PEGASUS 5 Current Status

Version 5.1k Modifications

1. Fixed bug: solid walls in phantom meshes were not used in auto hole cutting surfaces. Any solid wall specified in part of a phantom mesh is now utilized by the auto-hole cutting.

2. Fixed bug in default value of QCUTOFF: increased from 0.0 to 0.1. See detailed discussion on the implications of this in the section on Known Bugs

3. Fixed bug in MPI coding that limited the number of CPUs. The bug potentially caused problems when pegasus5mpi was used with more then 14 processors, due to an improperly dimensioned array.

4. Fixed bug in fixing of 2nd-fringe orphan ponts on periodic grid planes.

5. Made changes to allow REGION/VOLUME of type INTR to blank out a region extending to the boundary. Added this feature in the documentation for type INTR. Now it excludes any points in a VOLUME of type INTR from the list of single and double fringes. Fixed restart dependency: the list of fringe points was not rebuilt when the interior volume range input was changed.

6. Added --with-byteswapio option to the configure script for Portland group compilers. Added workaround for Portland compiler version 5.0-2: internal compiler error when using -fast flag; changed configure script to use "-fast -Mlre-noassoc" with pgi compilers; This compiler bug is supposed to be fixed in their next release; this does not slow the code down.

Version 5.1j Modifications

1. Modifications where made to the interpolation stencil searchs to make them more efficient and more robust for certain cases. This will cause 5.1j to produce slightly different results than previous versions. Extensive testing has shown that 5.1j produces the same or fewer orphan points, and runs in slightly less computing time.

2. Fixed several bugs, including: typo in index limit in do loop when creating the XINTOUT file which would cause wrong count of 1st or 2nd fringe orphans; fixed some coding errors inside some DEBUG statements; fixed bug in mod_spec_int2 which could cause the code to bomb for periodic o-grids in k; fixed memory allocation error in peg_hole_surf utility which would bomb for multimle hole cutters.

3. Modified peg_in for flow-through bc: simplified coding for first part which adds the ityp=51,52, or 53. Added more checks to detect a wingcap topology to fix cases where it was incorrectly adding ityp=-1 surfaces for a wingcap wake.

Version 5.1i Modifications

1. Improvements to the interpolation stencil searches, resulting in fewer ORPHANS for nearly all cases tested. These mostly involved grid topology checks such that the stencil walk could more readily jump across O-grid periodic boundaries and C-grid flow-through boundaries. With the use of rotated bounding boxes for each mesh, was able to eliminate stencil-walk searches for many points outside of donor meshes.

2. Fixed bug for projection which would cause bad problems for meshes with both viscous and inviscid walls. The projection subroutines were still using some data from the auto-hole cutting wall arrays. Added projection output to the log file indicating the number of points which were not projected due to ANGLE and DISTANCE restrictions.

3. Fixed bug in namelist input which would fail when reading lines greater than 80 characters. Can now read up to 132 characters on a line.

4. Added minor changes and fixes for Cray and HP compilers. This included the addition of a false Fortran 90 module for mpi, as these compilers fail on the Fortran 90 statement "use mpi".

5. Rewrote the mesh_lister and peg_orph programs: better output, more error checking.

Version 5.1h Modifications

1. A fix for the HP compiler and pgif90 compiler problems which resulted in extremely slow execution (this was present only in 5.1g).

2. Fixed a number of memory leaks, resulting in lower memory requirements.

3. Implmented a more graceful abort for fatal errors, including cleaning up and properly aborting all mpi processes;

4. The addition of the INTERNAL variable in the $HCUT namelist, which will enable to perform auto hole cuts for internal flow problems

5. Re-wrote peg_mem to give realistic estimates, and to be much easier to use: it gets its input from peg.in and X_DIR files instead of prompting the user for information.

6. Re-wrote peg_hole_surf so it gets its input from peg.in and X_DIR.

7. Wrote a new version of the peg_in program which will provide [jkl]include in the $MESH namelist which will exclude symmetry planes and solid walls automatically.

8. Fixed a bug in the implementation of $REGION TYPE='INTR'. In previous versions, This type of REGION did not work as documented: the interior points for these regions never got blanked out.

