Programming Obesity: A Code Health Epidemic
summary
Argues that generalized pointers are the main cause of systemic bloat and shows a working APL-based compiler that stores syntax trees as a depth vector, not pointer graphs. The compiler compiles the same code for GPU and CPU, shrinks ~1,012 lines to ~17, reduces AST nodes and names by multiple×, cuts memory for a 16.6M-node test from ~1–1.4GB down to ~63MB in one implementation, and delivers ~9× CPU speedup and ~50–200× on GPU in benchmarks.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:30Perfect programming environment
- 4:03We are producers
- 5:04What is programming obesity
- 6:32Examining complexity
- 7:48Breaking down complexity
- 10:31Impact
- 11:55Generalized pointers
- 12:58Trees
- 17:43What are we eliminating
- 19:11Can a compiler be GPU host
- 20:12A new approach
- 22:19Compiler
- 27:16Racket
- 46:27APL
related talks
56m“High Performance Code 201: Hybrid Data Structures"
40mLessons from two decades of writing bad code
26mWhen Idioms Become Bottlenecks
01h 07mSmall Is Beautiful
57mInvited Talk: Bridging Safety and Performance
02h 25mHow to Keep C++ Binaries Small - Techniques for C++ Optimization
45mDesigning your Data: The Bread & Butter of APL Performance
56mWrite Fast Code Like a Native