cublas函数使用问题

[FB1024]

使用cublas库函数cublasSgetrfBatched返回的结果为：CUBLAS_STATUS_EXECUTION_FAILED，改怎样解决呢？
注：头文件添加的“cublas_v2.h”,库链接正常。
部分代码：

float ** mat = new float *[num];//待求逆的矩阵组
float ** invMat = new float *[num];//存放逆矩阵的矩阵组
for(int i = 0; i< num; i++){
cudaMalloc((void **) & mat[i], sizeof(float) * nrcoeffs * nrcoeffs);
cudaMalloc((void **) & invMat[i], sizeof(float) * nrcoeffs * nrcoeffs);
// 这里将矩阵的数据载入mat[i]中,这里假设矩阵的数据在内存中是连续存放的
mat[i]=tmp1->dptr;
invMat[i]=tmp2->dptr;
}
float ** gpuMat;
cudaMalloc((void **) & gpuMat, sizeof(float *) * num);
cudaStatus=cudaMemcpy(gpuMat, mat, sizeof(float *) * num, cudaMemcpyHostToDevice);
if (cudaStatus != cudaSuccess) {
printf ("device mat memory cpy is failed");
}
cubStatus=cublasSgetrfBatched(cubHandle, tmp1->size1, gpuMat, tmp1->size1 , pivo, info, num);
if (cubStatus != CUBLAS_STATUS_SUCCESS) {
fprintf (stderr,"CUBLAS use cublasSgetrfBatched is failed and cubStatus is %ld\n",cubStatus);
if(cubStatus==CUBLAS_STATUS_EXECUTION_FAILED)
fprintf(stderr,"execution faile---------\n");
}