Enable Kokkos Kernels with host and GPU dependency configuration - #601
Merged
Conversation
Enable Kokkos Kernels by default, supporting installed packages and FetchContent builds. Configure host LAPACKE/MKL independently of CUDA/HIP libraries and validate provider capabilities, linkage, and integer ABI. Add OPALX-inspired linear-map tests covering analytic eigenvalues and eigenvector residuals. Use Kernels GEMM and direct host LAPACKE because the upstream 5.2.0 eigenanalysis header fails with GCC 15 and Clang 21. Fix build-tree/install dependency propagation, add a standalone consumer test, document configuration, and forward dependency options through CI. Validated Serial and OpenMP builds, GCC and Clang, 1/2/4 MPI ranks, disabled configurations, and build-tree/install consumers. CUDA, HIP, MKL, and a working ILP64 provider remain unverified. Refs #600
aaadelmann
marked this pull request as draft
September 8, 2026 19:24
aaadelmann
added
enhancement
Member
Author
|
cscs-ci run cscs-ci-gh200, cscs-ci-mi300, cscs-ci-openmp |
Fetch and build reference LAPACK when a compatible LAPACKE installation cannot be found. Build the host BLAS, LAPACK, and LAPACKE libraries in an isolated C/Fortran project so CUDA and HIP compiler settings do not leak into the host dependency. Export and install the fallback libraries through a relocatable CMake package, support LP64 and ILP64 interfaces, preserve explicitly supplied LAPACKE libraries, and allow automatic fetching to be disabled with IPPL_FETCH_LAPACKE.
Preserve linker options beginning with -. A clean configure using an explicit LAPACKE link line containing -lm now passes, including the native eigenvalue probes. All five linear-map tests pass locally.
Placing Kokkos’s bundled desul headers first. I reproduced the CSCS collision locally with an intentionally invalid desul header beside lapacke.h; Kokkos Kernels still compiled successfully.
aaadelmann
marked this pull request as ready for review
September 8, 2026 21:02
srikrrish
reviewed
Sep 9, 2026
srikrrish
left a comment
srikrrish
left a comment
Member
There was a problem hiding this comment.
Actually, we need this only for very small matrices (like 6x6 etc.) if I am not wrong right? Isn't it a bit overkill to expose these many options? I am also worried about the dependency chain but since we need it and I currently don't see a work around at least to be consistent with rest of IPPL enables it can be OFF by default?
Describe the opt-in setting, host LAPACKE and MKL providers, LP64 and ILP64 selection, source-build fallback, offline FetchContent overrides, compiler selection, and automatic CUDA and HIP TPL configuration.
srikrrish
approved these changes
Sep 9, 2026
aaadelmann
enabled auto-merge
September 9, 2026 18:50
Member
Author
|
cscs-ci run cscs-ci-openmp |
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 10, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Enable Kokkos Kernels by default, supporting installed packages and FetchContent builds. Configure host LAPACKE/MKL independently of CUDA/HIP libraries and validate provider capabilities, linkage, and integer ABI.
Add OPALX-inspired linear-map tests covering analytic eigenvalues and eigenvector residuals. Use Kernels GEMM and direct host LAPACKE because the upstream 5.2.0 eigenanalysis header fails with GCC 15 and Clang 21.
Fix build-tree/install dependency propagation, add a standalone consumer test, document configuration, and forward dependency options through CI.
Validated Serial and OpenMP builds, GCC and Clang, 1/2/4 MPI ranks, disabled configurations, and build-tree/install consumers. CUDA, HIP, MKL, and a working ILP64 provider remain unverified.
Refs #600