src/gallium/auxiliary/gallivm/lp_bld_misc.cpp

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  18  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  19  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  20  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * The above copyright notice and this permission notice (including the
  23  * next paragraph) shall be included in all copies or substantial portions
  24  * of the Software.
  25  *
  26  **************************************************************************/
  27
  28
  29 /**
  30  * The purpose of this module is to expose LLVM functionality not available
  31  * through the C++ bindings.
  32  */
  33
  34
  35 #ifndef __STDC_LIMIT_MACROS
  36 #define __STDC_LIMIT_MACROS
  37 #endif
  38
  39 #ifndef __STDC_CONSTANT_MACROS
  40 #define __STDC_CONSTANT_MACROS
  41 #endif
  42
  43 // Undef these vars just to silence warnings
  44 #undef PACKAGE_BUGREPORT
  45 #undef PACKAGE_NAME
  46 #undef PACKAGE_STRING
  47 #undef PACKAGE_TARNAME
  48 #undef PACKAGE_VERSION
  49
  50
  51 #include <stddef.h>
  52
  53 #include <llvm-c/Core.h>
  54 #include <llvm-c/ExecutionEngine.h>
  55 #include <llvm/Target/TargetOptions.h>
  56 #include <llvm/ExecutionEngine/ExecutionEngine.h>
  57 #include <llvm/ADT/Triple.h>
  58 #include <llvm/ExecutionEngine/JITMemoryManager.h>
  59 #include <llvm/Support/CommandLine.h>
  60 #include <llvm/Support/Host.h>
  61 #include <llvm/Support/PrettyStackTrace.h>
  62
  63 #include <llvm/Support/TargetSelect.h>
  64
  65 #if HAVE_LLVM >= 0x0303
  66 #include <llvm/IR/IRBuilder.h>
  67 #include <llvm/IR/Module.h>
  68 #include <llvm/Support/CBindingWrapping.h>
  69 #endif
  70
  71 #include "pipe/p_config.h"
  72 #include "util/u_debug.h"
  73 #include "util/u_cpu_detect.h"
  74
  75 #include "lp_bld_misc.h"
  76
  77 namespace {
  78
  79 class LLVMEnsureMultithreaded {
  80 public:
  81    LLVMEnsureMultithreaded()
  82    {
  83 #if HAVE_LLVM < 0x0303
  84       if (!llvm::llvm_is_multithreaded()) {
  85          llvm::llvm_start_multithreaded();
  86       }
  87 #else
  88       if (!LLVMIsMultithreaded()) {
  89          LLVMStartMultithreaded();
  90       }
  91 #endif
  92    }
  93 };
  94
  95 static LLVMEnsureMultithreaded lLVMEnsureMultithreaded;
  96
  97 }
  98
  99 extern "C" void
 100 lp_set_target_options(void)
 101 {
 102 #if HAVE_LLVM < 0x0304
 103    /*
 104     * By default LLVM adds a signal handler to output a pretty stack trace.
 105     * This signal handler is never removed, causing problems when unloading the
 106     * shared object where the gallium driver resides.
 107     */
 108    llvm::DisablePrettyStackTrace = true;
 109 #endif
 110
 111    // If we have a native target, initialize it to ensure it is linked in and
 112    // usable by the JIT.
 113    llvm::InitializeNativeTarget();
 114
 115    llvm::InitializeNativeTargetAsmPrinter();
 116
 117    llvm::InitializeNativeTargetDisassembler();
 118 }
 119
 120
 121 extern "C"
 122 LLVMValueRef
 123 lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
 124                        const char *Name)
 125 {
 126    return llvm::wrap(llvm::unwrap(B)->CreateLoad(llvm::unwrap(PointerVal), true, Name));
 127 }
 128
 129
 130 extern "C"
 131 void
 132 lp_set_load_alignment(LLVMValueRef Inst,
 133                        unsigned Align)
 134 {
 135    llvm::unwrap<llvm::LoadInst>(Inst)->setAlignment(Align);
 136 }
 137
 138 extern "C"
 139 void
 140 lp_set_store_alignment(LLVMValueRef Inst,
 141                        unsigned Align)
 142 {
 143    llvm::unwrap<llvm::StoreInst>(Inst)->setAlignment(Align);
 144 }
 145
 146
 147 /*
 148  * Delegating is tedious but the default manager class is hidden in an
 149  * anonymous namespace in LLVM, so we cannot just derive from it to change
 150  * its behavior.
