OpenCL tests

boredland

Hi there,

I just got my Firefly and want to start over by testing the Opencl-Power it has. Nevertheless I'm not able to compile the sample applications from the Mali_OpenCL_SDK_v1.1.0. I think that is caused by not having mali drivers, but I'm not so sure about that. I'd be very grateful, if someone could lead me into a direction with this.

My next goal would be to try to compile tesseract (https://code.google.com/p/tesseract-ocr/) with OpenCl.

Yours

Bored (of course I use the newest Ubuntu-Image available).

isle

Now， Our ubuntu has only mali kernel driver.and the userspace driver was not ready  .

But Good new is .it will be  arrived very soon.

boredland

Aren't that the drivers?
http://malideveloper.arm.com/develop-fo ... e-drivers/

It seems that the only problem is, that the manufacturer (the firefly team) need to come over with the platform specific binaries, since Mali doesn't offer the source to end-users, isn't it?

Yours

Jonas

isle

Yeh,that's the kernel driver.

what we need is the Lib(.so) ,also named User-Space Binary Drivers.

dewet

"To help with your scheduling, the fbdev version should be out this week"

It doesn't look like ARM is keeping to schedule, it's already the end of the week.

They also promised previously to release it at the end of Jan.

Can wait longer.

jas-rk

In theory Rockchip could provide the binaries (even earlier) because they have access to the Mali DDK but sadly aren't willing to do so. So I appreciate the fact that ARM are willing to assist even though the timescales aren't guaranteed.

jas-rk

fb drivers are available see here .

dewet

fb drivers are available see here .

Now someone or you just have to tell us how to load it into Firefly Ubuntu.

Also the Mali kernel drivers, how to integrate the latest ones, compatible with the userspace drivers into the kernel.

boredland

Did you come further?

jas-rk

I did some initial testing with fbdev drivers with the 3.14 kernel and they seems to work for GLES 2/3.0 samples. However the main issue seems to be lack of performance so the drivers may be not be that useful in their current state . The initial response from ARM is that the drivers are built for h/w verification not performance. The issue is raised with ARM and I wait for a response to clarify the situation. In the meantime if I get time will try to build against the 3.10 kernel.