|
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Hello POV Team,
I hope you can give me some hint about this topic...
I've tried to compile PovWin with MinGW 3.3 (gcc 3.2.3).
I had to comment out a couple of functions from pvupdate.cpp (reg_printf and
GetInstallTime) because of error messages complaining about redefinition.
(I didn't try to investigate this more deeply).
After this, I obtained a working pvengine.exe.
Rendering a sample scene with this compiled-by-me version takes about 90'',
while with the official executable rendering time is about 30''!
My compile was done with -O3 option (full optimisation for speed, I think).
I had to add this in the makefile as it was not present in the "release"
build options. Without this the rendering time was even longer.
Now, I know from reading some posts in this group that the POV team uses the
Intel compiler, that produces very fast executables, but a difference of
three times seems a bit strange to me.
Can you tell me whether this is just what you expect from these compilers or
I should adjust optimisation options differently?
Can you give me suggestions about what optimisations I should use?
Thanks in advance,
acasta
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
acasta wrote:
> [...]
>
> Can you tell me whether this is just what you expect from these compilers or
> I should adjust optimisation options differently?
> Can you give me suggestions about what optimisations I should use?
It is well known that MinGW produces very slow executables. This is not
due to the compiler itself though (since on other platforms this is not
a problem). On Intel systems you should always make sure you use
-march=pentium4 (or whatever suits).
Christoph
--
POV-Ray tutorials, include files, Sim-POV,
HCR-Edit and more: http://www.tu-bs.de/~y0013390/
Last updated 06 Jul. 2004 _____./\/^>_*_<^\/\.______
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Wed, 28 Jul 2004 04:22:39 EDT, "acasta" <acastagnini AT yahoo DOT com>
wrote:
> I've tried to compile PovWin with MinGW 3.3 (gcc 3.2.3).
> I had to comment out a couple of functions from pvupdate.cpp (reg_printf and
> GetInstallTime) because of error messages complaining about redefinition.
> (I didn't try to investigate this more deeply).
> After this, I obtained a working pvengine.exe.
>
> Rendering a sample scene with this compiled-by-me version takes about 90'',
> while with the official executable rendering time is about 30''!
What's comparison between rendering times under MinGW GUI and MinGW console
builds ?
> Now, I know from reading some posts in this group that the POV team uses the
> Intel compiler, that produces very fast executables, but a difference of
> three times seems a bit strange to me.
BTW, I did not tried it myself but there is free console version of VisualC
somewhere at Microsoft site. You could optionally try it.
> Can you tell me whether this is just what you expect from these compilers or
> I should adjust optimisation options differently?
The difference in speed is "a little" large but OTOH note POV-Ray is threading
software with exception handling. I do not know how GCC of MinGW handles
threads and exceptions in details but there can be source of difference.
> Can you give me suggestions about what optimisations I should use?
I never focused the optimisations under MinGW but I imagine you could find
some help within povray.unix group since they are more experienced in
optimisations of GCC.
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
ABX <abx### [at] abxartpl> wrote:
> > Rendering a sample scene with this compiled-by-me version takes about 90'',
> > while with the official executable rendering time is about 30''!
>
> What's comparison between rendering times under MinGW GUI and MinGW console
> builds ?
There is not much difference (80'' instead of 90'').
> BTW, I did not tried it myself but there is free console version of VisualC
> somewhere at Microsoft site. You could optionally try it.
I'll have a look at it. I also tried with Digital Mars, but I couldn't work
it out (some language compatibility problem, maybe).
> I never focused the optimisations under MinGW but I imagine you could find
> some help within povray.unix group since they are more experienced in
> optimisations of GCC.
OK,
thanks to everyone for your comments.
acasta
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Christoph Hormann <chr### [at] gmxde> wrote:
> > Can you tell me whether this is just what you expect from these compilers or
> > I should adjust optimisation options differently?
> > Can you give me suggestions about what optimisations I should use?
>
> It is well known that MinGW produces very slow executables.
I can't confirm that.
I compiled Povray (more precise: MegaPov 1.0) with MingGW some time ago (I used
an older version though, for some reason I couldn't get it to compile with
recent versions), IIRC the speed difference was less than 10%.
Lutz-Peter
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
<"acasta" <acastagnini AT yahoo DOT com>> wrote:
> Can you tell me whether this is just what you expect from these compilers or
> I should adjust optimisation options differently?
> Can you give me suggestions about what optimisations I should use?
There shouldn't be such a big difference.
Some results from a couple of winpov binaries I have available:
Chess2.pov, 320x240, no AA
My version (MingW): 39s
My version (BCC5): 55s
MegaPov 1.0 (MSVC): 40s
Povray 3.5 (ICL): 32s
I used these flags with mingw:
-DPRECISION_TIMER_AVAILABLE=0 -ffast-math -fomit-frame-pointer -malign-double
-march=pentium2 -mcpu=athlon-xp -O3 -Wno-multichar -mno-cygwin
-Wunused-variable -static -mwindows -mno-cygwin
I don't remember wich version of MingW/Gcc version I used, though.
Lutz-Peter
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Thu, 29 Jul 2004 00:44:33 +0200, Lutz-Peter Hooge <lpv### [at] gmxde> wrote:
> Chess2.pov, 320x240, no AA
> My version (MingW): 39s
> My version (BCC5): 55s
> MegaPov 1.0 (MSVC): 40s
> Povray 3.5 (ICL): 32s
Is 'My version' related to 3.5 sources? You probably missed frontend/backend
architecture introduced by 3.6 which is a big difference to internal world of
GUI work.
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
ABX <abx### [at] abxartpl> wrote:
> Is 'My version' related to 3.5 sources?
It's basically megapov 1.0 with a few small modifications
(bicubic interpolation, better png support).
> You probably missed frontend/backend architecture introduced by 3.6
> which is a big difference to internal world of GUI work.
But the problem was rendering performance, not GUI.
Lutz-Peter
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
I've compiled the unix/linux source distribution with the Cygwin environment
and results have been quite different. A summary:
Official povWin binary: about 30''
MinGW binary (GUI): 90''
MinGW binary (console, tried different optimisation options): 80''
Cygwin binary: 47''
I'll try to replicate exactly the same options in MinGW ans see what happens
(but I think most options were the same).
By the way, what compiler is used for the official linux binary?
acasta
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Thu, 29 Jul 2004 11:18:48 EDT, "acasta" <acastagnini AT yahoo DOT com>
wrote:
> MinGW binary (console, tried different optimisation options): 80''
> Cygwin binary: 47''
Have you tried to apply the same compiler options from Cygwin build into flags
in makefile dedicated to MinGW build ? That would be worth to compare.
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |