Present: abma, jK, Tobi, zerver
- FYI single threaded spring is 1.1% faster than MT with 1 thread (1.1% higher FPS, 1.1% faster Sim).
- can INTERNAL_ORDER be removed?
- target lock should be customizable somehow either modkey or config
>> zerver joined!
>> jK joined!
>> abma joined!
FYI single threaded spring is 1.1% faster than MT with 1 thread (1.1% higher FPS, 1.1% faster Sim).
jK: we need a objective benchmark buildbot
abma: + we need a benchmark?
jK: I got some work on that
abma: imo its hard to get always the same result. some simple background task (auto-updates, some disk sync, lock rotation, cron jobs, ...) could lead to completely different results...
abma: but.... better as no benchmark, yep
can INTERNAL_ORDER be removed?
abma: https://github.com/spring/spring/blob/d ... mand.h#L94
abma: someone responded "maybe remove DONT_REPEAT instead"
abma: uhm... ideas/hints?
jK: with non-deterministic script.txt
abma: aah, nice
abma: at least thats some start point
jK: with deterministic replay it converges very good
jK: the given benchmark was non-deterministic cause I wanted to check the LuaJit performance
jK: and LuaJit isn't syncing w/o additional work
jK: @INTERNAL_ORDER, imo both make sense, I more wonder why they translate to the same bit o_O
>> Tobi joined!
jK: e.g. it's all commands spawned from area commands (& patrol) should be internal (even when kloot is breaking that atm afaik)
jK: so lua and engine can identify 2nd class commands (automatic ones) and 1st class ones (given by user)
jK: offtopic again can someone update buildmaster?
abma: i'lll do that...sure
jK: if everything goes right I am near zero library dependencies
jK: even when I am unsure if getting rid of libm, libc & libpthread make sense >_<
jK: but it might be possibly some distros don't got .6 ones
abma: [LCC]jK: updated
abma: that means we need an upload script? :D
jK: binary itself is <20MB
jK: debugsymbols will make again 200MB uncompressed (duno yet how well it compresses)
zerver: hi, sry I was afk
jK: wonder if it is even faster (static libs `can be` faster than shared ones)
jK: perhaps just with LTO
jK: didn't tried 4.6.3 LTO yet
abma: [LCC]jK: do you have enough bandwith for uploading?
zerver: If 1.1% performance diff for MT/ST is acceptable we could merge them as suggested.
jK: MT still got a lot special code
zerver: Also, I used CPU affinity during the MT vs ST benchmark. The performance boost with affinity set is so huge that I am wondering if we should make it default.
jK: I fixed the problem without affinity on linux
zerver: I tried to disable the "special" codes when only 1 thread is active.
jK: windows can be fixed w/o affinity too
zerver: If you are referring to healthbars and such...
jK: healthbars, command processing, model loading, ...
zerver: Yeah, all that is fixed
jK: nice to hear
zerver: Plz try it, and we can vote about what to do later
jK: still want a benchmark before :/
jK: @affinity: http://msdn.microsoft.com/en-us/library/ms684247.aspx
zerver: I have my benchmark in detail, but if you want to make your own, sure
jK: that's the way to go
jK: problem is it needs mingw64
jK: in particular http://msdn.microsoft.com/en-us/library ... s.85).aspx
zerver: Yeah, saw your comments about it in the spring sources
zerver: so what about release?
jK: ppl seem to unhappy with auto force target
zerver: so we ask kloot to revert it / make it optional?
jK: revert is impossible already
zerver: yeah, it was the root cause of the majority of all bugs in 89.0
zerver: huge number of fixes, hard to revert
jK: so yes, needs a way to be optional (either by pressing a modkey or my setting)
jK: -my +by
zerver: I agree. I somewhat like the feature for artillery and such, but I also feel it may be very confusing for noobs
abma: have to sleep, gn8! :)
<< abma left!