 151  */
 152 class DelegatingJITMemoryManager : public llvm::JITMemoryManager {
 153
 154    protected:
 155       virtual llvm::JITMemoryManager *mgr() const = 0;
 156
 157    public:
 158       /*
 159        * From JITMemoryManager
 160        */
 161       virtual void setMemoryWritable() {
 162          mgr()->setMemoryWritable();
 163       }
 164       virtual void setMemoryExecutable() {
 165          mgr()->setMemoryExecutable();
 166       }
 167       virtual void setPoisonMemory(bool poison) {
 168          mgr()->setPoisonMemory(poison);
 169       }
 170       virtual void AllocateGOT() {
 171          mgr()->AllocateGOT();
 172          /*
 173           * isManagingGOT() is not virtual in base class so we can't delegate.
 174           * Instead we mirror the value of HasGOT in our instance.
 175           */
 176          HasGOT = mgr()->isManagingGOT();
 177       }
 178       virtual uint8_t *getGOTBase() const {
 179          return mgr()->getGOTBase();
 180       }
 181       virtual uint8_t *startFunctionBody(const llvm::Function *F,
 182                                          uintptr_t &ActualSize) {
 183          return mgr()->startFunctionBody(F, ActualSize);
 184       }
 185       virtual uint8_t *allocateStub(const llvm::GlobalValue *F,
 186                                     unsigned StubSize,
 187                                     unsigned Alignment) {
 188          return mgr()->allocateStub(F, StubSize, Alignment);
 189       }
 190       virtual void endFunctionBody(const llvm::Function *F,
 191                                    uint8_t *FunctionStart,
 192                                    uint8_t *FunctionEnd) {
 193          mgr()->endFunctionBody(F, FunctionStart, FunctionEnd);
 194       }
 195       virtual uint8_t *allocateSpace(intptr_t Size, unsigned Alignment) {
 196          return mgr()->allocateSpace(Size, Alignment);
 197       }
 198       virtual uint8_t *allocateGlobal(uintptr_t Size, unsigned Alignment) {
 199          return mgr()->allocateGlobal(Size, Alignment);
 200       }
 201       virtual void deallocateFunctionBody(void *Body) {
 202          mgr()->deallocateFunctionBody(Body);
 203       }
 204 #if HAVE_LLVM < 0x0304
 205       virtual uint8_t *startExceptionTable(const llvm::Function *F,
 206                                            uintptr_t &ActualSize) {
 207          return mgr()->startExceptionTable(F, ActualSize);
 208       }
 209       virtual void endExceptionTable(const llvm::Function *F,
 210                                      uint8_t *TableStart,
 211                                      uint8_t *TableEnd,
 212                                      uint8_t *FrameRegister) {
 213          mgr()->endExceptionTable(F, TableStart, TableEnd,
 214                                   FrameRegister);
 215       }
 216       virtual void deallocateExceptionTable(void *ET) {
 217          mgr()->deallocateExceptionTable(ET);
 218       }
 219 #endif
 220       virtual bool CheckInvariants(std::string &s) {
 221          return mgr()->CheckInvariants(s);
 222       }
 223       virtual size_t GetDefaultCodeSlabSize() {
 224          return mgr()->GetDefaultCodeSlabSize();
 225       }
 226       virtual size_t GetDefaultDataSlabSize() {
 227          return mgr()->GetDefaultDataSlabSize();
 228       }
 229       virtual size_t GetDefaultStubSlabSize() {
 230          return mgr()->GetDefaultStubSlabSize();
 231       }
 232       virtual unsigned GetNumCodeSlabs() {
 233          return mgr()->GetNumCodeSlabs();
 234       }
 235       virtual unsigned GetNumDataSlabs() {
 236          return mgr()->GetNumDataSlabs();
 237       }
 238       virtual unsigned GetNumStubSlabs() {
 239          return mgr()->GetNumStubSlabs();
 240       }
 241
 242       /*
 243        * From RTDyldMemoryManager
 244        */
 245 #if HAVE_LLVM >= 0x0304
 246       virtual uint8_t *allocateCodeSection(uintptr_t Size,
 247                                            unsigned Alignment,
 248                                            unsigned SectionID,
 249                                            llvm::StringRef SectionName) {
 250          return mgr()->allocateCodeSection(Size, Alignment, SectionID,
 251                                            SectionName);
 252       }
 253 #else
 254       virtual uint8_t *allocateCodeSection(uintptr_t Size,
 255                                            unsigned Alignment,
 256                                            unsigned SectionID) {
