Hi Dan,

Dan Williams wrote:

quoted

Why does it?  I don't see why you'd need to cover the vector-vector case
at all, since it doesn't appear anywhere in the algorithms.  Certainly
going backwards from a vector-vector set to derive if you can do a
scalar-vector multiply when you should have known that in the first
place is not really useful.

    

quoted

Looking closer, the only other caller, async_pq_zero_sum, can be
deleted because it has no users.  So async_pq can become a static
routine in async_r6recov.

      

That's another issue entirely, of course :)

    

...ah, but now I see your point and the root cause of the problem.
The caller needs to be modified to only require scalar multiplication
even in this corner case.
  

Argh... There is no vector-vector multiplication at all! You just need
to swap the 'for' cycles back (as it was in the original patch by me and
Yuri) to get rid of two-dimensional lookup at each step.

Regards, Ilya.