Scalable checkpointing and restarting

The salient feature of this release is support for checkpointing and restarting for the quasi-Newton solver. These features use the axom's scalable sidre data manager.

What's Changed

• Update cmake paths be relative to hiop project by @chapman39 in #684
• Patch two memory leaks that cause more leaks by @tepperly in #685
• add new function for warmstart by @nychiang in #688
• Fix CI pipeline by @nychiang in #694
• Add working spack build with build cache. by @cameronrutherford in #695
• Checkpointing (quasi-Newton solver) by @cnpetra in #693

New Contributors

• @chapman39 made their first contribution in #684

Full Changelog: v1.0.3...v1.1.0

NLP Moving limits and misc fixes

What's Changed

• Update modules for CI tests on LLNL LC by @nychiang in #679
• Update cmake build system to require RAJA when GPU compute mode is used by @nychiang in #676
• Moving limits options for NLP IPM solvers by @cnpetra in #681

Full Changelog: v1.0.2...v1.0.3

v1.0.2

What's Changed

• Removed deprecated ALG2 for cusparseCsr2cscEx2 by @cnpetra in #671
• Addressed fixed buffer size vulnerability for vsnprintf by @nychiang in #673
• Removed stringent -Wall and -Werror from release builds to avoid downstream compilation errors

Full Changelog: v1.0.1...v1.0.2

C++17 compatible and misc fixes

What's Changed

Default C++ standard remains C++14

• Fix LLNL CI by @nychiang in #663
• C++17 support: compile -Wall/-Werror proof by @tepperly in #653
• Fix a bug in copying an empty matrix into a bigger matrix by @nychiang in #666
• Fix the approach used to update mu by @nychiang in #664

Full Changelog: v1.0.0...v1.0.1

Mature solvers interfaces and execution backends

Notable new features

Interfaces of various solvers reached an equilibrium point after HiOp was interfaced with multiple optimization front-ends (e.g., power grid ACOPF and SC-ACOPF problems and topology optimization) both on CPUs and GPUs. The PriDec solver reached exascale on Frontier after minor communication optimizations. The quasi-Newton interior-point solver received a couple of updates that increase robustness. The Newton interior-point solver can fully operate on GPUs with select GPU linear solvers (CUSOLVER-LU and Gingko).

• Instrumentation of RAJA sparse matrix class with execution spaces by @cnpetra in #589
• Fix Assignment Typo in hiopMatrixSparseCsrCuda.cpp by @pate7 in #612
• Use failure not failed in PNNL commit status posting by @cameronrutherford in #609
• rebuild modules on quartz by @nychiang in #619
• Use constraint violation in checkTermination by @nychiang in #617
• MPI communication optimization by @rothpc in #613
• fix memory leaks in inertia-free alg and condensed linsys by @nychiang in #622
• Update IPM algorithm for the dense solver by @nychiang in #616
• Use integer preprocessor macros for version information by @tepperly in #627
• use compound vec in bicg IR by @nychiang in #621
• Use bicg ir in the quasi-Newton solver by @nychiang in #620
• Add support to MPI in C/Fortran examples by @nychiang in #633
• Refactor CUSOLVER-LU module and interface by @pelesh in #634
• Add MPI unit test for DenseEx4 by @nychiang in #644
• Add more options to control NLP scaling by @nychiang in #649
• Development of the feasibility restoration in the quasi-Newton solver by @nychiang in #647
• GPU linear solver interface by @pelesh in #650

New Contributors

• @pate7 made their first contribution in #612
• @rothpc made their first contribution in #613

Execution spaces abstractions and misc fixes

This release hosts a series of comprehensive internal developments and software re-engineering to improve the portability and performance on accelerators/GPU platforms. No changes to the user interface permeated under this release.

Notable new features

A new execution space abstraction is introduced to allow multiple hardware backends to run concurrently. The proposed design differentiates between "memory backend" and "execution policies" to allow using RAJA with Umpire-managed memory, RAJA with Cuda- or Hip-managed memory, RAJA with std memory, Cuda/Hip kernels with Cuda-/Hip- or Umpire-managed memory, etc.

• Execution spaces: support for memory backends and execution policies by @cnpetra in #543
• Build: Cuda without raja by @cnpetra in #579
• Update of RAJA-based dense matrix to support runtime execution spaces by @cnpetra in #580
• Reorganization of device namespace by @cnpetra in #582
• RAJA Vector int with ExecSpace by @cnpetra in #583
• Instrumentation of host vectors with execution spaces by @cnpetra in #584
• Remove copy from/to device methods in vector classes by @cnpetra in #587
• Add support for Raja with OpenMP into LLNL CI by @nychiang in #566

New vector classes using vendor-provided API were introduced and documentation was updated/improved

• Development of hiopVectorCuda by @nychiang in #572
• Implementation of hiopVectorHip by @nychiang in #590
• Update user manual by @nychiang in #591
• Update the code comments in hiopVector classes by @nychiang in #592

Refinement of triangular solver implementation for Ginkgo by @fritzgoebel in #585

Bug fixes

• Refine the computation in normal equation system by @nychiang in #530
• Fix static culibos issue #567 by @nychiang in #568
• Fix segfault, remove nonsymmetric ginkgo solver by @fritzgoebel in #548
• Calculate the inverse objective scale correctly. by @tepperly in #570
• Fix hiopVectorRajaPar::copyToStartingAt_w_pattern by @nychiang in #569
• Gitlab pipeline refactor by @cameronrutherford in #597

New Contributors

• @tepperly made their first contribution in #570

Misc build system fixes

This minor release fixes a couple of issues found in the build system after the major release 0.7 of HiOp.

What's Changed

• fixing memory leaks in cuSolver by @kswirydo in #554
• moves pnnl_cleanup to .pre stage and removes files based on age by @jdmacam in #555
• Condensed/normaleqn linear system without CUDA/RAJA by @nychiang in #557
• Fix mds raja example for raja without GPU by @nychiang in #558
• git submodule fixes by @cnpetra in #561
• Misc build system fixes related to HIOP_SPARSE by @cnpetra in #560
• Disable openmp when raja is not used by @nychiang in #562
• Pre stage pipeline fix by @ryandanehy in #564
• New tag release - 0.7.1 by @cnpetra in #565

New Contributors

• @jdmacam made their first contribution in #555
• @ryandanehy made their first contribution in #564

Full Changelog: v0.7.0...v0.7.1

Fortran interface and misc fixes and improvements

• Fortran interface and examples
• Bug fixing for sparse device linear solvers
• Implementation of CUDA CSR matrices
• Iterative refinement within CUSOLVER linear solver class
• Improved robustness and performance of mixed dense-sparse solver for AMD/HIP

ginko integration and misc fixes

This tag provides an initial integration with ginko, fixes a couple of issues, and add options for (outer) iterative refinement.

HIP linear algebra workaround and update for RAJA > v0.14

This version/tag provides a workaround for an issue in the HIP BLAS and updates the RAJA code to better operate with the newer versions of RAJA.