Multiple Cublas functions on single GPU

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

Is it possible to run cublas functions for different matrices at the same time on the same GPU?

For example,

1st block will make an operation on matrix A

and

2nd block will make an operation on matrix B

thanks in advance,

2

Hello,

You want to use cublas function in a kernel func?

I don’t think you can because cublas already uses the cuda api. You may try by using a cublas function in a kernel ie :

#include "cublas.h"

__global__ void kernelfunc() {

   cublasInit();

}
3

I tried it but it caused an error about using a host function in a kernel. I think it does not allow cublas functions in a kernel. :(

4

You have to implement this in an iterative way. But you know that cublas functions are already optimised for speed so you don’t need to change anything.

5

CUBLAS Library is Host API Library (e.g that must be called from the Host) that internally launch GPU kernels.
However, pointers passed to cublas routines must be device pointers.

6

CUBLAS Library is Host API Library (e.g that must be called from the Host) that internally launch GPU kernels.
However, pointers passed to cublas routines must be device pointers.