tester_helper.h 28.5 KB
Newer Older
L
luotao1 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
// Copyright (c) 2018 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.

#pragma once

#include <gtest/gtest.h>
Y
Yan Chunwei 已提交
18

L
luotao1 已提交
19
#include <algorithm>
L
luotao1 已提交
20
#include <memory>
T
Tao Luo 已提交
21
#include <string>
L
luotao1 已提交
22
#include <thread>  // NOLINT
L
luotao1 已提交
23
#include <unordered_map>
L
luotao1 已提交
24
#include <vector>
Y
Yiqun Liu 已提交
25 26 27
#ifdef WITH_GPERFTOOLS
#include <gperftools/profiler.h>
#endif
L
luotao1 已提交
28
#include "paddle/fluid/framework/ir/fuse_pass_base.h"
29
#include "paddle/fluid/framework/scope.h"
L
luotao1 已提交
30 31 32
#include "paddle/fluid/inference/analysis/analyzer.h"
#include "paddle/fluid/inference/analysis/ut_helper.h"
#include "paddle/fluid/inference/api/analysis_predictor.h"
33
#include "paddle/fluid/inference/api/helper.h"
Y
Yan Chunwei 已提交
34
#include "paddle/fluid/inference/api/paddle_inference_pass.h"
35
#include "paddle/fluid/inference/tests/api/config_printer.h"
T
Tao Luo 已提交
36
#include "paddle/fluid/inference/tests/test_helper.h"
N
nhzlx 已提交
37
#include "paddle/fluid/inference/utils/benchmark.h"
L
luotao1 已提交
38 39
#include "paddle/fluid/platform/profiler.h"

N
nhzlx 已提交
40
DEFINE_string(model_name, "", "model name");
L
luotao1 已提交
41 42
DEFINE_string(infer_model, "", "model path");
DEFINE_string(infer_data, "", "data file");
T
Tao Luo 已提交
43
DEFINE_string(refer_result, "", "reference result for comparison");
44
DEFINE_int32(batch_size, 1, "batch size");
45 46
DEFINE_bool(enable_fp32, true, "Enable FP32 type prediction");
DEFINE_bool(enable_int8, true, "Enable INT8 type prediction");
47 48 49
DEFINE_int32(warmup_batch_size, 100, "batch size for quantization warmup");
// setting iterations to 0 means processing the whole dataset
DEFINE_int32(iterations, 0, "number of batches to process");
L
luotao1 已提交
50 51 52
DEFINE_int32(repeat, 1, "Running the inference program repeat times.");
DEFINE_bool(test_all_data, false, "Test the all dataset in data file.");
DEFINE_int32(num_threads, 1, "Running the inference program in multi-threads.");
T
Tao Luo 已提交
53 54
DEFINE_bool(use_analysis, true,
            "Running the inference program in analysis mode.");
N
nhzlx 已提交
55 56
DEFINE_bool(record_benchmark, false,
            "Record benchmark after profiling the model");
L
luotao1 已提交
57
DEFINE_double(accuracy, 1e-3, "Result Accuracy.");
58
DEFINE_double(quantized_accuracy, 1e-2, "Result Quantized Accuracy.");
L
luotao1 已提交
59
DEFINE_bool(zero_copy, false, "Use ZeroCopy to speedup Feed/Fetch.");
60 61 62
DEFINE_bool(warmup, false,
            "Use warmup to calculate elapsed_time more accurately. "
            "To reduce CI time, it sets false in default.");
L
luotao1 已提交
63

64
DECLARE_bool(profile);
L
luotao1 已提交
65
DECLARE_int32(paddle_num_threads);
66

L
luotao1 已提交
67 68 69
namespace paddle {
namespace inference {

70 71
using paddle::framework::proto::VarType;

72 73 74 75 76 77 78 79 80 81 82 83 84
template <typename T>
constexpr paddle::PaddleDType GetPaddleDType();

template <>
constexpr paddle::PaddleDType GetPaddleDType<int64_t>() {
  return paddle::PaddleDType::INT64;
}

template <>
constexpr paddle::PaddleDType GetPaddleDType<float>() {
  return paddle::PaddleDType::FLOAT32;
}

85
void PrintConfig(const PaddlePredictor::Config *config, bool use_analysis) {
86
  const auto *analysis_config =
87
      reinterpret_cast<const AnalysisConfig *>(config);
88
  if (use_analysis) {
89
    LOG(INFO) << *analysis_config;
90 91
    return;
  }
92
  LOG(INFO) << analysis_config->ToNativeConfig();
93
}
Y
Yan Chunwei 已提交
94

