| Commit message (Collapse) | Author | Age |
|
|
|
| |
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@307 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
|
|
|
|
|
| |
this is need for gfortran 4.1 which otherwise assumes that the arrays have
already been used (in the kind(gaa) initializations inside of
CCTK_DECLARE_ARGUMENTS).
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@278 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
| |
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@276 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
| |
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@275 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
|
|
|
|
|
|
|
| |
* not all features of GRHydro are supported yet, in particular only the
HLLE solver supports Mulitpatch yet.
Original commit by Christian Reisswig and Christian Ott
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@273 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
|
|
|
|
|
|
|
| |
* get rid of nonsensical re-computation of the conserved tracer
* remove support for shift_state = 0 and get rid of unnecessary
assignments.
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@256 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1) Further splitting of PPM Reconstruction routine into magnetic
and non-magnetic part.
2) Merge the divergence cleaning loop into the non-divergence one,
since profiling indicated no substantial difference between the
two cases. Indeed it was a little bit better after the merger
(0.14%), but that it is not significant. We decided then to apply
Occam's razor and choose the simplest form. Branch prediction
seems to work fine in this case.
3) Move reconstruction initialization statements to their repective
drivers.
4) Get rid of WORKSHARE in the reconstruction routines. Profiling
showed a 4.16% performance improvement for the hydro ppm reconstruction
routine when using 1 processor and 2 threads only. Expect a
better improvement for a larger number of threads.
5) Introduce a parameter to control characteristic speeds
for psidc in HLLE.
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@255 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|
|
Comments:
1) The reconstruct routine is the most expensive routine
in GRHydro and may require some serious thought in the
future regarding optimization.
2) This routine was merged by the end of the last year.
With too many loops, the compiler wasn't able to fully
optimize it (and it would warn about that).
3) With this split, the code became easier to read,
the compiler was able to fully optimize it, but we only
have a marginal gain in the routine performance (3.2%).
Further optimization and clean-up is still desirable.
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinEvolve/GRHydro/trunk@252 c83d129a-5a75-4d5a-9c4d-ed3a5855bf45
|