[Ksummit-discuss] Compute-offload devices and memory device

[Ksummit-discuss] Compute-offload devices and memory device

[Jerome Glisse]

[-- Attachment #1: Type: text/plain, Size: 879 bytes --]

Hi,

So to make sure all the points i wish to discuss during kernel summit
do not elude my mind i wrote them down in bunch of slides. Feel free to
use that during discussion.

I added some background into those slides so people agree on technology
we are talking about. Not everyone is necessarily familiar with GPU or
thing like PCIE ATS/PASID.

So topic i would like to discuss
  - Device threads/process lifetime
  - Scheduling on offload devices (GPU/FPGA/DSP)
  - Address space mirroring
  - Exposing device memory (this overlap with ZONE_DEVICE and persistent
    memory issue about struct page thought in case of GPU this is still a
    slightly different case)

Note that all information in thus slides is not necessarily accurate
especially when it comes to future technology that are not set in stone
and mostly only exist in VHDL/Verilog at this time.

Cheers,
Jérôme

[-- Attachment #2: device-scheduling-memory.odp --]
[-- Type: application/vnd.oasis.opendocument.presentation, Size: 31563 bytes --]

[-- Attachment #3: device-scheduling-memory.pdf --]
[-- Type: application/pdf, Size: 46848 bytes --]