mcpu options for AMCC440 with fpu

mcpu options for AMCC440 with fpu

[Patrice Bouchand]

[-- Attachment #1: Type: text/plain, Size: 2382 bytes --]

Hello,

   I was searching for the best mcpu option for amcc440, compiling and running a small program you can find here: 
http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core 6, and results are obviously the same using DENX4.0.

--------------------
-bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    1m15.059s
user    1m14.996s
sys     0m0.060s
-bash-3.00# gcc -O2 -mcpu=440 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    0m39.606s
user    0m39.552s
sys     0m0.028s
-bash-3.00# gcc -O2 -mcpu=440fp ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    1m15.016s
user    1m14.988s
sys     0m0.024s
-bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    0m39.577s
user    0m39.544s
sys     0m0.032s
--------------------

* I do not explain why not using FPU gives better results ?

* I did also a test with bzip2.c ( http://pag.csail.mit.edu/~smcc/projects/single-file-programs/bzip2.c ).

----------------------
-bash-3.00# gcc -O2 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2

real    0m18.478s
user    0m18.216s
sys     0m0.152s
-bash-3.00# gcc -O2 -mcpu=440 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2

real    0m18.786s
user    0m18.544s
sys     0m0.128s
-bash-3.00# gcc -O2 -mcpu=440fp ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2

real    0m18.811s
user    0m18.548s
sys     0m0.144s
-bash-3.00# gcc -O2 -mcpu=G3 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2

real    0m18.871s
user    0m18.644s
sys     0m0.104s
----------------------

* mcpu option seems to not have any effect ?

*  My third test was to compile and run mplayer.
- With default options (no mcpu specified), my test mpeg was smooth.
- With mcpu=440fp, my test mpeg was not smooth.

 Can anybody bring me some lights about the right gcc mcpu option for amcc440 ?

Best regards,

       Patrice Bouchand


 		
---------------------------------
 Découvrez une nouvelle façon d'obtenir des réponses à toutes vos questions ! Profitez des connaissances, des opinions et des expériences des internautes sur Yahoo! Questions/Réponses.

[-- Attachment #2: Type: text/html, Size: 3259 bytes --]

Re: mcpu options for AMCC440 with fpu

[Patrice Bouchand]

FYI, I got the following answer on mplayer mailing list:

> Apparently -mcpu=440fp is the one. However, it's possible, that GCC
> "optimizations" actually make the code slower.

Can someone confirme this possiblity ?

Any answer would be greatly appreciated ;) 


    Patrice Bouchand


> Hello,
> 
>    I was searching for the best mcpu option for amcc440, compiling and
> running a small program you can find here: 
> http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core
> 6, and results are obviously the same using DENX4.0.
> 
> --------------------
> -bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot &&
> time ./mandelbrot && rm -f ./mandelbrot
> 
> real    1m15.059s
> user    1m14.996s
> sys     0m0.060s
> -bash-3.00# gcc -O2 -mcpu=440 ./mandelbrot.c -o ./mandelbrot &&
> time ./mandelbrot && rm -f ./mandelbrot
> 
> real    0m39.606s
> user    0m39.552s
> sys     0m0.028s
> -bash-3.00# gcc -O2 -mcpu=440fp ./mandelbrot.c -o ./mandelbrot &&
> time ./mandelbrot && rm -f ./mandelbrot
> 
> real    1m15.016s
> user    1m14.988s
> sys     0m0.024s
> -bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot &&
> time ./mandelbrot && rm -f ./mandelbrot
> 
> real    0m39.577s
> user    0m39.544s
> sys     0m0.032s
> --------------------
> 
> * I do not explain why not using FPU gives better results ?
> 
> * I did also a test with bzip2.c
> ( http://pag.csail.mit.edu/~smcc/projects/single-file-programs/bzip2.c ).
> 
> ----------------------
> -bash-3.00# gcc -O2 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c
> && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> 
> real    0m18.478s
> user    0m18.216s
> sys     0m0.152s
> -bash-3.00# gcc -O2 -mcpu=440 ./bzip2.c -o ./bzip2 &&
> time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> 
> real    0m18.786s
> user    0m18.544s
> sys     0m0.128s
> -bash-3.00# gcc -O2 -mcpu=440fp ./bzip2.c -o ./bzip2 &&
> time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> 
> real    0m18.811s
> user    0m18.548s
> sys     0m0.144s
> -bash-3.00# gcc -O2 -mcpu=G3 ./bzip2.c -o ./bzip2 &&
> time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> 
> real    0m18.871s
> user    0m18.644s
> sys     0m0.104s
> ----------------------
> 
> * mcpu option seems to not have any effect ?
> 
> *  My third test was to compile and run mplayer.
> - With default options (no mcpu specified), my test mpeg was smooth.
> - With mcpu=440fp, my test mpeg was not smooth.
> 
>  Can anybody bring me some lights about the right gcc mcpu option for
> amcc440 ?
> 
> Best regards,
> 
>        Patrice Bouchand
> 
> 
> 
> 
> ______________________________________________________________________
> Découvrez une nouvelle façon d'obtenir des réponses à toutes vos
> questions ! Profitez des connaissances, des opinions et des
> expériences des internautes sur Yahoo! Questions/Réponses.
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded

Re: mcpu options for AMCC440 with fpu

[Magnus Hjorth]

Hi,

Have you tried using the -ffast-math option?

/Magnus


On Wed, 21 Feb 2007 20:47:50 +0100
Patrice Bouchand <pb_fwd_list@yahoo.fr> wrote:

>=20
> FYI, I got the following answer on mplayer mailing list:
>=20
> > Apparently -mcpu=3D440fp is the one. However, it's possible, that GCC
> > "optimizations" actually make the code slower.
>=20
> Can someone confirme this possiblity ?
>=20
> Any answer would be greatly appreciated ;)=20
>=20
>=20
>     Patrice Bouchand
>=20
>=20
> > Hello,
> >=20
> >    I was searching for the best mcpu option for amcc440, compiling and
> > running a small program you can find here:=20
> > http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core
> > 6, and results are obviously the same using DENX4.0.
> >=20
> > --------------------
> > -bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> >=20
> > real    1m15.059s
> > user    1m14.996s
> > sys     0m0.060s
> > -bash-3.00# gcc -O2 -mcpu=3D440 ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> >=20
> > real    0m39.606s
> > user    0m39.552s
> > sys     0m0.028s
> > -bash-3.00# gcc -O2 -mcpu=3D440fp ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> >=20
> > real    1m15.016s
> > user    1m14.988s
> > sys     0m0.024s
> > -bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> >=20
> > real    0m39.577s
> > user    0m39.544s
> > sys     0m0.032s
> > --------------------
> >=20
> > * I do not explain why not using FPU gives better results ?
> >=20
> > * I did also a test with bzip2.c
> > ( http://pag.csail.mit.edu/~smcc/projects/single-file-programs/bzip2.c =
).
> >=20
> > ----------------------
> > -bash-3.00# gcc -O2 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c
> > && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> >=20
> > real    0m18.478s
> > user    0m18.216s
> > sys     0m0.152s
> > -bash-3.00# gcc -O2 -mcpu=3D440 ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> >=20
> > real    0m18.786s
> > user    0m18.544s
> > sys     0m0.128s
> > -bash-3.00# gcc -O2 -mcpu=3D440fp ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> >=20
> > real    0m18.811s
> > user    0m18.548s
> > sys     0m0.144s
> > -bash-3.00# gcc -O2 -mcpu=3DG3 ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> >=20
> > real    0m18.871s
> > user    0m18.644s
> > sys     0m0.104s
> > ----------------------
> >=20
> > * mcpu option seems to not have any effect ?
> >=20
> > *  My third test was to compile and run mplayer.
> > - With default options (no mcpu specified), my test mpeg was smooth.
> > - With mcpu=3D440fp, my test mpeg was not smooth.
> >=20
> >  Can anybody bring me some lights about the right gcc mcpu option for
> > amcc440 ?
> >=20
> > Best regards,
> >=20
> >        Patrice Bouchand
> >=20
> >=20
> >=20
> >=20
> > ______________________________________________________________________
> > D=E9couvrez une nouvelle fa=E7on d'obtenir des r=E9ponses =E0 toutes vos
> > questions ! Profitez des connaissances, des opinions et des
> > exp=E9riences des internautes sur Yahoo! Questions/R=E9ponses.
> > _______________________________________________
> > Linuxppc-embedded mailing list
> > Linuxppc-embedded@ozlabs.org
> > https://ozlabs.org/mailman/listinfo/linuxppc-embedded
>=20
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded

RE : Re: mcpu options for AMCC440 with fpu

[Patrice Bouchand]

[-- Attachment #1: Type: text/plain, Size: 4411 bytes --]


Thanks for your answer but unfortunately, it makes just the thing a bit slower:

-bash-3.00# gcc -O2 -mcpu=440fp -ffast-math ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    1m17.901s
user    1m17.828s
sys     0m0.036s
-bash-3.00# gcc -O2 -mcpu=440 -ffast-math ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    0m41.180s
user    0m41.116s
sys     0m0.040s
-bash-3.00# gcc -O2 -mcpu=G3 -ffast-math ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot

real    1m14.820s
user    1m14.760s
sys     0m0.044s


Magnus Hjorth <mh@omnisys.se> a écrit : Hi,

Have you tried using the -ffast-math option?

/Magnus


On Wed, 21 Feb 2007 20:47:50 +0100
Patrice Bouchand 
 wrote:

> 
> FYI, I got the following answer on mplayer mailing list:
> 
> > Apparently -mcpu=440fp is the one. However, it's possible, that GCC
> > "optimizations" actually make the code slower.
> 
> Can someone confirme this possiblity ?
> 
> Any answer would be greatly appreciated ;) 
> 
> 
>     Patrice Bouchand
> 
> 
> > Hello,
> > 
> >    I was searching for the best mcpu option for amcc440, compiling and
> > running a small program you can find here: 
> > http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core
> > 6, and results are obviously the same using DENX4.0.
> > 
> > --------------------
> > -bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    1m15.059s
> > user    1m14.996s
> > sys     0m0.060s
> > -bash-3.00# gcc -O2 -mcpu=440 ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    0m39.606s
> > user    0m39.552s
> > sys     0m0.028s
> > -bash-3.00# gcc -O2 -mcpu=440fp ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    1m15.016s
> > user    1m14.988s
> > sys     0m0.024s
> > -bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot &&
> > time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    0m39.577s
> > user    0m39.544s
> > sys     0m0.032s
> > --------------------
> > 
> > * I do not explain why not using FPU gives better results ?
> > 
> > * I did also a test with bzip2.c
> > ( http://pag.csail.mit.edu/~smcc/projects/single-file-programs/bzip2.c ).
> > 
> > ----------------------
> > -bash-3.00# gcc -O2 ./bzip2.c -o ./bzip2 && time ./bzip2 ./oggenc.c
> > && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> > 
> > real    0m18.478s
> > user    0m18.216s
> > sys     0m0.152s
> > -bash-3.00# gcc -O2 -mcpu=440 ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> > 
> > real    0m18.786s
> > user    0m18.544s
> > sys     0m0.128s
> > -bash-3.00# gcc -O2 -mcpu=440fp ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> > 
> > real    0m18.811s
> > user    0m18.548s
> > sys     0m0.144s
> > -bash-3.00# gcc -O2 -mcpu=G3 ./bzip2.c -o ./bzip2 &&
> > time ./bzip2 ./oggenc.c && ./bzip2 -d ./oggenc.c.bz2 && rm ./bzip2
> > 
> > real    0m18.871s
> > user    0m18.644s
> > sys     0m0.104s
> > ----------------------
> > 
> > * mcpu option seems to not have any effect ?
> > 
> > *  My third test was to compile and run mplayer.
> > - With default options (no mcpu specified), my test mpeg was smooth.
> > - With mcpu=440fp, my test mpeg was not smooth.
> > 
> >  Can anybody bring me some lights about the right gcc mcpu option for
> > amcc440 ?
> > 
> > Best regards,
> > 
> >        Patrice Bouchand
> > 
> > 
> > 
> > 
> > ______________________________________________________________________
> > Découvrez une nouvelle façon d'obtenir des réponses à toutes vos
> > questions ! Profitez des connaissances, des opinions et des
> > expériences des internautes sur Yahoo! Questions/Réponses.
> > _______________________________________________
> > Linuxppc-embedded mailing list
> > Linuxppc-embedded@ozlabs.org
> > https://ozlabs.org/mailman/listinfo/linuxppc-embedded
> 
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded



 		
---------------------------------
 Découvrez une nouvelle façon d'obtenir des réponses à toutes vos questions ! Profitez des connaissances, des opinions et des expériences des internautes sur Yahoo! Questions/Réponses.

[-- Attachment #2: Type: text/html, Size: 6054 bytes --]

Re: mcpu options for AMCC440 with fpu

[Wolfgang Denk]

In message <600461.55072.qm@web23412.mail.ird.yahoo.com> you wrote:
>
>    I was searching for the best mcpu option for amcc440, compiling and running a small program you can find here: 
> http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core 6, and results are obviously the same using DENX4.0.

In general, you should need to meddle much with such options; the ELDK
default settings are usually OK.

> --------------------
> -bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> 
> real    1m15.059s
> user    1m14.996s
> sys     0m0.060s
> -bash-3.00# gcc -O2 -mcpu=440 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> 
> real    0m39.606s
> user    0m39.552s
> sys     0m0.028s
> -bash-3.00# gcc -O2 -mcpu=440fp ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> 
> real    1m15.016s
> user    1m14.988s
> sys     0m0.024s
> -bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> 
> real    0m39.577s
> user    0m39.544s
> sys     0m0.032s

This looks to me as if you have been runing all these  tests  in  the
SAME  environment,  probably  the /opt/eldk/ppc_4xx root file system,
which is soft-float based. You must not do this.

For tests with FP instructions, you MUST use the  /opt/eldk/ppc_4xxFP
root  file  system.  For  tests  with  soft-float,  you  MUST use the
/opt/eldk/ppc_4xx root file system.

You CANNOT link  an  application  build  with  one  settings  against
libraries using other settings, or you will run into trouble. You
would probably have seen the problems if you had used any test where
the results of the calculations would have been visible.


> * I do not explain why not using FPU gives better results ?

Because you did not run the tests you thought yoiu were running.

> * mcpu option seems to not have any effect ?

Use "gcc -v" to see exactly which options get passed to the  compiler
stages.

> *  My third test was to compile and run mplayer.
> - With default options (no mcpu specified), my test mpeg was smooth.
> - With mcpu=440fp, my test mpeg was not smooth.
> 
>  Can anybody bring me some lights about the right gcc mcpu option for amcc440 ?

Just use the  defaults, unless you know exactly what you are doing.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH, HRB 165235 Munich, CEO: Wolfgang Denk
Office:  Kirchenstr. 5,       D-82194 Groebenzell,            Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de
If there was anything that depressed him more than his own  cynicism,
it was that quite often it still wasn't as cynical as real life.
                                 - Terry Pratchett, _Guards! Guards!_

Re: mcpu options for AMCC440 with fpu

[Patrice Bouchand]

 
Indeed ! I was waiting for warnings or errors while compiling to tell me
I was doing something wrong but I should not.

 Thanks for your time.

   Patrice


> In message <600461.55072.qm@web23412.mail.ird.yahoo.com> you wrote:
> >
> >    I was searching for the best mcpu option for amcc440, compiling and running a small program you can find here: 
> > http://magnux.free.fr/gcc/mandelbrot.c. My distro is a ppc fedora core 6, and results are obviously the same using DENX4.0.
> 
> In general, you should need to meddle much with such options; the ELDK
> default settings are usually OK.
> 
> > --------------------
> > -bash-3.00# gcc -O2 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    1m15.059s
> > user    1m14.996s
> > sys     0m0.060s
> > -bash-3.00# gcc -O2 -mcpu=440 ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    0m39.606s
> > user    0m39.552s
> > sys     0m0.028s
> > -bash-3.00# gcc -O2 -mcpu=440fp ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    1m15.016s
> > user    1m14.988s
> > sys     0m0.024s
> > -bash-3.00# gcc -O2 -msoft-float ./mandelbrot.c -o ./mandelbrot && time ./mandelbrot && rm -f ./mandelbrot
> > 
> > real    0m39.577s
> > user    0m39.544s
> > sys     0m0.032s
> 
> This looks to me as if you have been runing all these  tests  in  the
> SAME  environment,  probably  the /opt/eldk/ppc_4xx root file system,
> which is soft-float based. You must not do this.
> 
> For tests with FP instructions, you MUST use the  /opt/eldk/ppc_4xxFP
> root  file  system.  For  tests  with  soft-float,  you  MUST use the
> /opt/eldk/ppc_4xx root file system.
> 
> You CANNOT link  an  application  build  with  one  settings  against
> libraries using other settings, or you will run into trouble. You
> would probably have seen the problems if you had used any test where
> the results of the calculations would have been visible.
> 
> 
> > * I do not explain why not using FPU gives better results ?
> 
> Because you did not run the tests you thought yoiu were running.
> 
> > * mcpu option seems to not have any effect ?
> 
> Use "gcc -v" to see exactly which options get passed to the  compiler
> stages.
> 
> > *  My third test was to compile and run mplayer.
> > - With default options (no mcpu specified), my test mpeg was smooth.
> > - With mcpu=440fp, my test mpeg was not smooth.
> > 
> >  Can anybody bring me some lights about the right gcc mcpu option for amcc440 ?
> 
> Just use the  defaults, unless you know exactly what you are doing.
> 
> Best regards,
> 
> Wolfgang Denk
> 

Re: mcpu options for AMCC440 with fpu

[Wolfgang Denk]

In message <1172299485.2993.8.camel@localhost.localdomain> you wrote:
>  
> Indeed ! I was waiting for warnings or errors while compiling to tell me
> I was doing something wrong but I should not.

There might be situations where what you did was fully intentional.

"UNIX was not designed to stop you from doing stupid things,  because
that would also stop you from doing clever things."       - Doug Gwyn

:-)


Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH, HRB 165235 Munich, CEO: Wolfgang Denk
Office:  Kirchenstr. 5,       D-82194 Groebenzell,            Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de
History tends to exaggerate.
	-- Col. Green, "The Savage Curtain", stardate 5906.4