src/amd/llvm/ac_llvm_helper.cpp

   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
  26 #include <cstring>
  27
  28 #include "ac_binary.h"
  29 #include "ac_llvm_util.h"
  30 #include "ac_llvm_build.h"
  31
  32 #include "util/macros.h"
  33
  34 #include <llvm-c/Core.h>
  35 #include <llvm/Target/TargetMachine.h>
  36 #include <llvm/IR/IRBuilder.h>
  37 #include <llvm/Analysis/TargetLibraryInfo.h>
  38 #include <llvm/Transforms/IPO.h>
  39
  40 #include <llvm/IR/LegacyPassManager.h>
  41
  42 void ac_add_attr_dereferenceable(LLVMValueRef val, uint64_t bytes)
  43 {
  44    llvm::Argument *A = llvm::unwrap<llvm::Argument>(val);
  45    A->addAttr(llvm::Attribute::getWithDereferenceableBytes(A->getContext(), bytes));
  46 }
  47
  48 bool ac_is_sgpr_param(LLVMValueRef arg)
  49 {
  50         llvm::Argument *A = llvm::unwrap<llvm::Argument>(arg);
  51         llvm::AttributeList AS = A->getParent()->getAttributes();
  52         unsigned ArgNo = A->getArgNo();
  53         return AS.hasAttribute(ArgNo + 1, llvm::Attribute::InReg);
  54 }
  55
  56 LLVMValueRef ac_llvm_get_called_value(LLVMValueRef call)
  57 {
  58         return LLVMGetCalledValue(call);
  59 }
  60
  61 bool ac_llvm_is_function(LLVMValueRef v)
  62 {
  63         return LLVMGetValueKind(v) == LLVMFunctionValueKind;
  64 }
  65
  66 LLVMModuleRef ac_create_module(LLVMTargetMachineRef tm, LLVMContextRef ctx)
  67 {
  68    llvm::TargetMachine *TM = reinterpret_cast<llvm::TargetMachine*>(tm);
  69    LLVMModuleRef module = LLVMModuleCreateWithNameInContext("mesa-shader", ctx);
  70
  71    llvm::unwrap(module)->setTargetTriple(TM->getTargetTriple().getTriple());
  72    llvm::unwrap(module)->setDataLayout(TM->createDataLayout());
  73    return module;
  74 }
  75
  76 LLVMBuilderRef ac_create_builder(LLVMContextRef ctx,
  77                                  enum ac_float_mode float_mode)
  78 {
  79         LLVMBuilderRef builder = LLVMCreateBuilderInContext(ctx);
  80
  81         llvm::FastMathFlags flags;
  82
  83         switch (float_mode) {
  84         case AC_FLOAT_MODE_DEFAULT:
  85         case AC_FLOAT_MODE_DENORM_FLUSH_TO_ZERO:
  86                 break;
  87         case AC_FLOAT_MODE_NO_SIGNED_ZEROS_FP_MATH:
  88                 flags.setNoSignedZeros();
  89                 llvm::unwrap(builder)->setFastMathFlags(flags);
  90                 break;
  91         case AC_FLOAT_MODE_UNSAFE_FP_MATH:
  92                 flags.setFast();
  93                 llvm::unwrap(builder)->setFastMathFlags(flags);
  94                 break;
  95         }
  96
  97         return builder;
  98 }
  99
 100 LLVMTargetLibraryInfoRef
 101 ac_create_target_library_info(const char *triple)
 102 {
 103         return reinterpret_cast<LLVMTargetLibraryInfoRef>(new llvm::TargetLibraryInfoImpl(llvm::Triple(triple)));
 104 }
 105
 106 void
 107 ac_dispose_target_library_info(LLVMTargetLibraryInfoRef library_info)
 108 {
 109         delete reinterpret_cast<llvm::TargetLibraryInfoImpl *>(library_info);
 110 }
 111
 112 /* Implementation of raw_pwrite_stream that works on malloc()ed memory for
 113  * better compatibility with C code. */
 114 struct raw_memory_ostream : public llvm::raw_pwrite_stream {
 115         char *buffer;
 116         size_t written;
 117         size_t bufsize;
 118
 119         raw_memory_ostream()
 120         {
 121                 buffer = NULL;
 122                 written = 0;
 123                 bufsize = 0;
 124                 SetUnbuffered();
 125         }
 126
 127         ~raw_memory_ostream()
 128         {
 129                 free(buffer);
 130         }
 131
 132         void clear()
 133         {
 134                 written = 0;
 135         }
 136
 137         void take(char *&out_buffer, size_t &out_size)
 138         {
 139                 out_buffer = buffer;
 140                 out_size = written;
 141                 buffer = NULL;
 142                 written = 0;
 143                 bufsize = 0;
 144         }
 145
 146         void flush() = delete;
 147
 148         void write_impl(const char *ptr, size_t size) override
 149         {
 150                 if (unlikely(written + size < written))
 151                         abort();
 152                 if (written + size > bufsize) {
 153                         bufsize = MAX3(1024, written + size, bufsize / 3 * 4);
 154                         buffer = (char *)realloc(buffer, bufsize);
 155                         if (!buffer) {
 156                                 fprintf(stderr, "amd: out of memory allocating ELF buffer\n");
 157                                 abort();
 158                         }
 159                 }
 160                 memcpy(buffer + written, ptr, size);
 161                 written += size;
 162         }
 163
 164         void pwrite_impl(const char *ptr, size_t size, uint64_t offset) override
 165         {
 166                 assert(offset == (size_t)offset &&
 167                        offset + size >= offset && offset + size <= written);
 168                 memcpy(buffer + offset, ptr, size);
 169         }
 170
 171         uint64_t current_pos() const override
 172         {
 173                 return written;
 174         }
 175 };
 176
 177 /* The LLVM compiler is represented as a pass manager containing passes for
 178  * optimizations, instruction selection, and code generation.
 179  */
 180 struct ac_compiler_passes {
 181         raw_memory_ostream ostream; /* ELF shader binary stream */
 182         llvm::legacy::PassManager passmgr; /* list of passes */
 183 };
 184
 185 struct ac_compiler_passes *ac_create_llvm_passes(LLVMTargetMachineRef tm)
 186 {
 187         struct ac_compiler_passes *p = new ac_compiler_passes();
 188         if (!p)
 189                 return NULL;
 190
 191         llvm::TargetMachine *TM = reinterpret_cast<llvm::TargetMachine*>(tm);
 192
 193         if (TM->addPassesToEmitFile(p->passmgr, p->ostream,
 194                                     nullptr,
 195 #if LLVM_VERSION_MAJOR >= 10
 196                                     llvm::CGFT_ObjectFile)) {
 197 #else
 198                                     llvm::TargetMachine::CGFT_ObjectFile)) {
 199 #endif
 200                 fprintf(stderr, "amd: TargetMachine can't emit a file of this type!\n");
 201                 delete p;
 202                 return NULL;
 203         }
 204         return p;
 205 }
 206
 207 void ac_destroy_llvm_passes(struct ac_compiler_passes *p)
 208 {
 209         delete p;
 210 }
 211
 212 /* This returns false on failure. */
 213 bool ac_compile_module_to_elf(struct ac_compiler_passes *p, LLVMModuleRef module,
 214                               char **pelf_buffer, size_t *pelf_size)
 215 {
 216         p->passmgr.run(*llvm::unwrap(module));
 217         p->ostream.take(*pelf_buffer, *pelf_size);
 218         return true;
 219 }
 220
 221 void ac_llvm_add_barrier_noop_pass(LLVMPassManagerRef passmgr)
 222 {
 223         llvm::unwrap(passmgr)->add(llvm::createBarrierNoopPass());
 224 }
 225
 226 void ac_enable_global_isel(LLVMTargetMachineRef tm)
 227 {
 228   reinterpret_cast<llvm::TargetMachine*>(tm)->setGlobalISel(true);
 229 }
 230
 231 LLVMValueRef ac_build_atomic_rmw(struct ac_llvm_context *ctx, LLVMAtomicRMWBinOp op,
 232                                  LLVMValueRef ptr, LLVMValueRef val,
 233                                  const char *sync_scope) {
 234         llvm::AtomicRMWInst::BinOp binop;
 235         switch (op) {
 236         case LLVMAtomicRMWBinOpXchg:
 237                 binop = llvm::AtomicRMWInst::Xchg;
 238                 break;
 239         case LLVMAtomicRMWBinOpAdd:
 240                 binop = llvm::AtomicRMWInst::Add;
 241                 break;
 242         case LLVMAtomicRMWBinOpSub:
 243                 binop = llvm::AtomicRMWInst::Sub;
 244                 break;
 245         case LLVMAtomicRMWBinOpAnd:
 246                 binop = llvm::AtomicRMWInst::And;
 247                 break;
 248         case LLVMAtomicRMWBinOpNand:
 249                 binop = llvm::AtomicRMWInst::Nand;
 250                 break;
 251         case LLVMAtomicRMWBinOpOr:
 252                 binop = llvm::AtomicRMWInst::Or;
 253                 break;
 254         case LLVMAtomicRMWBinOpXor:
 255                 binop = llvm::AtomicRMWInst::Xor;
 256                 break;
 257         case LLVMAtomicRMWBinOpMax:
 258                 binop = llvm::AtomicRMWInst::Max;
 259                 break;
 260         case LLVMAtomicRMWBinOpMin:
 261                 binop = llvm::AtomicRMWInst::Min;
 262                 break;
 263         case LLVMAtomicRMWBinOpUMax:
 264                 binop = llvm::AtomicRMWInst::UMax;
 265                 break;
 266         case LLVMAtomicRMWBinOpUMin:
 267                 binop = llvm::AtomicRMWInst::UMin;
 268                 break;
 269         default:
 270                 unreachable(!"invalid LLVMAtomicRMWBinOp");
 271            break;
 272         }
 273         unsigned SSID = llvm::unwrap(ctx->context)->getOrInsertSyncScopeID(sync_scope);
 274         return llvm::wrap(llvm::unwrap(ctx->builder)->CreateAtomicRMW(
 275                 binop, llvm::unwrap(ptr), llvm::unwrap(val),
 276                 llvm::AtomicOrdering::SequentiallyConsistent, SSID));
 277 }
 278
 279 LLVMValueRef ac_build_atomic_cmp_xchg(struct ac_llvm_context *ctx, LLVMValueRef ptr,
 280                                       LLVMValueRef cmp, LLVMValueRef val,
 281                                       const char *sync_scope) {
 282         unsigned SSID = llvm::unwrap(ctx->context)->getOrInsertSyncScopeID(sync_scope);
 283         return llvm::wrap(llvm::unwrap(ctx->builder)->CreateAtomicCmpXchg(
 284                           llvm::unwrap(ptr), llvm::unwrap(cmp), llvm::unwrap(val),
 285                           llvm::AtomicOrdering::SequentiallyConsistent,
 286                           llvm::AtomicOrdering::SequentiallyConsistent, SSID));
 287 }