Messages in this thread | | | Date | Tue, 9 Nov 1999 14:16:32 -0700 | From | yodaiken@chelm ... | Subject | Re: PATCH 2.3.26: kmalloc GFP_ZERO |
| |
On Tue, Nov 09, 1999 at 10:40:50PM +0100, Andrea Arcangeli wrote: > On Tue, 9 Nov 1999, David S. Miller wrote: > > >No, as Victor has stated you don't call standard memset, you inline > >code it in the architecture specific idle loop and there you can > >check need_resched in between every store instruction if you feel > >it necessary. > > So you'll waste some more instruction cache with the start, stop, restart > from last, stop, restart from last algorithm.
What a cost! Nearly all optimizations are tradeoffs. In this case, you have a easy to show significant decrease in wall clock time of an important benchmark versus the terrible cost of using 2 cache lines in the instruction cache during a time period in which you have nothing else to do.
See: http://www.cs.nmt.edu/~cort/papers/linuxppc-mm/
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |