LLVMTargetMachine.cpp revision a342f90c3157d12a13eff85ae5211164906e98bf 
1//===-- LLVMTargetMachine.cpp - Implement the LLVMTargetMachine class -----===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file implements the LLVMTargetMachine class.
11//
12//===----------------------------------------------------------------------===//
13
14#include "llvm/Target/TargetMachine.h"
15#include "llvm/PassManager.h"
16#include "llvm/Pass.h"
17#include "llvm/Analysis/Verifier.h"
18#include "llvm/Assembly/PrintModulePass.h"
19#include "llvm/CodeGen/AsmPrinter.h"
20#include "llvm/CodeGen/Passes.h"
21#include "llvm/CodeGen/GCStrategy.h"
22#include "llvm/CodeGen/MachineFunctionAnalysis.h"
23#include "llvm/Target/TargetOptions.h"
24#include "llvm/MC/MCAsmInfo.h"
25#include "llvm/MC/MCContext.h"
26#include "llvm/MC/MCStreamer.h"
27#include "llvm/Target/TargetData.h"
28#include "llvm/Target/TargetRegistry.h"
29#include "llvm/Transforms/Scalar.h"
30#include "llvm/ADT/OwningPtr.h"
31#include "llvm/Support/CommandLine.h"
32#include "llvm/Support/Debug.h"
33#include "llvm/Support/FormattedStream.h"
34using namespace llvm;
35
36namespace llvm {
37  bool EnableFastISel;
38}
39
40static cl::opt<bool> DisablePostRA("disable-post-ra", cl::Hidden,
41    cl::desc("Disable Post Regalloc"));
42static cl::opt<bool> DisableBranchFold("disable-branch-fold", cl::Hidden,
43    cl::desc("Disable branch folding"));
44static cl::opt<bool> DisableTailDuplicate("disable-tail-duplicate", cl::Hidden,
45    cl::desc("Disable tail duplication"));
46static cl::opt<bool> DisableEarlyTailDup("disable-early-taildup", cl::Hidden,
47    cl::desc("Disable pre-register allocation tail duplication"));
48static cl::opt<bool> DisableCodePlace("disable-code-place", cl::Hidden,
49    cl::desc("Disable code placement"));
50static cl::opt<bool> DisableSSC("disable-ssc", cl::Hidden,
51    cl::desc("Disable Stack Slot Coloring"));
52static cl::opt<bool> DisableMachineLICM("disable-machine-licm", cl::Hidden,
53    cl::desc("Disable Machine LICM"));
54static cl::opt<bool> DisableMachineSink("disable-machine-sink", cl::Hidden,
55    cl::desc("Disable Machine Sinking"));
56static cl::opt<bool> DisableLSR("disable-lsr", cl::Hidden,
57    cl::desc("Disable Loop Strength Reduction Pass"));
58static cl::opt<bool> DisableCGP("disable-cgp", cl::Hidden,
59    cl::desc("Disable Codegen Prepare"));
60static cl::opt<bool> PrintLSR("print-lsr-output", cl::Hidden,
61    cl::desc("Print LLVM IR produced by the loop-reduce pass"));
62static cl::opt<bool> PrintISelInput("print-isel-input", cl::Hidden,
63    cl::desc("Print LLVM IR input to isel pass"));
64static cl::opt<bool> PrintGCInfo("print-gc", cl::Hidden,
65    cl::desc("Dump garbage collector data"));
66static cl::opt<bool> VerifyMachineCode("verify-machineinstrs", cl::Hidden,
67    cl::desc("Verify generated machine code"),
68    cl::init(getenv("LLVM_VERIFY_MACHINEINSTRS")!=NULL));
69
70static cl::opt<bool> EnableMachineCSE("enable-machine-cse", cl::Hidden,
71    cl::desc("Enable Machine CSE"));
72
73static cl::opt<cl::boolOrDefault>
74AsmVerbose("asm-verbose", cl::desc("Add comments to directives."),
75           cl::init(cl::BOU_UNSET));
76
77static bool getVerboseAsm() {
78  switch (AsmVerbose) {
79  default:
80  case cl::BOU_UNSET: return TargetMachine::getAsmVerbosityDefault();
81  case cl::BOU_TRUE:  return true;
82  case cl::BOU_FALSE: return false;
83  }
84}
85
86// Enable or disable FastISel. Both options are needed, because
87// FastISel is enabled by default with -fast, and we wish to be
88// able to enable or disable fast-isel independently from -O0.
89static cl::opt<cl::boolOrDefault>
90EnableFastISelOption("fast-isel", cl::Hidden,
91  cl::desc("Enable the \"fast\" instruction selector"));
92
93// Enable or disable an experimental optimization to split GEPs
94// and run a special GVN pass which does not examine loads, in
95// an effort to factor out redundancy implicit in complex GEPs.
96static cl::opt<bool> EnableSplitGEPGVN("split-gep-gvn", cl::Hidden,
97    cl::desc("Split GEPs and run no-load GVN"));
98
99LLVMTargetMachine::LLVMTargetMachine(const Target &T,
100                                     const std::string &TargetTriple)
101  : TargetMachine(T) {
102  AsmInfo = T.createAsmInfo(TargetTriple);
103}
104
105// Set the default code model for the JIT for a generic target.
106// FIXME: Is small right here? or .is64Bit() ? Large : Small?
107void
108LLVMTargetMachine::setCodeModelForJIT() {
109  setCodeModel(CodeModel::Small);
110}
111
112// Set the default code model for static compilation for a generic target.
113void
114LLVMTargetMachine::setCodeModelForStatic() {
115  setCodeModel(CodeModel::Small);
116}
117
118bool LLVMTargetMachine::addPassesToEmitFile(PassManagerBase &PM,
119                                            formatted_raw_ostream &Out,
120                                            CodeGenFileType FileType,
121                                            CodeGenOpt::Level OptLevel,
122                                            bool DisableVerify) {
123  // Add common CodeGen passes.
124  if (addCommonCodeGenPasses(PM, OptLevel, DisableVerify))
125    return true;
126
127  OwningPtr<MCContext> Context(new MCContext());
128  OwningPtr<MCStreamer> AsmStreamer;
129
130  formatted_raw_ostream *LegacyOutput;
131  switch (FileType) {
132  default: return true;
133  case CGFT_AssemblyFile: {
134    const MCAsmInfo &MAI = *getMCAsmInfo();
135    MCInstPrinter *InstPrinter =
136      getTarget().createMCInstPrinter(MAI.getAssemblerDialect(), MAI, Out);
137    AsmStreamer.reset(createAsmStreamer(*Context, Out, MAI,
138                                        getTargetData()->isLittleEndian(),
139                                        getVerboseAsm(), InstPrinter,
140                                        /*codeemitter*/0));
141    // Set the AsmPrinter's "O" to the output file.
142    LegacyOutput = &Out;
143    break;
144  }
145  case CGFT_ObjectFile: {
146    // Create the code emitter for the target if it exists.  If not, .o file
147    // emission fails.
148    MCCodeEmitter *MCE = getTarget().createCodeEmitter(*this, *Context);
149    if (MCE == 0)
150      return true;
151
152    AsmStreamer.reset(createMachOStreamer(*Context, Out, MCE));
153
154    // Any output to the asmprinter's "O" stream is bad and needs to be fixed,
155    // force it to come out stderr.
156    // FIXME: this is horrible and leaks, eventually remove the raw_ostream from
157    // asmprinter.
158    LegacyOutput = new formatted_raw_ostream(errs());
159    break;
160  }
161  case CGFT_Null:
162    // The Null output is intended for use for performance analysis and testing,
163    // not real users.
164    AsmStreamer.reset(createNullStreamer(*Context));
165    // Any output to the asmprinter's "O" stream is bad and needs to be fixed,
166    // force it to come out stderr.
167    // FIXME: this is horrible and leaks, eventually remove the raw_ostream from
168    // asmprinter.
169    LegacyOutput = new formatted_raw_ostream(errs());
170    break;
171  }
172
173  // Create the AsmPrinter, which takes ownership of Context and AsmStreamer
174  // if successful.
175  FunctionPass *Printer =
176    getTarget().createAsmPrinter(*LegacyOutput, *this, *Context, *AsmStreamer,
177                                 getMCAsmInfo());
178  if (Printer == 0)
179    return true;
180
181  // If successful, createAsmPrinter took ownership of AsmStreamer and Context.
182  Context.take(); AsmStreamer.take();
183
184  PM.add(Printer);
185
186  // Make sure the code model is set.
187  setCodeModelForStatic();
188  PM.add(createGCInfoDeleter());
189  return false;
190}
191
192/// addPassesToEmitMachineCode - Add passes to the specified pass manager to
193/// get machine code emitted.  This uses a JITCodeEmitter object to handle
194/// actually outputting the machine code and resolving things like the address
195/// of functions.  This method should returns true if machine code emission is
196/// not supported.
197///
198bool LLVMTargetMachine::addPassesToEmitMachineCode(PassManagerBase &PM,
199                                                   JITCodeEmitter &JCE,
200                                                   CodeGenOpt::Level OptLevel,
201                                                   bool DisableVerify) {
202  // Make sure the code model is set.
203  setCodeModelForJIT();
204
205  // Add common CodeGen passes.
206  if (addCommonCodeGenPasses(PM, OptLevel, DisableVerify))
207    return true;
208
209  addCodeEmitter(PM, OptLevel, JCE);
210  PM.add(createGCInfoDeleter());
211
212  return false; // success!
213}
214
215static void printAndVerify(PassManagerBase &PM,
216                           const char *Banner,
217                           bool allowDoubleDefs = false) {
218  if (PrintMachineCode)
219    PM.add(createMachineFunctionPrinterPass(dbgs(), Banner));
220
221  if (VerifyMachineCode)
222    PM.add(createMachineVerifierPass(allowDoubleDefs));
223}
224
225/// addCommonCodeGenPasses - Add standard LLVM codegen passes used for both
226/// emitting to assembly files or machine code output.
227///
228bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM,
229                                               CodeGenOpt::Level OptLevel,
230                                               bool DisableVerify) {
231  // Standard LLVM-Level Passes.
232
233  // Before running any passes, run the verifier to determine if the input
234  // coming from the front-end and/or optimizer is valid.
235  if (!DisableVerify)
236    PM.add(createVerifierPass());
237
238  // Optionally, tun split-GEPs and no-load GVN.
239  if (EnableSplitGEPGVN) {
240    PM.add(createGEPSplitterPass());
241    PM.add(createGVNPass(/*NoLoads=*/true));
242  }
243
244  // Run loop strength reduction before anything else.
245  if (OptLevel != CodeGenOpt::None && !DisableLSR) {
246    PM.add(createLoopStrengthReducePass(getTargetLowering()));
247    if (PrintLSR)
248      PM.add(createPrintFunctionPass("\n\n*** Code after LSR ***\n", &dbgs()));
249  }
250
251  // Turn exception handling constructs into something the code generators can
252  // handle.
253  switch (getMCAsmInfo()->getExceptionHandlingType())
254  {
255  case ExceptionHandling::SjLj:
256    // SjLj piggy-backs on dwarf for this bit. The cleanups done apply to both
257    // Dwarf EH prepare needs to be run after SjLj prepare. Otherwise,
258    // catch info can get misplaced when a selector ends up more than one block
259    // removed from the parent invoke(s). This could happen when a landing
260    // pad is shared by multiple invokes and is also a target of a normal
261    // edge from elsewhere.
262    PM.add(createSjLjEHPass(getTargetLowering()));
263    PM.add(createDwarfEHPass(getTargetLowering(), OptLevel==CodeGenOpt::None));
264    break;
265  case ExceptionHandling::Dwarf:
266    PM.add(createDwarfEHPass(getTargetLowering(), OptLevel==CodeGenOpt::None));
267    break;
268  case ExceptionHandling::None:
269    PM.add(createLowerInvokePass(getTargetLowering()));
270    break;
271  }
272
273  PM.add(createGCLoweringPass());
274
275  // Make sure that no unreachable blocks are instruction selected.
276  PM.add(createUnreachableBlockEliminationPass());
277
278  if (OptLevel != CodeGenOpt::None && !DisableCGP)
279    PM.add(createCodeGenPreparePass(getTargetLowering()));
280
281  PM.add(createStackProtectorPass(getTargetLowering()));
282
283  if (PrintISelInput)
284    PM.add(createPrintFunctionPass("\n\n"
285                                   "*** Final LLVM Code input to ISel ***\n",
286                                   &dbgs()));
287
288  // All passes which modify the LLVM IR are now complete; run the verifier
289  // to ensure that the IR is valid.
290  if (!DisableVerify)
291    PM.add(createVerifierPass());
292
293  // Standard Lower-Level Passes.
294
295  // Set up a MachineFunction for the rest of CodeGen to work on.
296  PM.add(new MachineFunctionAnalysis(*this, OptLevel));
297
298  // Enable FastISel with -fast, but allow that to be overridden.
