add MatFDColoringSetUp(); rename MatFDColoringCreate_xxx -> MatFDColoringSetUp_xxx

set default bcols s.t. dy takes 50% of memory as matrix

Merge branch 'hzhang/snes-jacobiancoloropt' of bitbucket.org:petsc/petsc into hzhang/snes-jacobiancoloropt

optimize MatFDColoring routines for seqaij using brows and bcols

cleanup MatFDColoringApply

rm old members of _p_MatFDColoring: rows, columnsforrow and vscaleforrow (replaced by matentry)

replace old impl of MatFDColoring routines with new ones.

Use MatEntry struct for MatFDColoring routines for seqaij and seqbaij matrices

when htype=wp, dx is a constant scalar, thus does not need vscale

use struct MatEntry such that (row,col,valaddr) are stored contigously for efficient accessing

small cleanup of MatFDColoringCreate_MPIAIJ_new()

small cleanup

implement MatFDColoringCreate_MPIAIJ_new()

optimize MatFDColoringApply_SeqBAIJ() by inserting block column values to J

fix memory leak

cleanup MatFDColoringCrete and MatFDColoringApply for SeqAIJ

experimenting different ways of storing row, col and den2sp -- rowcolden2sp3 gives the best performance

mv redundant valications in MatFDColoringApply_xxx()

mv checking !mat->assembled from MatFDColoringCreate_xxx() to interface MatFDColoringCreate

new impl of MatFDColoringxxx_SeqAIJ works now :-)

optimize MatSetValues in fdcoloring using an array of spidx - see MatTransposeColoringCreate_SeqAIJ()
-------------------

when htype=wp, dx is a constant scalar, thus does not need vscale

use struct MatEntry such that (row,col,valaddr) are stored contigously for efficient accessing

small cleanup of MatFDColoringCreate_MPIAIJ_new()

small cleanup