llvm-6502/lib/CodeGen/LLVMTargetMachine.cpp

529 lines
20 KiB
C++
Raw Normal View History

//===-- LLVMTargetMachine.cpp - Implement the LLVMTargetMachine class -----===//
//
// 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 LLVMTargetMachine class.
//
//===----------------------------------------------------------------------===//
#include "llvm/Target/TargetMachine.h"
#include "llvm/PassManager.h"
#include "llvm/Analysis/Passes.h"
#include "llvm/Analysis/Verifier.h"
#include "llvm/Assembly/PrintModulePass.h"
#include "llvm/CodeGen/AsmPrinter.h"
#include "llvm/CodeGen/MachineFunctionAnalysis.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/MC/MCInstrInfo.h"
#include "llvm/MC/MCStreamer.h"
#include "llvm/MC/MCSubtargetInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtargetInfo.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/ADT/OwningPtr.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/FormattedStream.h"
#include "llvm/Support/TargetRegistry.h"
using namespace llvm;
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,
cl::desc("Disable branch folding"));
static cl::opt<bool> DisableTailDuplicate("disable-tail-duplicate", cl::Hidden,
cl::desc("Disable tail duplication"));
static cl::opt<bool> DisableEarlyTailDup("disable-early-taildup", cl::Hidden,
cl::desc("Disable pre-register allocation tail duplication"));
Implement a block placement pass based on the branch probability and block frequency analyses. This differs substantially from the existing block-placement pass in LLVM: 1) It operates on the Machine-IR in the CodeGen layer. This exposes much more (and more precise) information and opportunities. Also, the results are more stable due to fewer transforms ocurring after the pass runs. 2) It uses the generalized probability and frequency analyses. These can model static heuristics, code annotation derived heuristics as well as eventual profile loading. By basing the optimization on the analysis interface it can work from any (or a combination) of these inputs. 3) It uses a more aggressive algorithm, both building chains from tho bottom up to maximize benefit, and using an SCC-based walk to layout chains of blocks in a profitable ordering without O(N^2) iterations which the old pass involves. The pass is currently gated behind a flag, and not enabled by default because it still needs to grow some important features. Most notably, it needs to support loop aligning and careful layout of loop structures much as done by hand currently in CodePlacementOpt. Once it supports these, and has sufficient testing and quality tuning, it should replace both of these passes. Thanks to Nick Lewycky and Richard Smith for help authoring & debugging this, and to Jakob, Andy, Eric, Jim, and probably a few others I'm forgetting for reviewing and answering all my questions. Writing a backend pass is *sooo* much better now than it used to be. =D git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142641 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-21 06:46:38 +00:00
static cl::opt<bool> EnableBlockPlacement("enable-block-placement",
cl::Hidden, cl::desc("Enable probability-driven block placement"));
static cl::opt<bool> EnableBlockPlacementStats("enable-block-placement-stats",
cl::Hidden, cl::desc("Collect probability-driven block placement stats"));
static cl::opt<bool> DisableCodePlace("disable-code-place", cl::Hidden,
cl::desc("Disable code placement"));
static cl::opt<bool> DisableSSC("disable-ssc", cl::Hidden,
cl::desc("Disable Stack Slot Coloring"));
static cl::opt<bool> DisableMachineDCE("disable-machine-dce", cl::Hidden,
cl::desc("Disable Machine Dead Code Elimination"));
static cl::opt<bool> DisableMachineLICM("disable-machine-licm", cl::Hidden,
cl::desc("Disable Machine LICM"));
static cl::opt<bool> DisableMachineCSE("disable-machine-cse", cl::Hidden,
cl::desc("Disable Machine Common Subexpression Elimination"));
static cl::opt<bool> DisablePostRAMachineLICM("disable-postra-machine-licm",
cl::Hidden,
cl::desc("Disable Machine LICM"));
static cl::opt<bool> DisableMachineSink("disable-machine-sink", cl::Hidden,
