pd_config.cc 17.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
//
// 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
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// 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.

#include "paddle/fluid/inference/capi_exp/pd_config.h"
#include "paddle/fluid/inference/api/paddle_inference_api.h"
17 18
#include "paddle/fluid/inference/capi_exp/pd_types.h"
#include "paddle/fluid/inference/capi_exp/utils_internal.h"
19 20 21 22 23 24 25 26 27 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 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
#include "paddle/fluid/platform/enforce.h"

#define CHECK_NULL_POINTER_PARM(param)                  \
  PADDLE_ENFORCE_NOT_NULL(                              \
      param, paddle::platform::errors::InvalidArgument( \
                 "The pointer of " #param " shouldn't be nullptr"))

#define CHECK_AND_CONVERT_PD_CONFIG                                         \
  PADDLE_ENFORCE_NOT_NULL(                                                  \
      pd_config, paddle::platform::errors::InvalidArgument(                 \
                     "The pointer of paddle config shouldn't be nullptr")); \
  Config* config = reinterpret_cast<Config*>(pd_config)

using paddle_infer::Config;

static Config::Precision ConvertToCxxPrecisionType(PD_PrecisionType precision) {
  switch (precision) {
    case PD_PRECISION_FLOAT32:
      return Config::Precision::kFloat32;
    case PD_PRECISION_INT8:
      return Config::Precision::kInt8;
    case PD_PRECISION_HALF:
      return Config::Precision::kHalf;
    default:
      PADDLE_THROW(paddle::platform::errors::InvalidArgument(
          "Unsupport paddle precision type %d.", precision));
      return Config::Precision::kFloat32;
  }
}

extern "C" {
__pd_give PD_Config* PD_ConfigCreate() {
  return reinterpret_cast<PD_Config*>(new Config());
}

void PD_ConfigDestroy(__pd_take PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  delete reinterpret_cast<Config*>(config);
}

void PD_ConfigSetModel(__pd_keep PD_Config* pd_config,
                       const char* prog_file_path,
                       const char* params_file_path) {
  CHECK_AND_CONVERT_PD_CONFIG;
  CHECK_NULL_POINTER_PARM(prog_file_path);
  CHECK_NULL_POINTER_PARM(params_file_path);
  config->SetModel(prog_file_path, params_file_path);
}
void PD_ConfigSetProgFile(__pd_keep PD_Config* pd_config,
                          const char* prog_file_path) {
  CHECK_AND_CONVERT_PD_CONFIG;
  CHECK_NULL_POINTER_PARM(prog_file_path);
  config->SetProgFile(prog_file_path);
}
void PD_ConfigSetParamsFile(__pd_keep PD_Config* pd_config,
                            const char* params_file_path) {
  CHECK_AND_CONVERT_PD_CONFIG;
  CHECK_NULL_POINTER_PARM(params_file_path);
  config->SetParamsFile(params_file_path);
}
void PD_ConfigSetOptimCacheDir(__pd_keep PD_Config* pd_config,
                               const char* opt_cache_dir) {
  CHECK_AND_CONVERT_PD_CONFIG;
  CHECK_NULL_POINTER_PARM(opt_cache_dir);
  config->SetOptimCacheDir(opt_cache_dir);
}

void PD_ConfigSetModelDir(__pd_keep PD_Config* pd_config,
                          const char* model_dir) {
  CHECK_AND_CONVERT_PD_CONFIG;
  CHECK_NULL_POINTER_PARM(model_dir);
  config->SetModel(model_dir);
}
const char* PD_ConfigGetModelDir(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->model_dir().c_str();
}
const char* PD_ConfigGetProgFile(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->prog_file().c_str();
}
const char* PD_ConfigGetParamsFile(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->params_file().c_str();
}

void PD_ConfigDisableFCPadding(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->DisableFCPadding();
}
PD_Bool PD_ConfigUseFcPadding(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->use_fc_padding();
}

