bachelor-thesis/offloading-cacher/main.cpp

#include <iostream>
#include <random>
#include <vector>
#include <string>

#include <omp.h>

#include "cache.hpp"

dsacache::Cache CACHE;

void InitCache(const std::string& device) {
    if (device == "default") {
        auto cache_policy = [](const int numa_dst_node, const int numa_src_node, const size_t data_size) {
            return numa_dst_node;
        };

        auto copy_policy = [](const int numa_dst_node, const int numa_src_node) {
            return std::vector<int>{ numa_src_node, numa_dst_node };
        };

        CACHE.Init(cache_policy,copy_policy);
    }
    else if (device == "xeonmax") {
        auto cache_policy = [](const int numa_dst_node, const int numa_src_node, const size_t data_size) {
            return numa_dst_node < 8 ? numa_dst_node + 8 : numa_dst_node;
        };

        auto copy_policy = [](const int numa_dst_node, const int numa_src_node) {
            const bool same_socket = ((numa_dst_node ^ numa_src_node) & 4) == 0;
            if (same_socket) {
                const bool socket_number = numa_dst_node >> 2;
                if (socket_number == 0) return std::vector<int>{ 0, 1, 2, 3 };
                else return std::vector<int>{ 4, 5, 6, 7 };
            }
            else return std::vector<int>{ numa_src_node, numa_dst_node };
        };

        CACHE.Init(cache_policy,copy_policy);
    }
    else {
        std::cerr << "Given device '" << device << "' not supported!" << std::endl;
        exit(-1);
    }
}

double* GetRandomArray(const size_t size) {
    double* array = new double[size];

    std::uniform_real_distribution<double> unif(std::numeric_limits<double>::min(), std::numeric_limits<double>::max());
    std::default_random_engine re;

    for (size_t i = 0; i < size; i++) {
        array[i] = unif(re);
    }

    return array;
}

bool IsEqual(const double* a, const double* b, const size_t size) {
    for (size_t i = 0; i < size; i++) {
        try {
            if (a[i] != b[i]) return false;
        }
        catch (...) {
            return false;
        }
    }

    return true;
}

void PerformAccessAndTest(double* src, const size_t size, const int tid) {
    std::unique_ptr<dsacache::CacheData> data_cache = CACHE.Access(
            reinterpret_cast<uint8_t *>(src),
            size * sizeof(double)
    );

    double* cached_imm = reinterpret_cast<double *>(data_cache->GetDataLocation());

    // check the value immediately just to see if ram or cache was returned

    if (src == cached_imm) {
        std::cout << "[" << tid << "] Caching did not immediately yield different data location." << std::endl;
    }
    else if (cached_imm == nullptr) {
        std::cout << "[" << tid << "] Immediately got nullptr." << std::endl;
    }
    else {
        std::cout << "[" << tid << "] Immediately got different data location." << std::endl;
    }

    // waits for the completion of the asynchronous caching operation

    data_cache->WaitOnCompletion();

    // gets the cache-data-location from the struct

    double* cached = reinterpret_cast<double *>(data_cache->GetDataLocation());

    // tests on the resulting value

    if (src == cached) {
        std::cout << "[" << tid << "] Caching did not affect data location." << std::endl;
    }
    else if (cached == nullptr) {
        std::cerr << "[" << tid << "] Got nullptr from cache." << std::endl;
    }
    else {
        std::cout << "[" << tid << "] Got different data location from cache." << std::endl;
    }

    if (IsEqual(src,cached,size)) {
        std::cout << "[" << tid << "] Cached data is correct." << std::endl;
    }
    else {
        std::cerr << "[" << tid << "] Cached data is wrong." << std::endl;
    }
}

void RunTestST(const size_t size) {
    double* data = GetRandomArray(size);

    static constexpr int tid = 0;

    std::cout << "[" << tid << "]  first access --- " << std::endl;

    PerformAccessAndTest(data, size, tid);

    std::cout << "[" << tid << "]  second access --- " << std::endl;

    PerformAccessAndTest(data, size, tid);

    std::cout << "[" << tid << "]  end of application --- " << std::endl;
}

void RunTestMT(const size_t size) {
    double* data = GetRandomArray(size);

    #pragma omp parallel
    {
        const int tid = omp_get_thread_num();

        std::cout << "[" << tid << "] first access --- " << std::endl;

        PerformAccessAndTest(data, size, tid);

        std::cout << "[" << tid << "] second access --- " << std::endl;

        PerformAccessAndTest(data, size, tid);

        std::cout << "[" << tid << "] end of block --- " << std::endl;
    }
}

int main(int argc, char **argv) {
    if (argc != 4) {
        std::cerr << "This application requires four parameters!" << std::endl;

        std::cout << "Please provide the following positional arguments: [device] [mode] [size]" << std::endl;
        std::cout << "[device] from { default, xeonmax } which influences cache and execution placement" << std::endl;
        std::cout << "[mode] from { st, mt } or single and multi threaded respectively" << std::endl;
        std::cout << "[size] positive integral number, amount of float64 in data array" << std::endl;

        exit(-1);
    }

    const std::string device = argv[1];
    const std::string mode = argv[2];
    const std::string size_s = argv[3];

    uint32_t size = 0;

    try {
        size = std::stoul(size_s);
    }
    catch (...) {
        std::cerr << "Given Size '" << size_s << "' caused error during conversion to number!" << std::endl;
    }

    InitCache(device);

    if (mode == "st") {
        RunTestST(size);
    }
    else if (mode == "mt") {
        RunTestMT(size);
    }
    else {
        std::cerr << "Given Mode '" << mode << "' not supported!" << std::endl;
        exit(-1);
    }
}