From: Blue Lizard (webmaster@dofty.zzn.com)
Date: Wed Apr 10 2002 - 22:44:01 EDT
On Fri, 2002-04-05 at 18:09, Patrick Lam wrote:
>A smart C++ compiler would've automatically inlined them, but oh well.
>
Um...I should really know when to stick a fish in my mouth, stop
squirming, and let it go...but yes, IT WOULD HAVE. WOULD HAVE if
linux.mk didnt suck for not letting me use any normal flags with
profiling ones. Gcc, poor thing, was told not to compile with any
inlining, so it didnt. I suck too, for not investigating linux.mk
earlier, esp. after i was noticing some rather suspicious outputs in the
profile comparison.
Anyway, dom told me how to entirely override that evil, EVIL file. I
got a lot of retesting to do of old (thought broken) patches. Please
remove that inline. Smart c++ compilers will inline unless they are
told not to, so we dont need to tell them to. This is buggin' me.
I will give you some results more accurate to everyday use soon.
Auch, consider the small regression we did get when not consistantly
inlining down the call tree til hit the spot. The parent needed an
inline more.
My very first thought was, why in satan's domain are we not queuing or
at least buffering those pref calls if we truely must call thousands of
times/sec? Buffer would help by letting the cpu cache hold on longer.
Queuing and pushing the queued calls through as one though, mm mm good!
What say ye oh developers of the beast?! The concept feels right, even
if I don't have any (productive) experience mucking around in the ant's
guts.
la lala la laaaa
-MG
This archive was generated by hypermail 2.1.4 : Wed Apr 10 2002 - 03:45:06 EDT