Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1630950 > unrolled thread

Re: [PATCH 00/16] Intel FPGA Device Drivers

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH 00/16] Intel FPGA Device Drivers One Thousand Gnomes <gnomes@lxorguk.ukuu.org.uk> - 2017-04-25 22:10 +0200
    Re: [PATCH 00/16] Intel FPGA Device Drivers Jerome Glisse <jglisse@redhat.com> - 2017-05-01 18:50 +0200

#1630950 — Re: [PATCH 00/16] Intel FPGA Device Drivers

> That is where we disagree. I do not see bitstream as firmware. For instance
> now you can run OpenCL on some FPGA, so this is exactly like GPU we should
> request open source stack from OpenCL down to bitstream.

It's an accelerator with a bunch of firmwares where you load the right
one. We've got lots of those in Linux already. Your GPU probably needs
firmware as well in just the same way.

> For me this is not enough (tool to load bitstream).

Unfortunately that isn't likely to change for any major FPGA device in
the near future. If you could load arbitrary bit patterns into an FPGA
then in most cases that also means you could physically destroy the
hardware.

Alan

[toc] | [next] | [standalone]

#1633744

On Tue, Apr 25, 2017 at 09:02:25PM +0100, One Thousand Gnomes wrote:
> > That is where we disagree. I do not see bitstream as firmware. For instance
> > now you can run OpenCL on some FPGA, so this is exactly like GPU we should
> > request open source stack from OpenCL down to bitstream.
> 
> It's an accelerator with a bunch of firmwares where you load the right
> one. We've got lots of those in Linux already. Your GPU probably needs
> firmware as well in just the same way.

I stress again OpenCL on FPGA ie at runtime you load a program source
that is compiled at runtime and then lower to a bitstream which get
loaded on the FGPA at runtime.  You can go look at how OpenCL works it
is not like fix firmware. OpenCL was done for GPU and everything needed
to support OpenCL is open source when it comes to GPU. On contrary with
FPGA the bitstream part is close source.

Sure GPU have firmware, those firmware run on realtime micro-controller
to dispatch job, handle GPU power management and other janitorial stuff.
For some GPU we even have open source firmware.

So FPGA with thing like OpenCL are more like GPU than like a fix hardware
that is program once and use like a device for which we need a firmware.
That is where we disagree on the characterization of things.


> > For me this is not enough (tool to load bitstream).
> 
> Unfortunately that isn't likely to change for any major FPGA device in
> the near future. If you could load arbitrary bit patterns into an FPGA
> then in most cases that also means you could physically destroy the
> hardware.

I am just questioning the different standard in the kernel.

Jérôme 

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web