 257          return mgr()->allocateCodeSection(Size, Alignment, SectionID);
 258       }
 259 #endif
 260 #if HAVE_LLVM >= 0x0303
 261       virtual uint8_t *allocateDataSection(uintptr_t Size,
 262                                            unsigned Alignment,
 263                                            unsigned SectionID,
 264 #if HAVE_LLVM >= 0x0304
 265                                            llvm::StringRef SectionName,
 266 #endif
 267                                            bool IsReadOnly) {
 268          return mgr()->allocateDataSection(Size, Alignment, SectionID,
 269 #if HAVE_LLVM >= 0x0304
 270                                            SectionName,
 271 #endif
 272                                            IsReadOnly);
 273       }
 274 #if HAVE_LLVM >= 0x0304
 275       virtual void registerEHFrames(uint8_t *Addr, uint64_t LoadAddr, size_t Size) {
 276          mgr()->registerEHFrames(Addr, LoadAddr, Size);
 277       }
 278       virtual void deregisterEHFrames(uint8_t *Addr, uint64_t LoadAddr, size_t Size) {
 279          mgr()->deregisterEHFrames(Addr, LoadAddr, Size);
 280       }
 281 #else
 282       virtual void registerEHFrames(llvm::StringRef SectionData) {
 283          mgr()->registerEHFrames(SectionData);
 284       }
 285 #endif
 286 #else
 287       virtual uint8_t *allocateDataSection(uintptr_t Size,
 288                                            unsigned Alignment,
 289                                            unsigned SectionID) {
 290          return mgr()->allocateDataSection(Size, Alignment, SectionID);
 291       }
 292 #endif
 293       virtual void *getPointerToNamedFunction(const std::string &Name,
 294                                               bool AbortOnFailure=true) {
 295          return mgr()->getPointerToNamedFunction(Name, AbortOnFailure);
 296       }
 297 #if HAVE_LLVM == 0x0303
 298       virtual bool applyPermissions(std::string *ErrMsg = 0) {
 299          return mgr()->applyPermissions(ErrMsg);
 300       }
 301 #elif HAVE_LLVM > 0x0303
 302       virtual bool finalizeMemory(std::string *ErrMsg = 0) {
 303          return mgr()->finalizeMemory(ErrMsg);
 304       }
 305 #endif
 306 };
 307
 308
 309 /*
 310  * Delegate memory management to one shared manager for more efficient use
 311  * of memory than creating a separate pool for each LLVM engine.
 312  * Keep generated code until freeGeneratedCode() is called, instead of when
 313  * memory manager is destroyed, which happens during engine destruction.
 314  * This allows additional memory savings as we don't have to keep the engine
 315  * around in order to use the code.
 316  * All methods are delegated to the shared manager except destruction and
 317  * deallocating code.  For the latter we just remember what needs to be
 318  * deallocated later.  The shared manager is deleted once it is empty.
 319  */
 320 class ShaderMemoryManager : public DelegatingJITMemoryManager {
 321
 322    static llvm::JITMemoryManager *TheMM;
 323    static unsigned NumUsers;
 324
 325    struct GeneratedCode {
 326       typedef std::vector<void *> Vec;
 327       Vec FunctionBody, ExceptionTable;
 328
 329       GeneratedCode() {
 330          ++NumUsers;
 331       }
 332
 333       ~GeneratedCode() {
 334          /*
 335           * Deallocate things as previously requested and
 336           * free shared manager when no longer used.
 337           */
 338          Vec::iterator i;
 339
 340          assert(TheMM);
 341          for ( i = FunctionBody.begin(); i != FunctionBody.end(); ++i )
 342             TheMM->deallocateFunctionBody(*i);
 343 #if HAVE_LLVM < 0x0304
 344          for ( i = ExceptionTable.begin(); i != ExceptionTable.end(); ++i )
 345             TheMM->deallocateExceptionTable(*i);
 346 #endif
 347          --NumUsers;
 348          if (NumUsers == 0) {
 349             delete TheMM;
 350             TheMM = 0;
 351          }
 352       }
 353    };
 354
 355    GeneratedCode *code;
 356
 357    llvm::JITMemoryManager *mgr() const {
 358       if (!TheMM) {
 359          TheMM = CreateDefaultMemManager();
 360       }
 361       return TheMM;
 362    }
 363
 364    public:
 365
 366       ShaderMemoryManager() {
 367          code = new GeneratedCode;
 368       }
 369
 370       virtual ~ShaderMemoryManager() {
 371          /*
 372           * 'code' is purposely not deleted.  It is the user's responsibility
 373           * to call getGeneratedCode() and freeGeneratedCode().
