Functional Array Funhouse Intensive - Part 5 of 5
summary
Demonstrates an APL/Dialog array programming workflow with a working compiler that emits CPU and GPU code, shows function trains as the primary notation, explains stack vs heap allocation and parallel flattening, and adds GPU caching to keep data on the device. Claims order-of-magnitude reductions in code size and speedups, gives concrete examples (matrix reshapes, path matrices), describes a debugger/visualizer prototype, and admits iterative fixes and open complexity questions.
AI-written summary · about
chapters (19)
- 0:00Intro
- 6:42Readability
- 11:28Dependent types
- 12:07Decode
- 13:15Code
- 14:08Knut
- 16:18Overlaps
- 19:51Converting Objects
- 20:42Garbage Function
- 22:46Enter the Matrix
- 25:15Direct Development
- 29:11APL Community
- 32:42Array Data Storage
- 33:38Array Query Language
- 34:26APL
- 35:31Auto Conversion
- 36:55Debugging Trains
- 40:46MATLAB vs APL
- 41:39Aesthetics
related talks
01h 17mModern APL in the Real World: Theory, Practice, Case Studies
24mAccelerating Haskell GPGPU Programming with Haskell
24mShared Memory & Parallel Programming with JavaScript
47mRethinking compilers with live coding
32m"Making Elm Talk to Your Personal Supercomputer"
23mPattern-Based Optimization (...)
28mCommunication Avoiding Algorithms for Linear Algebra and Beyond
41mConcurrency and Parallel Computing in JavaScript