Parent Directory
| 
Revision Log
| Sticky Revision: |
Little restructuring for solving memory bug in MKL_free process and making sure LDU factorzation done only once.
Paso_MKL1 put back again
The n_F==n condition with potentially uninitialized value is fixed.
some unnecessary lines are removed
MKL test lines are removed
Direct solver switched to UMFPACK, till problem with MKL solved. When all unknows are eliminated then we switch to Jacobi preconditiner.
changes the default thresold, bug fixes in RS coupling and extra checks added in Solver
Bug that made MPI tests to fail is fixed.
OpenMP bug fixed and more openMP code added for cutting blocks.
Extra check is added for AMG levels. If coursening proccess does not eliminate at least 30% from the fine Matrix then it does not make sense to continue to the next level.
Migrating to block_sizes greather than 3 made easy.
Finially unit tests for AMG preconditioner for single equations as well as for systems. Minor bug in AMG is also fixed
first attempt of AMG on blocks
Paso_Pattern_coup changed to Paso_Pattern_YS for making clear that it refers to YAIR-SHAPIRA coarsenenig.
Saving sparse matrix used before for debuging is removed
Merging changes from new doco branch. All docstrings are now in reStructured text. A few email addresses have been fixes as well.
memory leack fixed
a problem with the sparse matrix unrolling fixed.
AMG throws an error now if block size not 1
Updating copyright notices
Minimum size of the coarsest level matrix option added to solver options
AMG now takes as an input Paso_options for selecting coarsenening algorithm and treshold variable.
Now one can choose coarsening method from python for AMG. Some commented lines are removed.
some profiling tools added
AMG now takes into account new SolverOptions class, namely one can select coarsening algorithms from python. Not all options are included into AMG, this will be implemented later on.
same adjustment of nameing
Some bugs are fixed.
AMG now works with blocksize 3. Logical block sizes replaced with real block size before starting AMG. But this happens only when we use AMG as a preconditioner.
Ruge Stuben and Aggregiation is added for testing performance of AMG solver.
AMG now checks how many elemens left. if n_F<500, then it does not coursen further.
AMG levels parameter moved to options file. Now one can set how many levels AMG will go in SetDefauls. Default value is 3.
some clean up. RS and Aggregiation are removed.
Some statistics changed in solver.c. In AMG stopping criteria now depend on levels specified. In loadMM conditions on pivot and lval reformulated to avoid numerical overflow.
Some statistical print outs are added in solver and loadMM to Paso_SparseMatrix method is added
AMG coupling parameter added to options file.
Openmp bug is fixed in Paso_cum_sum and some printouts removed from Solver_AMG
Codition is added in the corsening process to avoid rows with very small row sum.
Temporary solution for nusty rows: a condition added
Uses Jacobi as a smoother and beats any preconditioner in 2D(for some problems) where number of elemens in one direction is more that 200
Current version of AMG uses ILU for relaxation, however it is not stable when schur matrix becames more denser. For example it is not stable for -e paramenter is more 400 in 2D for convection problem
This commit cleans up the incompressible solver and adds a DarcyFlux solver in model module. Some documentation for both classes has been added. The convection code is only linear at the moment.
Some variables are initialized to get rid of compiler warnings
some unsed variables are removed
condition is added to check wheter row sum is ZERO and some new coarsening algorithm is also added
get Gauss-Seidel factorization moved solverAMG to getAMG. Stopping criterium for corsening changed to depend to the number of unknowns left
get Gauss-Seidel factorization moved solverAMG to getAMG. Stopping criterium for corsening changed to depend to the number of unknowns left
Direct solver is added to solve_AMG. It is now working but I guess need optimization.
Remove unused vars, do a bit of standardising of indentation, and use the new Pattern_coupling.h to eliminate some implicit function declarations in Solver_AMG.c
some bugs fixed in coresening methods
memory leak is fixed
minor
Current version of AMG. check levels for stoping applies presmoothing, but not ready yet
Fixed two typos that stopped the test suite from running. Also, gcc 4.3.2 issued several warnings not seen before: ignoring the return value of fscanf and using the wrong format with printf.
pattern multiply, binary operation are added
Closing the moreshared branch
Branch commit. Added files while the merge op did not add. Modified shake59 config for non-ken users.
This form allows you to request diffs between any two revisions of this file. For each of the two "sides" of the diff, enter a numeric revision.
| ViewVC Help | |
| Powered by ViewVC 1.1.26 |