Skip to main content
Przejdź do strony domowej Komisji Europejskiej (odnośnik otworzy się w nowym oknie)
polski pl
CORDIS - Wyniki badań wspieranych przez UE
CORDIS

DaCe GPT

Cel

Modern high-performance processors increasingly rely on parallelism and specialized memory hierarchies, but legacy codebases struggle to harness these advances. Optimizing such applications is essential for energy efficiency and performance but requires expertise in performance engineering. Domain scientists, faced with the growing demand for computational throughput, are forced to be low-level programming experts. Worse, performance tuning often leads to non-local changes across large codebases: this is unsustainable, inflates the maintenance burden, and obfuscates scientific logic.

Through previous ERC-funded research, we demonstrated that data-centric programming (DaCe) provides a path for separating scientific logic from performance optimization and specialization. DaCe was shown to work at scales ranging from small computational kernels to applications with up to a million lines of code when guided by performance experts.

Recent advances in large language models (LLMs) enable coding agents to excel at tasks like API usage and documentation-based pattern matching (“how do I use OpenSSL to do X in Python?”). However, they consistently fail at tuning performance-critical imperative code, as this requires understanding dataflow and how to correctly restructure it. Finally, the resulting changes span many lines of code, requiring more context than available to LLMs.

In this project, we aim to democratize performance tuning by integrating LLMs with DaCe — not as direct code optimizers, but as reasoning interfaces that use a DaCe backend. We propose a novel approach in which users provide plain imperative code (Python or Fortran). Our system, DGPT, powered by a reasoning model, compiles the code into DaCe, analyzes it, applies performance-improving transformations, and verifies the resulting code correctness and performance. Thus, users will see explanations, speedups, and a cleaner, more efficient version of their code.

Dziedzina nauki (EuroSciVoc)

Program(-y)

Temat(-y)

System finansowania

HORIZON-ERC-POC - HORIZON ERC Proof of Concept Grants

Wyświetl wszystkie projekty finansowane w ramach tego programu finansowania

Zaproszenie do składania wniosków

(odnośnik otworzy się w nowym oknie) ERC-2026-POC

Wyświetl wszystkie projekty finansowane w ramach tego zaproszenia

Instytucja przyjmująca

EIDGENOESSISCHE TECHNISCHE HOCHSCHULE ZUERICH
Wkład UE netto
€ 150 000,00
Koszt całkowity
Brak danych

Beneficjenci (1)