Sparse 2024
as the organisers put it
The Sparse workshop (formerly CTSTA) aims to bring together researchers interested in compiler techniques, programming abstractions, and hardware for sparse computing including sparse tensor algebra, relational algebra, and graph processing applications. Due to the large number of applications, optimization techniques, types of data structures, and specialized hardware, there is a need for automation. In recent years, there has been a lot of interest in compiler techniques to automatically generate sparse computing code. This workshop aims to bring together leading researchers from academi ...
who it's for
compiler engineers, programming language designers, systems architects, machine learning systems engineers, database engineers, hardware architects, high performance computing engineers
topics
sparse tensor algebra compilers; sparse run-length encoding formats; continuous tensor abstraction; mixture-of-experts offloading; equality saturation e-graphs; free join join planning; incremental view maintenance; array abstraction beyond sparsity; mapping sparse code to accelerators; xdsl Python MLIR reimplementation; PDE weak-form code generation; runtime sparse-format switching and fusion
| title | author | length |
|---|---|---|
| Four Languages for Portability | Fredrik Kjolstad | 17m |
Argues that four domain-agnostic collection languages — relations, graphs, tensors, space — map to a single set-iteration IR that composes, fuses and tiles… | ||
| PDEs and Stencils | Paul H J Kelly | 24m |
Claim: sparsity problems are really locality problems because metadata size becomes proportional to data size. Tools auto-generate code from the PDE weak form… | ||
| Continuous Arrays | Jaeyeon Won | 19m |
Proposes a Continuous Tensor abstraction with real-valued indices and continuous for loops that assume tensors are piecewise-constant (finite axis-aligned… | ||
| Offloading-Efficient Sparse AI Systems | Luo Mai | 19m |
Built a system called Infinity that uses request-level tracing to predict and prefetch only the experts actually used during inference, showing that four to… | ||
| Panel: Compilation Frameworks | Max Willsey, Avery Laird, Tobias Grosser, Willow Ahrens | 19m |
Discussion shows a concrete path for quantify over bitwidth proofs: Lean supports quantifying bitwidth but only offers high‑performance automation for fixed… | ||
| Compiling Control Flow in Sparse and Structured Array Programs | Willow Ahrens | 20m |
Finch is a new language and compiler that merges control flow and data-format descriptions so the compiler specializes loops, ifs, breaks, masks and… | ||
| Design DSLs with xDSL | Tobias Grosser | 22m |
xDSL is a Python reimplementation of MLIR that prioritizes approachability and textual IR interoperability, letting teams define ops and types in minutes and… | ||
| Equality Saturation and Joins | Max Willsey | 20m |
Equality saturation uses egraphs to represent many equivalent programs and extraction picks the cheapest term after saturating rewrites. A relational encoding… | ||
| Panel: Databases and Hardware | Amir Shaikhha, Olivia Hsu, Haozhe Zhang, Matthias Boehm | 18m |
Panel reports a measured ~1% utilization on sparse arrays when running real sparse kernels, describes an Onyx tape-out and a working Sam prototype, and shows… | ||
| Mapping Sparse Applications to Accelerated Computing Systems | Olivia Hsu | 24m |
Describes a compiler stack that matches sparse tensor algebra to accelerators: Mosaic external function interface finds and offloads matching subexpressions… | ||
| Recent Advances in Incremental View Maintenance | Haozhe Zhang | 21m |
Shows that a class of conjunctive queries called q-hierarchical admits constant update time and constant enumeration delay (assuming OM-style hardness… | ||
| Exploiting Sparsity and Redundancy | Matthias Boehm | 20m |
Builds on Apache SystemDS with a DSL and compiler that auto-switches sparse formats at runtime and adds GPU and federated backends, then generates fused C/Java… | ||
| Panel: What are we doing? | Saman Amarasinghe, Gilbert Louis Bernstein, Shoaib Kamil | 19m |
Discussion highlights a persistent gap between research and deployable systems, claiming many sparse advances never reach real workloads and arguing that… | ||
| Extending the Scope of the Array Abstraction beyond Sparsity | Shoaib Kamil | 11m |
Argues that the array abstraction extends past simple sparsity by operating directly on compressed formats (run‑length and implicit-value encodings)… | ||
| nothing matches | ||