95
// Compare result between two PaddleTensor
L
luotao1 已提交
96
void CompareResult(const std::vector<PaddleTensor> &outputs,
T
tensor-tang 已提交
97
                   const std::vector<PaddleTensor> &ref_outputs) {
T
Tao Luo 已提交
98
  EXPECT_GT(outputs.size(), 0UL);
T
tensor-tang 已提交
99
  EXPECT_EQ(outputs.size(), ref_outputs.size());
L
luotao1 已提交
100 101
  for (size_t i = 0; i < outputs.size(); i++) {
    auto &out = outputs[i];
T
tensor-tang 已提交
102
    auto &ref_out = ref_outputs[i];
103 104
    size_t size = VecReduceToInt(out.shape);
    size_t ref_size = VecReduceToInt(ref_out.shape);
105
    EXPECT_GT(size, 0UL);
T
tensor-tang 已提交
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
    EXPECT_EQ(size, ref_size);
    EXPECT_EQ(out.dtype, ref_out.dtype);
    switch (out.dtype) {
      case PaddleDType::INT64: {
        int64_t *pdata = static_cast<int64_t *>(out.data.data());
        int64_t *pdata_ref = static_cast<int64_t *>(ref_out.data.data());
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
      case PaddleDType::FLOAT32: {
        float *pdata = static_cast<float *>(out.data.data());
        float *pdata_ref = static_cast<float *>(ref_out.data.data());
        for (size_t j = 0; j < size; ++j) {
Y
Yan Chunwei 已提交
121
          CHECK_LE(std::abs(pdata_ref[j] - pdata[j]), FLAGS_accuracy);
T
tensor-tang 已提交
122 123 124
        }
        break;
      }
125 126 127 128 129 130 131 132
      case PaddleDType::INT32: {
        int32_t *pdata = static_cast<int32_t *>(out.data.data());
        int32_t *pdata_ref = static_cast<int32_t *>(ref_out.data.data());
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
133 134 135 136 137 138 139 140
      case PaddleDType::UINT8: {
        uint8_t *pdata = static_cast<uint8_t *>(out.data.data());
        uint8_t *pdata_ref = static_cast<uint8_t *>(ref_out.data.data());
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
L
luotao1 已提交
141 142 143 144
    }
  }
}

145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
// Compare result between a PaddleTensor and a ZeroCopyTensor
void CompareResult(const std::vector<PaddleTensor> &outputs,
                   const std::vector<ZeroCopyTensor> &ref_outputs) {
  EXPECT_GT(outputs.size(), 0UL);
  EXPECT_EQ(outputs.size(), ref_outputs.size());
  for (size_t i = 0; i < outputs.size(); i++) {
    auto &out = outputs[i];
    auto &ref_out = ref_outputs[i];
    size_t size = VecReduceToInt(out.shape);
    EXPECT_GT(size, 0UL);
    int ref_size = 0;  // this is the number of elements not memory size
    PaddlePlace place;
    switch (out.dtype) {
      case PaddleDType::INT64: {
        int64_t *pdata = static_cast<int64_t *>(out.data.data());
        int64_t *pdata_ref = ref_out.data<int64_t>(&place, &ref_size);
161
        EXPECT_EQ(size, static_cast<size_t>(ref_size));
162 163 164 165 166 167 168 169 170 171 172 173 174 175
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
      case PaddleDType::FLOAT32: {
        float *pdata = static_cast<float *>(out.data.data());
        float *pdata_ref = ref_out.data<float>(&place, &ref_size);
        EXPECT_EQ(size, ref_size);
        for (size_t j = 0; j < size; ++j) {
          CHECK_LE(std::abs(pdata_ref[j] - pdata[j]), FLAGS_accuracy);
        }
        break;
      }
L
luotao1 已提交
176 177 178 179 180 181 182 183 184
      case PaddleDType::INT32: {
        int32_t *pdata = static_cast<int32_t *>(out.data.data());
        int32_t *pdata_ref = ref_out.data<int32_t>(&place, &ref_size);
        EXPECT_EQ(size, ref_size);
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
185 186 187 188 189 190 191 192 193
      case PaddleDType::UINT8: {
        uint8_t *pdata = static_cast<uint8_t *>(out.data.data());
        uint8_t *pdata_ref = ref_out.data<uint8_t>(&place, &ref_size);
        EXPECT_EQ(size, ref_size);
        for (size_t j = 0; j < size; ++j) {
          EXPECT_EQ(pdata_ref[j], pdata[j]);
        }
        break;
      }
194 195 196 197
    }
  }
}

198
std::unique_ptr<PaddlePredictor> CreateTestPredictor(
199
    const PaddlePredictor::Config *config, bool use_analysis = true) {
200
  const auto *analysis_config =
201
      reinterpret_cast<const AnalysisConfig *>(config);
T
Tao Luo 已提交
202
  if (use_analysis) {
203
    return CreatePaddlePredictor<AnalysisConfig>(*analysis_config);
T
Tao Luo 已提交
204
  }
205 206
  auto native_config = analysis_config->ToNativeConfig();
  return CreatePaddlePredictor<NativeConfig>(native_config);
T
Tao Luo 已提交
207 208
}

209
size_t GetSize(const PaddleTensor &out) { return VecReduceToInt(out.shape); }
T
Tao Luo 已提交
210

211
std::unordered_map<std::string, int> GetFuseStatis(PaddlePredictor *predictor,
T
Tao Luo 已提交
212
                                                   int *num_ops) {
213
  std::unordered_map<std::string, int> res;
214
  auto *analysis_predictor = static_cast<AnalysisPredictor *>(predictor);
215 216 217 218 219 220
  auto *fusion_status =
      analysis_predictor->analysis_argument().fusion_statis_ptr();
  if (!fusion_status) {
    return res;
  }
  for (auto &item : *fusion_status) {
T
Tao Luo 已提交
221 222 223 224
    LOG(INFO) << "fused " << item.first << " " << item.second;
  }
  int num = 0;
  for (auto &node :
225 226
       analysis_predictor->analysis_argument().main_graph().Nodes()) {
    if (node->IsOp()) {
T
Tao Luo 已提交
227 228 229 230
      ++num;
    }
  }
  *num_ops = num;
231
  return *fusion_status;
T
Tao Luo 已提交
232 233
}

T
Tao Luo 已提交
234
void SetFakeImageInput(std::vector<std::vector<PaddleTensor>> *inputs,
235 236
                       const std::string &dirname, bool is_combined = true,
                       std::string model_filename = "model",
T
tensor-tang 已提交
237
                       std::string params_filename = "params",
N
nhzlx 已提交
238 239
                       const std::vector<std::string> *feed_names = nullptr,
                       const int continuous_inuput_index = 0) {
T
Tao Luo 已提交
240 241
  // Set fake_image_data
  PADDLE_ENFORCE_EQ(FLAGS_test_all_data, 0, "Only have single batch of data.");
242 243 244 245 246 247 248 249 250 251 252
  std::vector<std::vector<int64_t>> feed_target_shapes = GetFeedTargetShapes(
      dirname, is_combined, model_filename, params_filename);
  std::ostringstream os;
  for (size_t i = 0; i < feed_target_shapes.size(); ++i) {
    os << "feed target " << i << ": {" << feed_target_shapes[i][0];
    for (size_t j = 1; j < feed_target_shapes[i].size(); ++j) {
      os << ", " << feed_target_shapes[i][j];
    }
    os << "}\n";
  }
  LOG(INFO) << os.str();
T
tensor-tang 已提交
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268
  if (feed_names) {
    PADDLE_ENFORCE_EQ(feed_names->size(), feed_target_shapes.size());
  }
  std::vector<PaddleTensor> input_slots(feed_target_shapes.size());
  for (size_t i = 0; i < feed_target_shapes.size(); ++i) {
    const auto &feed_shape = feed_target_shapes[i];
    auto &input = input_slots[i];
    std::vector<int> shape({FLAGS_batch_size});
    for (size_t s = 1; s < feed_shape.size(); ++s) {
      shape.push_back(static_cast<int>(feed_shape[s]));
    }
    if (feed_names) {
      input.name = (*feed_names)[i];
    }
    input.shape = shape;
    input.dtype = PaddleDType::FLOAT32;
269
    size_t len = std::accumulate(shape.begin(), shape.end(), size_t{1},
T
tensor-tang 已提交
270 271 272 273 274 275
                                 [](int a, int b) { return a * b; });
    input.data.Resize(len * sizeof(float));
    input.lod.assign({{0, static_cast<size_t>(FLAGS_batch_size)}});
    float *input_data = static_cast<float *>(input.data.data());
    // fill input data, for profile easily, do not use random data here.
    for (size_t j = 0; j < len; ++j) {
N
nhzlx 已提交
276 277
      *(input_data + j) =
          static_cast<float>((j + continuous_inuput_index) % len) / len;
T
tensor-tang 已提交
278
    }
T
Tao Luo 已提交
279 280 281 282
  }
  (*inputs).emplace_back(input_slots);
}

283 284 285 286 287 288 289 290 291 292 293 294
void GetInputPerBatch(const std::vector<std::vector<int64_t>> &in,
                      std::vector<std::vector<int64_t>> *out,
                      std::vector<size_t> *lod, size_t batch_iter,
                      size_t batch_end) {
  lod->clear();
  lod->push_back(0);
  for (auto it = in.begin() + batch_iter; it < in.begin() + batch_end; it++) {
    out->push_back(*it);
    lod->push_back(lod->back() + (*it).size());  // calculate lod
  }
}

L
luotao1 已提交
295 296 297 298 299 300 301 302 303 304 305
void ConvertPaddleTensorToZeroCopyTensor(
    PaddlePredictor *predictor, const std::vector<PaddleTensor> &inputs) {
  for (size_t i = 0; i < inputs.size(); i++) {
    auto input = inputs[i];
    auto tensor = predictor->GetInputTensor(input.name);
    tensor->Reshape(input.shape);
    tensor->SetLoD({input.lod});
    if (input.dtype == PaddleDType::INT64) {
      ZeroCopyTensorAssignData<int64_t>(tensor.get(), input.data);
    } else if (input.dtype == PaddleDType::FLOAT32) {
      ZeroCopyTensorAssignData<float>(tensor.get(), input.data);
L
luotao1 已提交
306 307
    } else if (input.dtype == PaddleDType::INT32) {
      ZeroCopyTensorAssignData<int32_t>(tensor.get(), input.data);
308 309
    } else if (input.dtype == PaddleDType::UINT8) {
      ZeroCopyTensorAssignData<uint8_t>(tensor.get(), input.data);
L
luotao1 已提交
310 311 312 313 314
    } else {
      LOG(ERROR) << "unsupported feed type " << input.dtype;
    }
  }
}
315

L
luotao1 已提交
316 317
void PredictionWarmUp(PaddlePredictor *predictor,
                      const std::vector<std::vector<PaddleTensor>> &inputs,
318
                      std::vector<std::vector<PaddleTensor>> *outputs,
319 320
                      int num_threads, int tid,
                      const VarType::Type data_type = VarType::FP32) {
L
luotao1 已提交
321 322 323 324 325
  int batch_size = FLAGS_batch_size;
  LOG(INFO) << "Running thread " << tid << ", warm up run...";
  if (FLAGS_zero_copy) {
    ConvertPaddleTensorToZeroCopyTensor(predictor, inputs[0]);
  }
326
  outputs->resize(1);
L
luotao1 已提交
327 328 329
  Timer warmup_timer;
  warmup_timer.tic();
  if (!FLAGS_zero_copy) {
330
    predictor->Run(inputs[0], &(*outputs)[0], batch_size);
L
luotao1 已提交
331 332
  } else {
    predictor->ZeroCopyRun();
333
  }
334
  PrintTime(batch_size, 1, num_threads, tid, warmup_timer.toc(), 1, data_type);
L
luotao1 已提交
335 336 337 338
  if (FLAGS_profile) {
    paddle::platform::ResetProfiler();
  }
}
339

L
luotao1 已提交
340 341
void PredictionRun(PaddlePredictor *predictor,
                   const std::vector<std::vector<PaddleTensor>> &inputs,
342
                   std::vector<std::vector<PaddleTensor>> *outputs,
343
                   int num_threads, int tid,
344 345
                   const VarType::Type data_type = VarType::FP32,
                   float *sample_latency = nullptr) {
L
luotao1 已提交
346
  int num_times = FLAGS_repeat;
347
  int iterations = inputs.size();  // process the whole dataset ...
348 349
  if (FLAGS_iterations > 0 &&
      FLAGS_iterations < static_cast<int64_t>(inputs.size()))
350 351 352 353 354
    iterations =
        FLAGS_iterations;  // ... unless the number of iterations is set
  outputs->resize(iterations);
  LOG(INFO) << "Thread " << tid << ", number of threads " << num_threads
            << ", run " << num_times << " times...";
L
luotao1 已提交
355 356
  Timer run_timer;
  double elapsed_time = 0;
Y
Yiqun Liu 已提交
357
#ifdef WITH_GPERFTOOLS
L
luotao1 已提交
358
  ProfilerStart("paddle_inference.prof");
Y
Yiqun Liu 已提交
359
#endif
360
  int predicted_num = 0;
L
luotao1 已提交
361
  if (!FLAGS_zero_copy) {
362
    for (int i = 0; i < iterations; i++) {
363
      run_timer.tic();
L
luotao1 已提交
364
      for (int j = 0; j < num_times; j++) {
365
        predictor->Run(inputs[i], &(*outputs)[i], FLAGS_batch_size);
366
      }
367 368 369 370 371 372
      elapsed_time += run_timer.toc();

      predicted_num += FLAGS_batch_size;
      if (predicted_num % 100 == 0) {
        LOG(INFO) << predicted_num << " samples";
      }
L
luotao1 已提交
373
    }
L
luotao1 已提交
374
  } else {
375
    for (int i = 0; i < iterations; i++) {
L
luotao1 已提交
376 377 378 379 380 381
      ConvertPaddleTensorToZeroCopyTensor(predictor, inputs[i]);
      run_timer.tic();
      for (int j = 0; j < num_times; j++) {
        predictor->ZeroCopyRun();
      }
      elapsed_time += run_timer.toc();
382 383 384 385 386

      predicted_num += FLAGS_batch_size;
      if (predicted_num % 100 == 0) {
        LOG(INFO) << predicted_num << " samples";
      }
L
luotao1 已提交
387 388
    }
  }
389

Y
Yiqun Liu 已提交
390
#ifdef WITH_GPERFTOOLS
L
luotao1 已提交
391
  ProfilerStop();
Y
Yiqun Liu 已提交
392
#endif
N
nhzlx 已提交
393

394 395
  auto batch_latency = elapsed_time / (iterations * num_times);
  PrintTime(FLAGS_batch_size, num_times, num_threads, tid, batch_latency,
396
            iterations, data_type);
397 398 399 400

  if (sample_latency != nullptr)
    *sample_latency = batch_latency / FLAGS_batch_size;

L
luotao1 已提交
401 402 403
  if (FLAGS_record_benchmark) {
    Benchmark benchmark;
    benchmark.SetName(FLAGS_model_name);
404 405
    benchmark.SetBatchSize(FLAGS_batch_size);
    benchmark.SetLatency(batch_latency);
L
luotao1 已提交
406
    benchmark.PersistToFile("benchmark_record.txt");
L
luotao1 已提交
407 408 409
  }
}

L
luotao1 已提交
410 411 412
void TestOneThreadPrediction(
    const PaddlePredictor::Config *config,
    const std::vector<std::vector<PaddleTensor>> &inputs,
413
    std::vector<std::vector<PaddleTensor>> *outputs, bool use_analysis = true,
414 415
    const VarType::Type data_type = VarType::FP32,
    float *sample_latency = nullptr) {
L
luotao1 已提交
416
  auto predictor = CreateTestPredictor(config, use_analysis);
417
  if (FLAGS_warmup) {
418
    PredictionWarmUp(predictor.get(), inputs, outputs, 1, 0, data_type);
419
  }
420 421
  PredictionRun(predictor.get(), inputs, outputs, 1, 0, data_type,
                sample_latency);
L
luotao1 已提交
422 423
}

L
luotao1 已提交
424
void TestMultiThreadPrediction(
425
    const PaddlePredictor::Config *config,
426
    const std::vector<std::vector<PaddleTensor>> &inputs,
427
    std::vector<std::vector<PaddleTensor>> *outputs, int num_threads,
T
Tao Luo 已提交
428
    bool use_analysis = true) {
L
luotao1 已提交
429
  std::vector<std::thread> threads;
L
luotao1 已提交
430 431 432 433 434
  std::vector<std::unique_ptr<PaddlePredictor>> predictors;
  predictors.emplace_back(CreateTestPredictor(config, use_analysis));
  for (int tid = 1; tid < num_threads; tid++) {
    predictors.emplace_back(predictors.front()->Clone());
  }
435

L
luotao1 已提交
436 437 438 439
  for (int tid = 0; tid < num_threads; ++tid) {
    threads.emplace_back([&, tid]() {
      // Each thread should have local inputs and outputs.
      // The inputs of each thread are all the same.
440
      std::vector<std::vector<PaddleTensor>> outputs_tid;
L
luotao1 已提交
441
      auto &predictor = predictors[tid];
442 443 444 445
      if (FLAGS_warmup) {
        PredictionWarmUp(predictor.get(), inputs, &outputs_tid, num_threads,
                         tid);
      }
446
      PredictionRun(predictor.get(), inputs, &outputs_tid, num_threads, tid);
L
luotao1 已提交
447 448 449 450 451 452 453
    });
  }
  for (int i = 0; i < num_threads; ++i) {
    threads[i].join();
  }
}

454
void TestPrediction(const PaddlePredictor::Config *config,
455
                    const std::vector<std::vector<PaddleTensor>> &inputs,
456 457
                    std::vector<std::vector<PaddleTensor>> *outputs,
                    int num_threads, bool use_analysis = FLAGS_use_analysis) {
458
  PrintConfig(config, use_analysis);
L
luotao1 已提交
459
  if (num_threads == 1) {
T
Tao Luo 已提交
460
    TestOneThreadPrediction(config, inputs, outputs, use_analysis);
L
luotao1 已提交
461
  } else {
T
Tao Luo 已提交
462 463
    TestMultiThreadPrediction(config, inputs, outputs, num_threads,
                              use_analysis);
L
luotao1 已提交
464 465 466
  }
}

467 468
void SummarizeAccuracy(float avg_acc_fp32, float avg_acc_int8,
                       int compared_idx) {
469 470 471 472 473 474
  PADDLE_ENFORCE_LE(compared_idx, 2,
                    "Compare either top1 accuracy or mAP (top5), the "
                    "compared_idx is out of range");
  PADDLE_ENFORCE_GE(compared_idx, 1,
                    "Compare either top1 accuracy or mAP (top5), the "
                    "compared_idx is out of range");
475
  std::string prefix = (compared_idx == 1) ? "top1_accuracy " : "mAP ";
476
  LOG(INFO) << "--- Accuracy summary --- ";
477 478 479 480 481 482 483 484
  LOG(INFO) << "Accepted " << prefix
            << "drop threshold: " << FLAGS_quantized_accuracy
            << ". (condition: (FP32_" << prefix << " - INT8_" << prefix
            << ") <= threshold)";
  LOG(INFO) << "FP32: avg " << prefix << std::fixed << std::setw(6)
            << std::setprecision(4) << avg_acc_fp32;
  LOG(INFO) << "INT8: avg " << prefix << std::fixed << std::setw(6)
            << std::setprecision(4) << avg_acc_int8;
485 486
}

487 488 489 490 491 492 493 494
void SummarizePerformance(const char *title, float sample) {
  CHECK_GT(sample, 0.0);
  auto throughput = 1000.0 / sample;
  LOG(INFO) << title << ": avg fps: " << std::fixed << std::setw(6)
            << std::setprecision(4) << throughput << ", avg latency: " << sample
            << " ms";
}

495 496
void SummarizePerformance(float sample_latency_fp32,
                          float sample_latency_int8) {
497 498
  if (FLAGS_enable_fp32) SummarizePerformance("FP32", sample_latency_fp32);
  if (FLAGS_enable_int8) SummarizePerformance("INT8", sample_latency_int8);
499 500
}

501 502
float CompareAccuracyOne(
    const std::vector<std::vector<PaddleTensor>> &output_slots,
503
    int compared_idx) {
504
  if (output_slots.size() == 0)
505
    throw std::invalid_argument(
506
        "CompareAccuracy: output_slots vector is empty.");
507

508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
  float total_accs{0};

  for (size_t i = 0; i < output_slots.size(); ++i) {
    switch (compared_idx) {
      case 1:
        PADDLE_ENFORCE_GE(
            output_slots[i].size(), 2UL,
            "To achieve top 1 accuracy, output_slots_quant[i].size()>=2");
        break;
      case 2:
        PADDLE_ENFORCE_GE(
            output_slots[i].size(), 2UL,
            "To achieve top 1 accuracy, output_slots_ref[i].size()>=2");
        break;
      default:
        throw std::invalid_argument(
            "CompareAccuracy: compared_idx is out of range.");
525 526
    }

527
    if (output_slots[i][compared_idx].lod.size() > 0)
528
      throw std::invalid_argument("CompareAccuracy: output has nonempty LoD.");
529 530

    if (output_slots[i][compared_idx].dtype != paddle::PaddleDType::FLOAT32)
531
      throw std::invalid_argument(
532
          "CompareAccuracy: output is of a wrong type.");
533 534 535

    total_accs +=
        *static_cast<float *>(output_slots[i][compared_idx].data.data());
536
  }
537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559

  CHECK_GT(output_slots.size(), 0);

  return total_accs / output_slots.size();
}

void CompareAccuracy(
    const std::vector<std::vector<PaddleTensor>> &output_slots_quant,
    const std::vector<std::vector<PaddleTensor>> &output_slots_ref,
    int compared_idx) {
  if ((FLAGS_enable_fp32 && FLAGS_enable_int8) &&
      (output_slots_quant.size() == 0 || output_slots_ref.size()) == 0)
    throw std::invalid_argument(
        "CompareAccuracy: output_slots vector is empty.");

  float avg_acc_quant = 0.0;
  float avg_acc_ref = 0.0;

  if (FLAGS_enable_int8)
    avg_acc_quant = CompareAccuracyOne(output_slots_quant, compared_idx);

  if (FLAGS_enable_fp32)
    avg_acc_ref = CompareAccuracyOne(output_slots_ref, compared_idx);
560

561
  SummarizeAccuracy(avg_acc_ref, avg_acc_quant, compared_idx);
562 563 564 565 566 567 568

  if (FLAGS_enable_fp32) CHECK_GT(avg_acc_ref, 0.0);

  if (FLAGS_enable_int8) CHECK_GT(avg_acc_quant, 0.0);

  if (FLAGS_enable_fp32 && FLAGS_enable_int8)
    CHECK_LE(avg_acc_ref - avg_acc_quant, FLAGS_quantized_accuracy);
569 570
}

L
luotao1 已提交
571 572 573 574 575 576 577 578 579
void CompareDeterministic(
    const PaddlePredictor::Config *config,
    const std::vector<std::vector<PaddleTensor>> &inputs) {
  int batch_size = FLAGS_batch_size;
  int num_times = FLAGS_repeat;
  auto predictor = CreateTestPredictor(config, FLAGS_use_analysis);

  std::vector<PaddleTensor> warmup_outputs, outputs;
  // run num_times to Compare Deterministic Result.
580 581 582 583
  for (size_t j = 0; j < inputs.size(); j++) {
    // warmup run
    predictor->Run(inputs[j], &warmup_outputs, batch_size);
    for (int i = 0; i < num_times; i++) {
L
luotao1 已提交
584 585 586 587 588 589
      predictor->Run(inputs[j], &outputs, batch_size);
      CompareResult(outputs, warmup_outputs);
    }
  }
}

T
Tao Luo 已提交
590
void CompareNativeAndAnalysis(
591
    const PaddlePredictor::Config *config,
592
    const std::vector<std::vector<PaddleTensor>> &inputs) {
593
  PrintConfig(config, true);
594
  std::vector<std::vector<PaddleTensor>> native_outputs, analysis_outputs;
595
  TestOneThreadPrediction(config, inputs, &native_outputs, false);
T
Tao Luo 已提交
596
  TestOneThreadPrediction(config, inputs, &analysis_outputs, true);
597 598
  PADDLE_ENFORCE_GT(native_outputs.size(), 0, "Native output is empty.");
  PADDLE_ENFORCE_GT(analysis_outputs.size(), 0, "Analysis output is empty.");
599
  CompareResult(analysis_outputs.back(), native_outputs.back());
T
Tao Luo 已提交
600 601
}

602
void CompareQuantizedAndAnalysis(
603
    const AnalysisConfig *config, const AnalysisConfig *qconfig,
604 605
    const std::vector<std::vector<PaddleTensor>> &inputs,
    const int compared_idx = 1) {
606 607 608 609 610 611 612 613 614
  PADDLE_ENFORCE_EQ(inputs[0][0].shape[0], FLAGS_batch_size,
                    "Input data has to be packed batch by batch.");
  LOG(INFO) << "FP32 & INT8 prediction run: batch_size " << FLAGS_batch_size
            << ", warmup batch size " << FLAGS_warmup_batch_size << ".";

  LOG(INFO) << "--- FP32 prediction start ---";
  auto *cfg = reinterpret_cast<const PaddlePredictor::Config *>(config);
  PrintConfig(cfg, true);
  std::vector<std::vector<PaddleTensor>> analysis_outputs;
615
  float sample_latency_fp32{-1};
616 617 618 619 620

  if (FLAGS_enable_fp32) {
    TestOneThreadPrediction(cfg, inputs, &analysis_outputs, true, VarType::FP32,
                            &sample_latency_fp32);
  }
621 622 623 624 625

  LOG(INFO) << "--- INT8 prediction start ---";
  auto *qcfg = reinterpret_cast<const PaddlePredictor::Config *>(qconfig);
  PrintConfig(qcfg, true);
  std::vector<std::vector<PaddleTensor>> quantized_outputs;
626
  float sample_latency_int8{-1};
627

628 629 630 631
  if (FLAGS_enable_int8) {
    TestOneThreadPrediction(qcfg, inputs, &quantized_outputs, true,
                            VarType::INT8, &sample_latency_int8);
  }
632
  SummarizePerformance(sample_latency_fp32, sample_latency_int8);
633

634
  CompareAccuracy(quantized_outputs, analysis_outputs, compared_idx);
635 636
}

N
nhzlx 已提交
637 638 639 640 641 642 643 644 645 646
void CompareNativeAndAnalysis(
    PaddlePredictor *native_pred, PaddlePredictor *analysis_pred,
    const std::vector<std::vector<PaddleTensor>> &inputs) {
  int batch_size = FLAGS_batch_size;
  std::vector<PaddleTensor> native_outputs, analysis_outputs;
  native_pred->Run(inputs[0], &native_outputs, batch_size);
  analysis_pred->Run(inputs[0], &analysis_outputs, batch_size);
  CompareResult(analysis_outputs, native_outputs);
}

647
void CompareAnalysisAndZeroCopy(
648
    PaddlePredictor::Config *config, PaddlePredictor::Config *config1,
649 650 651 652 653 654 655 656 657
    const std::vector<std::vector<PaddleTensor>> &inputs,
    const std::vector<std::string> &outputs_name) {
  int batch_size = FLAGS_batch_size;
  // analysis
  std::vector<PaddleTensor> analysis_outputs;
  auto predictor = CreateTestPredictor(config, true);
  predictor->Run(inputs[0], &analysis_outputs, batch_size);
  // analysis + zero_copy
  std::vector<ZeroCopyTensor> zerocopy_outputs;
658 659
  reinterpret_cast<AnalysisConfig *>(config1)->SwitchUseFeedFetchOps(false);
  predictor = CreateTestPredictor(config1, true);
660 661 662 663 664 665
  ConvertPaddleTensorToZeroCopyTensor(predictor.get(), inputs[0]);
  predictor->ZeroCopyRun();
  for (size_t i = 0; i < outputs_name.size(); i++) {
    ZeroCopyTensor zerocopy_output =
        *predictor->GetOutputTensor(outputs_name[i]).get();
    zerocopy_outputs.emplace_back(zerocopy_output);
L
luotao1 已提交
666
    LOG(INFO) << "ZeroCopy output: " << DescribeZeroCopyTensor(zerocopy_output);
667 668 669 670 671
  }
  // compare
  CompareResult(analysis_outputs, zerocopy_outputs);
}

672 673 674 675 676 677 678
void SaveOptimModel(AnalysisConfig *cfg, const std::string &dstPath) {
  auto predictor = CreateTestPredictor(
      reinterpret_cast<const PaddlePredictor::Config *>(cfg),
      FLAGS_use_analysis);
  (static_cast<AnalysisPredictor *>(predictor.get()))->SaveOptimModel(dstPath);
}

L
luotao1 已提交
679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749
template <typename T>
std::string LoDTensorSummary(const framework::LoDTensor &tensor) {
  std::stringstream ss;
  ss << "\n---- tensor ---" << '\n';
  ss << "lod: [";
  for (const auto &level : tensor.lod()) {
    ss << "[ ";
    for (auto i : level) {
      ss << i << ", ";
    }
    ss << "]";
  }
  ss << "]\n";

  ss << "shape: [";
  int size = 1;
  for (int i = 0; i < tensor.dims().size(); i++) {
    int dim = tensor.dims()[i];
    ss << dim << ", ";
    size *= dim;
  }
  ss << "]\n";

  ss << "data: ";
  for (int i = 0; i < std::min(20, size); i++) {
    ss << tensor.data<T>()[i] << " ";
  }
  ss << "\n";

  return ss.str();
}

static bool CompareLoD(const framework::LoD &a, const framework::LoD &b) {
  if (a.size() != b.size()) {
    LOG(ERROR) << string::Sprintf("lod size not match %d != %d", a.size(),
                                  b.size());
    return false;
  }
  for (size_t i = 0; i < a.size(); i++) {
    auto &al = a[i];
    auto &bl = b[i];
    if (al.size() != bl.size()) {
      LOG(ERROR) << string::Sprintf("level size %d != %d", al.size(),
                                    bl.size());
      return false;
    }
  }
  return true;
}

static bool CompareShape(const std::vector<int64_t> &a,
                         const std::vector<int64_t> &b) {
  if (a.size() != b.size()) {
    LOG(ERROR) << string::Sprintf("shape size not match %d != %d", a.size(),
                                  b.size());
    return false;
  }
  for (size_t i = 0; i < a.size(); i++) {
    if (a[i] != b[i]) {
      LOG(ERROR) << string::Sprintf("shape %d-th element not match %d != %d", i,
                                    a[i], b[i]);
      return false;
    }
  }
  return true;
}

static bool CompareTensorData(const framework::LoDTensor &a,
                              const framework::LoDTensor &b) {
  auto a_shape = framework::vectorize(a.dims());
  auto b_shape = framework::vectorize(b.dims());
750
  size_t a_size = std::accumulate(a_shape.begin(), a_shape.end(), size_t{1},
L
luotao1 已提交
751
                                  [](int a, int b) { return a * b; });
752
  size_t b_size = std::accumulate(b_shape.begin(), b_shape.end(), size_t{1},
L
luotao1 已提交
753 754 755 756 757 758 759
                                  [](int a, int b) { return a * b; });
  if (a_size != b_size) {
    LOG(ERROR) << string::Sprintf("tensor data size not match, %d != %d",
                                  a_size, b_size);
  }

  for (size_t i = 0; i < a_size; i++) {
760
    if (a.type() == VarType::FP32) {
L
luotao1 已提交
761 762 763 764 765 766 767 768
      const auto *a_data = a.data<float>();
      const auto *b_data = b.data<float>();
      if (std::abs(a_data[i] - b_data[i]) > 1e-3) {
        LOG(ERROR) << string::Sprintf(
            "tensor data %d-th element not match, %f != %f", i, a_data[i],
            b_data[i]);
        return false;
      }
769
    } else if (a.type() == VarType::INT64) {
L
luotao1 已提交
770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800
      const auto *a_data = a.data<int64_t>();
      const auto *b_data = b.data<int64_t>();
      if (std::abs(a_data[i] - b_data[i]) > 1e-3) {
        LOG(ERROR) << string::Sprintf(
            "tensor data %d-th element not match, %f != %f", i, a_data[i],
            b_data[i]);
        return false;
      }
    }
  }

  return true;
}

static bool CompareTensor(const framework::LoDTensor &a,
                          const framework::LoDTensor &b) {
  if (!CompareLoD(a.lod(), b.lod())) {
    return false;
  }
  if (!CompareShape(framework::vectorize(a.dims()),
                    framework::vectorize(b.dims()))) {
    return false;
  }

  if (!CompareTensorData(a, b)) {
    return false;
  }

  return true;
}

L
luotao1 已提交
801 802
}  // namespace inference
}  // namespace paddle