Re: Code profiling

Chris,

You didn't mention Ian Bartholomew's profiler so I guess you haven't tried that
yet. I don't know whether it would provide precisely the information you are
looking for, but it would certainly help a lot.

It would help, but I don't think it would provide some of the
precision that Jeff is looking for. There are two main problems with
the Profiler as it stands...

- it can only sample at a maximum of 1mS intervals. This was probably
OK on the slower machines that were in use when the Profiler was first
written but is too slow now.

- it will only take a sample when Dolphin allows it to - basically
before and after a "normal" message send. Anything unusual, api calls
/ optimised methods, get included in the enclosing sample.

The original Dolphin Profiler, named Svengali, that appeared back in
the early Dolphin beta could do more (it had a full execution tracer
and a GC monitor IIRC) so it is possible to provide more information -
just a bit difficult to implement.

What I think it really needs is some sort of VM support. Something
like a VM interrupt that, optionally, occurs after every n opcodes are
evaluated?
--
Ian

The From address is valid - for the moment
.

Relevant Pages

  • Re: IDB Profiler
    ... Back in the early days of Dolphin David Jones, ... added a new framework around it - this became my Profiler package. ... Every so many mS the timer in the profiling process expires and the ... pointing at the leaf node of the stack snapshot. ...
    (comp.lang.smalltalk.dolphin)
  • Re: OT : Approximate / fast math libraries ?
    ... Also, I agree with you in general on the precision, just in one particular ... case I know I need much less than I am currently getting, and using atan2 ... Subject: OT: Approximate / fast math libraries? ... for the profiler output. ...
    (Fedora)