[gs-devel] AFPL GS 7.30 seems to be slower than ESP/GNU GS 7.05. Any ideas why?

[Vadim Plessky]

I have tested two different versions of GS  - ESP/GNU GS 7.05 and AFPL GS 
7.30, and it seems to me that AFPL GS is slower than *old* GS 7.05.

I compiled both versions of GS myself, using stock Linux-Mandrake 8.2 system, 
with gcc 3.0.4 compiler.
My system is Compaq Presario 1800 notebook, with Pentium III/600Mhz and 128MB 
of RAM.
I used PPST (PostScript Processing Speed Test) developed by Jean-Serge Gagnon 
<jsg at gaaj.com>
which you can download from http://www.gaaj.qc.ca/ppst/
Direct link to archive: http://www.gaaj.qc.ca/ppst/ppst56.zip

GS was invoked as:
#GS -dNOPAUSE -r100 ppst56.ps

Results:
- - - - - - - - - - - - - - -
ESP Ghostscript (4M,100,fm.705)      3010
673.2   >1000!   4623.5   179.7   123.0   >1000!   1500

 The total time for PPST is 1.9 and the individual times for each tests are: 
0.0 seconds for PPST-I, 0.2 seconds for PPST-G, 1.3 seconds for PPST-F, 0.1 
seconds for PPST-R, 0.0 seconds for PPST-M and the time taken between all 
pages was 0.2 seconds. As well as the standard PPST times, the data recep
tion and interpretation took 0.1 seconds.

Specifications: Resolution is 100x100dpi. Floating point precision is 6 
digits. Ram size 4 Megs. Postscript is Level 2, version 3010. Firmware 
revisio
n 705. Page count 4711. Printer name is "ESP Ghostscript". Product name is
"ESP Ghostscript". The logical engine speed (according to "between" time) is 
1500 page(s) per minute.

- - - - - - - - - - - - - - - - - - -
AFPL GS 7.30

AFPL Ghostscript ALPHA RELEASE (4M,100,fm.730)      3010
456.8   323.0   1541.2   146.0   123.0   424.0   750
***
The total time for PPST is 2.8 and the individual times for each tests are: 
0.1 seconds for PPST-I, 0.6 seconds for PPST-G, 1.6 seconds for PPST-F, 0
.1 seconds for PPST-R, 0.1 seconds for PPST-M and the time taken between all 
pages was 0.4 seconds. As well as the standard PPST times, the data recep
tion and interpretation took 0.1 seconds.

Specifications: Resolution is 100x100dpi. Floating point precision is 6 
digits. Ram size 4 Megs. Postscript is Level 2, version 3010. Firmware 
revision 730. Page count 4711. Printer name is "AFPL Ghostscript ALPHA 
RELEASE". Product name is "AFPL Ghostscript ALPHA RELEASE". The logical 
engine speed (according to "between" time) is 750 page(s) per minute.

- - - - - - - - - - - - - - -
I repeated tests several times, and all the time results were about the same.
So I used numbers obtained for the first run.

Any ideas what can cause such significant difference in performance?

As a side note:
GNU GS 7.05 performance on WIndows 98 (same computer!) was just *terrible*.

GNU Ghostscript  (4M,100,fm.705)      3010
82.0   15.4   289.0  28.1   41.0   141.3   214
Total time for PPST: 15.6 sec.

So, it's about 8 (eight) times slower than on Linux.
Any ideas why? I hardly believe compiler differences can explain such a HUGE 
gap in performance.

-- 

Vadim Plessky
http://kde2.newmail.ru  (English)
33 Window Decorations and 6 Widget Styles for KDE
http://kde2.newmail.ru/kde_themes.html
KDE mini-Themes
http://kde2.newmail.ru/themes/