On 05/05/2011 11:51 PM, Gordan Bobic wrote: > I think the key thing to consider is cost/benefit. If the code optimized > for A9 runs poorly on A8, but the code optimized for A8 runs on A9 > almost as well as the code optimized for the A9, then it would be better > to optimize for the A8. > > Since A8 doesn't have OoOE, it wouldn't surprise me if latter was the > case (A8 code running imperceptibly worse on A9 but A9 code running > much worse on A8). > > There is also the more philosophical issue - A9 is already faster than > A8, so it may be better to target the A8 on the basis that it is more in > need of that extra boost. Two other thoughts on backward/forward compatibility and tuning: Since there is still going to be an ARMv5 Fedora distribution, no device will be left behind. Some devices may not realize their full potential, but they will continue to work. There have been a number of iterations of what Fedora x86 tunes for as the years have gone by. Maybe Fedora 17+ should be tuned for A15, but we can make the decision when the time comes. As long as we aren't doing something that is going to require ABI breakage later, when A15 is common, why not support the widest range of available ARMv7 hardware there is for the next 6 months? With this in mind, I suggest tuning for A8 on F15 and reevaluating the target again in Fedora 16. -- Brendan Conoboy / Red Hat, Inc. / blc@xxxxxxxxxx _______________________________________________ arm mailing list arm@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/arm
