//
// The LLVM Compiler Infrastructure
//
-// This file was developed by Chris Lattner and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
#include "llvm/Target/TargetMachine.h"
#include "llvm/PassManager.h"
#include "llvm/Pass.h"
+#include "llvm/Assembly/PrintModulePass.h"
+#include "llvm/Analysis/LoopPass.h"
#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/Collector.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Transforms/Scalar.h"
+#include "llvm/Support/CommandLine.h"
using namespace llvm;
+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> PrintEmittedAsm("print-emitted-asm", cl::Hidden,
+ cl::desc("Dump emitter generated instructions as assembly"));
+static cl::opt<bool> PrintGCInfo("print-gc", cl::Hidden,
+ cl::desc("Dump garbage collector data"));
+
+// Hidden options to help debugging
+static cl::opt<bool>
+EnableSinking("enable-sinking", cl::init(false), cl::Hidden,
+ cl::desc("Perform sinking on machine code"));
+static cl::opt<bool>
+PerformLICM("machine-licm",
+ cl::init(false), cl::Hidden,
+ cl::desc("Perform loop-invariant code motion on machine code"));
+
+// When this works it will be on by default.
+static cl::opt<bool>
+DisablePostRAScheduler("disable-post-RA-scheduler",
+ cl::desc("Disable scheduling after register allocation"),
+ cl::init(true));
+
FileModel::Model
LLVMTargetMachine::addPassesToEmitFile(FunctionPassManager &PM,
std::ostream &Out,
// Standard LLVM-Level Passes.
// Run loop strength reduction before anything else.
- if (!Fast) PM.add(createLoopStrengthReducePass(getTargetLowering()));
-
- // FIXME: Implement efficient support for garbage collection intrinsics.
- PM.add(createLowerGCPass());
+ if (!Fast) {
+ PM.add(createLoopStrengthReducePass(getTargetLowering()));
+ if (PrintLSR)
+ PM.add(new PrintFunctionPass("\n\n*** Code after LSR *** \n", &cerr));
+ }
- // FIXME: Implement the invoke/unwind instructions!
+ PM.add(createGCLoweringPass());
+
if (!ExceptionHandling)
PM.add(createLowerInvokePass(getTargetLowering()));
-
+
// Make sure that no unreachable blocks are instruction selected.
PM.add(createUnreachableBlockEliminationPass());
+ if (!Fast)
+ PM.add(createCodeGenPreparePass(getTargetLowering()));
+
+ if (PrintISelInput)
+ PM.add(new PrintFunctionPass("\n\n*** Final LLVM Code input to ISel *** \n",
+ &cerr));
+
// Ask the target for an isel.
if (addInstSelector(PM, Fast))
return FileModel::Error;
// Print the instruction selected machine code...
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(cerr));
+
+ if (PerformLICM)
+ PM.add(createMachineLICMPass());
+ if (EnableSinking)
+ PM.add(createMachineSinkingPass());
+
// Perform register allocation to convert to a concrete x86 representation
PM.add(createRegisterAllocator());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(cerr));
+
+ PM.add(createLowerSubregsPass());
+
+ if (PrintMachineCode) // Print the subreg lowered code
+ PM.add(createMachineFunctionPrinterPass(cerr));
// Run post-ra passes.
if (addPostRegAlloc(PM, Fast) && PrintMachineCode)
// Insert prolog/epilog code. Eliminate abstract frame index references...
PM.add(createPrologEpilogCodeInserter());
+ // Second pass scheduler.
+ if (!Fast && !DisablePostRAScheduler)
+ PM.add(createPostRAScheduler());
+
// Branch folding must be run after regalloc and prolog/epilog insertion.
if (!Fast)
- PM.add(createBranchFoldingPass());
-
+ PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
+
+ PM.add(createGCMachineCodeAnalysisPass());
+ if (PrintMachineCode)
+ PM.add(createMachineFunctionPrinterPass(cerr));
+
+ if (PrintGCInfo)
+ PM.add(createCollectorMetadataPrinter(*cerr));
+
// Fold redundant debug labels.
PM.add(createDebugLabelFoldingPass());
MachineCodeEmitter *MCE,
bool Fast) {
if (MCE)
- addSimpleCodeEmitter(PM, Fast, *MCE);
+ addSimpleCodeEmitter(PM, Fast, PrintEmittedAsm, *MCE);
+
+ PM.add(createCollectorMetadataDeleter());
// Delete machine code for this function
PM.add(createMachineCodeDeleter());
// Standard LLVM-Level Passes.
// Run loop strength reduction before anything else.
- if (!Fast) PM.add(createLoopStrengthReducePass(getTargetLowering()));
+ if (!Fast) {
+ PM.add(createLoopStrengthReducePass(getTargetLowering()));
+ if (PrintLSR)
+ PM.add(new PrintFunctionPass("\n\n*** Code after LSR *** \n", &cerr));
+ }
- // FIXME: Implement efficient support for garbage collection intrinsics.
- PM.add(createLowerGCPass());
+ PM.add(createGCLoweringPass());
- // FIXME: Implement the invoke/unwind instructions!
- PM.add(createLowerInvokePass(getTargetLowering()));
+ if (!ExceptionHandling)
+ PM.add(createLowerInvokePass(getTargetLowering()));
// Make sure that no unreachable blocks are instruction selected.
PM.add(createUnreachableBlockEliminationPass());
+ if (!Fast)
+ PM.add(createCodeGenPreparePass(getTargetLowering()));
+
+ if (PrintISelInput)
+ PM.add(new PrintFunctionPass("\n\n*** Final LLVM Code input to ISel *** \n",
+ &cerr));
+
// Ask the target for an isel.
if (addInstSelector(PM, Fast))
return true;
// Print the instruction selected machine code...
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(cerr));
+
+ if (PerformLICM)
+ PM.add(createMachineLICMPass());
+ if (EnableSinking)
+ PM.add(createMachineSinkingPass());
+
// Perform register allocation to convert to a concrete x86 representation
PM.add(createRegisterAllocator());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(cerr));
+
+ PM.add(createLowerSubregsPass());
+
+ if (PrintMachineCode) // Print the subreg lowered code
+ PM.add(createMachineFunctionPrinterPass(cerr));
// Run post-ra passes.
if (addPostRegAlloc(PM, Fast) && PrintMachineCode)
if (PrintMachineCode) // Print the register-allocated code
PM.add(createMachineFunctionPrinterPass(cerr));
+ // Second pass scheduler.
+ if (!Fast)
+ PM.add(createPostRAScheduler());
+
// Branch folding must be run after regalloc and prolog/epilog insertion.
if (!Fast)
- PM.add(createBranchFoldingPass());
+ PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
+
+ PM.add(createGCMachineCodeAnalysisPass());
+ if (PrintMachineCode)
+ PM.add(createMachineFunctionPrinterPass(cerr));
+
+ if (PrintGCInfo)
+ PM.add(createCollectorMetadataPrinter(*cerr));
if (addPreEmitPass(PM, Fast) && PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(cerr));
- addCodeEmitter(PM, Fast, MCE);
+ addCodeEmitter(PM, Fast, PrintEmittedAsm, MCE);
+
+ PM.add(createCollectorMetadataDeleter());
// Delete machine code for this function
PM.add(createMachineCodeDeleter());