changed variable in copy down of compressed row matrix for CUDA

Hg-commit: ecdec0adc609d02872d7af9022de216ddfbc051f

fixed a couple of things in aijcuda.c now works with mpiaijcuda on one process with zero nonzero rows in off diagonal part of matrix. Also only checks for compressed row flag AFTER MatAssembly_SeqAIJ

fixed a couple of things in aijcuda.c now works with mpiaijcuda on one process with zero nonzero rows in off diagonal part of matrix. Also only checks for compressed row flag AFTER MatAssembly_SeqAIJ() still crashes in cudastruct->mat->column_indices.assign(thrust::make_permutation_iterator(a->j,ii),thrust::make_permutation_iterator(a->j,ii)+a->nz);
I'll leave that to Victor

Hg-commit: 5c1bb66acba9adc4bdad19b438688bf4829716a2

show more ...

small change matassemblyend cuda

Hg-commit: e73f690591f57e5189181f408988401ef7311f82

small fix cuda assemblyend

Hg-commit: 729aeeb248fd8209813f3b3ca05ddb9884ac9c7c

some chnges to matasemblyend cuda

Hg-commit: 1ad7a66ccceb4ec05d70cd8053883be4c58ade5d

edited matmultadd_seqaijcuda in the case of yy===zz

Hg-commit: 46e2ad6353362731ee5424153446c796e3454319

added workvector in case vectors are the same (for test)

Hg-commit: 2f2337fb3af42ef2d8c3968ea23d14c1a4359704

added first attempt at matmultadd and cprow GPU matrices

Hg-commit: 4e32de76122211cf17356f3ff1256a3a592c1404

changed cuda matrices' spptr to point to a struct which contains multiple pointers in preparation for the
scatter etc necessary for MatMultAdd_SeqAIJCUDA.

Hg-commit: e28b48f3c0fb2f2c8d3d7c9eb272cb23

changed cuda matrices' spptr to point to a struct which contains multiple pointers in preparation for the
scatter etc necessary for MatMultAdd_SeqAIJCUDA.

Hg-commit: e28b48f3c0fb2f2c8d3d7c9eb272cb23cca018ca

show more ...

put error handling around CUDA code
added implementation details to sacuda

Hg-commit: 58261034b163e1fe25d4288ba1a3e6a3b9495747

added base files and register functions for SACUDA preconditioner (dependent on PETSC_HAVE_CUSP_DEV)
changed seqaijcuda matrix to check for old GPU matrix and delete if necessary on assembly end

added base files and register functions for SACUDA preconditioner (dependent on PETSC_HAVE_CUSP_DEV)
changed seqaijcuda matrix to check for old GPU matrix and delete if necessary on assembly end

Hg-commit: deb4fdacb51236e6e6a5414f90d282b684cc0e0d

show more ...

Reordered MatLoad input arguments.
Updated custom fortran MatLoad stub.

Hg-commit: 1b34440d31d2a1eb45d0adecc1d80c3909b37054

merge

Hg-commit: 9f93e4cbb9c0506d07517ab591fff486f16e0acb

bugfix: call MatDuplicate_SeqAIJ() outside of PetscOptionsBegin/End()
otherwise, the help menu is truncated.

Hg-commit: 4c70fe6a2571553912153dcb11b45b3bbbf2ccf1

changed cudathreadsynchronize to a defined macro to make it easy to turn on and off

Hg-commit: ecc7c3e3a24314d1118394a5f2fde4e1a8507de3

put synchronization calls into seqcuda for correct timing tests

Hg-commit: a6ee1eacedaf03fa88dba92740088b7473cabacd

header

Hg-commit: 52d432c5960b5dffd7233025fcebd11127da2a11

fixed header

Hg-commit: f0611f7492909907b34c77502114cbfd6be95aee

cuda extern c problem fix aijcuda

Hg-commit: d7e029ecb9b4e2109f1f94a92fc302b89d065c19

added include of cudavecimpl to aijcuda

Hg-commit: 5bf05e0f850b0d40fc088a77f251d481c55e0abf

changed function call in matmult_seqcuda to static version

Hg-commit: 81c3f5d206529a294aefc90e815b3ab9c7260231

fixed const problem in matmult aijcuda

Hg-commit: 19a2b332ae6b6b3f9488fb6b140fe31ac264cd27

fixed compressed row multiply on cuda (note this doesn't use cuda currently)

Hg-commit: aaa061b638198d7be447734c5c90cfed74845a46

added MATAIJCUDA superclass
added ex19 test example for CUDA

Hg-commit: ed1acf7471d5105d9d9830786cf3112b5987d7cf

enable '-mat_superlu_equil no' for superlu ilu

Hg-commit: 4aa1d22d509f406ac625f7758bb8bbbb2675d48e