This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-ssa] PATCH: Feedback-based prefetching

On Fri, 12 Dec 2003, Diego Novillo wrote:
> Finally, there is no guarantee that tree-ssa will be in 3.5.
> We have not beaten the mainline compiler in all areas yet.

Hi Diego,

Is there any chance you could post infrequent tree-ssa branch status
reports, like Gaby does for the 3.3 branch and Mark does for 3.4, so
that its possible to track the progress towards tree-ssa's goals?

I suspect it might be reasonable to relax the mainline merge criteria
if we have some historical data indicating/predicting a significant
rate of improvement.  i.e. it may not be necessary for tree-ssa to
meet all of its goals at the point its merged, if it is likely it will
be an improvement over mainline by the time 3.5 is to be released.

Similarly, if some of the evaluation metrics aren't doing so well,
and after a few tree-ssa status reports its clear that things aren't
improving sufficiently rapidly, then things may need to be rethought.
Perhaps just reprioritizing tree-ssa developers to these deficiencies.
I appreciate "Gerald's testcase" has gone a long way to this already.

If nothing else, a status report provides a forum to publicize the
cumulative effects of the very impressive individual improvements
contributed Jeff and others.  I'm always impressed by the X-fold
improvement patches, but it would be good to know how many more X-fold
improvements are still required.

Just my two bits worth,


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]