v0.10.1

MadNLP v0.10.1

Diff since v0.10.0

Merged pull requests:

• [MadNLPTests] v0.6.2 (#619) (@michel2323)
• [MadNLP] Upgrade MUMPS to v5.9.0 (#620) (@amontoison)

v0.10.0

MadNLP v0.10.0

Diff since v0.9.2

Highlights: the main solve! loop is now allocation-free in steady state — ~99% of per-iteration allocations have been removed by replacing mapreduce-based reductions with explicit loops and extracting backend-specific kernels so that GPU code paths can plug in without touching the core. Plus a CUDA.jl v6 / CUDSS 0.7 / Julia 1.12 update for MadNLPGPU.

Performance

• Remove ~99% of per-iteration allocations from solve!. get_varphi, get_inf_du, get_inf_compl, get_min_complementarity, and friends in src/IPM/kernels.jl are rewritten as explicit loops instead of mapreduce. count_lu_bounds replaces the boolean-mask + sum pattern in get_vars_info / get_cons_info. Indexed updates in _set_aug_diagonal! are wrapped in @views to avoid temporary copies. The unused dense jac_buffer allocation in the NLPModels callback is removed. (#603)
• New populate_RR_nn! kernel for restoration-phase initialization, factored out so backends can override it. (#603)

Breaking changes

• RR field on MadNLPSolver is now typed Union{Nothing,RobustRestorer{T,VT}} (was RobustRestorer{T}). Downstream code that holds a typed reference to RobustRestorer needs the extra type parameter. (#603)
• get_vars_info and get_cons_info are now restricted to AbstractMadNLPSolver. (#603)
• New extension points (_get_varphi, get_varphi, get_inf_du, get_inf_compl, get_min_complementarity, get_varphi_d, get_alpha_max, get_alpha_z, restoration-phase get_*_R variants, populate_RR_nn!, count_lu_bounds, SubVector) are exported from MadNLP for backends to specialize. (#603)

---

MadNLPGPU v0.10.0

Highlights: bumped to CUDA.jl v6 and CUDSS 0.7, and GPU specializations of the new MadNLP core kernels.

Dependencies

• Require CUDA.jl ≥ 6.0.0 and CUDSS ≥ 0.7. (#615)
• Depend on the CUDA subpackages (CUDACore, cuBLAS, cuSOLVER, cuSPARSE) instead of the CUDA meta-package. The CUDA extension now activates from these subpackages directly, so MadNLPGPU works alongside any CUDA.jl v6.x without forcing the meta-package into the project. (#615)
• AMDGPU.jl ≥ 2.

New features

• GPU specializations of the IPM kernels refactored out in MadNLP #603 (get_varphi, get_inf_du, get_inf_compl, get_min_complementarity, restoration populate_RR_nn!, etc.), dispatched on AbstractGPUVectorOrSubVector. The GPU path keeps using mapreduce while the CPU path uses explicit loops — both avoid the per-iteration allocations that the prior code had. (#603)

Fixes

• Work around a Julia 1.12 regression with the global solver state in MadNLPGPU.jl. (#615)

Compatibility

• Requires MadNLP ≥ 0.10.0.

Merged pull requests:

• Removing 99% of allocations in solve! (#603) (@apozharski)
• compat CUDA@6.0.0 CUDSS@0.7 (#615) (@michel2323)

v0.9.2

MadNLP v0.9.2

Diff since v0.9.1

Highlights: new SchurComplementKKTSystem for two-stage stochastic programs (CPU + GPU), an API cleanup so solver and wrapper internals are accessed through getter functions, and a MUMPS 5.8.x fix.

New features

• Add SchurComplementKKTSystem for exploiting the block-arrowhead structure of two-stage stochastic programs. Reduces each Newton step to a dense nd × nd design-variable system via per-scenario eliminations. CPU variant uses Umfpack for the per-scenario blocks by default; GPU variant lives in the MadNLPGPU CUDA extension and uses batched cuDSS + strided-batched CUBLAS. (#598)

API changes

• Solver member access is now through getter functions instead of direct field access. Downstream code that reaches into MadNLPSolver fields will need to migrate. (#497)
• Wrapper models use getters for inner-model access. (#595)
• solve is now re-exported from SolverCore, aligning MadNLP with the broader JuliaSmoothOptimizers ecosystem. (#601)

Fixes

• Update MUMPS bindings for the 5.8.x ABI. (#613)

Documentation

• Fix mu_init in the warm-start tutorial. (#607)

---

MadNLPGPU v0.9.0

Highlights: CUDA/CUDSS and AMDGPU are now weak dependencies, and the new Schur-complement KKT system has a GPU backend.

• CUDA/CUDSS moved to weak deps, GPU backend code deduplicated. Installing MadNLPGPU no longer pulls CUDA on machines without it; the CUDA-specific code (LapackCUDASolver, CUDSSSolver, GPU Schur) lives in the MadNLPGPUCUDAExt extension and loads only when both CUDA.jl and CUDSS.jl are present. The AMDGPU backend follows the same pattern via MadNLPGPUAMDGPUExt. (#596)
• GPU SchurComplementKKTSystem: batched per-scenario factorizations via cuDSS, Schur accumulation via CUBLAS strided-batched GEMM. (#598)
• Requires MadNLP ≥ 0.9.2.

Known issue — V100 / Volta (sm_70):
CUDA Toolkit 13 dropped support for Volta GPUs. Starting at CUDA.jl v5.10 (March 2026), Pkg may resolve to a CUDA.jl that auto-selects the CUDA 13 runtime, which then fails on V100. MadNLPGPU's compat allows this. Until JuliaGPU/CUDA.jl#3134 ships in a tagged release, V100 users should pin to a CUDA 12.x toolkit by running once per environment:

using CUDA
CUDA.set_runtime_version!(v"12.9")    # any 12.x works

Alternatively, pin CUDA = "~5.9" in your project's Project.toml.

---

MadNLPTests v0.6.1

• Add Schur-complement two-stage stochastic test fixtures used by MadNLP and MadNLPGPU Schur tests. (#598)

Merged pull requests:

• Overhaul solver member access via functions (#497) (@apozharski)
• Wrappers should use getters for nlp models meta (#595) (@apozharski)
• Add SchurComplementKKTSystem for two-stage stochastic programs (#598) (@michel2323)
• Revert and use function solve from SolverCore (#601) (@frapac)
• [doc] Fix mu_init in warm-start tutorial (#607) (@frapac)
• [Mumps] Fix the MUMPS structure for v5.8.x (#613) (@frapac)

Closed issues:

• Documentation (#93)
• Restoration failed when using MadNLP for ODE parameter estimation (#264)
• Using weak dependencies (#265)
• lost of precision in constraints using MadNLP.SparseCondensedKKTSystem? (#593)
• Consider not exporting own solve! (#597)
• Performance regression in v0.9.1 compared to v0.8.12 with Mumps (#602)
• Warm-start Tutorial Appears to be Out-of-Date (#606)

v0.9.1

MadNLP v0.9.1

Diff since v0.9.0

Merged pull requests:

• AbstractWrapperModel api missing returns (#592) (@apozharski)

Closed issues:

• Release 0.9 (#545)
• QuadraticModel and multi precision (#591)

v0.9.0

MadNLP v0.9.0

Diff since v0.8.12

Breaking changes

• Update to MadNLP 0.9

Merged pull requests:

