allocate.cpp

void NetworkGenetic::allocateHostAndGPUObjects( size_t deviceRam, size_t hostRam){
    long int hostGenSize = hostRam;
    long int deviceGenSize = deviceRam;
    std::cerr<<"total free device ram : "<<deviceRam<<std::endl;
    std::cerr<<"total free host ram : "<<hostRam<<std::endl;

    //this block below makes sure that the number of objects in stream is exactly half of the total amount of allocatable space on the GPU
    _streamSize = deviceGenSize/(sizeof(double)*2);

    while(_streamSize%_hostParams.array[2] || (_streamSize/_hostParams.array[2])&(_streamSize/_hostParams.array[2]-1)) // get the largest number divisible by the individual size, the threads in a block, and the size of a double
        _streamSize= _streamSize -1;

 

    CUDA_SAFE_CALL(cudaHostAlloc((void**)&host_genetics.array, hostGenSize, cudaHostAllocWriteCombined));
    CUDA_SAFE_CALL(cudaMalloc((void**) &device_genetics.array, deviceGenSize));
    std::fill(host_genetics.array, host_genetics.array+host_genetics.size, 0);
    CUDA_SAFE_CALL(cudaMemset(device_genetics.array, 0, deviceGenSize));
    _stream.resize(_numOfStreams);

    for(int i=0; i<_numOfStreams; i++){
        CUDA_SAFE_CALL(cudaStreamCreate(&_stream.at(i)));
        CUDA_SAFE_CALL( cudaStreamQuery(_stream.at(i)));
    }
    this->confDeviceParams();
}