Re: Mainline 5.4 kernel with VPU acceleration on iMX6


Andreas Müller
 

On Mon, Apr 6, 2020 at 9:51 PM Fabio Estevam <festevam@...> wrote:

Hi Mark,

On Sun, Apr 5, 2020 at 7:17 PM Mark Farver <mfarver@...> wrote:

On Sun, Apr 5, 2020 at 1:19 AM Carlos Rafael Giani <crg7475@...> wrote:
For using etnaviv, just use glimagesink. For using the VPU, use the v4l2*dec elements.
Definitely doing CPU decoding. I don't see any kernel messages that
indicate the VPU has been detected even though I have
Try grepping for "coda", which is the VPU driver name in mainline:

# dmesg | grep coda
[ 5.483623] coda 2040000.vpu: Direct firmware load for
vpu_fw_imx6q.bin failed with error -2
[ 5.492236] coda 2040000.vpu: Falling back to sysfs fallback for:
vpu_fw_imx6q.bin
[ 69.619815] coda 2040000.vpu: Using fallback firmware vpu/vpu_fw_imx6q.bin
[ 69.632736] coda 2040000.vpu: Firmware code revision: 46076
[ 69.638440] coda 2040000.vpu: Initialized CODA960.
[ 69.643255] coda 2040000.vpu: Firmware version: 3.1.1
[ 69.649875] coda 2040000.vpu: encoder registered as video9
[ 69.656045] coda 2040000.vpu: encoder registered as video10
[ 69.662214] coda 2040000.vpu: decoder registered as video11

CONFIG_VIDEO_IMX_PXP=y in the kernel config. Is that the correct
CONFIG_VIDEO_IMX_PXP=y applies to other i.MX SoCs.

option? Are there others? The device tree entry for the VPU seems to
be included in the kernel configuration by default.

I grabbed OSSystems meta-gstreamer layer to get 1.16.2, but even with
that included I do not appear to get any v4l2 decode plugins:

# gst-inspect-1.0 --version
gst-inspect-1.0 version 1.16.2
GStreamer 1.16.2
Unknown package origi

# gst-inspect-1.0 |grep v4l
imxv4l2video: imxv4l2videosrc: V4L2 CSI Video Source
imxv4l2video: imxv4l2videosink: V4L2 CSI Video Sink
This is wrong. You should not install these imx elements when using a
mainline kernel. These elements are to be used with NXP kernel only.
Have no use case currently so just out of curiosity: Does this also
apply to community kernels - do they support vpu-acceleration?

Andreas

Join {meta-freescale@lists.yoctoproject.org to automatically receive all group messages.