| 783ef271 | 25-Sep-2009 |
Hong Zhang <hzhang@mcs.anl.gov> |
add icc(0) with matorderings under new data structure
Hg-commit: 5b99ce0799b371a259538425a489fed8c02ef64a |
| 8bfb2e2b | 24-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
MatLUFactorSymbolic_SeqBAIJ_newdatastruct: Memory bug fixed, no performance improvement?!
Hg-commit: 71e63303ca85bde65acefd2082bcf9625640a4b6 |
| faca2338 | 24-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
MatLUFactorSymbolic_SeqBAIJ_newdatastruct: Need to fix memory bug
Hg-commit: fb37f7e06dc3545c0ba5212b7ea619b47cc21a55 |
| 29b92fc1 | 18-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
Replaced while loops with for loops in MatSolve_SeqBAIJ_<bs>_newdatastruct routines
Replaced pointer++ notation with array indexing
Hg-commit: b4fbf631d6de706335546019b23efb288e9bf1ca |
| 6464896e | 18-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
Replaced while loop with for loop in MatSolve_SeqBAIJ_<bs>_NaturalOrdering_newdatastruct routines
Hg-commit: ad9b2ff056d66854a0d78a9c316c5ce9cc594d9d |
| ce3d78c0 | 18-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
MatLuFactorSymbolic_SeqAIJ_newdatastruct: Need to fix the memory error
Hg-commit: 094afb1681ce16b67b854caf2518c8292aac3f36 |
| 715a17b5 | 16-Sep-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
Barry, I'm getting this in petsc4py testsuite against petsc-dev
[0]Total space allocated 248 bytes
[ 0]116 bytes MatToSymmetricIJ_SeqAIJ() line 67 in src/mat/impls/aij/seq/ij.c
[0] MatGetRowIJ
Barry, I'm getting this in petsc4py testsuite against petsc-dev
[0]Total space allocated 248 bytes
[ 0]116 bytes MatToSymmetricIJ_SeqAIJ() line 67 in src/mat/impls/aij/seq/ij.c
[0] MatGetRowIJ_SeqBAIJ() line 1176 in src/mat/impls/baij/seq/baij.c
[0] MatGetRowIJ() line 6017 in src/mat/interface/matrix.c
[0] MatOrdering_ND() line 18 in src/mat/order/spnd.c
[0] MatGetOrdering() line 185 in src/mat/order/sorder.c
Could you please check
MatGetRowIJ_SeqBAIJ()/MatRestoreRowIJ_SeqBAIJ(), specifically for the
combination symmetryc=PETSC_TRUE and oshift=1 ?? ... After a VERY fast
look, it seems the stuff allocated in MatToSymmetricIJ_SeqAIJ() is
never deallocated...
It was not handling the symmetric + oshift case properly. If oshift was given it ignored the symmetric format that was created hence the memory leak PLUS a bug that it did not actually use the symmetric version created. Fixed by increasing the tia and tja directly in the symmetric plus oshift case.
Hg-commit: 7b2e613cf29c5a7a5a8d734070f335e5e8dd4cda
show more ...
|
| 8f690400 | 13-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
Added MatSolve_SeqBAIJ_<bs>_newdatastruct routines for bs = 2...N
Fixed linking issue with PetscFreeSpaceContiguous_newdatastruct
Added test cases for new data structure
Hg-commit: c4df67b1f5fafce28
Added MatSolve_SeqBAIJ_<bs>_newdatastruct routines for bs = 2...N
Fixed linking issue with PetscFreeSpaceContiguous_newdatastruct
Added test cases for new data structure
Hg-commit: c4df67b1f5fafce2826457ab83b7b702fe8e2215
show more ...
|
| 7fa3a6a0 | 11-Sep-2009 |
Hong Zhang <hzhang@mcs.anl.gov> |
Add MatILUFactorSymbolic_SeqBAIJ_newdatastruct()
ILU(p) works for baij matrix now. Next, implement MatSolve for non-natural ordering.
