baum@apple.com (Allen Baum) (04/24/91)
In article <1991Apr23.053619.13474@kithrup.COM>, sef@kithrup.COM (Sean Eric Fagan) writes: > Perhaps that's where a 'don't cache' attribute on a load/store might be > useful... You could probably do it now by having telling the OS to not > cache the page your data are in (although that won't work on all systems). > At least one paper I read advocated this as an optimization- it didn't try to cache data it knew would be used and then thrown away, or that it would store and throw away. It mentioned that it was more difficult if cache lines were longer than 1 word.....hm
meissner@osf.org (Michael Meissner) (04/24/91)
In article <1991Apr23.053619.13474@kithrup.COM>, sef@kithrup.COM (Sean Eric Fagan) writes: | Perhaps that's where a 'don't cache' attribute on a load/store might be | useful... You could probably do it now by having telling the OS to not | cache the page your data are in (although that won't work on all systems). Another place a don't cache attribute is useful is VERY large floating point array code (ie, if the loop is vectorizable, and you have enough registers to unroll the loop to get rid of latencies of bypassing the cache). That way you won't be continually flushing the cache in processing your mega-array. -- Michael Meissner email: meissner@osf.org phone: 617-621-8861 Open Software Foundation, 11 Cambridge Center, Cambridge, MA, 02142 Considering the flames and intolerance, shouldn't USENET be spelled ABUSENET?