Hacker News Reader

He is still missing profile guided optimizations (with his testcase), and bolt.

And in some/most cases -Os is faster than -O3

Boehm GC is usually slower than glibc with lots of allocs. MPS would be better, but needs lots of rewriting. https://github.com/Ravenbrook/mps

View on HN

BOLT is only good for ~2% in this scenario, which is about what I would have expected from glancing at the profiles.

-Os is much slower, as I would have expected. Nothing in the perf data suggested high L1i miss rate. The .text of the "size optimized" program is, for some reason, 16x larger.

> And in some/most cases -Os is faster than -O3 This surprises me. Can you please elaborate?

If the whole program can fit in the L1/L2 cache, it can speed up things quite a lot!

A IC cache miss costs 50x more on average. If your insn and data are small, the overhead of doing more work is less than the overhead of cache misses

Reply on news.ycombinator.com