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 // Undef these vars just to silence warnings
  36 #undef PACKAGE_BUGREPORT
  37 #undef PACKAGE_NAME
  38 #undef PACKAGE_STRING
  39 #undef PACKAGE_TARNAME
  40 #undef PACKAGE_VERSION
  41
  42
  43 #include <stddef.h>
  44
  45 // Workaround http://llvm.org/PR23628
  46 #if HAVE_LLVM >= 0x0307
  47 #  pragma push_macro("DEBUG")
  48 #  undef DEBUG
  49 #endif
  50
  51 #include <llvm-c/Core.h>
  52 #include <llvm-c/ExecutionEngine.h>
  53 #include <llvm/Target/TargetOptions.h>
  54 #include <llvm/ExecutionEngine/ExecutionEngine.h>
  55 #include <llvm/ADT/Triple.h>
  56 #if HAVE_LLVM >= 0x0307
  57 #include <llvm/Analysis/TargetLibraryInfo.h>
  58 #else
  59 #include <llvm/Target/TargetLibraryInfo.h>
  60 #endif
  61 #if HAVE_LLVM < 0x0306
  62 #include <llvm/ExecutionEngine/JITMemoryManager.h>
  63 #else
  64 #include <llvm/ExecutionEngine/SectionMemoryManager.h>
  65 #endif
  66 #include <llvm/Support/CommandLine.h>
  67 #include <llvm/Support/Host.h>
  68 #include <llvm/Support/PrettyStackTrace.h>
  69
  70 #include <llvm/Support/TargetSelect.h>
  71
  72 #if HAVE_LLVM >= 0x0305
  73 #include <llvm/IR/CallSite.h>
  74 #endif
  75 #include <llvm/IR/IRBuilder.h>
  76 #include <llvm/IR/Module.h>
  77 #include <llvm/Support/CBindingWrapping.h>
  78
  79 #include <llvm/Config/llvm-config.h>
  80 #if LLVM_USE_INTEL_JITEVENTS
  81 #include <llvm/ExecutionEngine/JITEventListener.h>
  82 #endif
  83
  84 // Workaround http://llvm.org/PR23628
  85 #if HAVE_LLVM >= 0x0307
  86 #  pragma pop_macro("DEBUG")
  87 #endif
  88
  89 #include "c11/threads.h"
  90 #include "os/os_thread.h"
  91 #include "pipe/p_config.h"
  92 #include "util/u_debug.h"
  93 #include "util/u_cpu_detect.h"
  94
  95 #include "lp_bld_misc.h"
  96 #include "lp_bld_debug.h"
  97
  98 namespace {
  99
 100 class LLVMEnsureMultithreaded {
 101 public:
 102    LLVMEnsureMultithreaded()
 103    {
 104       if (!LLVMIsMultithreaded()) {
 105          LLVMStartMultithreaded();
 106       }
 107    }
 108 };
 109
 110 static LLVMEnsureMultithreaded lLVMEnsureMultithreaded;
 111
 112 }
 113
 114 static once_flag init_native_targets_once_flag = ONCE_FLAG_INIT;
 115
 116 static void init_native_targets()
 117 {
 118    // If we have a native target, initialize it to ensure it is linked in and
 119    // usable by the JIT.
 120    llvm::InitializeNativeTarget();
 121
 122    llvm::InitializeNativeTargetAsmPrinter();
 123
 124    llvm::InitializeNativeTargetDisassembler();
 125 }
 126
 127 extern "C" void
 128 lp_set_target_options(void)
 129 {
 130 #if HAVE_LLVM < 0x0304
 131    /*
 132     * By default LLVM adds a signal handler to output a pretty stack trace.
 133     * This signal handler is never removed, causing problems when unloading the
 134     * shared object where the gallium driver resides.
 135     */
 136    llvm::DisablePrettyStackTrace = true;
 137 #endif
 138
 139    /* The llvm target registry is not thread-safe, so drivers and state-trackers
 140     * that want to initialize targets should use the lp_set_target_options()
 141     * function to safely initialize targets.
