>
>
> Brad Lucier writes:
> > It seems to be fairly fast; e.g., my second generation sparse
> > matrix--vector multiply with a single right hand side takes 17 cycles
> > per flop on my Alpha 21264.
>
> A lot depends on the size of the matrix for whether this really is
> fast. The Alpha is a darn fast chip overall, so even at 17 cycles per
> flop it is pretty quick.
>
> But if memory serves it ought to be able to do 0.5 cycles per flop if
> the data fits in cache. Or 1 cycle per flop, I forget. Either way,
> more than an order of magnitude, and you would see close the machine
> speed in either C or Fortran.
Your reply was quite helpful to me. I looked at the test loop again,
it is really a loop to find the eigenvector associated with the
largest eigenvalue, so it took an L2 norm and did a scaling, too,
each time through the loop. So, in fact, the matrix multiply took
13 cycles/flop, which is not much different, but is closer to the
performance I was aiming for.
If my aim is off, you're free to tell me ;-).
Brad
--------------------- Object Oriented Numerics List --------------------------
* To subscribe/unsubscribe: use the handy web form at
http://oonumerics.org/oon/
* If this doesn't work, please send a note to owner-oon-list@oonumerics.org
This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 03:20:11 EST