[overture] Re: New Overture release v24

  • From: "Kyle K. Chand" <chand1@xxxxxxxx>
  • To: "overture@xxxxxxxxxxxxx" <overture@xxxxxxxxxxxxx>
  • Date: Tue, 24 May 2011 14:27:08 -0700

Hi Dominic,

So the "s" in smesh stands for "simple" (and is silent :-). It does not reorder
the anything in any way.  In fact, the output file only contains the
vertices and the list of vertices for each zone/cell.  You would have
to reconstruct the edge data from the cell->vertex connectivity.
The file Mapping/entityConnectivityBuilder.C has some code that
can do that for you...

Regards,
Kyle

mailto: chand1@xxxxxxxx
phoneto: (925) 422 7740



On May 24, 2011, at 2:14 PM, Dominic Chandar wrote:

Hi Kyle,

Thanks for the update. Currently, I'm working on Unstructured meshes, mostly applicable to Incompressible/Compressible Flow solver on the GPU, very similar to cgins ( using same algorithms etc ). I'm very much interested to see the gpu performance on hybrid grids. I will try the 2D smesh initially.

Also does smesh reorder the nodes/edges so that the computations are cache efficient ?

Regards,
Dominic

On Tue, May 24, 2011 at 1:01 PM, Kyle K. Chand <chand1@xxxxxxxx> wrote:
Hi Dominic,

The 2D unstructured mesh generator smesh is now in the distribution.
It was supposed to be in the last one but by an oversight it did not make it.
You can get some documentation for smesh on or documents page.

As for 3D meshes, you can try generating a 3D hybrid grid from an overlapping grid. I also wrote a 3D high-order tet mesher that combines our CAD/ IGES stuff with the advancing front mesh generator. Do you have any interest in that? I never
thought about including it with the Overture release...

Kyle

mailto: chand1@xxxxxxxx
phoneto: (925) 422 7740




On May 23, 2011, at 3:22 PM, Dominic Chandar wrote:

Hi Joel,

It definitely is tempting to use Overture with Petsc+gpu, but you will have to transfer data back and forth between CPU and GPU every time step ( cudaMemcpy - I believe you are using cuda ) which will spoil the speed-up. I guess, when using GPU's, one can reap a lot of benefit only if the entire computation is done on the GPU.

Bill,
Thanks for a new release ! Do we have any unstructured grid generator shipped with Overture this time ?

regards,
Dominic


On Mon, May 23, 2011 at 3:27 PM, Joel Guerrero <joegi.geo@xxxxxxxxx> wrote:
Hi Bill,

Great work, I can't wait to try this new version. Just one question, lately I have been playing around with petsc + gpu, so I wonder if it will be possible to use th epetsc gpu accelerated solver with overture.

Regards,

Joel

From: Bill Henshaw <henshaw@xxxxxxxx>
To: freelists <overture@xxxxxxxxxxxxx>
Sent: Mon, May 23, 2011 9:39:55 PM
Subject: [overture] New Overture release v24

Dear Overture users:

Announcing the release of Overture.v24 and cg.v24.

These are now available from the Overture web page.

A list of some of the major changes to Overture and cg are
given below (from the Overture/CHANGES and cg/Changes files).

Please let us know if you have troubles building or using
the new version or if we have forgotten to fix some bug that
has been previously reported.


Regards,
#6.


= = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = ======================================================================
Overture.v24 - released  May 20, 2011

Changes:

Overture.v24
Changes:
NOTE: Due to a bug fix in the HDF file output (100329 below) you will need to regenerate your grids and showfiles
       for this new version.

o there is a new graphics option, "set home", (in the command window "File menu") that will make the current view the "home" view. The home view is the one that appears when using the "reset" command (the button with the "home" on it). The "reset" button (command GLOB:init view)
           will return the home view to the default.
o 090422 : fixed a bug in ogen that mainly applied to multi-domain grids -- sometimes interpolation was allowed between grids from different domains. (classify.C l.2615) o 090423 : added a check to mogl.C for NumLock being on -- a message is printed if it is. You should turn off NumLock since the mouse buttons won't work properly. o 090429 : fixed a bug in CompositeGrid.C for grids saved from parallel ogen, some of the interpolation info was not saved correctly (convertLocalInterpolationData). o 090529 : fixed plotting of sequences in plotStuff with multi- domain problems. The sequences to plot
           now appear on the main dialog.
o 090530 : NOTE: running remotely with the gnome window manager can be slow when running command files interactively, (not sure why?), KDE seems less likely to have this problem. See note below 090809. o 090531 : fixed a problem with rubber-band zooming (with middle mouse button) with extreme zooms (e.g. for fixing trim curves) and picking objects -- this works better now. o 090628 : contour2d, when picking values with the mouse, hidden grids will not be used. o 090705 : now when a moving grid computation fails with an invalid grid, the program enters the normal interactive update mode of Ogen so you can remake/query the grid (before the
           program entered some funny update mode that was confusing).
o 090705 : fixed bug with moving grids, the bounding box was not always computed correctly which led to ogen failing when it should have worked (this bug was caused by some code added for parallel). o 090804 : fixed a bug in the ogen's moving grid algorithm: the mask on ghost points was not always set correctly (the problem occurs when a grid "slides" along a shared boundary). o 090808 : there are new parallel copy routines to copy a serial array from one processor to another, see
           ParallelUtility.h and CopyArray.bC
 o 090809 : there is a new option for the .overturerc file:
showCommandHistory: 1 # set to 0 if interactive scripts are slow to display Sometimes running scripts interactively is very slow when running over a network and
           this may be fixed by setting this option to '0'.
o 090913 : The new plot option 'set plot bounds' (from the grid or contour plotter) allows one to set fixed bounds on x, y and z. Useful when the grid deforms and changes size.
           Use 'reset plot bounds' to turn off.
o 091127 : The new InterpolatePointsOnAGrid class can be used to interpolate a set of points in space from a composite grid function. This functionality works in parallel and supports higher order interpolation. It replaces the old InterpolatePoints class and even older interpolatePoints function. o 091127 : The ExposedPoints class now works in parallel (used to interpolate points that become "exposed"
           during a moving grid computation).
o 091127 : Fixed some bugs for the parallel version of "extrapolate interpolation neighbours" that is used to assign values to unused points next to interpolation points (allowing a wider stencil to be used). o 091209 : there are new parallel copy functions in the CopyArray class for copying serial arrays into a single distributed array (see ParallelUtility.h and CopyArray.bC). This is more general than the normal P++ copy since the serial arrays can be arbitrarily distributed. o 091209 : There are a variety of new grid generation examples in the sampleGrids directory including: cubeInABox.cmd, slider.cmd, squareHole.cmd, sphere.cmd, rotatedBoxBesideBox.cmd valveArg.cmd, plateWithHoles.cmd, plate3dWithHoles.cmd wing3d.cmd, backStepSmooth3d.cmd,
               divot.cmd, afm.cmd, lens.cmd, solidSphereInABox.cmd
 o 100226 : The Integrate class should now work in parallel.
o 100309 : There is a new LoftedSurfaceMapping that can be used to define a lofted surface such as the surface of a wing with a tip. This mapping was used to build some of the surfaces for the overlapping grid for the model wind-turbine with tower and nacelle. o 100329 : Bug fixed in the HDF file output of parallel distributed arrays when using the multiple file output option (which is the default). Space was being allocated in the main hdf file for the whole array even though no data was written there. This meant that the main hdf file was huge when it should not have been. Thanks to Jeff Banks for noticing this. o 100424 : In cleaning up the MappedGrid the "min and max edge length" arrays were removed since these are
           rarely used.
o 100722 : There is a new "exponential to linear" stretching option which might be a good choice for clustering grid points near boundaries. This function can be used to create very small grid
           cells near boundaries if desired.
o 110211 : There are two new functions in showFileFromDataFile.C that can be used to read and write a grid and solution from an 'ovText' data file. Use 'plotStuff -ovText' to read in a grid and solution from an "ovText" data file (choose the menu item 'data file format' to display the format of the file). From the plotStuff menu use 'save ovText file' to save a grid and solution from a show file into an ovText data file. This is a good way to import a grid and solution into Overture. The solution can subsequently be used as initial conditions for one of the CG solvers, for example. This format is better than the plot3d format since it has more information about boundary conditions, ghost points etc. o 110515 : There is a new "transferSolution" program in Overture/ bin that can be used to interpolate a solution from one overlapping grid to another overlapping grid (e.g. a finer grid, or a grid on a larger or smaller domain.). The program reads a solution from a "source" show file and outputs the
           results to a new "target" show file.
o 110515 : Some improvements have been made for reading and writing plot3d files. There is a new option "save plot3d file" in the plotStuff menu that allows one to save a solution and grid into plot3d files (these could then be shown in another graphics program for e.g.). plotStuff can read plot3d grid and solution ("q" or "f") files using the
           command "plotStuff -plot3d".

= = = = = = = = = = = = = = = = = = = = = = = = ======================================================================
cg.v24 -- released  May 20, 2011


Changes:

o There have been many changes to the CG solvers and I have not kept a careful list.

o a new "matrix motion" option has been added for moving grids which can be used to compose together different rotations and translations to form more complex motions. See the new document matrixMotion.pdf for more details. The test routine cg/user/bin/motion can be used to test different matrix motions.










Other related posts: