Releases: eth-cscs/spla
Releases · eth-cscs/spla
SPLA 1.2.1
SPLA 1.2.1 Release Notes
Features
- Added CMake option to specify the BLAS library
Bug fixes
- Fixed an internal error being thrown for certain large sizes with
pgemm_ssb
SPLA 1.2.0
SPLA 1.2.0 Release Notes
Features
- Improved performance with multiple threads and small matrix sizes.
- Added threshold parameter, below which gemm computations may be done on Host, even if GPU is set a processing unit.
Bug fixes
- If OpenBLAS is compiled without thread-safety, multi-threading is disabled for gemm computation.
SPLA 1.1.1
SPLA 1.1.1 Release Notes
Bug fixes
- Fixed CMake find module for AMD HIP libraries
SPLA 1.1.0
SPLA 1.1.0 Release Notes
Changes
- Changed tiling strategy for host and GPU computations for improved performance
- Rewrote and extended tests
Bug fixes
- Fixed issue with
pgemm_ssbif localk = 0 - Added workaround for bug in ROCm 3.5 related to device pointer detection
- Fixed bug with Intel MKL reporting invalid arguments with certain edge cases
- Fixed parameter check, such that null pointers are valid if local slice size is 0
SPLA 1.0.0
v1.0.0 change version to 1.0.0