Skip to content

Add memory model support for host-resident problem data #819

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
tmckayus wants to merge 9 commits into
base: release/26.02
Choose a base branch
from
Open

Add memory model support for host-resident problem data #819

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
0fee7c2
Add flexible memory model with GPU/CPU problem class hierarchy
tmckayus Jan 31, 2026
270bad0
add solution class for returning results on gpu or cpu
tmckayus Feb 1, 2026
cc3a262
Add warmstart data support and refactor Python bindings
tmckayus Feb 1, 2026
44f16ea
Cleanup and defer CUDA init for CPU-only host remote execution support
tmckayus Feb 1, 2026
fc3e360
try workaround for distro size check on cuda 12.9 rocky
tmckayus Feb 1, 2026
715f3e5
fix transfer of warmstart data from Python API
tmckayus Feb 2, 2026
30fffe0
cleanup branching on problem type and backend type with polymorphism
tmckayus Feb 2, 2026
3bcdbaa
additional branching cleanup on solution type
tmckayus Feb 2, 2026
7bad723
add CPU memory mode for local testing (CUOPT_USE_CPU_MEM_FOR_LOCAL)
tmckayus Feb 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion ci/run_ctests.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#!/bin/bash
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

set -euo pipefail
Expand All @@ -26,3 +26,7 @@ for gt in "${GTEST_DIR}"/*_TEST; do
echo "Running gtest ${test_name}"
"${gt}" "$@"
done

# Run C_API_TEST with CPU memory for local solves (excluding time limit tests)
echo "Running gtest C_API_TEST with CUOPT_USE_CPU_MEM_FOR_LOCAL"
CUOPT_USE_CPU_MEM_FOR_LOCAL=1 "${GTEST_DIR}/C_API_TEST" --gtest_filter=-c_api/TimeLimitTestFixture.* "$@"
Comment on lines +30 to +32
Copy link

@coderabbitai coderabbitai bot Feb 2, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟠 Major

Guard the extra C_API_TEST run when the binary is absent.

The new unconditional invocation will fail in configurations that skip building C_API_TEST. Add an existence check to keep CI/dev flows resilient.

🛡️ Proposed guard 
-echo "Running gtest C_API_TEST with CUOPT_USE_CPU_MEM_FOR_LOCAL"
-CUOPT_USE_CPU_MEM_FOR_LOCAL=1 "${GTEST_DIR}/C_API_TEST" --gtest_filter=-c_api/TimeLimitTestFixture.* "$@"
+if [[ -x "${GTEST_DIR}/C_API_TEST" ]]; then
+  echo "Running gtest C_API_TEST with CUOPT_USE_CPU_MEM_FOR_LOCAL"
+  CUOPT_USE_CPU_MEM_FOR_LOCAL=1 "${GTEST_DIR}/C_API_TEST" --gtest_filter=-c_api/TimeLimitTestFixture.* "$@"
+else
+  echo "Skipping C_API_TEST (binary not found)"
+fi
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
# Run C_API_TEST with CPU memory for local solves (excluding time limit tests)
echo "Running gtest C_API_TEST with CUOPT_USE_CPU_MEM_FOR_LOCAL"
CUOPT_USE_CPU_MEM_FOR_LOCAL=1 "${GTEST_DIR}/C_API_TEST" --gtest_filter=-c_api/TimeLimitTestFixture.* "$@"
# Run C_API_TEST with CPU memory for local solves (excluding time limit tests)
if [[ -x "${GTEST_DIR}/C_API_TEST" ]]; then
echo "Running gtest C_API_TEST with CUOPT_USE_CPU_MEM_FOR_LOCAL"
CUOPT_USE_CPU_MEM_FOR_LOCAL=1 "${GTEST_DIR}/C_API_TEST" --gtest_filter=-c_api/TimeLimitTestFixture.* "$@"
else
echo "Skipping C_API_TEST (binary not found)"
fi
🤖 Prompt for AI Agents 
In `@ci/run_ctests.sh` around lines 30 - 32, The added unconditional invocation of
C_API_TEST using CUOPT_USE_CPU_MEM_FOR_LOCAL can fail when the C_API_TEST binary
isn't built; update the script to check for the executable (e.g., test -x
"${GTEST_DIR}/C_API_TEST" || skip) before setting CUOPT_USE_CPU_MEM_FOR_LOCAL
and running the test, and emit a clear message or silently skip if missing;
ensure you still pass through "$@" and preserve the existing --gtest_filter
argument when guarded.

63 changes: 43 additions & 20 deletions cpp/cuopt_cli.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,12 +1,14 @@
/* clang-format off */
/*
* SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*/
/* clang-format on */

#include <cuopt/linear_programming/mip/solver_settings.hpp>
#include <cuopt/linear_programming/optimization_problem.hpp>
#include <cuopt/linear_programming/optimization_problem_interface.hpp>
#include <cuopt/linear_programming/optimization_problem_utils.hpp>
#include <cuopt/linear_programming/solve.hpp>
#include <mps_parser/parser.hpp>
#include <utilities/logger.hpp>
Expand Down Expand Up @@ -89,7 +91,6 @@ int run_single_file(const std::string& file_path,
bool solve_relaxation,
const std::map<std::string, std::string>& settings_strings)
{
const raft::handle_t handle_{};
cuopt::linear_programming::solver_settings_t<int, double> settings;

try {
Expand Down Expand Up @@ -122,13 +123,31 @@ int run_single_file(const std::string& file_path,
return -1;
}

auto op_problem =
cuopt::linear_programming::mps_data_model_to_optimization_problem(&handle_, mps_data_model);
// Determine memory backend and create problem using interface
// Create handle only for GPU memory backend (avoid CUDA init on CPU-only hosts)
auto memory_backend = cuopt::linear_programming::get_memory_backend_type();
std::unique_ptr<raft::handle_t> handle_ptr;
std::unique_ptr<cuopt::linear_programming::optimization_problem_interface_t<int, double>>
problem_interface;

if (memory_backend == cuopt::linear_programming::memory_backend_t::GPU) {
handle_ptr = std::make_unique<raft::handle_t>();
problem_interface =
std::make_unique<cuopt::linear_programming::gpu_optimization_problem_t<int, double>>(
handle_ptr.get());
} else {
problem_interface =
std::make_unique<cuopt::linear_programming::cpu_optimization_problem_t<int, double>>(nullptr);
}

// Populate the problem from MPS data model
cuopt::linear_programming::populate_from_mps_data_model(problem_interface.get(), mps_data_model);

const bool is_mip =
(op_problem.get_problem_category() == cuopt::linear_programming::problem_category_t::MIP ||
op_problem.get_problem_category() == cuopt::linear_programming::problem_category_t::IP) &&
!solve_relaxation;
const bool is_mip = (problem_interface->get_problem_category() ==
cuopt::linear_programming::problem_category_t::MIP ||
problem_interface->get_problem_category() ==
cuopt::linear_programming::problem_category_t::IP) &&
!solve_relaxation;

try {
auto initial_solution =
Expand Down Expand Up @@ -157,10 +176,10 @@ int run_single_file(const std::string& file_path,
try {
if (is_mip) {
auto& mip_settings = settings.get_mip_settings();
auto solution = cuopt::linear_programming::solve_mip(op_problem, mip_settings);
auto solution = cuopt::linear_programming::solve_mip(problem_interface.get(), mip_settings);
} else {
auto& lp_settings = settings.get_pdlp_settings();
auto solution = cuopt::linear_programming::solve_lp(op_problem, lp_settings);
auto solution = cuopt::linear_programming::solve_lp(problem_interface.get(), lp_settings);
}
} catch (const std::exception& e) {
CUOPT_LOG_ERROR("Error: %s", e.what());
Expand Down Expand Up @@ -334,19 +353,23 @@ int main(int argc, char* argv[])
const auto initial_solution_file = program.get<std::string>("--initial-solution");
const auto solve_relaxation = program.get<bool>("--relaxation");

// All arguments are parsed as string, default values are parsed as int if unused.
const auto num_gpus = program.is_used("--num-gpus")
? std::stoi(program.get<std::string>("--num-gpus"))
: program.get<int>("--num-gpus");

// Only initialize CUDA resources if using GPU memory backend (not remote execution)
auto memory_backend = cuopt::linear_programming::get_memory_backend_type();
std::vector<std::shared_ptr<rmm::mr::device_memory_resource>> memory_resources;

for (int i = 0; i < std::min(raft::device_setter::get_device_count(), num_gpus); ++i) {
cudaSetDevice(i);
memory_resources.push_back(make_async());
rmm::mr::set_per_device_resource(rmm::cuda_device_id{i}, memory_resources.back().get());
if (memory_backend == cuopt::linear_programming::memory_backend_t::GPU) {
// All arguments are parsed as string, default values are parsed as int if unused.
const auto num_gpus = program.is_used("--num-gpus")
? std::stoi(program.get<std::string>("--num-gpus"))
: program.get<int>("--num-gpus");

for (int i = 0; i < std::min(raft::device_setter::get_device_count(), num_gpus); ++i) {
cudaSetDevice(i);
memory_resources.push_back(make_async());
rmm::mr::set_per_device_resource(rmm::cuda_device_id{i}, memory_resources.back().get());
}
cudaSetDevice(0);
}
cudaSetDevice(0);

return run_single_file(file_name, initial_solution_file, solve_relaxation, settings_strings);
}
Loading