Add AMD GPU support via ROCm/HIP

[jeffdaily]

This adds optional AMD GPU support to cuPDLPx through ROCm/HIP, alongside the existing CUDA path. The CUDA build is unchanged when the new option is off.

What changed

The port keeps every device source in its existing CUDA spelling and routes the CUDA APIs to their HIP equivalents through a single compatibility header, internal/cuda_to_hip.h. On a HIP build that header maps the CUDA runtime, cuBLAS, cuSPARSE, and CUB symbols to hipRT, hipBLAS, hipSPARSE, and hipCUB; on a CUDA build it includes the standard CUDA headers, so nothing about the NVIDIA path changes. internal/cusparse_compat.h selects the standard hipsparseSpMV path on ROCm, since hipSPARSE does not provide the cusparseSpMVOp variant.

The build system gains a USE_HIP option (off by default). When enabled, the project is configured with the HIP language, the .cu sources are compiled as HIP, and the targets link hipBLAS, hipSPARSE, and hipCUB instead of the CUDA libraries. GPU architectures are chosen with CMAKE_HIP_ARCHITECTURES (defaulting to gfx90a). On Windows, the CLI-only mps_parser.c is excluded from the core library because it relies on strtok_r.

The interface test gains a case that runs the GPU solver path with presolve disabled, so the hipBLAS/hipSPARSE execution path is exercised end to end. Enabling the test suite (-DCUPDLPX_BUILD_TESTS=ON) to run the new case surfaced four stale zero_tolerance assignments in test_interface.c referencing a matrix_desc_t field that no longer exists; they are removed so the file compiles.

Building for AMD GPUs

cmake -B build -DUSE_HIP=ON -DCMAKE_HIP_ARCHITECTURES=gfx90a -DCMAKE_PREFIX_PATH=/opt/rocm
cmake --build build --clean-first

Set CMAKE_HIP_ARCHITECTURES to match the target GPU (for example gfx90a for MI200, gfx1100 for RDNA3 desktop, or gfx1201 for RDNA4). If the ROCm install is not on CMake's default search path, point -DCMAKE_PREFIX_PATH at it (e.g. /opt/rocm) so find_package can locate hip, hipBLAS, hipSPARSE, and hipCUB. The resulting ./build/cupdlpx binary is invoked exactly as in the CUDA build. The README documents this alongside the CUDA build instructions.

Validation

Built and run on an AMD Instinct MI200 (gfx90a) with ROCm 7.2.1: the full interface test suite passes, including the GPU solver case that exercises the hipBLAS and hipSPARSE paths (Status: OPTIMAL). The same configuration builds cleanly for gfx1100 (RDNA3) and gfx1201 (RDNA4). The CUDA build path is unaffected by these changes.

This work was prepared with assistance from Claude, an AI assistant by Anthropic.

[ZedongPeng]

Thanks a lot for adding ROCm/HIP support, @jeffdaily. This is a genuinely useful contribution, and I really appreciate the clean implementation. I left a few comments, mostly minor ones related to project consistency.

[ZedongPeng]

Thanks again for this, @jeffdaily.

One more thing before we merge: could you add a CI job that builds the HIP/ROCm path? Our workflows currently only exercise the CUDA build, so the HIP path can regress silently whenever someone touches the shared headers or CMake. A compile-only job (-DUSE_HIP=ON, configure + build, no GPU execution) would be enough to guard against that — the GitHub-hosted runners don't have AMD GPUs, so it would mirror how the CUDA jobs build-check without running on device.