Merge remote-tracking branch 'origin/release'

MatGetColumnVector_SeqDenseCUDA: handle case of unallocated device data

HDF5 I/O: make 'horizontal' member of PetscViewer_HDF5;

set now true not only on presence of HDF5 attribute "MATLAB_class" but
also always for MATDENSE

Remove all double blank lines from source

Commit-type: petsc-style
/2h

Fix unused variable warning in CUDA

Merge branch 'barry/2021-05-02/fix-logging-no-symbolic-factorization' into 'main'

For factorization types/packages that do not perform a seperate symbolic phase...

See merge request petsc/petsc!3943

For factorization types/packages that do not perform a seperate symbolic phase do not log the event.

Also log the "numerical" factorization as an entire factorization

An alternative would be to pro

For factorization types/packages that do not perform a seperate symbolic phase do not log the event.

Also log the "numerical" factorization as an entire factorization

An alternative would be to provide MatXXXFactorSetUp_XXX that is called in lieu of the symbolic function
for such types/packages but requires more changes to the code.

Commit-type: bug fix for logging
/spend 45m
Reported-by: Xiaoye S. Li <xsli@lbl.gov>
Reported-by: Stephen Jardin <jardin@pppl.gov>
Reported-by: Jin Chen <jchen@pppl.gov>
Reported-by: Nan Ding <nanding@lbl.gov>
Reported-by: Yang Liu <liuyangzhuan@lbl.gov>

show more ...

Merge branch 'tisaac/feature-qr-factorization-cuda' into 'main'

QR factorization: cusolver dense

See merge request petsc/petsc!3690

MATSEQDENSECUDA: optimize using cuda*2d API for matrices with LDA

Create PetscCuBLASInt

Use cusolverDn QR factorization

Squashme: make sure seqdense matrices can still be solved with cuda

Fix memory leak in densecuda.cu

Use workvec to avoid cudaMalloc in MatSolve_SeqDenseCUDA_Interna

Use cusolverDn QR factorization

Squashme: make sure seqdense matrices can still be solved with cuda

Fix memory leak in densecuda.cu

Use workvec to avoid cudaMalloc in MatSolve_SeqDenseCUDA_Internal

squashme: remove timings

squashme: simplify copy-back logic in MatMatSolve_SeqDenseCUDA_Internal

Remove WaitForCUDA()

show more ...

Remove unneeded WaitForCUDA()