Non-numeric optimizations focused on AIJ, MatFDColoring, and DMCreateMatrix_DA_*AIJ, looking to improve performance in GPU environments

1) PetscCalloc*() now uses system calloc()
2) Merged some Pets

Non-numeric optimizations focused on AIJ, MatFDColoring, and DMCreateMatrix_DA_*AIJ, looking to improve performance in GPU environments

1) PetscCalloc*() now uses system calloc()
2) Merged some PetscMalloc*()
3) Eliminated unneeded PetscCalloc*()
4) Removed some memory allocations and copies in MatFDColoringSetUp(), added local variables for better compiler optimization
5) Added MatSetValues_SeqAIJ_SortedFull(), added MatSetOption(MAT_SORTED_FULL)
6) Optimized DMCreateMatrix_DA_*AIJ for nonperiodic case to automatically have sorted columns (faster MatSetValues() times)
7) Eliminated call to PetscMemzero() in PetscFree()

Commit-type: style-fix, feature

show more ...

Added PetscCUDAInitialize() which the user may call in the body of the program, -cuda_initialize 0 prevents PetscInitialize() from initializing CUDA

Add -cuda_initialize 0 as options database for CU

Added PetscCUDAInitialize() which the user may call in the body of the program, -cuda_initialize 0 prevents PetscInitialize() from initializing CUDA

Add -cuda_initialize 0 as options database for CUDA tests so machines won't load cuBLAS unless actually
utilizing it on the machine, update examples to filter option-cuda_initialize

Reported-by: Karl Rupp <me@karlrupp.net>
Commit-type: feature

show more ...

Various improvements for GPUs (mostly for performance and CUDA)

1) Add VecPinToCPU() for CUDA vector and matrices
2) Move initialization of cuBLAS to PetscInitialize() since it takes 1/2 second and

Various improvements for GPUs (mostly for performance and CUDA)

1) Add VecPinToCPU() for CUDA vector and matrices
2) Move initialization of cuBLAS to PetscInitialize() since it takes 1/2 second and distorts timing with -log_view
3) Add logging for DMCreateMatrix (for large meshes this is very large)
4) Add VecGet/RestoreArrayWrite() to prevent unneeded copies from GPU (only implemented so far for CUDA);
added a small number of usages in the source so that snes tutorials ex19 does not do unneeded communication from the GPU
5) Automatically convert MAIJ matrices to AIJ for CUDA since they are not yet supported natively in PETSc's CUDA matrix implementation
6) Pinned objects should still use the CUDA/ViennaCL versions of Destroy to clean up the GPU stuff

Commit-type: feature

show more ...

Merge remote-tracking branch 'origin/barry/add-petscarrayxxx-ops' [PR #1771]

Merged in barry/add-petscoptionspushpop (pull request #1782)

Add PetscOptionsPush() and PetscOptionsPop()

Remove dead Matlab code, fix typo, fix caste needed for const free

Commit-type: portability-fix
Reported-by: Satish Balay <balay@mcs.anl.gov>

Merge remote-tracking branch 'origin/maint'

remove code that assumes MPI_2INT is a macro.

This is a continuation of the fix in bcb1dc63284765edb2c98905bc313bc2d01136a1

/p/work1/bctacct/sw/app/math/PETSc/3.11.2/include/petscmath.h:708:19: err

remove code that assumes MPI_2INT is a macro.

This is a continuation of the fix in bcb1dc63284765edb2c98905bc313bc2d01136a1

/p/work1/bctacct/sw/app/math/PETSc/3.11.2/include/petscmath.h:708:19: error: ‘MPI_2INT’ redeclared as different kind of symbol
#define MPIU_2INT MPI_2INT
^
/p/work1/bctacct/sw/app/math/PETSc/3.11.2/src/sys/objects/init.c:76:14: note: in expansion of macro ‘MPIU_2INT’
MPI_Datatype MPIU_2INT = 0;
^
In file included from /p/work1/bctacct/sw/app/math/PETSc/3.11.2/include/petscsys.h:134:0,
from /p/work1/bctacct/sw/app/math/PETSc/3.11.2/src/sys/objects/init.c:9:
/p/app/sgi/mpt/mpt-2.15/include/mpi.h:143:2: note: previous definition of ‘MPI_2INT’ was here
MPI_2INT = 34,
^

show more ...

Add PetscArraycmp(), PetscArraycpy(), PetscArrayzero(), PetscArraymove(), update code to use new forms

Replace a couple multi-mallocs with a single malloc
Replace backwards loops in MatSetValues_*()

Add PetscArraycmp(), PetscArraycpy(), PetscArrayzero(), PetscArraymove(), update code to use new forms

Replace a couple multi-mallocs with a single malloc
Replace backwards loops in MatSetValues_*() with PetscArraymove()

Commit-type: style-fix, feature

show more ...

Improve clarity of -help message value of argument form <old : new> to <now % : formerly %> as suggested by Karl Rupp

Commit-type: style-fix, feature
Thanks-to: Karl Rupp <me@karlrupp.net>

The -help message for PetscOptionInt(), PetscOptionsReal() etc now includes the newly set value as well as the previous value

Example:

./toy -help -tao_catol 100 now produces

-tao_catol <0. : 100.

The -help message for PetscOptionInt(), PetscOptionsReal() etc now includes the newly set value as well as the previous value

Example:

./toy -help -tao_catol 100 now produces

-tao_catol <0. : 100.>: Stop if constraints violations within (TaoSetConstraintTolerances)

Commit-type: feature
Reported-by: Robert Nr Nourgaliev <nourgaliev1@llnl.gov>

show more ...

Fix typo in PetscOptionsValidKey manual page

Commit-type: documentation
Reported-by: Jed Brown <jed@jedbrown.org>

Fix typos in PetscOptions manual pages

Commit-type: documentation
Reported-by: Patrick Sanan <patrick.sanan@gmail.com>

Add collectivity information to the PetscOptions methods

Commit-type: documentation
Reported-by: Jed Brown <jed@jedbrown.org>

Enhancements to PetscOptionsPush() and PetscOptionsPop()

1) PetscOptionsLeft() -options_left now report unpoped options
2) Completed manual pages with cross-links
3) Removed unnecessary double linke

Enhancements to PetscOptionsPush() and PetscOptionsPop()

1) PetscOptionsLeft() -options_left now report unpoped options
2) Completed manual pages with cross-links
3) Removed unnecessary double linked list construct
4) Additional error checking

Commit-type: feature, documentation, example
Thanks-to: Jed Brown <jed@jedbrown.org>
Thanks-to: Patrick Sanan <patrick.sanan@gmail.com>

show more ...

Add PetscOptionsPush() and PetscOptionsPop()

Commit-type: feature
Reported-by: Sophie Blondel <sblondel@utk.edu>

Merge remote-tracking branch 'origin/maint'

CHKERRQ should not be used for PetscInitialize().

Merged in barry/update-collective-on (pull request #1744)

Update the use of Collective on in the manual pages to reflect the new style

Merged in barry/more-openmp-work (pull request #1735)

Barry/more openmp work

Update the use of Collective on in the manual pages to reflect the new style

Commit-type: style-fix, documentation
Thanks-to: Patrick Sanan <patrick.sanan@gmail.com>

Man pages: remove Concepts: fields

These fields were previously stripped from the man pages by logic removed in
21a59cba2737d49dc2f0bd12c08db0d2a3f3f209

Remove these fields from all man pages (but

Man pages: remove Concepts: fields

These fields were previously stripped from the man pages by logic removed in
21a59cba2737d49dc2f0bd12c08db0d2a3f3f209

Remove these fields from all man pages (but not from examples).

This is accomplished with GNU sed (gsed on OS X), with the following commands.
*Warning* that this type of command can corrupt a .git directory,
so be cautious in reusing or modifying these commands. They first look
for and delete matching lines with a following line consisting of only whitespace,
and then delete any remaining matching lines.

find src -type f -not -path "*/examples/*" -not -name "*.html" -not -name "*.bib" -exec gsed -i '/Concepts:/ {N; /\n\s*$/d}' {} +
find src -type f -not -path "*/examples/*" -not -name "*.html" -not -name "*.bib" -exec gsed -i '/Concepts:/d' {} +
find include -type f -not -path "*/examples/*" -not -name "*.html" -not -name "*.bib" -exec gsed -i '/Concepts:/ {N; /\n\s*$/d}' {} +
find include -type f -not -path "*/examples/*" -not -name "*.html" -not -name "*.bib" -exec gsed -i '/Concepts:/d' {} +

Hints on the sed command obtained from: https://unix.stackexchange.com/questions/100754/how-to-delete-a-specific-line-and-the-following-blank-line-using-gnu-sed

show more ...

Attempt to select and determine MKL libraries that use OpenMP

PetscLogView() now prints the number of OpenMP threads

Using ex43 -n 10000000 -log_view | grep VecDot
On my Apple Mac using the free MK

Attempt to select and determine MKL libraries that use OpenMP

PetscLogView() now prints the number of OpenMP threads

Using ex43 -n 10000000 -log_view | grep VecDot
On my Apple Mac using the free MKL installed from Intel I got speed down on vector operations
Using the MKL that came with the Intel Fortran I also got speed down
But using the MKL that comes with Matlab I got some speedup

Commit-type: feature

show more ...

Added -omp_num_threads and -omp_view options to PetscInitialize()

Commit-type: feature

Merged in rmills/feature-cuda-show-device-assignment (pull request #1708)

Device assignments by rank are now printed by -cuda_show_devices.

Approved-by: BarryFSmith <bsmith@mcs.anl.gov>