• Revisit fixed variables treatment (#360) (@frapac)
• Update MadNLPPardiso (#434) (@frapac)
• Using PrecompileTools.jl (#452) (@sshin23)
• Norm replaced with custom-norm for julia v1.12 (#453) (@sshin23)
• Implement adaptive barrier updates (#480) (@frapac)
• Dual regularize diagonal (#481) (@sshin23)
• Mumps tuned and made as default linear solver (#482) (@sshin23)
• Implement minor changes for c interface building (#483) (@apozharski)
• Print improvement (#485) (@sshin23)
• [bug fix] del_c should be determined by zero eig (#486) (@sshin23)
• [bug fix] correct inertia for unreduced systems (#487) (@sshin23)
• [CUDSS] Return dummy inertia if factorization incomplete (#488) (@sshin23)
• update MadNLPTests to use new NL interface in JuMP (#489) (@frapac)
• option treatment and specialization for cudss (#490) (@sshin23)
• fix typos in MadNLP's logs (#492) (@frapac)
• Update MOI extension for the operators (#493) (@amontoison)
• Fix the interface of the operators in MOI (#494) (@amontoison)
• Precompile fix (#495) (@sshin23)
• Fix tests broken in v1.12 (#498) (@apozharski)
• Fixes for AMDGPU (#499) (@apozharski)
• [MadNLPTests] Release 0.5.3 (#500) (@amontoison)
• [MadNLPGPU] Support CUDSS.jl v0.6 (#501) (@amontoison)
• fix type instability bug in dummy_qp (#502) (@apozharski)
• [MadNLPGPU.jl] Upgrade the interface for CUDSS.jl (#507) (@amontoison)
• Refactor default_sparse_solver to only use MumpsSolver (#511) (@amontoison)
• Add formatter (#515) (@michel2323)
• Flip objective in stats when maximizing (#519) (@klamike)
• Add tutorial on how to use MadNLP on the GPU (#520) (@frapac)
• [MadNLPGPU] Support LBFGS on GPU (#524) (@amontoison)
• [MadNLPGPU] Support uniform batch in CUDSSSolver (#527) (@amontoison)
• Move {Sparse,Dense}WrapperModel from MadNLPGPU to MadNLP (#528) (@frapac)
• Abstract stats (#529) (@klamike)
• [MadNLP] Update the implementation of LBFGS (#530) (@amontoison)
• Fix barrier initialization with type conversion (#532) (@amontoison)
• [MadNLPGPU] Fix CUDSS uniform batch syntax (#533) (@klamike)
• Use syr! and symv! in BFGS (#534) (@amontoison)
• Update the implementation of LBFGS in MadNLP (#536) (@amontoison)
• [hotfix] remove symul! in favor of _symv! (#538) (@sshin23)
• Properly map NLP constraint index for MOI dual (#540) (@klamike)
• Fix hessian for dense KKT (#541) (@klamike)
• Removed synchronize(backend) (#542) (@sshin23)
• Remove extra build_kkt! call (#544) (@klamike)
• Update the signature of the function solve! in MadNLP (#546) (@frapac)
• Apply obj_scale in unpack_z in MakeParameter case (#547) (@klamike)
• Add support for MOI.VectorNonlinearOracle (#548) (@frapac)
• [MOI] Add support for MOI.LagrangeMultiplier (#549) (@amontoison)
• [MOI] Reuse existing data structures from MOI on re-solve (#550) (@amontoison)
• Support precompilation with MOI extension (#551) (@amontoison)
• Add tests for multipliers' values in scaled and max problems (#553) (@frapac)
• [MOI] Optimize MOI.eval_constraint_jacobian_transpose_product when we have oracles (#554) (@amontoison)
• LBFGS is only supported with the KKT formulation SparseKKTSystem (#555) (@amontoison)
• [MOI/MadNLPTests] Add array_type to support GPU solve from MOI (#557) (@klamike)
• Fix generation of documentation (#558) (@frapac)
• Update the documentation for MadNLP 0.9 (#560) (@amontoison)
• Remove NEWS.md (#561) (@amontoison)
• [MOI] Drop support for ConstraintDualStart for variable bounds (#564) (@frapac)
• Add a field additional_buffers in CompactLBFGS (#565) (@amontoison)
• Add a function madsuite (#566) (@amontoison)
• Fix LBFGS in single precision (#567) (@amontoison)
• Refactor badge links in README.md (#568) (@amontoison)
• Rename the dense linear solvers on GPU (#569) (@amontoison)
• Update the documentation for MadNLP 0.9 (#570) (@frapac)
• rename solve! function for KKT and linear systems (#571) (@frapac)
• [documentation] Update the tutorial related to the custom KKT system (#572) (@amontoison)
• Default primal-dual regularization (#573) (@sshin23)
• [documentation] Fix a few errors in kkt.md (#574) (@amontoison)
• [documentation] Update algorithm.md (#575) (@amontoison)
• Avoid passing array_type to MadNLPSolver in MOI (#576) (@klamike)
• Use solve_kkt! instead of solve_kkt_system! (#577) (@amontoison)
• Guard against Julia versions built without GPL libs (#578) (@asinghvi17)
• Use the new API of NLPModels.jl for dense Jacobians and Hessians (#580) (@amontoison)
• Support LBFGS in extended precision (#581) (@amontoison)
• Add suppport for user defined callbacks (#582) (@frapac)
• [MOI] Fix support for MOI.Parameter constraints (#584) (@amontoison)
• Use variable_bounds_analysis and constraint_bounds_analysis in MOIModel (#585) (@amontoison)
• Make options type concrete in MadNLPSolver to prevent some minor type instability (#588) (@apozharski)
• Report the number of KKT factorizations in MadNLP (#589) (@frapac)
• [MadNLP] Release 0.9.0 (#590) (@frapac)

Closed issues:

• Revisiting fixed variable treatment (#70)
• Add different strategies to update the barrier mu (#96)
• LapackGPUSolver goes into feasibility restoration when LapackCPUSolver does not (#205)
• Definition of primal solution? (#212)
• Fixed variable multiplier (#229)
• Wrong solution in QuickStart example (#238)
• README.md simplification (#324)
• Unable to get GPU solver working (#381)
• [MadNLPGPU] Use CHOLESKY insted of LDL by default? (#438)
• Passing x, y, zl, and zu, to a first call of solve! ignores these arguments. (#450)
• Allow for LinearSolve.jl linear solver choices? (#479)
• No change to objective over course of optimization (#484)
• [Bug] Linear solver backend (#505)
• [MOI] Support VectorNonlinearOracle (#508)
• Support for user given callback functions (#512)
• [documentation] Explain how to use an AbstractNLPModels to solve NLP problems on GPU (#513)
• Existing options keys (#516)
• Maximization problem and objective value (#517)
• Issue with max (instead of min) with CUDABackend (#518)
• How to use L-BFGS on NVIDIA / AMD GPUs? (#522)
• [documentation] Specify the packages to load for the tutorial on LBFGS (#523)
• [documentation] Explain how to use BFGS in MadNLP (#525)
• [MOI] GPU syntax in readme is not working (#556)
• Should we keep the file NEWS.md? (#559)
• Multi-precision and LBFGS (#562)
• mu_init parameter (#579)

v0.8.12

MadNLP v0.8.12

Diff since v0.8.11

Merged pull requests:

• Interface the dense EVD on GPU (#474) (@amontoison)
• [hotfix] Fix method overwriting in new Lapack interface (#478) (@frapac)

v0.8.11

MadNLP v0.8.11

Diff since v0.8.10

Merged pull requests:

• [MadNLPGPU] Add an extension for AMDGPU.jl (#464) (@amontoison)
• Support dense EVD as a LAPACK solver on CPU (#475) (@amontoison)
• Use TRSV instead of TRSM in LAPACK solvers (#477) (@amontoison)

Closed issues:

• Support AMDGPU.jl in MadNLPGPU.jl (#380)
• breakage in OptimalControl.jl v1.1 (#443)

v0.8.10

MadNLP v0.8.10

Diff since v0.8.9

Merged pull requests:

• [MadNLPGPU] Update the interface for CUSOLVER (#466) (@amontoison)
• Split the code for the GPU kernels and CUDA.jl (#472) (@amontoison)
• [MadNLP] Update the interface for LapackCPUSolver (#473) (@amontoison)

Closed issues:

• [MadNLPGPU] Update the interface for CUSOLVER (#460)

v0.8.9

MadNLP v0.8.9

Diff since v0.8.8

Merged pull requests:

• [MadNLPGPU] Upgrade CUDSS.jl (#439) (@amontoison)
• CI fix (#440) (@sshin23)
• Update docs.yml (#444) (@amontoison)
• Update test.yml (#445) (@amontoison)
• [MadNLPGPU] Expose more options in cuDSS wrapper (#446) (@frapac)
• [MadNLPGPU] Support the hybrid execute mode of cuDSS (#447) (@amontoison)
• [MadNLPGPU] Support more ordering (#448) (@amontoison)
• Fix option bound_relax_factor if set to zero (#449) (@frapac)
• Fix scaling of multipliers if nlp_scaling=true (#451) (@frapac)
• fix bug in re-initialize (#454) (@apozharski)
• Implement centering step as done in ipopt (#455) (@apozharski)
• Fix MOI.is_valid for ConstraintIndex{VariableIndex,Parameter} (#456) (@guimarqu)
• Update madnlpgpu_test.jl (#457) (@amontoison)
• Don't support CUDA 5.8.3 (#458) (@amontoison)
• CI fix and clean up (#462) (@sshin23)

Closed issues:

• Unable to use CompactLBGS in Float32 (#420)
• Allow Setting bound_relax_factor to zero (#430)

v0.8.8

MadNLP v0.8.8

Diff since v0.8.7

Merged pull requests:

• Add tutorial for implementing a custom KKT system in MadNLP (#413) (@frapac)
• Fix floating-point for options (#424) (@michel2323)
• Support MUMPS v5.8.0 (#427) (@amontoison)
• Factorize line-search in function filter_line_search! (#428) (@frapac)
• [MadNLPGPU] Update sparse.jl (#429) (@amontoison)
• add helper function to call reduce_rhs! (#431) (@frapac)
• Update utils.jl (#433) (@amontoison)
• Don't call jac_structure! if nnzj == 0 (#435) (@amontoison)
• Add loqo rule and hack arround missing fallback in robust! (#436) (@apozharski)
• Add a kernel for K2.5 IN MadNLPGPU.jl (#437) (@amontoison)

Closed issues:

• Issue with UMFPack for large sparse matrices (#368)
• Don't allocate vectors of size nnzj or nnzh if not needed (#422)
• Float32 in BFGS (#423)
• Add violations of complementary constraints to the logs (#432)