GPU support for sparse linear algebra#60
Draft
pelesh wants to merge 10 commits into
Draft
Conversation
pelesh
added
enhancement
nkoukpaizan
force-pushed
the
olcf-hackathon-2026-dev
branch
from
b332272 to
d023534
Compare
* Fix line constraint bug in HiOp MDS module. --------- Co-authored-by: pelesh <pelesh@users.noreply.github.com>
…ns (#31) * test_acopf with PBPOLRAJAHIOPSPARSE (verify equality constraint Jacobian) + bug fix for PBPOL. Co-authored-by: William A Perkins <william.perkins@pnnl.gov> * Cleanup changes to RAJAHIOPSPARSE nnz computation. * Workaround for RAJAHIOPSPARSE inequality constraint verification. * Handling cases where IPOPT or HIOP are not available in test_acopt + bug fixes. * HiOp-compatible checks for constraint Jacobian allocations. Will need a smarter way to verify the stacked inequality constraints. * Fix for inequality constraint offsets. --------- Co-authored-by: William A Perkins <william.perkins@pnnl.gov> Co-authored-by: nkoukpaizan <nkoukpaizan@users.noreply.github.com>
* Add more comments to HIOPRAJASPARSE implementation. * [skip ci] Add headers to the matpower file for Jacobian test. --------- Co-authored-by: pelesh <pelesh@users.noreply.github.com>
* I am not going to lie, Cursor agent heavily helped me with this. Replace PETSc-based inequality Jacobian with GPU RAJA kernels Move the inequality constraint Jacobian computation for the HiOp sparse GPU solver entirely to the device, eliminating the per-iteration host back and forth (copy to host, PETSc compute, MatGetRow extraction, values copy back to device). Elimiate PETSc use from this part of the code. Three RAJA kernels now compute directly into device memory: - Generator set-point constraints (AGC) - Voltage-reactive-power bounds (FIXED_WITHIN_QBOUNDS) - Line flow limits (Sf^2/St^2 derivatives + slack variables) Supporting changes: - Analytical NNZ counting replaces PETSc MatGetInfo at solver setup - New device-side parameter fields (apf, vs, xpdevidx, xslackidx, bus-to-gen mapping) added to *ParamsRajaHiop structs - Sparse position indices assigned at model setup for all three contribution types Includes validation test (test_ineqjac_gpu) that solves with IPOPT, then compares PETSc and GPU Jacobian values at the converged solution. Optional -benchmark flag for performance comparison. Made-with: Cursor --------- Co-authored-by: kswirydo <kasia.swirydowicz@gmail.com> Co-authored-by: pelesh <pelesh@users.noreply.github.com> Co-authored-by: Nicholson Koukpaizan <koukpaizannk@ornl.gov>
* Port equality constraint Jacobian from PETSc to RAJA GPU kernels Replace the PETSc-based equality constraint Jacobian computation in the PBPOLRAJAHIOPSPARSE model with direct GPU kernels using RAJA, eliminating the D2H-compute-H2D round trip. The sparsity pattern is now computed on the host during setup and the values are computed entirely on device. Key changes: - Add ComputeEqJacValuesGPU_PBPOLRAJAHIOPSPARSE in new gpu.cpp/hpp files - Add device arrays for flat-array indices (bus eqjacsp_selfidx, line eqjacsp_idx/eqjacsp_diag_idx/isdcline, gen xpdevidx/xpsetidx) - Fix nnz counting bugs (missing gen/load entries, off-by-one in line loop) and populate flat-array indices during model setup - Replace PETSc MatGetRow extraction in sparsity and values phases - Handle parallel lines by sharing off-diagonal positions with atomicAdd - Use pre-computed nnz in get_sparse_blocks_info instead of PETSc query - Add correctness test (test_eqjac_compare) and performance benchmark (test_eqjac_perf) Made-with: Cursor * Add comments and annotations to eq Jacobian GPU port * Sort equality constraints Jacobian entries for hiopsparse_gpu (#53) * Reset pbpolrajahiopsparse equality constraint function pointer. * Sort equality constraint indices on host before copying to device. * _selfidx --> _idx. * Store equality constraints Jacobian permutation. * Equality constraints Jacobian index permuation on device and use in setting the values. * Need to store the reverse permutation map. --------- Co-authored-by: nkoukpaizan <nkoukpaizan@users.noreply.github.com> Co-authored-by: kswirydo <kswirydo@users.noreply.github.com>
* Hessian --> hessian in unit test folders * ConstraintHessian --> ConstraintsHessian in hiopkernels. * UNIT_TEST_ --> UNIT_TESTS_. * Capitalize test names for consistency. * EQJAC --> EQUALITY_CONSTRAINT_JACOBIAN in test name. * Include OPFLOW_ in related objective and constraints Jacobian unit test names. * Cleanup a few CMake guards. * Reorganize opflow unit tests. * Remove option from test_ineqjac_gpu, i.e, always evaluate performance. * Minor edits to comments and print statements. * Fix test_eqjac NETFILES and remove unnecessary command line option. * Consolidate test_eqjac_gpu and test_eqjac_perf. * Minor comment/print statement fix. * Remove test_eqjac_perf from CMakeList. --------- Co-authored-by: nkoukpaizan <nkoukpaizan@users.noreply.github.com>
nkoukpaizan
force-pushed
the
olcf-hackathon-2026-dev
branch
from
d023534 to
f7e07c6
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Merge request type
Relates to
This MR updates
Summary
Evaluation of constraints Jacobians (equality and inequality) on GPUs. This is still work in progress and some code cleanup is needed. The stretch goal is to add sparse Hessian evaluation, as well.
Linked Issue(s)