Latest Updates and NotesΒΆ

23.11.2012: FIXED: mesh initialization overhead, which was seen FEniCS 1.0 (not 1.0+)

23.11.2012: NOTE: the preconditioned cg-iteration counts may differ for different FEniCS installations, depending on the used PETSc release. For example the poisson-unitsquare demo:

petsc-3.3-p3

refinements |    dofs | cg+hypre_amg iter | cg+hypre_amg time | cg iter |  cg time | cg+fmg iter | cg+fmg time
------------+---------+-------------------+-------------------+---------+----------+-------------+------------
          0 |      64 |                 3 |           0.00124 |      23 | 0.000145 |           1 |    0.000586
          1 |     225 |                 4 |           0.00141 |      44 | 0.000395 |           5 |     0.00144
          2 |     841 |                 4 |           0.00452 |      87 |  0.00247 |           5 |     0.00253
          3 |    3249 |                 4 |            0.0195 |     172 |   0.0187 |           5 |     0.00693
          4 |   12769 |                 4 |            0.0864 |     343 |    0.221 |           5 |      0.0210
          5 |   50625 |                 4 |             0.370 |     685 |     2.06 |           5 |      0.0773
          6 |  201601 |                 4 |              1.50 |    1368 |     16.1 |           5 |       0.294


petsc-3.2-p5 (different computer)

refinements |    dofs | cg+hypre_amg iter | cg+hypre_amg time | cg iter |  cg time | cg+fmg iter | cg+fmg time
------------+---------+-------------------+-------------------+---------+----------+-------------+------------
          0 |      64 |                 3 |          0.000780 |      23 | 6.70e-05 |           1 |    0.000343
          1 |     225 |                 4 |          0.000685 |      44 | 0.000237 |           5 |    0.000835
          2 |     841 |                 4 |           0.00242 |      87 |  0.00153 |           5 |     0.00146
          3 |    3249 |                 4 |           0.00971 |     172 |   0.0119 |           6 |     0.00349
          4 |   12769 |                 5 |            0.0426 |     343 |   0.0945 |           6 |     0.00931
          5 |   50625 |                 4 |             0.165 |     685 |    0.786 |           6 |      0.0329
          6 |  201601 |                 4 |             0.752 |    1368 |     7.46 |           6 |       0.148

Since the cg-iterations are equal there must have been something fixed in PETSc KS preconditioning.

Previous topic

Documentation for FMG-1.0 (C++)

Next topic

Getting Started

This Page