Open Mis012 opened 5 years ago
I was struggling with this issue too and it seems like the problem is in upstream libva, which does not have this tiled_to_planar function. I guess you should use libva from here: https://github.com/bootlin/cedrus. Personally I use another approach and took patched ffmpeg with support for cedrus(mentioned in another issue): https://github.com/Kwiboo/ffmpeg. I tested it with ffplay and a couple of videos and it works nicely.
The problem also exists on my H6 board.
Remove these lines in src/image.c
and the runtime error disappears, but I don't know if the library will still work...
tiled_to_planar(surface_object->destination_data[i],
buffer_object->data + image->offsets[i],
image->pitches[i], image->width,
i == 0 ? image->height :
image->height / 2);
I also get the same error:
[vaapi] libva: dlopen of /usr/lib/dri/v4l2_request_drv_video.so failed: /usr/lib/dri/v4l2_request_drv_video.so: undefined symbol: tiled_to_planar
@dossalab Would you be able tell us how you set up ffmpeg + your player of choice to work with Kwiboo's ffmpeg? I'm also using Kwiboo's ffmpeg but every video output mode + decoder combination in VLC results in an error and falls back to software decoding.
So I bumped into this as well. I'm now at kernel 5.4 with a gentoo rootfs. Kodi isn't decoding smoothly. Unfortunately, the document https://linux-sunxi.org/Sunxi-Cedrus is kind of outdated.
Anyone know how to reproduce successful hardware decoding for AllwinnerH6 recenly?
Aarch64 allows running 32-bit (ARMv7) code
You can have an aarch64 kernel with an armv7hl userspace in some cases where the aarch32 execution state is implemented in the CPU at EL0, but it's not necessarily the case for all aarch64 CPU. At least you cannot have aarch64/armv7 interop in the userspace level (like i686 with x86_64).
I'm trying to test this project using amlogic soc (la frite) on aarch64 (kernel/userspace), but still hit this. I wonder by the tiled_to_planar could be avoided on aarch64 and not on armv7hl. It seems to me that either a pure c implementation have to be done or a related aarch64 ASM.
Until then anyone with a PR to either fix or workaround the issue ?
See also my attempt https://github.com/bootlin/libva-v4l2-request/pull/32
issue
It seems I get a problem similar to #18. Although compilation runs fine, the same error is now thrown on runtime. Since the original issue was closed, I assume this is something new?
If aarch64 is not supported, it seems like a shame given this looks like the only immediate solution for vlc hw acceleration on most aarch64 platforms.
steps to reproduce:
on aarch64 device:
error log: