>On many (most?) high speed machines, memory-intensive functions like
>bzero or memcpy are memory bandwidth limited by a large margin. You
>may want to benchmark what you described against the "trivial"
>implementation; it may be that they have identical performance limited
>by the memory bus.
I think that the only reason to use MMX is to workaround the bus speed
limit. As you can see with my memcpy.c example this is true (improvement
x2) for large array but not for small array (or at least my test program
tell this).
Andrea[s] Arcangeli
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu