On Mon, Mar 5, 2018 at 10:15 PM, Eric Anholt eric@anholt.net wrote:
Stefan Schake stschake@gmail.com writes:
We allow alpha formats on the primary plane but a partially transparent framebuffer will cause a corrupted display. With this change black pixels are output instead, in line with the behavior for other DRM drivers.
Signed-off-by: Stefan Schake stschake@gmail.com
Test program is available at https://github.com/stschake/vc4-alpha-test
How about this as a suggestion for a patch series:
vc4_plane_mode_set() sets ALPHA_PREMULT (POS2 bit 29) if alpha is enabled.
vc4_plane_mode_set() sets a new vc4_plane->needs_bg_fill boolean to (format->has_alpha || !covers_screen) where covers_screenis the can_position logic from drm_atomic_helper.c
vc4_crtc_atomic_flush() updates DISPBKGND to enable background fill (before vc4_crtc_update_dlist()) if the first plane has needs_bg_fill set.
vc4_plane_mode_set() strips off the alpha blend bits if !vc4_plane->needs_bg_fill.
This lets us keep avoiding the background fill cost in the normal case, and fixes the case where the "primary" plane doesn't cover the screen. It doesn't get the background fill turned back off if you transition away from primary not covering the screen, but that seems unlikely and harder to handle (since you would need to wait for the flip to be done before disabling).
I've sent out the series:
https://patchwork.freedesktop.org/series/39411/
I don't think we need the final change since
!needs_bg_fill <=> !(has_alpha||!covers_screen) <=> covers_screen && !has_alpha
so we shouldn't be setting the alpha blend bits in the first place.
Thanks, Stefan