| Time |
Nick |
Message |
| 00:13 |
|
adfeno joined #minetest-dev |
| 01:18 |
|
pmp-p joined #minetest-dev |
| 01:24 |
|
Fractalis joined #minetest-dev |
| 01:30 |
|
Foz1 joined #minetest-dev |
| 05:00 |
|
MTDiscord joined #minetest-dev |
| 07:58 |
|
calcul0n joined #minetest-dev |
| 08:00 |
|
ShadowNinja joined #minetest-dev |
| 10:24 |
|
proller joined #minetest-dev |
| 10:45 |
|
lisac joined #minetest-dev |
| 11:12 |
|
clavi joined #minetest-dev |
| 11:24 |
|
Fixer joined #minetest-dev |
| 11:42 |
|
YuGiOhJCJ joined #minetest-dev |
| 14:09 |
MTDiscord |
<appguru> @exe_virus not a good idea as you'll need your own ipairs |
| 14:10 |
MTDiscord |
<appguru> and not all Lua implementations handle [0] = ... well (LuaJIT does tho) |
| 14:10 |
MTDiscord |
<appguru> (performance-wise) |
| 14:10 |
MTDiscord |
<appguru> if you always subtract 1, you can usually shift your variable |
| 14:10 |
MTDiscord |
<appguru> also a single arithmetic OP most likely won't be the bottleneck |
| 15:03 |
|
proller joined #minetest-dev |
| 15:18 |
|
Fractalis joined #minetest-dev |
| 15:30 |
|
proller joined #minetest-dev |
| 15:34 |
MTDiscord |
<exe_virus> Ipairs is slow anyways. And you're right, it's not the main bottleneck, but it does make the code messy |
| 15:37 |
specing |
what's the fastest, for i = 1, #list? |
| 15:37 |
specing |
Hmm, there was a springrts benchmark out there, but springrts doesen't use luajit to my knowledge |
| 15:39 |
|
pyrollo joined #minetest-dev |
| 16:30 |
|
Fractalis joined #minetest-dev |
| 16:48 |
|
Andrey01 joined #minetest-dev |
| 17:12 |
MTDiscord |
<appguru> > Ipairs is slow anyways. And you're right, it's not the main bottleneck, but it does make the code messy Not using ipairs makes the code messy for little to no performance gain |
| 17:41 |
|
Fractalis joined #minetest-dev |
| 18:01 |
sfan5 |
merging #10659 in a few minutes |
| 18:01 |
ShadowBot |
https://github.com/minetest/minetest/issues/10659 -- Fix Visual Studio build in Actions by MoNTE48 |
| 18:04 |
|
Fractalis joined #minetest-dev |
| 18:22 |
|
pyrollo joined #minetest-dev |
| 18:40 |
|
Fractalis joined #minetest-dev |
| 18:59 |
|
Andrey01 joined #minetest-dev |
| 19:08 |
|
homthack joined #minetest-dev |
| 20:11 |
|
kilbith joined #minetest-dev |
| 20:15 |
|
LoneWolfHT left #minetest-dev |
| 20:25 |
|
TC01 joined #minetest-dev |
| 20:25 |
|
Shara joined #minetest-dev |
| 20:25 |
|
cheapie joined #minetest-dev |
| 20:25 |
|
Kray joined #minetest-dev |
| 20:25 |
|
thePalindrome joined #minetest-dev |
| 20:39 |
|
Andrey01 joined #minetest-dev |
| 20:52 |
|
kilbith_ joined #minetest-dev |
| 20:56 |
|
Foz joined #minetest-dev |
| 23:08 |
|
lhofhansl joined #minetest-dev |
| 23:12 |
lhofhansl |
Looks like #10656 is hurting performance. All the lookup of v3s16 seems to slow things and the indirection through MapSector had helped with that. |
| 23:12 |
ShadowBot |
https://github.com/minetest/minetest/issues/10656 -- Remove MapSector completely. by lhofhansl |
| 23:12 |
lhofhansl |
Using an unordered_map (with a hash function on v3s16) does not improve things (in fact makes it much worse). |
| 23:13 |
lhofhansl |
So I think I'm going to drop that PR. |
| 23:14 |
sfan5 |
how does that hash function look? |
| 23:19 |
lhofhansl |
https://github.com/freeminer/freeminer/blob/master/src/util/unordered_map_hash.h#L31 |
| 23:19 |
lhofhansl |
(Lifted it from freeminer) |
| 23:20 |
|
Ritchie joined #minetest-dev |
| 23:22 |
sfan5 |
hm |
| 23:22 |
lhofhansl |
It seems iteration it |
| 23:22 |
lhofhansl |
is what is expensive. |
| 23:22 |
sfan5 |
that'd make sense |
| 23:23 |
lhofhansl |
I was a bit surprised myself. |
| 23:27 |
lhofhansl |
I |
| 23:27 |
lhofhansl |
(can't type today) |
| 23:27 |
lhofhansl |
I |
| 23:27 |
lhofhansl |
I'll work on multi-threading with with MapSectors. Would have been easier without those. |
| 23:29 |
lhofhansl |
And #10637 seems somewhat important to avoid generating the same map chunk multiple times (when multiple emerge threads are enabled) |
| 23:29 |
ShadowBot |
https://github.com/minetest/minetest/issues/10637 -- Avoid generating the same chunk more than once with multiple emerge threads by lhofhansl |
| 23:38 |
|
Fractalis joined #minetest-dev |
| 23:45 |
lhofhansl |
I could also introduce a hierarchical map without the need for MapSector. That would retain the structure from before. I'll try that. |
