HVM2: Iteraction Combinator Evaluator
summary
HVM2 is an implemented massively parallel GPU and CPU evaluator for extended interaction combinators that compiles non‑sequential high‑level terms to C and CUDA runtimes and shows large speedups as cores increase. It uses a textual net format, reference nodes, numeric primitives encoded in tagged integers, and a 64‑bit atomic lock‑free linker to grab redexes; current limits include a 29‑bit (≈4GB) address space and mostly self‑benchmarks, and formal semantics and broader comparisons remain incomplete.
AI-written summary · about
chapters (15)
- 0:00HVM2 introduction
- 2:10Interaction combinator runtime
- 7:07Lambda calculus encoding
- 16:23Reduction rules walkthrough
- 19:00Affinity constraints
- 20:12Strong confluence
- 26:18HVM2 syntax
- 35:38References node type
- 37:44Numeric primitives
- 42:29Memory layout and GPU
- 46:51Atomic operations
- 50:01GPU scheduling
- 51:26Compiling to HVM2
- 55:17HVM2 limitations
- 57:42Q&A
related talks
15mHEMVM: a Heterogeneous Blockchain Framework for Interoperable Virtual Machines
23mVellvm: Formalizing the Informal
01h 02mMemory and Compilers
16mAccelerate HPC and AI Applications with RVV Auto-Vectorization
44mBuild Your Own VM with JIT Compilation for Dummies
15mMetaprogramming with Combinators
18m(De/Re)-Composition of Data-Parallel Computations via Multi-Dimensional(…)
43mAuto-Vectorization in HotSpot