Support for Raspberry Pi 2 and 3

[Terminus-IMRC]

Hi. This pull request is for supporting Raspberry Pi 2 and 3. Here are the changes:

• The Mailbox interface is now on /dev/vcio.
• GPU-side memory address must be masked before mapping on /dev/mem on Raspberry Pi 2 and 3. This BUS_TO_PHYS macro was first seen in raspberrypi/userland@3b81b91 .

Please merge these changes to your repo. Thanks.

[Terminus-IMRC]

Along with this, please update README.md, RaspberryPiLibrary/libjpcnn.so and so on.

[guysoft]

Hey, why was this not merged still? just got stuck?

[Terminus-IMRC]

He now works for Google and seems very busy. In addition, there are much faster sgemm implementations which run on Raspberry Pi QPU such as py-videocore, so little motivation to manage this repo IMHO :-)

[guysoft]

QPU = GPU?
I was looking for a dnn for audio that would run on a raspberrypi GPU. If you have a direction to point me to (real-time speed)

[Terminus-IMRC]

Yes, QPU (quad processing unit) is a GPU implemented on Raspberry Pi.

We, Idein Inc., offer a fast deep learning (inference) library/software. If you're interested, feel free to contact us: https://idein.jp/

Our YouTube channel: https://www.youtube.com/channel/UCp0a6npOi89ksgks3ej7FPw

[guysoft]

@Terminus-IMRC thans for the offer, but I am currently looking for a solution that is open source. I am trying to build something people can replicate instantly around the world, a library that requires contact would slow that.

[Terminus-IMRC]

I see. In addition to py-videocore, QMKL is a good open-source example if you want to make something faster by using QPU.