/home/brandes/workspace/LAMA/src/lama/cuda/CUDABLAS1.hpp

Go to the documentation of this file.

#ifndef LAMA_CUDABLAS1_HPP_
#define LAMA_CUDABLAS1_HPP_

// for dll_import
#include <lama/config.hpp>

// others
#include <lama/LAMATypes.hpp>
#include <lama/SyncToken.hpp>

#include <cublas.h>
#include <cuda_runtime_api.h>

namespace lama
{

class LAMA_DLL_IMPORTEXPORT CUDABLAS1
{
public:

    template<typename T>
    static void scal(
        const IndexType n,
        const T alpha,
        T* x,
        const IndexType incX,
        SyncToken* syncToken );

    template<typename T>
    static T nrm2( const IndexType n, const T* x, const IndexType incX,
        SyncToken* syncToken );

    template<typename T>
    static T asum( const IndexType n, const T* x, const IndexType incX,
        SyncToken* syncToken );

    template<typename T>
    static IndexType iamax( const IndexType n, const T* x, const IndexType incX,
        SyncToken* syncToken );

    template<typename T>
    static void swap(
        const IndexType n,
        T* y,
        const IndexType incY,
        T* x,
        const IndexType incX,
        SyncToken* syncToken );

    template<typename T>
    static void copy(
        const IndexType n,
        const T* x,
        const IndexType incX,
        T* y,
        const IndexType incY,
        SyncToken* syncToken );

    template<typename T>
    static void axpy(
        const IndexType n,
        const T alpha,
        const T* x,
        const IndexType incX,
        T* y,
        const IndexType incY,
        SyncToken* syncToken );

    template<typename T>
    static T dot(
        const IndexType n,
        const T* x,
        const IndexType incX,
        const T* y,
        const IndexType incY,
        SyncToken* syncToken );

    template<typename T>
    static void sum(
        const IndexType n,
        T alpha,
        const T* x,
        T beta,
        const T* y,
        T* z,
        SyncToken* syncToken );

    template<typename T>
    static void rot(
        const IndexType n,
        T* x,
        const IndexType incX,
        T* y,
        const IndexType incY,
        const T c,
        const T s,
        SyncToken* syncToken );

    template<typename T>
    static void rotm(
        const IndexType n,
        T* x,
        const IndexType incX,
        T* y,
        const IndexType incY,
        const T* P,
        SyncToken* syncToken );

    template<typename T>
    static void ass( const IndexType n, const T value, T* x, SyncToken* syncToken );

    template<typename T>
    static T viamax( const IndexType n, const T* x_d, const IndexType incx, SyncToken* syncToken );

private:

    template <typename T>
    static void ass_launcher( const int n, const T value, T* x, cudaStream_t stream );

    template<typename T>
    static void sum_launcher( const int n, T alpha, const T* x, T beta, const T* y, T* z, cudaStream_t stream );
};

} /* namespace lama */

#endif // LAMA_CUDABLAS1_HPP_