Four Languages for Portability
summary
Argues that four domain-agnostic collection languages — relations, graphs, tensors, space — map to a single set-iteration IR that composes, fuses and tiles across data models and hardware. Demonstrates a compiler path that lowers collection ops to nested loops or space-tree traversals, generates code for CPUs, GPUs, distributed machines and TPU-like/dataflow accelerators, and avoids actual data reshuffles when projecting graphs to matrices by pushing algebra back onto original structures.
AI-written summary · about
chapters (10)
- 0:00Language portability challenges
- 1:03Data collections as key abstraction
- 3:08Modular compiler infrastructure
- 4:29Collection oriented programming
- 5:58Four languages for collections
- 7:21Operations across languages
- 10:10Transitioning between languages
- 11:58Unified intermediate representation
- 14:02Portability across hardware
- 15:03Question and answer session
related talks
48mCross-Domain Compilation: Exploiting Synergies Across the CS Community
32mMind the Gap: General-Purpose Programming Languages Impede Scientific Model(…)
26mInfinite Possibilities
13mA Family of Languages for Trustworthy Agent-Based Simulation
18mDeclarative Data Collections for Portable Parallelism
22mA Formal Model for Portable, Heterogeneous Accelerator Programming
01h 03mThe Language of the System
05mBreaking data identities: Making a case for language-agnosticity