| Commit message (Collapse) | Author | Age |
|
|
|
| |
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@1526 f88db872-0e4f-0410-b76b-b9085cfa78c5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
writing data to the SphericalSurface variables in a multiprocessor run.
In detail, the code writes data to the SphericalSurface variables
on each processor. For everything except the horizon shape this
is fine. However, the code formerly didn't broadcast the horizon shape
to all processors, so the values written to the SphericalSurface variables
were all 0 on processors other than the one that actually found that horizon.
The fix is to broadcast the horizon shape if any horizon info is
being written to the SphericalSurface variables. (In addition to
broadcasting it whenever an excision mask is being set, which the
code already did.)
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@1392 f88db872-0e4f-0410-b76b-b9085cfa78c5
|
|
|
|
|
|
|
|
|
| |
biggest change is that we now group all the per-Newton-iteration
interprocessor communication logic into a single call per iteration;
this eats more bandwidth, but saves on latency
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@971 f88db872-0e4f-0410-b76b-b9085cfa78c5
|
|
across multiple processors -- see src/driver/README.parallel for details
* drop convergence checks on ||Delta_h|| in param.ccl because they don't
fit well with parallelization changes
==> With this changes, AHFinderDirect is now (I think)
multiprocessor-ready!!
git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/AHFinderDirect/trunk@946 f88db872-0e4f-0410-b76b-b9085cfa78c5
|