Brad,
On Mon, 6 Jul 2020, Brad Chamberlain wrote:
Or to inject a helper routine that computes the `v[i] * A[i, ..]` product,
storing the result into an array, returning it, and reducing that:
const rows, cols = 1..10;
var A: [rows, cols] real;
var v: [cols] real;
proc dotprod(v, Ai) {
var res = v * Ai;
return res;
}
var z = + reduce [i in rows] dotprod(v, A[i, ..]);
Yes! Well, almost.
After pondering your suggestions during breaks between meetings, you have
given me an idea. I will give up on the 'reduce' and write my code with a
helper routine:
const z : [columns] R = [i in columns] ((dot(v, t[rows, j]) / h) / g);
[i in rows] t[i, columns] += v * v[i];
It works!
This will not be much slower than what I had originally planned when using
the reduce but is slightly more accurate. It is way better than doing
everything on a column-at-a-time basis which was my original approach,
this being a translation of the original algorithm which was written with
Fortran array access in mind. This new way will do writes to memory one
row-at-a-time which is the key performance driver. Now initializing the
vector 'z' will still read from memory one column-at-a-time, but I can
live with that for my current work.
Thanks for the inspiration - Damian
Pacific Engineering Systems International, 277-279 Broadway, Glebe NSW 2037
Ph:+61-2-8571-0847 .. Fx:+61-2-9692-9623 | unsolicited email not wanted here
Views & opinions here are mine and not those of any past or present employer
_______________________________________________
Chapel-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-developers
