Optimized Alpha-Blending in asm?

[AGS]

Dear Peter

please have a look at the attached image in that post again:

http://eab.abime.net/showpost.php?p=...&postcount=109


The left Image is the result after 1000 iterations. But as far I can see the transparency only vanishes where the image is black. The parts of the 'scratches' surrounded in gray look still somehow partially transparent. Do you get what i mean? Is this correct?

thanks
A

[PeterK]

Hi AGS,

please make a better example. It should consist of the foreground image, the background image, the alpha channel image, a composite image of all three images and maybe some more after several iterations. Use red and green for the foreground and different shades of blue for the background. The alpha channel should have some smooth fades from black to white and also some gray objects with values of 50, 100, 150 and 200.

[AGS]

PeterK, one image with all included you mean or separate images of everything?

[PeterK]

Sorry for my poor English, but I would like to have at least four seperate images to make it easier to explain what really happens and to see which formular works correct or wrong.

[AGS]

Your English is Ok, I just can't see which of the four images should contain what and what not. Please could you explain for me more detailed what image 1 should contain and what image 2 and so on.

[PeterK]

1. the foreground image, in red and green
2. the background image, in blue
3. the alpha channel image, with fades from black to white and some simple gray objects
4. a composite image of all three images (1-3) above made with an image processing program

[AGS]

Before I start, please let me check if I understand correctly. You would like to have some image using only the colors red and green, then some other image only using blue, then an alphachannel only image with a gradient and some gray figures, and then a combination of the three? That combination should combine all channels, right? But when I follow your instructions as far I can understand them, then there are not four images but only three (red and green is in one) ... for a destination that should have four channels? I do not understand what to do here.

[PeterK]

The foreground image is a 24 bit RGB, most parts in red and green
The background image is a 24 bit RGB, most parts in blue
The alpha channel is 8 bit gray scale

And the final PNG output brings all as a 32 bit ARGB together. So difficult?

Ok, I must admit to have some problems with PSP8 to create an ARGB. The attached combined image is only RGB now.

[Mrs Beanbag]

what i'd like to ask for, is comparison between different numbers of iterations. what are you actually doing here, drawing the grey scribbles over the top a thousand times? i'd like to see just one time, two times, ten times, a hundred times perhaps.

[AGS]

It was really hard to understand for me. So as we now have the images, what should I do with them or do next?

[Mrs Beanbag]

actually when i think about it the result is not entirely inexplicable, for instance look at this

[PeterK]

Originally Posted by AGS

It was really hard to understand for me. So as we now have the images, what should I do with them or do next?

Hmm, I thought that you wanted to check if your alpha blending formular in posting #109 works correctly?

Use the attached ARGB.png image for your input in A0
and the background.png image for the array in A1
and then calculate the result for 1, 10, 100 iterations etc ...

;==============================================================
; apply alpha blending from an ARGB buffer into an RGB buffer
;==============================================================

calc_pixels	; apply ARGB data onto RGB data

		; a0	in ARGB image
		; a1	in RGB array

Update:

Oh dear!
Now, I noticed that you have changed the code in your old posting #109 without telling me about these modifications. Sorry, but I didn't expect such changes in an old posting, because it destroys the chronological progress and leads to confusion if somebody corrects his historical statements after several months without adding an "update" note.

Of course, you can use both formulars in your routine (optional selectable). Nevertheless, the verification of your old code shows that half of the results have the wrong roundings.