enforce.cc 12.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
2

L
Luo Tao 已提交
3 4 5
Licensed 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
6

L
Luo Tao 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
8

L
Luo Tao 已提交
9 10 11 12 13
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. */
14

15
#include "paddle/phi/core/enforce.h"
16 17 18 19 20 21

#include <map>
#include <memory>
#include <unordered_map>
#include <vector>

22
#include "gflags/gflags.h"
23
#include "glog/logging.h"
24
#include "paddle/phi/common/scalar.h"
R
Ruibiao Chen 已提交
25
#include "paddle/utils/blank.h"
26

27 28 29 30
#ifdef PADDLE_WITH_CUDA
#include "paddle/phi/core/external_error.pb.h"
#endif  // PADDLE_WITH_CUDA

31
DECLARE_int32(call_stack_level);
R
Ruibiao Chen 已提交
32

33 34 35 36 37
namespace egr {
class EagerVariable;
}
namespace paddle {
namespace framework {
38
class VarDesc;
39
class BlockDesc;
R
Ruibiao Chen 已提交
40
using Attribute = paddle::variant<paddle::blank,
R
Ruibiao Chen 已提交
41 42 43 44 45 46 47 48 49 50 51 52
                                  int,
                                  float,
                                  std::string,
                                  std::vector<int>,
                                  std::vector<float>,
                                  std::vector<std::string>,
                                  bool,
                                  std::vector<bool>,
                                  BlockDesc*,
                                  int64_t,
                                  std::vector<BlockDesc*>,
                                  std::vector<int64_t>,
53 54
                                  std::vector<double>,
                                  VarDesc*,
55
                                  std::vector<VarDesc*>,
56 57 58
                                  double,
                                  paddle::experimental::Scalar,
                                  std::vector<paddle::experimental::Scalar>>;
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
using AttributeMap = std::unordered_map<std::string, Attribute>;
}  // namespace framework
namespace imperative {
class VariableWrapper;
class SavedVariableWrapperList;
class VarBase;

namespace details {
template <typename T>
struct NameVarMapTrait {};

template <>
struct NameVarMapTrait<VarBase> {
  using Type = std::map<std::string, std::vector<std::shared_ptr<VarBase>>>;
};

template <>
struct NameVarMapTrait<VariableWrapper> {
  using Type = std::map<std::string, SavedVariableWrapperList>;
};

template <>
struct NameVarMapTrait<egr::EagerVariable> {
  using Type =
      std::map<std::string, std::vector<std::shared_ptr<egr::EagerVariable>>>;
};

}  // namespace details

template <typename T>
using NameVarMap = typename details::NameVarMapTrait<T>::Type;

using NameVarBaseMap = NameVarMap<VarBase>;
using NameVariableWrapperMap = NameVarMap<VariableWrapper>;
using NameTensorMap = NameVarMap<egr::EagerVariable>;

}  // namespace imperative
}  // namespace paddle

namespace phi {
namespace enforce {

101 102
int GetCallStackLevel() { return FLAGS_call_stack_level; }

103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
template <typename T>
static std::string ReplaceComplexTypeStr(std::string str,
                                         const std::string& type_name) {
  auto demangle_type_str = demangle(typeid(T).name());
  size_t start_pos = 0;
  while ((start_pos = str.find(demangle_type_str, start_pos)) !=
         std::string::npos) {
    str.replace(start_pos, demangle_type_str.length(), type_name);
    start_pos += type_name.length();
  }
  return str;
}

#define __REPLACE_COMPLEX_TYPE_STR__(__TYPENAME, __STR)                      \
  do {                                                                       \
    __STR =                                                                  \
        phi::enforce::ReplaceComplexTypeStr<__TYPENAME>(__STR, #__TYPENAME); \
  } while (0)

static std::string SimplifyDemangleStr(std::string str) {
  // the older is important, you have to put complex types in front
  __REPLACE_COMPLEX_TYPE_STR__(paddle::framework::AttributeMap, str);
  __REPLACE_COMPLEX_TYPE_STR__(paddle::framework::Attribute, str);
  __REPLACE_COMPLEX_TYPE_STR__(paddle::imperative::NameVariableWrapperMap, str);
  __REPLACE_COMPLEX_TYPE_STR__(paddle::imperative::NameVarBaseMap, str);
  __REPLACE_COMPLEX_TYPE_STR__(paddle::imperative::NameTensorMap, str);
  __REPLACE_COMPLEX_TYPE_STR__(std::string, str);
  return str;
}

std::string GetCurrentTraceBackString(bool for_signal) {
  std::ostringstream sout;

  if (!for_signal) {
    sout << "\n\n--------------------------------------\n";
    sout << "C++ Traceback (most recent call last):";
    sout << "\n--------------------------------------\n";
  }
#if !defined(_WIN32) && !defined(PADDLE_WITH_MUSL)
  static constexpr int TRACE_STACK_LIMIT = 100;

  void* call_stack[TRACE_STACK_LIMIT];
  auto size = backtrace(call_stack, TRACE_STACK_LIMIT);
  auto symbols = backtrace_symbols(call_stack, size);
  Dl_info info;
  int idx = 0;
  // `for_signal` used to remove the stack trace introduced by
  // obtaining the error stack trace when the signal error occurred,
  // that is not related to the signal error self, remove it to
  // avoid misleading users and developers
  int end_idx = for_signal ? 2 : 0;
  for (int i = size - 1; i >= end_idx; --i) {
    if (dladdr(call_stack[i], &info) && info.dli_sname) {
      auto demangled = demangle(info.dli_sname);
      std::string path(info.dli_fname);
      // C++ traceback info are from core.so
      if (path.substr(path.length() - 3).compare(".so") == 0) {
        sout << paddle::string::Sprintf(
            "%-3d %s\n", idx++, SimplifyDemangleStr(demangled));
      }
    }
  }
  free(symbols);
#else
  sout << "Not support stack backtrace yet.\n";
#endif
  return sout.str();
}

172 173 174 175
void ThrowWarnInternal(const std::string& msg) {
  LOG(WARNING) << "WARNING :" << msg;
}

176 177 178 179 180 181 182 183 184 185 186 187 188
std::string SimplifyErrorTypeFormat(const std::string& str) {
  std::ostringstream sout;
  size_t type_end_pos = str.find(":", 0);
  if (type_end_pos == std::string::npos) {
    sout << str;
  } else {
    // Remove "Error:", add "()""
    sout << "(" << str.substr(0, type_end_pos - 5) << ")"
         << str.substr(type_end_pos + 1);
  }
  return sout.str();
}

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357
/**************************************************************************/
/**************************** NVIDIA ERROR ********************************/
#ifdef PADDLE_WITH_CUDA

namespace details {

template <typename T>
struct ExternalApiProtoType {};

#define DEFINE_EXTERNAL_API_PROTO_TYPE(type, proto_type)    \
  template <>                                               \
  struct ExternalApiProtoType<type> {                       \
    using Type = type;                                      \
    static constexpr const char* kTypeString = #proto_type; \
    static constexpr phi::proto::ApiType kProtoType =       \
        phi::proto::ApiType::proto_type;                    \
  }

DEFINE_EXTERNAL_API_PROTO_TYPE(cudaError_t, CUDA);
DEFINE_EXTERNAL_API_PROTO_TYPE(curandStatus_t, CURAND);
DEFINE_EXTERNAL_API_PROTO_TYPE(cudnnStatus_t, CUDNN);
DEFINE_EXTERNAL_API_PROTO_TYPE(cublasStatus_t, CUBLAS);
DEFINE_EXTERNAL_API_PROTO_TYPE(cusparseStatus_t, CUSPARSE);
DEFINE_EXTERNAL_API_PROTO_TYPE(cusolverStatus_t, CUSOLVER);
DEFINE_EXTERNAL_API_PROTO_TYPE(cufftResult_t, CUFFT);
DEFINE_EXTERNAL_API_PROTO_TYPE(CUresult, CU);

#if !defined(__APPLE__) && defined(PADDLE_WITH_NCCL)
DEFINE_EXTERNAL_API_PROTO_TYPE(ncclResult_t, NCCL);
#endif

#undef DEFINE_EXTERNAL_API_PROTO_TYPE

}  // namespace details

template <typename T>
inline const char* GetErrorMsgUrl(T status) {
  using __CUDA_STATUS_TYPE__ = decltype(status);
  phi::proto::ApiType proto_type =
      details::ExternalApiProtoType<__CUDA_STATUS_TYPE__>::kProtoType;
  switch (proto_type) {
    case phi::proto::ApiType::CUDA:
    case phi::proto::ApiType::CU:
      return "https://docs.nvidia.com/cuda/cuda-runtime-api/"
             "group__CUDART__TYPES.html#group__CUDART__TYPES_"
             "1g3f51e3575c2178246db0a94a430e0038";
      break;
    case phi::proto::ApiType::CURAND:
      return "https://docs.nvidia.com/cuda/curand/"
             "group__HOST.html#group__HOST_1gb94a31d5c165858c96b6c18b70644437";
      break;
    case phi::proto::ApiType::CUDNN:
      return "https://docs.nvidia.com/deeplearning/cudnn/api/"
             "index.html#cudnnStatus_t";
      break;
    case phi::proto::ApiType::CUBLAS:
      return "https://docs.nvidia.com/cuda/cublas/index.html#cublasstatus_t";
      break;
    case phi::proto::ApiType::CUSOLVER:
      return "https://docs.nvidia.com/cuda/cusolver/"
             "index.html#cuSolverSPstatus";
      break;
    case phi::proto::ApiType::NCCL:
      return "https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/api/"
             "types.html#ncclresult-t";
      break;
    case phi::proto::ApiType::CUFFT:
      return "https://docs.nvidia.com/cuda/cufft/index.html#cufftresult";
    case phi::proto::ApiType::CUSPARSE:
      return "https://docs.nvidia.com/cuda/cusparse/"
             "index.html#cusparseStatus_t";
      break;
    default:
      return "Unknown type of External API, can't get error message URL!";
      break;
  }
}

template <typename T>
std::string GetExternalErrorMsg(T status) {
  std::ostringstream sout;
  bool _initSucceed = false;
  phi::proto::ExternalErrorDesc externalError;
  if (externalError.ByteSizeLong() == 0) {
    std::string filePath;
#if !defined(_WIN32)
    Dl_info info;
    if (dladdr(reinterpret_cast<void*>(GetCurrentTraceBackString), &info)) {
      std::string strModule(info.dli_fname);
      const size_t last_slash_idx = strModule.find_last_of("/");
      std::string compare_path = strModule.substr(strModule.length() - 6);
      if (std::string::npos != last_slash_idx) {
        strModule.erase(last_slash_idx, std::string::npos);
      }
      if (compare_path.compare("avx.so") == 0) {
        filePath =
            strModule +
            "/../include/third_party/externalError/data/externalErrorMsg.pb";
      } else {
        filePath = strModule +
                   "/../../third_party/externalError/data/externalErrorMsg.pb";
      }
    }
#else
    char buf[512];
    MEMORY_BASIC_INFORMATION mbi;
    HMODULE h_module =
        (::VirtualQuery(GetCurrentTraceBackString, &mbi, sizeof(mbi)) != 0)
            ? (HMODULE)mbi.AllocationBase
            : NULL;
    GetModuleFileName(h_module, buf, 512);
    std::string strModule(buf);
    const size_t last_slash_idx = strModule.find_last_of("\\");
    std::string compare_path = strModule.substr(strModule.length() - 7);
    if (std::string::npos != last_slash_idx) {
      strModule.erase(last_slash_idx, std::string::npos);
    }
    if (compare_path.compare("avx.pyd") == 0) {
      filePath = strModule +
                 "\\..\\include\\third_"
                 "party\\externalerror\\data\\externalErrorMsg.pb";
    } else {
      filePath =
          strModule +
          "\\..\\..\\third_party\\externalerror\\data\\externalErrorMsg.pb";
    }
#endif
    std::ifstream fin(filePath, std::ios::in | std::ios::binary);
    _initSucceed = externalError.ParseFromIstream(&fin);
  }
  using __CUDA_STATUS_TYPE__ = decltype(status);
  phi::proto::ApiType proto_type =
      details::ExternalApiProtoType<__CUDA_STATUS_TYPE__>::kProtoType;
  if (_initSucceed) {
    for (int i = 0; i < externalError.errors_size(); ++i) {
      if (proto_type == externalError.errors(i).type()) {
        for (int j = 0; j < externalError.errors(i).messages_size(); ++j) {
          if (status == externalError.errors(i).messages(j).code()) {
            sout << "\n  [Hint: "
                 << externalError.errors(i).messages(j).message() << "]";
            return sout.str();
          }
        }
      }
    }
  }

  sout << "\n  [Hint: Please search for the error code(" << status
       << ") on website (" << GetErrorMsgUrl(status)
       << ") to get Nvidia's official solution and advice about "
       << details::ExternalApiProtoType<__CUDA_STATUS_TYPE__>::kTypeString
       << " Error.]";
  return sout.str();
}

template std::string GetExternalErrorMsg<cudaError_t>(cudaError_t);
template std::string GetExternalErrorMsg<curandStatus_t>(curandStatus_t);
template std::string GetExternalErrorMsg<cudnnStatus_t>(cudnnStatus_t);
template std::string GetExternalErrorMsg<cublasStatus_t>(cublasStatus_t);
template std::string GetExternalErrorMsg<cusparseStatus_t>(cusparseStatus_t);
template std::string GetExternalErrorMsg<cusolverStatus_t>(cusolverStatus_t);
template std::string GetExternalErrorMsg<cufftResult_t>(cufftResult_t);
template std::string GetExternalErrorMsg<CUresult>(CUresult);
#if !defined(__APPLE__) && defined(PADDLE_WITH_NCCL)
template std::string GetExternalErrorMsg<ncclResult_t>(ncclResult_t);
#endif

#endif  // PADDLE_WITH_CUDA

358 359
}  // namespace enforce
}  // namespace phi