 142     *
 143     * LLVM targets should be initialized before the driver or state-tracker tries
 144     * to access the registry.
 145     */
 146    call_once(&init_native_targets_once_flag, init_native_targets);
 147 }
 148
 149 extern "C"
 150 LLVMTargetLibraryInfoRef
 151 gallivm_create_target_library_info(const char *triple)
 152 {
 153    return reinterpret_cast<LLVMTargetLibraryInfoRef>(
 154 #if HAVE_LLVM < 0x0307
 155    new llvm::TargetLibraryInfo(
 156 #else
 157    new llvm::TargetLibraryInfoImpl(
 158 #endif
 159    llvm::Triple(triple)));
 160 }
 161
 162 extern "C"
 163 void
 164 gallivm_dispose_target_library_info(LLVMTargetLibraryInfoRef library_info)
 165 {
 166    delete reinterpret_cast<
 167 #if HAVE_LLVM < 0x0307
 168    llvm::TargetLibraryInfo
 169 #else
 170    llvm::TargetLibraryInfoImpl
 171 #endif
 172    *>(library_info);
 173 }
 174
 175
 176 #if HAVE_LLVM < 0x0304
 177
 178 extern "C"
 179 void
 180 LLVMSetAlignmentBackport(LLVMValueRef V,
 181                          unsigned Bytes)
 182 {
 183    switch (LLVMGetInstructionOpcode(V)) {
 184    case LLVMLoad:
 185       llvm::unwrap<llvm::LoadInst>(V)->setAlignment(Bytes);
 186       break;
 187    case LLVMStore:
 188       llvm::unwrap<llvm::StoreInst>(V)->setAlignment(Bytes);
 189       break;
 190    default:
 191       assert(0);
 192       break;
 193    }
 194 }
 195
 196 #endif
 197
 198
 199 #if HAVE_LLVM < 0x0306
 200 typedef llvm::JITMemoryManager BaseMemoryManager;
 201 #else
 202 typedef llvm::RTDyldMemoryManager BaseMemoryManager;
 203 #endif
 204
 205
 206 /*
 207  * Delegating is tedious but the default manager class is hidden in an
 208  * anonymous namespace in LLVM, so we cannot just derive from it to change
 209  * its behavior.
 210  */
 211 class DelegatingJITMemoryManager : public BaseMemoryManager {
 212
 213    protected:
 214       virtual BaseMemoryManager *mgr() const = 0;
 215
 216    public:
 217 #if HAVE_LLVM < 0x0306
 218       /*
 219        * From JITMemoryManager
 220        */
 221       virtual void setMemoryWritable() {
 222          mgr()->setMemoryWritable();
 223       }
 224       virtual void setMemoryExecutable() {
 225          mgr()->setMemoryExecutable();
 226       }
 227       virtual void setPoisonMemory(bool poison) {
 228          mgr()->setPoisonMemory(poison);
 229       }
 230       virtual void AllocateGOT() {
 231          mgr()->AllocateGOT();
 232          /*
 233           * isManagingGOT() is not virtual in base class so we can't delegate.
 234           * Instead we mirror the value of HasGOT in our instance.
 235           */
 236          HasGOT = mgr()->isManagingGOT();
 237       }
 238       virtual uint8_t *getGOTBase() const {
 239          return mgr()->getGOTBase();
 240       }
 241       virtual uint8_t *startFunctionBody(const llvm::Function *F,
 242                                          uintptr_t &ActualSize) {
 243          return mgr()->startFunctionBody(F, ActualSize);
 244       }
 245       virtual uint8_t *allocateStub(const llvm::GlobalValue *F,
 246                                     unsigned StubSize,
 247                                     unsigned Alignment) {
 248          return mgr()->allocateStub(F, StubSize, Alignment);
 249       }
 250       virtual void endFunctionBody(const llvm::Function *F,
 251                                    uint8_t *FunctionStart,
 252                                    uint8_t *FunctionEnd) {
 253          mgr()->endFunctionBody(F, FunctionStart, FunctionEnd);
 254       }
 255       virtual uint8_t *allocateSpace(intptr_t Size, unsigned Alignment) {
 256          return mgr()->allocateSpace(Size, Alignment);
 257       }
 258       virtual uint8_t *allocateGlobal(uintptr_t Size, unsigned Alignment) {
 259          return mgr()->allocateGlobal(Size, Alignment);
 260       }
 261       virtual void deallocateFunctionBody(void *Body) {
 262          mgr()->deallocateFunctionBody(Body);
 263       }
 264 #if HAVE_LLVM < 0x0304
 265       virtual uint8_t *startExceptionTable(const llvm::Function *F,
 266                                            uintptr_t &ActualSize) {
 267          return mgr()->startExceptionTable(F, ActualSize);
 268       }
 269       virtual void endExceptionTable(const llvm::Function *F,
 270                                      uint8_t *TableStart,
 271                                      uint8_t *TableEnd,
 272                                      uint8_t *FrameRegister) {
 273          mgr()->endExceptionTable(F, TableStart, TableEnd,
 274                                   FrameRegister);
 275       }
 276       virtual void deallocateExceptionTable(void *ET) {
 277          mgr()->deallocateExceptionTable(ET);
 278       }
 279 #endif
 280       virtual bool CheckInvariants(std::string &s) {
 281          return mgr()->CheckInvariants(s);
 282       }
 283       virtual size_t GetDefaultCodeSlabSize() {
 284          return mgr()->GetDefaultCodeSlabSize();
 285       }
 286       virtual size_t GetDefaultDataSlabSize() {
 287          return mgr()->GetDefaultDataSlabSize();
 288       }
 289       virtual size_t GetDefaultStubSlabSize() {
 290          return mgr()->GetDefaultStubSlabSize();
 291       }
 292       virtual unsigned GetNumCodeSlabs() {
 293          return mgr()->GetNumCodeSlabs();
 294       }
 295       virtual unsigned GetNumDataSlabs() {
 296          return mgr()->GetNumDataSlabs();
 297       }
 298       virtual unsigned GetNumStubSlabs() {
 299          return mgr()->GetNumStubSlabs();
 300       }
 301 #endif
 302
 303       /*
 304        * From RTDyldMemoryManager
 305        */
 306 #if HAVE_LLVM >= 0x0304
 307       virtual uint8_t *allocateCodeSection(uintptr_t Size,
 308                                            unsigned Alignment,
 309                                            unsigned SectionID,
 310                                            llvm::StringRef SectionName) {
 311          return mgr()->allocateCodeSection(Size, Alignment, SectionID,
 312                                            SectionName);
 313       }
 314 #else
 315       virtual uint8_t *allocateCodeSection(uintptr_t Size,
 316                                            unsigned Alignment,
 317                                            unsigned SectionID) {
 318          return mgr()->allocateCodeSection(Size, Alignment, SectionID);
 319       }
 320 #endif
 321       virtual uint8_t *allocateDataSection(uintptr_t Size,
 322                                            unsigned Alignment,
