Shared Virtual Memory Lands In Mesa Gallium3D's "Clover" OpenCL Implementation

[phoronix]

Phoronix: Shared Virtual Memory Lands In Mesa Gallium3D's "Clover" OpenCL Implementation

After being in code review the past half-year, support for Shared Virtual Memory (SVM) within Gallium3D's "Clover" OpenCL state tracker was just merged for Mesa 20.1...

http://www.phoronix.com/scan.php?pag...enCL-SVM-Lands

[ms178]

I wonder if this work is related to Nvidia's "big Linux announcement" as I am not sure why Red Hat would put so much ressources into Clover, HMM and other parts of the stack if Nvidia wouldn't provide the blobs which you need to make any practical use of this? By the way, wasn't this big announcement supposed to happen in March? Are there any follow ups on that front, Michael?

[TemplarGR]

Even though AMD does not contribute to Clover anymore, doesn't this work still benefit AMD's MESA driver as well? It is still somewhat usable. Though to be honest i don't understand why they are working to implement 2.0 features when even 1.1 support is lacking in Clover.

[Michael]

I wonder if this work is related to Nvidia's "big Linux announcement" as I am not sure why Red Hat would put so much ressources into Clover, HMM and other parts of the stack if Nvidia wouldn't provide the blobs which you need to make any practical use of this? By the way, wasn't this big announcement supposed to happen in March? Are there any follow ups on that front, Michael?

Nothing that I have to announce at the moment, but now that they weren't rushed for GTC, it may be a few months before an announcement...

[Qaridarium]

Nothing that I have to announce at the moment, but now that they weren't rushed for GTC, it may be a few months before an announcement...

we (and I) speculate for stuff like this for like 10 years.
Also the firmware situation with AMD hardware just opened a opportunity for NVIDIA means AMD does tiny&small firmware
Nvidia could do big-firmware. i wrote about this month or even years ago.
there is a possibility that they just move everything what is closed source into the firmware.
by this they could claim they have same level of opensource drivers like amd because closed source firmware is closed source firmware. and in the end they could hide all their closed source secrets inside the firmware.

[lucasbekker]

I wonder if this work is related to Nvidia's "big Linux announcement" as I am not sure why Red Hat would put so much ressources into Clover, HMM and other parts of the stack if Nvidia wouldn't provide the blobs which you need to make any practical use of this? By the way, wasn't this big announcement supposed to happen in March? Are there any follow ups on that front, Michael?

I don't think that nouveau is the intended user of Clover, just its showcase.

It would not be the first time that a Mesa project starts out targeting certain hardware, only to find that it forms the basis of a driver for completely different hardware, like ANV and RADV.

Clover might end up providing a good basis for an OpenCL driver for a lot of ARM SoCs, which would be very beneficial indeed!

[Nille_kungen]

Nothing that I have to announce at the moment, but now that they weren't rushed for GTC, it may be a few months before an announcement...

GTC was digital this year.
https://www.nvidia.com/en-us/gtc/
https://www.nvidia.com/en-us/gtc/session-catalog/

[Michael]

GTC was digital this year.
https://www.nvidia.com/en-us/gtc/
https://www.nvidia.com/en-us/gtc/session-catalog/

Right, but not all of the original presentations planned GTC talks were done for GTC Digital, the open-source talk included.

[abott]

If this is for anything, it'd be Vulkan related to MGPU, outside of just OpenCL compute use.

[darkbasic]

What about image support? Is it still not working?