Re: Execution Profiling

icmv!mtranle%caliph@uunet.UU.NET (Minh Tran-Le)
Mon, 27 Apr 1992 05:54:17 GMT

          From comp.compilers

Related articles
Execution Profiling sansom@dcs.glasgow.ac.uk (1992-04-25)
Re: Execution Profiling icmv!mtranle%caliph@uunet.UU.NET (1992-04-27)
| List of all articles for this month |

Newsgroups: comp.compilers
From: icmv!mtranle%caliph@uunet.UU.NET (Minh Tran-Le)
Keywords: experiment, functional
Organization: IntelliCorp Inc.
References: 92-04-132
Date: Mon, 27 Apr 1992 05:54:17 GMT

sansom@dcs.glasgow.ac.uk (Patrick Sansom) writes:


>I am doing some work profiling lazy functional languages and
>would like to get a better feel for what exists out there for
>more conventional languages.


>I have seen stuff on the Unix C profilers prof, gprof and mprof, and
>on the SML profiler, but little else.


>Could you please tell me about your favorite profiling systems (giving
>references to articles etc if known).


I think that the best profiling system that I have used was on the
xerox Interlisp-D Lisp machines with the SPY package.


The system will let you asynchronously start and stop spy with a window
menu on your screen or you can start it by calling a function.


The way that the spy works is that it will interrupt the system and look
on the stack to see what function was used and record it. And you did not
have to compile your code specially to help spy work.


For the analysis phase it will display graphs of all the function calls
and the size of each of the nodes was proportional to the time spent it
that function. This gives direct visual feedback and let you know where
your code spend most of its time and with which path. You could also
apply different filter (cumulative timing, noncumultative ...).


It was a great tool for doing function profiling and I haven't found
any equivalent for unix C profiling. Also I haven't found yet any tool
to do memory profiling.


Minh Tran-Le.
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.