helper.h

// 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 <glog/logging.h>
#include <sys/time.h>
#include <algorithm>
#include <numeric>
#include <sstream>
#include <string>
#include <vector>
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/inference/api/paddle_inference_api.h"
#include "paddle/fluid/inference/api/timer.h"
#include "paddle/fluid/string/printf.h"

namespace paddle {
namespace inference {

static void split(const std::string &str, char sep,
                  std::vector<std::string> *pieces) {
  pieces->clear();
  if (str.empty()) {
    return;
  }
  size_t pos = 0;
  size_t next = str.find(sep, pos);
  while (next != std::string::npos) {
    pieces->push_back(str.substr(pos, next - pos));
    pos = next + 1;
    next = str.find(sep, pos);
  }
  if (!str.substr(pos).empty()) {
    pieces->push_back(str.substr(pos));
  }
}
static void split_to_float(const std::string &str, char sep,
                           std::vector<float> *fs) {
  std::vector<std::string> pieces;
  split(str, sep, &pieces);
  std::transform(pieces.begin(), pieces.end(), std::back_inserter(*fs),
                 [](const std::string &v) { return std::stof(v); });
}
static void split_to_int64(const std::string &str, char sep,
                           std::vector<int64_t> *is) {
  std::vector<std::string> pieces;
  split(str, sep, &pieces);
  std::transform(pieces.begin(), pieces.end(), std::back_inserter(*is),
                 [](const std::string &v) { return std::stoi(v); });
}
template <typename T>
std::string to_string(const std::vector<T> &vec) {
  std::stringstream ss;
  for (const auto &c : vec) {
    ss << c << " ";
  }
  return ss.str();
}
template <>
std::string to_string<std::vector<float>>(
    const std::vector<std::vector<float>> &vec);

template <>
std::string to_string<std::vector<std::vector<float>>>(
    const std::vector<std::vector<std::vector<float>>> &vec);

template <typename T>
int VecReduceToInt(const std::vector<T> &v) {
  return std::accumulate(v.begin(), v.end(), 1, [](T a, T b) { return a * b; });
}

template <typename T>
static void TensorAssignData(PaddleTensor *tensor,
                             const std::vector<std::vector<T>> &data) {
  // Assign buffer
  int num_elems = VecReduceToInt(tensor->shape);
  tensor->data.Resize(sizeof(T) * num_elems);
  int c = 0;
  for (const auto &f : data) {
    for (T v : f) {
      static_cast<T *>(tensor->data.data())[c++] = v;
    }
  }
}

template <typename T>
static int ZeroCopyTensorAssignData(ZeroCopyTensor *tensor,
                                    const std::vector<std::vector<T>> &data) {
  int size{0};
  auto *ptr = tensor->mutable_data<T>(PaddlePlace::kCPU);
  int c = 0;
  for (const auto &f : data) {
    for (T v : f) {
      ptr[c++] = v;
    }
  }
  return size;
}

static std::string DescribeTensor(const PaddleTensor &tensor) {
  std::stringstream os;
  os << "Tensor [" << tensor.name << "]\n";
  os << " - type: ";
  switch (tensor.dtype) {
    case PaddleDType::FLOAT32:
      os << "float32";
      break;
    case PaddleDType::INT64:
      os << "int64";
      break;
    default:
      os << "unset";
  }
  os << '\n';

  os << " - shape: " << to_string(tensor.shape) << '\n';
  os << " - lod: ";
  for (auto &l : tensor.lod) {
    os << to_string(l) << "; ";
  }
  os << "\n";
  os << " - data: ";

  int dim = VecReduceToInt(tensor.shape);
  for (int i = 0; i < dim; i++) {
    os << static_cast<float *>(tensor.data.data())[i] << " ";
  }
  os << '\n';
  return os.str();
}

static void PrintTime(int batch_size, int repeat, int num_threads, int tid,
                      double latency, int epoch = 1) {
  LOG(INFO) << "====== batch_size: " << batch_size << ", repeat: " << repeat
            << ", threads: " << num_threads << ", thread id: " << tid
            << ", latency: " << latency << "ms ======";
  if (epoch > 1) {
    int samples = batch_size * epoch;
    LOG(INFO) << "====== sample number: " << samples
              << ", average latency of each sample: " << latency / samples
              << "ms ======";
  }
}

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());
  size_t a_size = std::accumulate(a_shape.begin(), a_shape.end(), 1,
                                  [](int a, int b) { return a * b; });
  size_t b_size = std::accumulate(b_shape.begin(), b_shape.end(), 1,
                                  [](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++) {
    if (a.type() == typeid(float)) {
      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;
      }
    } else if (a.type() == typeid(int64_t)) {
      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;
}

}  // namespace inference
}  // namespace paddle