HPC++ brings sequential C++ to OpenCL GPUs and CPUs
An LLVM-based framework converts sequential C++ programs into parallel implementations for multicore CPUs and OpenCL-compatible GPUs. Learn how to evaluate the approach on heterogeneous systems.
HPC++ is an LLVM-based framework that transforms sequential C++ programs into parallel implementations for multicore CPUs and OpenCL-compatible GPUs. The approach may interest engineers evaluating automatic parallelization on heterogeneous systems.
To test it, choose a representative program and record a sequential run as a baseline. Check that the output remains correct after transformation; performance does not replace functional validation.
Compare the sequential version with parallel implementations under equivalent conditions. Measure runtimes on a multicore CPU and a compatible OpenCL GPU, also noting data-transfer and setup costs where applicable.
Record hardware, configuration, inputs, and results so the comparison can be reproduced. If you use AI to study the code or interpret measurements, avoid submitting personal data, confidential code, or secrets; apply your organization’s policy and review what you share.