsic::ctpop: {
@@ -5064,7 +5070,7 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
}
case Intrinsic::trap: {
- StringRef TrapFuncName = getTrapFunctionName();
+ StringRef TrapFuncName = TM.Options.getTrapFunctionName();
if (TrapFuncName.empty()) {
DAG.setRoot(DAG.getNode(ISD::TRAP, dl,MVT::Other, getRoot()));
return 0;
@@ -5226,7 +5232,7 @@ void SelectionDAGBuilder::LowerCallTo(ImmutableCallSite CS, SDValue Callee,
// If there's a possibility that fast-isel has already selected some amount
// of the current basic block, don't emit a tail call.
- if (isTailCall && EnableFastISel)
+ if (isTailCall && TM.Options.EnableFastISel)
isTailCall = false;
std::pair<SDValue,SDValue> Result =
@@ -5510,7 +5516,9 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
// can't be a library call.
if (!F->hasLocalLinkage() && F->hasName()) {
StringRef Name = F->getName();
- if (Name == "copysign" || Name == "copysignf" || Name == "copysignl") {
+ if ((LibInfo->has(LibFunc::copysign) && Name == "copysign") ||
+ (LibInfo->has(LibFunc::copysignf) && Name == "copysignf") ||
+ (LibInfo->has(LibFunc::copysignl) && Name == "copysignl")) {
if (I.getNumArgOperands() == 2 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
@@ -5521,7 +5529,9 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
LHS.getValueType(), LHS, RHS));
return;
}
- } else if (Name == "fabs" || Name == "fabsf" || Name == "fabsl") {
+ } else if ((LibInfo->has(LibFunc::fabs) && Name == "fabs") ||
+ (LibInfo->has(LibFunc::fabsf) && Name == "fabsf") ||
+ (LibInfo->has(LibFunc::fabsl) && Name == "fabsl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType()) {
@@ -5530,7 +5540,9 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "sin" || Name == "sinf" || Name == "sinl") {
+ } else if ((LibInfo->has(LibFunc::sin) && Name == "sin") ||
+ (LibInfo->has(LibFunc::sinf) && Name == "sinf") ||
+ (LibInfo->has(LibFunc::sinl) && Name == "sinl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
@@ -5540,7 +5552,9 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "cos" || Name == "cosf" || Name == "cosl") {
+ } else if ((LibInfo->has(LibFunc::cos) && Name == "cos") ||
+ (LibInfo->has(LibFunc::cosf) && Name == "cosf") ||
+ (LibInfo->has(LibFunc::cosl) && Name == "cosl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
@@ -5550,7 +5564,9 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
Tmp.getValueType(), Tmp));
return;
}
- } else if (Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl") {
+ } else if ((LibInfo->has(LibFunc::sqrt) && Name == "sqrt") ||
+ (LibInfo->has(LibFunc::sqrtf) && Name == "sqrtf") ||
+ (LibInfo->has(LibFunc::sqrtl) && Name == "sqrtl")) {
if (I.getNumArgOperands() == 1 && // Basic sanity checks.
I.getArgOperand(0)->getType()->isFloatingPointTy() &&
I.getType() == I.getArgOperand(0)->getType() &&
@@ -5560,6 +5576,83 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
Tmp.getValueType(), Tmp));
return;
}
+ } else if ((LibInfo->has(LibFunc::floor) && Name == "floor") ||
+ (LibInfo->has(LibFunc::floorf) && Name == "floorf") ||
+ (LibInfo->has(LibFunc::floorl) && Name == "floorl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FFLOOR, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::nearbyint) && Name == "nearbyint") ||
+ (LibInfo->has(LibFunc::nearbyintf) && Name == "nearbyintf") ||
+ (LibInfo->has(LibFunc::nearbyintl) && Name == "nearbyintl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FNEARBYINT, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::ceil) && Name == "ceil") ||
+ (LibInfo->has(LibFunc::ceilf) && Name == "ceilf") ||
+ (LibInfo->has(LibFunc::ceill) && Name == "ceill")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FCEIL, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::rint) && Name == "rint") ||
+ (LibInfo->has(LibFunc::rintf) && Name == "rintf") ||
+ (LibInfo->has(LibFunc::rintl) && Name == "rintl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FRINT, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::trunc) && Name == "trunc") ||
+ (LibInfo->has(LibFunc::truncf) && Name == "truncf") ||
+ (LibInfo->has(LibFunc::truncl) && Name == "truncl")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FTRUNC, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::log2) && Name == "log2") ||
+ (LibInfo->has(LibFunc::log2f) && Name == "log2f") ||
+ (LibInfo->has(LibFunc::log2l) && Name == "log2l")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FLOG2, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
+ } else if ((LibInfo->has(LibFunc::exp2) && Name == "exp2") ||
+ (LibInfo->has(LibFunc::exp2f) && Name == "exp2f") ||
+ (LibInfo->has(LibFunc::exp2l) && Name == "exp2l")) {
+ if (I.getNumArgOperands() == 1 && // Basic sanity checks.
+ I.getArgOperand(0)->getType()->isFloatingPointTy() &&
+ I.getType() == I.getArgOperand(0)->getType()) {
+ SDValue Tmp = getValue(I.getArgOperand(0));
+ setValue(&I, DAG.getNode(ISD::FEXP2, getCurDebugLoc(),
+ Tmp.getValueType(), Tmp));
+ return;
+ }
} else if (Name == "memcmp") {
if (visitMemCmpCall(I))
return;
@@ -6516,10 +6609,10 @@ SelectionDAGBuilder::CopyValueToVirtualRegister(const Value *V, unsigned Reg) {
/// isOnlyUsedInEntryBlock - If the specified argument is only used in the
/// entry block, return true. This includes arguments used by switches, since
/// the switch may expand into multiple basic blocks.
-static bool isOnlyUsedInEntryBlock(const Argument *A) {
+static bool isOnlyUsedInEntryBlock(const Argument *A, bool FastISel) {
// With FastISel active, we may be splitting blocks, so force creation
// of virtual registers for all non-dead arguments.
- if (EnableFastISel)
+ if (FastISel)
return A->use_empty();
const BasicBlock *Entry = A->getParent()->begin();
@@ -6709,7 +6802,7 @@ void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
SDB->getCurDebugLoc());
SDB->setValue(I, Res);
- if (!EnableFastISel && Res.getOpcode() == ISD::BUILD_PAIR) {
+ if (!TM.Options.EnableFastISel && Res.getOpcode() == ISD::BUILD_PAIR) {
if (LoadSDNode *LNode =
dyn_cast<LoadSDNode>(Res.getOperand(0).getNode()))
if (FrameIndexSDNode *FI =
@@ -6719,7 +6812,7 @@ void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
// If this argument is live outside of the entry block, insert a copy from
// wherever we got it to the vreg that other BB's will reference it as.
- if (!EnableFastISel && Res.getOpcode() == ISD::CopyFromReg) {
+ if (!TM.Options.EnableFastISel && Res.getOpcode() == ISD::CopyFromReg) {
// If we can, though, try to skip creating an unnecessary vreg.
// FIXME: This isn't very clean... it would be nice to make this more
// general. It's also subtly incompatible with the hacks FastISel
@@ -6730,7 +6823,7 @@ void SelectionDAGISel::LowerArguments(const BasicBlock *LLVMBB) {
continue;
}
}
- if (!isOnlyUsedInEntryBlock(I)) {
+ if (!isOnlyUsedInEntryBlock(I, TM.Options.EnableFastISel)) {
FuncInfo->InitializeRegForValue(I);
SDB->CopyToExportRegsIfNeeded(I);
}
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
index 0a21ca3..5147b6c 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
@@ -67,6 +67,7 @@ class SIToFPInst;
class StoreInst;
class SwitchInst;
class TargetData;
+class TargetLibraryInfo;
class TargetLowering;
class TruncInst;
class UIToFPInst;
@@ -294,6 +295,7 @@ public:
SelectionDAG &DAG;
const TargetData *TD;
AliasAnalysis *AA;
+ const TargetLibraryInfo *LibInfo;
/// SwitchCases - Vector of CaseBlock structures used to communicate
/// SwitchInst code generation information.
@@ -338,7 +340,8 @@ public:
HasTailCall(false), Context(dag.getContext()) {
}
- void init(GCFunctionInfo *gfi, AliasAnalysis &aa);
+ void init(GCFunctionInfo *gfi, AliasAnalysis &aa,
+ const TargetLibraryInfo *li);
/// clear - Clear out the current SelectionDAG and the associated
/// state and prepare this SelectionDAGBuilder object to be used
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index 8cecc17..3c95059 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -41,6 +41,7 @@
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
@@ -61,6 +62,81 @@ STATISTIC(NumFastIselBlocks, "Number of blocks selected entirely by fast isel");
STATISTIC(NumDAGBlocks, "Number of blocks selected using DAG");
STATISTIC(NumDAGIselRetries,"Number of times dag isel has to try another path");
+#ifndef NDEBUG
+static cl::opt<bool>
+EnableFastISelVerbose2("fast-isel-verbose2", cl::Hidden,
+ cl::desc("Enable extra verbose messages in the \"fast\" "
+ "instruction selector"));
+ // Terminators
+STATISTIC(NumFastIselFailRet,"Fast isel fails on Ret");
+STATISTIC(NumFastIselFailBr,"Fast isel fails on Br");
+STATISTIC(NumFastIselFailSwitch,"Fast isel fails on Switch");
+STATISTIC(NumFastIselFailIndirectBr,"Fast isel fails on IndirectBr");
+STATISTIC(NumFastIselFailInvoke,"Fast isel fails on Invoke");
+STATISTIC(NumFastIselFailResume,"Fast isel fails on Resume");
+STATISTIC(NumFastIselFailUnwind,"Fast isel fails on Unwind");
+STATISTIC(NumFastIselFailUnreachable,"Fast isel fails on Unreachable");
+
+ // Standard binary operators...
+STATISTIC(NumFastIselFailAdd,"Fast isel fails on Add");
+STATISTIC(NumFastIselFailFAdd,"Fast isel fails on FAdd");
+STATISTIC(NumFastIselFailSub,"Fast isel fails on Sub");
+STATISTIC(NumFastIselFailFSub,"Fast isel fails on FSub");
+STATISTIC(NumFastIselFailMul,"Fast isel fails on Mul");
+STATISTIC(NumFastIselFailFMul,"Fast isel fails on FMul");
+STATISTIC(NumFastIselFailUDiv,"Fast isel fails on UDiv");
+STATISTIC(NumFastIselFailSDiv,"Fast isel fails on SDiv");
+STATISTIC(NumFastIselFailFDiv,"Fast isel fails on FDiv");
+STATISTIC(NumFastIselFailURem,"Fast isel fails on URem");
+STATISTIC(NumFastIselFailSRem,"Fast isel fails on SRem");
+STATISTIC(NumFastIselFailFRem,"Fast isel fails on FRem");
+
+ // Logical operators...