cl::desc("Disable Machine Sinking"));
static cl::opt<bool> DisableLSR("disable-lsr", cl::Hidden,
cl::desc("Disable Loop Strength Reduction Pass"));
static cl::opt<bool> DisableCGP("disable-cgp", cl::Hidden,
cl::desc("Disable Codegen Prepare"));
static cl::opt<bool> DisableCopyProp("disable-copyprop", cl::Hidden,
cl::desc("Disable Copy Propagation pass"));
static cl::opt<bool> PrintLSR("print-lsr-output", cl::Hidden,
cl::desc("Print LLVM IR produced by the loop-reduce pass"));
static cl::opt<bool> PrintISelInput("print-isel-input", cl::Hidden,
cl::desc("Print LLVM IR input to isel pass"));
static cl::opt<bool> PrintGCInfo("print-gc", cl::Hidden,
cl::desc("Dump garbage collector data"));
static cl::opt<bool> ShowMCEncoding("show-mc-encoding", cl::Hidden,
cl::desc("Show encoding in .s output"));
static cl::opt<bool> ShowMCInst("show-mc-inst", cl::Hidden,
cl::desc("Show instruction structure in .s output"));
static cl::opt<bool> VerifyMachineCode("verify-machineinstrs", cl::Hidden,
cl::desc("Verify generated machine code"),
cl::init(getenv("LLVM_VERIFY_MACHINEINSTRS")!=NULL));
static cl::opt<cl::boolOrDefault>
AsmVerbose("asm-verbose", cl::desc("Add comments to directives."),
cl::init(cl::BOU_UNSET));
static bool getVerboseAsm() {
switch (AsmVerbose) {
case cl::BOU_UNSET: return TargetMachine::getAsmVerbosityDefault();
case cl::BOU_TRUE: return true;
case cl::BOU_FALSE: return false;
}
llvm_unreachable("Invalid verbose asm state");
}
// Enable or disable FastISel. Both options are needed, because
// FastISel is enabled by default with -fast, and we wish to be
// able to enable or disable fast-isel independently from -O0.
static cl::opt<cl::boolOrDefault>
EnableFastISelOption("fast-isel", cl::Hidden,
cl::desc("Enable the \"fast\" instruction selector"));
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, 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,
// and if the old one gets included then MCAsmInfo will be NULL and
// we'll crash later.
// Provide the user with a useful error message about what's wrong.
assert(AsmInfo && "MCAsmInfo not initialized."
"Make sure you include the correct TargetSelect.h"
"and that InitializeAllTargetMCs() is being invoked!");
}
TargetPassConfig::TargetPassConfig(TargetMachine *tm, PassManagerBase &pm,
bool DisableVerifyFlag)
: TM(tm), PM(pm), DisableVerify(DisableVerifyFlag) {
// Register all target independent codegen passes to activate their PassIDs.
initializeCodeGen(*PassRegistry::getPassRegistry());
}
/// createPassConfig - Create a pass configuration object to be used by
/// addPassToEmitX methods for generating a pipeline of CodeGen passes.
TargetPassConfig *LLVMTargetMachine::createPassConfig(PassManagerBase &PM,
bool DisableVerify) {
return new TargetPassConfig(this, PM, DisableVerify);
}
bool LLVMTargetMachine::addPassesToEmitFile(PassManagerBase &PM,
formatted_raw_ostream &Out,
CodeGenFileType FileType,
bool DisableVerify) {
// Add common CodeGen passes.
MCContext *Context = 0;
OwningPtr<TargetPassConfig> PassConfig(createPassConfig(PM, DisableVerify));
if (PassConfig->addCodeGenPasses(Context))
return true;
assert(Context != 0 && "Failed to get MCContext");
if (hasMCSaveTempLabels())
Context->setAllowTemporaryLabels(false);
const MCAsmInfo &MAI = *getMCAsmInfo();
const MCSubtargetInfo &STI = getSubtarget<MCSubtargetInfo>();
OwningPtr<MCStreamer> AsmStreamer;
switch (FileType) {
case CGFT_AssemblyFile: {
MCInstPrinter *InstPrinter =
getTarget().createMCInstPrinter(MAI.getAssemblerDialect(), MAI, STI);
// Create a code emitter if asked to show the encoding.
MCCodeEmitter *MCE = 0;
MCAsmBackend *MAB = 0;
if (ShowMCEncoding) {
const MCSubtargetInfo &STI = getSubtarget<MCSubtargetInfo>();
MCE = getTarget().createMCCodeEmitter(*getInstrInfo(), STI, *Context);
MAB = getTarget().createMCAsmBackend(getTargetTriple());
}
MCStreamer *S = getTarget().createAsmStreamer(*Context, Out,
getVerboseAsm(),
hasMCUseLoc(),
hasMCUseCFI(),
hasMCUseDwarfDirectory(),
InstPrinter,
MCE, MAB,
ShowMCInst);
AsmStreamer.reset(S);
break;
}
case CGFT_ObjectFile: {
// Create the code emitter for the target if it exists. If not, .o file
// emission fails.