 323                                            unsigned SectionID,
 324 #if HAVE_LLVM >= 0x0304
 325                                            llvm::StringRef SectionName,
 326 #endif
 327                                            bool IsReadOnly) {
 328          return mgr()->allocateDataSection(Size, Alignment, SectionID,
 329 #if HAVE_LLVM >= 0x0304
 330                                            SectionName,
 331 #endif
 332                                            IsReadOnly);
 333       }
 334 #if HAVE_LLVM >= 0x0304
 335       virtual void registerEHFrames(uint8_t *Addr, uint64_t LoadAddr, size_t Size) {
 336          mgr()->registerEHFrames(Addr, LoadAddr, Size);
 337       }
 338 #else
 339       virtual void registerEHFrames(llvm::StringRef SectionData) {
 340          mgr()->registerEHFrames(SectionData);
 341       }
 342 #endif
 343 #if HAVE_LLVM >= 0x0500
 344       virtual void deregisterEHFrames() {
 345          mgr()->deregisterEHFrames();
 346       }
 347 #elif HAVE_LLVM >= 0x0304
 348       virtual void deregisterEHFrames(uint8_t *Addr, uint64_t LoadAddr, size_t Size) {
 349          mgr()->deregisterEHFrames(Addr, LoadAddr, Size);
 350       }
 351 #endif
 352       virtual void *getPointerToNamedFunction(const std::string &Name,
 353                                               bool AbortOnFailure=true) {
 354          return mgr()->getPointerToNamedFunction(Name, AbortOnFailure);
 355       }
 356 #if HAVE_LLVM <= 0x0303
 357       virtual bool applyPermissions(std::string *ErrMsg = 0) {
 358          return mgr()->applyPermissions(ErrMsg);
 359       }
 360 #else
 361       virtual bool finalizeMemory(std::string *ErrMsg = 0) {
 362          return mgr()->finalizeMemory(ErrMsg);
 363       }
 364 #endif
 365 };
 366
 367
 368 /*
 369  * Delegate memory management to one shared manager for more efficient use
 370  * of memory than creating a separate pool for each LLVM engine.
 371  * Keep generated code until freeGeneratedCode() is called, instead of when
 372  * memory manager is destroyed, which happens during engine destruction.
 373  * This allows additional memory savings as we don't have to keep the engine
 374  * around in order to use the code.
 375  * All methods are delegated to the shared manager except destruction and
 376  * deallocating code.  For the latter we just remember what needs to be
 377  * deallocated later.  The shared manager is deleted once it is empty.
 378  */
 379 class ShaderMemoryManager : public DelegatingJITMemoryManager {
 380
 381    BaseMemoryManager *TheMM;
 382
 383    struct GeneratedCode {
 384       typedef std::vector<void *> Vec;
 385       Vec FunctionBody, ExceptionTable;
 386       BaseMemoryManager *TheMM;
 387
 388       GeneratedCode(BaseMemoryManager *MM) {
 389          TheMM = MM;
 390       }
 391
 392       ~GeneratedCode() {
 393          /*
 394           * Deallocate things as previously requested and
 395           * free shared manager when no longer used.
 396           */
 397 #if HAVE_LLVM < 0x0306
 398          Vec::iterator i;
 399
 400          assert(TheMM);
 401          for ( i = FunctionBody.begin(); i != FunctionBody.end(); ++i )
 402             TheMM->deallocateFunctionBody(*i);
 403 #if HAVE_LLVM < 0x0304
 404          for ( i = ExceptionTable.begin(); i != ExceptionTable.end(); ++i )
 405             TheMM->deallocateExceptionTable(*i);
 406 #endif /* HAVE_LLVM < 0x0304 */
 407 #endif /* HAVE_LLVM < 0x0306 */
 408       }
 409    };
 410
 411    GeneratedCode *code;
 412
 413    BaseMemoryManager *mgr() const {
 414       return TheMM;
 415    }
 416
 417    public:
 418
 419       ShaderMemoryManager(BaseMemoryManager* MM) {
 420          TheMM = MM;
 421          code = new GeneratedCode(MM);
 422       }
 423
 424       virtual ~ShaderMemoryManager() {
 425          /*
 426           * 'code' is purposely not deleted.  It is the user's responsibility
 427           * to call getGeneratedCode() and freeGeneratedCode().