+STATISTIC(NumFastIselFailAnd,"Fast isel fails on And");
+STATISTIC(NumFastIselFailOr,"Fast isel fails on Or");
+STATISTIC(NumFastIselFailXor,"Fast isel fails on Xor");
+
+ // Memory instructions...
+STATISTIC(NumFastIselFailAlloca,"Fast isel fails on Alloca");
+STATISTIC(NumFastIselFailLoad,"Fast isel fails on Load");
+STATISTIC(NumFastIselFailStore,"Fast isel fails on Store");
+STATISTIC(NumFastIselFailAtomicCmpXchg,"Fast isel fails on AtomicCmpXchg");
+STATISTIC(NumFastIselFailAtomicRMW,"Fast isel fails on AtomicRWM");
+STATISTIC(NumFastIselFailFence,"Fast isel fails on Frence");
+STATISTIC(NumFastIselFailGetElementPtr,"Fast isel fails on GetElementPtr");
+
+ // Convert instructions...
+STATISTIC(NumFastIselFailTrunc,"Fast isel fails on Trunc");
+STATISTIC(NumFastIselFailZExt,"Fast isel fails on ZExt");
+STATISTIC(NumFastIselFailSExt,"Fast isel fails on SExt");
+STATISTIC(NumFastIselFailFPTrunc,"Fast isel fails on FPTrunc");
+STATISTIC(NumFastIselFailFPExt,"Fast isel fails on FPExt");
+STATISTIC(NumFastIselFailFPToUI,"Fast isel fails on FPToUI");
+STATISTIC(NumFastIselFailFPToSI,"Fast isel fails on FPToSI");
+STATISTIC(NumFastIselFailUIToFP,"Fast isel fails on UIToFP");
+STATISTIC(NumFastIselFailSIToFP,"Fast isel fails on SIToFP");
+STATISTIC(NumFastIselFailIntToPtr,"Fast isel fails on IntToPtr");
+STATISTIC(NumFastIselFailPtrToInt,"Fast isel fails on PtrToInt");
+STATISTIC(NumFastIselFailBitCast,"Fast isel fails on BitCast");
+
+ // Other instructions...
+STATISTIC(NumFastIselFailICmp,"Fast isel fails on ICmp");
+STATISTIC(NumFastIselFailFCmp,"Fast isel fails on FCmp");
+STATISTIC(NumFastIselFailPHI,"Fast isel fails on PHI");
+STATISTIC(NumFastIselFailSelect,"Fast isel fails on Select");
+STATISTIC(NumFastIselFailCall,"Fast isel fails on Call");
+STATISTIC(NumFastIselFailShl,"Fast isel fails on Shl");
+STATISTIC(NumFastIselFailLShr,"Fast isel fails on LShr");
+STATISTIC(NumFastIselFailAShr,"Fast isel fails on AShr");
+STATISTIC(NumFastIselFailVAArg,"Fast isel fails on VAArg");
+STATISTIC(NumFastIselFailExtractElement,"Fast isel fails on ExtractElement");
+STATISTIC(NumFastIselFailInsertElement,"Fast isel fails on InsertElement");
+STATISTIC(NumFastIselFailShuffleVector,"Fast isel fails on ShuffleVector");
+STATISTIC(NumFastIselFailExtractValue,"Fast isel fails on ExtractValue");
+STATISTIC(NumFastIselFailInsertValue,"Fast isel fails on InsertValue");
+STATISTIC(NumFastIselFailLandingPad,"Fast isel fails on LandingPad");
+#endif
+
static cl::opt<bool>
EnableFastISelVerbose("fast-isel-verbose", cl::Hidden,
cl::desc("Enable verbose messages in the \"fast\" "
@@ -177,7 +253,7 @@ TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
void TargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI,
SDNode *Node) const {
- assert(!MI->getDesc().hasPostISelHook() &&
+ assert(!MI->hasPostISelHook() &&
"If a target marks an instruction with 'hasPostISelHook', "
"it must implement TargetLowering::AdjustInstrPostInstrSelection!");
}
@@ -190,7 +266,7 @@ SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm,
CodeGenOpt::Level OL) :
MachineFunctionPass(ID), TM(tm), TLI(*tm.getTargetLowering()),
FuncInfo(new FunctionLoweringInfo(TLI)),
- CurDAG(new SelectionDAG(tm)),
+ CurDAG(new SelectionDAG(tm, OL)),
SDB(new SelectionDAGBuilder(*CurDAG, *FuncInfo, OL)),
GFI(),
OptLevel(OL),
@@ -198,6 +274,7 @@ SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm,
initializeGCModuleInfoPass(*PassRegistry::getPassRegistry());
initializeAliasAnalysisAnalysisGroup(*PassRegistry::getPassRegistry());
initializeBranchProbabilityInfoPass(*PassRegistry::getPassRegistry());
+ initializeTargetLibraryInfoPass(*PassRegistry::getPassRegistry());
}
SelectionDAGISel::~SelectionDAGISel() {
@@ -211,6 +288,7 @@ void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const {
AU.addPreserved<AliasAnalysis>();
AU.addRequired<GCModuleInfo>();
AU.addPreserved<GCModuleInfo>();
+ AU.addRequired<TargetLibraryInfo>();
if (UseMBPI && OptLevel != CodeGenOpt::None)
AU.addRequired<BranchProbabilityInfo>();
MachineFunctionPass::getAnalysisUsage(AU);
@@ -256,9 +334,9 @@ static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) {
bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) {
// Do some sanity-checking on the command-line options.
- assert((!EnableFastISelVerbose || EnableFastISel) &&
+ assert((!EnableFastISelVerbose || TM.Options.EnableFastISel) &&
"-fast-isel-verbose requires -fast-isel");
- assert((!EnableFastISelAbort || EnableFastISel) &&
+ assert((!EnableFastISelAbort || TM.Options.EnableFastISel) &&
"-fast-isel-abort requires -fast-isel");
const Function &Fn = *mf.getFunction();
@@ -268,6 +346,7 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) {
MF = &mf;
RegInfo = &MF->getRegInfo();
AA = &getAnalysis<AliasAnalysis>();
+ LibInfo = &getAnalysis<TargetLibraryInfo>();
GFI = Fn.hasGC() ? &getAnalysis<GCModuleInfo>().getFunctionInfo(Fn) : 0;
DEBUG(dbgs() << "\n\n\n=== " << Fn.getName() << "\n");
@@ -282,7 +361,7 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) {
else
FuncInfo->BPI = 0;
- SDB->init(GFI, *AA);
+ SDB->init(GFI, *AA, LibInfo);
SelectAllBasicBlocks(Fn);
@@ -346,7 +425,8 @@ bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) {
TII.get(TargetOpcode::DBG_VALUE))
.addReg(CopyUseMI->getOperand(0).getReg(), RegState::Debug)
.addImm(Offset).addMetadata(Variable);
- EntryMBB->insertAfter(CopyUseMI, NewMI);
+ MachineBasicBlock::iterator Pos = CopyUseMI;
+ EntryMBB->insertAfter(Pos, NewMI);
}
}
}
@@ -820,10 +900,88 @@ static bool isFoldedOrDeadInstruction(const Instruction *I,
!FuncInfo->isExportedInst(I); // Exported instrs must be computed.
}
+#ifndef NDEBUG
+static void collectFailStats(const Instruction *I) {
+ switch (I->getOpcode()) {
+ default: assert (0 && "<Invalid operator> ");
+
+ // Terminators
+ case Instruction::Ret: NumFastIselFailRet++; return;
+ case Instruction::Br: NumFastIselFailBr++; return;
+ case Instruction::Switch: NumFastIselFailSwitch++; return;
+ case Instruction::IndirectBr: NumFastIselFailIndirectBr++; return;
+ case Instruction::Invoke: NumFastIselFailInvoke++; return;
+ case Instruction::Resume: NumFastIselFailResume++; return;
+ case Instruction::Unwind: NumFastIselFailUnwind++; return;
+ case Instruction::Unreachable: NumFastIselFailUnreachable++; return;
+
+ // Standard binary operators...
+ case Instruction::Add: NumFastIselFailAdd++; return;
+ case Instruction::FAdd: NumFastIselFailFAdd++; return;
+ case Instruction::Sub: NumFastIselFailSub++; return;
+ case Instruction::FSub: NumFastIselFailFSub++; return;
+ case Instruction::Mul: NumFastIselFailMul++; return;
+ case Instruction::FMul: NumFastIselFailFMul++; return;
+ case Instruction::UDiv: NumFastIselFailUDiv++; return;
+ case Instruction::SDiv: NumFastIselFailSDiv++; return;
+ case Instruction::FDiv: NumFastIselFailFDiv++; return;
+ case Instruction::URem: NumFastIselFailURem++; return;
+ case Instruction::SRem: NumFastIselFailSRem++; return;
+ case Instruction::FRem: NumFastIselFailFRem++; return;
+
+ // Logical operators...
+ case Instruction::And: NumFastIselFailAnd++; return;
+ case Instruction::Or: NumFastIselFailOr++; return;
+ case Instruction::Xor: NumFastIselFailXor++; return;
+
+ // Memory instructions...
+ case Instruction::Alloca: NumFastIselFailAlloca++; return;
+ case Instruction::Load: NumFastIselFailLoad++; return;
+ case Instruction::Store: NumFastIselFailStore++; return;
+ case Instruction::AtomicCmpXchg: NumFastIselFailAtomicCmpXchg++; return;
+ case Instruction::AtomicRMW: NumFastIselFailAtomicRMW++; return;
+ case Instruction::Fence: NumFastIselFailFence++; return;
+ case Instruction::GetElementPtr: NumFastIselFailGetElementPtr++; return;
+
+ // Convert instructions...
+ case Instruction::Trunc: NumFastIselFailTrunc++; return;
+ case Instruction::ZExt: NumFastIselFailZExt++; return;
+ case Instruction::SExt: NumFastIselFailSExt++; return;
+ case Instruction::FPTrunc: NumFastIselFailFPTrunc++; return;
+ case Instruction::FPExt: NumFastIselFailFPExt++; return;
+ case Instruction::FPToUI: NumFastIselFailFPToUI++; return;
+ case Instruction::FPToSI: NumFastIselFailFPToSI++; return;
+ case Instruction::UIToFP: NumFastIselFailUIToFP++; return;
+ case Instruction::SIToFP: NumFastIselFailSIToFP++; return;
+ case Instruction::IntToPtr: NumFastIselFailIntToPtr++; return;
+ case Instruction::PtrToInt: NumFastIselFailPtrToInt++; return;
+ case Instruction::BitCast: NumFastIselFailBitCast++; return;
+
+ // Other instructions...
