An IR for Distributed Heterogeneous ComputingComputer Systems Workshop 2025This work introduces DHIR, an MLIR-based intermediate representation designed to unify distributed and heterogeneous execution. DHIR enables task parallelism, device placement, and communication within a single abstraction, improving programmability and compiler optimization across CPUs, GPUs, and clusters.
An IR for Distributed Heterogeneous ComputingPLDI SRC 2025This work introduces DHIR, an MLIR-based intermediate representation designed to unify distributed and heterogeneous execution. DHIR enables task parallelism, device placement, and communication within a single abstraction, improving programmability and compiler optimization across CPUs, GPUs, and clusters.