Minetest logo

IRC log for #minetest-dev, 2015-06-27

| Channels | #minetest-dev index | Today | | Google Search | Plaintext

All times shown according to UTC.

Time Nick Message
00:11 Lunatrius joined #minetest-dev
00:12 RealBadAngel est31, damn whole the code around have * and & on the left side
00:13 RealBadAngel i just added one parameter to that function
00:14 RealBadAngel am i supposed to edit whole drawscene.cpp or what?
00:15 est31 no RealBadAngel just edit the diff :)
00:15 est31 only the things you change matter
00:15 RealBadAngel then in function def one parameter will look different than others
00:16 RealBadAngel not to mention whole file
00:16 est31 so then, thats the correct style
00:16 RealBadAngel unintuitive one imho
00:17 RealBadAngel having ptr next to function name looks completely idiotic
00:18 est31 its in the kernel style https://www.kernel.org/doc​/Documentation/CodingStyle
00:18 Lunatrius joined #minetest-dev
00:20 est31 ok can you at least do it in the new files?
00:21 est31 like here: https://github.com/minetest/minetest/pull/2814/​files#diff-f20c82d93887e9a0f4d4eeb66cc7eaa7R419
00:22 RealBadAngel ive changed the lines im touching
00:24 est31 ok
00:26 RealBadAngel done
00:27 est31 ok so why di you make your own queue?
00:27 est31 do*
00:29 RealBadAngel you mean in minimap update thread?
00:29 est31 yes
00:29 est31 why not std::queue
00:29 RealBadAngel its not iterative
00:30 est31 ?
00:30 RealBadAngel you cant iterate over std:queue elements
00:30 RealBadAngel its fifo
00:30 RealBadAngel no access to anything in the middle
00:30 est31 and why do you iterate?
00:31 RealBadAngel because of multiple updates
00:31 RealBadAngel if that happens i do cancel previous one
00:31 est31 I see
00:32 RealBadAngel same thing is done in mapblock mesh
00:32 RealBadAngel and its all about cracks
00:32 RealBadAngel those suckers are causing block updates
00:33 RealBadAngel thats one of the reasons im against cracks (when done in mapblock mesh)
00:33 RealBadAngel but that is kinda religion, nobody can touch it ;)
00:34 est31 what I've seen in kilbith's screenshot is a bad way to do crack replacements
00:34 est31 cracks shouldnt be removed :)
00:34 RealBadAngel particles are way better
00:35 est31 yes at the end they are good
00:35 RealBadAngel what screenshot youre talking about?
00:35 RealBadAngel cracks have also one nasty effect
00:35 est31 #1685
00:35 ShadowBot https://github.com/minetest/minetest/issues/1685 -- Digging animations tweaks. by RealBadAngel
00:36 RealBadAngel each punching new node is causing growing memory usage
00:36 est31 ? its not freed?
00:36 RealBadAngel because for each step of crack animation new texture is made
00:37 RealBadAngel one should just cut that ^ operators code, burry it deep somewhere
00:37 est31 ?
00:38 RealBadAngel you think why minetest with dreambuilder and hdx needs 12gb of ram?
00:38 RealBadAngel because texture pack is 11gb?
00:39 est31 ah lol
00:39 RealBadAngel happily creating new textures everywhere
00:39 est31 yea
00:39 RealBadAngel rotating, cracks etc
00:40 est31 you added the rotating part didnt you?
00:40 RealBadAngel i made that with vertices
00:40 RealBadAngel ^ is not my sin
00:40 est31 whats a vertex
00:41 RealBadAngel point in 3d space
00:41 est31 so a vector?
00:41 RealBadAngel its a point, not a vector
00:41 est31 whats a vector then?
00:42 RealBadAngel https://en.wikipedia.org/wiki/Vec​tor_%28mathematics_and_physics%29
00:43 est31 for me thats a point in 3d space
00:43 est31 isnt it
00:43 RealBadAngel no, vector is (pos of point B - pos of point A)
00:43 est31 yes, but you have a 0
00:43 est31 whatever
00:44 est31 first entry
00:44 est31 ok, so how did you do it with vertices?
00:44 RealBadAngel hold on
00:45 RealBadAngel https://github.com/minetest/minetest/b​lob/master/src/mapblock_mesh.cpp#L455
00:46 RealBadAngel whole 6dfacedir is made this way
00:47 RealBadAngel instead of rotating textures (and generating new ones) face is roated in 3d space
00:47 est31 thats nice
00:47 RealBadAngel now lets try to make ores
00:47 est31 so whats generating new textures?
00:47 RealBadAngel we should display stone
00:48 RealBadAngel add offset for ore face
00:48 RealBadAngel and display it
00:48 RealBadAngel and voila, no new combined texture needed
00:50 est31 how do you mean "add offset"?
00:51 RealBadAngel imagine two sheets of paper
00:51 RealBadAngel put one on another
00:52 RealBadAngel cut holes in that one on top, you will see the bottom one
00:53 est31 why not do just a simple cache?
00:53 est31 just store the textures you need
00:53 est31 and if you dont have them, just remove them
00:53 RealBadAngel how could you tell whats needed and whats not?
00:53 est31 err if you dont need them
00:53 RealBadAngel you have already defined them all
00:53 RealBadAngel on startup
00:54 RealBadAngel everything is here and loaded into ram
00:54 est31 so change it
00:54 est31 make it not load everything on startup
00:54 est31 but when needed
00:54 RealBadAngel good luck with lags
00:54 RealBadAngel preload item visuals says hello
00:55 est31 "preload item visuals" is one of the RAM generators
00:55 RealBadAngel but it works to show both sides of the problem
00:55 RealBadAngel how slow will be such cache
00:56 RealBadAngel and how bad item visual is (new texture ofc)
00:56 est31 as slow as texture generation at the start
00:56 RealBadAngel instead of displaying just mesh
00:56 RealBadAngel its all because we are doing everything wrong
00:57 est31 true
00:57 RealBadAngel instead of drawtypes that do create needed meshes we generate textures on how the object will look like in game
00:57 RealBadAngel thats sick
00:57 RealBadAngel game is capable of displying it already
00:58 RealBadAngel why to take a picture of it for christ sake?
00:59 est31 I agree, taking a picture is bad.
00:59 est31 in that case
01:00 RealBadAngel btw, take a look at minimap draw
01:00 RealBadAngel its a mesh
01:00 RealBadAngel flat, but still a mesh
01:00 RealBadAngel you can freely rotate it
01:01 RealBadAngel and still display it as it would be a texture
01:01 est31 so?
01:01 RealBadAngel now, just take a nodes and display all of them at once
01:01 RealBadAngel at given positions
01:02 RealBadAngel selected one can even spin around for fancy look
01:02 RealBadAngel you have just imagined formspec inventory slots using real meshes
01:03 est31 I agree that that is the thing to do
01:03 est31 its barely adding additional draw time for the gpu
01:03 RealBadAngel one of the steps to reduce ram usage
01:04 est31 and it reduces ram usage
01:04 RealBadAngel next thing are specialized drawtypes
01:04 RealBadAngel for wires, tubes, panes etc
01:04 est31 yes
01:05 est31 wires and tubes have same drawtype I guess
01:05 RealBadAngel almost
01:05 RealBadAngel tubes are 6d
01:05 RealBadAngel wires are wallmounted
01:05 est31 ?
01:06 RealBadAngel btw, do you know how many nodefs wire that could climb the walls needs?
01:06 est31 we are speaking about technic cables, and pipeworks tubes?
01:06 RealBadAngel those are using the same
01:06 RealBadAngel wirelike is mesecons wire
01:06 est31 ok then I dont know the wire you speak about
01:07 RealBadAngel single mesecons wire would need circa 1k nodedefs
01:07 est31 ah isnt that raillike?
01:07 RealBadAngel no its not
01:08 RealBadAngel railike is specialized for rails
01:08 RealBadAngel not the wires
01:09 RealBadAngel but are good example. you need just one nodedef for rails
01:09 est31 ah I have the code now
01:09 est31 ok
01:09 RealBadAngel and they are forming connections live
01:09 est31 then we need wirelike too
01:09 RealBadAngel to simulate that behaviour modders predefine all possible connections
01:09 RealBadAngel each being a nodedef
01:10 RealBadAngel a nodebox, a mesh
01:10 RealBadAngel one frigging wire would need 1k meshes
01:11 RealBadAngel but what about if we wanted different couloured wire?
01:11 RealBadAngel another 1k
01:11 est31 yes I know
01:11 est31 adding drawtypes is good
01:12 RealBadAngel mod that makes such tricks is even in minetest_game
01:12 RealBadAngel panes mod
01:14 est31 btw do we really need to refer to MinimapPixel with pointers?
01:14 est31 if we have 64 bit program, then there is no advantage
01:15 est31 only disadvantage by free and malloc
01:15 est31 ah I see
01:15 est31 not needed
01:16 est31 disregard
01:16 est31 it would be good to be abled to specify a cursor for round maps too
01:16 est31 one which isnt rotated
01:16 est31 e.g. a crosshair
01:17 RealBadAngel rotated one you can do without problems
01:17 RealBadAngel just draw it on overlay image
01:17 est31 ah I see
01:17 est31 ok
01:18 RealBadAngel you can add there NESW on the borders, whatever
01:18 RealBadAngel those overlays can be modified by texture packs
01:19 est31 can you add the minimap to formspec too?
01:20 RealBadAngel since mapper stores info about all loaded blocks it could be used to draw bigger maps
01:20 RealBadAngel and display textures whenever you want
01:20 est31 ok I +1 it
01:20 est31 but I merge it
01:20 est31 because of the \r\n
01:20 RealBadAngel so for example full screen map could be possible
01:21 RealBadAngel hold on a sec, i will squash it
01:23 est31 also, if you want, you can specify your own name in the copyright boilerplate
01:23 est31 but I guess you know that already
01:24 est31 entirely your descision
01:24 Warr1024 what's the policy on licenses, btw?  does it need to be GPL, or just compat?
01:25 est31 no GPL isnt even allowed
01:25 RealBadAngel est31, void *MinimapUpdateThread::Thread()
01:25 Warr1024 er, lgpl or whatever?
01:25 RealBadAngel you commented that "space"? why?
01:25 est31 yes minetest is lgpl 2.1+
01:25 Warr1024 it was one of those gnu ones, wasn't it?
01:25 est31 yes
01:26 Warr1024 sry, I'm usually way down in MIT/ISC-land
01:26 est31 so we need compat yea. I guess nobody will object if they publish their code more liberally
01:26 est31 RealBadAngel, while(!StopRequested()) {
01:27 est31 that one?
01:27 RealBadAngel ah, that ones
01:27 RealBadAngel i thought it was bout ptrs again
01:32 RealBadAngel est31, ok, outdated all of your comments
01:49 deltib joined #minetest-dev
01:58 paramat joined #minetest-dev
02:05 est31 ok RealBadAngel found some other style issue
02:05 est31 I make a commit which also fixes the position update bug, ok?
02:12 paramat i'll push #2831 later
02:12 ShadowBot https://github.com/minetest/minetest/issues/2831 -- Mgv5/mgv7 caves: Fix sand in underground caves by paramat
02:19 hmmmm bug!?
02:19 hmmmm why was something patched with a bug in it
02:19 VanessaE bug!
02:19 VanessaE :P
02:19 hmmmm paramat:  i looked at that, seems good
02:19 paramat cool thanks
02:20 hmmmm also why did you not switch to std::queue for the queue
02:21 hmmmm meh
02:21 est31 <est31> ok so why di you make your own queue?
02:21 est31 <est31> do*
02:21 est31 <RealBadAngel> you mean in minimap update thread?
02:21 est31 <est31> yes
02:21 est31 <est31> why not std::queue
02:21 est31 <RealBadAngel> its not iterative
02:21 est31 <est31> ?
02:21 est31 <RealBadAngel> you cant iterate over std:queue elements
02:21 est31 <RealBadAngel> its fifo
02:21 est31 <RealBadAngel> no access to anything in the middle
02:21 est31 <est31> and why do you iterate?
02:21 est31 <RealBadAngel> because of multiple updates
02:21 est31 <RealBadAngel> if that happens i do cancel previous one
02:21 est31 <est31> I see
02:21 hmmmm why not use std::list then...
02:22 hmmmm is there any need for random access to the queue?
02:22 est31 vector is indeed a bad choice
02:22 hmmmm there's a lot of egregious code style errors too
02:22 est31 no
02:22 est31 m_queue and m_blocks_cache have to be renamed
02:22 hmmmm meh
02:23 est31 how can I convert vector3d<s16> to vector3d<float> ??
02:24 hmmmm v3s16 to v3f?
02:24 est31 yea
02:24 hmmmm there's no intToFloat?
02:24 hmmmm hrmm, well for the cave code it just uses v3f(pos.X, pos.Y, pos.Z) where pos is a v3s16
02:25 hmmmm v3f posf(pos.X, pos.Y, pos.Z)
02:25 est31 my code is m_minimap_update_thread->last_update_pos - data->pos
02:25 est31 I'll just assign a variable
02:34 paramat now pushing 2831
02:39 paramat complete
02:45 paramat i'm working on the fractal mapgen. it will be a 4D fractal with a 4th dimension co-ordinate parameter to choose the 3D slice generated, this will enable a big choice of different shapes. i've also added a mode to generate all the corresponding julia sets. possibly mergeable for 0.4.13
02:47 est31 RealBadAngel, does every pixel on the minimap get converted into one pixel on the screen?
02:51 VanessaE est31: no.
02:51 VanessaE the image is scaled according to the window size.
02:52 est31 argh dammit
02:52 est31 seems I need that scaling factor
02:52 est31 or find another way to do it
02:55 paramat woo this minimap is fun
02:56 VanessaE est31: what are you fixing anyway?
02:56 est31 that slow update issue
02:57 VanessaE ohhh
02:57 est31 position update I mean
03:08 deezl joined #minetest-dev
03:09 paramat ah i see the square map doesn't rotate, so they have complimentary uses. surprisingly i prefer the round one =)
03:21 paramat think i'll make my own non-solid player marker texture that doesn't cover the central area
03:34 RealBadAngel est31, dont try to skip the update if position remains the same
03:34 est31 ?
03:35 RealBadAngel i can see youre trying to do something like that, am i right?
03:35 est31 no, I want to update the position faster than the minimap
03:36 est31 so I've tried to update it with material.TextureLayer[0].setTextureMatrix
03:36 est31 inside Mapper::drawMinimap()
03:36 est31 but that doesnt work
03:36 est31 that method has no effect at all
03:37 RealBadAngel ? mimimap runs the same fps as whole game
03:37 RealBadAngel at least on my box
03:37 est31 not on mine, when  I turn shaders off
03:37 est31 RealBadAngel, have you seen my screencast?
03:37 est31 it updates the minimap perhaps thrice a second
03:38 RealBadAngel why would you disable shaders?
03:38 est31 RealBadAngel, thats not the point
03:38 RealBadAngel shaders (final color blend) are faster than c++ code
03:38 est31 If I enable them, the effect is not as strong, but still there
03:39 est31 also, rotation is not affected
03:39 est31 so thats very weird
03:39 est31 rotation working on one hand
03:39 est31 and position update not working
03:39 RealBadAngel rotation has nothing to do with scanning
03:39 est31 and it shouldnt
03:40 est31 but I want to have position updates be faster
03:40 RealBadAngel cant watch the screencast, im on mobile connection atm
03:41 est31 why doesnt material.TextureLayer[0].setTextureMatrix() work, do you have an idea?
03:41 RealBadAngel what you want to get with changing the  texture matrix?
03:42 est31 from what I see, the  Mapper::drawMinimap() is called more often than an actual update to the minimap texture
03:43 est31 now I just take the difference between last position the map was updated and the current position
03:43 RealBadAngel draw is called each frame
03:44 RealBadAngel it has to be
03:44 RealBadAngel oherwise you couldnt see it
03:44 est31 but the position doesnt update each frame
03:45 RealBadAngel you cant help that
03:45 RealBadAngel on my box minimap update is rdy faster than, so thread waits for texture to be grabbed
03:46 est31 not on mine
03:46 est31 only because you cant reproduce an issue doesnt mean others dont have it
03:47 RealBadAngel whats your cpu?
03:47 est31 amd
03:47 est31 is that importannt
03:47 RealBadAngel cores and speed
03:47 est31 2ghz 6 cores
03:47 RealBadAngel lookin for a reason, so yeah, its important
03:48 RealBadAngel ovearall fps?
03:48 est31 30
03:48 est31 to 40
03:48 RealBadAngel when you enable minimap, how much fps you are loosing?
03:49 est31 none noticable
03:50 RealBadAngel so why you think matrix operations have something to do with map update speed?
03:50 est31 I dont want to improve update spee
03:50 est31 d
03:50 est31 I want to work around the issue
03:50 RealBadAngel those are done for each and every object in scene node
03:50 est31 the texture isn't up to date yes
03:51 RealBadAngel this has nothing to do with draw code then
03:51 est31 but at least I want to translate it by the amount the player has moved in the time the last update was
03:51 est31 and that has to happen at the same place as the rotation is applied
03:52 RealBadAngel i will add some test code to find locate your problem
03:52 RealBadAngel will do that in my branch
03:52 est31 ok
03:53 RealBadAngel it looks like for some reason scan takes longer than it should
03:53 RealBadAngel which is strange because most heavy operations are done in mapblock_mesh
03:54 RealBadAngel and you dont see lags when loading meshes around, dont you?
03:54 est31 what do you mean
03:55 est31 they just appear, movement stays smooth
03:56 RealBadAngel so it means that most heavy scanning operations are done also fast
03:56 RealBadAngel minimap_mapblocks are created together with mapblock_mesh
03:57 RealBadAngel in mapblockmesh_update_thread
03:57 RealBadAngel does mode of minimap any effect on update speed for you?
03:58 RealBadAngel i mean the surfacex1 vs x4 ?
03:58 est31 x1 zoom is slower than x2 is slower than x4
03:59 RealBadAngel does x4 runs at acceptable speed?
03:59 est31 no
04:00 est31 err yes
04:00 est31 or almost
04:00 est31 its borderline
04:00 est31 but I'd say yet
04:00 RealBadAngel i see
04:00 est31 yes*
04:01 est31 there is a noticeable difference, but you can excuse it with the pixel-y design of minetest if you know what I mean
04:03 RealBadAngel what takes most of the time is MinimapUpdateThread::getMinimapPixel
04:03 RealBadAngel i mean getMap which calls above
04:04 RealBadAngel since blocks are 16x16 chunks of data we need to continuosly translate node pos to block pos
04:04 RealBadAngel for each and every pixel
04:05 RealBadAngel thats only thing i can think of to improve the speed
04:05 est31 just work around it
04:06 RealBadAngel if it wont be translated you will get movement step at a size of mapblock
04:07 est31 just translate by the delta pos since the last texture update
04:07 est31 the drawing is smooth, but the minimap doesnt update often enough
04:08 est31 so the rotation is smooth, thats my core problem
04:08 RealBadAngel because those two work in separate threads
04:08 RealBadAngel draw only takes new texture if scan is ready
04:08 est31 so just make a position adjustment in the same thread as the rotation
04:09 RealBadAngel that wont help
04:09 RealBadAngel same as delta
04:09 est31 why
04:09 RealBadAngel because for same reason youre getting scan ready too slow
04:10 RealBadAngel draw rate is 30-40 times per second
04:10 RealBadAngel if mapper stays behind, it means its ready with scan less frequent
04:10 RealBadAngel and you cant get another scan in the middle of it
04:11 est31 I dont want to re-scan
04:11 RealBadAngel so you see the last one complete
04:11 est31 I just want to move the texture as far as you moved since the last scan
04:12 paramat http://i.imgur.com/VXqVQZf.png testing new mountain parameters in mgv7, found this floatland rising to y = 945 (hmmmm)
04:12 paramat left #minetest-dev
04:12 RealBadAngel first of all you do have a few pixels margin left for that
04:12 RealBadAngel those which are covered by overlays
04:13 est31 ok
04:13 est31 now how to move it
04:13 hmmmm paramat, i hope that isn't too common
04:13 hmmmm that's a little too crazy for a default mainstream mapgen
04:14 VanessaE hmmmm: maybe but wow
04:15 VanessaE there's something strangely appealing about the idea of playing on such a crazy map
04:16 est31 RealBadAngel, https://github.com/est31/minetest/commit/​81586b58ed2be68f02f2c0b8e91e3d785120f330
04:16 hmmmm then edit your parameters
04:16 hmmmm don't make them the defaults
04:17 VanessaE I didn't say I wanted it, just that it's "strangely appealing" :)
04:17 * est31 feels the urge to finally implement VAEs so that he can give those floatlands a float-ore and when you mine too much of it, the whole floatland falls down and crashes into the ground
04:18 VanessaE heh
04:18 est31 and you can make your buildings fly with that ore too
04:21 RealBadAngel http://irrlicht.sourceforge.net/docu​/classirr_1_1core_1_1_c_matrix4.html​#a2bab9633697a892f08d89c7aeee6daf6
04:22 RealBadAngel maybe you should try this one
04:25 RealBadAngel est31, anyway i will add there later code to measure time for it, something runs definitely way too slow for you
04:25 est31 doesnt help either
04:31 VanessaE RealBadAngel: the map shifts at about 4 pixels per second for me when walking, while the world is refreshing at ~35 fps.
04:32 VanessaE so, similar to est31's result
04:32 est31 its biased by up to 10 when I'm flying with fast
04:32 VanessaE while turning is at full fps
04:32 VanessaE biased?
04:32 est31 you know lagging behing
04:32 est31 d*
04:33 VanessaE OH
04:33 VanessaE yeah I see that effect too, but it's somewhat subtle
04:33 VanessaE maybe 5 pixels behind
04:34 est31 try this commit to get accurate numbers https://github.com/est31/minetest/commit/​8fab7f74f09680705a519c4609a559aae08a3dc1
04:35 est31 oh, I didnt mean pixels, I meant nodes
04:35 est31 it will print the delta in life
04:35 VanessaE ok, sec.
04:36 VanessaE when talking "pixels" I mean the rendered map, so 1 pixel == ~1 node
04:36 VanessaE in any case the map doesn't move perfectly fluid for me, but it's not a huge jump either.
04:38 VanessaE it's mostly printing D 0 0 0 ... sometimes 0 0 (1 or -1) while I walk or fast fly
04:39 * VanessaE turns shaders on..
04:41 est31 ok its about as same as bad as for without shaders
04:42 VanessaE mostly -1 to +1 there, but LOTS more of non-zero values; occasionally it hits -2 or +2
04:42 est31 I have it going to 10 at most times, then jumping to 0 again
04:42 est31 you are on x1?
04:42 VanessaE maybe a CPU and GPU speed issue?
04:43 VanessaE yes, x1 mode
04:43 est31 perhaps. whats your fps?
04:45 VanessaE in the "benchmark scene" RBA and I have been using, 34 fps @ 87.6m range (drawtime of 28-29) with the minimap off.  With it on, 34 fps @ 72.9m, drawtime of around 28.
04:46 VanessaE (the minimap costs me about 15m of view distance, or about 5 or so fps if I lock the distance)
04:46 est31 I have even ~57 fps occasionally
04:46 VanessaE (maybe 10 fps, need to test)
04:47 est31 when I move, its around 60, but when I stand still, it falls down to ~30
04:47 est31 very weird
04:48 VanessaE ok, 102 on the drawtime with the minimap off, if I lock my view range to 245m. 122 with the minimap on in that state.
04:48 VanessaE so a 20% cost for me :-/
04:48 est31 hmmmm, can caves really cut into jungles, like here: http://pasteboard.co/1wJ4gPdi.png
04:50 hmmmm est31:  I think that's caused by paramat's recent is_ground_content patch
04:51 est31 ok
05:09 est31 ahhh
05:09 est31 I seem to know why it doesnt work
05:09 est31 lemme try sth
05:25 est31 wow this irrlicht is weird
05:25 est31 if you do matrix.setTextureScale(3.0,1.0);
05:25 est31 then it works for the quadratic, and doesnt for the round
05:26 est31 now when I remove the rotate call for the round one, it works again
05:26 est31 just why is it overriding?
05:27 err404 joined #minetest-dev
05:38 Calinou joined #minetest-dev
05:41 Megaf_ joined #minetest-dev
05:41 err404 left #minetest-dev
05:49 paramat joined #minetest-dev
05:51 paramat hmmmm yes i should have added: it's too crazy, using this location i will tune parameters and limit the highest mountains to around 512, similar to the highest summits found before in mgv7
05:53 paramat est31, your screenshot of chopped jungles is bemusing, in my recent commit i set trees to be ground_content = false to avoid that, you're using latest mtgame?
05:53 est31 no minimal
05:53 paramat oh
05:53 est31 and it was most likely freshly generated
05:54 paramat okay yes my fault
05:55 paramat will fix
05:55 paramat when i last worked on minimal i edited some ground_content settings
06:00 paramat however tree was set as ground content true before my commit, minimal is a rather neglected mess
06:02 paramat it's a bit frustrating to work on because there's so much updating needed but i don't know how much i should do because it is after all 'minimal'. for example jungletrees and pines don't drop saplings etc..
