The Mysterious Swift Performance
summary
CryptoSwift development surfaces concrete speed levers: build with -O and whole-module optimizations, prefer release over debug (debug ~30x slower), use unchecked arithmetic for ~2–3% gains and inout/address-passing to avoid copies (one refactor yields 5x faster), prefer simple for-in loops to map/reduce, reserveCapacity for arrays, and prefer stack or static data over heap; an example goes 0.5s -> 0.065s -> 0.0116s -> 0.007s as storage moves from heap to stack to static data. Use @specialize on hot generics and run perf tests to detect regressions.
AI-written summary · about
chapters (14)
- 0:00Welcome and introduction
- 1:27Swift performance background
- 6:27Compiled vs script Swift
- 11:56Compiler flow and optimizations
- 13:03Basic Swift optimizations
- 16:44Struct vs class performance
- 18:17Loop and collection optimizations
- 22:05Function inlining techniques
- 26:33Stack vs heap allocation
- 33:34Real-world optimization example
- 39:23Array vs ArraySlice trade-offs
- 42:57Generic specialization impact
- 49:16Debug vs release performance testing
- 50:23Q&A session
