[Mesa-dev] Status of Clover (AMDGPU)

[Marek Olšák]

On Wed, May 31, 2017 at 5:55 PM, Luke A. Guest <laguest at archeia.com> wrote:
>
>> Nobody has worked on RadeonSI OpenCL for quite some time. Even the
>> main developer of R600 and RadeonSI OpenCL (Tom Stellard) was mostly
>> working on ROCm when he was at AMD.
>>
>>> So far I've found that gl sharing is missing as are some entry points in
>>> dispatch.cpp.
>> ROCm OpenCL has interop support with RadeonSI OpenGL. We committed it
>> into Mesa when ROCm OpenCL was still closed source. I think it can
>> also be used to pass VDPAU surfaces through OpenGL into OpenCL that
>> way. It's pretty flexible.
> I'm not sure my machine can handle ROCm due to the PCIe intrisics.
>>
>>> 1) Would it be worth continuting and if so would we be allowed to shared
>>> clc files from ROCm to Mesa?
>>>
>>> 2) If not, is it worth considering porting Beignet instead, that at
>>> least gives OpenCL 2.0.
>> Clover and its driver backends are unmaintained and there is no
>> production quality Mesa OpenCL driver. If you are interested in
>> working on it, feel free, but if you upgrade to a ROCm-capable
>> machine, you might lose interest.
>>
> I'm not going to upgrading for another year, I want to wait it out.
>
> Who are the best people to ask for help?

You can ask us hw-specific questions or use the radeonsi compute
shader support and the AMDGPU LLVM backend as an example of working
compute support. RadeonSI uses the same LLVM interface for OpenCL as
ROCm. Generally, the code is the documentation. There are also 3D
register and shader ISA PDFs that you'll need.

Marek