[MLton-devel] cvs commit: source-level profiling

[Matthew Fluet]

>  All of this is how profiling of "leaf" source-level profiling works.
>  That is, the counts (be they clock ticks or bytes allocated) are
>  associated with leaves in the call graph.  Coming soon, I will add the
>  ability to have the (alloc or time) profiler walk the stack and bump a
>  counter for all the functions on the call stack.  Then, the deeper
>  sources in each sourceSeq and the frameSources will be used to map the
>  SSA call stack into the source call stack.  The mlmon file will still
>  be the same, the counts will just denote how often the function was on
>  the call stack.  This should give a clearer picture of where time is
>  being spent by the major functions program (kind of like our pass
>  timings with MLton -v2).

Is there an ETA on stack walking profiling?  I'm looking at the Basis2002
IO functors and trying to speed up the stream and buffered IO, but the
granularity of the profiling info isn't really helpful right now.  I see
70% of the time in two wc-input1 lambdas, 30% in <unknown> (which seem to
be all equalities looking at the SSA), and the rest in sequence.fun and
list.sml.  None of which helps me narrow in on which part of TextIO.input1
is taking up the majority of the time.

Also, I'm not quite sure which mlprof options are still valid.



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel