Renamed MainJob to PetscThreadsRunKernel and MainWait to PetscThreadsWait

Hg-commit: 44b125778f815ce49647dd42cf739fc68ee59169

Fixes for CUDA matrix assembly

Hg-commit: 0707079090e7635c2821aebf7507a3312bd3e9ea

Source code, makefiles, and output should not be executable

Hg-commit: b7a5317c393eff1636dfaf28894f18ed4657b2ca

Better handling of data ownership between Petsc and txpetscgpu package. The code in VecCUSPAllocateCheck (veccusp.cu) is much cleaner. Also, correct and efficient parallel-GPU SpMV is handled in the

Better handling of data ownership between Petsc and txpetscgpu package. The code in VecCUSPAllocateCheck (veccusp.cu) is much cleaner. Also, correct and efficient parallel-GPU SpMV is handled in the latest txpetscgpu tarball for ELL, COO, and DIA formats. Lastly, we also have initial usage of cusparse SpMV for CSR format. This works well for some matrices. Later, we'll have CUSPARSE ELL, COO, and HYB formats.

Hg-commit: d52e04f97453e9e2790982a5f6208d3f353cab08

show more ...

Remove MatSetUpPreallocation() and MatPreallocated(). MatSetUp() required if not preallocated. Unify error reporting.

Hg-commit: 9dd5fa086e48409435f8bacb456e7397da336895

Use rmap->N instead of rmap->n for setting the rows assigned to each thread.

Hg-commit: a2a12faf486ec289403636cd21b5dc5d009743bc

Fixed stupid bug in VecMDot_SeqPThread.
Trying to figure out the bug in pthread matrix class causing snes to have more iterations.

Hg-commit: 2ee5b725da5e34e60a1dbd9f275c8b17e67d8198

Disabled used of MatAssemblyEnd_SeqAIJPThread()

Hg-commit: 706dc04e26536ef1ad825c5b37012e405ba80fc7

merge release-3.2

Hg-commit: 92ecc738af2e54343bf34f9c473146d1479f768d

Fix a few format strings

Hg-commit: 71f429175a71a799d01a1cefdb2806fbcdd7ae81

turned off confusing logging for eigen estimator, cleaned up repartitioning code: removed my hack to protect parmetis and aded dumb processor aggregation w/o repartitioning.

Hg-commit: 668967087563a

turned off confusing logging for eigen estimator, cleaned up repartitioning code: removed my hack to protect parmetis and aded dumb processor aggregation w/o repartitioning.

Hg-commit: 668967087563aae3b1328539102ddbe5b38852f4

show more ...

Fixes so that GPU codes works on multiple configurations (1) CUDA 4.0/Thrust 1.4/CUSP 0.2 and (2) CUDA 4.1RC2/Thrust 1.5/Mercurial CUSP. Also some cleanup of the setup code for the GPU-enabled ILU p

Fixes so that GPU codes works on multiple configurations (1) CUDA 4.0/Thrust 1.4/CUSP 0.2 and (2) CUDA 4.1RC2/Thrust 1.5/Mercurial CUSP. Also some cleanup of the setup code for the GPU-enabled ILU preconditioner. Tests pass with and without txpetscgpu for both configurations listed above.

Hg-commit: ba236d6d587e2ce145bf01dd74dec900aeace9dc

show more ...

Fix MatInvertBlockDiagonal_SeqAIJ, the old code produced the transpose so it was worthless for nonsymmetric problems

Hg-commit: e0126a204d8c55dd75b3ab1dd2812c5c7fd8a2c9

changing values in seqaij matrices means must turn off valid flag for diagonal and block diagonal

Hg-commit: 5891761562790363cf7fce8d1e33071e9ee65b57

Many fixes for // comments in C source

Hg-commit: 6087d5ff9c9480e0ff8b9d44422a6be84041fc11

add level info in manpages

Hg-commit: 3bba9593b9f6541d89b764edd2655e02309f98e3

Fix MatLUFactorNumeric_UMFPACK for use with a different matrix than symbolic factorization was done with, eliminate raw pointers in favor of reference counted Mat

Hg-commit: 6fe2be003849e79921432030

Fix MatLUFactorNumeric_UMFPACK for use with a different matrix than symbolic factorization was done with, eliminate raw pointers in favor of reference counted Mat

Hg-commit: 6fe2be003849e799214320306a24a1947982a71e

show more ...

fix to MatGetRowIJ_SeqAIJ_Inode_Symmetric() to prevent looking past the end of the ja[] in the sparse matrix data strucre which could happen if the final row of the matrix did not have a diagonal ent

fix to MatGetRowIJ_SeqAIJ_Inode_Symmetric() to prevent looking past the end of the ja[] in the sparse matrix data strucre which could happen if the final row of the matrix did not have a diagonal entry

Hg-commit: b51e2c6bacd63d4c536eb6ae91fc92efac64d391

show more ...

free memory allocated by MatInvertBlockDiagonal_SeqAIJ()

Hg-commit: d7cde46437c5fbb6c3de4b21d8b095069e36c231

MatInvertBlockDiagonal_SeqAIJ() now properly returns values if they are already valide

Hg-commit: 8709ce6f8fb07a8c87f43691b3bbe87074d9789d

bugfix of MatLoad() for seqaij and seqbaij: user might provide local size instead of global size

Hg-commit: 47523900f112a8d3c822027cf87f709ec6f83131

fix c++/matlab link error

Hg-commit: 4ae152c822e0f891ec98d838d04da4db12bc8894

fix prototype for fortranmultaddaij()

Hg-commit: 84af361fbd6f72fdb257a654b7f1f73e39bf0ad0

fixed bad merge

Hg-commit: d81cbcaf7d9958449fb47f9094433f68325f8ffd

lots of fixed for nightly build

Hg-commit: 2a2d0cde08a4182a8c0c4b22619801648220ba54