Add Compressed Basis GMRES (CB-GMRES)#693

Merged

thoasm merged 87 commits into

Feb 21, 2021

Commits on Feb 19, 2021

First commit for the GmresMixed class:

aliaga@uji.es
authored and
Thomas Grützmacher
committed
Inclusion of the omp executor in the repository:

aliaga@uji.es
authored and
Thomas Grützmacher
committed
Inclusion of the cuda executor in the repository:

aliaga@uji.es
authored and
Thomas Grützmacher
committed
The test files are finally included, but:

aliaga@uji.es
authored and
Thomas Grützmacher
committed
The first unoptimized version is done. Next tasks:

aliaga@uji.es
authored and
Thomas Grützmacher
committed
The default value for ValueTypeKrylovBasis has been changed to avoid some errors which

josealiaga
authored and
Thomas Grützmacher
committed
Definition of the CG2 variant of the finish_arnoldi routine for omp and cuda executors,

josealiaga
authored and
Thomas Grützmacher
committed
Add GMRES_mixed to benchmark
Thomas Grützmacher
committed
Definition of the CGS2 version of finish_arnoldi method, for omp and cuda executors.

josealiaga
authored and
Thomas Grützmacher
committed
Finally a good implementation of the multidot_kernels_num_iters_1 is done.

josealiaga
authored and
Thomas Grützmacher
committed
Add Accessor support and extend reference test
Thomas Grützmacher
committed
Made GmresMixed compile with complex types
Thomas Grützmacher
committed
The update routines have been improved. Now the computational time is close to 75s for 6221 iters.

josealiaga
authored and
Thomas Grützmacher
committed
Add specialization for integer types for Accessor
Thomas Grützmacher
committed
Make the scale work with integer types
Thomas Grützmacher
committed
Add helper to determine if we need a scale or not
Thomas Grützmacher
committed
Add a helper structure to manage the scale writing in common
Thomas Grützmacher
committed
Testing the push command

josealiaga
authored and
Thomas Grützmacher
committed
Definition of norm2 and norminf routines in CUDA. Only the first one is used because the second one needs the correct management of norms as noncomplex values.

josealiaga
authored and
Thomas Grützmacher
committed
remove_complex has been added to the norms variables, and multinorm2_inf_kernel has been created.

josealiaga
authored and
Thomas Grützmacher
committed
Fixed cuda step2 to take a view into account
Thomas Grützmacher
committed
Change const accessor to non-const in check_arnoldi_norms_new
Thomas Grützmacher
committed
The set_scale method finally works!!

josealiaga
authored and
Thomas Grützmacher
committed
Change storage layout of krylov_bases
Thomas Grützmacher
committed
Make memory access to krylov_bases coalesced again
Thomas Grützmacher
committed
Transpose grid when launching singledot kernel
Thomas Grützmacher
committed
Reversed the transpose of the grid dim
Thomas Grützmacher
committed
Add half precision support to GmresMixed
Thomas Grützmacher
committed
Hopefully improve singledot performance
Thomas Grützmacher
committed
Infinity norm only computed when scale is present
Thomas Grützmacher
committed
Add another RHS generation in the benchmark
Thomas Grützmacher
committed
Fix residual_norm calculation in GmresMixed
Thomas Grützmacher
committed
Make sure GmresMixed does not exit early
Thomas Grützmacher
committed
Add benchmark parameter for GMRES krylov_dim
Thomas Grützmacher
committed
Add forced iterations when convergence is detected
Thomas Grützmacher
committed
Add debug output to forced iterations
Thomas Grützmacher
committed
Fix reference bug in GmresMixed
Thomas Grützmacher
committed
DEBUG: Add write output for integral accessor
Thomas Grützmacher
committed
DEBUG: Move towards `at` with accessor
Thomas Grützmacher
committed
Remove Accessor3dConst
Thomas Grützmacher
committed
Adopt OpenMP support to new Accessor
Thomas Grützmacher
committed
Remove unused GMRES_mixed code from Ref & OMP
Thomas Grützmacher
committed
Adopt CUDA to the new accessor format (NOT `at`)
Thomas Grützmacher
committed
Make HIP and CUDA work with new accessor (NOT at)
Thomas Grützmacher
committed
Remove unused code from CUDA
Thomas Grützmacher
committed
CUDA implementation is now using `at`
Thomas Grützmacher
committed
Re-add ConstAccessor
Thomas Grützmacher
committed
Fix accessor by adding additional __restrict__
Thomas Grützmacher
committed
GmresMixed storage prec is now a factory parameter
Thomas Grützmacher
committed
Improve reference test and include the enum there
Thomas Grützmacher
committed
Fix the reference test to pass
Thomas Grützmacher
committed
Adopt to new parameter macros
Thomas Grützmacher
committed
Update the helper to throw when complex
Thomas Grützmacher
committed
Make GmresMixed work properly with multiple RHS
Thomas Grützmacher
committed
Fix benchmark to work with new GmresMixed layout
Thomas Grützmacher
committed
Use new reduced_row_major Accessor in GmresMixed
Thomas Grützmacher
committed
Remove unnecessary code from CUDA GmresMixed
Thomas Grützmacher
committed
Add HIP kernels
Thomas Grützmacher
committed
Fix GmresMixed core problem
Thomas Grützmacher
committed
Improve force-reset behavior
Thomas Grützmacher
committed
Rename GmresMixed to CbGmres
Thomas Grützmacher
committed
Format files

authored and
Thomas Grützmacher
committed
Add DPCPP stubs to allow compilation
Thomas Grützmacher
committed
Make cb-gmres benchmarks dependent on etype
Thomas Grützmacher
committed
Fix implementation and reference test for CB-GMRES
Thomas Grützmacher
committed
Update tolerance for one reference CB-GMRES test
Thomas Grützmacher
committed
Update atomic_max
Thomas Grützmacher
committed
Remove unnecessary kernels and properly name them

committed
Review update

committed
Add Helper INSTANTIATE macro for CB-GMRES

Thomas Grützmacher
and
Terry Cojean
committed
Remove CB-GMRES and GMRES example
Thomas Grützmacher
committed
Review update

committed
Remove unnecessary includes of iostream and time.h
Thomas Grützmacher
committed
Remove circular dependency of compute_norm2 in (CB)-GMRES
Thomas Grützmacher
committed
Update solver generation in benchmark
Thomas Grützmacher
committed
Update eta and arnoldi_norms in CB-GMRES
Thomas Grützmacher
committed
Remove CUDA 9.0 exception for constexpr parameter
Thomas Grützmacher
committed
Review Update

Thomas Grützmacher
and
pratikvn
committed
Sonarcloud update
Thomas Grützmacher
committed
Review update; Improve run_all_benchmarks.sh
Thomas Grützmacher
committed
Put storage_precision enum into cb_gmres namespace
Thomas Grützmacher
committed
Add CB-GMRES example
Thomas Grützmacher
committed
Remove unnecessary included files for CB-GMRES
Thomas Grützmacher
committed
Review update

Thomas Grützmacher
and
pratikvn
committed
Review update

Thomas Grützmacher
and
pratikvn
committed
Update contributors.txt

josealiaga
authored and
Thomas Grützmacher
committed
Update contributors.txt

josealiaga
authored and
Thomas Grützmacher
committed