[petsc-users] How to compute y = y - A*b

[Timothée Nicolas]

What exactly is the problem? Don't you get good performances with the first
option? I think it is good enough. However, I think it is safer to make
sure that the scalars have the same type. i.e. sometimes you will get
problems passing numerical values to PETSc functions if they are written as
integers when a scalar is expected and vice versa. So I think you should
rather define

one=1.0d0

and then

VecAXPY(y, -one, r);

Best

Timothée


2016-12-22 14:07 GMT+01:00 Florian Lindner <mailinglists at xgm.de>:

> Hello,
>
> what is the best / most efficient way to compute:
>
> y = y - A * b
>
> with vectors b, y and matrix A:
>
> * VecAXPY: I need to compute A*b first
>
> MatMult(A, b, r);
> VecAXPY(y, -1, r);
>
> * VecWAXPY: Same case, but I don't reuse y
>
> MatMult(A, b, r);
> VecWAXPY(w, -1, r, y);
>
> * VecAYPX: Don't work, because I need to multiply r = A*b with -1
>
> Is there anything else I have overseen, or should I just go with VecAXPY?
>
> Best,
> Florian
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20161222/126d62b1/attachment.html>