cuPDLP-C-ROCm

中文主页: README.zh-CN.md
ROCm/gfx1150 quick start: README_ROCM_gfx1150.md
Documentation map: docs/README.md
Validation index: validation/README.md
Benchmark index: docs/benchmarks/README.md

cuPDLP-C-ROCm is a ROCm/HIP port and validation fork of upstream cuPDLP-C for AMD GPUs/APUs. The project keeps the CPU path and upstream-compatible CUDA path, and adds a ROCm/HIP backend for AMD Radeon-class hardware.

Item	Current value
Primary ROCm target	AMD Radeon 890M
ROCm architecture	gfx1150
ROCm version used in local validation	7.2.1
Planned larger AMD target	AMD Radeon PRO W7900 / gfx1100
CUDA baseline devices	RTX 3090, RTX 4090D, H100

Status: experimental but buildable. The ROCm/HIP backend has passed smoke validation, Netlib validation, cross-device benchmark checks, and large-MPS baseline testing on AMD Radeon 890M / gfx1150. It is not yet a production-ready or fully tuned ROCm solver release.

Start here

Need	English	中文
ROCm/gfx1150 quick start	README_ROCM_gfx1150.md	README_ROCM_gfx1150.zh-CN.md
Full documentation map	docs/README.md	docs/README.md
Validation data index	validation/README.md	validation/README.zh-CN.md
Benchmark index	docs/benchmarks/README.md	docs/benchmarks/README.md

Documentation

Topic	English	中文
Build / run / validation workflow	docs/ROCM_WORKFLOW.md	docs/ROCM_WORKFLOW.zh-CN.md
CPU vs ROCm validation semantics	docs/VALIDATION.md	docs/VALIDATION.zh-CN.md
Backend modes and naming policy	docs/BACKEND_MODES_AND_NAMING.md	docs/BACKEND_MODES_AND_NAMING.zh-CN.md
CUDA to ROCm migration case study	docs/CUDA_TO_ROCM_MIGRATION_CASE_STUDY.md	docs/CUDA_TO_ROCM_MIGRATION_CASE_STUDY.zh-CN.md
ROCm porting guide	docs/ROCM_PORTING_GUIDE.md	docs/ROCM_PORTING_GUIDE.zh-CN.md
ROCm profiling notes	docs/ROCM_PROFILING_NOTES.md	docs/ROCM_PROFILING_NOTES.zh-CN.md
ROCm tuning history	docs/ROCM_TUNING_HISTORY.md	docs/ROCM_TUNING_HISTORY.zh-CN.md
ROCm tuning guide	docs/TUNING_GUIDE_ROCM.md	docs/TUNING_GUIDE_ROCM.zh-CN.md
Cross-device Netlib benchmarks	docs/CROSS_DEVICE_BENCHMARKS.md	docs/CROSS_DEVICE_BENCHMARKS.zh-CN.md
Large MPS benchmark plan	docs/LARGE_MPS_BENCHMARK_PLAN.md	docs/LARGE_MPS_BENCHMARK_PLAN.zh-CN.md
greenbea numerical behavior	docs/NUMERICAL_BEHAVIOR_GREENBEA.md	docs/NUMERICAL_BEHAVIOR_GREENBEA.zh-CN.md
Upstream reference snapshot	README_UPSTREAM.md	—

README_UPSTREAM.md is intentionally kept as an upstream reference snapshot and is not translated or rewritten as project documentation.

Benchmarks

The raw .mps benchmark files are not committed. Curated result CSVs and explanation documents are committed instead.

Topic	English	中文	Raw CSV
Large MPS CUDA/ROCm baseline	summary	中文版	platform summary, per-case timing
cuPDLPx vs cuPDLP-C short13	comparison	中文版	comparison CSV

What this repository provides

• CPU-only cuPDLP-C build path.
• Upstream-compatible CUDA build path for NVIDIA baselines.
• ROCm/HIP backend built from migrated CUDA backend code.
• plc executable linked against the ROCm/HIP backend.
• CPU-vs-ROCm smoke validation scripts.
• Extended Netlib validation cases.
• Cross-device benchmark workflows and summaries for RTX 3090, RTX 4090D, H100, and Radeon 890M.
• Large MPS benchmark documents and curated CSV summaries.
• rocprofv3 profiling workflow and ROCm tuning notes.
• Migration documentation for CUDA-to-ROCm/HIP scientific-computing projects.

Backend modes

Mode	CMake options	Role
CPU	BUILD_CUDA=OFF, BUILD_ROCM=OFF	Correctness and portability baseline
CUDA	BUILD_CUDA=ON, BUILD_ROCM=OFF	Upstream-compatible NVIDIA backend and benchmark baseline
ROCm/HIP	BUILD_CUDA=OFF, BUILD_ROCM=ON	AMD Radeon ROCm/HIP target backend

BUILD_CUDA and BUILD_ROCM must not be enabled at the same time. Use separate build directories such as build-cpu, build-cuda, and build-rocm-plc.

Current validation and benchmark status

Large MPS baseline status:

Platform	Backend	Result
RTX 3090	CUDA upstream	25/26 OPTIMAL, 1/26 TIMELIMIT
Radeon 890M	ROCm/HIP baseline	24/26 OPTIMAL, 2/26 TIMELIMIT
RTX 4090D	CUDA upstream	26/26 OPTIMAL
H100	CUDA upstream	26/26 OPTIMAL

cuPDLPx short13 comparison status:

Solver	Platform	Result
cuPDLP-C upstream	RTX 4090D CUDA	13/13 OPTIMAL on selected short/medium cases
cuPDLPx v0.2.9	RTX 4090D CUDA	13/13 OPTIMAL on the same selected cases

Quick start: ROCm/HIP build

cmake -S . -B build-rocm-plc -G Ninja \
  -DCMAKE_BUILD_TYPE=Release \
  -DBUILD_CUDA=OFF \
  -DBUILD_ROCM=ON \
  -DBUILD_APPS=OFF \
  -DBUILD_PYTHON=OFF \
  -DBUILD_TESTING=ON \
  -DCMAKE_PREFIX_PATH=/opt/rocm \
  -DCMAKE_HIP_ARCHITECTURES=gfx1150

cmake --build build-rocm-plc --target plc -j"$(nproc)"

Run a smoke example:

./build-rocm-plc/bin/plc \
  -fname ./example/afiro.mps \
  -out /tmp/afiro_rocm_sum.json \
  -nIterLim 200

Validation

./scripts/check_rocm_port.sh
ctest --test-dir build-rocm-plc --output-on-failure

Extended validation:

RESULT_ROOT=validation/results/extended_netlib \
  ./scripts/run_validation.sh validation/cases_extended_netlib.txt

See validation/README.md for curated validation summaries and CSV files.

Profiling and tuning

RESULT_ROOT=profiling/results/current ./scripts/profile_rocm_smoke.sh

python3 scripts/summarize_rocm_profile.py \
  --input profiling/results/current \
  --output profiling/results/current/profile_summary.md

See docs/ROCM_PROFILING_NOTES.md, docs/ROCM_TUNING_HISTORY.md, and docs/TUNING_GUIDE_ROCM.md.

Adapting to another ROCm GPU

Identify the GPU architecture:

rocminfo | grep -E "Name:|Marketing Name|gfx"
rocm_agent_enumerator

Then set the matching architecture, for example:

-DCMAKE_HIP_ARCHITECTURES=gfx1100

for AMD Radeon PRO W7900, depending on ROCm support.