Re: [ublas] numeric bindings - computing workspace requirements

Subject: Re: [ublas] numeric bindings - computing workspace requirements
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2011-01-12 08:22:33

On 01/12/2011 01:30 PM, Peter Man wrote:
>
> However, the user of the wrapped lapack functions doesn't have access to
> the size of this temporary workspace. Since i'm making repeated calls to
> these lapack routines, it seems wasteful that I'm repeatedly creating
> and destroying this temporary workspace. My question is this: have I
> missed something and that the user actually does have some easy way of
> determining the optimal workspace sizes, or should I just create an
> overly large workspace object myself? The problem is that I'm not sure
> how large I should make this workspace object if I choose the latter
> option.
>
> Thanks,
>
> Peter

Hello Peter,

at the moment, this is not supported through the bindings interface. I 
guess we would need to establish a comfortable interface for that, first.

You could inspect Lapack's Fortran code to acquire the original 
optimal-size requirements, and use those to compute the dimensions of a 
custom workspace object, and pass those to the bindings.

Cheers,

Rutger