On 03/07/2012 02:49 PM, Andre Smit wrote:
Is it 2012.2 or git master?
git master. I haven't been keeping track but did cython help the speed any?
It helped not in terms of speed, but it did help in terms of code simplicity. I have actually written some code that was in C directly in cython. So e.g. handling of complex numbers is for free now.
Overall, the assembling of vector/matrix is slower than it was two years ago, but the code is much more readable/simpler now (and more work is done during the assembling now - lazy evaluation of reference mappings etc.). There are certainly areas where the speed could be re-gained without much pain. Also the speed of other, non-assembling, parts of the code is better now. For example
time ./simple.py examples/diffusion/cube.py
with meshes/3d/cube_big_tetra.mesh mesh gives now
sfepy: nls: iter: 0, residual: 2.110578e+01 (rel: 1.000000e+00) sfepy: cg convergence: 0 (successful exit) sfepy: rezidual: 0.13 [s] sfepy: solve: 0.49 [s] sfepy: matrix: 0.16 [s] sfepy: linear system not solved! (err = 1.851779e-11 < 1.000000e-12) sfepy: nls: iter: 1, residual: 1.851784e-11 (rel: 8.773824e-13)
real 0m10.983s user 0m10.489s sys 0m0.368s
and with two years old code
sfepy: nls: iter: 0, residual: 2.110578e+01 (rel: 1.000000e+00) sfepy: rezidual: 0.03 [s] sfepy: solve: 0.49 [s] sfepy: matrix: 0.12 [s] sfepy: linear system not solved! (err = 1.851779e-11) sfepy: nls: iter: 1, residual: 1.851784e-11 (rel: 8.773824e-13)
real 0m19.853s user 0m19.265s sys 0m0.492s
r.