Domain-Specific languages (DSLs) such as Triton enable developers to write high-performance GPU kernels in a Python-friendly manner; however, profiling these kernels with existing tools often incurs runtime and storage overhead while failing to deliver actionable insights for both kernel authors and framework developers. We present Proton, a multi-level, adaptive profiler tailored for the Triton programming language and compiler. Proton provides frontend APIs to selectively profile relevant regions, aggregate results, capture custom metrics not available through hardware counters, and query profiles using a SQL-like language. Proton’s backend design unifies vendor profiling APIs with instrumentation-based profiling, ensuring portability and extensibility. Using Proton, users are able to query custom and hardware metrics across the relevant levels of abstraction—full end-to-end model execution, isolated neural network layers, language-specific Triton operators, and compiler intermediate representations. We demonstrate the tool’s effectiveness through case studies on production-grade kernel development, continuous integration, multi-GPU analysis, language model inference, and intra-kernel profiling. Our evaluations on end-to-end workloads, as well as standalone Triton kernels, demonstrate that Proton imposes lower runtime overhead and delivers significant reductions in profile sizes relative to existing framework and vendor profilers while being fully open source.

In this paper, we propose Mercury, a multi-GPU operator compiler based on a loop-based intermediate representation, CommIR. At the core of Mercury is an abstraction that treats remote GPU memory as an explicitly managed extension of the memory hierarchy, expanding the available storage and communication resources beyond local HBM. This unified view enables the compiler to reason holistically about data placement and inter-device communication, unlocking a vastly larger design space that encompasses and extends beyond existing manual strategies. As a result, Mercury is able to automatically reproduce the performance of hand-optimized baselines like RingAttention and Ulysses, and in some configurations, even discovers more effective strategies that manual designs have overlooked. Our implementation is open-sourced at https://github.com/ChandlerGuan/mercury_artifact.

Graph neural network (GNN) is a promising emerging application for link prediction, recommendation, etc. Existing hardware innovation is limited to single-machine GNN (SM-GNN), however, the enterprises usually adopt huge graph with large-scale distributed GNN (LSD-GNN) that has to be carried out with distributed in-memory storage. The LSD-GNN is very different from SM-GNN in terms of system architecture demand, workflow and operators, and hence characterizations.

The data science revolution and growing popularity of data lakes make efficient processing of raw data increasingly important. To address this, we propose the ACCelerated Operators for Raw Data Analysis (ACCORDA) architecture. By extending the operator interface (subtype with encoding) and employing a uniform runtime worker model, ACCORDA integrates data transformation acceleration seamlessly, enabling a new class of encoding optimizations and robust high-performance raw data processing. Together, these key features preserve the software system architecture, empowering state-of-art heuristic optimizations to drive flexible data encoding for performance. ACCORDA derives performance from its software architecture, but depends critically on the acceleration of the Unstructured Data Processor (UDP) that is integrated into the memory-hierarchy, and accelerates data transformation tasks by 16x-21x (parsing, decompression) to as much as 160x (deserialization) compared to an x86 core. We evaluate ACCORDA using TPC-H queries on tabular data formats, exercising raw data properties such as parsing and data conversion. The ACCORDA system achieves 2.9x-13.2x speedups when compared to SparkSQL, reducing raw data processing overhead to a geomean of 1.2x (20%). In doing so, ACCORDA robustly matches or outperforms prior systems that depend on caching loaded data, while computing on raw, unloaded data. This performance benefit is robust across format complexity, query predicates, and selectivity (data statistics). ACCORDA's encoding-extended operator interface unlocks aggressive encoding-oriented optimizations that deliver 80% average performance increase over the 7 affected TPC-H queries.

Big data analytic applications give rise to large-scale extract-transform-load (ETL) as a fundamental step to transform new data into a native representation. ETL workloads pose significant performance challenges on conventional architectures, so we propose the design of the unstructured data processor (UDP), a software programmable accelerator that includes multi-way dispatch, variable-size symbol support, Flexible-source dispatch (stream buffer and scalar registers), and memory addressing to accelerate ETL kernels both for current and novel future encoding and compression. Specifically, UDP excels at branch-intensive and symbol and pattern-oriented workloads, and can offload them from CPUs.

We propose the Unified Automata Processor (UAP), a new architecture that provides general and efficient support for finite automata (FA). The UAP supports a wide range of existing finite automata models (DFAs, NFAs, A-DFAs, JFAs, counting-DFAs, and counting-NFAs), and future novel FA models. Evaluation on realistic workloads shows that UAP implements all models efficiently, achieving line rates 94.5% of ideal. A single UAP lane delivers line rates 50x greater than software approaches on CPUs and GPUs. Scaling UAP to 64 lanes achieves FA transition throughputs as high as 295 Gbps, more than 100x higher than CPUs and GPUs, and even exceeding ASIC approaches such as IBM's RegX by 6x. With efficient support for multiple input streams, UAP achieves throughputs that saturate even high speed stacked DRAM memory systems.