+ case Instruction::ICmp: NumFastIselFailICmp++; return;
+ case Instruction::FCmp: NumFastIselFailFCmp++; return;
+ case Instruction::PHI: NumFastIselFailPHI++; return;
+ case Instruction::Select: NumFastIselFailSelect++; return;
+ case Instruction::Call: NumFastIselFailCall++; return;
+ case Instruction::Shl: NumFastIselFailShl++; return;
+ case Instruction::LShr: NumFastIselFailLShr++; return;
+ case Instruction::AShr: NumFastIselFailAShr++; return;
+ case Instruction::VAArg: NumFastIselFailVAArg++; return;
+ case Instruction::ExtractElement: NumFastIselFailExtractElement++; return;
+ case Instruction::InsertElement: NumFastIselFailInsertElement++; return;
+ case Instruction::ShuffleVector: NumFastIselFailShuffleVector++; return;
+ case Instruction::ExtractValue: NumFastIselFailExtractValue++; return;
+ case Instruction::InsertValue: NumFastIselFailInsertValue++; return;
+ case Instruction::LandingPad: NumFastIselFailLandingPad++; return;
+ }
+ return;
+}
+#endif
+
void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) {
// Initialize the Fast-ISel state, if needed.
FastISel *FastIS = 0;
- if (EnableFastISel)
+ if (TM.Options.EnableFastISel)
FastIS = TLI.createFastISel(*FuncInfo);
// Iterate over all basic blocks in the function.
@@ -931,6 +1089,11 @@ void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) {
continue;
}
+#ifndef NDEBUG
+ if (EnableFastISelVerbose2)
+ collectFailStats(Inst);
+#endif
+
// Then handle certain instructions as single-LLVM-Instruction blocks.
if (isa<CallInst>(Inst)) {
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index d7bad43..9ced1ac 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -577,16 +577,26 @@ TargetLowering::TargetLowering(const TargetMachine &tm,
setOperationAction(ISD::ConstantFP, MVT::f80, Expand);
// These library functions default to expand.
- setOperationAction(ISD::FLOG , MVT::f64, Expand);
- setOperationAction(ISD::FLOG2, MVT::f64, Expand);
- setOperationAction(ISD::FLOG10,MVT::f64, Expand);
- setOperationAction(ISD::FEXP , MVT::f64, Expand);
- setOperationAction(ISD::FEXP2, MVT::f64, Expand);
- setOperationAction(ISD::FLOG , MVT::f32, Expand);
- setOperationAction(ISD::FLOG2, MVT::f32, Expand);
- setOperationAction(ISD::FLOG10,MVT::f32, Expand);
- setOperationAction(ISD::FEXP , MVT::f32, Expand);
- setOperationAction(ISD::FEXP2, MVT::f32, Expand);
+ setOperationAction(ISD::FLOG , MVT::f64, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f64, Expand);
+ setOperationAction(ISD::FEXP , MVT::f64, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f64, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f64, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f64, Expand);
+ setOperationAction(ISD::FRINT, MVT::f64, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG , MVT::f32, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f32, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f32, Expand);
+ setOperationAction(ISD::FEXP , MVT::f32, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f32, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f32, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f32, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f32, Expand);
+ setOperationAction(ISD::FRINT, MVT::f32, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f32, Expand);
// Default ISD::TRAP to expand (which turns it into abort).
setOperationAction(ISD::TRAP, MVT::Other, Expand);
@@ -1473,9 +1483,8 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
if (InOp.getNode()->getOpcode() == ISD::ANY_EXTEND) {
SDValue InnerOp = InOp.getNode()->getOperand(0);
EVT InnerVT = InnerOp.getValueType();
- if ((APInt::getHighBitsSet(BitWidth,
- BitWidth - InnerVT.getSizeInBits()) &
- DemandedMask) == 0 &&
+ unsigned InnerBits = InnerVT.getSizeInBits();
+ if (ShAmt < InnerBits && NewMask.lshr(InnerBits) == 0 &&
isTypeDesirableForOp(ISD::SHL, InnerVT)) {
EVT ShTy = getShiftAmountTy(InnerVT);
if (!APInt(BitWidth, ShAmt).isIntN(ShTy.getSizeInBits()))
@@ -1545,7 +1554,7 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
// always convert this into a logical shr, even if the shift amount is
// variable. The low bit of the shift cannot be an input sign bit unless
// the shift amount is >= the size of the datatype, which is undefined.
- if (DemandedMask == 1)
+ if (NewMask == 1)
return TLO.CombineTo(Op,
TLO.DAG.getNode(ISD::SRL, dl, Op.getValueType(),
Op.getOperand(0), Op.getOperand(1)));
@@ -1783,7 +1792,8 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
case ISD::BITCAST:
// If this is an FP->Int bitcast and if the sign bit is the only
// thing demanded, turn this into a FGETSIGN.
- if (!Op.getValueType().isVector() &&
+ if (!TLO.LegalOperations() &&
+ !Op.getValueType().isVector() &&
!Op.getOperand(0).getValueType().isVector() &&
NewMask == APInt::getSignBit(Op.getValueType().getSizeInBits()) &&
Op.getOperand(0).getValueType().isFloatingPoint()) {
diff --git a/lib/CodeGen/ShrinkWrapping.cpp b/lib/CodeGen/ShrinkWrapping.cpp
index 13f269e..70fcf55 100644
--- a/lib/CodeGen/ShrinkWrapping.cpp
+++ b/lib/CodeGen/ShrinkWrapping.cpp
@@ -124,7 +124,7 @@ MachineLoop* PEI::getTopLevelLoopParent(MachineLoop *LP) {
}
bool PEI::isReturnBlock(MachineBasicBlock* MBB) {
- return (MBB && !MBB->empty() && MBB->back().getDesc().isReturn());
+ return (MBB && !MBB->empty() && MBB->back().isReturn());
}
// Initialize shrink wrapping DFA sets, called before iterations.
diff --git a/lib/CodeGen/SjLjEHPrepare.cpp b/lib/CodeGen/SjLjEHPrepare.cpp
index c865192..8e2f74f 100644
--- a/lib/CodeGen/SjLjEHPrepare.cpp
+++ b/lib/CodeGen/SjLjEHPrepare.cpp
@@ -69,6 +69,8 @@ namespace {
private:
bool setupEntryBlockAndCallSites(Function &F);
+ void substituteLPadValues(LandingPadInst *LPI, Value *ExnVal,
+ Value *SelVal);
Value *setupFunctionContext(Function &F, ArrayRef<LandingPadInst*> LPads);
void lowerIncomingArguments(Function &F);
void lowerAcrossUnwindEdges(Function &F, ArrayRef<InvokeInst*> Invokes);
@@ -138,6 +140,38 @@ static void MarkBlocksLiveIn(BasicBlock *BB,
MarkBlocksLiveIn(*PI, LiveBBs);
}
+/// substituteLPadValues - Substitute the values returned by the landingpad
+/// instruction with those returned by the personality function.
+void SjLjEHPass::substituteLPadValues(LandingPadInst *LPI, Value *ExnVal,
+ Value *SelVal) {
+ SmallVector<Value*, 8> UseWorkList(LPI->use_begin(), LPI->use_end());
+ while (!UseWorkList.empty()) {
+ Value *Val = UseWorkList.pop_back_val();
+ ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(Val);
+ if (!EVI) continue;
+ if (EVI->getNumIndices() != 1) continue;
+ if (*EVI->idx_begin() == 0)
+ EVI->replaceAllUsesWith(ExnVal);
+ else if (*EVI->idx_begin() == 1)
+ EVI->replaceAllUsesWith(SelVal);
+ if (EVI->getNumUses() == 0)
+ EVI->eraseFromParent();
+ }
+
+ if (LPI->getNumUses() == 0) return;
+
+ // There are still some uses of LPI. Construct an aggregate with the exception
+ // values and replace the LPI with that aggregate.
+ Type *LPadType = LPI->getType();
+ Value *LPadVal = UndefValue::get(LPadType);
+ IRBuilder<>
+ Builder(llvm::next(BasicBlock::iterator(cast<Instruction>(SelVal))));
+ LPadVal = Builder.CreateInsertValue(LPadVal, ExnVal, 0, "lpad.val");
+ LPadVal = Builder.CreateInsertValue(LPadVal, SelVal, 1, "lpad.val");
+
+ LPI->replaceAllUsesWith(LPadVal);
+}
+
/// setupFunctionContext - Allocate the function context on the stack and fill
/// it with all of the data that we know at this point.
Value *SjLjEHPass::
@@ -189,12 +223,7 @@ setupFunctionContext(Function &F, ArrayRef<LandingPadInst*> LPads) {
ExnVal = Builder.CreateIntToPtr(ExnVal, Type::getInt8PtrTy(F.getContext()));
Value *SelVal = Builder.CreateLoad(SelectorAddr, true, "exn_selector_val");
- Type *LPadType = LPI->getType();
- Value *LPadVal = UndefValue::get(LPadType);
- LPadVal = Builder.CreateInsertValue(LPadVal, ExnVal, 0, "lpad.val");
- LPadVal = Builder.CreateInsertValue(LPadVal, SelVal, 1, "lpad.val");
-
- LPI->replaceAllUsesWith(LPadVal);
+ substituteLPadValues(LPI, ExnVal, SelVal);
}
// Personality function
diff --git a/lib/CodeGen/SplitKit.cpp b/lib/CodeGen/SplitKit.cpp
index 751d604..c086073 100644
--- a/lib/CodeGen/SplitKit.cpp
+++ b/lib/CodeGen/SplitKit.cpp
@@ -80,7 +80,7 @@ SlotIndex SplitAnalysis::computeLastSplitPoint(unsigned Num) {
for (MachineBasicBlock::const_iterator I = MBB->end(), E = MBB->begin();
I != E;) {
--I;
- if (I->getDesc().isCall()) {
+ if (I->isCall()) {
LSP.second = LIS.getInstructionIndex(I);
break;
}
diff --git a/lib/CodeGen/Splitter.cpp b/lib/CodeGen/Splitter.cpp
deleted file mode 100644
index 16cf9b8..0000000
--- a/lib/CodeGen/Splitter.cpp
+++ /dev/null
@@ -1,827 +0,0 @@
-//===-- llvm/CodeGen/Splitter.cpp - Splitter -----------------------------===//
-//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "loopsplitter"
-
-#include "Splitter.h"
-
-#include "llvm/Module.h"
-#include "llvm/CodeGen/CalcSpillWeights.h"
-#include "llvm/CodeGen/LiveIntervalAnalysis.h"
-#include "llvm/CodeGen/LiveStackAnalysis.h"
-#include "llvm/CodeGen/MachineDominators.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/SlotIndexes.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetInstrInfo.h"
-
-using namespace llvm;
-
-char LoopSplitter::ID = 0;
-INITIALIZE_PASS_BEGIN(LoopSplitter, "loop-splitting",
- "Split virtual regists across loop boundaries.", false, false)
-INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
-INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
-INITIALIZE_PASS_DEPENDENCY(SlotIndexes)
-INITIALIZE_PASS_DEPENDENCY(LiveIntervals)
-INITIALIZE_PASS_END(LoopSplitter, "loop-splitting",
- "Split virtual regists across loop boundaries.", false, false)
-
-namespace llvm {
-
- class StartSlotComparator {
- public:
- StartSlotComparator(LiveIntervals &lis) : lis(lis) {}
- bool operator()(const MachineBasicBlock *mbb1,
- const MachineBasicBlock *mbb2) const {
- return lis.getMBBStartIdx(mbb1) < lis.getMBBStartIdx(mbb2);
- }
- private:
- LiveIntervals &lis;
- };
-
- class LoopSplit {
- public:
- LoopSplit(LoopSplitter &ls, LiveInterval &li, MachineLoop &loop)
- : ls(ls), li(li), loop(loop), valid(true), inSplit(false), newLI(0) {
- assert(TargetRegisterInfo::isVirtualRegister(li.reg) &&
- "Cannot split physical registers.");
- }
-
- LiveInterval& getLI() const { return li; }
-
- MachineLoop& getLoop() const { return loop; }
-
- bool isValid() const { return valid; }
-
- bool isWorthwhile() const { return valid && (inSplit || !outSplits.empty()); }
-
- void invalidate() { valid = false; }
-
- void splitIncoming() { inSplit = true; }
-
- void splitOutgoing(MachineLoop::Edge &edge) { outSplits.insert(edge); }
-
- void addLoopInstr(MachineInstr *i) { loopInstrs.push_back(i); }
-
- void apply() {
- assert(valid && "Attempt to apply invalid split.");
- applyIncoming();
- applyOutgoing();
- copyRanges();
- renameInside();
- }
-
- private:
- LoopSplitter &ls;
- LiveInterval &li;
- MachineLoop &loop;
- bool valid, inSplit;
- std::set<MachineLoop::Edge> outSplits;
- std::vector<MachineInstr*> loopInstrs;
-
- LiveInterval *newLI;
- std::map<VNInfo*, VNInfo*> vniMap;
-
- LiveInterval* getNewLI() {
- if (newLI == 0) {
- const TargetRegisterClass *trc = ls.mri->getRegClass(li.reg);
- unsigned vreg = ls.mri->createVirtualRegister(trc);
- newLI = &ls.lis->getOrCreateInterval(vreg);
- }
- return newLI;
- }
-
- VNInfo* getNewVNI(VNInfo *oldVNI) {
- VNInfo *newVNI = vniMap[oldVNI];
-
- if (newVNI == 0) {
- newVNI = getNewLI()->createValueCopy(oldVNI,
- ls.lis->getVNInfoAllocator());
- vniMap[oldVNI] = newVNI;
- }
-
- return newVNI;
- }
-
- void applyIncoming() {
- if (!inSplit) {
- return;
- }
-
- MachineBasicBlock *preHeader = loop.getLoopPreheader();
- if (preHeader == 0) {
- assert(ls.canInsertPreHeader(loop) &&
- "Can't insert required preheader.");
- preHeader = &ls.insertPreHeader(loop);
- }
-
- LiveRange *preHeaderRange =
- ls.lis->findExitingRange(li, preHeader);
- assert(preHeaderRange != 0 && "Range not live into preheader.");
-
- // Insert the new copy.
- MachineInstr *copy = BuildMI(*preHeader,
- preHeader->getFirstTerminator(),
- DebugLoc(),
- ls.tii->get(TargetOpcode::COPY))
- .addReg(getNewLI()->reg, RegState::Define)
- .addReg(li.reg, RegState::Kill);
-
- ls.lis->InsertMachineInstrInMaps(copy);
-
- SlotIndex copyDefIdx = ls.lis->getInstructionIndex(copy).getRegSlot();
-
- VNInfo *newVal = getNewVNI(preHeaderRange->valno);
- newVal->def = copyDefIdx;
- newVal->setCopy(copy);
- li.removeRange(copyDefIdx, ls.lis->getMBBEndIdx(preHeader), true);
-
- getNewLI()->addRange(LiveRange(copyDefIdx,
- ls.lis->getMBBEndIdx(preHeader),
- newVal));
- }
-
- void applyOutgoing() {
-
- for (std::set<MachineLoop::Edge>::iterator osItr = outSplits.begin(),
- osEnd = outSplits.end();
- osItr != osEnd; ++osItr) {
- MachineLoop::Edge edge = *osItr;
- MachineBasicBlock *outBlock = edge.second;
- if (ls.isCriticalEdge(edge)) {
- assert(ls.canSplitEdge(edge) && "Unsplitable critical edge.");
- outBlock = &ls.splitEdge(edge, loop);
- }
- LiveRange *outRange = ls.lis->findEnteringRange(li, outBlock);
- assert(outRange != 0 && "No exiting range?");
-
- MachineInstr *copy = BuildMI(*outBlock, outBlock->begin(),
- DebugLoc(),
- ls.tii->get(TargetOpcode::COPY))
- .addReg(li.reg, RegState::Define)
- .addReg(getNewLI()->reg, RegState::Kill);
-
- ls.lis->InsertMachineInstrInMaps(copy);
-
- SlotIndex copyDefIdx = ls.lis->getInstructionIndex(copy).getRegSlot();
-
- // Blow away output range definition.
- outRange->valno->def = ls.lis->getInvalidIndex();
- li.removeRange(ls.lis->getMBBStartIdx(outBlock), copyDefIdx);
-
- SlotIndex newDefIdx = ls.lis->getMBBStartIdx(outBlock);
- assert(ls.lis->getInstructionFromIndex(newDefIdx) == 0 &&
- "PHI def index points at actual instruction.");
- VNInfo *newVal =
- getNewLI()->getNextValue(newDefIdx, 0, ls.lis->getVNInfoAllocator());
-
- getNewLI()->addRange(LiveRange(ls.lis->getMBBStartIdx(outBlock),
- copyDefIdx, newVal));
-
- }
- }
-
- void copyRange(LiveRange &lr) {
- std::pair<bool, LoopSplitter::SlotPair> lsr =
- ls.getLoopSubRange(lr, loop);
-
- if (!lsr.first)
- return;
-
- LiveRange loopRange(lsr.second.first, lsr.second.second,
- getNewVNI(lr.valno));
-
- li.removeRange(loopRange.start, loopRange.end, true);
-
- getNewLI()->addRange(loopRange);
- }
-
- void copyRanges() {
- for (std::vector<MachineInstr*>::iterator iItr = loopInstrs.begin(),
- iEnd = loopInstrs.end();
- iItr != iEnd; ++iItr) {
- MachineInstr &instr = **iItr;
- SlotIndex instrIdx = ls.lis->getInstructionIndex(&instr);
- if (instr.modifiesRegister(li.reg, 0)) {
- LiveRange *defRange =
- li.getLiveRangeContaining(instrIdx.getRegSlot());
- if (defRange != 0) // May have caught this already.
- copyRange(*defRange);
- }
- if (instr.readsRegister(li.reg, 0)) {
- LiveRange *useRange =
- li.getLiveRangeContaining(instrIdx.getRegSlot(true));
- if (useRange != 0) { // May have caught this already.
- copyRange(*useRange);
- }
- }
- }
-
- for (MachineLoop::block_iterator bbItr = loop.block_begin(),
- bbEnd = loop.block_end();
- bbItr != bbEnd; ++bbItr) {
- MachineBasicBlock &loopBlock = **bbItr;
- LiveRange *enteringRange =
- ls.lis->findEnteringRange(li, &loopBlock);
- if (enteringRange != 0) {
- copyRange(*enteringRange);
- }
- }
- }
-
- void renameInside() {
- for (std::vector<MachineInstr*>::iterator iItr = loopInstrs.begin(),
- iEnd = loopInstrs.end();
- iItr != iEnd; ++iItr) {
- MachineInstr &instr = **iItr;
- for (unsigned i = 0; i < instr.getNumOperands(); ++i) {
- MachineOperand &mop = instr.getOperand(i);
- if (mop.isReg() && mop.getReg() == li.reg) {
- mop.setReg(getNewLI()->reg);
- }
- }
- }
- }
-
- };
-
- void LoopSplitter::getAnalysisUsage(AnalysisUsage &au) const {
- au.addRequired<MachineDominatorTree>();
- au.addPreserved<MachineDominatorTree>();
- au.addRequired<MachineLoopInfo>();
- au.addPreserved<MachineLoopInfo>();
- au.addPreservedID(RegisterCoalescerPassID);
- au.addPreserved<CalculateSpillWeights>();
- au.addPreserved<LiveStacks>();
- au.addRequired<SlotIndexes>();
- au.addPreserved<SlotIndexes>();
- au.addRequired<LiveIntervals>();
- au.addPreserved<LiveIntervals>();
- MachineFunctionPass::getAnalysisUsage(au);
- }
-
- bool LoopSplitter::runOnMachineFunction(MachineFunction &fn) {
-
- mf = &fn;
- mri = &mf->getRegInfo();
- tii = mf->getTarget().getInstrInfo();
- tri = mf->getTarget().getRegisterInfo();
- sis = &getAnalysis<SlotIndexes>();
- lis = &getAnalysis<LiveIntervals>();
- mli = &getAnalysis<MachineLoopInfo>();
- mdt = &getAnalysis<MachineDominatorTree>();
-
- fqn = mf->getFunction()->getParent()->getModuleIdentifier() + "." +
- mf->getFunction()->getName().str();
-
- dbgs() << "Splitting " << mf->getFunction()->getName() << ".";
-
- dumpOddTerminators();
-
-// dbgs() << "----------------------------------------\n";
-// lis->dump();
-// dbgs() << "----------------------------------------\n";
-
-// std::deque<MachineLoop*> loops;
-// std::copy(mli->begin(), mli->end(), std::back_inserter(loops));
-// dbgs() << "Loops:\n";
-// while (!loops.empty()) {
-// MachineLoop &loop = *loops.front();
-// loops.pop_front();
-// std::copy(loop.begin(), loop.end(), std::back_inserter(loops));
-
-// dumpLoopInfo(loop);
-// }
-
- //lis->dump();
- //exit(0);
-
- // Setup initial intervals.
- for (LiveIntervals::iterator liItr = lis->begin(), liEnd = lis->end();
- liItr != liEnd; ++liItr) {
- LiveInterval *li = liItr->second;
-
- if (TargetRegisterInfo::isVirtualRegister(li->reg) &&
- !lis->intervalIsInOneMBB(*li)) {
- intervals.push_back(li);
- }
- }
-
- processIntervals();
-
- intervals.clear();
-
-// dbgs() << "----------------------------------------\n";
-// lis->dump();
-// dbgs() << "----------------------------------------\n";
-
- dumpOddTerminators();
-
- //exit(1);
-
- return false;
- }
-
- void LoopSplitter::releaseMemory() {
- fqn.clear();
- intervals.clear();
- loopRangeMap.clear();
- }
-
- void LoopSplitter::dumpOddTerminators() {
- for (MachineFunction::iterator bbItr = mf->begin(), bbEnd = mf->end();
- bbItr != bbEnd; ++bbItr) {
- MachineBasicBlock *mbb = &*bbItr;
- MachineBasicBlock *a = 0, *b = 0;
- SmallVector<MachineOperand, 4> c;
- if (tii->AnalyzeBranch(*mbb, a, b, c)) {
- dbgs() << "MBB#" << mbb->getNumber() << " has multiway terminator.\n";
- dbgs() << " Terminators:\n";
- for (MachineBasicBlock::iterator iItr = mbb->begin(), iEnd = mbb->end();
- iItr != iEnd; ++iItr) {
- MachineInstr *instr= &*iItr;
- dbgs() << " " << *instr << "";
- }
- dbgs() << "\n Listed successors: [ ";
- for (MachineBasicBlock::succ_iterator sItr = mbb->succ_begin(), sEnd = mbb->succ_end();
- sItr != sEnd; ++sItr) {
- MachineBasicBlock *succMBB = *sItr;
- dbgs() << succMBB->getNumber() << " ";
- }
- dbgs() << "]\n\n";
- }
- }
- }
-
- void LoopSplitter::dumpLoopInfo(MachineLoop &loop) {
- MachineBasicBlock &headerBlock = *loop.getHeader();
- typedef SmallVector<MachineLoop::Edge, 8> ExitEdgesList;
- ExitEdgesList exitEdges;
- loop.getExitEdges(exitEdges);
-
- dbgs() << " Header: BB#" << headerBlock.getNumber() << ", Contains: [ ";
- for (std::vector<MachineBasicBlock*>::const_iterator
- subBlockItr = loop.getBlocks().begin(),
- subBlockEnd = loop.getBlocks().end();
- subBlockItr != subBlockEnd; ++subBlockItr) {
- MachineBasicBlock &subBlock = **subBlockItr;
- dbgs() << "BB#" << subBlock.getNumber() << " ";
- }
- dbgs() << "], Exit edges: [ ";
- for (ExitEdgesList::iterator exitEdgeItr = exitEdges.begin(),
- exitEdgeEnd = exitEdges.end();
- exitEdgeItr != exitEdgeEnd; ++exitEdgeItr) {
- MachineLoop::Edge &exitEdge = *exitEdgeItr;
- dbgs() << "(MBB#" << exitEdge.first->getNumber()
- << ", MBB#" << exitEdge.second->getNumber() << ") ";
- }
- dbgs() << "], Sub-Loop Headers: [ ";
- for (MachineLoop::iterator subLoopItr = loop.begin(),
- subLoopEnd = loop.end();
- subLoopItr != subLoopEnd; ++subLoopItr) {
- MachineLoop &subLoop = **subLoopItr;
- MachineBasicBlock &subLoopBlock = *subLoop.getHeader();
- dbgs() << "BB#" << subLoopBlock.getNumber() << " ";
- }
- dbgs() << "]\n";
- }
-
- void LoopSplitter::updateTerminators(MachineBasicBlock &mbb) {
- mbb.updateTerminator();
-
- for (MachineBasicBlock::iterator miItr = mbb.begin(), miEnd = mbb.end();
- miItr != miEnd; ++miItr) {
- if (lis->isNotInMIMap(miItr)) {
- lis->InsertMachineInstrInMaps(miItr);
- }
- }
- }
-
- bool LoopSplitter::canInsertPreHeader(MachineLoop &loop) {
- MachineBasicBlock *header = loop.getHeader();
- MachineBasicBlock *a = 0, *b = 0;
- SmallVector<MachineOperand, 4> c;
-
- for (MachineBasicBlock::pred_iterator pbItr = header->pred_begin(),
- pbEnd = header->pred_end();
- pbItr != pbEnd; ++pbItr) {
- MachineBasicBlock *predBlock = *pbItr;
- if (!!tii->AnalyzeBranch(*predBlock, a, b, c)) {
- return false;
- }
- }
-
- MachineFunction::iterator headerItr(header);
- if (headerItr == mf->begin())
- return true;
- MachineBasicBlock *headerLayoutPred = llvm::prior(headerItr);
- assert(headerLayoutPred != 0 && "Header should have layout pred.");
-
- return (!tii->AnalyzeBranch(*headerLayoutPred, a, b, c));
- }
-
- MachineBasicBlock& LoopSplitter::insertPreHeader(MachineLoop &loop) {
- assert(loop.getLoopPreheader() == 0 && "Loop already has preheader.");
-
- MachineBasicBlock &header = *loop.getHeader();
-
- // Save the preds - we'll need to update them once we insert the preheader.
- typedef std::set<MachineBasicBlock*> HeaderPreds;
- HeaderPreds headerPreds;
-
- for (MachineBasicBlock::pred_iterator predItr = header.pred_begin(),
- predEnd = header.pred_end();
- predItr != predEnd; ++predItr) {
- if (!loop.contains(*predItr))
- headerPreds.insert(*predItr);
- }
-
- assert(!headerPreds.empty() && "No predecessors for header?");
-
- //dbgs() << fqn << " MBB#" << header.getNumber() << " inserting preheader...";
-
- MachineBasicBlock *preHeader =
- mf->CreateMachineBasicBlock(header.getBasicBlock());
-
- assert(preHeader != 0 && "Failed to create pre-header.");
-
- mf->insert(header, preHeader);
-
- for (HeaderPreds::iterator hpItr = headerPreds.begin(),
- hpEnd = headerPreds.end();
- hpItr != hpEnd; ++hpItr) {
- assert(*hpItr != 0 && "How'd a null predecessor get into this set?");
- MachineBasicBlock &hp = **hpItr;
- hp.ReplaceUsesOfBlockWith(&header, preHeader);
- }
- preHeader->addSuccessor(&header);
-
- MachineBasicBlock *oldLayoutPred =
- llvm::prior(MachineFunction::iterator(preHeader));
- if (oldLayoutPred != 0) {
- updateTerminators(*oldLayoutPred);
- }
-
- lis->InsertMBBInMaps(preHeader);
-
- if (MachineLoop *parentLoop = loop.getParentLoop()) {
- assert(parentLoop->getHeader() != loop.getHeader() &&
- "Parent loop has same header?");
- parentLoop->addBasicBlockToLoop(preHeader, mli->getBase());
-
- // Invalidate all parent loop ranges.
- while (parentLoop != 0) {
- loopRangeMap.erase(parentLoop);
- parentLoop = parentLoop->getParentLoop();
- }
- }
-
- for (LiveIntervals::iterator liItr = lis->begin(),
- liEnd = lis->end();
- liItr != liEnd; ++liItr) {
- LiveInterval &li = *liItr->second;
-
- // Is this safe for physregs?
- // TargetRegisterInfo::isPhysicalRegister(li.reg) ||
- if (!lis->isLiveInToMBB(li, &header))
- continue;
-
- if (lis->isLiveInToMBB(li, preHeader)) {
- assert(lis->isLiveOutOfMBB(li, preHeader) &&
- "Range terminates in newly added preheader?");
- continue;
- }
-
- bool insertRange = false;
-
- for (MachineBasicBlock::pred_iterator predItr = preHeader->pred_begin(),
- predEnd = preHeader->pred_end();
- predItr != predEnd; ++predItr) {
- MachineBasicBlock *predMBB = *predItr;
- if (lis->isLiveOutOfMBB(li, predMBB)) {
- insertRange = true;
- break;
- }
- }
-
- if (!insertRange)
- continue;
-
- SlotIndex newDefIdx = lis->getMBBStartIdx(preHeader);
- assert(lis->getInstructionFromIndex(newDefIdx) == 0 &&
- "PHI def index points at actual instruction.");
- VNInfo *newVal = li.getNextValue(newDefIdx, 0, lis->getVNInfoAllocator());
- li.addRange(LiveRange(lis->getMBBStartIdx(preHeader),
- lis->getMBBEndIdx(preHeader),
- newVal));
- }
-
-
- //dbgs() << "Dumping SlotIndexes:\n";
- //sis->dump();
-
- //dbgs() << "done. (Added MBB#" << preHeader->getNumber() << ")\n";
-
- return *preHeader;
- }
-
- bool LoopSplitter::isCriticalEdge(MachineLoop::Edge &edge) {
- assert(edge.first->succ_size() > 1 && "Non-sensical edge.");
- if (edge.second->pred_size() > 1)
- return true;
- return false;
- }
-
- bool LoopSplitter::canSplitEdge(MachineLoop::Edge &edge) {
- MachineFunction::iterator outBlockItr(edge.second);
- if (outBlockItr == mf->begin())
- return true;
- MachineBasicBlock *outBlockLayoutPred = llvm::prior(outBlockItr);
- assert(outBlockLayoutPred != 0 && "Should have a layout pred if out!=begin.");
- MachineBasicBlock *a = 0, *b = 0;
- SmallVector<MachineOperand, 4> c;
- return (!tii->AnalyzeBranch(*outBlockLayoutPred, a, b, c) &&
- !tii->AnalyzeBranch(*edge.first, a, b, c));
- }
-
- MachineBasicBlock& LoopSplitter::splitEdge(MachineLoop::Edge &edge,
- MachineLoop &loop) {
-
- MachineBasicBlock &inBlock = *edge.first;
- MachineBasicBlock &outBlock = *edge.second;
-
- assert((inBlock.succ_size() > 1) && (outBlock.pred_size() > 1) &&
- "Splitting non-critical edge?");
-
- //dbgs() << fqn << " Splitting edge (MBB#" << inBlock.getNumber()
- // << " -> MBB#" << outBlock.getNumber() << ")...";
-
- MachineBasicBlock *splitBlock =
- mf->CreateMachineBasicBlock();
-
- assert(splitBlock != 0 && "Failed to create split block.");
-
- mf->insert(&outBlock, splitBlock);
-
- inBlock.ReplaceUsesOfBlockWith(&outBlock, splitBlock);
- splitBlock->addSuccessor(&outBlock);
-
- MachineBasicBlock *oldLayoutPred =
- llvm::prior(MachineFunction::iterator(splitBlock));
- if (oldLayoutPred != 0) {
- updateTerminators(*oldLayoutPred);
- }
-
- lis->InsertMBBInMaps(splitBlock);
-
- loopRangeMap.erase(&loop);
-
- MachineLoop *splitParentLoop = loop.getParentLoop();
- while (splitParentLoop != 0 &&
- !splitParentLoop->contains(&outBlock)) {
- splitParentLoop = splitParentLoop->getParentLoop();
- }
-
- if (splitParentLoop != 0) {
- assert(splitParentLoop->contains(&loop) &&
- "Split-block parent doesn't contain original loop?");
- splitParentLoop->addBasicBlockToLoop(splitBlock, mli->getBase());
-
- // Invalidate all parent loop ranges.
