add old aggressive coarsening algorithm as option, switch default MIS ordering algorithm back to old one
Replace LINPACKcgtql1 with LAPACKstev for computing sym tri eigenvaluesSpecifically, removes translated Fortran and Algol routines which had non-terminating loops if assumptions were violated with
Replace LINPACKcgtql1 with LAPACKstev for computing sym tri eigenvaluesSpecifically, removes translated Fortran and Algol routines which had non-terminating loops if assumptions were violated with a standard LAPACK eigenvalue solver routine. Add new alternatives due to minor differences in eigenvalue estimates.
show more ...
SNESNEWTONTR: stol should be zero, since we use the trust region parameter
SNESNEWTONTR: add support for NPC and sufficient decreaseadd tests
SNES: support Application context in python codeWe keep a global registry for contexts set via Python and cast to object only if SNESGetApplicationContext returns a void* in the setOtherwise, we r
SNES: support Application context in python codeWe keep a global registry for contexts set via Python and cast to object only if SNESGetApplicationContext returns a void* in the setOtherwise, we return the address
Add implementation of 4th-kind Chebyshev polynomial smoother
SNESNEWTONTR: add dogleg and Cauchy fallbacks from NTRDC
Merge branch 'danofinn/swarm-example-landau-damping' into 'main'Swarm: Add swarm ex9 Landau damping + tests (ex6, ex8, snes/ex15)See merge request petsc/petsc!6026
SNES ex15: Mixed Poisson example
Minor additions for viewing image draw viewers
PCVPBJACOBI: record min/max block sizes and number of blocks for viewing
print N in ex13
CI: add --download-amgx to linux-cuda-pkgsAnd fix related issues
Add PC interface to AmgXhttps://developer.nvidia.com/amgx
Merge branch 'jczhang/feature-vpbjacobi-cuda' into 'main'Implement PCApply_VPBJacobi() for AIJ CUDA and Kokkos matricesSee merge request petsc/petsc!5466
PC: implement PCApply_VPBJacobi() for AIJ CUDA and Kokkos matrices
Fix test output
cleanup ex13 tests, switch GMRES to cg - in an attempt to fix this errorHowever there appears to be a non-deterministic bug in parallel MKL, in this older version*** Error in `../ex13': free(): i
cleanup ex13 tests, switch GMRES to cg - in an attempt to fix this errorHowever there appears to be a non-deterministic bug in parallel MKL, in this older version*** Error in `../ex13': free(): invalid next size (normal): 0x0000000000e210d0 ***======= Backtrace: =========/lib64/libc.so.6(+0x81329)[0x2b3c7c755329]/homes/intel/18u2/compilers_and_libraries_2018.2.199/linux/mkl/lib/intel64_lin/libmkl_core.so(mkl_serv_free+0x2c4)[0x2b3c725d6504]/homes/intel/18u2/compilers_and_libraries_2018.2.199/linux/mkl/lib/intel64_lin/libmkl_sequential.so(mkl_sparse_d_inspector_gen_mv_i4+0x4af)[0x2b3c767c166f]/homes/intel/18u2/compilers_and_libraries_2018.2.199/linux/mkl/lib/intel64_lin/libmkl_sequential.so(mkl_sparse_d_optimize_csr_mv_i4+0xa5)[0x2b3c767c5515]/homes/intel/18u2/compilers_and_libraries_2018.2.199/linux/mkl/lib/intel64_lin/libmkl_avx512_mic.so(mkl_sparse_optimize_i4_avx512_mic+0x220)[0x2b3ca3476bc0]/home/petsc/builds/6i15ewNf/2/petsc/petsc/arch-ci-linux-knl/lib/libpetsc.so.3.017(+0x65e06a)[0x2b3c6f4f006a]/home/petsc/builds/6i15ewNf/2/petsc/petsc/arch-ci-linux-knl/lib/libpetsc.so.3.017(MatMultTranspose_SeqAIJMKL_SpMV2+0x110)[0x2b3c6f4f14d0]/home/petsc/builds/6i15ewNf/2/petsc/petsc/arch-ci-linux-knl/lib/libpetsc.so.3.017(MatMultTranspose_MPIAIJ+0x49)[0x2b3c6f3529e9]/home/petsc/builds/6i15ewNf/2/petsc/petsc/arch-ci-linux-knl/lib/libpetsc.so.3.017(MatMultTranspose+0x15e)[0x2b3c6f8c828e]/home/petsc/builds/6i15ewNf/2/petsc/petsc/arch-ci-linux-knl/lib/libpetsc.so.3.017(MatRestrict+0x7f)[0x2b3c6f8bcb4f]
added distance k MIS MatCoarsen type
Change matrix factor database options that occur in KSP/PC to always use KSP/PC options prefixAdd MatSetOptionsPrefixFactor() and MatAppendOptionsPrefixFactor() to allow controlling the options pre
Change matrix factor database options that occur in KSP/PC to always use KSP/PC options prefixAdd MatSetOptionsPrefixFactor() and MatAppendOptionsPrefixFactor() to allow controlling the options prefixused by factors created from this matrixChange MatSetOptionsPrefix() to no longer affect the options prefix used by factors created from this matrixAs part of the above work the handling of the factor matrix options is now done in code in the factor matrixnot in the code that originates the factor matrixUpdate examples output to the new approach where the factored matrix prefix is set from the KSP/PCMuch of this work was performed by Hong ZhangCommit-type: usability/spend 5h
Remove extra spaces and unwanted plurals
PC_Factor: set prefix from PC in pc->pmat
Enable PETSC_ATTRIBUTE_FORMAT()
fix format and update ex13
PCGAMG: update tests* formatting of Chebyshev information* use of GAMG SA eigenvalue estimates
123