[PATCH v3 0/5] GPU workload hints for better performance

Sharma, Shashank shashank.sharma at amd.com
Tue Sep 27 17:25:45 UTC 2022



On 9/27/2022 7:13 PM, Michel Dänzer wrote:
> On 2022-09-27 18:59, Sharma, Shashank wrote:
>> Hey Michel,
>> Thanks for the review coments.
>>
>> On 9/27/2022 6:24 PM, Michel Dänzer wrote:
>>> On 2022-09-26 23:40, Shashank Sharma wrote:
>>>> AMDGPU SOCs supports dynamic workload based power profiles, which can
>>>> provide fine-tuned performance for a particular type of workload.
>>>> This patch series adds an interface to set/reset these power profiles
>>>> based on the workload type hints. A user can set a hint of workload
>>>> type being submistted to GPU, and the driver can dynamically switch
>>>> the power profiles which is best suited to this kind of workload.
>>>>
>>>> Currently supported workload profiles are:
>>>> "None", "3D", "Video", "VR", "Compute"
>>>>
>>>> V2: This version addresses the review comment from Christian about
>>>> chaning the design to set workload mode in a more dynamic method
>>>> than during the context creation.
>>>>
>>>> V3: Addressed review comment from Christian, Removed the get_workload()
>>>>       calls from UAPI, keeping only the set_workload() call.
>>>>
>>>> Shashank Sharma (5):
>>>>     drm/amdgpu: add UAPI for workload hints to ctx ioctl
>>>>     drm/amdgpu: add new functions to set GPU power profile
>>>>     drm/amdgpu: set GPU workload via ctx IOCTL
>>>>     drm/amdgpu: switch GPU workload profile
>>>>     drm/amdgpu: switch workload context to/from compute
>>>
>>> Where are the corresponding Mesa changes?
>>>
>>>
>> This series here was to get the feedback on the kernel side design first. As you can see from the patch history, we have already changed the design once and this is V2. So I thought it would be a good idea to get the feedback on kernel UAPI, before starting sending patches to mesa.
> 
> In general, it's not possible to review UAPI without the corresponding user-space code. I don't think this is an exception.
> 
> 

Sure, good that we already have got the kernel inputs we wanted, now the 
next version will be with corresponding MESA changes.

- Shashank


More information about the amd-gfx mailing list