cblas_wrappers/fupermod_cblas.h

#ifndef FUPERMOD_CBLAS_H_
#define FUPERMOD_CBLAS_H_

#include "config.h"
#include "fupermod/fupermod_conf.h"
#include "fupermod/fupermod_bind.h"
#include <gsl/gsl_cblas.h>

#ifdef __cplusplus
extern "C" {
#endif

#ifdef ENABLE_BLAS_SP

  typedef float fupermod_float;
#define FUPERMOD_MPI_FLOAT MPI_FLOAT
#else

  typedef double fupermod_float;
#define FUPERMOD_MPI_FLOAT MPI_DOUBLE
#endif

  typedef struct fupermod_gemm {
      fupermod_process_conf* conf;
    /* time recording*/
    float time;
    void* params;
  } fupermod_gemm;

  fupermod_gemm* fupermod_gemm_alloc(fupermod_process_conf* conf);

  void fupermod_gemm_execute(struct fupermod_gemm* gemm,
      const enum CBLAS_ORDER Order, const enum CBLAS_TRANSPOSE TransA, const enum CBLAS_TRANSPOSE TransB,
      const int M, const int N, const int K,
      const fupermod_float alpha, const fupermod_float *A, const int lda,
      const fupermod_float *B, const int ldb,  const fupermod_float beta,
      fupermod_float *C, const int ldc);

  void fupermod_gemm_free(fupermod_gemm* gemm);

#ifdef __cplusplus
}
#endif

#endif /* FUPERMOD_CBLAS_H_ */