 374           */
 375       }
 376
 377       struct lp_generated_code *getGeneratedCode() {
 378          return (struct lp_generated_code *) code;
 379       }
 380
 381       static void freeGeneratedCode(struct lp_generated_code *code) {
 382          delete (GeneratedCode *) code;
 383       }
 384
 385 #if HAVE_LLVM < 0x0304
 386       virtual void deallocateExceptionTable(void *ET) {
 387          // remember for later deallocation
 388          code->ExceptionTable.push_back(ET);
 389       }
 390 #endif
 391
 392       virtual void deallocateFunctionBody(void *Body) {
 393          // remember for later deallocation
 394          code->FunctionBody.push_back(Body);
 395       }
 396 };
 397
 398 llvm::JITMemoryManager *ShaderMemoryManager::TheMM = 0;
 399 unsigned ShaderMemoryManager::NumUsers = 0;
 400
 401
 402 /**
 403  * Same as LLVMCreateJITCompilerForModule, but:
 404  * - allows using MCJIT and enabling AVX feature where available.
 405  * - set target options
 406  *
 407  * See also:
 408  * - llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp
 409  * - llvm/tools/lli/lli.cpp
 410  * - http://markmail.org/message/ttkuhvgj4cxxy2on#query:+page:1+mid:aju2dggerju3ivd3+state:results
 411  */
 412 extern "C"
 413 LLVMBool
 414 lp_build_create_jit_compiler_for_module(LLVMExecutionEngineRef *OutJIT,
 415                                         lp_generated_code **OutCode,
 416                                         LLVMModuleRef M,
 417                                         unsigned OptLevel,
 418                                         int useMCJIT,
 419                                         char **OutError)
 420 {
 421    using namespace llvm;
 422
 423    std::string Error;
 424    EngineBuilder builder(unwrap(M));
 425
 426    /**
 427     * LLVM 3.1+ haven't more "extern unsigned llvm::StackAlignmentOverride" and
 428     * friends for configuring code generation options, like stack alignment.
 429     */
 430    TargetOptions options;
 431 #if defined(PIPE_ARCH_X86)
 432    options.StackAlignmentOverride = 4;
 433 #if HAVE_LLVM < 0x0304
 434    options.RealignStack = true;
 435 #endif
 436 #endif
 437
 438 #if defined(DEBUG)
 439    options.JITEmitDebugInfo = true;
 440 #endif
 441
 442 #if defined(DEBUG) || defined(PROFILE)
 443 #if HAVE_LLVM < 0x0304
 444    options.NoFramePointerElimNonLeaf = true;
 445 #endif
 446    options.NoFramePointerElim = true;
 447 #endif
 448
 449    builder.setEngineKind(EngineKind::JIT)
 450           .setErrorStr(&Error)
 451           .setTargetOptions(options)
 452           .setOptLevel((CodeGenOpt::Level)OptLevel);
 453
 454    if (useMCJIT) {
 455        builder.setUseMCJIT(true);
 456 #ifdef _WIN32
 457        /*
 458         * MCJIT works on Windows, but currently only through ELF object format.
 459         */
 460        std::string targetTriple = llvm::sys::getProcessTriple();
 461        targetTriple.append("-elf");
 462        unwrap(M)->setTargetTriple(targetTriple);
 463 #endif
 464    }
 465
 466    llvm::SmallVector<std::string, 1> MAttrs;
 467    if (util_cpu_caps.has_avx) {
 468       /*
 469        * AVX feature is not automatically detected from CPUID by the X86 target
 470        * yet, because the old (yet default) JIT engine is not capable of
 471        * emitting the opcodes.  But as we're using MCJIT here, it is safe to
 472        * add set this attribute.
 473        */
 474       MAttrs.push_back("+avx");
 475       if (util_cpu_caps.has_f16c) {
 476          MAttrs.push_back("+f16c");
 477       }
 478       builder.setMAttrs(MAttrs);
 479    }
 480
 481    ShaderMemoryManager *MM = new ShaderMemoryManager();
 482    *OutCode = MM->getGeneratedCode();
 483
 484    builder.setJITMemoryManager(MM);
 485
 486    ExecutionEngine *JIT;
 487 #if HAVE_LLVM >= 0x0302
 488    JIT = builder.create();
 489 #else
 490    /*
 491     * Workaround http://llvm.org/PR12833
 492     */
 493    StringRef MArch = "";
 494    StringRef MCPU = "";
 495    Triple TT(unwrap(M)->getTargetTriple());
 496    JIT = builder.create(builder.selectTarget(TT, MArch, MCPU, MAttrs));
 497 #endif
 498    if (JIT) {
 499       *OutJIT = wrap(JIT);
 500       return 0;
 501    }
 502    lp_free_generated_code(*OutCode);
 503    *OutCode = 0;
 504    delete MM;
 505    *OutError = strdup(Error.c_str());
 506    return 1;
 507 }
 508
 509
 510 extern "C"
 511 void
 512 lp_free_generated_code(struct lp_generated_code *code)
 513 {
 514    ShaderMemoryManager::freeGeneratedCode(code);
 515 }