- while (splitParentLoop != 0) {
- loopRangeMap.erase(splitParentLoop);
- splitParentLoop = splitParentLoop->getParentLoop();
- }
- }
-
-
- for (LiveIntervals::iterator liItr = lis->begin(),
- liEnd = lis->end();
- liItr != liEnd; ++liItr) {
- LiveInterval &li = *liItr->second;
- bool intersects = lis->isLiveOutOfMBB(li, &inBlock) &&
- lis->isLiveInToMBB(li, &outBlock);
- if (lis->isLiveInToMBB(li, splitBlock)) {
- if (!intersects) {
- li.removeRange(lis->getMBBStartIdx(splitBlock),
- lis->getMBBEndIdx(splitBlock), true);
- }
- } else if (intersects) {
- SlotIndex newDefIdx = lis->getMBBStartIdx(splitBlock);
- assert(lis->getInstructionFromIndex(newDefIdx) == 0 &&
- "PHI def index points at actual instruction.");
- VNInfo *newVal = li.getNextValue(newDefIdx, 0,
- lis->getVNInfoAllocator());
- li.addRange(LiveRange(lis->getMBBStartIdx(splitBlock),
- lis->getMBBEndIdx(splitBlock),
- newVal));
- }
- }
-
- //dbgs() << "done. (Added MBB#" << splitBlock->getNumber() << ")\n";
-
- return *splitBlock;
- }
-
- LoopSplitter::LoopRanges& LoopSplitter::getLoopRanges(MachineLoop &loop) {
- typedef std::set<MachineBasicBlock*, StartSlotComparator> LoopMBBSet;
- LoopRangeMap::iterator lrItr = loopRangeMap.find(&loop);
- if (lrItr == loopRangeMap.end()) {
- LoopMBBSet loopMBBs((StartSlotComparator(*lis)));
- std::copy(loop.block_begin(), loop.block_end(),
- std::inserter(loopMBBs, loopMBBs.begin()));
-
- assert(!loopMBBs.empty() && "No blocks in loop?");
-
- LoopRanges &loopRanges = loopRangeMap[&loop];
- assert(loopRanges.empty() && "Loop encountered but not processed?");
- SlotIndex oldEnd = lis->getMBBEndIdx(*loopMBBs.begin());
- loopRanges.push_back(
- std::make_pair(lis->getMBBStartIdx(*loopMBBs.begin()),
- lis->getInvalidIndex()));
- for (LoopMBBSet::iterator curBlockItr = llvm::next(loopMBBs.begin()),
- curBlockEnd = loopMBBs.end();
- curBlockItr != curBlockEnd; ++curBlockItr) {
- SlotIndex newStart = lis->getMBBStartIdx(*curBlockItr);
- if (newStart != oldEnd) {
- loopRanges.back().second = oldEnd;
- loopRanges.push_back(std::make_pair(newStart,
- lis->getInvalidIndex()));
- }
- oldEnd = lis->getMBBEndIdx(*curBlockItr);
- }
-
- loopRanges.back().second =
- lis->getMBBEndIdx(*llvm::prior(loopMBBs.end()));
-
- return loopRanges;
- }
- return lrItr->second;
- }
-
- std::pair<bool, LoopSplitter::SlotPair> LoopSplitter::getLoopSubRange(
- const LiveRange &lr,
- MachineLoop &loop) {
- LoopRanges &loopRanges = getLoopRanges(loop);
- LoopRanges::iterator lrItr = loopRanges.begin(),
- lrEnd = loopRanges.end();
- while (lrItr != lrEnd && lr.start >= lrItr->second) {
- ++lrItr;
- }
-
- if (lrItr == lrEnd) {
- SlotIndex invalid = lis->getInvalidIndex();
- return std::make_pair(false, SlotPair(invalid, invalid));
- }
-
- SlotIndex srStart(lr.start < lrItr->first ? lrItr->first : lr.start);
- SlotIndex srEnd(lr.end > lrItr->second ? lrItr->second : lr.end);
-
- return std::make_pair(true, SlotPair(srStart, srEnd));
- }
-
- void LoopSplitter::dumpLoopRanges(MachineLoop &loop) {
- LoopRanges &loopRanges = getLoopRanges(loop);
- dbgs() << "For loop MBB#" << loop.getHeader()->getNumber() << ", subranges are: [ ";
- for (LoopRanges::iterator lrItr = loopRanges.begin(), lrEnd = loopRanges.end();
- lrItr != lrEnd; ++lrItr) {
- dbgs() << "[" << lrItr->first << ", " << lrItr->second << ") ";
- }
- dbgs() << "]\n";
- }
-
- void LoopSplitter::processHeader(LoopSplit &split) {
- MachineBasicBlock &header = *split.getLoop().getHeader();
- //dbgs() << " Processing loop header BB#" << header.getNumber() << "\n";
-
- if (!lis->isLiveInToMBB(split.getLI(), &header))
- return; // Not live in, but nothing wrong so far.
-
- MachineBasicBlock *preHeader = split.getLoop().getLoopPreheader();
- if (!preHeader) {
-
- if (!canInsertPreHeader(split.getLoop())) {
- split.invalidate();
- return; // Couldn't insert a pre-header. Bail on this interval.
- }
-
- for (MachineBasicBlock::pred_iterator predItr = header.pred_begin(),
- predEnd = header.pred_end();
- predItr != predEnd; ++predItr) {
- if (lis->isLiveOutOfMBB(split.getLI(), *predItr)) {
- split.splitIncoming();
- break;
- }
- }
- } else if (lis->isLiveOutOfMBB(split.getLI(), preHeader)) {
- split.splitIncoming();
- }
- }
-
- void LoopSplitter::processLoopExits(LoopSplit &split) {
- typedef SmallVector<MachineLoop::Edge, 8> ExitEdgesList;
- ExitEdgesList exitEdges;
- split.getLoop().getExitEdges(exitEdges);
-
- //dbgs() << " Processing loop exits:\n";
-
- for (ExitEdgesList::iterator exitEdgeItr = exitEdges.begin(),
- exitEdgeEnd = exitEdges.end();
- exitEdgeItr != exitEdgeEnd; ++exitEdgeItr) {
- MachineLoop::Edge exitEdge = *exitEdgeItr;
-
- LiveRange *outRange =
- split.getLI().getLiveRangeContaining(lis->getMBBStartIdx(exitEdge.second));
-
- if (outRange != 0) {
- if (isCriticalEdge(exitEdge) && !canSplitEdge(exitEdge)) {
- split.invalidate();
- return;
- }
-
- split.splitOutgoing(exitEdge);
- }
- }
- }
-
- void LoopSplitter::processLoopUses(LoopSplit &split) {
- std::set<MachineInstr*> processed;
-
- for (MachineRegisterInfo::reg_iterator
- rItr = mri->reg_begin(split.getLI().reg),
- rEnd = mri->reg_end();
- rItr != rEnd; ++rItr) {
- MachineInstr &instr = *rItr;
- if (split.getLoop().contains(&instr) && processed.count(&instr) == 0) {
- split.addLoopInstr(&instr);
- processed.insert(&instr);
- }
- }
-
- //dbgs() << " Rewriting reg" << li.reg << " to reg" << newLI->reg
- // << " in blocks [ ";
- //dbgs() << "]\n";
- }
-
- bool LoopSplitter::splitOverLoop(LiveInterval &li, MachineLoop &loop) {
- assert(TargetRegisterInfo::isVirtualRegister(li.reg) &&
- "Attempt to split physical register.");
-
- LoopSplit split(*this, li, loop);
- processHeader(split);
- if (split.isValid())
- processLoopExits(split);
- if (split.isValid())
- processLoopUses(split);
- if (split.isValid() /* && split.isWorthwhile() */) {
- split.apply();
- DEBUG(dbgs() << "Success.\n");
- return true;
- }
- DEBUG(dbgs() << "Failed.\n");
- return false;
- }
-
- void LoopSplitter::processInterval(LiveInterval &li) {
- std::deque<MachineLoop*> loops;
- std::copy(mli->begin(), mli->end(), std::back_inserter(loops));
-
- while (!loops.empty()) {
- MachineLoop &loop = *loops.front();
- loops.pop_front();
- DEBUG(
- dbgs() << fqn << " reg" << li.reg << " " << li.weight << " BB#"
- << loop.getHeader()->getNumber() << " ";
- );
- if (!splitOverLoop(li, loop)) {
- // Couldn't split over outer loop, schedule sub-loops to be checked.
- std::copy(loop.begin(), loop.end(), std::back_inserter(loops));
- }
- }
- }
-
- void LoopSplitter::processIntervals() {
- while (!intervals.empty()) {
- LiveInterval &li = *intervals.front();
- intervals.pop_front();
-
- assert(!lis->intervalIsInOneMBB(li) &&
- "Single interval in process worklist.");
-
- processInterval(li);
- }
- }
-
-}
diff --git a/lib/CodeGen/Splitter.h b/lib/CodeGen/Splitter.h
deleted file mode 100644
index 9fb1b8b..0000000
--- a/lib/CodeGen/Splitter.h
+++ /dev/null
@@ -1,101 +0,0 @@
-//===-- llvm/CodeGen/Splitter.h - Splitter -*- C++ -*----------------------===//
-//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_CODEGEN_SPLITTER_H
-#define LLVM_CODEGEN_SPLITTER_H
-
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineLoopInfo.h"
-#include "llvm/CodeGen/SlotIndexes.h"
-
-#include <deque>
-#include <map>
-#include <string>
-#include <vector>
-
-namespace llvm {
-
- class LiveInterval;
- class LiveIntervals;
- struct LiveRange;
- class LoopSplit;
- class MachineDominatorTree;
- class MachineRegisterInfo;
- class SlotIndexes;
- class TargetInstrInfo;
- class VNInfo;
-
- class LoopSplitter : public MachineFunctionPass {
- friend class LoopSplit;
- public:
- static char ID;
-
- LoopSplitter() : MachineFunctionPass(ID) {
- initializeLoopSplitterPass(*PassRegistry::getPassRegistry());
- }
-
- virtual void getAnalysisUsage(AnalysisUsage &au) const;
-
- virtual bool runOnMachineFunction(MachineFunction &fn);
-
- virtual void releaseMemory();
-
-
- private:
-
- MachineFunction *mf;
- LiveIntervals *lis;
- MachineLoopInfo *mli;
- MachineRegisterInfo *mri;
- MachineDominatorTree *mdt;
- SlotIndexes *sis;
- const TargetInstrInfo *tii;
- const TargetRegisterInfo *tri;
-
- std::string fqn;
- std::deque<LiveInterval*> intervals;
-
- typedef std::pair<SlotIndex, SlotIndex> SlotPair;
- typedef std::vector<SlotPair> LoopRanges;
- typedef std::map<MachineLoop*, LoopRanges> LoopRangeMap;
- LoopRangeMap loopRangeMap;
-
- void dumpLoopInfo(MachineLoop &loop);
-
- void dumpOddTerminators();
-
- void updateTerminators(MachineBasicBlock &mbb);
-
- bool canInsertPreHeader(MachineLoop &loop);
- MachineBasicBlock& insertPreHeader(MachineLoop &loop);
-
- bool isCriticalEdge(MachineLoop::Edge &edge);
- bool canSplitEdge(MachineLoop::Edge &edge);
- MachineBasicBlock& splitEdge(MachineLoop::Edge &edge, MachineLoop &loop);
-
- LoopRanges& getLoopRanges(MachineLoop &loop);
- std::pair<bool, SlotPair> getLoopSubRange(const LiveRange &lr,
- MachineLoop &loop);
-
- void dumpLoopRanges(MachineLoop &loop);
-
- void processHeader(LoopSplit &split);
- void processLoopExits(LoopSplit &split);
- void processLoopUses(LoopSplit &split);
-
- bool splitOverLoop(LiveInterval &li, MachineLoop &loop);
-
- void processInterval(LiveInterval &li);
-
- void processIntervals();
- };
-
-}
-
-#endif
diff --git a/lib/CodeGen/TailDuplication.cpp b/lib/CodeGen/TailDuplication.cpp
index 3a6211a..031377b 100644
--- a/lib/CodeGen/TailDuplication.cpp
+++ b/lib/CodeGen/TailDuplication.cpp
@@ -553,7 +553,7 @@ TailDuplicatePass::shouldTailDuplicate(const MachineFunction &MF,
bool HasIndirectbr = false;
if (!TailBB.empty())
- HasIndirectbr = TailBB.back().getDesc().isIndirectBranch();
+ HasIndirectbr = TailBB.back().isIndirectBranch();
if (HasIndirectbr && PreRegAlloc)
MaxDuplicateCount = 20;
@@ -561,22 +561,21 @@ TailDuplicatePass::shouldTailDuplicate(const MachineFunction &MF,
// Check the instructions in the block to determine whether tail-duplication
// is invalid or unlikely to be profitable.
unsigned InstrCount = 0;
- for (MachineBasicBlock::const_iterator I = TailBB.begin(); I != TailBB.end();
- ++I) {
+ for (MachineBasicBlock::iterator I = TailBB.begin(); I != TailBB.end(); ++I) {
// Non-duplicable things shouldn't be tail-duplicated.
- if (I->getDesc().isNotDuplicable())
+ if (I->isNotDuplicable())
return false;
// Do not duplicate 'return' instructions if this is a pre-regalloc run.
// A return may expand into a lot more instructions (e.g. reload of callee
// saved registers) after PEI.
- if (PreRegAlloc && I->getDesc().isReturn())
+ if (PreRegAlloc && I->isReturn())
return false;
// Avoid duplicating calls before register allocation. Calls presents a
// barrier to register allocation so duplicating them may end up increasing
// spills.
- if (PreRegAlloc && I->getDesc().isCall())
+ if (PreRegAlloc && I->isCall())
return false;
if (!I->isPHI() && !I->isDebugValue())
@@ -611,7 +610,7 @@ TailDuplicatePass::isSimpleBB(MachineBasicBlock *TailBB) {
++I;
if (I == E)
return true;
- return I->getDesc().isUnconditionalBranch();
+ return I->isUnconditionalBranch();
}
static bool
diff --git a/lib/CodeGen/TargetFrameLoweringImpl.cpp b/lib/CodeGen/TargetFrameLoweringImpl.cpp
new file mode 100644
index 0000000..cadb878
--- /dev/null
+++ b/lib/CodeGen/TargetFrameLoweringImpl.cpp
@@ -0,0 +1,45 @@
+//===----- TargetFrameLoweringImpl.cpp - Implement target frame interface --==//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Implements the layout of a stack frame on the target machine.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+
+#include <cstdlib>
+using namespace llvm;
+
+TargetFrameLowering::~TargetFrameLowering() {
+}
+
+/// getFrameIndexOffset - Returns the displacement from the frame register to
+/// the stack frame of the specified index. This is the default implementation
+/// which is overridden for some targets.
+int TargetFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
+ int FI) const {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ return MFI->getObjectOffset(FI) + MFI->getStackSize() -
+ getOffsetOfLocalArea() + MFI->getOffsetAdjustment();
+}
+
+int TargetFrameLowering::getFrameIndexReference(const MachineFunction &MF,
+ int FI, unsigned &FrameReg) const {
+ const TargetRegisterInfo *RI = MF.getTarget().getRegisterInfo();
+
+ // By default, assume all frame indices are referenced via whatever
+ // getFrameRegister() says. The target can override this if it's doing
+ // something different.
+ FrameReg = RI->getFrameRegister(MF);
+ return getFrameIndexOffset(MF, FI);
+}
diff --git a/lib/CodeGen/TargetInstrInfoImpl.cpp b/lib/CodeGen/TargetInstrInfoImpl.cpp
index 08e2b16..7ed9455 100644
--- a/lib/CodeGen/TargetInstrInfoImpl.cpp
+++ b/lib/CodeGen/TargetInstrInfoImpl.cpp
@@ -24,6 +24,7 @@
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/ScoreboardHazardRecognizer.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/MC/MCInstrItineraries.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
@@ -121,6 +122,9 @@ MachineInstr *TargetInstrInfoImpl::commuteInstruction(MachineInstr *MI,
bool TargetInstrInfoImpl::findCommutedOpIndices(MachineInstr *MI,
unsigned &SrcOpIdx1,
unsigned &SrcOpIdx2) const {
+ assert(!MI->isBundle() &&
+ "TargetInstrInfoImpl::findCommutedOpIndices() can't handle bundles");
+
const MCInstrDesc &MCID = MI->getDesc();
if (!MCID.isCommutable())
return false;
@@ -136,11 +140,28 @@ bool TargetInstrInfoImpl::findCommutedOpIndices(MachineInstr *MI,
}
+bool
+TargetInstrInfoImpl::isUnpredicatedTerminator(const MachineInstr *MI) const {
+ if (!MI->isTerminator()) return false;
+
+ // Conditional branch is a special case.
+ if (MI->isBranch() && !MI->isBarrier())
+ return true;
+ if (!MI->isPredicable())
+ return true;
+ return !isPredicated(MI);
+}
+
+
bool TargetInstrInfoImpl::PredicateInstruction(MachineInstr *MI,
const SmallVectorImpl<MachineOperand> &Pred) const {
bool MadeChange = false;
+
+ assert(!MI->isBundle() &&
+ "TargetInstrInfoImpl::PredicateInstruction() can't handle bundles");
+
const MCInstrDesc &MCID = MI->getDesc();
- if (!MCID.isPredicable())
+ if (!MI->isPredicable())
return false;
for (unsigned j = 0, i = 0, e = MI->getNumOperands(); i != e; ++i) {
@@ -218,7 +239,7 @@ TargetInstrInfoImpl::produceSameValue(const MachineInstr *MI0,
MachineInstr *TargetInstrInfoImpl::duplicate(MachineInstr *Orig,
MachineFunction &MF) const {
- assert(!Orig->getDesc().isNotDuplicable() &&
+ assert(!Orig->isNotDuplicable() &&
"Instruction cannot be duplicated");
return MF.CloneMachineInstr(Orig);
}
@@ -288,10 +309,10 @@ TargetInstrInfo::foldMemoryOperand(MachineBasicBlock::iterator MI,
if (MachineInstr *NewMI = foldMemoryOperandImpl(MF, MI, Ops, FI)) {
// Add a memory operand, foldMemoryOperandImpl doesn't do that.
assert((!(Flags & MachineMemOperand::MOStore) ||
- NewMI->getDesc().mayStore()) &&
+ NewMI->mayStore()) &&
"Folded a def to a non-store!");
assert((!(Flags & MachineMemOperand::MOLoad) ||
- NewMI->getDesc().mayLoad()) &&
+ NewMI->mayLoad()) &&
"Folded a use to a non-load!");
const MachineFrameInfo &MFI = *MF.getFrameInfo();
assert(MFI.getObjectOffset(FI) != -1);
@@ -331,7 +352,7 @@ MachineInstr*
TargetInstrInfo::foldMemoryOperand(MachineBasicBlock::iterator MI,
const SmallVectorImpl<unsigned> &Ops,
MachineInstr* LoadMI) const {
- assert(LoadMI->getDesc().canFoldAsLoad() && "LoadMI isn't foldable!");
+ assert(LoadMI->canFoldAsLoad() && "LoadMI isn't foldable!");
#ifndef NDEBUG
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
assert(MI->getOperand(Ops[i]).isUse() && "Folding load into def!");
@@ -382,10 +403,8 @@ isReallyTriviallyReMaterializableGeneric(const MachineInstr *MI,
MF.getFrameInfo()->isImmutableObjectIndex(FrameIdx))
return true;
- const MCInstrDesc &MCID = MI->getDesc();
-
// Avoid instructions obviously unsafe for remat.
- if (MCID.isNotDuplicable() || MCID.mayStore() ||
+ if (MI->isNotDuplicable() || MI->mayStore() ||
MI->hasUnmodeledSideEffects())
return false;
@@ -395,7 +414,7 @@ isReallyTriviallyReMaterializableGeneric(const MachineInstr *MI,
return false;
// Avoid instructions which load from potentially varying memory.
- if (MCID.mayLoad() && !MI->isInvariantLoad(AA))
+ if (MI->mayLoad() && !MI->isInvariantLoad(AA))
return false;
// If any of the registers accessed are non-constant, conservatively assume
@@ -456,7 +475,7 @@ bool TargetInstrInfoImpl::isSchedulingBoundary(const MachineInstr *MI,
const MachineBasicBlock *MBB,
const MachineFunction &MF) const{
// Terminators and labels can't be scheduled around.
- if (MI->getDesc().isTerminator() || MI->isLabel())
+ if (MI->isTerminator() || MI->isLabel())
return true;
// Don't attempt to schedule around any instruction that defines
@@ -492,3 +511,32 @@ CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
return (ScheduleHazardRecognizer *)
new ScoreboardHazardRecognizer(II, DAG, "post-RA-sched");
}
+
+int
+TargetInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
+ SDNode *DefNode, unsigned DefIdx,
+ SDNode *UseNode, unsigned UseIdx) const {
+ if (!ItinData || ItinData->isEmpty())
+ return -1;
+
+ if (!DefNode->isMachineOpcode())
+ return -1;
+
+ unsigned DefClass = get(DefNode->getMachineOpcode()).getSchedClass();
+ if (!UseNode->isMachineOpcode())
+ return ItinData->getOperandCycle(DefClass, DefIdx);
+ unsigned UseClass = get(UseNode->getMachineOpcode()).getSchedClass();
+ return ItinData->getOperandLatency(DefClass, DefIdx, UseClass, UseIdx);
+}
+
+int TargetInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
+ SDNode *N) const {
+ if (!ItinData || ItinData->isEmpty())
+ return 1;
+
+ if (!N->isMachineOpcode())
+ return 1;
+
+ return ItinData->getStageLatency(get(N->getMachineOpcode()).getSchedClass());
+}
+
diff --git a/lib/CodeGen/TargetLoweringObjectFileImpl.cpp b/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
index c43e5b6..7fe164a 100644
--- a/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
+++ b/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
@@ -536,9 +536,7 @@ getCFIPersonalitySymbol(const GlobalValue *GV, Mangler *Mang,
// Add information about the stub reference to MachOMMI so that the stub
// gets emitted by the asmprinter.
MCSymbol *SSym = getContext().GetOrCreateSymbol(Name.str());
- MachineModuleInfoImpl::StubValueTy &StubSym =
- GV->hasHiddenVisibility() ? MachOMMI.getHiddenGVStubEntry(SSym) :
- MachOMMI.getGVStubEntry(SSym);
+ MachineModuleInfoImpl::StubValueTy &StubSym = MachOMMI.getGVStubEntry(SSym);
if (StubSym.getPointer() == 0) {
MCSymbol *Sym = Mang->getSymbol(GV);
StubSym = MachineModuleInfoImpl::StubValueTy(Sym, !GV->hasLocalLinkage());
diff --git a/lib/CodeGen/TargetOptionsImpl.cpp b/lib/CodeGen/TargetOptionsImpl.cpp
new file mode 100644
index 0000000..0f59d01
--- /dev/null
+++ b/lib/CodeGen/TargetOptionsImpl.cpp
@@ -0,0 +1,52 @@
+//===-- TargetOptionsImpl.cpp - Options that apply to all targets ----------==//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the methods in the TargetOptions.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/Target/TargetOptions.h"
+using namespace llvm;
+
+/// DisableFramePointerElim - This returns true if frame pointer elimination
+/// optimization should be disabled for the given machine function.
+bool TargetOptions::DisableFramePointerElim(const MachineFunction &MF) const {
+ // Check to see if we should eliminate non-leaf frame pointers and then
+ // check to see if we should eliminate all frame pointers.
+ if (NoFramePointerElimNonLeaf && !NoFramePointerElim) {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ return MFI->hasCalls();
+ }
+
+ return NoFramePointerElim;
+}
+
+/// LessPreciseFPMAD - This flag return true when -enable-fp-mad option
+/// is specified on the command line. When this flag is off(default), the
+/// code generator is not allowed to generate mad (multiply add) if the
+/// result is "less precise" than doing those operations individually.
+bool TargetOptions::LessPreciseFPMAD() const {
+ return UnsafeFPMath || LessPreciseFPMADOption;
+}
+
+/// HonorSignDependentRoundingFPMath - Return true if the codegen must assume
+/// that the rounding mode of the FPU can change from its default.
+bool TargetOptions::HonorSignDependentRoundingFPMath() const {
+ return !UnsafeFPMath && HonorSignDependentRoundingFPMathOption;
+}
+
+/// getTrapFunctionName - If this returns a non-empty string, this means isel
+/// should lower Intrinsic::trap to a call to the specified function name
+/// instead of an ISD::TRAP node.
+StringRef TargetOptions::getTrapFunctionName() const {
+ return TrapFuncName;
+}
+
diff --git a/lib/CodeGen/TwoAddressInstructionPass.cpp b/lib/CodeGen/TwoAddressInstructionPass.cpp
index a2e8134..6a63335 100644
--- a/lib/CodeGen/TwoAddressInstructionPass.cpp
+++ b/lib/CodeGen/TwoAddressInstructionPass.cpp
@@ -242,7 +242,7 @@ bool TwoAddressInstructionPass::Sink3AddrInstruction(MachineBasicBlock *MBB,
// appropriate location, we can try to sink the current instruction
// past it.
if (!KillMI || KillMI->getParent() != MBB || KillMI == MI ||
- KillMI->getDesc().isTerminator())
+ KillMI->isTerminator())
return false;
// If any of the definitions are used by another instruction between the
@@ -498,8 +498,7 @@ MachineInstr *findLocalKill(unsigned Reg, MachineBasicBlock *MBB,
MachineInstr *UseMI = &*UI;
if (UseMI == MI || UseMI->getParent() != MBB)
continue;
- DenseMap<MachineInstr*, unsigned>::iterator DI = DistanceMap.find(UseMI);
- if (DI != DistanceMap.end())
+ if (DistanceMap.count(UseMI))
continue;
if (!UI.getOperand().isKill())
return 0;
@@ -817,10 +816,9 @@ void TwoAddressInstructionPass::ProcessCopy(MachineInstr *MI,
static bool isSafeToDelete(MachineInstr *MI,
const TargetInstrInfo *TII,
SmallVector<unsigned, 4> &Kills) {
- const MCInstrDesc &MCID = MI->getDesc();
- if (MCID.mayStore() || MCID.isCall())
+ if (MI->mayStore() || MI->isCall())
return false;
- if (MCID.isTerminator() || MI->hasUnmodeledSideEffects())
+ if (MI->isTerminator() || MI->hasUnmodeledSideEffects())
return false;
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
@@ -918,9 +916,8 @@ TwoAddressInstructionPass::RescheduleMIBelowKill(MachineBasicBlock *MBB,
// Don't mess with copies, they may be coalesced later.
return false;
- const MCInstrDesc &MCID = KillMI->getDesc();
- if (MCID.hasUnmodeledSideEffects() || MCID.isCall() || MCID.isBranch() ||
- MCID.isTerminator())
+ if (KillMI->hasUnmodeledSideEffects() || KillMI->isCall() ||
+ KillMI->isBranch() || KillMI->isTerminator())
// Don't move pass calls, etc.
return false;
@@ -975,9 +972,8 @@ TwoAddressInstructionPass::RescheduleMIBelowKill(MachineBasicBlock *MBB,
if (NumVisited > 10) // FIXME: Arbitrary limit to reduce compile time cost.
return false;
++NumVisited;
- const MCInstrDesc &OMCID = OtherMI->getDesc();
- if (OMCID.hasUnmodeledSideEffects() || OMCID.isCall() || OMCID.isBranch() ||
- OMCID.isTerminator())
+ if (OtherMI->hasUnmodeledSideEffects() || OtherMI->isCall() ||
+ OtherMI->isBranch() || OtherMI->isTerminator())
// Don't move pass calls, etc.
return false;
for (unsigned i = 0, e = OtherMI->getNumOperands(); i != e; ++i) {
@@ -1119,9 +1115,8 @@ TwoAddressInstructionPass::RescheduleKillAboveMI(MachineBasicBlock *MBB,
if (NumVisited > 10) // FIXME: Arbitrary limit to reduce compile time cost.
return false;
++NumVisited;
- const MCInstrDesc &MCID = OtherMI->getDesc();
- if (MCID.hasUnmodeledSideEffects() || MCID.isCall() || MCID.isBranch() ||
- MCID.isTerminator())
+ if (OtherMI->hasUnmodeledSideEffects() || OtherMI->isCall() ||
+ OtherMI->isBranch() || OtherMI->isTerminator())
// Don't move pass calls, etc.
return false;
SmallVector<unsigned, 2> OtherDefs;
@@ -1201,7 +1196,6 @@ TryInstructionTransform(MachineBasicBlock::iterator &mi,
return false;
MachineInstr &MI = *mi;
- const MCInstrDesc &MCID = MI.getDesc();
unsigned regA = MI.getOperand(DstIdx).getReg();
unsigned regB = MI.getOperand(SrcIdx).getReg();
@@ -1223,7 +1217,7 @@ TryInstructionTransform(MachineBasicBlock::iterator &mi,
unsigned regCIdx = ~0U;
bool TryCommute = false;
bool AggressiveCommute = false;
- if (MCID.isCommutable() && MI.getNumOperands() >= 3 &&
+ if (MI.isCommutable() && MI.getNumOperands() >= 3 &&
TII->findCommutedOpIndices(&MI, SrcOp1, SrcOp2)) {
if (SrcIdx == SrcOp1)
regCIdx = SrcOp2;
@@ -1261,7 +1255,7 @@ TryInstructionTransform(MachineBasicBlock::iterator &mi,
if (TargetRegisterInfo::isVirtualRegister(regA))
ScanUses(regA, &*mbbi, Processed);
- if (MCID.isConvertibleTo3Addr()) {
+ if (MI.isConvertibleTo3Addr()) {
// This instruction is potentially convertible to a true
// three-address instruction. Check if it is profitable.
if (!regBKilled || isProfitableToConv3Addr(regA, regB)) {
@@ -1288,7 +1282,7 @@ TryInstructionTransform(MachineBasicBlock::iterator &mi,
// movq (%rax), %rcx
// addq %rdx, %rcx
// because it's preferable to schedule a load than a register copy.
- if (MCID.mayLoad() && !regBKilled) {
+ if (MI.mayLoad() && !regBKilled) {
// Determine if a load can be unfolded.
unsigned LoadRegIndex;
unsigned NewOpc =
@@ -1531,7 +1525,7 @@ bool TwoAddressInstructionPass::runOnMachineFunction(MachineFunction &MF) {
// If it's safe and profitable, remat the definition instead of
// copying it.
if (DefMI &&
- DefMI->getDesc().isAsCheapAsAMove() &&
+ DefMI->isAsCheapAsAMove() &&
DefMI->isSafeToReMat(TII, AA, regB) &&
isProfitableToReMat(regB, rc, mi, DefMI, mbbi, Dist)){
DEBUG(dbgs() << "2addr: REMATTING : " << *DefMI << "\n");