[MLton] Question on profile.fun

Wesley W. Terpstra wesley@terpstra.ca
Fri, 3 Jun 2005 12:18:09 +0200

On Thu, Jun 02, 2005 at 08:45:55PM -0400, Matthew Fluet wrote:
> I note that the difference between
>   -profile mark -profile-dummy false
> and
>   -profile no 
>   27 incur less than 10%
>   34 incur less than 20%
>   36 incur less than 30%
>   38 incur less than 40%
>   40 incur less than 50%
>   and the last two are at 65% and 603%.

If you are using the move of a constant value into a gc variable, why do you
still need to keep the profiling mark data? It seems to me this is an unfair
comparison. Right now, those extra labels prevent some optimization, right?

If you put in a 'mov' at the profiling entry point _instead_ of a label,
that may incur entirely different penalties than a label and a mov...?

Wesley W. Terpstra