void PD_ConfigEnableUseGpu(__pd_keep PD_Config* pd_config,
                           uint64_t memory_pool_init_size_mb,
                           int32_t device_id) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableUseGpu(memory_pool_init_size_mb, device_id);
}
void PD_ConfigDisableGpu(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->DisableGpu();
}
PD_Bool PD_ConfigUseGpu(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->use_gpu();
}

void PD_ConfigEnableXpu(__pd_keep PD_Config* pd_config,
130 131 132
                        int32_t l3_workspace_size, PD_Bool locked,
                        PD_Bool autotune, const char* autotune_file,
                        const char* precision, PD_Bool adaptive_seqlen) {
133
  CHECK_AND_CONVERT_PD_CONFIG;
134 135
  config->EnableXpu(l3_workspace_size, locked, autotune, autotune_file,
                    precision, adaptive_seqlen);
136
}
137

W
Wilber 已提交
138 139 140 141 142
void PD_ConfigEnableNpu(__pd_keep PD_Config* pd_config, int32_t device_id) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableNpu(device_id);
}

143 144 145 146 147
PD_Bool PD_ConfigUseXpu(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->use_xpu();
}

W
Wilber 已提交
148 149 150 151 152
PD_Bool PD_ConfigUseNpu(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->use_npu();
}

153 154 155 156 157 158 159 160
int32_t PD_ConfigGpuDeviceId(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->gpu_device_id();
}
int32_t PD_ConfigXpuDeviceId(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->xpu_device_id();
}
W
Wilber 已提交
161 162 163 164
int32_t PD_ConfigNpuDeviceId(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->npu_device_id();
}
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
int32_t PD_ConfigMemoryPoolInitSizeMb(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->memory_pool_init_size_mb();
}
float PD_ConfigFractionOfGpuMemoryForPool(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->fraction_of_gpu_memory_for_pool();
}
void PD_ConfigEnableCudnn(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableCUDNN();
}
PD_Bool PD_ConfigCudnnEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->cudnn_enabled();
}

void PD_ConfigSwitchIrOptim(__pd_keep PD_Config* pd_config, PD_Bool x) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SwitchIrOptim(x);
}
PD_Bool PD_ConfigIrOptim(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->ir_optim();
}

void PD_ConfigEnableTensorRtEngine(__pd_keep PD_Config* pd_config,
                                   int32_t workspace_size,
                                   int32_t max_batch_size,
                                   int32_t min_subgraph_size,
                                   PD_PrecisionType precision,
                                   PD_Bool use_static, PD_Bool use_calib_mode) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableTensorRtEngine(
      workspace_size, max_batch_size, min_subgraph_size,
      ConvertToCxxPrecisionType(precision), use_static, use_calib_mode);
}
PD_Bool PD_ConfigTensorRtEngineEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->tensorrt_engine_enabled();
}

void PD_ConfigSetTrtDynamicShapeInfo(__pd_keep PD_Config* pd_config,
                                     size_t tensor_num,
                                     const char** tensor_name,
                                     size_t* shapes_num, int32_t** min_shape,
                                     int32_t** max_shape, int32_t** optim_shape,
                                     PD_Bool disable_trt_plugin_fp16) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::map<std::string, std::vector<int>> min_input_shapes;
  std::map<std::string, std::vector<int>> max_input_shapes;
  std::map<std::string, std::vector<int>> optim_input_shapes;
  for (size_t tensor_index = 0; tensor_index < tensor_num; ++tensor_index) {
    std::string name(tensor_name[tensor_index]);
    std::vector<int> min_input_shape, max_input_shape, optim_input_shape;
    for (size_t shape_index = 0; shape_index < shapes_num[tensor_index];
         ++shape_index) {
      min_input_shape.emplace_back(min_shape[tensor_index][shape_index]);
      max_input_shape.emplace_back(max_shape[tensor_index][shape_index]);
      optim_input_shape.emplace_back(optim_shape[tensor_index][shape_index]);
    }
    min_input_shapes[name] = std::move(min_input_shape);
    max_input_shapes[name] = std::move(max_input_shape);
    optim_input_shapes[name] = std::move(optim_input_shape);
  }
  config->SetTRTDynamicShapeInfo(min_input_shapes, max_input_shapes,
                                 optim_input_shapes, disable_trt_plugin_fp16);
}

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
PD_Bool PD_ConfigTensorRtDynamicShapeEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->tensorrt_dynamic_shape_enabled();
}

void PD_ConfigEnableTunedTensorRtDynamicShape(__pd_keep PD_Config* pd_config,
                                              const char* shape_range_info_path,
                                              PD_Bool allow_build_at_runtime) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableTunedTensorRtDynamicShape(shape_range_info_path,
                                          allow_build_at_runtime);
}

PD_Bool PD_ConfigTunedTensorRtDynamicShape(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->tuned_tensorrt_dynamic_shape();
}

PD_Bool PD_ConfigTrtAllowBuildAtRuntime(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->trt_allow_build_at_runtime();
}

void PD_ConfigCollectShapeRangeInfo(__pd_keep PD_Config* pd_config,
                                    const char* shape_range_info_path) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->CollectShapeRangeInfo(shape_range_info_path);
}

const char* PD_ConfigShapeRangeInfoPath(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  auto shape_str = config->shape_range_info_path();
  char* c = reinterpret_cast<char*>(malloc(shape_str.length() + 1));
  snprintf(c, shape_str.length() + 1, "%s", shape_str.c_str());
  return c;
}

PD_Bool PD_ConfigShapeRangeInfoCollected(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->shape_range_info_collected();
}

276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402
void PD_ConfigDisableTensorRtOPs(__pd_keep PD_Config* pd_config, size_t ops_num,
                                 const char** ops_name) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::vector<std::string> ops_list;
  for (size_t index = 0; index < ops_num; ++index) {
    ops_list.emplace_back(ops_name[index]);
  }
  config->Exp_DisableTensorRtOPs(ops_list);
}

void PD_ConfigEnableTensorRtOSS(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableTensorRtOSS();
}
PD_Bool PD_ConfigTensorRtOssEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->tensorrt_oss_enabled();
}

void PD_ConfigEnableTensorRtDla(__pd_keep PD_Config* pd_config,
                                int32_t dla_core) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableTensorRtDLA(dla_core);
}
PD_Bool PD_ConfigTensorRtDlaEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->tensorrt_dla_enabled();
}

void PD_ConfigEnableLiteEngine(__pd_keep PD_Config* pd_config,
                               PD_PrecisionType precision, PD_Bool zero_copy,
                               size_t passes_filter_num,
                               const char** passes_filter,
                               size_t ops_filter_num, const char** ops_filter) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::vector<std::string> passes_filters, ops_filters;
  for (size_t index = 0; index < passes_filter_num; ++index) {
    passes_filters.emplace_back(passes_filter[index]);
  }
  for (size_t index = 0; index < ops_filter_num; ++index) {
    ops_filters.emplace_back(ops_filter[index]);
  }
  config->EnableLiteEngine(ConvertToCxxPrecisionType(precision), zero_copy,
                           passes_filters, ops_filters);
}
PD_Bool PD_ConfigLiteEngineEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->lite_engine_enabled();
}

void PD_ConfigSwitchIrDebug(__pd_keep PD_Config* pd_config, PD_Bool x) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SwitchIrDebug(x);
}
void PD_ConfigEnableMKLDNN(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableMKLDNN();
}
void PD_ConfigSetMkldnnCacheCapacity(__pd_keep PD_Config* pd_config,
                                     int32_t capacity) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SetMkldnnCacheCapacity(capacity);
}
PD_Bool PD_ConfigMkldnnEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->mkldnn_enabled();
}
void PD_ConfigSetCpuMathLibraryNumThreads(
    __pd_keep PD_Config* pd_config, int32_t cpu_math_library_num_threads) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SetCpuMathLibraryNumThreads(cpu_math_library_num_threads);
}
int32_t PD_ConfigGetCpuMathLibraryNumThreads(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->cpu_math_library_num_threads();
}

void PD_ConfigSetMkldnnOp(__pd_keep PD_Config* pd_config, size_t ops_num,
                          const char** op_list) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::unordered_set<std::string> op_names;
  for (size_t index = 0; index < ops_num; ++index) {
    op_names.emplace(op_list[index]);
  }
  config->SetMKLDNNOp(std::move(op_names));
}
void PD_ConfigEnableMkldnnQuantizer(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableMkldnnQuantizer();
}
void PD_ConfigEnableMkldnnBfloat16(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableMkldnnBfloat16();
}
PD_Bool PD_ConfigMkldnnBfloat16Enabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->mkldnn_bfloat16_enabled();
}
void PD_ConfigSetBfloat16Op(__pd_keep PD_Config* pd_config, size_t ops_num,
                            const char** op_list) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::unordered_set<std::string> op_names;
  for (size_t index = 0; index < ops_num; ++index) {
    op_names.emplace(op_list[index]);
  }
  config->SetBfloat16Op(std::move(op_names));
}
PD_Bool PD_ConfigThreadLocalStreamEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->thread_local_stream_enabled();
}
PD_Bool PD_ConfigMkldnnQuantizerEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->mkldnn_quantizer_enabled();
}
void PD_ConfigSetModelBuffer(__pd_keep PD_Config* pd_config,
                             const char* prog_buffer, size_t prog_buffer_size,
                             const char* params_buffer,
                             size_t params_buffer_size) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SetModelBuffer(prog_buffer, prog_buffer_size, params_buffer,
                         params_buffer_size);
}
PD_Bool PD_ConfigModelFromMemory(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->model_from_memory();
}
403
void PD_ConfigEnableMemoryOptim(__pd_keep PD_Config* pd_config, PD_Bool x) {
404
  CHECK_AND_CONVERT_PD_CONFIG;
405
  config->EnableMemoryOptim(x);
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
}
PD_Bool PD_ConfigMemoryOptimEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->enable_memory_optim();
}
void PD_ConfigEnableProfile(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableProfile();
}
PD_Bool PD_ConfigProfileEnabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->profile_enabled();
}
void PD_ConfigDisableGlogInfo(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->DisableGlogInfo();
}
PD_Bool PD_ConfigGlogInfoDisabled(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->glog_info_disabled();
}
void PD_ConfigSetInvalid(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->SetInValid();
}
PD_Bool PD_ConfigIsValid(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  return config->is_valid();
}
void PD_ConfigEnableGpuMultiStream(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->EnableGpuMultiStream();
}
void PD_ConfigPartiallyRelease(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->PartiallyRelease();
}
443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461
void PD_ConfigDeletePass(__pd_keep PD_Config* pd_config, const char* pass) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->pass_builder()->DeletePass(pass);
}
void PD_ConfigInsertPass(__pd_keep PD_Config* pd_config, size_t idx,
                         const char* pass) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->pass_builder()->InsertPass(idx, pass);
}
void PD_ConfigAppendPass(__pd_keep PD_Config* pd_config, const char* pass) {
  CHECK_AND_CONVERT_PD_CONFIG;
  config->pass_builder()->AppendPass(pass);
}
__pd_give PD_OneDimArrayCstr* PD_ConfigAllPasses(
    __pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  std::vector<std::string> passes = config->pass_builder()->AllPasses();
  return paddle_infer::CvtVecToOneDimArrayCstr(passes);
}
462 463 464 465 466 467 468
const char* PD_ConfigSummary(__pd_keep PD_Config* pd_config) {
  CHECK_AND_CONVERT_PD_CONFIG;
  auto sum_str = config->Summary();
  char* c = reinterpret_cast<char*>(malloc(sum_str.length() + 1));
  snprintf(c, sum_str.length() + 1, "%s", sum_str.c_str());
  return c;
}
469 470

}  // extern "C"