Hg-commit: 0266bb75bef2e7d0fc654ca27018f9469c7e4067 |
| 16a2bf60 | 11-Sep-2009 |
Hong Zhang <hzhang@mcs.anl.gov> |
add MatILUFactorSymbolic_SeqBAIJ()
ilu(0) works. ilu(p), p>0 is not done yet
Hg-commit: fca544fa7099605d92c6e4a4a1ba1211b5314d50 |
| cee9d6f2 | 04-Sep-2009 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
Added MatSolve Natural ordering routines with new data structure for bs = 2,3,4,5,6,7
Hg-commit: cf50a8af7b89ac881c3de178a4b43c433037203c |
| 27019359 | 02-Sep-2009 |
Hong Zhang <hzhang@mcs.anl.gov> |
Shri's 1st dev: MatSolve_SeqBAIJ_2_NaturalOrdering_newdatastruct()
Hg-commit: a24a3be84921dd9594077a92234c6f7d90dfc1b7 |
| 0c4e8012 | 19-Aug-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
removed Contributed by: in documentation and source for PETSc developers
Hg-commit: 19546b65bd4769308070770659a516cb6b59906a |
| a1c3900f | 19-Aug-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
finished adding -ksp_cg_single_reduction
fixed bug in MatDuplicate() for SBAIJ with jshort
Hg-commit: ab3bd021969125c78e64ec0e5509cc65844f43bc |
| ee54c7ee | 15-Aug-2009 |
Hong Zhang <hzhang@mcs.anl.gov> |
bugfix for MatMult_SeqBAIJ_1()
Hg-commit: 80a8368f0ed203da6f4b0b349cf49decc5a3006d |
| 2162cab8 | 13-Aug-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
minor optimization to MatMult_SeqBAIJ_1()
Hg-commit: f8fed164bdb43e3e06c3e548229441ccb2f45bb5 |
| 8397fe1a | 27-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
inclined PetscMemcpy() and PetscMemzero(), stupid maybe but let's see how it goes
Hg-commit: a0c866eb70319ddc2a5d86cf7b851aed3de2afd0 |
| f6d58c54 | 25-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
finished support for MatFDColoringCreate() etc for BAIJ matrices
Hg-commit: 61c9609b5ffa140392733ce982b1260ca27cbd92 |
| 3acb8795 | 23-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
more work on adding FD computation of Jacobian for BAIJ matrices
Hg-commit: 89075e85186765069f32cddc814471a4b6a51957 |
| ab040260 | 20-Jul-2009 |
Jed Brown <jed@59A2.org> |
Enable SOR for BAIJ with block size 1.
Includes bug fix in MatPBRelax_SeqBAIJ_1, but this function was never
called before.
Hg-commit: b6761fec08841f1e9f0b5b70e2ab44dbd28856cd |
| ae3dafa0 | 17-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
pull bug fix for The situation is the following: I'm calling MatSetValues with the flag ADD_VALUES and with matrix option MAT_IGNORE_ZERO_ENTRIES. I'm inserting multiple values at once, multiple colu
pull bug fix for The situation is the following: I'm calling MatSetValues with the flag ADD_VALUES and with matrix option MAT_IGNORE_ZERO_ENTRIES. I'm inserting multiple values at once, multiple columns and rows, so I provide a rank-2 matrix of values. As I'm calling this from fortran I'm also using MAT_COLUMN_ORIENTED. Now for provided rows that are not owned by the process, it jumps to mpiaij.c:394 (line numbers as in petsc-dev). On line 399, it checks for zero entries, but only checks the very first entry of the (non-owned) row. If however other entries of that same row are nonzero, the entire row is still dropped. Note that this is independent of row_oriented/column_oriented as line 396 does exactly the same.
If I don't set the option MAT_IGNORE_ZERO_ENTRIES the problem disappears. In that case however we would either have to preallocate substantially more nonzeros, or complicate the matrix assembly in our code by taking out the zero entries ourselves and call MatSetValues for each entry seperately.
Hg-commit: f09f0ab443e582d40fb2b646b2461bcc61f6e436
show more ...
|
| b400d20c | 17-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
From: s.kramer@imperial.ac.uk
Subject: non-local values being dropped in MatSetValues
Date: July 17, 2009 7:44:09 AM CDT
To: petsc-users@mcs.anl.gov
Reply-To: petsc-users@mcs.anl.gov
The si
From: s.kramer@imperial.ac.uk
Subject: non-local values being dropped in MatSetValues
Date: July 17, 2009 7:44:09 AM CDT
To: petsc-users@mcs.anl.gov
Reply-To: petsc-users@mcs.anl.gov
The situation is the following: I'm calling MatSetValues with the flag ADD_VALUES and with matrix option MAT_IGNORE_ZERO_ENTRIES. I'm inserting multiple values at once, multiple columns and rows, so I provide a rank-2 matrix of values. As I'm calling this from fortran I'm also using MAT_COLUMN_ORIENTED. Now for provided rows that are not owned by the process, it jumps to mpiaij.c:394 (line numbers as in petsc-dev). On line 399, it checks for zero entries, but only checks the very first entry of the (non-owned) row. If however other entries of that same row are nonzero, the entire row is still dropped. Note that this is independent of row_oriented/column_oriented as line 396 does exactly the same.
If I don't set the option MAT_IGNORE_ZERO_ENTRIES the problem disappears. In that case however we would either have to preallocate substantially more nonzeros, or complicate the matrix assembly in our code by taking out the zero entries ourselves and call MatSetValues for each entry seperately.
Hg-commit: d28deff5e1b8c4e99cf2d919d60e6073ac99448f
show more ...
|
| 8a31d241 | 17-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
fixed up bad #if !defined(__FNORM in a bunch of new fxxxx.h headers
Hg-commit: e4f7d048668222896b45e4296bf4a151702b411e |
| d4bf62d1 | 17-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
finished removing dead src/inline code and putting Fortran kernels in subdirectories where they are used. got rid of BGL fortran kernels, now handled by a macro
Hg-commit: b3833a5be95213a4ef66922f90
finished removing dead src/inline code and putting Fortran kernels in subdirectories where they are used. got rid of BGL fortran kernels, now handled by a macro
Hg-commit: b3833a5be95213a4ef66922f90e9b536af5e3fb6
show more ...
|
| a4005a5d | 17-Jul-2009 |
Barry Smith <bsmith@mcs.anl.gov> |
more moving fortran kernels
Hg-commit: 5ed9048967492d9347972f95e32e80dabb44f7ae |