 428           */
 429       }
 430
 431       struct lp_generated_code *getGeneratedCode() {
 432          return (struct lp_generated_code *) code;
 433       }
 434
 435       static void freeGeneratedCode(struct lp_generated_code *code) {
 436          delete (GeneratedCode *) code;
 437       }
 438
 439 #if HAVE_LLVM < 0x0304
 440       virtual void deallocateExceptionTable(void *ET) {
 441          // remember for later deallocation
 442          code->ExceptionTable.push_back(ET);
 443       }
 444 #endif
 445
 446       virtual void deallocateFunctionBody(void *Body) {
 447          // remember for later deallocation
 448          code->FunctionBody.push_back(Body);
 449       }
 450 };
 451
 452
 453 /**
 454  * Same as LLVMCreateJITCompilerForModule, but:
 455  * - allows using MCJIT and enabling AVX feature where available.
 456  * - set target options
 457  *
 458  * See also:
 459  * - llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp
 460  * - llvm/tools/lli/lli.cpp
 461  * - http://markmail.org/message/ttkuhvgj4cxxy2on#query:+page:1+mid:aju2dggerju3ivd3+state:results
 462  */
 463 extern "C"
 464 LLVMBool
 465 lp_build_create_jit_compiler_for_module(LLVMExecutionEngineRef *OutJIT,
 466                                         lp_generated_code **OutCode,
 467                                         LLVMModuleRef M,
 468                                         LLVMMCJITMemoryManagerRef CMM,
 469                                         unsigned OptLevel,
 470                                         int useMCJIT,
 471                                         char **OutError)
 472 {
 473    using namespace llvm;
 474
 475    std::string Error;
 476 #if HAVE_LLVM >= 0x0306
 477    EngineBuilder builder(std::unique_ptr<Module>(unwrap(M)));
 478 #else
 479    EngineBuilder builder(unwrap(M));
 480 #endif
 481
 482    /**
 483     * LLVM 3.1+ haven't more "extern unsigned llvm::StackAlignmentOverride" and
 484     * friends for configuring code generation options, like stack alignment.
 485     */
 486    TargetOptions options;
 487 #if defined(PIPE_ARCH_X86)
 488    options.StackAlignmentOverride = 4;
 489 #if HAVE_LLVM < 0x0304
 490    options.RealignStack = true;
 491 #endif
 492 #endif
 493
 494 #if defined(DEBUG) && HAVE_LLVM < 0x0307
 495    options.JITEmitDebugInfo = true;
 496 #endif
 497
 498    /* XXX: Workaround http://llvm.org/PR21435 */
 499 #if defined(DEBUG) || defined(PROFILE) || \
 500     (HAVE_LLVM >= 0x0303 && (defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)))
 501 #if HAVE_LLVM < 0x0304
 502    options.NoFramePointerElimNonLeaf = true;
 503 #endif
 504 #if HAVE_LLVM < 0x0307
 505    options.NoFramePointerElim = true;
 506 #endif
 507 #endif
 508
 509    builder.setEngineKind(EngineKind::JIT)
 510           .setErrorStr(&Error)
 511           .setTargetOptions(options)
 512           .setOptLevel((CodeGenOpt::Level)OptLevel);
 513
 514    if (useMCJIT) {
 515 #if HAVE_LLVM < 0x0306
 516        builder.setUseMCJIT(true);
 517 #endif
 518 #ifdef _WIN32
 519        /*
 520         * MCJIT works on Windows, but currently only through ELF object format.
 521         *
 522         * XXX: We could use `LLVM_HOST_TRIPLE "-elf"` but LLVM_HOST_TRIPLE has
 523         * different strings for MinGW/MSVC, so better play it safe and be
 524         * explicit.
 525         */
 526 #  ifdef _WIN64
 527        LLVMSetTarget(M, "x86_64-pc-win32-elf");
 528 #  else
 529        LLVMSetTarget(M, "i686-pc-win32-elf");
 530 #  endif
 531 #endif
 532    }
 533
 534    llvm::SmallVector<std::string, 16> MAttrs;
 535
 536 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
 537 #if HAVE_LLVM >= 0x0400
 538    /* llvm-3.7+ implements sys::getHostCPUFeatures for x86,
 539     * which allows us to enable/disable code generation based
 540     * on the results of cpuid.
 541     */
 542    llvm::StringMap<bool> features;
 543    llvm::sys::getHostCPUFeatures(features);
 544
 545    for (StringMapIterator<bool> f = features.begin();
 546         f != features.end();
 547         ++f) {
 548       MAttrs.push_back(((*f).second ? "+" : "-") + (*f).first().str());
 549    }
 550 #else
 551    /*
 552     * We need to unset attributes because sometimes LLVM mistakenly assumes
 553     * certain features are present given the processor name.
 554     *
 555     * https://bugs.freedesktop.org/show_bug.cgi?id=92214
 556     * http://llvm.org/PR25021
 557     * http://llvm.org/PR19429
 558     * http://llvm.org/PR16721
 559     */
 560    MAttrs.push_back(util_cpu_caps.has_sse    ? "+sse"    : "-sse"   );
 561    MAttrs.push_back(util_cpu_caps.has_sse2   ? "+sse2"   : "-sse2"  );
 562    MAttrs.push_back(util_cpu_caps.has_sse3   ? "+sse3"   : "-sse3"  );
 563    MAttrs.push_back(util_cpu_caps.has_ssse3  ? "+ssse3"  : "-ssse3" );
 564 #if HAVE_LLVM >= 0x0304
 565    MAttrs.push_back(util_cpu_caps.has_sse4_1 ? "+sse4.1" : "-sse4.1");
 566 #else
 567    MAttrs.push_back(util_cpu_caps.has_sse4_1 ? "+sse41"  : "-sse41" );
 568 #endif
 569 #if HAVE_LLVM >= 0x0304
 570    MAttrs.push_back(util_cpu_caps.has_sse4_2 ? "+sse4.2" : "-sse4.2");
 571 #else
 572    MAttrs.push_back(util_cpu_caps.has_sse4_2 ? "+sse42"  : "-sse42" );
 573 #endif
 574    /*
 575     * AVX feature is not automatically detected from CPUID by the X86 target
 576     * yet, because the old (yet default) JIT engine is not capable of
 577     * emitting the opcodes. On newer llvm versions it is and at least some
 578     * versions (tested with 3.3) will emit avx opcodes without this anyway.
 579     */
 580    MAttrs.push_back(util_cpu_caps.has_avx  ? "+avx"  : "-avx");
 581    MAttrs.push_back(util_cpu_caps.has_f16c ? "+f16c" : "-f16c");
 582    if (HAVE_LLVM >= 0x0304) {
 583       MAttrs.push_back(util_cpu_caps.has_fma  ? "+fma"  : "-fma");
 584    } else {
 585       /*
 586        * The old JIT in LLVM 3.3 has a bug encoding llvm.fmuladd.f32 and
 587        * llvm.fmuladd.v2f32 intrinsics when FMA is available.
 588        */
 589       MAttrs.push_back("-fma");
 590    }
 591    MAttrs.push_back(util_cpu_caps.has_avx2 ? "+avx2" : "-avx2");
 592    /* disable avx512 and all subvariants */
 593 #if HAVE_LLVM >= 0x0304
 594    MAttrs.push_back("-avx512cd");
 595    MAttrs.push_back("-avx512er");
 596    MAttrs.push_back("-avx512f");
 597    MAttrs.push_back("-avx512pf");
 598 #endif
 599 #if HAVE_LLVM >= 0x0305
 600    MAttrs.push_back("-avx512bw");
 601    MAttrs.push_back("-avx512dq");
 602    MAttrs.push_back("-avx512vl");
 603 #endif
 604 #endif
 605 #endif
 606
 607 #if defined(PIPE_ARCH_PPC)
 608    MAttrs.push_back(util_cpu_caps.has_altivec ? "+altivec" : "-altivec");
 609 #if (HAVE_LLVM >= 0x0304)
 610 #if (HAVE_LLVM <= 0x0307) || (HAVE_LLVM == 0x0308 && MESA_LLVM_VERSION_PATCH == 0)
 611    /*
 612     * Make sure VSX instructions are disabled
 613     * See LLVM bug https://llvm.org/bugs/show_bug.cgi?id=25503#c7
 614     */
 615    if (util_cpu_caps.has_altivec) {
 616       MAttrs.push_back("-vsx");
 617    }
 618 #else
 619    /*
 620     * However, bug 25503 is fixed, by the same fix that fixed
 621     * bug 26775, in versions of LLVM later than 3.8 (starting with 3.8.1):
 622     * Make sure VSX instructions are ENABLED
 623     * See LLVM bug https://llvm.org/bugs/show_bug.cgi?id=26775
 624     */
 625    if (util_cpu_caps.has_altivec) {
 626       MAttrs.push_back("+vsx");
 627    }
 628 #endif
 629 #endif
 630 #endif
 631
 632    builder.setMAttrs(MAttrs);
 633
 634    if (gallivm_debug & (GALLIVM_DEBUG_IR | GALLIVM_DEBUG_ASM | GALLIVM_DEBUG_DUMP_BC)) {
 635       int n = MAttrs.size();
 636       if (n > 0) {
 637          debug_printf("llc -mattr option(s): ");
 638          for (int i = 0; i < n; i++)
 639             debug_printf("%s%s", MAttrs[i].c_str(), (i < n - 1) ? "," : "");
 640          debug_printf("\n");
 641       }
 642    }
 643
 644 #if HAVE_LLVM >= 0x0305
 645    StringRef MCPU = llvm::sys::getHostCPUName();
 646    /*
 647     * The cpu bits are no longer set automatically, so need to set mcpu manually.
 648     * Note that the MAttrs set above will be sort of ignored (since we should
 649     * not set any which would not be set by specifying the cpu anyway).
 650     * It ought to be safe though since getHostCPUName() should include bits
 651     * not only from the cpu but environment as well (for instance if it's safe
 652     * to use avx instructions which need OS support). According to
 653     * http://llvm.org/bugs/show_bug.cgi?id=19429 however if I understand this
 654     * right it may be necessary to specify older cpu (or disable mattrs) though
 655     * when not using MCJIT so no instructions are generated which the old JIT
 656     * can't handle. Not entirely sure if we really need to do anything yet.
 657     */
 658 #if defined(PIPE_ARCH_LITTLE_ENDIAN)  && defined(PIPE_ARCH_PPC_64)
 659    /*
 660     * Versions of LLVM prior to 4.0 lacked a table entry for "POWER8NVL",
 661     * resulting in (big-endian) "generic" being returned on
 662     * little-endian Power8NVL systems.  The result was that code that
 663     * attempted to load the least significant 32 bits of a 64-bit quantity
 664     * from memory loaded the wrong half.  This resulted in failures in some
 665     * Piglit tests, e.g.
 666     * .../arb_gpu_shader_fp64/execution/conversion/frag-conversion-explicit-double-uint
 667     */
 668    if (MCPU == "generic")
 669       MCPU = "pwr8";
 670 #endif
 671    builder.setMCPU(MCPU);
 672    if (gallivm_debug & (GALLIVM_DEBUG_IR | GALLIVM_DEBUG_ASM | GALLIVM_DEBUG_DUMP_BC)) {
 673       debug_printf("llc -mcpu option: %s\n", MCPU.str().c_str());
 674    }
 675 #endif
 676
 677    ShaderMemoryManager *MM = NULL;
 678    if (useMCJIT) {
 679        BaseMemoryManager* JMM = reinterpret_cast<BaseMemoryManager*>(CMM);
 680        MM = new ShaderMemoryManager(JMM);
 681        *OutCode = MM->getGeneratedCode();
 682
 683 #if HAVE_LLVM >= 0x0306
 684        builder.setMCJITMemoryManager(std::unique_ptr<RTDyldMemoryManager>(MM));
 685        MM = NULL; // ownership taken by std::unique_ptr
 686 #elif HAVE_LLVM > 0x0303
 687        builder.setMCJITMemoryManager(MM);
 688 #else
 689        builder.setJITMemoryManager(MM);
 690 #endif
 691    } else {
 692 #if HAVE_LLVM < 0x0306
 693        BaseMemoryManager* JMM = reinterpret_cast<BaseMemoryManager*>(CMM);
 694        MM = new ShaderMemoryManager(JMM);
 695        *OutCode = MM->getGeneratedCode();
 696
 697        builder.setJITMemoryManager(MM);
 698 #else
 699        assert(0);
 700 #endif
 701    }
 702
 703    ExecutionEngine *JIT;
 704
 705    JIT = builder.create();
 706 #if LLVM_USE_INTEL_JITEVENTS
 707    JITEventListener *JEL = JITEventListener::createIntelJITEventListener();
 708    JIT->RegisterJITEventListener(JEL);
 709 #endif
 710    if (JIT) {
 711       *OutJIT = wrap(JIT);
 712       return 0;
 713    }
 714    lp_free_generated_code(*OutCode);
 715    *OutCode = 0;
 716    delete MM;
 717    *OutError = strdup(Error.c_str());
 718    return 1;
 719 }
 720
 721
 722 extern "C"
 723 void
 724 lp_free_generated_code(struct lp_generated_code *code)
 725 {
 726    ShaderMemoryManager::freeGeneratedCode(code);
 727 }
 728
 729 extern "C"
 730 LLVMMCJITMemoryManagerRef
 731 lp_get_default_memory_manager()
 732 {
 733    BaseMemoryManager *mm;
 734 #if HAVE_LLVM < 0x0306
 735    mm = llvm::JITMemoryManager::CreateDefaultMemManager();
 736 #else
 737    mm = new llvm::SectionMemoryManager();
 738 #endif
 739    return reinterpret_cast<LLVMMCJITMemoryManagerRef>(mm);
 740 }
 741
 742 extern "C"
 743 void
 744 lp_free_memory_manager(LLVMMCJITMemoryManagerRef memorymgr)
 745 {
 746    delete reinterpret_cast<BaseMemoryManager*>(memorymgr);
 747 }
 748
 749 extern "C" LLVMValueRef
 750 lp_get_called_value(LLVMValueRef call)
 751 {
 752 #if HAVE_LLVM >= 0x0309
 753         return LLVMGetCalledValue(call);
 754 #elif HAVE_LLVM >= 0x0305
 755         return llvm::wrap(llvm::CallSite(llvm::unwrap<llvm::Instruction>(call)).getCalledValue());
 756 #else
 757         return NULL; /* radeonsi doesn't support so old LLVM. */
 758 #endif
 759 }
 760
 761 extern "C" bool
 762 lp_is_function(LLVMValueRef v)
 763 {
 764 #if HAVE_LLVM >= 0x0309
 765         return LLVMGetValueKind(v) == LLVMFunctionValueKind;
 766 #else
 767         return llvm::isa<llvm::Function>(llvm::unwrap(v));
 768 #endif
 769 }
 770
 771 extern "C" LLVMBuilderRef
 772 lp_create_builder(LLVMContextRef ctx, enum lp_float_mode float_mode)
 773 {
 774    LLVMBuilderRef builder = LLVMCreateBuilderInContext(ctx);
 775
 776 #if HAVE_LLVM >= 0x0308
 777    llvm::FastMathFlags flags;
 778
 779    switch (float_mode) {
 780    case LP_FLOAT_MODE_DEFAULT:
 781       break;
 782    case LP_FLOAT_MODE_NO_SIGNED_ZEROS_FP_MATH:
 783       flags.setNoSignedZeros();
 784       llvm::unwrap(builder)->setFastMathFlags(flags);
 785       break;
 786    case LP_FLOAT_MODE_UNSAFE_FP_MATH:
 787       flags.setUnsafeAlgebra();
 788       llvm::unwrap(builder)->setFastMathFlags(flags);
 789       break;
 790    }
 791 #endif
 792
 793    return builder;
 794 }