06:05 paramat then when i do work on it i feel like i should be doing more important things, because it's such a mess =)
06:21 asl97 joined #minetest-dev
06:45 Jbb joined #minetest-dev
06:58 paramat left #minetest-dev
07:11 nore joined #minetest-dev
07:37 Hunterz joined #minetest-dev
07:40 kilb joined #minetest-dev
07:41 asl97 joined #minetest-dev
07:42 kilbith joined #minetest-dev
07:53 Krock joined #minetest-dev
08:04 Yepoleb joined #minetest-dev
08:29 Hunterz hi, please is there any windows build with latest commits (minimap) ?
08:30 VanessaE probably not yet
08:30 VanessaE minimap just went in some hours ago
08:30 est31 at least not here: https://forum.minetest.net/viewforum.php?f=42
08:30 Hunterz I know :) its very nice
09:40 SopaXorzTaker joined #minetest-dev
09:50 MinetestForFun joined #minetest-dev
10:00 Krock Yay! Minimap *recompiles*
10:31 Krock Hunterz, there's a windows build with minimap now.. I'm testing it right now
10:33 Krock somehow it takes years to load the item definitions
10:43 kilbith joined #minetest-dev
11:04 selat joined #minetest-dev
11:07 Krock joined #minetest-dev
11:08 Calinou joined #minetest-dev
11:24 kilbith joined #minetest-dev
12:07 Puma_rc joined #minetest-dev
12:12 RealBadAngel Krock, it can take a bit longer, minimap needs data from all the nodes
12:13 Krock RealBadAngel, I've set enable_minimap to false to see if it's the minimap but it doesn't change anything
12:13 RealBadAngel whats your game?
12:15 RealBadAngel ah, indeed, i havent disabled fetching the colors with minimap disabled
12:15 RealBadAngel i will push fix for that ASAP
12:15 Krock could that color loading cause a packet time-out? I doupt that..
12:16 RealBadAngel no way
12:16 RealBadAngel its only client side and after media received
12:16 Krock okay, thanks. perhaps there's an other fail
12:16 RealBadAngel so the delay you have must be something else
12:17 RealBadAngel when i tested the time cost it was like 4-5s longer for dreambuilder game
12:17 est31 joined #minetest-dev
12:18 RealBadAngel hi est31, any luck with textures shifting?
12:18 est31 nope
12:18 FR^2 joined #minetest-dev
12:20 RealBadAngel i can think of another solution for you
12:20 RealBadAngel when moving scanning code (most of it) to mapblock mesh, ive decided to double scanning height
12:21 RealBadAngel you could propably try old values
12:22 RealBadAngel https://github.com/minetest/minetes​t/blob/master/src/minimap.cpp#L283
12:22 RealBadAngel second value is scan height, try setting it to 128
12:29 est31 RealBadAngel, its faster, but far from good
12:30 RealBadAngel ok, lets try another thing
12:30 RealBadAngel https://github.com/minetest/minetes​t/blob/master/src/minimap.cpp#L141
12:30 RealBadAngel uncomment that and see if it changes anything
12:32 est31 I've set it to 1000 and its gotten slower
12:33 RealBadAngel i mean some low value
12:33 RealBadAngel not 1k :) try 1
12:33 est31 btw that loop definitely needs improvement
12:33 est31 nope not faster
12:34 RealBadAngel ok
12:34 Krock btw, est31, why did you merge that commit with master? RealBadAngel could do it aswell
12:34 Krock just wonder
12:34 est31 Krock, RBA had evil \r\n line endings
12:35 Krock that depends on the git settings, they can be set, so they commit as \n and read as \r\n
12:35 RealBadAngel what should i set?
12:35 est31 yes and a repo can override those
12:35 est31 which minetest does
12:36 est31 this is the strange part
12:43 est31 RealBadAngel, list of things hmmm and I havve seen are bad. #2844
12:43 ShadowBot https://github.com/minetest/minetest/issues/2844 -- Minimap code and behaviour problems
12:43 est31 bye for 2day
12:44 est31 joined #minetest-dev
12:46 Krock lol
12:58 RealBadAngel well, i just measured how long get map takes
12:58 Krock is it a pain to look at that number?
12:59 RealBadAngel im not sure if im seeing it right
13:01 RealBadAngel yeah, its a bit different, hold on
13:02 RealBadAngel 13-14ms
13:02 RealBadAngel thats how long full scan in surface mode takes for me
13:03 RealBadAngel so theoretically, on my box minimap could run at 77fps
13:04 RealBadAngel and in practice minimap updates are faster than overall FPS
13:04 RealBadAngel 3,1Ghz dual core here
13:06 RealBadAngel i need some feedback and others to test the time on different CPUs
13:08 RealBadAngel est31, come back into the channel :)
13:11 kilbith i use an old single core CPU, showing the minimap doesn't impact the FPS at all and the map is shown as soon as the mapblocks are generated
13:12 RealBadAngel can you add some testing code to the minimap?
13:13 kilbith sure
13:13 RealBadAngel http://pastebin.com/tP5ukU2d
13:13 RealBadAngel https://github.com/minetest/minetes​t/blob/master/src/minimap.cpp#L135
13:14 kilbith the output is shown in the console ?
13:14 kilbith i don't see a 'cout'
13:14 RealBadAngel yes, its dumped to the console
13:19 RealBadAngel kahrl, nore, Calinou here?
13:35 RealBadAngel anybody to help me testing code above?
13:40 * Krock tries to find the source of #2842
13:40 ShadowBot https://github.com/minetest/minetest/issues/2842 -- RE-SENDING timed-out RELIABLE to 127.0.0.1
13:41 RealBadAngel Krock, can you apply the code i pasted? i need feedback
13:42 selat joined #minetest-dev
13:43 Krock RealBadAngel, sure, I must check an other build first
13:43 RealBadAngel thx
13:44 red1 joined #minetest-dev
13:44 Krock gawd, if there just would be a faster way to link those .obj files.. no mater if just 1 line was changed or not, it always takes 5 minutes or longer just to create an exe
13:45 kilbith .obj files are meshes
13:45 Krock .obj are generally just binary files
13:53 kilbith http://pastie.org/10261678
13:53 kilbith ^ with crappy old CPU
13:53 Krock kilbith, define "crappy old CPU"
13:54 kilbith intel celeron 900, single core
13:54 Krock k
13:55 red1 joined #minetest-dev
14:05 Krock RealBadAngel, where exactly should those messages appear?
14:05 RealBadAngel in console when minimap is on
14:06 RealBadAngel copy for me circa 500 lines and pastebin them
14:06 Krock I enabled the minimap but when I joined the singleplayer world after waiting 2 minutes, there was no minimap and no debug text
14:06 Krock blue sky and flying at 0, 0.5, 0
14:08 Krock perhaps I should un-comment the line "sleep_ms(10);"
14:08 RealBadAngel no leave it commented out
14:09 Krock so what should I do? I can't play in any singleplayer world
14:10 RealBadAngel weird
14:10 kilbith fresh new git clone maybe
14:10 Krock unicode is broken again \o/ öäü are gone
14:10 RealBadAngel i think so
14:11 RealBadAngel Krock, indeed, try from scratch
14:11 Krock RealBadAngel, already tried that - doesn't help
14:11 Krock *didn't help
14:12 RealBadAngel i think that you already checked minimap and it worked?
14:13 Krock that's wrong
14:13 Krock I tried to find the fail in my compiler but now I think something's messed up in the code
14:14 RealBadAngel you are trying to compile unmodified master right?
14:15 Krock yes and it compiled fine
14:16 RealBadAngel but?
14:16 Krock as already said, joining a world takes minutes and when I get in, the terrain doens't load
14:17 RealBadAngel im checkin master now too
14:17 kilbith certainly windows-specific issue then
14:17 kilbith on linux everything's fine
14:18 Krock I tested c6766b2 from Jun 23, (without building from scratch) and the problem didn't exist anymore
14:18 Krock yes, perhaps a porting issue
14:19 kilbith too bad you've no verbose
14:19 Krock I got minetest's verbose mode.. doesn't help very much
14:21 est31 joined #minetest-dev
14:23 RealBadAngel est31, whats your cpu?
14:23 est31 hardware is pretty recent
14:23 est31 from this decade
14:23 est31 AMD Phenom(tm) II X6 1055T
14:24 est31 Krock, have you pressed f9?
14:25 RealBadAngel est31, yet its 25 times slower than my 2 core intel
14:25 RealBadAngel such difference is not the matter of the scanning algorithm i used
14:25 est31 that cant be
14:25 kilbith est31, my CPU is like 800% slower than your and performs twice more better on minimap, how possible ?
14:26 Krock est31, oooh.
14:26 RealBadAngel i do get 17ms average, kilbith 193ms (single core). and you with amd 6cores get 427ms
14:27 Krock well, ther are no nodes to render, so it's no wonder why the times are that tiny http://pastebin.com/tcKCgRU9
14:27 est31 rule confirmed the better your hardware, the worse your experience
14:27 RealBadAngel but to be sure we need more testers
14:27 est31 RealBadAngel, whats your setup?
14:27 est31 square map?
14:27 RealBadAngel for me it looks like amd threading is fucked up
14:27 RealBadAngel shape doesnt matter
14:27 RealBadAngel map is always a square
14:28 est31 also is the map always loaded?
14:28 RealBadAngel what we do measure now is performace of a routine in a thread
14:28 RealBadAngel forget fps, shape and everything else
14:29 RealBadAngel we should get similar results,  i do have 3,1ghz cpu vs 2,8ghz. 2 cores vs 6 cores
14:29 RealBadAngel intel vs amd
14:29 RealBadAngel 17 ms vs 423
14:30 est31 intel is always better, and you have larger ghz
14:30 RealBadAngel by 300mhz
14:30 est31 but that only accounts for 3x-6x factor
14:30 est31 I guess
14:30 est31 you have windows RealBadAngel, dont you?
14:30 RealBadAngel looking at time? i should have 70 Ghz cpu
14:31 kilbith frequence is not all, a recent Atom at 1200 Mhz is more powerful than my old celeron at 2200 mhz
14:31 RealBadAngel ^^  that was a joke
14:31 celeron55 stop hilighting me god damnit
14:31 RealBadAngel est31, atm only linux
14:31 kilbith oops, sorry
14:31 celeron55 or alternatively make it so that people don't call me "celeron" so that i can disable that hilight
14:32 est31 celeron55, what have you set to ping you?
14:32 est31 c55 too?
14:32 celeron55 c55 doesn't
14:32 kilbith i dunno how i can name my CPU differently than celero*
14:33 celeron55 it's not worth mentioning
14:33 celeron55 stop talking about it!
14:33 kilbith ok sorry
14:33 celeron55 ok good 8-)
14:33 RealBadAngel actually we have to talk bout processors
14:33 RealBadAngel looks like threading doesnt work for amd (just a guess)
14:33 celeron55 well i guess i'll disable the hilight then
14:33 celeron55 done
14:34 est31 RealBadAngel, before saying its cpu fault, perhaps we should try to ensure its no code fault?
14:34 est31 you run windows, right?
14:34 RealBadAngel no, linux here
14:34 est31 next thing: do you try the map in singleplayer?
14:35 est31 which db backend?
14:35 RealBadAngel and i havent said its cpu fault. it start to look like threading code fault
14:35 RealBadAngel est31 all those things are irrevelant to this code
14:35 RealBadAngel i just want to know its execution time, thats all
14:36 RealBadAngel it is working on data stored in ram. runs in thread. end of conditions
14:36 est31 so, confirm your theory
14:36 est31 but its not always hw faul
14:36 est31 t
14:37 RealBadAngel atm i dont check anything else just how long it takes to execute this very specific function
14:37 est31 so lets test other things too
14:37 RealBadAngel how it gathered needed data, whats the weather outside doesnt matter
14:38 RealBadAngel i just wanna know why your amd is 25 times slower than my intel
14:39 est31 ok
14:39 est31 RealBadAngel, can you confirm something for me
14:39 RealBadAngel sure
14:40 est31 can you open your cpu overview, and search for whetrher a cpu is at 100% when minimap is activated
14:42 RealBadAngel 70-75%
14:43 est31 because I get 100% CPU even when minimap is turned off
14:43 est31 I do nothing
14:43 est31 nothing at all
14:43 est31 and its only since the minimap commit
14:43 RealBadAngel minimap adds another thread
14:44 RealBadAngel it works no matter you are sitting duck or flying
14:44 RealBadAngel same as the other threads
14:44 est31 do you have the 10 ms line commented in?
14:44 RealBadAngel its inactive
14:45 RealBadAngel i do have an idea
14:45 RealBadAngel lets run a simple loop in a thread
14:46 RealBadAngel doing just a simple fun math inside
14:46 est31 thats the result of your while (true) stuff
14:46 RealBadAngel what are you talkin about?
14:46 est31 and imho a pretty serious regression
14:46 RealBadAngel then you should be able to play at all since ages
14:46 RealBadAngel *nt
14:47 est31 one of my CPU cores is at 100%
14:47 RealBadAngel its how mapblock mesh updates are done
14:47 RealBadAngel since always
14:47 RealBadAngel but yet you dont complain on meshes
14:47 est31 constantly, even if I don't have minimap activated
14:47 est31 lets see how its done there
14:47 RealBadAngel the very same way
14:47 est31 you have the file
14:48 est31 ?
14:48 RealBadAngel open client.cpp and look at the start of the file
14:48 RealBadAngel find mapblock mesh update thread
14:48 RealBadAngel that while is made to empty the queue
14:48 RealBadAngel it doest stop the thread or something
14:53 est31 its using MutexedQueue
14:54 RealBadAngel where?
14:55 RealBadAngel i can see std::vector :P
14:55 RealBadAngel the very same one you and hmmm dont like
14:58 est31 ok
14:59 est31 but still there is a 100% CPU
14:59 est31 with minimap turned off
14:59 RealBadAngel threads are running in loop
14:59 RealBadAngel doing nothing also takes time
14:59 est31 but you agree 100% CPU is bad isnt it?
14:59 est31 for doing nothing
15:00 RealBadAngel even when doing nothing cpu is always busy
15:00 RealBadAngel doing nothing is a command
15:01 RealBadAngel but still, we are not talking on the subject
15:02 RealBadAngel i need to test something. you will have to test the same
15:02 est31 agreed
15:05 est31 ShadowNinja, you there?
15:12 deltib joined #minetest-dev
15:31 est31 but better some minimap than no minimap
15:32 est31 I am writing a patch adding semaphores
15:32 est31 which will remove the 100% issue
15:32 est31 not the slow drawing issue
15:33 ShadowNinja est31: Yep.
15:34 est31 ShadowNinja, I wonder how the MeshUpdateThread works without running hot, by re-requesting and re-requesting again
15:35 red1 how can I kick all players on server crash?
15:36 est31 that doesnt work
15:36 est31 for regular shutdowns, try this mod: https://github.com/est31/kickall
15:36 red1 ok
15:37 est31 but there is no feature for triggering non-regular shutdowns
15:37 RealBadAngel ShadowNinja, https://github.com/minetest/minetest/issues/2844
15:37 RealBadAngel what do you think about difference in execution time?
15:38 ShadowNinja RealBadAngel: IIRC you have a pretty good gfx card.
15:38 RealBadAngel we are not talkin GPU now
15:38 ShadowNinja CPU then.
15:38 RealBadAngel the measured thing works in thread, on data stored in ram and thats cpu
15:39 RealBadAngel look at kilbith, est31 and mine results
15:39 ShadowNinja You might have different settings since yours is so much faster.
15:40 RealBadAngel settings doesnt matter here
15:41 RealBadAngel it will be the same for any kind of game, textures used and weather outside
15:41 est31 really RealBadAngel?
15:41 RealBadAngel what counts here is pure cpu speed
15:41 est31 even if I say that i have disabled mesh cache?
15:41 RealBadAngel ofc
15:41 red1 how can i dectect an error from lua?
15:42 RealBadAngel its not butterfly effect ;)
15:43 RealBadAngel you could propably instert there simple loop repated 1kk times and measure that
15:43 RealBadAngel and you still will ask one answer, why AMD is slower
15:43 RealBadAngel * question
15:46 RealBadAngel ShadowNinja, can you test that too?
15:47 est31 also, I get one CPU core with 100%, even  when minimap is disabled
15:47 est31 can you test that too?
15:49 RealBadAngel another thread are here
15:49 RealBadAngel minimap's one is just one of them
15:49 hmmmm joined #minetest-dev
15:49 RealBadAngel you have saw that mapblock mesh updates thread works the very same way
15:49 est31 I dont know whats happening but I dont get it for the other threads
15:50 est31 You are right, there are almost no differences
15:50 est31 but thats not the way to code
15:50 est31 its busy polling
15:50 * est31 guesses we need a new queuedworkerthread class
15:51 RealBadAngel propapbly, my knowledge bout threads is next to nothing
15:51 RealBadAngel i just copied for minimap one we are already using
15:52 est31 ok then you arent to blame
15:52 est31 but I'm fixing it for the minimap first
15:52 est31 and then I do that class
15:52 hmmmm so who was the second person to +1 the minimap feature
15:52 RealBadAngel and it looks like everythin is fine for Intel cpus
15:52 hmmmm just curious
15:52 est31 hmmmm, it was me
15:52 hmmmm who was the first person to +1 it
15:52 est31 now I clean up (part of the) mess
15:52 est31 RBA
15:53 hmmmm ... that
15:53 est31 "mess"
15:53 hmmmm is definitely not what was intended
15:53 hmmmm i wouldn't agree to something so unstable being added to minetest
15:53 est31 me neither if I knew it back then
15:53 hmmmm this is such a massive performance regression for something so incredibly trivial
15:54 hmmmm you shouldn't be comparing cpus
15:54 hmmmm "oh well it's only 70% on my i7"
15:54 est31 there are two issues here
15:54 hmmmm it should be completely unnoticable on any cpu later than athlon 64
15:54 est31 the first one being why its slow for me
15:54 hmmmm if it's not there's a problem
15:54 RealBadAngel hmmm, no its working fine on single core intel cpu
15:55 est31 and the second one why there is 100% CPU
15:55 RealBadAngel always was
15:55 est31 the first one has a big question mark
15:55 est31 no idea why
15:55 kilbith there was always 100% CPU
15:55 RealBadAngel because of mapblock mesh update thread
15:55 est31 the second one, well I'm writing a patch for it right now
15:55 RealBadAngel new thread doesnt change anything
15:55 est31 mapblock mesh update thread does that horrible thing too, but somehow it doesnt get cpu to 100%
15:56 RealBadAngel and it itself is not a regression
15:56 est31 at least for me
15:56 RealBadAngel it just shows that threading for amd cpus is broken at all
15:56 * est31 stops debating whether its a regression or not and continues writing the patch
15:56 RealBadAngel agree
15:57 hmmmm "threading for AMD cpus"
15:57 RealBadAngel yes
15:57 hmmmm you're trying to minimize the problem
15:57 RealBadAngel VanessaE reported the same, long time before minimap was there
15:57 hmmmm AMD gpus
15:57 RealBadAngel no, im not minimizing it, im looking for a solution
15:57 hmmmm good coders would admit there's a problem and work on figuring it out rather than denying
15:58 est31 ^
15:58 RealBadAngel hmmmm, i just figured out that it depends on CPU kind
15:58 hmmmm it doesn't.
15:58 hmmmm that's a bunch of bullshit.
15:58 RealBadAngel what the code does its not important
15:58 RealBadAngel only matters it runs in a thread
15:59 RealBadAngel damn replace whole getMap code, throw it and put there 1kk loop
15:59 RealBadAngel you will get the same effect
16:02 RealBadAngel hmmmm, what CPU do you have?
16:02 RealBadAngel can you run the same test?
16:08 ElectronLibre joined #minetest-dev
16:08 est31 ok my theory wasn't right, my patch doesnt help
16:09 RealBadAngel est31, what have you tried?
16:12 est31 RealBadAngel, https://github.com/est31/minetest/commit/​f33cd5261bec9c94437a87213e24907b4c450710
16:13 est31 either way I think that should be pushed
16:13 est31 because its the right way to do it
16:13 RealBadAngel it couldnt had any impact on performance
16:14 RealBadAngel as queue has usually just a few entries per second
16:14 est31 yes but the for loop will rage on either way
16:14 est31 so that has been optimized
16:15 RealBadAngel it just collects data which another thread has made
16:15 RealBadAngel data transferred are as small as v3s16 and a pointer
16:15 est31 thats totally irrelevant
16:15 est31 a while true stays a while true
16:16 RealBadAngel time spent on it is usually 30-40ns
16:16 est31 I made it a while true { block()
16:16 est31 RealBadAngel, you still dont get it?
16:16 RealBadAngel yes, i get it
16:17 RealBadAngel but still it has no impact on performance
16:18 est31 you dont get it.
16:24 RealBadAngel i may not. what i can see it is doing just the same
16:24 est31 ok then I try to explain
16:25 est31 do you know semaphores?
16:25 est31 they are a data structure to communicate between threads
16:25 est31 basically, its an int
16:25 est31 you have two operations: increment, and decrement
16:25 est31 now the increment works instantaneously, and the decrement too, unless the int is 0
16:26 est31 if its 0, it will wait until the int is 1 (by another thread incrementing the semaphore), and then decrement
16:26 est31 so what people can use is that "wait" part
16:26 est31 its ideal for queues
16:27 est31 every time you add to the queue, you increase, every time you remove, you decrease
16:27 est31 and when you can't remove, you wait until stuff comes in
16:27 est31 and not spend CPU power checking the same value again and again
16:28 hmmmm maybe it's time to check out what's going on under valgrind
16:28 est31 yup
16:28 est31 5 mins ago I've started callgrind
16:28 est31 now its at initializing nodes 61%
16:28 est31 64$
16:29 RealBadAngel ok
16:29 RealBadAngel i will try to optimize getMap itself to work faster
16:30 RealBadAngel math here is pretty complex because i do have data in chunks, not as before got pixel by pixel
16:30 est31 so! I have a scene
16:31 Puma_rc joined #minetest-dev
16:31 est31 now lets let it render into nothingness for some time
16:31 RealBadAngel and hmmmm, getting a full scan of 8,5kk nodes real time is not trivial
16:31 est31 ok now I have it
16:31 TeTpaAka joined #minetest-dev
16:32 TeTpaAka I also get around 400 ms on Intel.
16:33 RealBadAngel TeTpaAka, whats your cpu?
16:33 TeTpaAka I posted it on github
16:33 TeTpaAka Its a core i5
16:33 est31 can you reproduce the "big 100%" bug, where one cpu core is at 100%, with minimap turned off?
16:33 TeTpaAka Around 5 years old
16:33 est31 (and not there before)
16:34 RealBadAngel TeTpaAka, thx
16:35 RealBadAngel est31, if you would like remove busy polling from threads you should modify other threads too
16:35 est31 RealBadAngel, agreed
16:35 RealBadAngel start with mesh update thread
16:36 RealBadAngel client.cpp line 165
16:36 RealBadAngel i will look if we are using such mechanism elsewhere
16:37 Puma_rc joined #minetest-dev
16:40 TeTpaAka Confirmed, one core is always at 100%, even in pause menu, which wasn't the case before.
16:40 RealBadAngel est31, httpfetch.cpp line 641
16:40 est31 ok, callgrind sais it is spending 60% of its time inside minimapupdatethread::thread()
16:41 est31 inclusive
16:41 CraigyDavi joined #minetest-dev
16:42 Puma_rc joined #minetest-dev
16:42 est31 out of some reason the size requesting plays a big role
16:44 Puma_rc_ joined #minetest-dev
16:45 RealBadAngel est31, maybe that sleep should be there to free time for others?
16:45 est31 thats what the semaphore is for
16:46 est31 its sleeping until something happens
16:46 est31 much better than sleeping a bit, checking, sleeping again, checking again
16:47 RealBadAngel have you checked cpu usage with you patch then?
16:47 est31 yes, and I wonder why it didnt improve
16:48 RealBadAngel <est31> ok, callgrind sais it is spending 60% of its time inside minimapupdatethread::thread()
16:48 RealBadAngel it was with or without the patch?
16:48 est31 with the patch
16:49 RealBadAngel can you check how much takes meshupdatethread?
16:49 RealBadAngel it is using sleep
16:49 est31 11
16:49 VanessaE RealBadAngel: told you it was thread-related. :)
16:50 RealBadAngel VanessaE, hi
16:50 VanessaE hi
16:50 RealBadAngel you were getting 100% load before minimap, right?
16:51 VanessaE variably.
16:51 VanessaE but you already know turning the minimap on costs me a good 5 or so fps when it should theoretically be immesurable.
16:51 VanessaE (well 5+ fps or 10-15 m draw distance, depending on which is capped)
16:51 RealBadAngel that 5fps cost is preparing new textures, and is done in main thread
16:52 VanessaE but it's nearly always stuck using up a single core
16:52 RealBadAngel so its not the cause of cpu load
16:52 VanessaE (i.e. almost never expands into the other cores)
16:53 est31 ahhh
16:53 est31 I'm stupid
16:53 est31 darmmm
16:53 Puma_rc joined #minetest-dev
16:53 VanessaE ?
