gpu_getting_started.cpp

This C++ API example demonstrates programming for Intel(R) Processor Graphics with Intel(R) MKL-DNN.

Annotated version: Getting started on GPU

/*******************************************************************************
* Copyright 2019 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 <sstream>

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

// Optional header to access debug functions like `mkldnn_status2str()`
#include "mkldnn_debug.h"

using namespace mkldnn;

using namespace std;
// [Prologue]

size_t product(const memory::dims adims) {
    size_t n_elems = 1;
    for (size_t d = 0; d < adims.size(); ++d) {
        n_elems *= (size_t)adims[d];
    }
    return n_elems;
}

void fill(const memory &mem, const memory::dims adims) {
    float *array = mem.map_data<float>();

    for (size_t e = 0; e < adims.size(); ++e) {
        array[e] = e % 7 ? 1.0f : -1.0f;
    }

    mem.unmap_data(array);
}

int find_negative(const memory &mem, const memory::dims adims) {
    int negs = 0;

    float *array = mem.map_data<float>();

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

    mem.unmap_data(array);
    return negs;
}

void gpu_getting_started_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,
            { { MKLDNN_ARG_SRC, m_gpu }, { MKLDNN_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) {
        std::stringstream ss;
        ss << "Unexpected output, find a negative value after the ReLU execution";
        throw ss.str();
    }
    // [Check the results]
}


// [Main]
int main(int argc, char **argv) {
    try {
        gpu_getting_started_tutorial();
    } catch (mkldnn::error &e) {
        std::cerr << "Intel MKL-DNN error: " << e.what() << std::endl
            << "Error status: " << mkldnn_status2str(e.status) << std::endl;
        return 1;
    } catch (std::string &e) {
        std::cerr << "Error in the example: " << e << std::endl;
        return 2;
    }

    std::cout << "Example passes" << std::endl;
    return 0;
}
// [Main]