From Prompts to Performance: Evaluating LLMs for Task-based Parallel Code Generation
2026-03-04T19:54:26Z•6ba4188f5da543a2ccc94784c48156d5c726e35570724ccbaf5842ffe48ba338
ACSLC++-parallelismFrama-CHPXIEEE-754LLVMOpenMPRustTorchLeanarrayscode-generationconcurrencydata-racefloating-point-semantics','certificate-checking','IBP','LiRPA','formal-verificationfunction-contractslarge-language-modelsmemory-safetyneural-network-verificationparallelismscalabilitystatic-analysissymbolic-executionundefined-behaviorunsafe-code
What happened
This collection of PL/security research papers covers: (1) evaluation of LLMs for generating task-based parallel code across OpenMP Tasking, C++ standard parallelism, and HPX, showing correctness and scalability limitations that can induce concurrency and performance bugs; (2) a novel symbolic-execution approach that carries array-segment invariants to generate stronger function contracts (pre/post/assigns) and is implemented in LLVM/ACSL/Frama‑C; (3) TorchLean, a Lean4 framework giving a single precise semantics for PyTorch-style models including explicit IEEE‑754 binary32 semantics and proof
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_pl
- Record identifier
- 6ba4188f5da543a2ccc94784c48156d5c726e35570724ccbaf5842ffe48ba338
- Enrichment time
- 2026-03-04T19:54:26Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.