cross_engine_reorder.cpp

This C++ API example demonstrates programming flow when reordering memory between CPU and GPU engines.

Annotated version: Reorder between CPU and GPU engines

/*******************************************************************************
* Copyright 2019-2020 Intel Corporation
*
* 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 <iostream>
#include <numeric>
#include <stdexcept>
#include <vector>

// [Prologue]
#include "dnnl.hpp"

#include "example_utils.hpp"

using namespace dnnl;

using namespace std;
// [Prologue]

memory::dim product(const memory::dims &dims) {
    return std::accumulate(dims.begin(), dims.end(), (memory::dim)1,
            std::multiplies<memory::dim>());
}

void fill(memory &mem, const memory::dims &adims) {
    std::vector<float> array(product(adims));
    for (size_t e = 0; e < array.size(); ++e) {
        array[e] = e % 7 ? 1.0f : -1.0f;
    }
    write_to_dnnl_memory(array.data(), mem);
}

int find_negative(memory &mem, const memory::dims &adims) {
    int negs = 0;
    std::vector<float> array(product(adims));
    read_from_dnnl_memory(array.data(), mem);

    for (size_t e = 0; e < adims.size(); ++e) {
        negs += array[e] < 0.0f;
    }
    return negs;
}

void cross_engine_reorder_tutorial() {
    // [Initialize engine]
    auto cpu_engine = engine(engine::kind::cpu, 0);
    auto gpu_engine = engine(engine::kind::gpu, 0);
    // [Initialize engine]

    // [Initialize stream]
    auto stream_gpu = stream(gpu_engine);
    // [Initialize stream]

    //  [reorder cpu2gpu]
    const auto tz = memory::dims {2, 16, 1, 1};
    auto m_cpu
            = memory({{tz}, memory::data_type::f32, memory::format_tag::nchw},
                    cpu_engine);
    auto m_gpu
            = memory({{tz}, memory::data_type::f32, memory::format_tag::nchw},
                    gpu_engine);
    fill(m_cpu, tz);
    auto r1 = reorder(m_cpu, m_gpu);
    //  [reorder cpu2gpu]

    // [Create a ReLU primitive]
    //  ReLU op descriptor (uses a GPU memory as source memory.
    //  no engine- or implementation-specific information)
    auto relu_d = eltwise_forward::desc(prop_kind::forward,
            algorithm::eltwise_relu, m_gpu.get_desc(), 0.0f);
    // ReLU primitive descriptor, which corresponds to a particular
    // implementation in the library. Specify engine type for the ReLU
    // primitive. Use a GPU engine here.
    auto relu_pd = eltwise_forward::primitive_desc(relu_d, gpu_engine);
    // ReLU primitive
    auto relu = eltwise_forward(relu_pd);
    // [Create a ReLU primitive]

    //  [reorder gpu2cpu]
    auto r2 = reorder(m_gpu, m_cpu);
    //  [reorder gpu2cpu]

    // [Execute primitives]
    // wrap source data from CPU to GPU
    r1.execute(stream_gpu, m_cpu, m_gpu);
    // Execute ReLU on a GPU stream
    relu.execute(stream_gpu, {{DNNL_ARG_SRC, m_gpu}, {DNNL_ARG_DST, m_gpu}});
    // Get result data from GPU to CPU
    r2.execute(stream_gpu, m_gpu, m_cpu);

    stream_gpu.wait();
    // [Execute primitives]

    // [Check the results]
    if (find_negative(m_cpu, tz) != 0)
        throw std::logic_error(
                "Unexpected output, find a negative value after the ReLU "
                "execution.");
    // [Check the results]
}

int main(int argc, char **argv) {
    return handle_example_errors({engine::kind::cpu, engine::kind::gpu},
            cross_engine_reorder_tutorial);
}