Pedro Alves a écrit :
> Danny Backx escreveu:
>> On Wed, 2007-09-19 at 17:51 +0200, pgeorges wrote:
>>>> Crosswords, compiled with cegcc, runs noticably slower than compiled
>>>> with the m$ compiler. Eric House also reported this. I've not
>>>> investigated this yet. We may have inefficiencies other than the code
>>>> generated by the compiler. Something with the coredll perhaps ?
>>>>
>>>>     Danny
>>>>   
>>> Is it possible to know in which cases the slowdowns are encountered 
>>> ? CPU intensive code, I/O , system calls ?
>>> What is the amount of slow down ? 10 % , 50 % ?
>>>
>>> Pascal
>>
>> I don't have an eVC4 installed (only my children have Windows PC's) so I
>> have no simple way to do comparisons.
>>
>> Xwords shows a noticable difference if you get it to refresh its screen.
>> (It's a scrabble game application.) When compiled with cegcc, you can
>> see how it draws the game board - takes one or 1.5 seconds. With the
>> executable as you can download it from http://xwords.sf.net , the redraw
>> is instantaneous.
>>
>> So that example could lead you to the conclusion that it's either
>> graphics or system (DLL?) access.
>>
>
> I doubt it.  There shouldn't be any difference in calling system 
> functions.
> It's just a bl anyway.
>
> Your example is about floating point arithmetic.  Is Xwords doing it a 
> lot?
>
> Pgeorges, is you chess engine doing a lot of fp math?
>
My application uses a lot integer and floating point calculation, but 
more integer than floating point, I guess. But if I remember well the 
first SPARC processors, they used 120 CPU cycles for a mere division, 
because of the lack of FPU, so even if the FP instructions are used 1% 
of the time (compared to integer), they can lead to a use of 50% of the CPU.
> We currently generate code that assumes that the device *does not* 
> have an fpu.
  The processor of my PPC is a PXA 270 which does not have a FPU 
(correct me if I am wrong).

Pascal

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Cegcc-devel mailing list
Cegcc-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/cegcc-devel

Reply via email to