Merge branch 'hzhang/update-networkex'
Merge branch 'pr395/dmeiser/fix-veccuspgetarraywrite/master'
fix --with-cuda and --with-cusp to compile with --with-64-bit-indicesDo not know if cuda and cusp will work with --64-bit-indices since my test box cuda/cusp does not workReported-by: Rajeev Das <
fix --with-cuda and --with-cusp to compile with --with-64-bit-indicesDo not know if cuda and cusp will work with --64-bit-indices since my test box cuda/cusp does not workReported-by: Rajeev Das <rajeev.das@cnl.ca>
show more ...
Deprecate PetscViewerSetFormat() since it better to use PetscViewerPush/PopFormat() and usingPetscViewerSetFormat() can lead to confusing situations for users when later viewers do not recognize the
Deprecate PetscViewerSetFormat() since it better to use PetscViewerPush/PopFormat() and usingPetscViewerSetFormat() can lead to confusing situations for users when later viewers do not recognize the setformat and thus do not produce any information.Reported-by: Jed Brown <jed@jedbrown.org>
veccusp: add VecCUSP(Get/Restore)CUDAArray(Read/Write/ReadWrite).The main purpose of this patch is to add a family of functions to getraw device pointers from Vec's of type VECCUSP. The patch cor
veccusp: add VecCUSP(Get/Restore)CUDAArray(Read/Write/ReadWrite).The main purpose of this patch is to add a family of functions to getraw device pointers from Vec's of type VECCUSP. The patch corrects anddocuments the synchronization of the vector's device data with the host.This patch also moves all non-cuda code from veccusp.cu to a newcompilation unit veccusp.c.List of changes:- Add VecCUSPGetCUDAArrayRead().- Add VecCUSPRestoreCUDAArrayRead().- Add VecCUSPGetCUDAArrayWrite().- Add VecCUSPRestoreCUDAArrayWrite().- Rename VecCUSPGetCUDAArray() -> VecCUSPGetCUDAArrayReadWrite() to be consistent with VecCUSPGetArray*() functions.- Rename VecCUSPRestoreCUDAArray() -> VecCUSPRestoreCUDAArrayReadWrite() to be consistent with VecCUSPGetArray*() functions.- Fix host-device synchronization in VecCUSP(Get/Restore)CUDAArrayReadWrite().- Improve documentation of all functions for access to CUDA and CUSP device pointers to better explain the host-device synchronization behavior.- Remove redundant calls to VecCUSPAllocateCheck() in VecCUSPGetCUDAArray*() functions. Device memory allocation is handled in the VecCUSPGetArray*() functions.- Mark VecCUSPGetCUDAArray*() functions as advanced.- Remove a duplicate declaration in petscvec.h- Move non-cuda code from veccusp.cu to veccusp.c.- Rename veccusp.cu to veccusp2.cu so that there are no duplicate *.o targets.- Restructure cuspvecimpl.h header so that it can be used with NVCC and CC- Move several function declarations to cuspvecimpl.h so that they can be used in veccusp2.cu and veccusp.c.- Add a documentation comments regarding missing Fortran bindings for several functions.
Merge branch 'master' into tisaac/dmforest* master: (143 commits) Revert "revolve: c++ compile fix" - as revolve now does not require 'bool' anymore revolve: c++ compile fix revolve: enable t
Merge branch 'master' into tisaac/dmforest* master: (143 commits) Revert "revolve: c++ compile fix" - as revolve now does not require 'bool' anymore revolve: c++ compile fix revolve: enable tests and add to nightlybuilds fix download url for revolve remove outdated unused script removed outdated use of the term petsc-dev for the development version of PETSc streams: precompute NPMAX - and use this value. And now user can use it optionally as: mpich: default to using ver 3.2 on OSX/ElCapitan as 3.1.3 does not build on Barry's laptop Revert "mpich: update version to mpich-3.2" superlu/dist: disable build on freebsd - as this test is with bsdmake [and superlu requires gnumake] superlu/dist: enable in complex nightly builds fortran: fix typo with MATSEQAIJRESTOREARRAYF90 Documentation for Dense and SeqAIJ Get/RestoreArrayf90 had incorrect dimensions Docs for MatGetRowIJF90() cut and paste and never fixed Bib: NPC refs superlu: no longer requires xerbla slamch dlamch ? $ nm -Ao libsuperlu_dist_4.3.a |egrep \(xerbla\|slamch\|dlamch\) superlu_dist: update to version 4.3 Bib: More of our refs added whitespaces to align numbers in verbatim environment Added verbatim environment to special tokens ...
Fix remaining valgrind complains: Calloc data to make sure it is initialized to zero
Merge branch 'pr388/mapdes/fix-petscsf-create-section-sf/master'
Merge branch 'master' into tisaac/dmforest* master: (177 commits) DT ex3: The test should check the number of digits requested DT ex3: One test fails in single precision Plex: Do not automati
Merge branch 'master' into tisaac/dmforest* master: (177 commits) DT ex3: The test should check the number of digits requested DT ex3: One test fails in single precision Plex: Do not automatically create a default section if no fields are defined Plex: DMPlexCopyCoordinates() needs to create the coordinate section if it is missing superlu: no longer requires xerbla slamch dlamch ? superlu: update to v5.1 skip MatSolve() if matrix factorization fails replacing SETERRQ with error output info and continue execution SNES ex12: Added parallel tests - These test parallel, unstructured interpolation/restriction using FAS Plex+SNES: Now in DMPlexComputeJacobian_Internal(), we also form the Jacobian preconditioner matrix FE: Now we can form either the Jacobian or the Jacobian preconditioned matrix using a flag DS: Now have point wise function slots for the Jacobian preconditioned matrix - This will allow us to form the "right" operator for preconditioning, like M in the (2,2) block for Stokes Plex: When swapping Plexes, also swap point SF Plex: Fix parallel error in Injector Mat+Plex: Added MatPreallocator which can be called with MatSetValues() and preallocates another Mat - This could have been another "mode" of Mat, but I thought this was cleaner - Use MatPreallocator to preallocator Plex interpolators DT ex3: Change tolerance and eliminate tests which need MPFR from regular runs superlu: fix complex build superlu: update to using 5.0 tarball with the fix for -Werror=format-security superlu: update to use tarball with -Werror=format-security fix DM: Clone the coordinate DM instead of reusing in Clone() ...Conflicts: config/builder.py src/dm/impls/plex/plexcreate.c
PetscSF: compile even if MPI_Reduce_local() is not available
Fix bad function rename and missed rename related to DMProject routines
Merge remote-tracking branch 'origin/knepley/feature-plex-add-time' into tisaac/dmforestBoth branches changeds DM projection behavior, so I wanted to sort itout before merging* origin/knepley/fe
Merge remote-tracking branch 'origin/knepley/feature-plex-add-time' into tisaac/dmforestBoth branches changeds DM projection behavior, so I wanted to sort itout before merging* origin/knepley/feature-plex-add-time: (59 commits) SNES ex69: Updated to new function API, fixed test output SNES ex75: Fixed for new function API add fortran stub for TSMonitorLGSetTransform() and fix TSMonitorLG for multiple fields but no names provided for fields add -binary_write_double so __float128 precision can save as double to binary files TSClone() did not clone Fortran function pointers hence some TS methods would fail if functions provided in Fortan petsc4py: update to snapshot with 'barry/petscoptions-object' changes fix bad merge that introduced 2 copies of the MPI_Comm_free() update example to use MAT_INPLACE_MATRIX, not REUSE incorrect addition of float to PetscInt update for new PetscOptions API TS ex18: Fixed function API and test output Function now takes PetscOptions as first argument, this was missed during the first round of updates bugfix for MatSetOption_Elemental() fix test example no change output that was missing a line fixes for new PetscOptions.. API that are needed based on problems with nightly build Plex ex3: Use new function API SNES ex62: Use new function API SNES ex12: Use new function API Plex: Fixed unused options in examples Plex+DMDA: Add time to signatures ...
PetscSF_Basic: implement PetscSFReduceEnd() for generic MPI_Op using MPI_Reduce_local()Loop over all of the unpack sites and call MPI_Reduce_local(): betterthan no implementation.
typo reported by Morten Nobel-Jørgensen
Merge branch 'maint'
Section: Doc fixReported-by: Morten Nobel-Jørgensen <mono@mek.dtu.dk>
AOFinalizePackage: reset AORegisterAllCalledReported-by: Kathrin Burckhardt <burckhardt@itis.ethz.ch>
PetscSFCreateSectionSF: fix docstringThe leafSection argument is an input, not an output parameter.
PetscSFCreateSectionSF: don't free input parameterThe remoteOffsets arrays is an input parameter to this function, sodon't free it. Fix all callers.
fixes for new PetscOptions.. API that are needed based on problems with nightly build
Merge branch 'jed/process-options-handler' into barry/petscoptions-object
Complete update to new PetscOptions interface
Begin renaming the PetscOptions objects for the future where PetscOptions will not be a singleton
detected location in code where MPIU_Allreduce() cannot be used
Introduce MPIU_Allreduce() to detect cases where all processes that share an MPI comm do not call the same sequence of MPI_Allreduce()This is a bug that can be difficult to track down.Suggested-by
Introduce MPIU_Allreduce() to detect cases where all processes that share an MPI comm do not call the same sequence of MPI_Allreduce()This is a bug that can be difficult to track down.Suggested-by: Eric Chamberland <Eric.Chamberland@giref.ulaval.ca>Suggested-by: Patrick Lacasse <placasse@giref.ulaval.ca>
1...<<111112113114115116117118119120>>...278