Skip to content
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

Use the RoCM/HIP device to accelerate certain DPLASMA kernels #57

Merged
merged 50 commits into from
Aug 9, 2024

Conversation

abouteiller
Copy link
Contributor

This PR adds RoCM enabled kernels to the GEMM, PORTF and memory-aware GEMM operations.

@bosilca
Copy link
Contributor

bosilca commented Mar 31, 2023

As discussed on 03/31/23 we need to rebase and check the result. This will be tested next week on Frontier, we need it to be ready.

Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>

Conflicts:
	src/CMakeLists.txt
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Signed-off-by: Aurelien Bouteiller <[email protected]>
Use proper error checks instead of asserts
Signed-off-by: Aurelien Bouteiller <[email protected]>
@abouteiller
Copy link
Contributor Author

This is in ready to merge state beside the 'squash to less commits'.

@abouteiller abouteiller enabled auto-merge (squash) May 21, 2024 03:57
@abouteiller
Copy link
Contributor Author

I decided to go for a squash merge, this is ready for final review @bosilca @devreal

Error in ctest is #115, preexisting and unrelated to hip.

Copy link
Contributor

@bosilca bosilca left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why only certain kernels are HIP-enabled instead of having all CUDA-enabled kernel also be HIP-enabled ?

src/dplasmaaux.h Show resolved Hide resolved
src/zgemm_TN.jdf Show resolved Hide resolved
share/help-dplasma.txt Show resolved Hide resolved
src/dplasmaaux_hip.h Show resolved Hide resolved
use cublasv2 by default (req. for cusolver anyway) except if cublas(v1)
is included before the dplasmajdf/aux.h include
This reverts the logic for cublas.h determination, and is a first steps
toward deprecating cublas(v1) usage entirely
@abouteiller abouteiller force-pushed the feature/hip branch 5 times, most recently from e5ed5f0 to a985fa6 Compare August 9, 2024 20:07
@abouteiller abouteiller merged commit edf5be3 into ICLDisco:master Aug 9, 2024
2 of 3 checks passed
@abouteiller abouteiller deleted the feature/hip branch October 24, 2024 14:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants