Re: [PATCH 3/2] Avoid unnecessary strlen() calls

Linus Torvalds [off-list ref] writes:

Dscho's bigger patch probably helps more on an in-order architecture, and 
should be equally good on a P4 (or Opteron). On Core 2, neither of the 
patches seem to make a huge difference.

Because hoisting stable test outside loop is always better for
any architecture, I thought picking between Gitte and Gitney
patches is a no brainer, and I didn't bother to compare-bench,
but I got curious.

(plain)
7.89user 0.15system 0:08.08elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41608minor)pagefaults 0swaps
7.93user 0.18system 0:08.14elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41608minor)pagefaults 0swaps

(gitte -- separate function for slow path)
6.98user 0.18system 0:07.17elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41606minor)pagefaults 0swaps
7.14user 0.12system 0:07.26elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41607minor)pagefaults 0swaps

(gitney -- cheap test first before isspace)
7.23user 0.18system 0:07.42elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41608minor)pagefaults 0swaps
7.32user 0.14system 0:07.48elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+41607minor)pagefaults 0swaps

So it does not seem to make much difference on Athlon 64x2 either.

Will apply the "stupid hashcmp() removal" and Dscho's patch and
call it a day.