home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.lang.c++
- Path: sparky!uunet!mcsun!ub4b!info-sparc1.info.ucl.ac.be!fynu.ucl.ac.be!vijghen
- From: vijghen@fynu3.fynu.ucl.ac.be (Philippe Vijghen)
- Subject: Re: BC 3.1 TV Running slow ?
- Message-ID: <1992Aug27.073530.9986@info.ucl.ac.be>
- Sender: news@info.ucl.ac.be (News Administrator)
- Nntp-Posting-Host: topo2
- Organization: Free University of Brussels (ULB) - Astrophysic Dpt.
- References: <1992Aug19.232224.3557@cssc-woll.tansu.com.au>
- Date: Thu, 27 Aug 1992 07:35:30 GMT
- Lines: 46
-
- In article <1992Aug19.232224.3557@cssc-woll.tansu.com.au> greg@cssc-woll.tansu.com.au (Pengelly Greg) writes:
- >
- >
- >Problems with Turbo Vision V3.1
- >
- >We have a reasonably large TV application developed under BC V3.0
- >Compiled under BC V3.1 it runs like a dog.
- >The application now runs noticably slower.
- >Compiler switches and source code have not been changed.
- >
- >Has anyone else had this problem. Better still can anyone provide the
- >solution. Help from Borland Australia has sofar been nonexistant.
- >
-
- I had the same problem with a Turbo Vision application and BC++ 3.0
- compiler. I looked around for a long time and I found that the source
- of the problem is the memory allocation mechanism: it uses a kind of
- "safe pool" and a lot of verifications; this certainbly provide advantages
- but the main result is a important reduction of the speed inside the
- routines using dynamic memory allocation (I was doing some symbolic
- computation to implement a fuzzy logic controller).
-
- The only solution I found was to do an "execvp(...)" call to a non
- Turbo Vision excutable! I think another one should be to rewrite the
- allocation function new (Yes, it's possible because Borland provides
- the source of every single function of his libraries!) ...but the
- code should'nt be as portable.
-
- The execvp() was not a problem for my applications because the computation
- time was in days on a 486!!! (because of a Genetic Algorithm optimisation
- with a trcky fitness function...) I understand that for other applications
- it should not be a good solution.
-
- I think that Borland should put an option in his library: speed/security
- for the mem. alloc., don't you?
-
- Anyway, I really enjoy the Borland C++ package (I am waiting for an upgrade
- to the version 3.1) and I think that no compiler is going better an better
- as it does.
-
- Philippe VIJGHEN
- Ingenieur Civil Mecanicien et Electricien (AIrBr)
- Free University of Brussels (ULB)
- Astrophysic -Dpt.
-
- e-mail: vijghen@fynu.ucl.ac.be
-