I'm looking for good compiler settings to made fastest win32 console
application which heavily uses floating point operations. Currently I use:
  CFLAGS= -o -v0 -cpp -mn -6 -ff -WA -I$(CPATH) -Ae
but I'm not sure they are best. I have this application compiled with 5
free-like compilers and DMC++ version is slowest :-( Does linker settings
influence speed much ? If so what can be improved in
  LFLAGS= /B /N /NOL /P:32

I know I can use profiling for optimizing code but this would influence all compilations. I first want to get best settings specific for DMC++.

Thanks in advance, ABX

"W³odzimierz Skiba" <abx@abx.art.pl> wrote in message news:b3ldsp$1dn9$1@digitaldaemon.com...
> I'm looking for good compiler settings to made fastest win32 console
> application which heavily uses floating point operations. Currently I use:
>   CFLAGS= -o -v0 -cpp -mn -6 -ff -WA -I$(CPATH) -Ae
> but I'm not sure they are best. I have this application compiled with 5
> free-like compilers and DMC++ version is slowest :-( Does linker settings
> influence speed much ? If so what can be improved in
>   LFLAGS= /B /N /NOL /P:32

-o -ff -6

should do the trick.

> I know I can use profiling for optimizing code but this would influence
all
> compilations. I first want to get best settings specific for DMC++.

If you could track down just what is slower about it, I can have a go at fixing it.

"Walter" <walter@digitalmars.com> wrote in news:b3lo73$1k95$1@digitaldaemon.com:
> >   CFLAGS= -o -v0 -cpp -mn -6 -ff -WA -I$(CPATH) -Ae
> 
> -o -ff -6
> should do the trick.

This is already used.

> > I know I can use profiling for optimizing code but this would
> > influence  all
> > compilations. I first want to get best settings specific for DMC++.
> 
> If you could track down just what is slower about it, I can have a go at fixing it.

I really would like to help making DMC++ faster in float operations but I
have no idea how to extract slowest part. The application is large and
complicated. First of all I would like to play with recompilation of
snn.lib as suggested at http://cmeerw.org/prog/dm/rtl.html . Can it improve
speed? What are the other ways I can follow to track speed differences
between various compilations? The summary is available at
http://news.povray.org/povray.unofficial.patches/30656/
(skip setup page).

ABX

I think that problems could be due to -Ae switch. Code with exceptions handling is usually slower than that without exceptions.

And what CPU you run tests on? I think DMC is much faster on Intels than on AMDs or others.

Nic Tiger.

"W³odzimierz Skiba" <abx@abx.art.pl> ÓÏÏÂÝÉÌ/ÓÏÏÂÝÉÌÁ × ÎÏ×ÏÓÔÑÈ ÓÌÅÄÕÀÝÅÅ: news:b3n7mn$2f6p$1@digitaldaemon.com...
> "Walter" <walter@digitalmars.com> wrote in news:b3lo73$1k95$1@digitaldaemon.com:
> > >   CFLAGS= -o -v0 -cpp -mn -6 -ff -WA -I$(CPATH) -Ae
> >
> > -o -ff -6
> > should do the trick.
>
> This is already used.
>
> > > I know I can use profiling for optimizing code but this would
> > > influence  all
> > > compilations. I first want to get best settings specific for DMC++.
> >
> > If you could track down just what is slower about it, I can have a go at fixing it.
>
> I really would like to help making DMC++ faster in float operations but I have no idea how to extract slowest part. The application is large and complicated. First of all I would like to play with recompilation of snn.lib as suggested at http://cmeerw.org/prog/dm/rtl.html . Can it
improve
> speed? What are the other ways I can follow to track speed differences
> between various compilations? The summary is available at
> http://news.povray.org/povray.unofficial.patches/30656/
> (skip setup page).
>
> ABX

"Nic Tiger" <nictiger@progtech.ru> wrote in news:b3npni$2obf$1@digitaldaemon.com:

> I think that problems could be due to -Ae switch. Code with exceptions handling is usually slower than that without exceptions.

Are you refering to exception handling in DMC or in general? Exception was necessary due to features in code. I can't remove it from code. It is part of official binaries.

> And what CPU you run tests on? I think DMC is much faster on Intels than on AMDs or others.

As written at mentioned url, I used Intel Pentium II 233 MHz with NT4.

ABX

"W³odzimierz Skiba" <abx@abx.art.pl> wrote in message news:b3n7mn$2f6p$1@digitaldaemon.com...
> I really would like to help making DMC++ faster in float operations but I have no idea how to extract slowest part. The application is large and complicated.

One way is to try the profiler, www.digitalmars.com/ctg/trace.html


> First of all I would like to play with recompilation of
> snn.lib as suggested at http://cmeerw.org/prog/dm/rtl.html . Can it
improve
> speed?

No, I don't think that will help.

> What are the other ways I can follow to track speed differences between various compilations?

Are you using a high speed timer?

> The summary is available at http://news.povray.org/povray.unofficial.patches/30656/

"Walter" <walter@digitalmars.com> wrote in news:b3o8c1$98$4@digitaldaemon.com:
> "W³odzimierz Skiba" <abx@abx.art.pl> wrote in message news:b3n7mn$2f6p$1@digitaldaemon.com...
> > I really would like to help making DMC++ faster in float operations but I have no idea how to extract slowest part. The application is large and complicated.
> 
> One way is to try the profiler, www.digitalmars.com/ctg/trace.html

Ok. I have output of profiled MinGW console compilation and output of profiler used in DMC. Is there easy way to compare it or I have to explore it myself line by line ?

> > What are the other ways I can follow to track speed differences between various compilations?
> 
> Are you using a high speed timer?

I think "yes".

> > The summary is available at http://news.povray.org/povray.unofficial.patches/30656/

latest measurement at http://news.povray.org/ou566vgg0gqih2vvlsovillshbf09t7r4g%404ax.com

ABX

"W³odzimierz Skiba" <abx@abx.art.pl> wrote in message news:b44i9u$1gis$1@digitaldaemon.com...
> Ok. I have output of profiled MinGW console compilation and output of profiler used in DMC. Is there easy way to compare it or I have to explore it myself line by line ?

I think you'll have to explore it line by line.