#ifndef __AMSCU_CUDAFUNCTIONS_IMPL_HPP__
#define __AMSCU_CUDAFUNCTIONS_IMPL_HPP__

namespace amscuda
{

//frees devbuffer if it is not already NULL, and sets devbuffer to NULL
//wrapper to cudaFree
template<typename T> int cuda_free(T **devptr)
{
    int ret = 0;
    cudaError_t err = cudaSuccess;

    if(*devptr==NULL)
    {
        return ret; //devbuffer is already NULL/freed
    }

    err = cudaFree(*devptr);
    if(err!=cudaSuccess)
    {
        ret = -1; //failed to free device pointer
        *devptr = NULL; // - ? should only happen if I'm trying to double-free something
    }
    else
    {
        ret = 1;
        *devptr = NULL;
    }

    return ret;
}

//copies hostbuffer to devbuffer
//initializes devbuffer from NULL if devbuffer is NULL
//if overwrite is true, deletes and reallocates devbuffer on device (for resizing)
template<typename T> int buffer_copytodevice(T *hostbuffer, T **devbuffer, long N, bool overwrite)
{
    int ret = 0;
    cudaError_t err = cudaSuccess;

    if(N<=0)
    {
        ret = 0;
        return ret;
    }

    if(hostbuffer==NULL)
    {
        ret = -2; //host buffer is NULL
        return ret;
    }

    if(overwrite==1)
    {
        if(*devbuffer !=NULL)
        {
            cuda_free(devbuffer);
        }
    }

    if(*devbuffer==NULL)
    {
        err = cudaMalloc(devbuffer,sizeof(T)*N);
        if(err!=cudaSuccess)
        {
            ret = -3; //failed to allocate
            *devbuffer = NULL;
            return ret;
        }
    }

    err = cudaMemcpy(*devbuffer,hostbuffer,sizeof(T)*N,cudaMemcpyHostToDevice);
    if(err!=cudaSuccess)
    {
        ret = -4; //failed to copy
    }
    else
    {
        ret = 1;
    }


    return ret;
}

//copies info from devbuffer to hostbuffer
//initialzies hostbuffer from NULL if NULL
//if overwrite is true, deletes and reallocates hostbuffer on host (for resizing)
template<typename T> int buffer_copyfromdevice(T *devbuffer, T **hostbuffer, long N, bool overwrite)
{
    int ret = 0;
    cudaError_t err = cudaSuccess;

    if(N<=0)
    {
        ret = 0;
        return ret;
    }

    if(devbuffer==NULL)
    {
        ret = -5; //null dev buffer
        return ret;
    }

    if(overwrite==1 && *hostbuffer!=NULL)
    {
        delete[] (*hostbuffer); hostbuffer = NULL;
    }

    if(*hostbuffer==NULL)
    {
        *hostbuffer = new(std::nothrow) T[N];
        if(*hostbuffer==NULL)
        {
            ret = -6; //failed to allocate host buffer
            return ret;
        }
    }

    err = cudaMemcpy(*hostbuffer, devbuffer, sizeof(T)*N, cudaMemcpyDeviceToHost);
    if(err!=cudaSuccess)
    {
        ret = -7; //failed to copy
    }
    else
    {
        ret = 1;
    }
    
    return ret;
}

//wrapper for cudaMemcpy - copies an item or struct (count 1) to the device
//initializes devptr from NULL if not already initialized
template<typename T> int cuda_copytodevice(T *hostptr, T **devptr)
{
    int ret = 0;
    cudaError_t err = cudaSuccess;
    bool overwrite = 1;

    if(hostptr==NULL)
    {
        ret = -2; //host buffer is NULL
        return ret;
    }

    if(overwrite==1)
    {
        if(*devptr !=NULL)
        {
            cuda_free(devptr);
        }
    }

    if(*devptr==NULL)
    {
        err = cudaMalloc(devptr,sizeof(T));
        if(err!=cudaSuccess)
        {
            ret = -3; //failed to allocate
            *devptr = NULL;
            return ret;
        }
    }

    err = cudaMemcpy(*devptr,hostptr,sizeof(T),cudaMemcpyHostToDevice);
    if(err!=cudaSuccess)
    {
        ret = -4; //failed to copy
    }
    else
    {
        ret = 1;
    }


    return ret;
}

//wrapper for cudaMemcpy - copies an item or struct (count 1) from device
//initializes hostptr from NULL with new if not already initialized
template<typename T> int cuda_copyfromdevice(T *devptr, T **hostptr)
{
    int ret = 0;
    cudaError_t err = cudaSuccess;
    bool overwrite = 1;

    if(devptr==NULL)
    {
        ret = -5; //null dev buffer
        return ret;
    }

    if(overwrite==1 && *hostptr!=NULL)
    {
        delete (*hostptr); hostptr = NULL;
    }

    if(*hostptr==NULL)
    {
        *hostptr = new(std::nothrow) T;
        if(*hostptr==NULL)
        {
            ret = -6; //failed to allocate host buffer
            return ret;
        }
    }

    err = cudaMemcpy(*hostptr, devptr, sizeof(T), cudaMemcpyDeviceToHost);
    if(err!=cudaSuccess)
    {
        ret = -7; //failed to copy
    }
    else
    {
        ret = 1;
    }
    
    return ret;
}


};

#endif