Time  Nick       Message
17:43 hmmmm      alright
17:43 hmmmm      i'll try highlighting him
17:44 hmmmm      celeron55_:  what is the rationale behind making datatypes in noise double instead of float?  why do you add 0.5 to the X,Y position in noise2d_perlin calls?
17:45 celeron55_ the addition is because otherwise the nouse would be always zero at zero
17:45 celeron55_ noise*
17:45 celeron55_ or something like that
17:46 celeron55_ dunno about the doubles
17:46 celeron55_ it probably doesn't make much of a difference
18:05 hmmmm      it does make a difference, if they're floats i can get twice as much throughput
18:06 hmmmm      also, the noise at 0 won't be 0 unless the seed were 0
18:06 hmmmm      according to noise2d()
18:09 celeron55_ haven't touched it in a good while; i guess you might know better
18:10 hmmmm      yes i've decided that no matter how the mapgen is redone it's going to take a lot of perlin calls
18:10 hmmmm      so the focus right now is on SSE-ifying the perlin noise
18:13 hmmmm      it might be cleaner to divide this up into perlin_asm_msvc.c and perlin_asm_gcc.c, and which file is added is chosen in the makefile
18:40 hmmmm      this is just a misc. problem that should not exist
18:41 hmmmm      1 * 4096 + 2048 = 2 * 4096 - 2048
18:41 hmmmm      you know, if y = 1, or y = 2, and x = 2048, or x = -2048
18:41 hmmmm      blocks can be replaced
18:42 VanessaE   especially if the server's HDD runs out of space.
18:42 hmmmm      this doesn't have to do with that
18:42 hmmmm      anyway this could be fixed by treating the LBA as a 48-bit field of 16 bit unsigned integers
18:42 VanessaE   oh, sounded like our previous convo.
18:44 celeron55_ hmmmm: i'd imagine making a perlin noise implementation that is built from the ground up to make arbitrary chunks of noise would work well
18:44 celeron55_ hmmmm: the current one calculates the same things again and again
18:45 hmmmm      i reckon you can fix this without any negative side effects by changing getBlockAsInteger to (sqlite3_uint64)((sqlite3_uint64)p.Z << 32) | ((sqlite3_uint64)p.Y << 16) | ((sqlite3_uint64)p.X << 0));
18:45 hmmmm      (and getIntegerAsBlock too for that matter)
18:47 celeron55_ also, no assembly; C is close enough to that
18:47 hmmmm      assembly is necessary here
18:47 celeron55_ (C is a cross-platform macro assembler)
18:47 hmmmm      it's easy enough to fall back on C if the assembly isn't supported
18:47 hmmmm      so i don't see why not
18:48 celeron55_ yeah and work unpleasantly slow?
18:48 celeron55_ you'd need to make an ARM one at least too
18:48 celeron55_ and then what about clang?
18:48 hmmmm      clang supports GCC-style inline assembly
18:48 hmmmm      as does ICC
18:49 hmmmm      this would be nothing but a strict speed _increase_
18:49 hmmmm      ARM wouldn't be unpleasantly slow, it'll just be normal, and the x86 version would be really fast
18:49 celeron55_ anyway, the current implementation "polls" each noise position from the noise functions for a big chunk; it could just calculate each noise level one time, buffer them and then add the buffers
18:49 hmmmm      I can do NEON too, I won't really be able to test it well though.
18:49 celeron55_ it'd take out SO many calculations
18:53 hmmmm      definitely, the entire thing is getting a redesign to be the most optimal
18:55 celeron55_ if it ends up good, then maybe 3d noise could be used for generation again without some silly tricks
18:55 hmmmm      i am already using 3d noise for Nether and Aether biomes
18:56 hmmmm      so it's an absolute must to get 3d perlin noise in a good state
18:57 celeron55_ you'd get more bang for the effort by adding Lua APIs for bulkier generation of things
18:58 hmmmm      what, the biomes?
18:58 hmmmm      i guess, i'd rather just have it in C for now so there's no performance problem from the start
19:00 celeron55_ but that is basically uneditable for most people, which is kind of not how things are supposed to be
19:00 hmmmm      Lua api for biomes might be like this:  if a custom node-placement function is not required, then noise type and noise parameters are specified along with the temperature and height ranges for when it occurs
19:00 hmmmm      at the same time i also aim to make it friendly for people who'd like to add it right in the engine as well
19:32 hmmmm      in map.cpp, ServerMap::saveBlock:  block->resetModified(); should not be called on error like it is.
20:52 VanessaE   celeron55_: what's your opinion on having the client cache the received map data on disk (instead of just memory) for future use?