codegen_nvptx.cc 9.47 KB
Newer Older
1 2 3 4 5 6 7 8
/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
9
 *
10
 *   http://www.apache.org/licenses/LICENSE-2.0
11
 *
12 13 14 15 16 17 18 19
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
 * KIND, either express or implied.  See the License for the
 * specific language governing permissions and limitations
 * under the License.
 */

20 21 22 23 24 25 26
/*!
 * \file codegen_nvptx.cc
 * \brief NVPTX code generator.
 */
#ifdef TVM_LLVM_VERSION

#include <tvm/runtime/device_api.h>
27
#include "codegen_llvm.h"
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
#include "../build_common.h"
#include "../../pass/ir_util.h"
#include "../../runtime/cuda/cuda_module.h"

namespace tvm {
namespace codegen {

// NVPTX code generator.
class CodeGenNVPTX : public CodeGenLLVM {
 public:
  void AddFunction(const LoweredFunc& f) final {
    // add function as void return value
    CodeGenLLVM::AddFunctionInternal(f, true);
    // annotate as kernel function
    module_->getOrInsertNamedMetadata("nvvm.annotations")
        ->addOperand(llvm::MDNode::get(*ctx_, {
              llvm::ValueAsMetadata::get(function_),
              llvm::MDString::get(*ctx_, "kernel"),
              llvm::ValueAsMetadata::get(ConstInt32(1)) }));
  }

  void VisitStmt_(const Allocate* op) final {
    CHECK(!is_zero(op->condition));
    llvm::Value* buf = nullptr;
    if (op->new_expr.defined()) {
      CHECK_EQ(op->free_function, "nop");
      buf = MakeValue(op->new_expr);
    } else {
      int32_t constant_size = op->constant_allocation_size();
      CHECK_GT(constant_size, 0)
          << "Can only handle constant size stack allocation in GPU";
      StorageInfo& info = alloc_storage_info_[op->buffer_var.get()];
      if (constant_size % 4 == 0 && info.alignment == 0) {
61
        info.alignment = GetTempAllocaAlignment(op->dtype, constant_size);
62 63 64 65 66
      }
      // maximum necessary alignment in the NV devices
      if (info.alignment > 16) {
        info.alignment = 16;
      }
67
      if (info.scope.rank == runtime::StorageRank::kLocal) {
68 69
        // const int local_address_space = 5;
        // TODO(tqchen): for higher version of LLVM, local address space can be set.
70 71
        llvm::AllocaInst* alloca = WithFunctionEntry([&]() {
            return builder_->CreateAlloca(
72
                LLVMType(op->dtype), ConstInt32(constant_size));
73
          });
74
        if (alloca->getAlignment() < static_cast<uint32_t>(info.alignment)) {
75 76 77
#if TVM_LLVM_VERSION >= 100
          alloca->setAlignment(llvm::Align(info.alignment));
#else
78
          alloca->setAlignment(info.alignment);
79
#endif
80 81 82
        }
        buf = alloca;
      } else {
83
        CHECK(info.scope.rank == runtime::StorageRank::kShared)
84 85 86
            << "Can only allocate shared or local memory inside kernel";
        // Shared memory: address space  == 3
        const unsigned shared_address_space = 3;
87
        llvm::Type* type = llvm::ArrayType::get(LLVMType(op->dtype), constant_size);
88 89 90 91
        // Allocate shared memory in global, address_space = 3
        llvm::GlobalVariable *global = new llvm::GlobalVariable(
            *module_, type, false, llvm::GlobalValue::PrivateLinkage, 0, ".shared",
            nullptr, llvm::GlobalValue::NotThreadLocal, shared_address_space);
92 93 94
#if TVM_LLVM_VERSION >= 100
        global->setAlignment(llvm::Align(info.alignment));
#else
95
        global->setAlignment(info.alignment);
96
#endif
97 98 99 100
        buf = global;
      }
    }
    buf = builder_->CreatePointerCast(
101
        buf, LLVMType(op->dtype)->getPointerTo(
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
            buf->getType()->getPointerAddressSpace()));
    CHECK(!var_map_.count(op->buffer_var.get()));
    var_map_[op->buffer_var.get()] = buf;
    this->VisitStmt(op->body);
  }

  // Return the thread index via intrinsics.
  llvm::Value* GetThreadIndex(const IterVar& iv) final {
    runtime::ThreadScope ts = runtime::ThreadScope::make(iv->thread_tag);
    llvm::Intrinsic::ID intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_tid_x;
    if (ts.rank == 1) {
      switch (ts.dim_index) {
        case 0: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_tid_x; break;
        case 1: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_tid_y; break;
        case 2: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_tid_z; break;
        default: LOG(FATAL) << "unknown thread idx";
      }
    } else {
      CHECK_EQ(ts.rank, 0);
      switch (ts.dim_index) {
        case 0: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_ctaid_x; break;
        case 1: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_ctaid_y; break;
        case 2: intrin_id = ::llvm::Intrinsic::nvvm_read_ptx_sreg_ctaid_z; break;
        default: LOG(FATAL) << "unknown thread idx";
      }
    }
    llvm::Function* f = llvm::Intrinsic::getDeclaration(module_.get(), intrin_id);
    return builder_->CreateCall(f, {});
  }

  llvm::Value* CreateStorageSync(const Call* op) final {
    const std::string& sync = op->args[0].as<StringImm>()->value;
    if (sync == "warp") {
      // TODO(tqchen) warp sync in CUDA9
      return nullptr;
    } else if (sync == "shared") {
      llvm::Function* f = llvm::Intrinsic::getDeclaration(
          module_.get(),
          ::llvm::Intrinsic::nvvm_barrier0);
      return builder_->CreateCall(f, {});
    } else {
      LOG(FATAL) << "Do not support sync " << sync;
      return nullptr;
    }
  }

  void InitPassManagerBuilder(llvm::PassManagerBuilder* builder) final {
    // Additional optimization hook to tweak the builder.
  }

152 153 154 155 156 157 158 159 160 161 162 163 164 165
  void Optimize() final {
    for (auto& f : *module_) {
      auto fname = static_cast<std::string>(f.getName());
      if (fname.substr(0, 4) != "__nv") continue;
      // This is to strip off unused __nv_* functions from the final module
      // The one that is actually used will be inlined at call site
      // Adapted from Halide's runtime linker
      if (!f.isDeclaration() && !f.hasFnAttribute(llvm::Attribute::NoInline)) {
        f.setLinkage(llvm::GlobalValue::AvailableExternallyLinkage);
      }
    }
    CodeGenLLVM::Optimize();
  }

166 167 168 169 170 171 172 173
 protected:
  void InitTarget(llvm::TargetMachine* tm) final {
    // Maximum vector lane = float4
    native_vector_bits_ = 4 * 32;
    CodeGenLLVM::InitTarget(tm);
  }
};

174 175
inline int DetectCUDAComputeVersion() {
  TVMContext tvm_ctx;
176
  tvm_ctx.device_type = kDLGPU;
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193
  tvm_ctx.device_id = 0;
  TVMRetValue val;
  tvm::runtime::DeviceAPI::Get(tvm_ctx)->GetAttr(
      tvm_ctx, tvm::runtime::kExist, &val);
  if (val.operator int() == 1) {
    tvm::runtime::DeviceAPI::Get(tvm_ctx)->GetAttr(
        tvm_ctx, tvm::runtime::kComputeVersion, &val);
    std::string version = val;
    std::istringstream is(version);
    double ver;
    is >> ver;
    return static_cast<int>(ver * 10);
  } else {
    return 20;
  }
}

194
runtime::Module BuildNVPTX(Array<LoweredFunc> funcs, std::string target) {
195
  InitializeLLVM();
196
  CHECK(target.length() >= 5 &&
197
        target.substr(0, 5) == "nvptx");
198
  int compute_ver = DetectCUDAComputeVersion();
199 200
  std::ostringstream config;
  config << "-mtriple=nvptx64-nvidia-cuda -mcpu=sm_"
201
         << compute_ver
202
         << target.substr(5, target.length() - 5);
203
  std::unique_ptr<llvm::TargetMachine> tm = GetLLVMTargetMachine(config.str());
204 205
  std::unique_ptr<CodeGenNVPTX> cg(new CodeGenNVPTX());
  std::unique_ptr<llvm::LLVMContext> ctx(new llvm::LLVMContext());
206
  cg->Init(funcs[0]->name, tm.get(), ctx.get(), false, false);
207 208 209
  for (LoweredFunc f :  funcs) {
    cg->AddFunction(f);
  }
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224

  const auto* flibdevice_path =
      tvm::runtime::Registry::Get("tvm_callback_libdevice_path");
  if (flibdevice_path != nullptr) {
    std::string path = (*flibdevice_path)(compute_ver);
    if (path.length() != 0) {
      llvm::SMDiagnostic err;
      std::unique_ptr<llvm::Module> mlib = llvm::parseIRFile(path, err, *ctx);
      if (mlib.get() == nullptr) {
        std::string msg = err.getMessage();
        LOG(FATAL) << "Fail to load bitcode file " << path << "\n"
                   << "line " << err.getLineNo() << ":" << msg;
      }
      mlib->setTargetTriple(tm->getTargetTriple().str());
      mlib->setDataLayout(tm->createDataLayout());
225
      cg->AddLinkModule(std::move(mlib));
226 227
    }
  }
228
  std::unique_ptr<llvm::Module> module = cg->Finish();
229 230 231 232
  llvm::SmallString<8> data_ptx, data_ll;
  llvm::raw_svector_ostream dest_ptx(data_ptx), dest_ll(data_ll);
  dest_ptx.SetUnbuffered();
  dest_ll.SetUnbuffered();
233 234 235 236
  // print ll
  module->print(dest_ll, nullptr);
  std::string ll(data_ll.begin(), data_ll.end());
  // emit ptx
237
  llvm::legacy::PassManager pass;
238
#if TVM_LLVM_VERSION <= 60
239
  CHECK(tm->addPassesToEmitFile(
240
      pass, dest_ptx, llvm::TargetMachine::CGFT_AssemblyFile) == 0)
241
      << "Cannot emit target CGFT_ObjectFile";
242
#elif TVM_LLVM_VERSION <= 90
243 244 245
  CHECK(tm->addPassesToEmitFile(
      pass, dest_ptx, nullptr, llvm::TargetMachine::CGFT_AssemblyFile) == 0)
      << "Cannot emit target CGFT_ObjectFile";
246 247
#else
  CHECK(tm->addPassesToEmitFile(
248
      pass, dest_ptx, nullptr, llvm::CGFT_AssemblyFile) == 0)
249
      << "Cannot emit target CGFT_ObjectFile";
250
#endif
251
  pass.run(*module);
252 253
  std::string ptx(data_ptx.begin(), data_ptx.end());
  return CUDAModuleCreate(ptx, "ptx", ExtractFuncInfo(funcs), ll);
254 255 256
}

TVM_REGISTER_API("codegen.build_nvptx")
257
.set_body_typed(BuildNVPTX);
258 259 260 261

}  // namespace codegen
}  // namespace tvm
#endif  // TVM_LLVM_VERSION