Cedrus Video Decode Driver Moving Along With Allwinner H5/A64 Support

[phoronix]

Phoronix: Cedrus Video Decode Driver Moving Along With Allwinner H5/A64 Support

With the Linux 4.20 kernel the Cedrus VPU decoder driver was mainlined that was developed this year over at Bootlin for providing open-source accelerated video support for Allwinner SoCs. That driver continues to be ramped up to increase its usefulness...

http://www.phoronix.com/scan.php?page=news_item&px=Cedrus-Allwinner-H5-A64

[Ardje]

Nice! using v4l2-m2m (originally from Samsung to augment the v4l2 system for codecs), they added a slew other things to v4l2. I hope that will force all other SoC drivers to a unified v4l2 interface for bitstream codecs. And yes, that includes changes to the v4l2 API to incorporate all the features.
Compressed framebuffer output for instance would be nice.
Then we only need to have an interface for the XV like hardware most SoC's have (multi framebuffer multi colorspace alpha blending mixers, to mix camera input, video and graphics on-the-fly).

[tuxd3v]

Nice! using v4l2-m2m (originally from Samsung to augment the v4l2 system for codecs), they added a slew other things to v4l2. I hope that will force all other SoC drivers to a unified v4l2 interface for bitstream codecs. And yes, that includes changes to the v4l2 API to incorporate all the features.
Compressed framebuffer output for instance would be nice.
Then we only need to have an interface for the XV like hardware most SoC's have (multi framebuffer multi colorspace alpha blending mixers, to mix camera input, video and graphics on-the-fly).

This is good news,
I hope that they sort out support also for H6 which is a very nice chip...its a pity that Pcie have problems on linux.
The H6 is in there need of attention, this processor seems really good

[anarsoul]

its a pity that Pcie have problems on linux.

It's not really about Linux. AW HW engineers made a stupid decision to make MMIO accessible through 64kb window, everything larger have to switch banks. It's hardware issue, not Linux issue. Basically you can't mmap large PCIe MMIO on H6.

[jernej]

Then we only need to have an interface for the XV like hardware most SoC's have (multi framebuffer multi colorspace alpha blending mixers, to mix camera input, video and graphics on-the-fly).

All of that can be already implemented through DRM and it is for many SoCs.

[tuxd3v]

It's not really about Linux. AW HW engineers made a stupid decision to make MMIO accessible through 64kb window, everything larger have to switch banks. It's hardware issue, not Linux issue. Basically you can't mmap large PCIe MMIO on H6.

I haven t looked at it in detail,
I saw some emails around, and read them fast some time ago..
Indeed the memory mapping seems not consecutive, or linear..

By some reason that I don't know they haven't mapped all address space continuously, ind worst, it seems that only a small part of it appears mapped..
The kernel drivers only access memory maps that are mapped already...

For what I know, there are 2 Ways for pcie,
1) were all memory appears linearly mapped
2) were addressing and sending/receiving data, are separated..

I don't know if H6 implement the second method..the information around is also very small, its like guessing...

If we got more info about, maybe it would be nice and then maybe..., appears some candidate to add support for it, because that pcie even thought its only 1 lane 2.0 is still nice, for example for 1 sata disk...
That chip could became the A20 Successor, very nice indeed, 4 core 1.8Ghz and lower fab process, that CPU consumes little power, I already tested it..