Skip to content

Releases: icl-utk-edu/magma

MAGMA-2.9.0

28 Jan 14:01
@abdelfattah83 abdelfattah83
v2.9.0
48a32d7
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
MAGMA-2.9.0 Latest
Latest
  • [IMPORTANT]: The official tarball can be downloaded here: https://icl.utk.edu/magma. Releases available on GitHub run MAGMA's precision generator first before building the library.
  • [DEPRECATED] The MAGMA-sparse package is considered to be "Legacy Support Mode" only, meaning the MAGMA-sparse component is not under active development. Many sparse routines have been deprecated. Users are encouraged to consider switching to the Ginkgo library (https://ginkgo-project.github.io/) to find newer and actively-maintained implementations of deprecated functionality. Deprecated routines will be removed in the next release.
  • [DEPRECATED] The MAGMA V1 interface is deprecated and will be removed in the next release. This includes the "magma.h" header and all routines specified in the headers under the "include" subdirectory with the suffix "_v1.h"
  • [NEW] Added expert interfaces for the following routines:
    • getrs_gpu (magma_?getrs_expert_gpu_work)
    • getri_gpu (magma_?getri_expert_gpu_work)
    • trtri_gpu (magma_?getri_expert_gpu_work)
    • potrs_gpu (magma_?potrs_expert_gpu_work)
    • gegqr_gpu (magma_?gegqr_expert_gpu_work)
    • geqp3_gpu (magma_?geqp3_expert_gpu_work)
  • [IMPORTANT] The workspace requirements of the following routines have been updated. Users are advised to check the documentation for the new workspace sizes.
    • magma_?gegqr_gpu
    • magma_?geqp3_gpu (still backward compatible)
  • The magma_?gbtf2_native and magma_?gbtf2_native_v2 routines have been disabled for CUDA <11.0.
  • Bug fixes and improvements:
    • magma_zgeqp3_gpu: Fix failures at certain sizes (only double-complex)
    • magma_?gesv_rbt_batched: Fix incorrect results for certain sizes
    • Fix failures at certain sizes for several Hermitian eigensolver routines in double-complex precision
    • Fix 64-bit integer builds for CUDA-12.6
    • Fixes and improvements to CMake
    • Add support for Ada Lovelace architecture in Makefile
    • Fixes to documentation
    • Fix shpotrf_gpu to work for CUDA <11.0
    • Sparse: fix a bug in the use of cuSPARSE CSR -> CSC conversion for CUDA >=11.0
    • Sparse: fix initialization in LSQR solver for non-zero initial guesses
Assets 2
Loading