On 08/02/2013 21:53, Brian Smith wrote:
The assumption here seems to be that it would be OK to ship a significant performance regression for 32-bit users as long as 64-bit users don't see a regression.
It's not. The assumption is that it will be unavoidable to turn off PGO for 32 bit builds (and probably suffer a significant performance regression). Therefore it'd be better than nothing if around 33% of users can keep it through 64 bit builds.
It seems to me that if we don't really have any alternative plans to turning off PGO then it will become a high priority to find all the performance regressions caused by disabling PGO and resolve them.
It's probably possible to compensate the regression, but I don't know if it's achievable to cancel the regression, in other word to make a non-PGO build as fast as a PGO build of the same code. And if possible, it might take a lot of efforts and be fragile.
However if the focus will be on XP users in the next year or two, as Asa describes, then it's probably a good reason not to spend time on 64-bits builds.
_______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
