Forward-merge release/26.04 into main

ci/test_self_hosted_service.sh

@@ -14,8 +14,18 @@ CUOPT_SERVER_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cuopt_server_${RAPIDS_PY_CUDA_SU
 CUOPT_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cuopt_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github python)
 LIBCUOPT_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcuopt_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 
-# echo to expand wildcard before adding `[extra]` requires for pip
+# generate constraints (possibly pinning to oldest support versions of dependencies)
+rapids-generate-pip-constraints test_python "${PIP_CONSTRAINT}"
+
+# notes:
+#
+#   * echo to expand wildcard before adding `[test]` requires for pip
+#   * just providing --constraint="${PIP_CONSTRAINT}" to be explicit, and because
+#     that environment variable is ignored if any other --constraint are passed via the CLI
+#
 rapids-pip-retry install \
+    --prefer-binary \
+    --constraint "${PIP_CONSTRAINT}" \
     "${CUOPT_MPS_PARSER_WHEELHOUSE}"/cuopt_mps_parser*.whl \
     "${CUOPT_WHEELHOUSE}"/cuopt*.whl \
     "${LIBCUOPT_WHEELHOUSE}"/libcuopt*.whl \

ci/test_wheel_cuopt.sh

@@ -25,8 +25,18 @@ cuopt-sh-client @ file://$(echo ${CUOPT_SH_CLIENT_WHEELHOUSE}/cuopt_sh_client-*.
 libcuopt-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo ${LIBCUOPT_WHEELHOUSE}/libcuopt_${RAPIDS_PY_CUDA_SUFFIX}-*.whl)
 EOF
 
-# echo to expand wildcard before adding `[extra]` requires for pip
+# generate constraints (possibly pinning to oldest support versions of dependencies)
+rapids-generate-pip-constraints test_python "${PIP_CONSTRAINT}"
+
+# notes:
+#
+#   * echo to expand wildcard before adding `[test]` requires for pip
+#   * just providing --constraint="${PIP_CONSTRAINT}" to be explicit, and because
+#     that environment variable is ignored if any other --constraint are passed via the CLI
+#
 rapids-pip-retry install \
+    --prefer-binary \
+    --constraint "${PIP_CONSTRAINT}" \
     --constraint "${PIP_CONSTRAINT}" \
     "${CUOPT_MPS_PARSER_WHEELHOUSE}"/cuopt_mps_parser*.whl \
     "$(echo "${CUOPT_WHEELHOUSE}"/cuopt*.whl)[test]" \
@@ -67,4 +77,6 @@ timeout 30m ./ci/run_cuopt_pytests.sh --verbose --capture=no
 if [[ "${RAPIDS_BUILD_TYPE}" == "nightly" ]]; then
     ./ci/thirdparty-testing/run_jump_tests.sh
     ./ci/thirdparty-testing/run_cvxpy_tests.sh
+    ./ci/thirdparty-testing/run_pulp_tests.sh
+    ./ci/thirdparty-testing/run_pyomo_tests.sh
 fi

ci/test_wheel_cuopt_server.sh

@@ -15,8 +15,18 @@ CUOPT_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cuopt_${RAPIDS_PY_CUDA_SUFFIX}" rapids-
 CUOPT_SH_CLIENT_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cuopt_sh_client" RAPIDS_PY_WHEEL_PURE="1" rapids-download-wheels-from-github python)
 LIBCUOPT_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcuopt_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 
-# echo to expand wildcard before adding `[extra]` requires for pip
+# generate constraints (possibly pinning to oldest support versions of dependencies)
+rapids-generate-pip-constraints test_python "${PIP_CONSTRAINT}"
+
+# notes:
+#
+#   * echo to expand wildcard before adding `[test]` requires for pip
+#   * just providing --constraint="${PIP_CONSTRAINT}" to be explicit, and because
+#     that environment variable is ignored if any other --constraint are passed via the CLI
+#
 rapids-pip-retry install \
+    --prefer-binary \
+    --constraint "${PIP_CONSTRAINT}" \
     "${CUOPT_MPS_PARSER_WHEELHOUSE}"/cuopt_mps_parser*.whl \
     "$(echo "${CUOPT_SERVER_WHEELHOUSE}"/cuopt_server*.whl)[test]" \
     "${CUOPT_WHEELHOUSE}"/cuopt*.whl \