Version 5.1g Modifications

This release includes an MPI parallel option, fixes several bugs, a modification to the level-2 fringe algorithm, and implements the mesh scaling, translation, and rotation capability. NOTE: one should not attempt to perform a restart execution of Version 5.1g using a WORK directory from a previous verison.

Because of the bug-fixes to the interpolation stencil search, and improvements to the level 2 fringe selection (changes 3-5 below), version 5.1g will produce a slightly different XINTOUT files than previous versions. Several OVERFLOW runs comparing the grid systems computed by 5.1f and 5.1g show only a slight changes in the flow solution. For two different aircraft configurations, changes in computed lift and drag coefficients were on the order of 0.1%.

NOTE for Portland Group Compiler Users: change in version 5.1g have created very large performance degradation (factor of 100 slower) in a test of this version on a Red Hat 6.2 Linux PC. UPDATE: This has been fixed in the 5.1h release. This was identified as a compiler bug by the Portland Group support, but a coding work-around was enables in version 5.1h.

Specific changes in version 5.1g include:

1. Added MPI parallelization option. See the User's Guide sections for details on compiling and running this option.

2. Modified output: all output is written to the log.dddd.tttt file (where dddd is the date, tttt is the time). Only error messages are written to stdout.

3. Fixed EPS interpolation stencil bug: for cases where a non-default value of the GLOBAL or MESH namelist input variable EPS was specified, some of the interpolation stencils output in the XINTOUT file would have errors as large as one grid cell. This could have produced significant errors in the solution.

4. Fixed interpolation stencil bug: for certain degenerate grid cells, such as those around a singular axis, bogus interpolation stencils were being generated. The code was not performing a check on the convergence of a newton stencil walk when computing interpolation weights for level 1 and level 2 interpolation. Rewrote the interpolation routines, changing from Fortran 77 to Fortran 90 coding. Replaced two different implmentations of the newton stencil walk with a single version, which is now used in several parts of the code.

5. Modified the level2 fringe point algorithm. When computing cell-difference, compute the donor-cell weights, and then shift the donor j,k,l index so that the donor cell which is being measured is the one closest to the fringe point under consideration. This removes a randomness and directional bias in the level 2 fringes for pointwise matching grids, and fixes a problem which caused the INS3D flow solver stability problems.

6. Implemented the scaling, translation, and rotation of mesh coordinates using the MESH namelist input variables SCALE, X0, Y0, Z0, XR, YR, ZR, ALPHA, BETA, GAMMA. In previous versions of the code these values were read in but never used. Changed the input names from ALFA to ALPHA and GAMA to GAMMA. Implemented mesh coordinate transformations in the tool programs peg_plot, peg_proj, peg_hole, and XINtegrity.

7. Added projection displacements to coordinates in XINtegrity so that this utility can be used for the first time with problems which include projection.

8. Fixed projection bug: no longer use inviscid walls or ibtyp=-1 solid walls for projection surfaces.

9. Fixed restart bug: modification of an existing input volume grid file (X_DIR/*.x) did not cause the code to update the internal alternating-digital-tree (adt) map for the modified mesh. This could cause failed searches or segmentation violations.

10. Changed default name of $HCUT NAME from "HOLE CUTTER" to "HOLE_CUTTER".

11. Fixed restart bug for manual hole-cutting: was not re-cutting when adding/changing $BOX namelist inputs.

12. Changed reading of logical namelist value to check for both upper and lower case T or F.

Version 5.1f Modifications

This version includes several modifications to the automatic hole-cutting routines, which results in an increase in computer time in the autocut process, adding about 25% more to the total execution time. Also, one should not attempt to perform a restart execution of Version 5.1f using a previous verison 5.1e run. Specific changes include:
1. Implemented several modifications to the automatic hole-cutting algorithm. In regions of overlapping solid-wall surfaces, attempt to combine overlapping surface elements (which are represented as triangles) into a unique set of non-overlapping Delauny trianges used for the clear-line-of-sight (CLOS) test. This fixes most, but not all, instances of the bad holes being cut into overlapping solid walls. Also forced CLOS test to examine all of the neighboring Cartesian hole-map elements, and then use majority to determine inside versus outside.

2. Added step to automatic hole-cutting process, which blanks points on a solid wall when their neigbhoring point just above the wall is blanked. This improves the hole cutting for intersecting bodies and collar grids.

3. Added step to automatic hole-cutting process which will blank certain unresolved points with no clear line of sight to an inside or outside element. This examines the normal vectors of near-by walls, and blanks the point if it appears to be on the inside of all the walls. This dramatically improves problems with ORPHAN points left inside of thin bodies, such as thin trailing edges in wings.

4. Added -sp option to the utility programs peg_hole, peg_hole_surf, peg_diag, and peg_proj. The -sp flag forces these utilities to write single-precision output grid and/or solution files.

5. Fixed bug in execute_comp_hole: implementation of REGION TYPE='UNBL' (which has the effect of unblanking a specified subset of a grid) never worked correctly.

6. For 2D problems, added check to ensure that holes will be the same on all three of the computational planes in the third direction.

7. Fixed restart bug with projections: previously, changes to input variables which control projection did not force the code to re-do the interpolations, and thus the change in the projections was never used.

8. Fixed error checking for $BOUNDARY namelist: made it OK for BOUNDARY to not specify ISPARTOF, as long as SURFACE specifies MESHNAME. Added error checking for $SURFACE namelist, checked that MESHNAME matches an existing mesh name. Changed MESH NAME to MESHNAME in documentation for SURFACE namelist.

9. Fixed bug when checking manual hole-cutting inputs. If the number of surfaces was zero, it got stuck in an infinite loop

10. Fixed bug specifying outer boundaries: it was setting second fringe points at outer boundaries, in some cases setting boundary points as fringe points, resulting in ORPHANS on solid walls. Changed the index limits: only need to go all the way to the edge for periodic grids.

Version 5.1e Modifications

1. Additional automatic hole-cutting operation for Cartesian box grids: This additional hole-cutting step will blank out cells of a Cartesian box grid which contain a solid-wall boundary point. This fixes the problem of PEGASUS auto hole cutting unable to blank out large Cartesian box-grid cells which completely surround a solid body and which do not have any vertices inside the solid body. The previous workaround, which did not always work, was to increase the OFFSET input parameter to values in the range 3 to 7 for Cartesian box grids. This is no longer necessary, and thus, an appropriate value for OFFSET for a Cartesian box grid is in the range 0 to 2, which is the same appropriate range for body-fitted volume grids.

2. Fixed large-hole restart BUG in automatic hole cutting: This bug would cause a failure of the automatic hole cutting during a restart run. The failure was characterized by a very large rectangular hole: all points inside the min/max box of the hole cutter would be blanked out. The bug was a failure to initialize the variable ilen in mod_hole_octree during a restart run. This caused all of the cartesian cells in the hole map to be labeled as INSIDE the solid body.

3. Fixed bug with Level 2 interpolation for periodic O-grids: Previously, the Level 2 interpolation would not provide interpolation stencils for points on a periodic plane. For example, consider a volume grid which is periodic in K, and has significant overlap with a neighboring grid, and which a body-normal direction in L, with an outer boundary at L=LMAX. Level 2 interpolation might provide stencils at points L=-10,-1 for all J and all K, except points on the K=1 and K=KMAX planes. This bug has been corrected such that Level 2 interpolation will provide stencils for all K points in a periodic grid. This bug was causing major problems with the flow solution in the vicinity of a the periodic planes.

4. Fixed bug with periodic O-grid outer boundaries: The code will now produce outer-boundary interpolation stencils on the edges of a periodic plane. For example, consider an O-grid which is periodic in the K-direction, and has an outer-boundary at L=LMAX which requires interpolation stencils. Previously, PEGASUS would not provide stencils on the edges given by {J=1,JMAX;K=1;L=LMAX} and {J=1,JMAX;K=KMAX;L=LMAX}. This could cause a minor glitch in the flow solution in this vicinity.

5. Modified the clean_holes script: No longer remove contents of WORK/HCUT_BC_DIR, since this forces the code to redo the projections, and thus also the interpolations. If this level of restart is required, the entire problem should be re-run from the beginning by removing the entire WORK directory.

6. Modification for LINUX: Modified interpolate.F90 to change strange behavior on LINUX with pgf90 compiler: in first pass through interpolate, zmaxg and zming were always computed to be nan after this statement:

         zmaxg = zmaxg + dztol
         zming = zming - dztol
   

   even though before this point, zming and zmaxg and dztol are all valid numbers. Mysteriously, the problem goes away if the previous lines:

         dxtol = rvap*(xmaxg - xming)
         dytol = rvap*(ymaxg - yming)
         dztol = rvap*(zmaxg - zming)
   

   are switched to this ordering:

         dztol = rvap*(zmaxg - zming)
         dytol = rvap*(ymaxg - yming)
         dxtol = rvap*(xmaxg - xming)
   

   With this change, version 5.1e seems to work OK on LINUX with the pgf90 compiler, after only limited testing, including the examples cases naca_airfoil, 30p30n_airfoil, and sphere.

Version 5.1d Modifications

1. Fixed bug in automatic hole cutting: Was never allowed to blank points which were part of a surface which had a specified boundary condition. This caused major problems with O-grid periodic surfaces, and some other cases. This, however, magnifies the problem with erroneous hole cuts at overlapping surfaces.

2. Fixed bug with GLOBAL namelist CARTX,CARTY,CARTZ inputs: These user input values were not overriding those calculated internally. Added CARTX, CARTY, CARTZ to HCUT namelist variables. Added description of CART[XYZ] to the Users Guide (these were previously undocumented).

3. Set default value of $GLOBAL orphfix to .TRUE.

4. Fixed bugs in peg_in.F90: was adding wrong IBDIR for duplicate boundary condition on O-grid periodic plane, and for IBTYP=54-59; was not processing IBTYP=51-59 properly; fixed IBTYP=-1 bug, adding of IBTYP=-1 extensions to TE was failing because logical variable wingcap was never initialized. Added -no_add_te flag to peg_in.

5. Fixed bug in peg_plot.F90: -sp and -dp flags were not coded properly. Caused CRAY or single-precision executables to not write the x,y,z,iblank data unless using -sp,-dp flags. Removed -dp flag; -sp flag can be used if the code was compiled as double-precision. Added description of -sp to Users Guide.

6. Fixed bug in peg_plot.F90: was assigning all ORPHAN points an iblank value of 101, ignoring the effect of ORPHFIX. By default, it now only sets iblank=101 for level ORPHANs; added the -noorphfix flag, which causes level 2 ORPHANS to be flagged with iblank=101.

7. Changed q5 field in peg_diag Changed output q file from 0.0 everywhere, to 0.0 at the field points, -1.0 for orphan points, and connecting mesh number at all legal l1 and l2 boundary points.

8. Modifications for compiling on LINUX: modified a number of files to fix compiler problems on LINUX systems: both Compaq Alpha Fortran 90 compiler, and Portland Group pgf90 compiler. Added --with-pgi flap to configure script.

9. Additional error checks: ensured that boundary condition indicies and IBDIR are within an acceptable range; ensure that mesh names given in BCINP and BOUNDARY input refer to existing meshes.

1. Investigate pruning away fringe levels greater than 3 (or 2) after the level 2 interpolation.

2. Investigate letting level 2 interpolation find stencils on boundary points, not just interior points.

3. For cases with a symmetry plane, add a feature to automatically cut away grid points which stray onto the wrong side of a symmetry plane

4. Projection issues: need to report number of points not projected due to the DISTANCE and ANGLE restriction.

5. Routines for manual hole cuts are still written in Fortran 77. Does this slow it down?

6. Execution times on a Cray C90 are about 5 times slower than an SGI Octane R12K 300 Mhz processor, which is about 10 to 20 times slower than they ought to be. Investigate and improve C90 performance.

7. Spew table showing how many stencils come from each zone in the peg_diag utility.

8. For symmetry boundary condition IBTYP=17 BC, it should read in grid to determine which physical plane is the symmetry plane, and accordingly automatically set XSYM, YSYM, or ZSYM.

Updated: Monday, 20-Dec-2004 14:09:43 PST WebWork: Stuart E. Rogers <rogers@nas.nasa.gov> NASA Responsible Official: Stuart Rogers