Robert K Samuel, Rupesh NasreData sizes are growing rapidly, making XML processing increasingly slow at large scales. Although XML remains a popular structured data format, existing tools struggle with long parsing and query times. We introduce a parallel XML parser and query engine that can handle complex XPath patterns, including regular expressions and Kleene closures. Our approach overcomes the sequential and memory challenges of traditional XML processing. Experiments show significant performance improvements over libXML, reaching over 32× speedup.
[HiPC 2025] Accepted, Distinguished Paper Nominee 🏆
Robert K Samuel, Rupesh NasreDistributed heterogeneous systems that combine multi-core CPUs, GPUs, and multi-node clusters are now common, but coordinating computation across such diverse hardware remains challenging. A key issue is the absence of a unified intermediate representation that can model both distributed and heterogeneous execution. We present Distributed Heterogeneous IR (DHIR), an MLIR-based IR designed to express task parallelism, device placement, communication, and scheduling in a single abstraction. DHIR enables compilers to optimize distribution while preserving programmability. Experiments on 15 PolyBench/C kernels show that DHIR delivers performance comparable to expert-tuned MPI and OpenMP implementations.
Under Review