MCCodeEmitter *MCE = getTarget().createMCCodeEmitter(*getInstrInfo(), STI,
*Context);
MCAsmBackend *MAB = getTarget().createMCAsmBackend(getTargetTriple());
if (MCE == 0 || MAB == 0)
return true;
AsmStreamer.reset(getTarget().createMCObjectStreamer(getTargetTriple(),
*Context, *MAB, Out,
MCE, hasMCRelaxAll(),
hasMCNoExecStack()));
AsmStreamer.get()->InitSections();
break;
}
case CGFT_Null:
// The Null output is intended for use for performance analysis and testing,
// not real users.
AsmStreamer.reset(createNullStreamer(*Context));
break;
}
// Create the AsmPrinter, which takes ownership of AsmStreamer if successful.
FunctionPass *Printer = getTarget().createAsmPrinter(*this, *AsmStreamer);
if (Printer == 0)
return true;
// If successful, createAsmPrinter took ownership of AsmStreamer.
AsmStreamer.take();
PM.add(Printer);
PM.add(createGCInfoDeleter());
return false;
}
/// addPassesToEmitMachineCode - Add passes to the specified pass manager to
/// get machine code emitted. This uses a JITCodeEmitter object to handle
/// actually outputting the machine code and resolving things like the address
/// of functions. This method should returns true if machine code emission is
/// not supported.
///
bool LLVMTargetMachine::addPassesToEmitMachineCode(PassManagerBase &PM,
JITCodeEmitter &JCE,
bool DisableVerify) {
// Add common CodeGen passes.
MCContext *Ctx = 0;
OwningPtr<TargetPassConfig> PassConfig(createPassConfig(PM, DisableVerify));
if (PassConfig->addCodeGenPasses(Ctx))
return true;
addCodeEmitter(PM, JCE);
PM.add(createGCInfoDeleter());
return false; // success!
}
/// addPassesToEmitMC - Add passes to the specified pass manager to get
/// machine code emitted with the MCJIT. This method returns true if machine
/// code is not supported. It fills the MCContext Ctx pointer which can be
/// used to build custom MCStreamer.
///
bool LLVMTargetMachine::addPassesToEmitMC(PassManagerBase &PM,
MCContext *&Ctx,
raw_ostream &Out,
bool DisableVerify) {
// Add common CodeGen passes.
OwningPtr<TargetPassConfig> PassConfig(createPassConfig(PM, DisableVerify));
if (PassConfig->addCodeGenPasses(Ctx))
return true;
if (hasMCSaveTempLabels())
Ctx->setAllowTemporaryLabels(false);
// Create the code emitter for the target if it exists. If not, .o file
// emission fails.
const MCSubtargetInfo &STI = getSubtarget<MCSubtargetInfo>();
MCCodeEmitter *MCE = getTarget().createMCCodeEmitter(*getInstrInfo(),STI, *Ctx);
MCAsmBackend *MAB = getTarget().createMCAsmBackend(getTargetTriple());
if (MCE == 0 || MAB == 0)
return true;
OwningPtr<MCStreamer> AsmStreamer;
AsmStreamer.reset(getTarget().createMCObjectStreamer(getTargetTriple(), *Ctx,
*MAB, Out, MCE,
hasMCRelaxAll(),
hasMCNoExecStack()));
AsmStreamer.get()->InitSections();
// Create the AsmPrinter, which takes ownership of AsmStreamer if successful.
FunctionPass *Printer = getTarget().createAsmPrinter(*this, *AsmStreamer);
if (Printer == 0)
return true;
// If successful, createAsmPrinter took ownership of AsmStreamer.
AsmStreamer.take();
PM.add(Printer);
return false; // success!
}
void TargetPassConfig::printNoVerify(const char *Banner) const {
if (TM->shouldPrintMachineCode())
PM.add(createMachineFunctionPrinterPass(dbgs(), Banner));
}
void TargetPassConfig::printAndVerify(const char *Banner) const {
if (TM->shouldPrintMachineCode())
PM.add(createMachineFunctionPrinterPass(dbgs(), Banner));
if (VerifyMachineCode)
PM.add(createMachineVerifierPass(Banner));
}
/// addCodeGenPasses - Add standard LLVM codegen passes used for both
/// emitting to assembly files or machine code output.
///
bool TargetPassConfig::addCodeGenPasses(MCContext *&OutContext) {
// Standard LLVM-Level Passes.
// Basic AliasAnalysis support.
// Add TypeBasedAliasAnalysis before BasicAliasAnalysis so that
// BasicAliasAnalysis wins if they disagree. This is intended to help
// support "obvious" type-punning idioms.
PM.add(createTypeBasedAliasAnalysisPass());
PM.add(createBasicAliasAnalysisPass());
// Before running any passes, run the verifier to determine if the input
// coming from the front-end and/or optimizer is valid.
if (!DisableVerify)
PM.add(createVerifierPass());
// Run loop strength reduction before anything else.
if (getOptLevel() != CodeGenOpt::None && !DisableLSR) {
PM.add(createLoopStrengthReducePass(getTargetLowering()));
if (PrintLSR)
PM.add(createPrintFunctionPass("\n\n*** Code after LSR ***\n", &dbgs()));
}
PM.add(createGCLoweringPass());
// Make sure that no unreachable blocks are instruction selected.
PM.add(createUnreachableBlockEliminationPass());
Add a new codegen pass that normalizes dwarf exception handling code in preparation for code generation. The main thing it does is handle the case when eh.exception calls (and, in a future patch, eh.selector calls) are far away from landing pads. Right now in practice you only find eh.exception calls close to landing pads: either in a landing pad (the common case) or in a landing pad successor, due to loop passes shifting them about. However future exception handling improvements will result in calls far from landing pads: (1) Inlining of rewinds. Consider the following case: In function @f: ... invoke @g to label %normal unwind label %unwinds ... unwinds: %ex = call i8* @llvm.eh.exception() ... In function @g: ... invoke @something to label %continue unwind label %handler ... handler: %ex = call i8* @llvm.eh.exception() ... perform cleanups ... "rethrow exception" Now inline @g into @f. Currently this is turned into: In function @f: ... invoke @something to label %continue unwind label %handler ... handler: %ex = call i8* @llvm.eh.exception() ... perform cleanups ... invoke "rethrow exception" to label %normal unwind label %unwinds unwinds: %ex = call i8* @llvm.eh.exception() ... However we would like to simplify invoke of "rethrow exception" into a branch to the %unwinds label. Then %unwinds is no longer a landing pad, and the eh.exception call there is then far away from any landing pads. (2) Using the unwind instruction for cleanups. It would be nice to have codegen handle the following case: invoke @something to label %continue unwind label %run_cleanups ... handler: ... perform cleanups ... unwind This requires turning "unwind" into a library call, which necessarily takes a pointer to the exception as an argument (this patch also does this unwind lowering). But that means you are using eh.exception again far from a landing pad. (3) Bugpoint simplifications. When bugpoint is simplifying exception handling code it often generates eh.exception calls far from a landing pad, which then causes codegen to assert. Bugpoint then latches on to this assertion and loses sight of the original problem. Note that it is currently rare for this pass to actually do anything. And in fact it normally shouldn't do anything at all given the code coming out of llvm-gcc! But it does fire a few times in the testsuite. As far as I can see this is almost always due to the LoopStrengthReduce codegen pass introducing pointless loop preheader blocks which are landing pads and only contain a branch to another block. This other block contains an eh.exception call. So probably by tweaking LoopStrengthReduce a bit this can be avoided. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72276 91177308-0d34-0410-b5e6-96231b3b80d8
2009-05-22 20:36:31 +00:00
// Turn exception handling constructs into something the code generators can
// handle.
switch (TM->getMCAsmInfo()->getExceptionHandlingType()) {
case ExceptionHandling::SjLj:
// SjLj piggy-backs on dwarf for this bit. The cleanups done apply to both
// Dwarf EH prepare needs to be run after SjLj prepare. Otherwise,
// catch info can get misplaced when a selector ends up more than one block
// removed from the parent invoke(s). This could happen when a landing
// pad is shared by multiple invokes and is also a target of a normal
// edge from elsewhere.
