ViennaCL: Added command-line switches for backend selection.

User can now switch backends conveniently from command line.
Available options:
- viennacl_view: Lists available and selected backends

ViennaCL: Added command-line switches for backend selection.

User can now switch backends conveniently from command line.
Available options:
- viennacl_view: Lists available and selected backends
- viennacl_backed: One out of CUDA, OpenCL, OpenMP (if enabled in configure)
- viennacl_opencl_device_type: cpu, gpu, accelerator (OpenCL device type selection)

Note: The current implementation couples petscsys with petscvec.
This will be addressed in a later commit.

show more ...

ViennaCL: Added support for mtdot on GPU.

Since ViennaCL does not support complex, this is for the time being just the same as mdot.

PetscSF: clean up example syntax

PetscSF: Hook up ex3 to test examples

PetscSF: Add example using derived data types

PetscSF: Free newly allocated MPI datatype objects

If the returned types from MPI_Type_get_contents are derived
datatypes (or dupped from base types) then the caller is responsible for
calling MPI_T

PetscSF: Free newly allocated MPI datatype objects

If the returned types from MPI_Type_get_contents are derived
datatypes (or dupped from base types) then the caller is responsible for
calling MPI_Type_free on them.

show more ...

ViennaCL: Added hybrid capabilities (CUDA, OpenCL, OpenMP)

Depending on whether the user configures with CUDA, OpenCL, or OpenMP,
the respective backends will be available in ViennaCL.
CUDA has prec

ViennaCL: Added hybrid capabilities (CUDA, OpenCL, OpenMP)

Depending on whether the user configures with CUDA, OpenCL, or OpenMP,
the respective backends will be available in ViennaCL.
CUDA has precedence over OpenCL, which has precedence over OpenMP.
If none of the backends is selected, ViennaCL is built in sequential mode.

Remaining tasks to better leverage these hybrid capabilities:
- Select backend at runtime (API + options)
- Eliminate copy-overhead for OpenMP backend
- Documentation, examples, and tests to demonstrate usage

show more ...

Merge branch 'jose/veccuda-witharray'

jose/veccuda-witharray:
The aim of this PR is to be able to create a VecCUDA vector
where the user provides the array on the GPU side.

Also, this PR (#466) inc

Merge branch 'jose/veccuda-witharray'

jose/veccuda-witharray:
The aim of this PR is to be able to create a VecCUDA vector
where the user provides the array on the GPU side.

Also, this PR (#466) includes a few fixes:
- Fix use of not initialized variable
- Fix CUDA_VERSION check in configure
- Increase the required CUDA_VERSION >= 7.5 for complex builds
(CUDA version 7.0 contains thrust/complex.h but it produces compilation errors)
- Stop using CUDA_SUCCESS and CUBLAS_STATUS_SUCCESS in CHKERRCUDA macro (otherwise the user must include cuda.h)

show more ...

Merge branch 'maint'

Fix formatting issues

Merge branch 'barry/housekeeping-1'

add missing PETSC_EXTERN to custom Fortran wrappers

removed a bunch of unneeded inclusions of xxximpl.h which exposed private data where it did not have to be exposed

also made more systematic the inclusion of the base xxximpl.h in the derived yyyimp

removed a bunch of unneeded inclusions of xxximpl.h which exposed private data where it did not have to be exposed

also made more systematic the inclusion of the base xxximpl.h in the derived yyyimpl.h
so the inclusion is done in the include file rather than in each source code that includes it

show more ...

Merge branch 'jose/maint/tabs-print-class' into maint

There appear to be bugs in VecAssemblyBegin/End_MPI_BTS() [petsc-maint] Undeterministic behaviour in VecSetValuesBlocked

So turning it off until this is resolved
Reported-by: Daniele Prada <dprada@u

There appear to be bugs in VecAssemblyBegin/End_MPI_BTS() [petsc-maint] Undeterministic behaviour in VecSetValuesBlocked

So turning it off until this is resolved
Reported-by: Daniele Prada <dprada@umail.iu.edu>

show more ...

Update output files after change in PetscObjectPrintClassNamePrefixType

Merge branch 'barry/fix-some-clang-warnings'

Merge branch 'barry/fix-source-clang'

Merge branch 'barry/cleanup-external-mat-solver-packages'

Merge branch 'barry/make-functions-static'

Example had print statements of complex numbers

Reported-by: nightly tests

Fix redundant/wrong VecDestroy assignment in VecCreate_MPICUDA_Private

Updated script for detecting visible symbols

Removed unneeded visibility of some symbols
Fixed some namespacing
Removed the use of unneeded implementation includes in the TAO code

fixed some naming spacing problems, some visibile function problems

changed findvisibleprivatefunctions to use shared library since static library still exposes PETSC_INTERN functions

fix compile warning