>>>>> "Sergey" == Sergey Kvachonok <[email protected]> writes:
> On Sun, Jan 2, 2011 at 5:22 PM, David Kuehling <[email protected]> wrote: >> hi, >> >> I managed to compile GNU Octave for the NanoNote. The package source >> is available from git://projects.qi-hardware.com/openwrt-packages.git >> , binary package can be downloaded here: >> > That's a great achievement, but is octave really functional on NN? > AFAIK it has fortran/c++ core and uses double precision math for > everything. GCC softfloat performance does not shine in most cases... > Could you please give some reference how fast it is? For example > inverting 16x16 matrix, does it take days? No, I think even with softfloats its much faster than the average pocket calculator :) 16x16 inversion: tic; inv(rand(16)); toc Elapsed time is 0.02449 seconds. 32x32, 100x100 inversion: tic; inv(rand(32)); toc Elapsed time is 0.1179 seconds. tic; inv(rand(100)); toc Elapsed time is 2.937 seconds. The memory consumption can propably be tuned by changing the memory increment of octave's allocators. But since compilation takes about 1h that's going to take some time to find out. Also linking liboctave liboctaveinterp and everything statically into one binary /might/ help to save memory (I wonder what'd be the downside?). BTW some of the functionality that matlab users would expect resides in the separate libraries that are part of octave-forge [1]. These are not part of the octave package and it's going to take some more time and effort to package them separately. cheers, David [1] http://octave.sourceforge.net/ -- GnuPG public key: http://user.cs.tu-berlin.de/~dvdkhlng/dk.gpg Fingerprint: B17A DC95 D293 657B 4205 D016 7DEF 5323 C174 7D40
pgp1rHBDIHHfq.pgp
Description: PGP signature
_______________________________________________ Qi Hardware Discussion List Mail to list (members only): [email protected] Subscribe or Unsubscribe: http://lists.en.qi-hardware.com/mailman/listinfo/discussion