ci/thirdparty-testing/run_pulp_tests.sh

@@ -0,0 +1,43 @@
+#!/bin/bash
+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+set -e -u -o pipefail
+
+rapids-logger "building 'pulp' from source and running cuOpt tests"
+
+if [ -z "${PIP_CONSTRAINT:-}" ]; then
+    rapids-logger "PIP_CONSTRAINT is not set; ensure ci/test_wheel_cuopt.sh (or equivalent) has set it so cuopt wheels are used."
+    exit 1
+fi
+
+git clone --depth 1 https://github.com/coin-or/pulp.git
+pushd ./pulp || exit 1
+
+# Install PuLP in editable form so it uses the environment's cuopt (from PIP_CONSTRAINT)
+python -m pip install \
+    --constraint "${PIP_CONSTRAINT}" \
+    --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple \
+    pytest \
+    -e .
+
+pip check
+
+rapids-logger "running PuLP tests (cuOpt-related)"
+# PuLP uses pytest; run only tests that reference cuopt/CUOPT
+# Exit code 5 = no tests collected; then try run_tests.py which detects solvers (including cuopt)
+pytest_rc=0
+timeout 5m python -m pytest \
+    --verbose \
+    --capture=no \
+    -k "cuopt or CUOPT" \
+    pulp/tests/ || pytest_rc=$?
+
+if [ "$pytest_rc" -eq 5 ]; then
+    rapids-logger "No pytest -k cuopt tests found; running PuLP run_tests.py (solver auto-detection, includes cuopt)"
+    timeout 5m python pulp/tests/run_tests.py
+    pytest_rc=$?
+fi
+
+popd || exit 1
+exit "$pytest_rc"

ci/thirdparty-testing/run_pyomo_tests.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+set -e -u -o pipefail
+
+rapids-logger "building 'pyomo' from source and running cuOpt tests"
+
+if [ -z "${PIP_CONSTRAINT:-}" ]; then
+    rapids-logger "PIP_CONSTRAINT is not set; ensure ci/test_wheel_cuopt.sh (or equivalent) has set it so cuopt wheels are used."
+    exit 1
+fi
+
+git clone --depth 1 https://github.com/Pyomo/pyomo.git
+pushd ./pyomo || exit 1
+
+# Install Pyomo in editable form so it uses the environment's cuopt (from PIP_CONSTRAINT)
+python -m pip install \
+    --constraint "${PIP_CONSTRAINT}" \
+    --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple \
+    pytest \
+    -e .
+
+pip check
+
+rapids-logger "running Pyomo tests (cuopt_direct / cuOpt-related)"
+# Run only tests that reference cuopt (cuopt_direct solver)
+timeout 5m python -m pytest \
+    --verbose \
+    --capture=no \
+    -k "cuopt or CUOPT" \
+    pyomo/solvers/tests/
+
+popd || exit 1

cpp/include/cuopt/linear_programming/constants.h

@@ -74,6 +74,7 @@
 #define CUOPT_NUM_CPU_THREADS                 "num_cpu_threads"
 #define CUOPT_NUM_GPUS                        "num_gpus"
 #define CUOPT_USER_PROBLEM_FILE               "user_problem_file"
+#define CUOPT_PRESOLVE_FILE                   "presolve_file"
 #define CUOPT_RANDOM_SEED                     "random_seed"
 #define CUOPT_PDLP_PRECISION                  "pdlp_precision"
 

cpp/include/cuopt/linear_programming/mip/solver_settings.hpp

@@ -105,6 +105,7 @@ class mip_solver_settings_t {
   std::string log_file;
   std::string sol_file;
   std::string user_problem_file;
+  std::string presolve_file;
 
   /** Initial primal solutions */
   std::vector<std::shared_ptr<rmm::device_uvector<f_t>>> initial_solutions;

cpp/include/cuopt/linear_programming/pdlp/solver_settings.hpp

@@ -245,6 +245,7 @@ class pdlp_solver_settings_t {
   std::string log_file{""};
   std::string sol_file{""};
   std::string user_problem_file{""};
+  std::string presolve_file{""};
   bool per_constraint_residual{false};
   bool crossover{false};
   bool cudss_deterministic{false};

cpp/src/math_optimization/solver_settings.cu

@@ -131,7 +131,9 @@ solver_settings_t<i_t, f_t>::solver_settings_t() : pdlp_settings(), mip_settings
     {CUOPT_SOLUTION_FILE,  &mip_settings.sol_file, ""},
     {CUOPT_SOLUTION_FILE,  &pdlp_settings.sol_file, ""},
     {CUOPT_USER_PROBLEM_FILE, &mip_settings.user_problem_file, ""},
-    {CUOPT_USER_PROBLEM_FILE, &pdlp_settings.user_problem_file, ""}
+    {CUOPT_USER_PROBLEM_FILE, &pdlp_settings.user_problem_file, ""},
+    {CUOPT_PRESOLVE_FILE, &mip_settings.presolve_file, ""},
+    {CUOPT_PRESOLVE_FILE, &pdlp_settings.presolve_file, ""}
   };
   // clang-format on
 }

cpp/src/mip_heuristics/solve.cu

@@ -283,6 +283,7 @@ mip_solution_t<i_t, f_t> solve_mip(optimization_problem_t<i_t, f_t>& op_problem,
                                      settings.tolerances.relative_tolerance,
                                      presolve_time_limit,
                                      settings.num_cpu_threads);
+
       if (!result.has_value()) {
         return mip_solution_t<i_t, f_t>(mip_termination_status_t::Infeasible,
                                         solver_stats_t<i_t, f_t>{},
@@ -306,6 +307,10 @@ mip_solution_t<i_t, f_t> solve_mip(optimization_problem_t<i_t, f_t>& op_problem,
       CUOPT_LOG_INFO("Writing user problem to file: %s", settings.user_problem_file.c_str());
       op_problem.write_to_mps(settings.user_problem_file);
     }
+    if (run_presolve && settings.presolve_file != "") {
+      CUOPT_LOG_INFO("Writing presolved problem to file: %s", settings.presolve_file.c_str());
+      presolve_result->reduced_problem.write_to_mps(settings.presolve_file);
+    }
 
     auto sol = run_mip(problem, settings, timer);
 

cpp/src/pdlp/solve.cu

@@ -1453,6 +1453,10 @@ optimization_problem_solution_t<i_t, f_t> solve_lp(
       CUOPT_LOG_INFO("Writing user problem to file: %s", settings.user_problem_file.c_str());
       op_problem.write_to_mps(settings.user_problem_file);
     }
+    if (run_presolve && settings.presolve_file != "") {
+      CUOPT_LOG_INFO("Writing presolved problem to file: %s", settings.presolve_file.c_str());
+      result->reduced_problem.write_to_mps(settings.presolve_file);
+    }
 
     // Set the hyper-parameters based on the solver_settings
     if (use_pdlp_solver_mode) { set_pdlp_solver_mode(settings); }

dependencies.yaml

@@ -633,22 +633,16 @@ dependencies:
           - matrix:
             packages:
               - *pylibraft_unsuffixed
+  # 'cuda_version' intentionally does not contain fallback entries... we want
+  # a loud error if an unsupported 'cuda' value is passed
   cuda_version:
     specific:
       - output_types: conda
         matrices:
-          - matrix:
-              cuda: "12.0"
-            packages:
-              - cuda-version=12.0
           - matrix:
               cuda: "12.2"
             packages:
               - cuda-version=12.2
-          - matrix:
-              cuda: "12.4"
-            packages:
-              - cuda-version=12.4
           - matrix:
               cuda: "12.5"
             packages:
@@ -669,6 +663,51 @@ dependencies:
               cuda: "13.1"
             packages:
               - cuda-version=13.1
+      - output_types: requirements
+        matrices:
+          # if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
+          # (e.g. for DLFW and pip devcontainers)
+          - matrix:
+              use_cuda_wheels: "false"
+            packages:
+          - matrix:
+              arch: aarch64
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              # some components (like nvidia-cublas-cu12 and nvidia-cuda-nvcc-cu12) didn't have
+              # aarch64 wheels until CTK 12.3, so allow a slightly looser bound here
+              - cuda-toolkit>=12.2,<12.4
+          - matrix:
+              cuda: "12.2"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.2.*
+          - matrix:
+              cuda: "12.5"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.5.*
+          - matrix:
+              cuda: "12.8"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.8.*
+          - matrix:
+              cuda: "12.9"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==12.9.*
+          - matrix:
+              cuda: "13.0"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.0.*
+          - matrix:
+              cuda: "13.1"
+              use_cuda_wheels: "true"
+            packages:
+              - cuda-toolkit==13.1.*
   cuda:
     common:
       - output_types: [conda]
@@ -682,37 +721,40 @@ dependencies:
 
   cuda_wheels:
     specific:
+      # cuOpt needs 'nvJitLink>={whatever-cuopt-was-built-against}' at runtime, and mixing
+      # old-CTK with new-nvJitLink is supported, so maintain 1 entry here per CUDA major.minor
+      # cuOpt is built against.
+      #
+      # nvJitLink has historically roughly followed the CTK's minor versioning scheme,
+      # but check at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
       - output_types: [requirements, pyproject]
         matrices:
+          # if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
+          # (e.g. for DLFW and pip devcontainers)
+          - matrix:
+              use_cuda_wheels: "false"
+            packages:
           - matrix:
               cuda: "12.*"
               use_cuda_wheels: "true"
             packages:
-              - cuda-toolkit[cublas,cudart,curand,cusolver,cusparse,nvjitlink,nvtx]==12.*
+              - cuda-toolkit[cublas,cudart,curand,cusolver,cusparse,nvtx]==12.*
               - nvidia-cudss-cu12
+              - nvidia-nvjitlink-cu12>=12.9,<13
           - matrix:
               cuda: "13.*"
               use_cuda_wheels: "true"
             packages:
-              - cuda-toolkit[cublas,cudart,curand,cusolver,cusparse,nvjitlink,nvtx]==13.*
-              - nvidia-cudss-cu13
-          # if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
-          # (e.g. for DLFW and pip devcontainers)
-          - matrix:
-              use_cuda_wheels: "false"
-            packages: []
-          # if no matching matrix selectors passed, list the unsuffixed packages
+              - &ctk_cu13 cuda-toolkit[cublas,cudart,curand,cusolver,cusparse,nvtx]==13.*
+              - &cudss_cu13 nvidia-cudss-cu13
+              - &nvjitlink_cu13 nvidia-nvjitlink>=13.0,<14
+          # if no matching matrix selectors passed, list the CUDA 13 requirement
           # (just as a source of documentation, as this populates pyproject.toml in source control)
           - matrix:
             packages:
-              - nvidia-cudart
-              - nvidia-cublas
-              - nvidia-cudss
-              - nvidia-curand
-              - nvidia-cusparse
-              - nvidia-nvjitlink
-              - nvidia-cusolver
-              - nvidia-nvtx
+              - *ctk_cu13
+              - *cudss_cu13
+              - *nvjitlink_cu13
   develop:
     common:
       - output_types: [conda, requirements]

docs/cuopt/source/conf.py

@@ -94,6 +94,11 @@
 nbsphinx_execute = "never"
 ipython_mplbackend = "str"
 
+# GPU routing example: Sphinx execution can fail when CUDA/CuPy don't match the docs
+# environment. Listed paths are skipped by myst-nb; this notebook is rendered from
+# checked-in cell outputs.
+nb_execution_excludepatterns = ["cuopt-python/routing/routing-example.ipynb"]
+
 # Add any files to exclude from the build
 exclude_patterns = ["hidden"]
 

python/libcuopt/pyproject.toml

@@ -30,16 +30,11 @@ classifiers = [
     "Programming Language :: Python :: 3.14",
 ]
 dependencies = [
+    "cuda-toolkit[cublas,cudart,curand,cusolver,cusparse,nvtx]==13.*",
     "cuopt-mps-parser==26.4.*,>=0.0.0a0",
     "librmm==26.4.*,>=0.0.0a0",
-    "nvidia-cublas",
-    "nvidia-cudart",
-    "nvidia-cudss",
-    "nvidia-curand",
-    "nvidia-cusolver",
-    "nvidia-cusparse",
-    "nvidia-nvjitlink",
-    "nvidia-nvtx",
+    "nvidia-cudss-cu13",
+    "nvidia-nvjitlink>=13.0,<14",
     "rapids-logger==0.2.*,>=0.0.0a0",
 ] # This list was generated by `rapids-dependency-file-generator`. To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.