PM.add(createSjLjEHPass(getTargetLowering()));
// FALLTHROUGH
case ExceptionHandling::DwarfCFI:
case ExceptionHandling::ARM:
case ExceptionHandling::Win64:
PM.add(createDwarfEHPass(TM));
break;
case ExceptionHandling::None:
PM.add(createLowerInvokePass(getTargetLowering()));
// The lower invoke pass may create unreachable code. Remove it.
PM.add(createUnreachableBlockEliminationPass());
break;
}
Add a new codegen pass that normalizes dwarf exception handling code in preparation for code generation. The main thing it does is handle the case when eh.exception calls (and, in a future patch, eh.selector calls) are far away from landing pads. Right now in practice you only find eh.exception calls close to landing pads: either in a landing pad (the common case) or in a landing pad successor, due to loop passes shifting them about. However future exception handling improvements will result in calls far from landing pads: (1) Inlining of rewinds. Consider the following case: In function @f: ... invoke @g to label %normal unwind label %unwinds ... unwinds: %ex = call i8* @llvm.eh.exception() ... In function @g: ... invoke @something to label %continue unwind label %handler ... handler: %ex = call i8* @llvm.eh.exception() ... perform cleanups ... "rethrow exception" Now inline @g into @f. Currently this is turned into: In function @f: ... invoke @something to label %continue unwind label %handler ... handler: %ex = call i8* @llvm.eh.exception() ... perform cleanups ... invoke "rethrow exception" to label %normal unwind label %unwinds unwinds: %ex = call i8* @llvm.eh.exception() ... However we would like to simplify invoke of "rethrow exception" into a branch to the %unwinds label. Then %unwinds is no longer a landing pad, and the eh.exception call there is then far away from any landing pads. (2) Using the unwind instruction for cleanups. It would be nice to have codegen handle the following case: invoke @something to label %continue unwind label %run_cleanups ... handler: ... perform cleanups ... unwind This requires turning "unwind" into a library call, which necessarily takes a pointer to the exception as an argument (this patch also does this unwind lowering). But that means you are using eh.exception again far from a landing pad. (3) Bugpoint simplifications. When bugpoint is simplifying exception handling code it often generates eh.exception calls far from a landing pad, which then causes codegen to assert. Bugpoint then latches on to this assertion and loses sight of the original problem. Note that it is currently rare for this pass to actually do anything. And in fact it normally shouldn't do anything at all given the code coming out of llvm-gcc! But it does fire a few times in the testsuite. As far as I can see this is almost always due to the LoopStrengthReduce codegen pass introducing pointless loop preheader blocks which are landing pads and only contain a branch to another block. This other block contains an eh.exception call. So probably by tweaking LoopStrengthReduce a bit this can be avoided. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72276 91177308-0d34-0410-b5e6-96231b3b80d8
2009-05-22 20:36:31 +00:00
if (getOptLevel() != CodeGenOpt::None && !DisableCGP)
PM.add(createCodeGenPreparePass(getTargetLowering()));
PM.add(createStackProtectorPass(getTargetLowering()));
addPreISel();
if (PrintISelInput)
PM.add(createPrintFunctionPass("\n\n"
"*** Final LLVM Code input to ISel ***\n",
&dbgs()));
// All passes which modify the LLVM IR are now complete; run the verifier
// to ensure that the IR is valid.
if (!DisableVerify)
PM.add(createVerifierPass());
// Standard Lower-Level Passes.
// Install a MachineModuleInfo class, which is an immutable pass that holds
// all the per-module stuff we're generating, including MCContext.
MachineModuleInfo *MMI =
new MachineModuleInfo(*TM->getMCAsmInfo(), *TM->getRegisterInfo(),
&getTargetLowering()->getObjFileLowering());
PM.add(MMI);
OutContext = &MMI->getContext(); // Return the MCContext specifically by-ref.
// Set up a MachineFunction for the rest of CodeGen to work on.
PM.add(new MachineFunctionAnalysis(*TM));
// Enable FastISel with -fast, but allow that to be overridden.
if (EnableFastISelOption == cl::BOU_TRUE ||
(getOptLevel() == CodeGenOpt::None &&
EnableFastISelOption != cl::BOU_FALSE))
TM->setFastISel(true);
// Ask the target for an isel.
if (addInstSelector())
return true;
// Print the instruction selected machine code...
printAndVerify("After Instruction Selection");
// Expand pseudo-instructions emitted by ISel.
