| 00:36.59 | *** join/#brlcad fgmwbcwkyxfdqobw (~armin@dslb-088-064-046-039.088.064.pools.vodafone-ip.de) |
| 01:35.44 | starseeker | huh, interesting: https://github.com/arrayfire/arrayfire |
| 01:59.06 | *** join/#brlcad amarjeet (~Amarjeet@101.216.63.182) |
| 02:13.06 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 02:56.30 | brlcad | starseeker: Nice! arrayfire looks really interesting actually, definitely something to try on the cl branch |
| 02:56.54 | brlcad | would allow some nice cl vs cuda vs cpu comparisons/debugging |
| 02:59.53 | Stragus | brlcad, for the mesh decimation, it seems the only extra check still missing would be to make sure that no two triangles share the same edge in the same direction |
| 03:00.00 | Stragus | (unless that's included somewhere in that last) |
| 03:00.03 | Stragus | that list* |
| 03:00.37 | brlcad | sync does that |
| 03:01.03 | Stragus | And CUDA is only significantly faster than CL if you use the extra CUDA-only capabilities, like warp shuffles |
| 03:01.12 | brlcad | it just doesn't know if it's inverted the whole thing or not, and of course only works if the geometry is actually syncable with no degenerate faces |
| 03:04.00 | Stragus | And I believe any "framework" to run code on both CPUs and GPUs can't be usable, optimization for one is totally different from the other... |
| 03:05.37 | Stragus | And proper code makes a huge difference on GPUs (like 10x), such as pulling "work" coherently into the on-chip shared memory |
| 03:05.45 | brlcad | looks like they provide their own library of operations you perform on the data containers, which they also define |
| 03:06.08 | brlcad | how to pack and process the data is still up to you |
| 03:06.46 | brlcad | looks like you can compile lock your code to a particular backend, or you can make it bind at runtime to one of the three |
| 03:07.42 | Stragus | So it's still bound and targeted to a backend, if you want performance. Can't really see the advantage |
| 03:08.06 | brlcad | it can be bound, doesn't have to be |
| 03:08.58 | Stragus | Well, if it's 10 times slower because it doesn't use shared memory, that's not really usable |
| 03:10.00 | brlcad | who said it's 10x slower? they actually claim to be faster than most people's gpu code |
| 03:10.36 | brlcad | looks like they make that claim based on their specific function kernels they provide for specific common operations |
| 03:10.54 | brlcad | which they presumably have tailored for each backend |
| 03:11.05 | Stragus | Depends on the problem, obviously. But I know very well that if you don't target GPU specifically and write CPU-style code, it can easily be 20 times slower than it could be |
| 03:11.23 | brlcad | sure |
| 03:23.13 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 06:45.45 | *** join/#brlcad teepee (~teepee@unaffiliated/teepee) |
| 07:50.44 | *** join/#brlcad sniok (~sniok@89.252.2.135) |
| 08:57.25 | *** join/#brlcad teepee` (bc5c2134@gateway/web/freenode/ip.188.92.33.52) |
| 11:21.57 | *** join/#brlcad Notify (~notify@66-118-151-70.static.sagonet.net) |
| 11:23.13 | Notify | 03BRL-CAD:brlcad * 67823 brlcad/trunk/src/nirt/parse_fmt.c: plug a tiny leak. we're done with the filename. |
| 11:23.15 | Notify | 03BRL-CAD:brlcad * 67824 brlcad/trunk/src/nirt/nirt.c: make nirt report the current output style as well as a listing of the available output formats, with hints on what command line options change the setting or give more info. this is implemented specifically in response to user requests (e.g., ajem scr 1942) where several people have asked to have gaps or exit hitpoints reported, which are features of the other |
| 11:23.17 | Notify | output formats. |
| 11:23.19 | Notify | ... |
| 11:23.28 | Notify | 03BRL-CAD:brlcad * 67825 brlcad/trunk/NEWS: nirt now reports the current output format setting, itemizes the available formats, and provides parenthetical help for what options pertain to output formatting. this was implemented in regards to prior mailing list discussion and ajem scr 1942 requesting gaps be reported (which is already one of the available output formats. |
| 12:11.21 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 12:25.55 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 12:46.14 | *** join/#brlcad yorik (~yorik@177.189.140.33) |
| 13:46.14 | Notify | 03BRL-CAD:starseeker * 67826 brlcad/trunk/src/tclscripts/rtwizard/CMakeLists.txt: remove rtwizard.bat |
| 13:54.54 | Notify | 03BRL-CAD:starseeker * 67827 brlcad/trunk/misc/CMake/NSIS.template.in: Tell NSIS to go for the .exe files, not the bat files - MGED is the only bat launcher left. |
| 15:09.20 | ``Erik | is it wwdc yet? O.o |
| 15:29.27 | *** join/#brlcad Mathnerd314 (~quassel@supertux/Mathnerd314) |
| 15:56.37 | Notify | 03BRL-CAD:n_reed * 67828 brlcad/branches/brep-debug/doc/docbook/system/implementation/en/bool_eval_development.xml: add descriptions of remaining dplot subcommands |
| 16:37.14 | *** join/#brlcad sniok (~sniok@89.252.2.135) |
| 17:01.35 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 17:06.37 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 19:02.32 | *** join/#brlcad amarjeet (~Amarjeet@101.216.73.71) |
| 19:07.13 | *** join/#brlcad amarjeet (~amarjeet@101.216.73.71) |
| 19:15.18 | Notify | 03BRL-CAD:n_reed * 67829 brlcad/branches/brep-debug/doc/docbook/system/implementation/en/bool_eval_development.xml: document how to plot evaluation curves not automatically recorded by dplot |
| 19:22.07 | *** join/#brlcad yorik (~yorik@177.189.140.33) |
| 19:48.37 | *** join/#brlcad amarjeet (~Amarjeet@101.216.73.71) |
| 20:05.43 | *** join/#brlcad ickby (~stefan@HSI-KBW-046-005-253-113.hsi8.kabel-badenwuerttemberg.de) |
| 20:27.30 | *** join/#brlcad ickby (~stefan@HSI-KBW-046-005-253-113.hsi8.kabel-badenwuerttemberg.de) |
| 20:38.30 | *** join/#brlcad ickby (~stefan@HSI-KBW-046-005-253-113.hsi8.kabel-badenwuerttemberg.de) |
| 20:41.15 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 23:26.05 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 23:58.58 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |