src/gallium/auxiliary/gallivm/lp_bld_init.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 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 above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include "pipe/p_config.h"
  30 #include "pipe/p_compiler.h"
  31 #include "util/u_cpu_detect.h"
  32 #include "util/u_debug.h"
  33 #include "util/u_memory.h"
  34 #include "util/u_simple_list.h"
  35 #include "os/os_time.h"
  36 #include "lp_bld.h"
  37 #include "lp_bld_debug.h"
  38 #include "lp_bld_misc.h"
  39 #include "lp_bld_init.h"
  40
  41 #include <llvm-c/Analysis.h>
  42 #include <llvm-c/Transforms/Scalar.h>
  43 #include <llvm-c/BitWriter.h>
  44
  45
  46 /* Only MCJIT is available as of LLVM SVN r216982 */
  47 #if HAVE_LLVM >= 0x0306
  48 #  define USE_MCJIT 1
  49 #elif defined(PIPE_ARCH_PPC_64) || defined(PIPE_ARCH_S390) || defined(PIPE_ARCH_ARM) || defined(PIPE_ARCH_AARCH64)
  50 #  define USE_MCJIT 1
  51 #else
  52 #  define USE_MCJIT 0
  53 #endif
  54
  55 #if USE_MCJIT
  56 void LLVMLinkInMCJIT();
  57 #endif
  58
  59 #ifdef DEBUG
  60 unsigned gallivm_debug = 0;
  61
  62 static const struct debug_named_value lp_bld_debug_flags[] = {
  63    { "tgsi",   GALLIVM_DEBUG_TGSI, NULL },
  64    { "ir",     GALLIVM_DEBUG_IR, NULL },
  65    { "asm",    GALLIVM_DEBUG_ASM, NULL },
  66    { "nopt",   GALLIVM_DEBUG_NO_OPT, NULL },
  67    { "perf",   GALLIVM_DEBUG_PERF, NULL },
  68    { "no_brilinear", GALLIVM_DEBUG_NO_BRILINEAR, NULL },
  69    { "no_rho_approx", GALLIVM_DEBUG_NO_RHO_APPROX, NULL },
  70    { "no_quad_lod", GALLIVM_DEBUG_NO_QUAD_LOD, NULL },
  71    { "gc",     GALLIVM_DEBUG_GC, NULL },
  72    DEBUG_NAMED_VALUE_END
  73 };
  74
  75 DEBUG_GET_ONCE_FLAGS_OPTION(gallivm_debug, "GALLIVM_DEBUG", lp_bld_debug_flags, 0)
  76 #endif
  77
  78
  79 static boolean gallivm_initialized = FALSE;
  80
  81 unsigned lp_native_vector_width;
  82
  83
  84 /*
  85  * Optimization values are:
  86  * - 0: None (-O0)
  87  * - 1: Less (-O1)
  88  * - 2: Default (-O2, -Os)
  89  * - 3: Aggressive (-O3)
  90  *
  91  * See also CodeGenOpt::Level in llvm/Target/TargetMachine.h
  92  */
  93 enum LLVM_CodeGenOpt_Level {
  94    None,        // -O0
  95    Less,        // -O1
  96    Default,     // -O2, -Os
  97    Aggressive   // -O3
  98 };
  99
 100
 101 /**
 102  * Create the LLVM (optimization) pass manager and install
 103  * relevant optimization passes.
 104  * \return  TRUE for success, FALSE for failure
 105  */
 106 static boolean
 107 create_pass_manager(struct gallivm_state *gallivm)
 108 {
 109    char *td_str;
 110    assert(!gallivm->passmgr);
 111    assert(gallivm->target);
 112
 113    gallivm->passmgr = LLVMCreateFunctionPassManagerForModule(gallivm->module);
 114    if (!gallivm->passmgr)
 115       return FALSE;
 116
 117    // Old versions of LLVM get the DataLayout from the pass manager.
 118    LLVMAddTargetData(gallivm->target, gallivm->passmgr);
 119
 120    // New ones from the Module.
 121    td_str = LLVMCopyStringRepOfTargetData(gallivm->target);
 122    LLVMSetDataLayout(gallivm->module, td_str);
 123    free(td_str);
 124
 125    if ((gallivm_debug & GALLIVM_DEBUG_NO_OPT) == 0) {
 126       /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
 127        * but there are more on SVN.
 128        * TODO: Add more passes.
 129        */
 130       LLVMAddScalarReplAggregatesPass(gallivm->passmgr);
 131       LLVMAddLICMPass(gallivm->passmgr);
 132       LLVMAddCFGSimplificationPass(gallivm->passmgr);
 133       LLVMAddReassociatePass(gallivm->passmgr);
 134       LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr);
 135       LLVMAddConstantPropagationPass(gallivm->passmgr);
 136       LLVMAddInstructionCombiningPass(gallivm->passmgr);
 137       LLVMAddGVNPass(gallivm->passmgr);
 138    }
 139    else {
 140       /* We need at least this pass to prevent the backends to fail in
 141        * unexpected ways.
 142        */
 143       LLVMAddPromoteMemoryToRegisterPass(gallivm->passmgr);
 144    }
 145
 146    return TRUE;
 147 }
 148
 149
 150 /**
 151  * Free gallivm object's LLVM allocations, but not any generated code
 152  * nor the gallivm object itself.
 153  */
 154 void
 155 gallivm_free_ir(struct gallivm_state *gallivm)
 156 {
 157    if (gallivm->passmgr) {
 158       LLVMDisposePassManager(gallivm->passmgr);
 159    }
 160
 161    if (gallivm->engine) {
 162       /* This will already destroy any associated module */
 163       LLVMDisposeExecutionEngine(gallivm->engine);
 164    } else if (gallivm->module) {
 165       LLVMDisposeModule(gallivm->module);
 166    }
 167
 168 #if !USE_MCJIT
 169    /* Don't free the TargetData, it's owned by the exec engine */
 170 #else
 171    if (gallivm->target) {
 172       LLVMDisposeTargetData(gallivm->target);
 173    }
 174 #endif
 175
 176    if (gallivm->builder)
 177       LLVMDisposeBuilder(gallivm->builder);
 178
 179    /* The LLVMContext should be owned by the parent of gallivm. */
 180
 181    gallivm->engine = NULL;
 182    gallivm->target = NULL;
 183    gallivm->module = NULL;
 184    gallivm->passmgr = NULL;
 185    gallivm->context = NULL;
 186    gallivm->builder = NULL;
 187 }
 188
 189
 190 /**
 191  * Free LLVM-generated code.  Should be done AFTER gallivm_free_ir().
 192  */
 193 static void
 194 gallivm_free_code(struct gallivm_state *gallivm)
 195 {
 196    assert(!gallivm->module);
 197    assert(!gallivm->engine);
 198    lp_free_generated_code(gallivm->code);
 199    gallivm->code = NULL;
 200    lp_free_memory_manager(gallivm->memorymgr);
 201    gallivm->memorymgr = NULL;
 202 }
 203
 204
 205 static boolean
 206 init_gallivm_engine(struct gallivm_state *gallivm)
 207 {
 208    if (1) {
 209       enum LLVM_CodeGenOpt_Level optlevel;
 210       char *error = NULL;
 211       int ret;
 212
 213       if (gallivm_debug & GALLIVM_DEBUG_NO_OPT) {
 214          optlevel = None;
 215       }
 216       else {
 217          optlevel = Default;
 218       }
 219
 220       ret = lp_build_create_jit_compiler_for_module(&gallivm->engine,
 221                                                     &gallivm->code,
 222                                                     gallivm->module,
 223                                                     gallivm->memorymgr,
 224                                                     (unsigned) optlevel,
 225                                                     USE_MCJIT,
 226                                                     &error);
 227       if (ret) {
 228          _debug_printf("%s\n", error);
 229          LLVMDisposeMessage(error);
 230          goto fail;
 231       }
 232    }
 233
 234 #if !USE_MCJIT
 235    gallivm->target = LLVMGetExecutionEngineTargetData(gallivm->engine);
 236    if (!gallivm->target)
 237       goto fail;
 238 #else
 239    if (0) {
 240        /*
 241         * Dump the data layout strings.
 242         */
 243
 244        LLVMTargetDataRef target = LLVMGetExecutionEngineTargetData(gallivm->engine);
 245        char *data_layout;
 246        char *engine_data_layout;
 247
 248        data_layout = LLVMCopyStringRepOfTargetData(gallivm->target);
 249        engine_data_layout = LLVMCopyStringRepOfTargetData(target);
 250
 251        if (1) {
 252           debug_printf("module target data = %s\n", data_layout);
 253           debug_printf("engine target data = %s\n", engine_data_layout);
 254        }
 255
 256        free(data_layout);
 257        free(engine_data_layout);
 258    }
 259 #endif
 260
 261    return TRUE;
 262
 263 fail:
 264    return FALSE;
 265 }
 266
 267
 268 /**
 269  * Allocate gallivm LLVM objects.
 270  * \return  TRUE for success, FALSE for failure
 271  */
 272 static boolean
 273 init_gallivm_state(struct gallivm_state *gallivm, const char *name,
 274                    LLVMContextRef context)
 275 {
 276    assert(!gallivm->context);
 277    assert(!gallivm->module);
 278
 279    if (!lp_build_init())
 280       return FALSE;
 281
 282    gallivm->context = context;
 283
 284    if (!gallivm->context)
 285       goto fail;
 286
 287    gallivm->module = LLVMModuleCreateWithNameInContext(name,
 288                                                        gallivm->context);
 289    if (!gallivm->module)
 290       goto fail;
 291
 292    gallivm->builder = LLVMCreateBuilderInContext(gallivm->context);
 293    if (!gallivm->builder)
 294       goto fail;
 295
 296    gallivm->memorymgr = lp_get_default_memory_manager();
 297    if (!gallivm->memorymgr)
 298       goto fail;
 299
 300    /* FIXME: MC-JIT only allows compiling one module at a time, and it must be
 301     * complete when MC-JIT is created. So defer the MC-JIT engine creation for
 302     * now.
 303     */
 304 #if !USE_MCJIT
 305    if (!init_gallivm_engine(gallivm)) {
 306       goto fail;
 307    }
 308 #else
 309    /*
 310     * MC-JIT engine compiles the module immediately on creation, so we can't
 311     * obtain the target data from it.  Instead we create a target data layout
 312     * from a string.
 313     *
 314     * The produced layout strings are not precisely the same, but should make
 315     * no difference for the kind of optimization passes we run.
 316     *
 317     * For reference this is the layout string on x64:
 318     *
 319     *   e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-f128:128:128-n8:16:32:64
 320     *
 321     * See also:
 322     * - http://llvm.org/docs/LangRef.html#datalayout
 323     */
 324
 325    {
 326       const unsigned pointer_size = 8 * sizeof(void *);
 327       char layout[512];
 328       util_snprintf(layout, sizeof layout, "%c-p:%u:%u:%u-i64:64:64-a0:0:%u-s0:%u:%u",
 329 #ifdef PIPE_ARCH_LITTLE_ENDIAN
 330                     'e', // little endian
 331 #else
 332                     'E', // big endian
 333 #endif
 334                     pointer_size, pointer_size, pointer_size, // pointer size, abi alignment, preferred alignment
 335                     pointer_size, // aggregate preferred alignment
 336                     pointer_size, pointer_size); // stack objects abi alignment, preferred alignment
 337
 338       gallivm->target = LLVMCreateTargetData(layout);
 339       if (!gallivm->target) {
 340          return FALSE;
 341       }
 342    }
 343 #endif
 344
 345    if (!create_pass_manager(gallivm))
 346       goto fail;
 347
 348    return TRUE;
 349
 350 fail:
 351    gallivm_free_ir(gallivm);
 352    gallivm_free_code(gallivm);
 353    return FALSE;
 354 }
 355
 356
 357 boolean
 358 lp_build_init(void)
 359 {
 360    if (gallivm_initialized)
 361       return TRUE;
 362
 363 #ifdef DEBUG
 364    gallivm_debug = debug_get_option_gallivm_debug();
 365 #endif
 366
 367    lp_set_target_options();
 368
 369 #if USE_MCJIT
 370    LLVMLinkInMCJIT();
 371 #else
 372    LLVMLinkInJIT();
 373 #endif
 374
 375    util_cpu_detect();
 376
 377    /* AMD Bulldozer AVX's throughput is the same as SSE2; and because using
 378     * 8-wide vector needs more floating ops than 4-wide (due to padding), it is
 379     * actually more efficient to use 4-wide vectors on this processor.
