[petsc-dev] Implementing longer pipelines with VecDotBegin and VecDotEnd

[Jed Brown]

"Smith, Barry F." <bsmith at mcs.anl.gov> writes:

>   Wim,
>
>  Both approaches are acceptable. For expediency making a pull request with the current code is probably best.

Yup.

>  How dramatic a change is needed in the structure of the VecDotBegin/End constructs? Enough that it is best to start with an entirely new design?

I think it shouldn't require any public interface changes and the
implementation changes would be modest -- just need to hold a list of
batched reductions instead of a single one.