CUDA: Get PETSc to compile with CUDA 11.CUSPARSE has changed significantly in CUDA 11.This has lead to a lot of broken code.This commit disables the AIJCUSPARSE class so that PETSc at least compi
CUDA: Get PETSc to compile with CUDA 11.CUSPARSE has changed significantly in CUDA 11.This has lead to a lot of broken code.This commit disables the AIJCUSPARSE class so that PETSc at least compiles with other GPU features enabled.
show more ...
Turn on checkbadSource test to generate an error when found; fix all source code that causes errorsCommit-type: portability-fix, testing-fix, style-fix, feature, maintainability/spend 1.5h
Add support of CUDA/HIP initialization at the same time
Merge remote-tracking branch 'origin/maint'
cudavecimpl.h move to include/petsc/private as its used by private/sfimpl.hReported-by: Nidish <nb25@rice.edu>
Do lazy petsc cuda initialization and validationSo that we can avoid overburdening the GPU in CI.See more at https://gitlab.com/petsc/petsc/-/issues/360#note_373723821
Most external factorization packages do not use the ordering provided by PETSc, therefor only provide it when needed.Adds MatOrderingType external to indicate not to generate an ordering and use wh
Most external factorization packages do not use the ordering provided by PETSc, therefor only provide it when needed.Adds MatOrderingType external to indicate not to generate an ordering and use what the package needsNow -pc_view will not print the wrong PETSc ordering when the ordering is done externallySaves some memory and compute time.Commit-type: optimizationTime: 1.2 hoursReported-by: Junchao Zhang <junchao.zhang@gmail.com>Thanks-to: Stefano Zampini <stefano.zampini@gmail.com>
MATSEQAIJCUSPARSE: test multiple transpose generation flags
MATSEQAIJCUSPARSE: log transpose generation
MATSEQAIJCUSPARSE: remove wrong CPUtoGPU memory logging
MatMultHermitianTranspose: use MatTranspose directly for real builds
MATSEQAIJCUSPARSE: implement hermitian transpose operationSimplify the class, using a single routine for all the MatMult variantsMake transpose generation optional (not default)
MATSEQAIJCUSPARSE: allow MatConvert with reuse != MAT_INPLACE_MATRIX
MATSEQAIJCUSPARSE: add logging support for MatProduct with dense matrix
Fix compile warning
MATSEQAIJCUSPARSE: support for MatMat operationsfix issue with ILU reusage (leaks and not setting the offdiagonal flag properly)add missing nullify of function composition
MatConvert_SeqAIJ_SeqAIJCUSPARSE: fix bug with reuse != MAT_INPLACE_MATRIX
Single petscdir.mk
Fix cuSPARSE index type with 64-bit PetscIntFix issue #612
MAT: do not force compressed row on GPU, instead use whatever on CPU
Use cusparse version number instead of cudart'sThanks-to: Stanimire Tomov <tomov@icl.utk.edu>
build: add more support for not needing to set PETSC_DIR yet build with PETSc, also support for GNU make to directly use "make test" instead of "make -f gmakefile.test"Still needs -include ../petsc
build: add more support for not needing to set PETSC_DIR yet build with PETSc, also support for GNU make to directly use "make test" instead of "make -f gmakefile.test"Still needs -include ../petscdir.mk in many of the makefilesStill needs the link for ../petscdir.mk in many directoriesGNUMakefile: call legacy makefile recursively instead of including itGNUMakefile: promote default "all" target; "libs" builds libpetsc*GNUMakefile: If gnumake < 4.2 prefer makefile wrapper interface over direct gmakefile interface to avoid the following warningsgmakefile.test:92: arch-ci-linux-cxx-cmplx-pkgs-64idx/tests/testfiles: No such file or directorygmakefile:67: arch-ci-linux-cxx-cmplx-pkgs-64idx/lib/petsc/conf/files: No such file or directoryCommit-type: testing-fix, feature, usabilityThanks-to: Jed Brown jed@jedbrown.org
Fix errors in MatBindToCPU_SeqAIJCUSPARSE().Reported-by: Junchao Zhang <jczhang@mcs.anl.gov>
1...<<11121314151617