[gs-devel] Understanding clamp_pattern_bbox()

[Leonardo]

See also log messages for revisions 7892 - 7895.

Leo.

----- Original Message ----- 
From: "Kevin Shanahan" <kmshanah at disenchant.net>
To: "Leonardo" <leonardo at artifex.com>; <gs-devel at ghostscript.com>
Sent: Thursday, May 03, 2007 12:49 AM
Subject: Re: [gs-devel] Understanding clamp_pattern_bbox()


> On Wed, May 02, 2007 at 11:14:49AM +0400, Leonardo wrote:
>> >when the
>> >function exists the following should be true:
>> >
>> >    pbbox->q.x - pbbox->p.x <= width
>> > && pbbox->q.y - pbbox->p.y <= height
>>
>> No, it should not.
>> It computes an union of *all* possible intersections while tiling.
>> For example, if a tile size is page size + w,
>> and if all 4 tile corners appear at the center of the page,
>> the result covers entire tile size+w.
>
> Ah, I see now. Yes, looks like the function is behaving as intended
> after all. Thanks for the pointers. :-)
>
>> I agree that in this case the performance may be improved
>> with representing the tile as 4 bitmaps, so that
>> the summary bitmap area is same as page size rather than
>> size+w. However it needs a non-trivial architectural change,
>> because now it's a single bitmap.
>>
>> I recommend you to look at the bug 688396 Comment #10
>> and its referents before taking further actions.
>
> Right, makes sense. I'll try a build from current svn and post the
> results to my bug report #688991. It certainly does sound like mine is
> a duplicate.
>
> Cheers,
> Kevin.
>
>> ----- Original Message ----- 
>> From: "Kevin Shanahan" <kmshanah at disenchant.net>
>> To: "Leonardo" <leonardo at artifex.com>; <gs-devel at ghostscript.com>
>> Sent: Wednesday, May 02, 2007 4:14 AM
>> Subject: Re: [gs-devel] Understanding clamp_pattern_bbox()
>>
>>
>> >On Tue, May 01, 2007 at 07:36:50PM +0400, Leonardo wrote:
>> >>>    /*
>> >>>     * Check if the pattern bounding box intersects the page.
>> >>>     */
>> >>>    xlower = (xdev + pbbox->p.x > 0) ? pbbox->p.x : -xdev;
>> >>>    xupper = (xdev + pbbox->q.x < width) ? pbbox->q.x : -xdev + width;
>> >>>    ylower = (ydev + pbbox->p.y > 0) ? pbbox->p.y : -ydev;
>> >>>    yupper = (ydev + pbbox->q.y < height) ? pbbox->q.y : -ydev + 
>> >>> height;
>> >>>
>> >>>I wonder if someone could spell out for me the logic behind these
>> >>>calculations?
>> >>
>> >>It looks as a superposition of 2 operations :
>> >>
>> >>1. Intersect xdev + pbbox and [0:width][0:height].
>> >>2. Shift coords to [-xdev, -ydev].
>> >
>> >Ah, thanks - I can see it now. However, I still think there might be a
>> >bug lurking here somewhere. Am I right to understand that when the
>> >function exists the following should be true:
>> >
>> >    pbbox->q.x - pbbox->p.x <= width
>> > && pbbox->q.y - pbbox->p.y <= height
>> >
>> >In this instance I have:
>> >
>> > (gdb) print width
>> > $2 = 2479
>> > (gdb) print height
>> > $3 = 3508
>> >
>> >And, at function exit:
>> >
>> > (gdb) print *pbbox
>> > $5 = {p = {x = 155.63597106933594, y = -6677.2569427490234}, q = {x =
>> >117745.96476745605, y = 380.97372436523438}}
>> >
>> >Regards,
>> >Kevin.