Merge branch 'hzhang/mat-coloring'

rename Mat_MatMatTransMult *art -> *abt; *multtrans -> *abt

remove container from MatMatTransposeMult_SeqAIJ_SeqAIJ()

cleanup matmatmult.c

replace previous impl of '-matden2sp_brows >0' with the new one which only checks non-zeros of Cden in MatTransColoringApplyDenToSp_SeqAIJ()

optimize MatTransColoringApplyDenToSp_SeqAIJ() for the case '-matden2sp_brows -1'

add a new implementation of '-matrart_brow' with matrix coloring: only loop over
sparse non-zeros entries of C_dense.

bugfix for the case brows=0

cleanup routines for MatTransposeColoring; set MatTransposeColoring->brows>0 as default

implement MatTransColoringApplyDenToSp_SeqAIJ() with rowblock-wise sweeping which shows 40% faster for pflotran tests

trivial cleanup of MatTransposeColoringCreate_SeqAIJ

enable runtime switch between A^TA and AA^T when using gamg

raise erorr flag when inode is used for coloring of RARt and ARt

PetscGetTime -> PetscTime

timing each segment of A*R^T when coloring is used

rm INEFFICIENT_ALGORITHM(sparse inner-product) in MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ();
rm dead routines MatApplyPAPt_Symbolic_SeqAIJ_SeqAIJ() and MatApplyPAPt_Numeric_SeqAIJ_SeqAIJ()

Merged petsc/petsc into rmills/petsc master.

implement MatTransposeMatMult_SeqAIJ_SeqDense() - works, but not done yet

new dev on MatTransposeMatMult_aij_dense

Merge branch 'jed/matmatmult-sort'

New version provides in the neighborhood of 20% speedup for MatMatMult
as used in PCGAMG.

* jed/matmatmult-sort:
MatMatMultSymbolic_SeqAIJ_SeqAIJ: switch to "so

Merge branch 'jed/matmatmult-sort'

New version provides in the neighborhood of 20% speedup for MatMatMult
as used in PCGAMG.

* jed/matmatmult-sort:
MatMatMultSymbolic_SeqAIJ_SeqAIJ: switch to "sort" version as default
segbuffer: g++ does not like casting int*restrict* to void*, void** okay
MatMatMult_SeqAIJ_SeqAIJ_Sort: new implementation using char flags and sorting
segbuffer: add PetscSegBufferGetInts() to encourage use of PETSC_RESTRICT
segbuffer: create wrapper structure
segbuffer: add PetscSegBufferGetSize() and PetscSegBufferUnuse()
segbuffer: fix bug in which seg->used was not reset

show more ...

MatMatMultSymbolic_SeqAIJ_SeqAIJ: switch to "sort" version as default

This version is showing approximately 25% performance improvement in
GAMG tests and so far is never slower.

Merge branch 'jed/matmatmult-sort' into jed/fix-optionsinsertfile-large

Branch contains segbuffer enhancements in addition to
MatMatMult_SeqAIJ_SeqAIJ_Sort. I need the segbuffer enhancements, one of

Merge branch 'jed/matmatmult-sort' into jed/fix-optionsinsertfile-large

Branch contains segbuffer enhancements in addition to
MatMatMult_SeqAIJ_SeqAIJ_Sort. I need the segbuffer enhancements, one of
which is a portability fix that was applied after the
MatMatMult_SeqAIJ_SeqAIJ_Sort commit.

* jed/matmatmult-sort:
segbuffer: g++ does not like casting int*restrict* to void*, void** okay
MatMatMult_SeqAIJ_SeqAIJ_Sort: new implementation using char flags and sorting
segbuffer: add PetscSegBufferGetInts() to encourage use of PETSC_RESTRICT
segbuffer: create wrapper structure
segbuffer: add PetscSegBufferGetSize() and PetscSegBufferUnuse()
segbuffer: fix bug in which seg->used was not reset

show more ...

fix errors made in previous change

log events of symbolic and numeric mat-mat operations

Merge branch 'madams/petsc-mem'

* madams/petsc-mem:
MatView_SeqAIJ_ASCII: fix out-of-bounds access when matrix has 0 rows
MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ: fix out-of-bounds memory access