I am new to cuda, and am trying to understand how cublas works. I read saxpy.cu and saxpy.h in cublas, but have two question:
In saxpy.cu, the declaration is: “host void CUBLASAPI cublasSaxpy (int n, float alpha, const float *x, int incx, float *y, int incy)”. What does “CUBLASAPI” do? I can’t its definition.
Also, there is another subroutine: " cublasVectorSplay (n, CUBLAS_SAXPY_THREAD_MIN, CUBLAS_SAXPY_THREAD_MAX, CUBLAS_SAXPY_CTAS_MAX, &nbrCtas, &elemsPerCta, &threadsPerCta)". I can’t find it in cuda documents, but it seems that cublasVectorSplay computes the size of block and grid. Does anyone know how it works?
Thanks in advance for the questions.