[ODE] ODE and SIMD
simon at bf2engine.com
Tue Nov 28 02:59:43 MST 2006
> indeed, I was trying to make faster code for matrix and vectors
> operations, but in result I was very close to c++ version with
> compiler optimizations on (it was vc7) - imho you can get very slight
> improvement, but the work might be huge...
Currently I'm trying to optimize SOR_LCP solver with SIMD. I profiled
test_crash example with about 1800 boxes and more than 80% of simulation
time was spent in this solver. The main problem that I hit was data
misalignment, but this should be easy to resolve (I hope ;)).
I'm also thinking about another approach - maybe better - moving the most
expensive computations from this solver to GPU.
More information about the ODE