-
Notifications
You must be signed in to change notification settings - Fork 228
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add ck mha fp16 solver #3277
Add ck mha fp16 solver #3277
Conversation
It would be nice to have some tests as well. |
Absolutely! Working on those now, and this will live in an integration branch until those are ready. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
changes requested related to ExecutePrimitive function call
- Update requirements to CK that generates MHA for gfx90a - Fix compilation issue for No CK - Use provided tensor strides instead of calculating
ROCm/composable_kernel@9c0811f39a2262dbe4d71b81898187951c1e11ba -DCMAKE_BUILD_TYPE=Release -DINSTANCES_ONLY=ON | ||
ROCm/composable_kernel@294cb823142a815170cf1faa63e01a431a557a04 -DCMAKE_BUILD_TYPE=Release -DINSTANCES_ONLY=ON |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Note, this is pointing to a dev commit where I added generating the device_mha_operations lib for gfx90a as well.
This will need to be updated to point to an amd-develop commit instead once it's been promoted.
Since this is living in a integration branch it should be fine for now, but we shouldn't merge the integration into develop until this is resolved.
95186b8
into
bharriso/integrate-ck-mha-fp16
Purpose of review:
Note: