[XeTeX] performance on Windows machines

[Hans Hagen]

Jonathan Kew wrote:
> On 12 Sep 2006, at 7:33 am, Akira Kakuto wrote:
>
>   
>>> To test performance with legacy TeX stuff (not OpenType, etc), I
>>> sometimes run texbook.tex (with a suitable prefix file to make it
>>> actually typeset). The Mac Pro takes about 1.1-1.2 seconds (real
>>> time) to generate the 494-page PDF using xetex -output-
>>> driver="xdvipdfmx -q -E". Wow! Actually, that's faster than pdftex
>>> can do it. :)
>>>       
>> I'm amazed at the speed of xetex. XeTeX for W32 binary is
>> not optimized, while pdftex for W32 is optimized:
>>
>> xetex --proctimes texbook
>> Time statistics: user mode=2328 ms, kernel mode=125 ms, total=2453 ms.
>>
>> pdftex --proctimes texbook
>> Time statistics: user mode=2828 ms, kernel mode=187 ms, total=3015 ms.
>>     
>
> No, this isn't the whole story: --proctimes is (presumably) only  
> reporting the time for the actual xetex process.
>
> For a fair comparison, you also need to take xdvipdfmx into account,  
> and that will make the total times for xetex somewhat greater than  
> pdftex. But the actual "wall time" to complete the job can still be  
> faster if you have a multiprocessor machine, because the two  
> processes run in parallel on separate CPUs.
>   
the texbook is not really representing the average document

- there are no complex, time consuming macros involved
- the font system is relatively simple
- the document is coded in raw tex
- no color, graphics etc are used

since pdftex and xetex share the same tex machinery, the difference most 
indeed be related to the backend; there it are influences by

- special versus literals (not here)
- resolving font info (map files, encodings etc)
- including fonts
- including graphics (inclusion as well as determining dimensions)

but that does not play a real role in the tex book; different areas may 
compensate each other

the two processor approach is interesting since it's one of the few 
areas where tex can still be sped up (now that processors have reached 
their max); it would be interesting to see what happens if xetex can 
deliver more data then dvipdfmx can process (large graphics and such); i 
wonder if there are advanced timing programs to determine that kind of 
things

Hans

-- 

-----------------------------------------------------------------
                                          Hans Hagen | PRAGMA ADE
              Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
     tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com
                                             | www.pragma-pod.nl
-----------------------------------------------------------------