[petsc-dev] (no subject)

[Matthew Knepley]

On Sat, Oct 1, 2011 at 9:08 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
>   Matt,
>
>     I assume you wrote this? Does it make sense ti initialize the
> cublasInit() before setting the device? Seems like the cublasInit() should
> go after the other code?
>

I think you are probably right. I will move it.

  Matt


>   Barry
>
> #if defined(PETSC_HAVE_CUDA)
>  cublasInit();
>
>  ierr =
> PetscOptionsHasName(PETSC_NULL,"-cuda_show_devices",&flg);CHKERRQ(ierr);
>  if (flg) {
>    struct cudaDeviceProp prop;
>    int devCount;
>    int device;
>
>    ierr = cudaGetDeviceCount(&devCount);CHKERRQ(ierr);
>    for(device = 0; device < devCount; ++device) {
>      ierr = cudaGetDeviceProperties(&prop, device);CHKERRQ(ierr);
>      ierr = PetscPrintf(PETSC_COMM_WORLD, "CUDA device %d: %s\n", device,
> prop.name);CHKERRQ(ierr);
>    }
>  }
>  {
>    int device;
>
>    ierr = PetscOptionsGetInt(PETSC_NULL,"-cuda_set_device", &device,
> &flg);CHKERRQ(ierr);
>    if (flg) {
>      ierr = cudaSetDevice(device);CHKERRQ(ierr);
>    }
>  }
> #endif
>
>


-- 
What most experimenters take for granted before they begin their experiments
is infinitely more interesting than any results to which their experiments
lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20111001/0789d47a/attachment.html>