 380     *
 381     * See also:
 382     * - http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/2
 383     */
 384    if (util_cpu_caps.has_avx &&
 385        util_cpu_caps.has_intel) {
 386       lp_native_vector_width = 256;
 387    } else {
 388       /* Leave it at 128, even when no SIMD extensions are available.
 389        * Really needs to be a multiple of 128 so can fit 4 floats.
 390        */
 391       lp_native_vector_width = 128;
 392    }
 393
 394    lp_native_vector_width = debug_get_num_option("LP_NATIVE_VECTOR_WIDTH",
 395                                                  lp_native_vector_width);
 396
 397    if (lp_native_vector_width <= 128) {
 398       /* Hide AVX support, as often LLVM AVX intrinsics are only guarded by
 399        * "util_cpu_caps.has_avx" predicate, and lack the
 400        * "lp_native_vector_width > 128" predicate. And also to ensure a more
 401        * consistent behavior, allowing one to test SSE2 on AVX machines.
 402        * XXX: should not play games with util_cpu_caps directly as it might
 403        * get used for other things outside llvm too.
 404        */
 405       util_cpu_caps.has_avx = 0;
 406       util_cpu_caps.has_avx2 = 0;
 407    }
 408
 409 #ifdef PIPE_ARCH_PPC_64
 410    /* Set the NJ bit in VSCR to 0 so denormalized values are handled as
 411     * specified by IEEE standard (PowerISA 2.06 - Section 6.3). This guarantees
 412     * that some rounding and half-float to float handling does not round
 413     * incorrectly to 0.
 414     * XXX: should eventually follow same logic on all platforms.
 415     * Right now denorms get explicitly disabled (but elsewhere) for x86,
 416     * whereas ppc64 explicitly enables them...
 417     */
 418    if (util_cpu_caps.has_altivec) {
 419       unsigned short mask[] = { 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
 420                                 0xFFFF, 0xFFFF, 0xFFFE, 0xFFFF };
 421       __asm (
 422         "mfvscr %%v1\n"
 423         "vand   %0,%%v1,%0\n"
 424         "mtvscr %0"
 425         :
 426         : "r" (*mask)
 427       );
 428    }
 429 #endif
 430
 431    gallivm_initialized = TRUE;
 432
 433 #if 0
 434    /* For simulating less capable machines */
 435    util_cpu_caps.has_sse3 = 0;
 436    util_cpu_caps.has_ssse3 = 0;
 437    util_cpu_caps.has_sse4_1 = 0;
 438    util_cpu_caps.has_avx = 0;
 439    util_cpu_caps.has_f16c = 0;
 440 #endif
 441
 442    return TRUE;
 443 }
 444
 445
 446
 447 /**
 448  * Create a new gallivm_state object.
 449  */
 450 struct gallivm_state *
 451 gallivm_create(const char *name, LLVMContextRef context)
 452 {
 453    struct gallivm_state *gallivm;
 454
 455    gallivm = CALLOC_STRUCT(gallivm_state);
 456    if (gallivm) {
 457       if (!init_gallivm_state(gallivm, name, context)) {
 458          FREE(gallivm);
 459          gallivm = NULL;
 460       }
 461    }
 462
 463    return gallivm;
 464 }
 465
 466
 467 /**
 468  * Destroy a gallivm_state object.
 469  */
 470 void
 471 gallivm_destroy(struct gallivm_state *gallivm)
 472 {
 473    gallivm_free_ir(gallivm);
 474    gallivm_free_code(gallivm);
 475    FREE(gallivm);
 476 }
 477
 478
 479 /**
 480  * Validate a function.
 481  * Verification is only done with debug builds.
 482  */
 483 void
 484 gallivm_verify_function(struct gallivm_state *gallivm,
 485                         LLVMValueRef func)
 486 {
 487    /* Verify the LLVM IR.  If invalid, dump and abort */
 488 #ifdef DEBUG
 489    if (LLVMVerifyFunction(func, LLVMPrintMessageAction)) {
 490       lp_debug_dump_value(func);
 491       assert(0);
 492       return;
 493    }
 494 #endif
 495
 496    if (gallivm_debug & GALLIVM_DEBUG_IR) {
 497       /* Print the LLVM IR to stderr */
 498       lp_debug_dump_value(func);
 499       debug_printf("\n");
 500    }
 501 }
 502
 503
 504 /**
 505  * Compile a module.
 506  * This does IR optimization on all functions in the module.
 507  */
 508 void
 509 gallivm_compile_module(struct gallivm_state *gallivm)
 510 {
 511    LLVMValueRef func;
 512    int64_t time_begin;
 513
 514    assert(!gallivm->compiled);
 515
 516    if (gallivm->builder) {
 517       LLVMDisposeBuilder(gallivm->builder);
 518       gallivm->builder = NULL;
 519    }
 520
 521    if (gallivm_debug & GALLIVM_DEBUG_PERF)
 522       time_begin = os_time_get();
 523
 524    /* Run optimization passes */
 525    LLVMInitializeFunctionPassManager(gallivm->passmgr);
 526    func = LLVMGetFirstFunction(gallivm->module);
 527    while (func) {
 528       if (0) {
 529          debug_printf("optimizing func %s...\n", LLVMGetValueName(func));
 530       }
 531       LLVMRunFunctionPassManager(gallivm->passmgr, func);
 532       func = LLVMGetNextFunction(func);
 533    }
 534    LLVMFinalizeFunctionPassManager(gallivm->passmgr);
 535
 536    if (gallivm_debug & GALLIVM_DEBUG_PERF) {
 537       int64_t time_end = os_time_get();
 538       int time_msec = (int)(time_end - time_begin) / 1000;
 539       debug_printf("optimizing module %s took %d msec\n",
 540                    lp_get_module_id(gallivm->module), time_msec);
 541    }
 542
 543    /* Dump byte code to a file */
 544    if (0) {
 545       LLVMWriteBitcodeToFile(gallivm->module, "llvmpipe.bc");
 546       debug_printf("llvmpipe.bc written\n");
 547       debug_printf("Invoke as \"llc -o - llvmpipe.bc\"\n");
 548    }
 549
 550 #if USE_MCJIT
 551    assert(!gallivm->engine);
 552    if (!init_gallivm_engine(gallivm)) {
 553       assert(0);
 554    }
 555 #endif
 556    assert(gallivm->engine);
 557
 558    ++gallivm->compiled;
 559 }
 560
 561
 562
 563 func_pointer
 564 gallivm_jit_function(struct gallivm_state *gallivm,
 565                      LLVMValueRef func)
 566 {
 567    void *code;
 568    func_pointer jit_func;
 569
 570    assert(gallivm->compiled);
 571    assert(gallivm->engine);
 572
 573    code = LLVMGetPointerToGlobal(gallivm->engine, func);
 574    assert(code);
 575    jit_func = pointer_to_func(code);
 576
 577    if (gallivm_debug & GALLIVM_DEBUG_ASM) {
 578       lp_disassemble(func, code);
 579    }
 580
 581 #if defined(PROFILE)
 582    lp_profile(func, code);
 583 #endif
 584
 585    return jit_func;
 586 }