[Detrius]

Hello All!

I'm needing to do quite a bit of work with matrices. Specifically, I'm needing to calculate inverses of matrices and to convert vectors between bases (which typically would involve inverting a matrix).

I've written the code to do this by hand. However, minor errors in the accuracy of double precision floating point numbers are resulting in non-zero values where they should be zero... which wreaks MAJOR havoc on a matrix inversion (as any non-zero value needs to be fixed). I've done some research into OS X's vecLib and have found LAPACK. However, these are hideously ugly fotran routines converted to C using f2c... it doesn't mesh too well with my C++ and Objective-C programming style.

Any input on this issue would be greatly appreciated--this includes tips on getting around the errors in the accuracy of a double as well as other OS X frameworks that might work better.


At this point, I'm running under the assumption that LAPACK is my best option (as it is AltiVec accelerated).


BTW, as soon as I get this major-minor bug worked out, I'll release the vector calculator that I promised a long time ago. Sorry... I finally got around to doing it--to figure out why other things weren't working *quite* like they were supposed to.

[smeger]

I'm no longer in a field where I need this stuff, so my info is probably somewhat out of date.

That said, you can check out TNT or Blitz++. These are both object-oriented linear algebra packages similar to LAPACK (which, I agree, is vile). I haven't really checked out TNT, but it sounds nice from their description. Blitz++ seems to have sorta stagnated, but it's the object oriented package I've heard the most about.

Good luck on the calc!