src/amd/common/ac_llvm_util.c

   1 /*
   2  * Copyright 2014 Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sub license, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  13  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  14  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  15  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  16  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  17  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  18  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  19  *
  20  * The above copyright notice and this permission notice (including the
  21  * next paragraph) shall be included in all copies or substantial portions
  22  * of the Software.
  23  *
  24  */
  25 /* based on pieces from si_pipe.c and radeon_llvm_emit.c */
  26 #include "ac_llvm_util.h"
  27 #include "ac_llvm_build.h"
  28 #include "util/bitscan.h"
  29 #include <llvm-c/Core.h>
  30 #include <llvm-c/Support.h>
  31 #include <llvm-c/Transforms/IPO.h>
  32 #include <llvm-c/Transforms/Scalar.h>
  33 #if HAVE_LLVM >= 0x0700
  34 #include <llvm-c/Transforms/Utils.h>
  35 #endif
  36 #include "c11/threads.h"
  37 #include "gallivm/lp_bld_misc.h"
  38 #include "util/u_math.h"
  39
  40 #include <assert.h>
  41 #include <stdio.h>
  42 #include <string.h>
  43
  44 static void ac_init_llvm_target()
  45 {
  46         LLVMInitializeAMDGPUTargetInfo();
  47         LLVMInitializeAMDGPUTarget();
  48         LLVMInitializeAMDGPUTargetMC();
  49         LLVMInitializeAMDGPUAsmPrinter();
  50
  51         /* For inline assembly. */
  52         LLVMInitializeAMDGPUAsmParser();
  53
  54         /* Workaround for bug in llvm 4.0 that causes image intrinsics
  55          * to disappear.
  56          * https://reviews.llvm.org/D26348
  57          *
  58          * "mesa" is the prefix for error messages.
  59          *
  60          * -global-isel-abort=2 is a no-op unless global isel has been enabled.
  61          * This option tells the backend to fall-back to SelectionDAG and print
  62          * a diagnostic message if global isel fails.
  63          */
  64         const char *argv[3] = { "mesa", "-simplifycfg-sink-common=false", "-global-isel-abort=2" };
  65         LLVMParseCommandLineOptions(3, argv, NULL);
  66 }
  67
  68 static once_flag ac_init_llvm_target_once_flag = ONCE_FLAG_INIT;
  69
  70 void ac_init_llvm_once(void)
  71 {
  72         call_once(&ac_init_llvm_target_once_flag, ac_init_llvm_target);
  73 }
  74
  75 static LLVMTargetRef ac_get_llvm_target(const char *triple)
  76 {
  77         LLVMTargetRef target = NULL;
  78         char *err_message = NULL;
  79
  80         if (LLVMGetTargetFromTriple(triple, &target, &err_message)) {
  81                 fprintf(stderr, "Cannot find target for triple %s ", triple);
  82                 if (err_message) {
  83                         fprintf(stderr, "%s\n", err_message);
  84                 }
  85                 LLVMDisposeMessage(err_message);
  86                 return NULL;
  87         }
  88         return target;
  89 }
  90
  91 const char *ac_get_llvm_processor_name(enum radeon_family family)
  92 {
  93         switch (family) {
  94         case CHIP_TAHITI:
  95                 return "tahiti";
  96         case CHIP_PITCAIRN:
  97                 return "pitcairn";
  98         case CHIP_VERDE:
  99                 return "verde";
 100         case CHIP_OLAND:
 101                 return "oland";
 102         case CHIP_HAINAN:
 103                 return "hainan";
 104         case CHIP_BONAIRE:
 105                 return "bonaire";
 106         case CHIP_KABINI:
 107                 return "kabini";
 108         case CHIP_KAVERI:
 109                 return "kaveri";
 110         case CHIP_HAWAII:
 111                 return "hawaii";
 112         case CHIP_MULLINS:
 113                 return "mullins";
 114         case CHIP_TONGA:
 115                 return "tonga";
 116         case CHIP_ICELAND:
 117                 return "iceland";
 118         case CHIP_CARRIZO:
 119                 return "carrizo";
 120         case CHIP_FIJI:
 121                 return "fiji";
 122         case CHIP_STONEY:
 123                 return "stoney";
 124         case CHIP_POLARIS10:
 125                 return "polaris10";
 126         case CHIP_POLARIS11:
 127         case CHIP_POLARIS12:
 128         case CHIP_VEGAM:
 129                 return "polaris11";
 130         case CHIP_VEGA10:
 131                 return "gfx900";
 132         case CHIP_RAVEN:
 133                 return "gfx902";
 134         case CHIP_VEGA12:
 135                 return HAVE_LLVM >= 0x0700 ? "gfx904" : "gfx902";
 136         case CHIP_VEGA20:
 137                 return HAVE_LLVM >= 0x0700 ? "gfx906" : "gfx902";
 138         case CHIP_RAVEN2:
 139                 return "gfx902"; /* TODO: use gfx909 when it's available */
 140         default:
 141                 return "";
 142         }
 143 }
 144
 145 static LLVMTargetMachineRef ac_create_target_machine(enum radeon_family family,
 146                                                      enum ac_target_machine_options tm_options,
 147                                                      LLVMCodeGenOptLevel level,
 148                                                      const char **out_triple)
 149 {
 150         assert(family >= CHIP_TAHITI);
 151         char features[256];
 152         const char *triple = (tm_options & AC_TM_SUPPORTS_SPILL) ? "amdgcn-mesa-mesa3d" : "amdgcn--";
 153         LLVMTargetRef target = ac_get_llvm_target(triple);
 154
 155         snprintf(features, sizeof(features),
 156                  "+DumpCode,+vgpr-spilling,-fp32-denormals,+fp64-denormals%s%s%s%s",
 157                  tm_options & AC_TM_SISCHED ? ",+si-scheduler" : "",
 158                  tm_options & AC_TM_FORCE_ENABLE_XNACK ? ",+xnack" : "",
 159                  tm_options & AC_TM_FORCE_DISABLE_XNACK ? ",-xnack" : "",
 160                  tm_options & AC_TM_PROMOTE_ALLOCA_TO_SCRATCH ? ",-promote-alloca" : "");
 161
 162         LLVMTargetMachineRef tm = LLVMCreateTargetMachine(
 163                                      target,
 164                                      triple,
 165                                      ac_get_llvm_processor_name(family),
 166                                      features,
 167                                      level,
 168                                      LLVMRelocDefault,
 169                                      LLVMCodeModelDefault);
 170
 171         if (out_triple)
 172                 *out_triple = triple;
 173         if (tm_options & AC_TM_ENABLE_GLOBAL_ISEL)
 174                 ac_enable_global_isel(tm);
 175         return tm;
 176 }
 177
 178 static LLVMPassManagerRef ac_create_passmgr(LLVMTargetLibraryInfoRef target_library_info,
 179                                             bool check_ir)
 180 {
 181         LLVMPassManagerRef passmgr = LLVMCreatePassManager();
 182         if (!passmgr)
 183                 return NULL;
 184
 185         if (target_library_info)
 186                 LLVMAddTargetLibraryInfo(target_library_info,
 187                                          passmgr);
 188
 189         if (check_ir)
 190                 LLVMAddVerifierPass(passmgr);
 191         LLVMAddAlwaysInlinerPass(passmgr);
 192         /* Normally, the pass manager runs all passes on one function before
 193          * moving onto another. Adding a barrier no-op pass forces the pass
 194          * manager to run the inliner on all functions first, which makes sure
 195          * that the following passes are only run on the remaining non-inline
 196          * function, so it removes useless work done on dead inline functions.
 197          */
 198         ac_llvm_add_barrier_noop_pass(passmgr);
 199         /* This pass should eliminate all the load and store instructions. */
 200         LLVMAddPromoteMemoryToRegisterPass(passmgr);
 201         LLVMAddScalarReplAggregatesPass(passmgr);
 202         LLVMAddLICMPass(passmgr);
 203         LLVMAddAggressiveDCEPass(passmgr);
 204         LLVMAddCFGSimplificationPass(passmgr);
 205         /* This is recommended by the instruction combining pass. */
 206         LLVMAddEarlyCSEMemSSAPass(passmgr);
 207         LLVMAddInstructionCombiningPass(passmgr);
 208         return passmgr;
 209 }
 210
 211 static const char *attr_to_str(enum ac_func_attr attr)
 212 {
 213    switch (attr) {
 214    case AC_FUNC_ATTR_ALWAYSINLINE: return "alwaysinline";
 215    case AC_FUNC_ATTR_INREG: return "inreg";
 216    case AC_FUNC_ATTR_NOALIAS: return "noalias";
 217    case AC_FUNC_ATTR_NOUNWIND: return "nounwind";
 218    case AC_FUNC_ATTR_READNONE: return "readnone";
 219    case AC_FUNC_ATTR_READONLY: return "readonly";
 220    case AC_FUNC_ATTR_WRITEONLY: return "writeonly";
 221    case AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY: return "inaccessiblememonly";
 222    case AC_FUNC_ATTR_CONVERGENT: return "convergent";
 223    default:
 224            fprintf(stderr, "Unhandled function attribute: %x\n", attr);
 225            return 0;
 226    }
 227 }
 228
 229 void
 230 ac_add_function_attr(LLVMContextRef ctx, LLVMValueRef function,
 231                      int attr_idx, enum ac_func_attr attr)
 232 {
 233    const char *attr_name = attr_to_str(attr);
 234    unsigned kind_id = LLVMGetEnumAttributeKindForName(attr_name,
 235                                                       strlen(attr_name));
 236    LLVMAttributeRef llvm_attr = LLVMCreateEnumAttribute(ctx, kind_id, 0);
 237
 238    if (LLVMIsAFunction(function))
 239       LLVMAddAttributeAtIndex(function, attr_idx, llvm_attr);
 240    else
 241       LLVMAddCallSiteAttribute(function, attr_idx, llvm_attr);
 242 }
 243
 244 void ac_add_func_attributes(LLVMContextRef ctx, LLVMValueRef function,
 245                             unsigned attrib_mask)
 246 {
 247         attrib_mask |= AC_FUNC_ATTR_NOUNWIND;
 248         attrib_mask &= ~AC_FUNC_ATTR_LEGACY;
 249
 250         while (attrib_mask) {
 251                 enum ac_func_attr attr = 1u << u_bit_scan(&attrib_mask);
 252                 ac_add_function_attr(ctx, function, -1, attr);
 253         }
 254 }
 255
 256 void
 257 ac_dump_module(LLVMModuleRef module)
 258 {
 259         char *str = LLVMPrintModuleToString(module);
 260         fprintf(stderr, "%s", str);
 261         LLVMDisposeMessage(str);
 262 }
 263
 264 void
 265 ac_llvm_add_target_dep_function_attr(LLVMValueRef F,
 266                                      const char *name, unsigned value)
 267 {
 268         char str[16];
 269
 270         snprintf(str, sizeof(str), "0x%x", value);
 271         LLVMAddTargetDependentFunctionAttr(F, name, str);
 272 }
 273
 274 unsigned
 275 ac_count_scratch_private_memory(LLVMValueRef function)
 276 {
 277         unsigned private_mem_vgprs = 0;
 278
 279         /* Process all LLVM instructions. */
 280         LLVMBasicBlockRef bb = LLVMGetFirstBasicBlock(function);
 281         while (bb) {
 282                 LLVMValueRef next = LLVMGetFirstInstruction(bb);
 283
 284                 while (next) {
 285                         LLVMValueRef inst = next;
 286                         next = LLVMGetNextInstruction(next);
 287
 288                         if (LLVMGetInstructionOpcode(inst) != LLVMAlloca)
 289                                 continue;
 290
 291                         LLVMTypeRef type = LLVMGetElementType(LLVMTypeOf(inst));
 292                         /* No idea why LLVM aligns allocas to 4 elements. */
 293                         unsigned alignment = LLVMGetAlignment(inst);
 294                         unsigned dw_size = align(ac_get_type_size(type) / 4, alignment);
 295                         private_mem_vgprs += dw_size;
 296                 }
 297                 bb = LLVMGetNextBasicBlock(bb);
 298         }
 299
 300         return private_mem_vgprs;
 301 }
 302
 303 bool
 304 ac_init_llvm_compiler(struct ac_llvm_compiler *compiler,
 305                       bool okay_to_leak_target_library_info,
 306                       enum radeon_family family,
 307                       enum ac_target_machine_options tm_options)
 308 {
 309         const char *triple;
 310         memset(compiler, 0, sizeof(*compiler));
 311
 312         compiler->tm = ac_create_target_machine(family, tm_options,
 313                                                 LLVMCodeGenLevelDefault,
 314                                                 &triple);
 315         if (!compiler->tm)
 316                 return false;
 317
 318         if (tm_options & AC_TM_CREATE_LOW_OPT) {
 319                 compiler->low_opt_tm =
 320                         ac_create_target_machine(family, tm_options,
 321                                                  LLVMCodeGenLevelLess, NULL);
 322                 if (!compiler->low_opt_tm)
 323                         goto fail;
 324         }
 325
 326         if (okay_to_leak_target_library_info || (HAVE_LLVM >= 0x0700)) {
 327                 compiler->target_library_info =
 328                         ac_create_target_library_info(triple);
 329                 if (!compiler->target_library_info)
 330                         goto fail;
 331         }
 332
 333         compiler->passmgr = ac_create_passmgr(compiler->target_library_info,
 334                                               tm_options & AC_TM_CHECK_IR);
 335         if (!compiler->passmgr)
 336                 goto fail;
 337
 338         return true;
 339 fail:
 340         ac_destroy_llvm_compiler(compiler);
 341         return false;
 342 }
 343
 344 void
 345 ac_destroy_llvm_compiler(struct ac_llvm_compiler *compiler)
 346 {
 347         if (compiler->passmgr)
 348                 LLVMDisposePassManager(compiler->passmgr);
 349 #if HAVE_LLVM >= 0x0700
 350         /* This crashes on LLVM 5.0 and 6.0 and Ubuntu 18.04, so leak it there. */
 351         if (compiler->target_library_info)
 352                 ac_dispose_target_library_info(compiler->target_library_info);
 353 #endif
 354         if (compiler->low_opt_tm)
 355                 LLVMDisposeTargetMachine(compiler->low_opt_tm);
 356         if (compiler->tm)
 357                 LLVMDisposeTargetMachine(compiler->tm);
 358 }