16:53 est31 grr
16:53 est31 I did the wait call inside the second while
16:53 est31 where it makes totally no sense
16:53 est31 because then its guaranteed already there is an element in there
16:54 est31 ah yes
16:54 est31 ok
16:55 RealBadAngel so, lets wait again for that grind thingy
16:55 est31 now it seems to work
16:55 est31 no 100% CPU
16:55 RealBadAngel but?
16:55 est31 nothing but
16:55 est31 CPU is calm again
16:55 RealBadAngel whats the load now
16:55 RealBadAngel that but
16:55 est31 big regression is gone
16:56 RealBadAngel can you apply the same to meshes?
16:56 est31 yes
16:56 est31 while it would be interesting why I can't see any issues there
16:57 RealBadAngel it sleeps
16:57 est31 that has no difference
16:57 RealBadAngel propapbly thats why its freeing time for others
16:57 est31 I've removed the sleep call, and it worked still
16:57 RealBadAngel but no freeing time anymore
16:59 RealBadAngel but, does freeing cpu time had any impact on getMap execution time?
16:59 RealBadAngel the test we are doing before
16:59 RealBadAngel were
16:59 est31 lemme see
17:00 est31 no
17:00 est31 and it was unlikely
17:00 est31 because the CPU is only occupied if it has no tasks
17:00 RealBadAngel there has to be reason why code is 25 times faster for me
17:01 RealBadAngel cpu speed is comparable to yours
17:02 VanessaE time spent communicating with the GPU, I'd guess.  you issue a GPU call, it blocks for us but not for you? so for you you can go on and do something else right after.
17:03 RealBadAngel VanessaE, this code is not gpu related
17:03 RealBadAngel its pure cpu and memory ops
17:03 VanessaE hm
17:03 VanessaE *shrug* not sure then
17:03 RealBadAngel VanessaE, you have the same cpu as est31, can you run the test too?
17:04 VanessaE sure.  link to the correct patch?
17:04 RealBadAngel grab the mainstream and open minimap.cpp
17:04 VanessaE open.
17:05 RealBadAngel https://github.com/minetest/minetest/issues/2844
17:05 RealBadAngel here you have code in my comment
17:05 VanessaE https://github.com/minetest/minetest​/issues/2844#issuecomment-116043264
17:06 VanessaE that?
17:06 est31 btw RealBadAngel there is autmatically average generating code in this commit : https://github.com/minetest/minetest/commi​t/0b6b87154503f35d1d1cfd55c4352ff24efed4bb
17:06 est31 its easy to adjust to print the sample size too
17:06 RealBadAngel VanessaE, yes
17:07 RealBadAngel VanessaE, it doesnt matter you will try it with minimal or dreambuilder
17:07 RealBadAngel times will be the same
17:07 VanessaE ok
17:07 VanessaE building...
17:07 RealBadAngel est31, in this commit you have messed with texture matices
17:08 est31 yes but no visible change
17:08 est31 also, its another thread
17:09 VanessaE RealBadAngel: ok, what am I supposed to do?
17:09 RealBadAngel turn minimap on
17:10 RealBadAngel and copy like 500 lines from console and pastebin it to me
17:11 VanessaE http://pastebin.com/GCWbRDyT
17:11 VanessaE looks like it's averaging around 27ms
17:11 RealBadAngel rotfl
17:11 VanessaE this is an AMD Phenom II X6 1055T (2.8 GHz, 6 cores)
17:11 RealBadAngel and thats funny
17:12 VanessaE default window size, default textures, and I think I had shaders turned on.
17:12 RealBadAngel my theory on vendor related problems is officially thrown into dustbin
17:12 VanessaE lol
17:12 RealBadAngel now we had one intel i5 with 400ms
17:13 RealBadAngel and same kind amds one 27ms and another 427
17:13 VanessaE fwiw, the time went up a bit while I was turning around slowly to let the whole minimap fill up with loaded blocks.
17:13 VanessaE (into the low 30's)
17:14 RealBadAngel est31, how can you explain that? she has the same cpu as you do
17:15 est31 its obviously, like I and hmmmm std::deque
17:15 est31 errr
17:15 est31 like i and hmmm said no cpu problem
17:15 VanessaE I always compile with -O3, in Release mode, and always a full compile (make clean; rm CMakeCache; cmake . -Dblahblah... ; make -j8)
17:16 est31 perhaps we had a dust of grain during cpu production, while you people had none?
17:16 VanessaE I doubt it.
17:16 est31 I compile debug mode
17:17 VanessaE I still suspect a GPU-threading issue; race conditions resulting in a spinlock somewhere maybe
17:17 RealBadAngel its weird
17:18 est31 grain of dust*
17:18 est31 wow sometimes I speak funny stuff
17:18 VanessaE est31: right, I doubt there's a speck of wheat in your CPU ;)
17:19 RealBadAngel est31, anyway i think polling patch is not related to that issue and should be merged
17:19 RealBadAngel hmmmm, do you agree on that?
17:21 est31 https://github.com/est31/minetest/commit/​420125debd3d010fcd3e7738c43f25c8f407ab63
17:23 TeTpaAka joined #minetest-dev
17:24 TeTpaAka Mine was a debug build. Now, as release, I get 29,5
17:24 VanessaE est31: ^^^ interesting.
17:24 est31 yea
17:24 est31 ahaaaaa
17:24 VanessaE ?
17:24 est31 lemme try
17:24 est31 too
17:24 VanessaE also Debug mode compiles with -O1 doesn't it?
17:25 RealBadAngel TeTpaAka, 29,5 is the average time for my test?
17:25 TeTpaAka Yes, over 316 samples
17:26 RealBadAngel est31, how do you build your game?
17:26 est31 debug
17:26 RealBadAngel :PPPPPPPPP
17:26 RealBadAngel i think i should kick you in the ass ;)
17:27 est31 hah
17:27 RealBadAngel because of you cpu vendors were blamed :P
17:27 est31 I didnt raise the blames
17:27 est31 still I notice lag
17:27 est31 only occasionally
17:27 RealBadAngel timings please
17:27 est31 but larger than before
17:27 est31 umm lemme see
17:28 Wuzzy joined #minetest-dev
17:29 VanessaE btw it's ~28.79 ms over 604 samples, if you want an exact figure.
17:29 est31 the problem is it doesnt show up in the statistics
17:29 RealBadAngel i mean the test for getMap
17:29 Wuzzy Any comments on <https://github.com/minetest/minetest/pull/2438>? Many people complained the tooltips were too long, but I don't know how to shorten them any further without losing information. :-/
17:29 RealBadAngel wuzzy, break lines
17:30 RealBadAngel tooltips support multiline texts
17:30 Wuzzy ok, and when? After 80 chars or so?
17:30 RealBadAngel just use /n
17:30 Wuzzy yes, sure
17:30 est31 lag is most times > 1 second
17:30 Wuzzy Would be a line break after 80 chars ok for you?
17:31 RealBadAngel Wuzzy, its not automatic, you decide when to break the line
17:31 Wuzzy That's what I am saying ...
17:31 RealBadAngel btw, youre going to put book chapters in tooltips or what? ;)
17:32 est31 yes
17:32 RealBadAngel est, so what timings you got?
17:33 est31 all < 50
17:33 est31 enough to make me satisfied :)
17:33 VanessaE that's still rather slower than mine
17:33 RealBadAngel thats acceptable
17:34 RealBadAngel est31, now change scan height to 128
17:34 VanessaE est31: what else you got on that box?
17:34 VanessaE (that could impact your speed)
17:34 est31 onboard gfx
17:34 RealBadAngel indeed, thats 2x slower than the very same cpu
17:34 RealBadAngel est31, for that gpu doesnt count at all
17:35 est31 RealBadAngel, with < 50 I can also mean average of 10
17:35 RealBadAngel copy like 500 lines for me
17:35 RealBadAngel i will get an average
17:36 RealBadAngel TeTpaAka, btw, thx for solving the issue :)
17:36 est31 http://pastebin.com/n5VZW97V
17:36 TeTpaAka No Problem. It was VanessaEs hint that made me think of the build type.
17:36 RealBadAngel otherwise we could be sitting whole day lookin for solution lol
17:36 est31 so the main problem is now that there is occasional, but larger lag
17:36 est31 > 1 second
17:36 est31 at least all 10 seconds once
17:36 est31 and it doesnt show up in those stats
17:37 VanessaE est31: this sounds like the main mesh thread lag I get from time to time
17:37 VanessaE the whole thing just.... stalls... for about one second, not really sure what the pattern is
17:37 VanessaE (most noticable with HDX, but I've seen it happen with defaults too)
17:38 RealBadAngel 39ms average
17:38 RealBadAngel that should be enough for 25fps minimap
17:40 RealBadAngel est31, can you check times with scan height set to 128?
17:40 RealBadAngel im thinkin on makin a setting for that, for slower boxes
17:41 VanessaE what you should be doing here is timing how long each map update takes and auto-tuning the scan height accordingly.
17:41 RealBadAngel no
17:41 RealBadAngel scan time will highly depend on terrain
17:42 VanessaE hence auto-tuning
17:42 VanessaE if it can be done (poorly) with the in-game terrain, it can be done with the minimap as well
17:42 RealBadAngel that would make things appear and dissapear without reason
17:42 est31 ok, can I merge the 100% CPU away commit now?
17:43 VanessaE not if you apply a bit of slop to the rate at which it tunes.
17:43 RealBadAngel you have my +1, one more is needed
17:43 est31 its an early hotfix for a more proper solution
17:43 est31 why one more?
17:43 est31 thats three core maintainers needed or what?
17:44 RealBadAngel i thought 2 apart from author
17:44 RealBadAngel but anyway its bugfix
17:44 VanessaE for a regular patch I understood it to require two +1's from other devs, but for a quick bugfix I guess that requirement is relaxed somewhat.
17:45 est31 pushed
17:45 RealBadAngel also please do mention in the issue you created that most regression was caused by debug build :)
17:46 RealBadAngel and please let me know how 128 scan height works for you
17:46 RealBadAngel i would like to know its worth to make a setting for that
17:47 est31 yes it works but not that large speedup
17:47 RealBadAngel the timings?
17:48 est31 yes
17:48 RealBadAngel btw, nice mod imho, gonna try it now: https://github.com/technomancy/calandria
17:53 RealBadAngel est31, i will code now the cosmetic changes, like m_ name problem and a setting for height so we can have minimap issues closed for now
17:53 RealBadAngel then since we could check VBO patch, wanna try?
17:53 VanessaE ooo gimme
17:53 est31 VBO?
17:54 Calinou Vertex Buffer Objects
17:54 Calinou makes for faster jungles
17:54 RealBadAngel not only jungles
17:54 VanessaE est31: keeps meshes on the GPU instead of continuously streaming them
17:54 RealBadAngel lotsa of vertices in general
17:54 est31 nice
17:54 est31 but today, I'm out
17:54 RealBadAngel that alone doubles fps in most cases
17:54 est31 bye
17:55 RealBadAngel :P
17:55 VanessaE RealBadAngel: have you figured out a good algorithm for discarding obsolete meshes?
17:56 RealBadAngel what for? removing them from hardware buffers?
17:56 VanessaE because if you don't discard obsolete meshes with VBO running, you'll run out of memory.
17:56 RealBadAngel it doesnt work that way
17:56 VanessaE it did before.
17:57 VanessaE that's precisely why the old VBO patch never got merged
17:57 RealBadAngel simply if you no longer need the mesh you shall flush it from the buffer manually
17:57 VanessaE that's my point
17:57 RealBadAngel either by checkin when it was last time used
17:57 VanessaE have you worked out when to flush? :)
17:57 VanessaE that's what I thought
17:57 RealBadAngel yes
17:57 VanessaE a mesh that's more than ~ 5 mins old is probably old enough to flush
17:57 RealBadAngel i do delete the mesh when block got unloaded
17:58 VanessaE that may be too long
17:58 VanessaE an active client can have thousands of loaded blocks.
17:58 RealBadAngel circa 1k
17:58 VanessaE I've seen it go somewhat higher than that in the distant past
17:58 VanessaE around 1800 if memory serves
17:58 RealBadAngel depends
17:59 RealBadAngel lets say no more than 2k
17:59 VanessaE how much will 2K of VBO-side meshes need?
17:59 RealBadAngel no idea and no way to check it
17:59 RealBadAngel irrlicht does that, its not our business
17:59 VanessaE well how big is a raw minetest mesh, at worst case?
18:00 VanessaE worst case I can think of is an entire 16^3 mapblock full of homedecor wine racks
18:00 RealBadAngel irrlicht tries to send it to VRAM, if it fails it keep using RAM one
18:00 RealBadAngel so telling irrlicht to use VBO is just a hint
18:00 RealBadAngel not a directive
18:00 VanessaE right
18:01 VanessaE I think before, the patch was actually doing the VBO stuff somewhat more.. "manually".
18:01 red-001 joined #minetest-dev
18:01 RealBadAngel hint is even in the name that turns it on
18:02 RealBadAngel setHardwareMappingHint(scene::EHM_STATIC);
18:02 red-001 how can I get connected players in c++
18:02 VanessaE heh
18:07 Krock Has anyone noticed the special character issue? A commit about converting strings or using them in formspecs remove öäü-like characters now
18:07 Krock *removes
18:08 red-001 anyone?
18:11 Krock yes, anyone?
18:12 Krock red-001, what do you plan to do?
18:14 VanessaE red-001:
18:14 VanessaE oops
18:14 VanessaE RealBadAngel: https://github.com/freeminer/freeminer/comm​it/d4a711869ff339724136e4059db799e19ef956e3
18:15 VanessaE there's PilzAdam's VBO patch as added to fm.
18:15 VanessaE (the repo that originally contained it has since been deleted)
18:15 RealBadAngel yes,  with all the flaws
18:15 RealBadAngel you can find the patch in kahrl's and SN's repos
18:16 VanessaE and this old comment from the related thread:  https://forum.minetest.net/vi​ewtopic.php?p=102847#p102847
18:16 VanessaE summary:  VBO plus a shader tweak doubled my fps at the time.
18:17 VanessaE (I'm sure the shader tweak no longer applies)
18:19 RealBadAngel its here
18:19 VanessaE right, I notice you wrote about that in a later post.
18:19 VanessaE point was to make sure the old VBO patch is still kept in mind
18:37 RealBadAngel #2848
18:37 ShadowBot https://github.com/minetest/minetest/issues/2848 -- Minimap update by RealBadAngel
18:37 RealBadAngel ok, im done with it, now will make PR for VBO
18:41 asl97 joined #minetest-dev
18:49 VanessaE RealBadAngel: meanwhile, see your /msgs
19:21 RealBadAngel hmmmm, here?
19:29 zat joined #minetest-dev
19:31 ElectronLibre joined #minetest-dev
19:58 Gethiox left #minetest-dev
20:07 Darcidride joined #minetest-dev
20:09 Darcidride_ joined #minetest-dev
20:28 neoascetic joined #minetest-dev
20:29 neoascetic Hi all. I want to track all map changes (including by players and by mods, abms etc). Is this possible? I mean, maybe some debugging level, or any other way
20:30 Krock minetest.register_on_placenode()
20:30 Krock minetest.register_on_dignode()
20:30 neoascetic Great, what with ABMs?
20:31 neoascetic I tried to use Redis' pub/sub but it doesn't give me a block id and data beign modified
20:32 asl joined #minetest-dev
20:32 neoascetic Ok, it seems I may patch "register_abm" and sniff all action changes... is there a better way?
20:48 Lunatrius joined #minetest-dev
21:12 selat joined #minetest-dev
21:18 asl97 abm using dig/place should trigger the function registed using register.on_placenode and rgister.on_dignode
21:19 asl97 oh wait, left already :|
21:25 daswort when was the fsaa option introduced?
21:26 asl97 Apr 3, 2012
21:26 asl97 https://github.com/minetest/minetest​/blame/master/minetest.conf.example
21:26 asl97 https://github.com/minetest/minetest/commi​t/e8e73d37fb1834208dabac207593030b87db135a
21:27 asl97 you can easily find out when something has been added/change using the blame feature on github
21:31 daswort thx, but why isn't there a gui switch if it is present for so long?
21:32 daswort *for fsaa
21:32 asl97 because it still Experimental?
21:34 daswort Ah thx again
21:34 asl97 idk
21:37 neoascetic joined #minetest-dev
22:00 chchjesus joined #minetest-dev
22:01 SudoAptGetPlay joined #minetest-dev
22:14 chchjesus joined #minetest-dev
23:25 prozacgod joined #minetest-dev
23:51 proller joined #minetest-dev

| Channels | #minetest-dev index | Today | | Google Search | Plaintext