caml-list - the Caml user's mailing list
 help / color / mirror / Atom feed
From: Richard Jones <rich@annexia.org>
To: Arthur Chan <baguasquirrel@gmail.com>
Cc: Jake Donham <jake@donham.org>,
	caml-list@yquem.inria.fr, Raj Bandyopadhyay <rajb@rice.edu>
Subject: Re: [Caml-list] Q: Profiling ocaml using gprof
Date: Tue, 15 Jul 2008 10:12:18 +0100	[thread overview]
Message-ID: <20080715091218.GA20823@annexia.org> (raw)
In-Reply-To: <74cabd9e0807142242n3854dcf7w6d0cfacb0ca6d6a3@mail.gmail.com>

On Mon, Jul 14, 2008 at 10:42:39PM -0700, Arthur Chan wrote:
> Is gprof better for profiling ocaml than ocaml's own profilers?

They are slightly different.  I use 'gprof' all the time because I
tend to only use natively compiled executables.  'gprof' is the
ordinary GNU profiling tool that tells you which function is being run
most often and some limited information about the call path into that
function.  It's pretty useful for simple profiling where you're
looking for obvious problems.

'ocamlprof' is a bit different.  Last time I used it [which was a few
years ago, so maybe it's different now], it only worked on bytecode.
It outputs your original code with annotations telling you how often
each expression was run.  So this isn't time taken (each expression
can take a different amount of time to execute, and this time isn't
shown), but how often a particular path through the code is taken.

> How would you go about figuring out that that particular function stub is
> string concat?

> 'camlPervasives__$5e_136'.

In 'gprof' there's a simple name mangling used to map OCaml function
names to assembler symbols.  Once you understand it, you'll find it
easy to follow.  First of all note that OCaml function names aren't
unique, eg in the following code:

  let f () = printf "this is the first f()\n"
  let f () = printf "this is the second f()\n"; f () ;;
  
  f ()

The assembler symbol is:

  "caml" ^ Modulename ^ "__" ^ functionname ^ "_" ^ uniquenumber

'uniquenumber' is just a counter added to function names by the
compiler to make sure that functions which have the same name will
have different symbols.

So when I compiled the code above in a file called 'test.ml' (hence a
module called Test), in my case I ended up with two symbols called:

  camlTest__f_77
  camlTest__f_78

where '77' and '78' are arbitrary.  You can check this by looking at
the assembler output from the compiler (ocamlopt -S).

If a function name contains an operator symbol (eg. (^) ) then a $xx
hex code is used.

I guess in theory one could write an OCaml symbol filter, similar to
c++filt [http://sourceware.org/binutils/docs/binutils/c_002b_002bfilt.html]

Rich.

-- 
Richard Jones
Red Hat


  reply	other threads:[~2008-07-15  9:12 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-07-15  1:52 Raj Bandyopadhyay
2008-07-15  4:09 ` [Caml-list] " Jake Donham
2008-07-15  5:42   ` Arthur Chan
2008-07-15  9:12     ` Richard Jones [this message]
2008-07-15  7:11   ` Vincent Hanquez

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20080715091218.GA20823@annexia.org \
    --to=rich@annexia.org \
    --cc=baguasquirrel@gmail.com \
    --cc=caml-list@yquem.inria.fr \
    --cc=jake@donham.org \
    --cc=rajb@rice.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).