[sane-devel] genesys backend

[Pierre Willenbrock]

Hi

As you might guess, i'm having another patch ready for review.

Stéphane VOLTZ schrieb:
> 	Hello,
> 
> 	I have applied your slope table patch in the experimental/genesys. I tested 
> it wtih my MD6571 and everything I tried worked fine.
> 	The 'stagger' pattern is indeed :
> ...0...0
> ..0...0.
> .0...0..
> 0...0...
> 

Currently the pattern is restricted to this one:
.0.0
0.0.
because the old code seemed to do it that way, and there is no variable
storing the actual width.

> 	I've reread the 'genesys_read_ordered_data'. It has 2 phases. First it fills 
> buffer with enough raw data for the second stage, which then copies data to 
> the backend. The complexity comes from all the possible cases when copying 
> raw data in the form required by the backend and frontend combination. One 
> thing that come to my mind is that splitting that copy in a dozen of simple 
> functions (one for each case) would be enough. But, since you're going after 
> something better, I'll wait for it.

I doubt it is that much better. I produced about 6 new functions, all in
two variants for single and double byte data.

There are 5 steps now:

step 0: Reading data from scanner. Delegated into its own function, but
may be moved back.

step 1:
genesys_reorder_components_(cis|cis_bgr|bgr)_(8|16) convert byte order,
planar and bgr data.

genesys_reorder_components_endian_16 converts byte order, and is only
available on big endian systems.

step 2:
genesys_reverse_ccd_(8|16) reverses the effects of ccd layout. It takes
line offsets per component for multiple pixels, thus doing line distance
correction and unstaggering in one pass. We could convert any stagger
pattern (up to 4 pixels width currently) that way.

step 3:
genesys_shrink_lines_(8|16) shrinks/enlarges lines.

step 4: copy result to frontend buffer. This is needed because the
conversion functions work on line aligned data, and we may have lines
longer than 32768 bytes, thus needing a output buffer.

I used the preprocessor to generate the 8/16 bit versions, and i think
it is a good idea to put that into an extra file, leading to two extra
files. I could expand genesys_conv.c to contain all the mentioned
functions, but then there would be two very similar implementations to
be kept in sync.
The functions may be useful for other backends, too.

I changed the way the information about the format of the scanner output
is stored. I think it is more understandable this way.

I needed 4 buffers. To handle that efficiently i added a set of buffer
management functions.

Then i completely removed dev->exposure_time. It was still used at quite
some places. If that is not okay, i can revert that.

Finally i moved the byte order detection from run time detection to
compile time detection, using the WORDS_BIGENDIAN macro defined(or
undefined) in include/config.h.
> 
> 	I think it is high time you get CVS access to SANE. The one thing we will 
> have to mind is to discuss or warn for changes in the common code. Simple 
> fixes and chipset specific changes should be OK.
How do i get cvs access? I am already having a user on alioth.
> 
> 	I am going to tune the gl646 part for the HP2400. I believe that the changes 
> I have in mind (mostly constants tweaking) shouldn't disturb your current 
> work.
> 
> 	Regards,
> 		Stef
> 

If this patch is okay, we need to discuss calibration. My scanner has as
black and a white strip at the beginning of the scanning area, while
other scanners seem to be different. I also need additional calibration
steps to calibrate the led exposure times to get nearly equal
scanning results for all three colors before doing shading. That
improves the quality of 16bit scans.


The attached patch still is not moving the gl841 part to any working
state. As soon as there are checkins into experimental cvs i will update
the patch on my website. It still is too large for this mailing list.

Regards,
  Pierre
-------------- next part --------------
A non-text attachment was scrubbed...
Name: genesys_conversion.diff.gz
Type: application/x-gzip
Size: 16097 bytes
Desc: not available
Url : http://lists.alioth.debian.org/pipermail/sane-devel/attachments/20050904/ff511c15/genesys_conversion.diff-0001.bin