[Gs-devel] Re: [Gs-code-review] Re: Omni integrated into GS_6_5 branch

[Raph Levien]

On Tue, Jun 05, 2001 at 05:23:20PM -0400, Pete Zannucci wrote:
> Thanks for the discussion and info earlier Raph.  It will help us figure
> out what needs to happen from a distribution point of view.
> I have included two patches, one for gomni.c (gomni.patch) and one for the
> makefile updates that work on my Redhat 7.0 system.  With the changes that
> I have included there is no segment fault when exiting Ghostscript.  It
> could of course be the system level or possibly the c++ libraries.
> The reason that I am including the two patches is that the makefile. patch
> may help a lot of people that don't want to go in and make the changes by
> hand to get the Omni built.

Included the gomni.c patch, but not the makefile one. That will of course
vary widely from system to system.

What I'd suggest is that you put a page together containing directions
on how to use Ghostscript with Omni. Once the distros start shipping
6.51, these directions can all be RPM-based rather than "how to build
source". The directions will also need to include hints on how to set
the properties, etc.

> I read the diary update about the 870 performance and output.  One of our
> to-do's is to go back into the Epson code for the devices that are using
> the High Res Blitter and add white space skip back in along with four-color
> support on six-color printers.  Right now the Omni has to render the white
> space right along with the color data when using this blitter.  Also it
> would be interesting to know what resolution was tested.  You might want to
> try
> -sproperties="resolution=RESOLUTION_720_X_720 media=MEDIA_SPECIAL_360".
> This should give you good output on coated paper.
> A couple of other settings would be RESOLUTION_1440_X_720 and
> MEDIA_SPECIAL_720.  The gammas are adjusted for 360 media as coated and 720
> media as glossy.  We'll be adding in other medias along with increasing
> usability in the near future.  I gammas were adjusted on an 875DC since I
> didn't have an 870 here.

Ok. I believe 720x720 is the default, as the speed seems to be
identical with the setting and without. The MEDIA_SPECIAL_360 made a
very subtle difference.

Here are some performance results you might find enlightening. The
test system is a dual Celeron 400, parallel port. I tested an HP 990C
with hpijs 0.96 against an Epson Photo 870 with Omni 0.2.0. With both
printers, I used Ghostscript 6.51. My main test file was colorcir.ps
from the Ghostscript distribution, although I also used tiger.ps and
vancebarn.ppm (the latter is a 2677 x 2129 color image, available on
request).

Performance differences were dramatic. The 990C clocked colorcir.ps in
44.5 seconds at 600dpi. The 970, by contrast, took 6:19 at 720 x
720. When the resolution was increased to 1440 x 720, speed plunged to
26:25, a factor of 35 slower.

Overall quality on both printers was comparable. The Epson has
somewhat brighter colors, and considerably less noticeable dither
patterns. However, the Epson output suffered from a number of defects,
while the HP output was consistently very good.

First, the Epson suffered from weaving artifacts at the top of the
page. These were particularly noticeable with tiger.ps.

Second, the Omni error diffusion algorithms occasionally suffer from
"worming" and "tearing" effects. These effects are common in error
diffusion algorithms unless some effort has been made to counteract
them. The worming is particularly visible in the light gray samples in
colorcir.ps, as well as greenish yellow and orangish yellow. The
tearing effects were most severe in the "White" text sample on
colorcir.ps - the characters appear eroded, and entire stems and
serifs are missing. Tearing is also visible in the greenish yellow
samples. These artifacts are not significantly different in the 720 x
720 and 1440 x 720 dpi samples.

Third, color correction is way off on the Epson. This is most dramatic
in the light gray samples of colorcir.ps, which have a pronounced
yellowish cast, and are also "burned out", ie all very light grays
appear as white, so no contrast is visible. Further, the 1440 x 720
print had dramatically more saturated colors than the 720 x 720 one
(the MEDIA_SPECIAL_360 setting was used for both, which of course may
not have been appropriate).

I attempted to try the 360 x 360 resolution, but that caused the
printer to take multiple passes over the same area, soaking the paper
and forcing me to abort the job in fear of having the printheads
damaged.

I still get coredumps on exit, likely due to my use of a more-or-less
stock RedHat 6.2 system. These don't interfere with pages being
printed, however, so it's not really critical.

I'm cc'ing gs-devel rather than gs-code-review on this, as it seems
the more relevant list.

Thanks for all your work on this!

Raph