Function BM_ReluFloat

tensorflow/core/kernels/nn_ops_test.cc:1198–1246

Relu Op Run benchmark with: */

Run benchmark with:
*/
static void BM_ReluFloat(int iters, int batch_size, int rows, int cols,
                         int depth, int num_threads, const string& label) {
  tensorflow::testing::StopTiming();
  std::unique_ptr<Device> device(
      DeviceFactory::NewDevice("CPU", {}, "/job:a/replica:0/task:0"));

  thread::ThreadPool threadpool(Env::Default(), "test", num_threads);
  Eigen::ThreadPoolDevice eigen_cpu_device(threadpool.AsEigenThreadPool(),
                                           num_threads);
  device->set_eigen_cpu_device(&eigen_cpu_device);

  gtl::InlinedVector<TensorValue, 4> inputs;
  TensorShape shape1({batch_size, rows, cols, depth});
  Tensor input1(DT_FLOAT, shape1);
  test::FillIota<float>(&input1, 1.0);
  inputs.push_back({nullptr, &input1});

  // Reluing op.
  NodeDef relu_node_def;
  Status status = NodeDefBuilder("relu_op", "Relu")
                      .Input(FakeInput(DT_FLOAT))
                      .Finalize(&relu_node_def);
  TF_CHECK_OK(status);
  std::unique_ptr<OpKernel> op(CreateOpKernel(DEVICE_CPU, device.get(),
                                              cpu_allocator(), relu_node_def,
                                              TF_GRAPH_DEF_VERSION, &status));
  TF_CHECK_OK(status);
  OpKernelContext::Params params;
  params.device = device.get();
  params.frame_iter = FrameAndIter(0, 0);
  params.inputs = &inputs;
  params.op_kernel = op.get();
  std::vector<AllocatorAttributes> attrs;
  test::SetOutputAttrs(&params, &attrs);

  std::unique_ptr<OpKernelContext> relu_context(new OpKernelContext(&params));

  op->Compute(relu_context.get());
  testing::UseRealTime();
  tensorflow::testing::StartTiming();
  for (int i = 0; i < iters; ++i) {
    delete relu_context->release_output(0).tensor;
    op->Compute(relu_context.get());
  }
  tensorflow::testing::StopTiming();
  testing::ItemsProcessed(relu_context->mutable_output(0)->NumElements() *
                          iters);
  testing::SetLabel(label);
}

// BS: batch_size
// IR: input_rows