Update APIs - SOCP support by Iroy30 · Pull Request #1979 · NVIDIA/cuopt

Iroy30 · 2026-09-22T22:59:51Z

Description

follows #1941 and #1913
Barrier sequence solves (update_linear_objective and update_rhs) now cover second-order cone programs built from quadratic constraints.
The cache stores the expansion: the pre-expansion sizes, the column permutation, the cone start before and after inequality slacks are inserted, the appended cone-row right-hand sides, and the rows that proved a cone head nonnegative. The crush replays that map, the slack shift, and the free-variable equality substitution so the update lands on the cached barrier problem. The quadratic-constraint constant is kept in the cone-row tail and is not overwritten.
A warm reuse restores the cone block to the initial diagonal before the first factorization. Reuse is refused when cone variables were aliased, and the cone layout must match the cached one.

Issue

Checklist

I am familiar with the Contributing Guidelines.
Testing
- New or existing tests cover these changes
- Added tests
- Created an issue to follow-up
- NA
Documentation
- The documentation is up to date with these changes
- Added new documentation
- NA

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

This PR fixes build after the latest RMM merge broke our pipeline (rapidsai/rmm@6646d15). device_scalar no longer accepts a r-value constructor. Replaced with common constants as inline constexpr that are passed instead of r-value constants.   - [ ] I am familiar with the [Contributing Guidelines](https://github.com/NVIDIA/cuopt/blob/HEAD/CONTRIBUTING.md). - Testing - [ ] New or existing tests cover these changes - [ ] Added tests - [ ] Created an issue to follow-up - [ ] NA - Documentation - [ ] The documentation is up to date with these changes - [ ] Added new documentation - [ ] NA

The cache-reuse rebind left one destructor check as settings_. instead of settings_->, which only compiles on CU13 wheels. Signed-off-by: root <root@ipp1-3302.aselab.nvidia.com>

Keep cache-reuse symbolic_done_ and main's explicit CUstream initialization. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

…augmented path, and the never-read device_A_x_values snapshot in iteration_data_t; no numerical change. Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

Crush a new user-space constraint RHS into the cached barrier workspace so a sequence re-solve can skip convert/presolve/scaling, mirroring update_linear_objective. - crush_user_rhs in barrier_transform.hpp negates 'G' rows, checks the rows presolve dropped as empty, gathers remaining_constraints and divides by row_scales. rhs_shift and rhs_update_supported are recorded on the first solve; range rows and folding are refused. - Empty rows dropped at t=0 are tested against the solve's primal_tol rather than exact zero, and an infeasible one short-circuits the next Solve to INFEASIBLE without running IPM. - The single c_dirty flag becomes dirty()/mark_clean() over separate c/b flags so further update APIs can reuse the same gate. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

Recovered from 7d61502; it was deleted by the 698afbe log cleanup. Updated for the update_rhs naming and the deferred setter cache-invalidation gap. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

…e_apis Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com> # Conflicts: # cpp/src/barrier/barrier.cu # cpp/src/barrier/device_sparse_matrix.cuh

new_slacks.empty() was far too strict: convert_less_than_to_equal adds a slack for every inequality row, so any model with an inequality was refused. Only convert_range_rows destroys the RHS (it zeroes rhs[i] and moves the bounds onto the slack); artificials leave rhs alone and convert_greater_to_less negates it, which the crush already mirrors. Gate on num_range_rows instead. Verified with a QP over a G row: two successive update_rhs re-solves take the reuse path, skip presolve / reordering / symbolic factorization, and match a fresh full solve. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

The repo had no sequence_solve coverage at all. These compare every cached re-solve against a fresh full solve of the same model, so no assertion depends on a hand-derived optimum, and each asserts the reuse log line so a test cannot pass while the gate quietly rejects the model and falls back to a full solve. Models force the crush paths a one-row QP leaves as no-ops: mixed E/L/G senses, row norms seven orders of magnitude apart (non-unit row_scales), nonzero variable lower bounds (rhs_shift of -7; dropping it moves the optimum 115%), and an empty row presolve drops, covering both the feasible case and the short-circuit to PrimalInfeasible with no IPM and a surviving cache. Checked by mutation: removing barrier_presolve_bound_free_variables=0 fails 5 of the 6, all reporting the fallback. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

