/home/brandes/workspace/LAMA/src/lama/cuda/CUDABLAS3.hpp

Go to the documentation of this file.

#ifndef LAMA_CUDABLAS3_HPP_
#define LAMA_CUDABLAS3_HPP_

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

// others
#include <lama/openmp/BLASHelper.hpp>
#include <lama/LAMATypes.hpp>

// logging
#include <logging/logging.hpp>

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

namespace lama
{

class LAMA_DLL_IMPORTEXPORT CUDABLAS3
{
public:
    template<typename T>
    static void gemm(
        const enum CBLAS_ORDER order,
        const enum CBLAS_TRANSPOSE TransA,
        const enum CBLAS_TRANSPOSE TransB,
        const IndexType M,
        const IndexType N,
        const IndexType K,
        const T alpha,
        const T* A,
        const IndexType lda,
        const T* B,
        const IndexType ldb,
        const T beta,
        T* C,
        const IndexType ldc,
        class SyncToken* syncToken );

//    template<typename T>
//    static void symm(
//        const enum CBLAS_ORDER order,
//        const enum CBLAS_SIDE side,
//        const enum CBLAS_UPLO uplo,
//        const IndexType m,
//        const IndexType n,
//        const T alpha,
//        const T *A,
//        const IndexType lda,
//        const T *B,
//        const IndexType ldb,
//        const T beta,
//        T *C,
//        const IndexType ldc);

//    template<typename T>
//    static void trmm(
//        const enum CBLAS_ORDER order,
//        const enum CBLAS_SIDE Side,
//        const enum CBLAS_UPLO Uplo,
//        const enum CBLAS_TRANSPOSE TransA,
//        const enum CBLAS_DIAG Diag,
//        const IndexType M,
//        const IndexType N,
//        const T alpha,
//        const T *A,
//        const IndexType lda,
//        T *B,
//        const IndexType ldb);

    template<typename T>
    static void trsm(
        const enum CBLAS_ORDER order,
        const enum CBLAS_SIDE Side,
        const enum CBLAS_UPLO Uplo,
        const enum CBLAS_TRANSPOSE TransA,
        const enum CBLAS_DIAG Diag,
        const IndexType M,
        const IndexType N,
        const T alpha,
        const T* A,
        const IndexType lda,
        T* B,
        const IndexType ldb,
        class SyncToken* syncToken );

//    template<typename T>
//    static void syrk(
//        const enum CBLAS_ORDER order,
//        const enum CBLAS_UPLO uplo,
//        const enum CBLAS_TRANSPOSE trans,
//        const IndexType n,
//        const IndexType k,
//        const T alpha,
//        const T* A,
//        const IndexType lda,
//        const T beta,
//        T* C,
//        const IndexType ldc );

//    template<typename T>
//    static void syrk2(
//        const enum CBLAS_ORDER order,
//        const enum CBLAS_UPLO uplo,
//        const enum CBLAS_TRANSPOSE trans,
//        const IndexType n,
//        const IndexType k,
//        const T alpha,
//        const T* A,
//        const IndexType lda,
//        const T* B,
//        const IndexType ldb,
//        const T beta,
//        T* C,
//        const IndexType ldc );
private:

    LAMA_LOG_DECL_STATIC_LOGGER( logger );

    template <typename T>
    void gemm_launcher(
        const char transA_char,
        const char transB_char,
        const int m,
        const int n,
        const int k,
        const T alpha,
        const T* const A,
        const int lda,
        const T* const B,
        const int ldb,
        const T beta,
        T* const C,
        const int ldc,
        cudaStream_t cs );
};

} /* namespace lama */

#endif // LAMA_CUDABLAS3_HPP_