cublasp.h

Nividia提供的CUDA的BLAS库源码

    int   texYOfs;
};

struct cublasSnrm2Params {
    const float *sx;
    float *result;
    int   n;
    int   incx;
    int   texXOfs;
};

struct cublasScnrm2Params {
    const cuComplex *cx;
    float *result;
    int   n;
    int   incx;
};

struct cublasSrotParams {
    float *sx;
    float *sy;
    int   n;
    int   incx;
    int   incy;
    float sc;
    float ss;
    int   texXOfs;
    int   texYOfs;
};

struct cublasCsrotParams {
    cuComplex *cx;
    cuComplex *cy;
    int   n;
    int   incx;
    int   incy;
    float sc;
    float ss;
};

struct cublasCrotParams {
    cuComplex cs;
    cuComplex *cx;
    cuComplex *cy;
    int   n;
    int   incx;
    int   incy;
    float sc;
};

struct cublasSrotmParams {
    float *sx;                 
    float *sy;                  
    int   n;                   
    int   incx;                 
    int   incy;   
    int   texXOfs;
    int   texYOfs;
    float sparams[CUBLAS_SROTM_PARAM_VEC_LEN]; 
};

struct cublasSgemmParams {
    const float *A;
    const float *B;
    float *C;
    unsigned int   transa;
    unsigned int   transb;
    unsigned int   m;
    unsigned int   n;
    unsigned int   k;
    float alpha;
    unsigned int   lda;
    unsigned int   ldb;
    float beta;
    unsigned int   ldc;
    int texAOfs;
    int texBOfs;
};

struct cublasSsymmParams {
    const float *A;
    const float *B;
    float *C;
    unsigned int   lside;
    unsigned int   upper;
    unsigned int   m;
    unsigned int   n;
    unsigned int   k;
    float alpha;
    unsigned int   lda;
    unsigned int   ldb;
    float beta;
    unsigned int   ldc;
};

struct cublasSsyrkParams {
    const float *A;
    const float *B;
    float *C;
    int   upper;
    int   transpose;
    int   n;
    int   k;
    float alpha;
    int   lda;
    int   ldb;
    float beta;
    int   ldc;
};

struct cublasStrsmParams {
    const float *A;
    float *B;
    int   lside;
    int   upper;
    int   notrans;
    int   nounit;
    int   m;
    int   n;
    float alpha;
    int   lda;
    int   ldb;
};

struct cublasStrmmParams {
    const float *A;
    float *B;
    int   lside;
    int   upper;
    int   notrans;
    int   unit;
    int   m;
    int   n;
    float alpha;
    int   lda;
    int   ldb;
};

struct cublasSgemvParams {
    const float *A;
    const float *x;
    float *y;
    int   m;
    int   n;
    float alpha;
    int   lda;
    int   incx;
    float beta;
    int   incy;
};

struct cublasSsymvParams {
    const float *A;
    const float *x;
    float *y;
    int   up;
    int   n;
    float alpha;
    int   lda;
    int   incx;
    float beta;
    int   incy;
};

struct cublasSgbmvParams {
    const float *A;
    const float *x;
    float *y;
    int   trans;
    int   m;
    int   n;
    int   kl;
    int   ku;
    float alpha;
    int   lda;
    int   incx;
    float beta;
    int   incy;
};

struct cublasSspmvParams {
    const float *AP;
    const float *x;
    float *y;
    int   up;
    int   n;
    float alpha;
    int   incx;
    float beta;
    int   incy;
};

struct cublasStpmvParams {
    const float *AP;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   incx;
};

struct cublasSsyrParams {
    const float *x;
    float *A;
    int   up;
    int   n;
    float alpha;
    int   incx;
    int   lda;
};

struct cublasSsprParams {
    const float *x;
    float *AP;
    int   up;
    int   n;
    float alpha;
    int   incx;
};

struct cublasSgerParams {
    const float *x;
    const float *y;
    float *A;
    int   m;
    int   n;
    float alpha;
    int   incx;
    int   incy;
    int   lda;
};

struct cublasSsyr2Params {
    const float *x;
    const float *y;
    float *A;
    int   up;
    int   m;
    int   n;
    float alpha;
    int   incx;
    int   incy;
    int   lda;
};

struct cublasSspr2Params {
    const float *x;
    const float *y;
    float *AP;
    int   up;
    int   m;
    int   n;
    float alpha;
    int   incx;
    int   incy;
};

struct cublasStrsvParams {
    const float *A;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   lda;
    int   incx;
};

struct cublasStpsvParams {
    const float *AP;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   incx;
};

struct cublasStbsvParams {
    const float *A;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   k;
    int   lda;
    int   incx;
};

struct cublasStrmvParams {
    const float *A;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   lda;
    int   incx;
};

struct cublasStbmvParams {
    const float *A;
    float *x;
    int   up;
    int   trans;
    int   unit;
    int   n;
    int   k;
    int   lda;
    int   incx;
};

struct cublasSsbmvParams {
    const float *A;
    const float *x;
    float *y;
    int   up;
    int   n;
    int   k;
    float alpha;
    int   lda;
    int   incx;
    float beta;
    int   incy;
};

struct cublasCcopyParams {
    const cuComplex *cx; 
    cuComplex *cy;
    int   n;
    int   incx;
    int   incy;
};

struct cublasCaxpyParams {
    cuComplex ca;
    const cuComplex *cx; 
    cuComplex *cy;
    int   n;
    int   incx;
    int   incy;
};

struct cublasCscalParams {
    cuComplex ca;
    cuComplex *cx;
    int   n;
    int   incx;
};

struct cublasCsscalParams {
    cuComplex *cx;
    float sa;
    int   n;
    int   incx;
};

struct cublasCswapParams {
    cuComplex *cx;
    cuComplex *cy;
    int   n;
    int   incx;
    int   incy;
};

struct cublasScasumParams {
    const cuComplex *cx;
    float *result;
    int   n;
    int   incx;
    int   texXOfs;
};

struct cublasCgemmParams {
    cuComplex alpha;
    cuComplex beta;
    const cuComplex *A;
    const cuComplex *B;
    cuComplex *C;
    unsigned int   m;
    unsigned int   n;
    unsigned int   k;
    unsigned int   lda;
    unsigned int   ldb;
    unsigned int   ldc;
    int texAOfs;
    int texBOfs;
};

/* CUBLAS internal functions */
int cublasInitialized (const struct cublasContext *ctx);
void cublasShutDownCtx (struct cublasContext *ctx);
__tlsHookStatus cublasInitCtx (struct cublasContext *ctx, void *_status);
void cublasSetError (struct cublasContext *ctx, cublasStatus error);
void cublasVectorSplay (int n, int tMin, int tMax, int gridW, int *nbrCtas, 
                        int *elemsPerCta, int *threadsPerCta);
void cublasSmallSgemm (struct cublasContext *ctx, char transa, char transb, 
                       int m, int n, int k, float alpha, const float *A, 
                       int lda, const float *B, int ldb, float beta, float *C, 
                       int ldc);
void cublasFastSgemm (struct cublasContext *ctx, char transa, char transb,
                      int m, int n, int k, float alpha, const float *A, 
                      int lda, const float *B, int ldb, float beta, float *C, 
                      int ldc);
void cublasLargeSgemm (struct cublasContext *ctx, char transa, char transb, 
                       int m, int n, int k, float alpha, const float *A, 
                       int lda, const float *B, int ldb, float beta, float *C, 
                       int ldc);
void cublasFastCgemm (struct cublasContext *ctx, char transa, char transb, 
                      int m, int n, int k, cuComplex alpha, const cuComplex *A,
                      int lda, const cuComplex *B, int ldb, cuComplex beta, 
                      cuComplex *C, int ldc);

static int imax(int x, int y)
{
    return (x > y) ? x : y;
}
static int imin(int x, int y)
{
    return (x < y) ? x : y;
}

#include "cublasDblP.h"

#if defined(__cplusplus)
}
#endif /* __cplusplus */


#endif /* !defined(CUBLAS_P_H_) */