Skip to content

Release 1.9.0

Latest
Latest
Compare
Choose a tag to compare
@MarcelKoch MarcelKoch released this 09 Dec 14:45
v1.9.0
20cfd68
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Learn about vigilant mode.

The Ginkgo team is proud to announce the new Ginkgo minor release 1.9.0.
This release brings new features such as:

  • Support for half precision (IEEE FP16). The type gko::half can now be selected in most instances as the value type
    of a matrix, solver, preconditioner, etc. If the selected backend supports FP16 as a native type, the native type is
    used within the kernels, otherwise an overhead might occur. The new behavior is enabled by default, but it can be
    turned off during configuration.
  • New implementations of the ILU and IC factorization for CUDA, HIP, OpenMP, and Reference backends. These are
    available in addition to the existing implementations based on the vendor libraries cuSPARSE and hipSPARSE.
  • New (S)SOR and Gauss-Seidel preconditioners.
  • Simplified distributed matrix assembly by exchanging local rows between neighboring processes.

And more!

If you face an issue, please first check our known issues page and the open issues list and if you do not
find a solution, feel free to open a new issue or ask a question using the github discussions.

Supported systems and requirements:

  • For all platforms, CMake 3.16+
  • C++17 compliant compiler
  • Linux and macOS
    • GCC: 7.0+
    • clang: 5.0+
    • Intel compiler: 2019+
    • Apple Clang: 15.0 is tested. Earlier versions might also work.
    • NVHPC: 22.7+
    • Cray Compiler: 14.0.1+
    • CUDA module: CMake 3.18+, and CUDA 11.0+ or NVHPC 22.7+, Compute Capability 5.3+
    • HIP module: CMake 3.21+, and ROCm 4.5+
    • DPC++ module: Intel oneAPI 2023.1+ with oneMKL and oneDPL. Set the CXX compiler to dpcpp or icpx.
    • MPI: standard version 3.1+, ideally GPU Aware, for best performance
  • Windows
    • MinGW: GCC 7.0+
    • Microsoft Visual Studio: VS 2019+
    • CUDA module: CUDA 11.0+, Microsoft Visual Studio
    • OpenMP module: MinGW.

Version support changes

  • Ginkgo now requires a compiler with C++ 17 support #1603

Deprecations

  • The Executor::run overload taking in multiple functions without a name as first parameter has been deprecated #1667
  • The master branch has been deprecated in favor of a new branch named main #1739.

Summary of previous deprecations

  • The device_reset parameter of CUDA and HIP executors no longer has an effect, and its allocation_mode parameters have been deprecated in favor of the Allocator interface.
  • The CMake parameter GINKGO_BUILD_DPCPP has been deprecated in favor of GINKGO_BUILD_SYCL.
  • The gko::reorder::Rcm interface has been deprecated in favor of gko::experimental::reorder::Rcm based on Permutation.
  • The Permutation class' permute_mask functionality.
  • Multiple functions with typos (set_complex_subpsace(), range functions such as conj_operaton etc).
  • gko::lend() is not necessary anymore.
  • The classes RelativeResidualNorm and AbsoluteResidualNorm are deprecated in favor of ResidualNorm.
  • The class AmgxPgm is deprecated in favor of Pgm.
  • Default constructors for the CSR load_balance and automatical strategies
  • The PolymorphicObject's move-semantic copy_from variant
  • The templated SolverBase class.
  • The class MachineTopology is deprecated in favor of machine_topology.
  • Logger constructors and create functions with the executor parameter.
  • The virtual, protected, Dense functions compute_norm1_impl, add_scaled_impl, etc.
  • Logger events for solvers and criterion without the additional implicit_tau_sq parameter.
  • The global gko::solver::default_krylov_dim, use instead gko::solver::gmres_default_krylov_dim.
  • array::get_num_elems() has been renamed to get_size()
  • matrix_data::ensure_row_major_order() has been renamed to sort_row_major()
  • device_matrix_data::get_num_elems() has been renamed to get_num_stored_elements()
  • The CMake parameter GINKGO_COMPILER_FLAGS has been superseded by CMAKE_CXX_FLAGS, and GINKGO_CUDA_COMPILER_FLAGS has been superseded by CMAKE_CUDA_FLAGS
  • The std::initializer_list overloads of matrix create methods and constructors are deprecated in favor of explicit array parameters

Added features

  • Add Executor::get_description() for textual representation of the device #1615
  • Add row and column scaling functionality to the distributed matrix #1640
  • Add SolverProgress logger printing out or storing to disk the individual scalars (and vectors) of an iterative solver after each iteration #1620
  • Add new ortho_method parameter for GMRES, with classical Gram-Schmidt and classical Gram-Schmidt with re-orthogonalization options in addition to previously-available modified Gram-Schmidt #1646
  • Add file config support for Schwarz #1658
  • Add overload for Executor::run which accepts a name and a closure for the ReferenceExecutor as the first two arguments #1667
  • Add function to fill device_matrix_data with zeros #1683
  • Add (S)SOR and Gauss-Seidel preconditioner #1633, #1634
  • Add support for additive read_distributed for the distributed matrix #1650
  • Add Ginkgo's own ILU and IC implementation #1684
  • Add NVIDIA Ada architecture #1733
  • Add half precision support #1706, #1708, #1711, #1712, #1713, #1716, #1710, #1736

Improvements

  • Add workspace in residual norm check #1687, which reduces the alloc/free and corresponding overhead.
  • Add distributed VectorCache and use it as workspace in Schwarz #1688.
  • Add example to show the file config usage #1662
  • Improve compile time for batched solvers #1629
  • Reduce conflicting thrust symbols when linking with different thrust libraries by adding a custom thrust namespace #1730

Fixes

  • Fix using the same algorithm as the original triangular solver when creating the transposed of the solver #1641
  • Fix the inconsistent behavior on the zero diagonal value in scalar Jacobi #1642
  • Fix an issue related to GCR and non-default strides in the rhs vector #1656
  • Fix an issue related to triangular solvers with CUDA on Windows #1665
  • Fix an issue where non-conforming MatrixMarket files were parsed without an error #1628
  • Fix finding rocthrust if it's not installed paths included by default #1668
  • Fix an issue related to casting between vectors of different value types in the mixed-precision multigrid setup #1663
  • Fix some test failures with ROCm 6.x #1670
  • Fix a race condition in bicgstab #1676
  • Fix an issue with MGS GMRES for complex numbers #1678
  • Fix finding ROCm on recent ROCm version (5.0+) #1673
  • Fix a compiler error when using NVHPC with MPI enabled #1697
  • Fix build issues of OMP backend when using HIPCC as C++ compiler #1695
  • Fix build issues for Intel OneAPI 2025.0 #1718
  • Fix inconsistencies between declaration and definition of functions and classes/structs, which mainly fixes clang-cl #1725
  • Fix undefined symbols in shared library in msys2/clang #1724
  • Fix page fault issues when running on multiple Intel GPUs in parallel #1723
  • Fix data races in several OMP kernels #1743
Assets 2
Loading