New profile approach

Yannick PERRET <yperret@ligim.univ-lyon1.fr>
8 Dec 2000 22:21:01 -0500

          From comp.compilers

Related articles
New profile approach yperret@ligim.univ-lyon1.fr (Yannick PERRET) (2000-12-08)
| List of all articles for this month |
From: Yannick PERRET <yperret@ligim.univ-lyon1.fr>
Newsgroups: comp.compilers
Date: 8 Dec 2000 22:21:01 -0500
Organization: UFR Info - Universite Claude Bernard Lyon1
Keywords: testing , GCC
Posted-Date: 08 Dec 2000 22:21:01 EST

Hi,


I was playing with the new (recent) feature of gcc:
-finstrument-functions
which insert a call at each begin and end of functions.


I make a short lib which count the number of entries in each
functions, the time spend inside and this time minus the time spend in
child functions. In other word : prof/gprof


But it appear that their is no time lost for functions that are not
compiled with the good option (-g/-pg for prof/gprof). When you
profile a program that call external functions (not compiled with -pg)
the corresponding time is loose. In consequence it change the time
balance of the program.


With the -finstrument-functions, their is no time lost. This kime of
time is considered as local time for the caller function.




My questions are:
    - how do I know if it is still done by someone else ?
    - is it really interesting (perhaps I dont understand gprof at all...)
?
    - if it is interesting, how to let people contact me, in order to make
better
              treatments (in particular I dont manage functions call-tree) in
my lib
            and not to re-create existing algorithms...






Thank you for your help.
I hope it's the good topic this time...


---
Yannick PERRET, LIGIM, Tel: 72448000 (4210 ou 4211),
Batiment 710, Universite Claude Bernard Lyon 1,
43 boulevard du 11 novembre 1918, 69622 Villeurbanne cedex, France.


Post a followup to this message

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