ROCm / Tensile

Stretching GPU performance for GEMMs and tensor contractions.
MIT License
220 stars 149 forks source link

Add gfx1151 support #2013

Closed amd-garydeng closed 2 months ago

amd-garydeng commented 2 months ago

resolves #___

Summary:

enable gfx1151 support

*What is being changed and why?

Outcomes:

What is the result of this change? What components of the project does it affect?

Notable changes:

Are there any changes that are of particular importance?

Testing and Environment:

What environment are you targeting (OS, ROCm version, Python versions, etc.)?

*What testing did you do to ensure this change will integrate successfully?

nakajee commented 2 months ago

I think you need to add skip-gfx1151 to all yaml files under Tensile/Tests with skip-gfx1100.

nakajee commented 2 months ago

Please run precheckin + extended tests on gfx1151.

nakajee commented 2 months ago

I grepped "gfx1100" in Tensile and found some hits under Tensile\Source\lib. Don't we need to add gfx1151 to these files?

nakajee commented 2 months ago

I grepped (11, In Common.py isasWithDisabledHWMonitor = ((9,4,1), (9,4,2), (11,0,0), (11,0,1), (11,0,2), (12,0,0), (12,0,1)) Do we need to disable HW monitor for 1151?

eidenyoshida commented 2 months ago

@yoichiyoshida can you answer @nakajee 's questions?

yoichiyoshida commented 2 months ago

we're only adding hip fallback support for this arch, we are not running any assembly tests