PM.add(createExpandISelPseudosPass());
// Pre-ra tail duplication.
if (getOptLevel() != CodeGenOpt::None && !DisableEarlyTailDup) {
PM.add(createTailDuplicatePass(true));
printAndVerify("After Pre-RegAlloc TailDuplicate");
}
// Optimize PHIs before DCE: removing dead PHI cycles may make more
// instructions dead.
if (getOptLevel() != CodeGenOpt::None)
PM.add(createOptimizePHIsPass());
// If the target requests it, assign local variables to stack slots relative
// to one another and simplify frame index references where possible.
PM.add(createLocalStackSlotAllocationPass());
if (getOptLevel() != CodeGenOpt::None) {
// With optimization, dead code should already be eliminated. However
// there is one known exception: lowered code for arguments that are only
// used by tail calls, where the tail calls reuse the incoming stack
// arguments directly (see t11 in test/CodeGen/X86/sibcall.ll).
if (!DisableMachineDCE)
PM.add(createDeadMachineInstructionElimPass());
printAndVerify("After codegen DCE pass");
if (!DisableMachineLICM)
PM.add(createMachineLICMPass());
if (!DisableMachineCSE)
PM.add(createMachineCSEPass());
if (!DisableMachineSink)
PM.add(createMachineSinkingPass());
printAndVerify("After Machine LICM, CSE and Sinking passes");
PM.add(createPeepholeOptimizerPass());
printAndVerify("After codegen peephole optimization pass");
}
// Run pre-ra passes.
if (addPreRegAlloc())
printAndVerify("After PreRegAlloc passes");
// Perform register allocation.
PM.add(createRegisterAllocator(getOptLevel()));
printAndVerify("After Register Allocation");
// Perform stack slot coloring and post-ra machine LICM.
if (getOptLevel() != CodeGenOpt::None) {
// FIXME: Re-enable coloring with register when it's capable of adding
// kill markers.
if (!DisableSSC)
PM.add(createStackSlotColoringPass(false));
// Run post-ra machine LICM to hoist reloads / remats.
if (!DisablePostRAMachineLICM)
PM.add(createMachineLICMPass(false));
printAndVerify("After StackSlotColoring and postra Machine LICM");
}
// Run post-ra passes.
if (addPostRegAlloc())
printAndVerify("After PostRegAlloc passes");
// Insert prolog/epilog code. Eliminate abstract frame index references...
PM.add(createPrologEpilogCodeInserter());
printAndVerify("After PrologEpilogCodeInserter");
// Branch folding must be run after regalloc and prolog/epilog insertion.
if (getOptLevel() != CodeGenOpt::None && !DisableBranchFold) {
PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
printNoVerify("After BranchFolding");
}
// Tail duplication.
if (getOptLevel() != CodeGenOpt::None && !DisableTailDuplicate) {
PM.add(createTailDuplicatePass(false));
printNoVerify("After TailDuplicate");
}
// Copy propagation.
if (getOptLevel() != CodeGenOpt::None && !DisableCopyProp) {
PM.add(createMachineCopyPropagationPass());
printNoVerify("After copy propagation pass");
}
// Expand pseudo instructions before second scheduling pass.
PM.add(createExpandPostRAPseudosPass());
printNoVerify("After ExpandPostRAPseudos");
// Run pre-sched2 passes.
if (addPreSched2())
printNoVerify("After PreSched2 passes");
// Second pass scheduler.
if (getOptLevel() != CodeGenOpt::None && !DisablePostRA) {
PM.add(createPostRAScheduler(getOptLevel()));
printNoVerify("After PostRAScheduler");
}
PM.add(createGCMachineCodeAnalysisPass());
if (PrintGCInfo)
PM.add(createGCInfoPrinter(dbgs()));
if (getOptLevel() != CodeGenOpt::None && !DisableCodePlace) {
Implement a block placement pass based on the branch probability and block frequency analyses. This differs substantially from the existing block-placement pass in LLVM: 1) It operates on the Machine-IR in the CodeGen layer. This exposes much more (and more precise) information and opportunities. Also, the results are more stable due to fewer transforms ocurring after the pass runs. 2) It uses the generalized probability and frequency analyses. These can model static heuristics, code annotation derived heuristics as well as eventual profile loading. By basing the optimization on the analysis interface it can work from any (or a combination) of these inputs. 3) It uses a more aggressive algorithm, both building chains from tho bottom up to maximize benefit, and using an SCC-based walk to layout chains of blocks in a profitable ordering without O(N^2) iterations which the old pass involves. The pass is currently gated behind a flag, and not enabled by default because it still needs to grow some important features. Most notably, it needs to support loop aligning and careful layout of loop structures much as done by hand currently in CodePlacementOpt. Once it supports these, and has sufficient testing and quality tuning, it should replace both of these passes. Thanks to Nick Lewycky and Richard Smith for help authoring & debugging this, and to Jakob, Andy, Eric, Jim, and probably a few others I'm forgetting for reviewing and answering all my questions. Writing a backend pass is *sooo* much better now than it used to be. =D git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142641 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-21 06:46:38 +00:00
if (EnableBlockPlacement) {
// MachineBlockPlacement is an experimental pass which is disabled by
// default currently. Eventually it should subsume CodePlacementOpt, so
// when enabled, the other is disabled.
PM.add(createMachineBlockPlacementPass());
printNoVerify("After MachineBlockPlacement");
Implement a block placement pass based on the branch probability and block frequency analyses. This differs substantially from the existing block-placement pass in LLVM: 1) It operates on the Machine-IR in the CodeGen layer. This exposes much more (and more precise) information and opportunities. Also, the results are more stable due to fewer transforms ocurring after the pass runs. 2) It uses the generalized probability and frequency analyses. These can model static heuristics, code annotation derived heuristics as well as eventual profile loading. By basing the optimization on the analysis interface it can work from any (or a combination) of these inputs. 3) It uses a more aggressive algorithm, both building chains from tho bottom up to maximize benefit, and using an SCC-based walk to layout chains of blocks in a profitable ordering without O(N^2) iterations which the old pass involves. The pass is currently gated behind a flag, and not enabled by default because it still needs to grow some important features. Most notably, it needs to support loop aligning and careful layout of loop structures much as done by hand currently in CodePlacementOpt. Once it supports these, and has sufficient testing and quality tuning, it should replace both of these passes. Thanks to Nick Lewycky and Richard Smith for help authoring & debugging this, and to Jakob, Andy, Eric, Jim, and probably a few others I'm forgetting for reviewing and answering all my questions. Writing a backend pass is *sooo* much better now than it used to be. =D git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142641 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-21 06:46:38 +00:00
} else {
PM.add(createCodePlacementOptPass());
printNoVerify("After CodePlacementOpt");
Implement a block placement pass based on the branch probability and block frequency analyses. This differs substantially from the existing block-placement pass in LLVM: 1) It operates on the Machine-IR in the CodeGen layer. This exposes much more (and more precise) information and opportunities. Also, the results are more stable due to fewer transforms ocurring after the pass runs. 2) It uses the generalized probability and frequency analyses. These can model static heuristics, code annotation derived heuristics as well as eventual profile loading. By basing the optimization on the analysis interface it can work from any (or a combination) of these inputs. 3) It uses a more aggressive algorithm, both building chains from tho bottom up to maximize benefit, and using an SCC-based walk to layout chains of blocks in a profitable ordering without O(N^2) iterations which the old pass involves. The pass is currently gated behind a flag, and not enabled by default because it still needs to grow some important features. Most notably, it needs to support loop aligning and careful layout of loop structures much as done by hand currently in CodePlacementOpt. Once it supports these, and has sufficient testing and quality tuning, it should replace both of these passes. Thanks to Nick Lewycky and Richard Smith for help authoring & debugging this, and to Jakob, Andy, Eric, Jim, and probably a few others I'm forgetting for reviewing and answering all my questions. Writing a backend pass is *sooo* much better now than it used to be. =D git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142641 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-21 06:46:38 +00:00
}
// Run a separate pass to collect block placement statistics.
if (EnableBlockPlacementStats) {
PM.add(createMachineBlockPlacementStatsPass());
printNoVerify("After MachineBlockPlacementStats");
}
}
if (addPreEmitPass())
printNoVerify("After PreEmit passes");
return false;
}