…gmented sort, reuse it for a new on-device transpose so the augmented path drops the host A^T build and upload, and cover both with a unit test. Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com> # Conflicts: # cpp/src/barrier/device_sparse_matrix.cuh

chris-maes · 2026-10-09T01:28:54Z

-    original[static_cast<std::size_t>(i)] =
-      xf.row_sense[static_cast<std::size_t>(i)] == 'G' ? -b[i] : b[i];
+  const std::size_t user_num_rows = xf.user_num_rows;
+  for (std::size_t i = 0; i < user_num_rows; ++i) {


const i_t user_num_rows = xf.user_num_rows;
for (i_t i = 0; i < user_num_rows; ++i)

chris-maes · 2026-10-09T01:31:42Z

-                                                          op_problem.has_quadratic_objective(),
-                                                          false);
+    // Must stay in lockstep with the gate in solve_linear_program_with_barrier: this path swaps
+    // in the slim user_problem_from_transform, so disagreement runs presolve on a fabricated


What is a fabricated problem?

user_problem_from_transform builds a stand-in problem, not the converted model. The wording is changed now

Iroy30 · 2026-10-09T02:45:12Z

/ok to test 2e1a10e

github-actions · 2026-10-09T03:43:23Z

CI Test Summary

✅ All 32 test job(s) passed.

Iroy30 · 2026-10-09T05:33:31Z

/ok to test 2f022ba

yuwenchen95

Overall is good to me, with some minor comments. Thanks!

yuwenchen95 · 2026-10-09T08:10:09Z

-  bool barrier;               // true to use barrier method, false to use dual simplex method
+  // Equality substitution of zero-cost free variables. A barrier cache cannot refresh the
+  // stored pivot RHS, so a solve that fills the cache turns this off.
+  bool barrier_eliminate_free_variables = true;


nit: every other field here is initialized in the constructor's init list; could this follow the same pattern (barrier_eliminate_free_variables(true) right after barrier_presolve(false)) with a one-line trailing comment like its neighbors?

yuwenchen95 · 2026-10-09T08:42:23Z

 {
  lp_status_t status                                   = lp_status_t::UNSET;
  simplex_solver_settings_t<i_t, f_t> barrier_settings = settings;
+  // The cached presolve records cannot be replayed after an RHS update, so the fill solve


It is to disable a presolve step automatically. Can we address it like effective_bound_free_variables?

added effective_eliminate_free_variables like the effective_bound_free_variables

yuwenchen95 · 2026-10-09T09:25:05Z

  impl_->rhs_dirty = true;
 }

+#ifdef DUAL_SIMPLEX_INSTANTIATE_DOUBLE


Do we really need #ifdef here?

Was included to addtress #1979 (comment)

The #ifdef is how we decide to instantiate the int, double template. It is needed.

yuwenchen95 · 2026-10-09T09:26:53Z

  const i_t problem_rows = user_problem.original_num_rows;
  std::vector<i_t> row_nz(problem_rows, 0);
  for (i_t j = 0; j < user_problem.num_cols; ++j) {
-    for (i_t p = A.col_start[j]; p < A.col_start[j + 1]; ++p) {


It looks the same as before. Can we undo the change here?

It was addressing a previous review
#1979 (comment)

yuwenchen95 · 2026-10-09T09:27:12Z

      cone_head_bound_t<i_t, f_t> bound;
      bound.head_col = head;
      // A is CSC, so the head's own column already lists every row it appears in.
-      for (i_t p = A.col_start[head]; p < A.col_start[head + 1]; ++p) {


The same as above.

it was addressing a previous review
#1979 (comment)

Iroy30 · 2026-10-09T19:08:52Z

/ok to test 6562707

chris-maes · 2026-10-09T20:27:44Z

+    }
+    f_t const pivot_rhs = values[elimination.pivot_row];
+    for (std::size_t k = 0; k < elimination.affected_rows.size(); ++k) {
+      i_t const row = elimination.affected_rows[k];


i_t const row -> const i_t row

chris-maes · 2026-10-09T20:28:14Z

+  auto const& keep = xf.presolve_info.free_elimination_remaining_constraints;
+  std::vector<f_t> reduced(keep.size());
+  for (std::size_t k = 0; k < keep.size(); ++k) {
+    i_t const row = keep[k];


i_t const row -> const i_t row

chris-maes · 2026-10-09T20:31:19Z

+    // can_reuse_barrier_cache skips that comparison: the cache stores the expanded size.
+    const bool user_has_soc = op_problem.has_quadratic_constraints();
+    const bool reuse_from_cache =
+      settings.user_problem_file.empty() &&


Why do we care about user_problem_file here?

chris-maes · 2026-10-10T00:00:41Z

/merge

yuwenchen95 and others added 30 commits August 18, 2026 05:29

direct KKT matrix support

1038138

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

cache reuse for linear objective updates

e2572f7

cleanup

2c31258

Fix CUDA 13 settings_ pointer access in sparse_cholesky.

62d595b

The cache-reuse rebind left one destructor check as settings_. instead of settings_->, which only compiles on CU13 wheels. Signed-off-by: root <root@ipp1-3302.aselab.nvidia.com>

fix merge conflicts

13a5215

resolve conflicts

4270b5c

move rmm to gpu path

050783d

remove log files

6e20206

address reviews part 1

7d61502

remove logs

698afbe

address reviews part 2

5cf8c5f

copyright update

0ae8190

address reviews part 3

d1b4b73

restructure and rename barrier.cu solve functions

43b9b29

formatting checks fix

0499d81

fix comments

260d6a1

comment cleanup

faf0aa2

Merge origin/main into cuopt_cache_reuse_FSI

bf7d115

Keep cache-reuse symbolic_done_ and main's explicit CUstream initialization. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

Add GPU ruiz scaling

854b857

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

barrier: skip the A^T transpose on the ADAT path, the AD copy on the …

8711e6b

…augmented path, and the never-read device_A_x_values snapshot in iteration_data_t; no numerical change. Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

address reviews part 3

bf7f479

Restore the barrier sequence-update API design notes

d9a579b

Recovered from 7d61502; it was deleted by the 698afbe log cleanup. Updated for the update_rhs naming and the deferred setter cache-invalidation gap. Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

Merge branch 'main' of https://github.com/NVIDIA/cuopt into add_updat…

a1a936f

…e_apis Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com> # Conflicts: # cpp/src/barrier/barrier.cu # cpp/src/barrier/device_sparse_matrix.cuh

Point the update API notes at the new sequence_solve tests

2782e35

Signed-off-by: Ishika Roy <iroy@ipp1-3302.aselab.nvidia.com>

barrier: replace the merge-sort device CSC→CSR with a scatter plus se…

69191a6

…gmented sort, reuse it for a new on-device transpose so the augmented path drops the host A^T build and upload, and cover both with a unit test. Signed-off-by: yuwenchen95 <yuwchen@nvidia.com>

Merge branch 'main' into gpu-scaling

9ea7635

Signed-off-by: yuwenchen95 <yuwchen@nvidia.com> # Conflicts: # cpp/src/barrier/device_sparse_matrix.cuh

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/barrier_transform.hpp Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Comment thread cpp/src/barrier/second_order_cone_kernels.cuh Outdated

chris-maes reviewed Oct 9, 2026

View reviewed changes

Iroy30 added 2 commits October 9, 2026 02:06

address reviews Oct 8

cc9209b

Merge origin/main into main.

2e1a10e

Iroy30 added 2 commits October 9, 2026 04:03

address reviews

5fec35c

address format review and record cone head bounds in translate_soc

2f022ba

yuwenchen95 reviewed Oct 9, 2026

View reviewed changes

address reviews and style formatting

17173b9

yuwenchen95 approved these changes Oct 9, 2026

View reviewed changes

Iroy30 added 2 commits October 9, 2026 17:25

reuse for LP barrier

d6429df

make a pass on templatization

6562707

chris-maes reviewed Oct 9, 2026

View reviewed changes

rapids-bot Bot merged commit 57e6f0b into NVIDIA:main Oct 10, 2026
183 of 185 checks passed

Conversation

Iroy30 commented Sep 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Issue

Checklist

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Iroy30 commented Oct 9, 2026

Uh oh!

github-actions Bot commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Test Summary

Uh oh!

Iroy30 commented Oct 9, 2026

Uh oh!

yuwenchen95 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Iroy30 Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Iroy30 Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Iroy30 commented Oct 9, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chris-maes commented Oct 10, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Iroy30 commented Sep 22, 2026 •

edited

Loading

github-actions Bot commented Oct 9, 2026 •

edited

Loading

Iroy30 Oct 9, 2026 •

edited

Loading

Iroy30 Oct 9, 2026 •

edited

Loading