ср, 8 июн. 2022 г. в 11:17, Maxime Ripard maxime@cerno.tech:
On Mon, Jun 06, 2022 at 01:16:06PM +0300, Roman Stratiienko wrote:
вс, 5 июн. 2022 г. в 23:23, Jernej Škrabec jernej.skrabec@gmail.com:
Dne nedelja, 05. junij 2022 ob 11:40:18 CEST je Roman Stratiienko napisal(a):
Otherwise alpha value is discarded, resulting incorrect pixel apperance on the display.
This also fixes missing transparency for the most bottom layer.
Can you explain that a bit more?
Well... I would recommend reading Bartosz Ciechanowski's blog https://ciechanow.ski/alpha-compositing/ or the Porter-Duff's 1984 whitepaper itself.
HINT: That magic numbers from sun8i_mixer.h ( 0x03010301 ) corresponds to SOURCE OVER mode.
As you can see from the blending equation it outputs both pixel value and alpha value (non-premultiplied data mode).
Also single-layer non-premultiplied buffers may have for example (R255,G255,B255,A2) pixel value, which should be sent as {R2, G2, B2} through the physical display interface.
When OUTCTL.PREMULTI disabled pixel, the RGB values passes as is, and even 100% transparent data {R255, G255, B255, A0} will appear as 100% opaque white.
Without going into the full explanation about what alpha is, your commit log must describe what the bug is exactly, and most importantly how do you trigger it.
I do not understand what you want me to add. I checked alpha appearance manually by preparing framebuffers with data and presenting it on the display in various combinations.
I attached the videos and tests as a proof. If you don't believe me you can always check.
If you find something missing in the commit message or don't like to see external links feel free to amend it. From my point of view the patch is complete.
Maxime
Roman