Conversation
NVIDIA publishes official redistrib manifests for cuSPARSELt, so the component follows the cuDNN/NVSHMEM model: a new cusparselt_version attribute on cuda_ext.redist(...) pins a cuSPARSELt release per registered CUDA version, exposed as @cuda//cusparselt. Manifests are prefetched asynchronously alongside the CUDA/cuDNN/NVSHMEM ones. The version catalog covers every release NVIDIA publishes a manifest for (0.3.0 through 0.9.1); all manifest sha256s were computed from the downloaded files. Releases up to 0.7.1 publish one archive per platform (linux-x86_64 and linux-sbsa), while 0.8.0+ publish per-CUDA-major cuda_variant archives; the existing variant selection picks the archive matching the registered CUDA major, and requesting a version with no archive for that major fails during module extension evaluation. The cusparselt.BUILD.bazel template exports the full target family: cusparselt / cusparselt_shared_library, cusparselt_system / cusparselt_interface_library (linking through the archive's unversioned libcusparseLt.so, since cuSPARSELt ships no lib/stubs/ directory), cusparselt_static / cusparselt_static_library, plus headers, header_list, and shared_library_files. cusparselt_static also links the hermetic static CUDA runtime because libcusparseLt_static.a leaves CUDA runtime symbols undefined (the shared library embeds the runtime statically). headers depends on the cusparse and cudart header closures, since cusparseLt.h includes cusparse.h, driver_types.h, and library_types.h. shared_library_files stages only the fully versioned shared library: the SONAME (libcusparseLt.so.0) is a truncation of the versioned file name, so the runtime symlink is derivable. e2e: cusparselt_smoke, cusparselt_system_smoke, and (manual, like the other static smokes) cusparselt_static_smoke link against cusparseLtGetVersion; every registered CUDA version pins a cusparselt_version, covering both manifest shapes.
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
1 participant
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.
NVIDIA publishes official redistrib manifests for cuSPARSELt, so the component follows the cuDNN/NVSHMEM model: a new cusparselt_version attribute on cuda_ext.redist(...) pins a cuSPARSELt release per registered CUDA version, exposed as @cuda//cusparselt. Manifests are prefetched asynchronously alongside the CUDA/cuDNN/NVSHMEM ones.
The version catalog covers every release NVIDIA publishes a manifest for (0.3.0 through 0.9.1); all manifest sha256s were computed from the downloaded files. Releases up to 0.7.1 publish one archive per platform (linux-x86_64 and linux-sbsa), while 0.8.0+ publish per-CUDA-major cuda_variant archives; the existing variant selection picks the archive matching the registered CUDA major, and requesting a version with no archive for that major fails during module extension evaluation.
The cusparselt.BUILD.bazel template exports the full target family: cusparselt / cusparselt_shared_library, cusparselt_system / cusparselt_interface_library (linking through the archive's unversioned libcusparseLt.so, since cuSPARSELt ships no lib/stubs/ directory), cusparselt_static / cusparselt_static_library, plus headers, header_list, and shared_library_files. cusparselt_static also links the hermetic static CUDA runtime because libcusparseLt_static.a leaves CUDA runtime symbols undefined (the shared library embeds the runtime statically). headers depends on the cusparse and cudart header closures, since cusparseLt.h includes cusparse.h, driver_types.h, and library_types.h. shared_library_files stages only the fully versioned shared library: the SONAME (libcusparseLt.so.0) is a truncation of the versioned file name, so the runtime symlink is derivable.
e2e: cusparselt_smoke, cusparselt_system_smoke, and (manual, like the other static smokes) cusparselt_static_smoke link against cusparseLtGetVersion; every registered CUDA version pins a cusparselt_version, covering both manifest shapes.