Re: Can this type of cache miss be reduced?

Eric Fisher <joefoxreal@gmail.com> wrote:


< I'm looking at the data prefetching. There's an example
< of loop spliting in "The compiler Design Handbook"
(snip)


< So we can insert prefetching as


< for (i=0; i<m; i+=n){
< prefetch(a[i+pd]);
< sum[0] += a[i];
< for(i1 = i+1; i1 < min(m, i+n); i1++){
< sum[0] += a[i1];
< }
< }


< I tried this method in my test program. The surprising thing is that the
< performance is degrading due to the loop splitting. Even though the data
< prefetching can get back some benefit, the overall performance is lower
< than before.


What is pd? It would seem to depend on where in a cache line
a[0] is, but you usually don't know that. If you can modify
your allocation routine to allocate on cache line boundaries
it might work better.


Also, there is a fair amount of extra loop overhead that
you have to make up for, including the min calculation
inside the inner loop. I would have written (i1<m && i1<i+n),
though that may not be any better.


-- glen

Return to the comp.compilers page.
Search the comp.compilers archives again.