I've just began work on a NEON optimised cmath like library. You can
see what I've done so far here:
I don't have any hardware to test on so I've mainly been focusing on
branchless C implementations (except where the branch can be
precaclulated on the ARM). I want it to be more game targeted than
DSP, so I've made some large approximations to gain performance. All
of the currently implemented functions i've written from scratch using
Mathematica to generate minimax polynomials.
In C: sinf, cosf, tanf, atanf, powf, logf and expf.
In NEON asm: sinf, tanf and atanf.
I've been considering entering it into the contest i just found on the
wiki. Is this a good idea? Might be an easy way to get some
I'm very new to this stuff, this is the first real assembly I've
written. So any tips or comments on bugs in code would be much
PS: I've also began work on a OGL to OGLESv2 translator. You can find
Haven't worked on it alot recently, but it does implement a useful
subset. For instance i dynamically generate fragment shaders for OGL2
compliant multitexturing combinors, fog, alpha test support.