[MLton-devel] icfp?

[Matthew Fluet]

> > Don't the Enter/Leave statements cause changes in the earlier passes stuff?
> > I guess perhaps not much.
>
> It can, but as evidence that it's not often much, here are the latest
> benchmark runtime ratios for -profile time -profile-stack false.

I'm not sure whether or not the run time ratio's are really evidence for
your claim.  There should be some (linear in the length of the program
execution time) overhead in time profiling (without stack walking) just
because of the nature of the time profiling.  So, while I would expect the
run time ratios to all be > 1.00, I wouldn't necessarily expect the
run time ratio - 1.00  to be a constant.  The more interesting, IMHO, test
would be

-mlton "mlton -profile {no, time -profile-stack false} {-drop-pass Contify1,}"

where I would want to see the ratio of the profiled runs to equal the
ratio of the unprofiled runs.

Any guesses as to why some programs run faster with profiling enabled?
Is it really some early pass effect, or it just Henry's "very strong
Heisenburg effects"?




-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel