This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Slowness problem due to sjlj-exceptions for Octave

Hello Brian

This is Tatsuro Matsuoka replying.

>Here I meant to add that of course we have discussed deprecating
>-mno-cygwin and offering a real MinGW cross in its place, which would
>have its own back-end, and thus could have a different exception
>handling model.  But by the time you've done all that, you might as well
>just package a separate gcc named "gcc-dw2" or whatever and have them
>both installable in parallel, to please everybody.  This is what Danny
>did with his 4.2 MinGW releases.

To whom you wrote this message? To me?

I am now using three different compliers on the cygwin.

gcc-4.2.1 (only gcc, gfortran, g++).
gcc-3.4.4-3 configured with '--disable-sjlj-exceptions' (only gcc, 
gfortran, g++)
standard gcc-3.4.4-3 in cygwin package.

Usually first I try gcc-4.2.1 because its optimizing ability is good. 
I have build the cvs version gnuplot, it works well and very fast.

The building by gcc-4.2.1 sometimes fails. Second I use standard 
cygwin gcc complier.  Because the ordinary programs do not use the 
unwind exceptions, the gcc configured with sjlj-exceptions has no 
slowness problem.  It is favorable to use standard complier if it 
makes no slowness.

In octave case, I am using the complier gcc-3.4.4-3 configured with '-
-disable-sjlj-exceptions'.  Building octave by standard complier, the 
slowness appears.   Especially the interpreter speed slows very much. 
When most part of the program to be vectorized, the  calculation 
speed is not so slow.  But I have to use iteration or iterated 
estimation the function value inside octave (for example ODE solver) 
occurs, the slowness is not allowable.  So I tried make octave by gcc 
configured with ?disable-sjlj-exception because I found the following 

>just package a separate gcc named "gcc-dw2" or whatever and have them
>both installable in parallel, to please everybody.

I install of course two my own GCCs placed at the different 
directories from the standard one.   If the "gcc-dw2" will be packaged,
where is the suitable directory it will be placed.  I think the 
directories where the standard complier used is to be avoided.




Easy + Joy + Powerful = Yahoo! Bookmarks x Toolbar

Unsubscribe info:
Problem reports:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]