299  if (EnableFastISelOption == cl::BOU_TRUE ||
300      (OptLevel == CodeGenOpt::None && EnableFastISelOption != cl::BOU_FALSE))
301    EnableFastISel = true;
302
303  // Ask the target for an isel.
304  if (addInstSelector(PM, OptLevel))
305    return true;
306
307  // Print the instruction selected machine code...
308  printAndVerify(PM, "After Instruction Selection",
309                 /* allowDoubleDefs= */ true);
310
311  // Optimize PHIs before DCE: removing dead PHI cycles may make more
312  // instructions dead.
313  if (OptLevel != CodeGenOpt::None)
314    PM.add(createOptimizePHIsPass());
315
316  // Delete dead machine instructions regardless of optimization level.
317  PM.add(createDeadMachineInstructionElimPass());
318  printAndVerify(PM, "After codegen DCE pass",
319                 /* allowDoubleDefs= */ true);
320
321  if (OptLevel != CodeGenOpt::None) {
322    PM.add(createOptimizeExtsPass());
323    if (EnableMachineCSE)
324      PM.add(createMachineCSEPass());
325    if (!DisableMachineLICM)
326      PM.add(createMachineLICMPass());
327    if (!DisableMachineSink)
328      PM.add(createMachineSinkingPass());
329    printAndVerify(PM, "After MachineLICM and MachineSinking",
330                   /* allowDoubleDefs= */ true);
331  }
332
333  // Pre-ra tail duplication.
334  if (OptLevel != CodeGenOpt::None && !DisableEarlyTailDup) {
335    PM.add(createTailDuplicatePass(true));
336    printAndVerify(PM, "After Pre-RegAlloc TailDuplicate",
337                   /* allowDoubleDefs= */ true);
338  }
339
340  // Run pre-ra passes.
341  if (addPreRegAlloc(PM, OptLevel))
342    printAndVerify(PM, "After PreRegAlloc passes",
343                   /* allowDoubleDefs= */ true);
344
345  // Perform register allocation.
346  PM.add(createRegisterAllocator());
347  printAndVerify(PM, "After Register Allocation");
348
349  // Perform stack slot coloring.
350  if (OptLevel != CodeGenOpt::None && !DisableSSC) {
351    // FIXME: Re-enable coloring with register when it's capable of adding
352    // kill markers.
353    PM.add(createStackSlotColoringPass(false));
354    printAndVerify(PM, "After StackSlotColoring");
355  }
356
357  // Run post-ra passes.
358  if (addPostRegAlloc(PM, OptLevel))
359    printAndVerify(PM, "After PostRegAlloc passes");
360
361  PM.add(createLowerSubregsPass());
362  printAndVerify(PM, "After LowerSubregs");
363
364  // Insert prolog/epilog code.  Eliminate abstract frame index references...
365  PM.add(createPrologEpilogCodeInserter());
366  printAndVerify(PM, "After PrologEpilogCodeInserter");
367
368  // Run pre-sched2 passes.
369  if (addPreSched2(PM, OptLevel))
370    printAndVerify(PM, "After PreSched2 passes");
371
372  // Second pass scheduler.
373  if (OptLevel != CodeGenOpt::None && !DisablePostRA) {
374    PM.add(createPostRAScheduler(OptLevel));
375    printAndVerify(PM, "After PostRAScheduler");
376  }
377
378  // Branch folding must be run after regalloc and prolog/epilog insertion.
379  if (OptLevel != CodeGenOpt::None && !DisableBranchFold) {
380    PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
381    printAndVerify(PM, "After BranchFolding");
382  }
383
384  // Tail duplication.
385  if (OptLevel != CodeGenOpt::None && !DisableTailDuplicate) {
386    PM.add(createTailDuplicatePass(false));
387    printAndVerify(PM, "After TailDuplicate");
388  }
389
390  PM.add(createGCMachineCodeAnalysisPass());
391
392  if (PrintGCInfo)
393    PM.add(createGCInfoPrinter(dbgs()));
394
395  if (OptLevel != CodeGenOpt::None && !DisableCodePlace) {
396    PM.add(createCodePlacementOptPass());
397    printAndVerify(PM, "After CodePlacementOpt");
398  }
399
400  if (addPreEmitPass(PM, OptLevel))
401    printAndVerify(PM, "After PreEmit passes");
402
403  return false;
404}
405