Beating the compiler at signed 32-bit division
Web page · https://lemire.me/blog/2026/10/31/beating-the-compiler-at-signed-32-bit-division
First seen by Oriel on .
AI summary
The article exploits extra range when dividing signed 32-bit integers on 64-bit processors, shortening constant division and making fastmod 0.2.0 branchless for runtime divisors.
- Among divisors from 3 through 1000 excluding powers of two, 376 of 990 require the compiler’s multiply-and-add correction because the ideal multiplier needs 33 bits.
- The div32 template uses a 33-bit multiplier in signed 64-bit arithmetic, followed by a shift and sign-based rounding; it assumes a positive, non-power-of-two divisor.
- For divisors 7, 60, and 365, Intel latency fell from roughly 10 cycles to 8, while Apple M4 latency improved by 12%.
- For runtime divisor 7 on Intel, fastmod 0.2.0 achieved about 0.78 ns per division, versus 1.72 ns for division and over 3 ns previously.
- The optimization should only replace compiler output requiring correction; for divisor 1000 on Apple M4, div32 was slower because it needed a separate sxtw instruction.
#integer division #compiler optimization #fastmod #cpu performance
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Lobsters: 13, latest observation ; +12 over observed period since
Comments
Comment count reported by the source.
- Lobsters: 5, latest observation ; +5 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Lobsters | ep5vsq | https://lemire.me/blog/2026/10/31/beating-the-compiler-at-signed-32-bit-division/ Read discussion |