Hmmm, I think you are better at words than I am, I generally come out a bit rough :-D ... when I was running 15 ns latency, I was running on the NMI and had disabled half the cache for the microcontroller, this one made it possible to disable and memory map it, so I had plenty of space to stuff the 'fun part' in , about 8 or 16 k .... which made the time for the interupt deterministic so we prescheduled a bit, end result we always came out with the same latency.
Is it possible to do something similar on the Arduino ? I might check if adm64 permitts something similar, I will be back on that. It would be nice to use the arduino as backend, stepgenerator or so, and linuxcnc as frontend for heavy stuff .... Floating point is nice but people tend to trust it :-D .... so I have actualy had more problems getting things running in fixed point code, but more problems getting stuff 'safe' or bugfree in floatingpoint, it's so easy to loose precision. / regards, Lars Segerlund. 2013/7/6 EBo <e...@sandien.com>: > I agree. If you can stuff your entire calculation into fixed point > then you can do some wicked cool stuff -- I worked on a project where we > embedded a 4-level wavelet decomposition and filter into an FPGA that > processed ultrasonic scans of train rail tracks at 35MPH in real-time > (we were scanning for cracks). > > I do not consider this a pissing match (yet), but I agree with the > sentiment. Adding to my prior points (of rt-preempt, jerk, and real > NURBS instead of bi-arcs), I would also ask in the redesign if we can > break things out enough so that we could actually use other parts of > LinuxCNC in embedded systems (like path planning, tool compensation, > etc.) As Lars said, if we can get 15ns latencies out of them then for > certain applications they are the right kind of thing. > > EBo -- > > On Jul 5 2013 2:41 PM, Lars Segerlund wrote: >> Hi guys, >> >> Floating point is moot, if you use fixed point in machine >> coordinates >> and add a 2^x bits to that, and you can do a lot of things on simple >> hardware :-D >> >> Linuxcnc flaunts a lot of hardware, but I think a microcontroller >> can >> beat it any day, I did 14.7 ns jitter on one once :-D ... go figure >> that out will ya, ( interrupt jitter, scheduled timer ) .... all >> that >> said I work on linuxcnc and rt-preempt ..... >> >> So if both sides in this debate state what they would like out of >> this perhaps some progress can be made instead of pissing contest ? >> >> I think linuxcnc could do wonders for 3d printing, and I think the >> 3d >> printing folks can teach us a bit about hardware and software, and I >> think the linuxcnc folks knows a bit about motion controll they can >> share :-D >> >> / Lars Segerlund >> >> >> ------------------------------------------------------------------------------ >> This SF.net email is sponsored by Windows: >> >> Build for Windows Store. >> >> http://p.sf.net/sfu/windows-dev2dev >> _______________________________________________ >> Emc-developers mailing list >> Emc-developers@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/emc-developers > > > ------------------------------------------------------------------------------ > This SF.net email is sponsored by Windows: > > Build for Windows Store. > > http://p.sf.net/sfu/windows-dev2dev > _______________________________________________ > Emc-developers mailing list > Emc-developers@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-developers ------------------------------------------------------------------------------ This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
