Investigate hardware rendering on RPi2



Firefox OS
2 years ago
2 years ago


(Reporter: gerard, Unassigned)


Firefox Tracking Flags

(Not tracked)




2 years ago
Currently, upstream (AOSP/RPi2) hardware rendering looks to be in pretty bad shape. Let's investigate this and help fixing upstream if needed.


2 years ago
Blocks: 1225139
UPDATE: I could finally make GFX stack work (mostly). This is how it works.
* We have a kernel (4.2.1) with patches for the VC4 GPU, via DRM drivers. I needed to patch it [1] to support one of the pixel format we use in Gecko with the boot animation [2].
* We are using Eric Anholt's drm_gralloc fork, because it has the patches for VC4 support, but the drm_gralloc fork version is outdated. I needed to patch this gralloc [3] version: to support newer kernel DRM drivers mostly, to add support for other pixel formats and to accommodate our build system.
* We are using freedesktop Mesa library implementation for OpenGLES on it's latest release. In this release, there's not a good Android support yet (is coming for next releases..) so I needed to patch it [4] because Gecko is relying on this support for the GFX stack. I fixed some other pixel format related stuff and build system integration as well.

Problems detected so far:
* There are glitches rendering some of the components in Gaia. These are probably happening because there's not a good support for pitch alignment in the gralloc component [5].
* We are not using hardware VSYNC because this is implemented via HWComposer on Android, but the HWComposer we have is v1.0, which is not supported by Gecko.

You need to log in before you can comment on or make changes to this bug.