matrix operation optimization

Hi

I am trying to implement a code that requires 4 x 4, 4x 1 matrix operations at each grid node on a 2-D computational space.

Are there C/C++ libraries available that can do handle this type of matrix calculations?

My understanding is that libraries like lapack etc. are suited for large matrix inversions.

cheers
shuo

LU decomposition might be relevant to your interests.

google yields many results.

http://www.alglib.net/matrixops/general/lu.php

SIMD instructions (MMX, SSE, etc) might help. GCC and the Intel compilers have intrinsics (built-in non-standard functions) that you can use to achieve this. Don't even need to go into assembly. There is probably something equivalent on the Microsoft compiler.

Quote:

---

Originally Posted by cyberfish

SIMD instructions (MMX, SSE, etc) might help. GCC and the Intel compilers have intrinsics (built-in non-standard functions) that you can use to achieve this. Don't even need to go into assembly. There is probably something equivalent on the Microsoft compiler.

---

Yes, and both do not do that great a job of translating the intrinsics into good code.

Also, SIMD requires a little bit of planning so that data that should be processed together sits together in memory - otherwise the code ends up spending as much time shuffling data around as you save on doing multiple data in one instruction

--
Mats