1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 17:25:37 +01:00
Commit Graph

3622 Commits

Author SHA1 Message Date
paboyle
1dc86efd26 Finalize protection 2018-03-05 12:22:18 +00:00
30391cb2eb
Merge pull request #155 from fionnoh/develop
Some changes needed for deflation interface
2018-03-03 13:43:59 +00:00
Fionn O hOgain
2e88408f5c Some changes needed for deflation interface 2018-03-02 22:27:41 +00:00
paboyle
4790e99817 Extra communicator free that I had missed.
Hard to audit them all as this is complex
2018-02-20 15:12:31 +00:00
paboyle
2dd63aa7a4 Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2018-02-20 14:29:26 +00:00
paboyle
559a501140 Deflation interface for solvers 2018-02-20 14:29:08 +00:00
paboyle
945684c470 updates for deflation in the RB solver 2018-02-20 14:28:38 +00:00
Christopher Kelly
e30a80a234 Relaxed constraints on MPI thread mode when not using multiple comms threads 2018-02-15 17:13:36 +00:00
paboyle
c96483e3bd Whitespace only change 2018-02-13 11:39:07 +00:00
paboyle
ae31a6a760 Move deflate to right class 2018-02-13 02:11:37 +00:00
paboyle
dd8f2a64fe INterface to suit hadrons on Lanczos 2018-02-13 02:08:49 +00:00
paboyle
7b8b2731e7 Conj error for complex coeffs 2018-02-12 16:06:31 +00:00
paboyle
237a8ec918 Communicator leak fixed (I think) 2018-02-12 13:27:20 +00:00
Daniel Richtmann
6ab60c5b70
Merge remote-tracking branch 'upstream/develop' into feature/wilsonmg 2018-02-08 23:59:07 +01:00
Daniel Richtmann
8c692b7ffd
WilsonMG: Comment assertion on hermiticity of coarse operator for now
TODO: Think of a way to not break dwf_hdcr by doing that. It's only an assertion
but it still interferes with it.
2018-02-08 23:55:05 +01:00
Daniel Richtmann
2976132bdd
Add first version of multigrid for wilson clover analogous to wilson one
Just like the wilson one, this algorithm

• is currently only a 2-level method since I don't have correct implementations
  for Mdir and Mdiag in CoarsenedMatrix yet (needed for further coarsening)
• needs levelization and refactoring into a proper algorithm
2018-02-08 23:52:10 +01:00
Daniel Richtmann
48177f2f2d
Add tests for all MR|GMRES solvers with wilson clover action 2018-02-08 23:52:09 +01:00
Daniel Richtmann
c4ce70a821
WilsonMG: Major cleanup 2018-02-08 23:52:08 +01:00
Daniel Richtmann
a3e009ba54
Add tests for CAGMRES solvers with staggered action 2018-02-08 17:46:28 +01:00
Daniel Richtmann
eb7cf239d9
Print warning messages in CAGMRES solvers
Currently, the implementation of these algorithms doesn't differ from their non
communication-avoiding versions.
2018-02-08 17:43:47 +01:00
Daniel Richtmann
13ae371ef8
Make solver parameters match in all MR|GMRES solver tests 2018-02-08 17:33:10 +01:00
Daniel Richtmann
9f79a87102
Fix bugs in Flexible GMRES solvers
Somehow I got the left and right-preconditioned versions of GMRES mixed up. As
of now this is right-preconditioned version, which is what we want.
2018-02-08 16:00:31 +01:00
Daniel Richtmann
4ded1ceeb0
Make GMRES solvers perform no more than MaxIterations steps
I noticed that it was possible to overrun this number.
2018-02-08 15:29:44 +01:00
Daniel Richtmann
8bc12e0ce1
Remove superfluous comments in MR solver 2018-02-07 18:09:09 +01:00
Daniel Richtmann
cc2f00f827
Remove test for MR solver with dwf action as it doesn't converge 2018-02-07 18:09:08 +01:00
Daniel Richtmann
cd61e2e6d6
Increase max iterations in test of MR solver with staggered action 2018-02-07 18:09:07 +01:00
Daniel Richtmann
323ed1a588
Add an overrelaxation parameter to the MR solver 2018-02-07 18:09:06 +01:00
Daniel Richtmann
68c66d2e4b
Remove empty line in output of *Residual* solvers 2018-02-07 18:08:56 +01:00
Daniel Richtmann
1671adfd49
WilsonMG: Add some tests for linear operators 2018-02-07 17:15:22 +01:00
Peter Boyle
896f3a8002 Fix to MPI for Hokusai system 2018-02-01 18:51:51 +00:00
Daniel Richtmann
871649238c
WilsonMG: Stricter naming for linear operators 2018-02-01 14:43:08 +01:00
Daniel Richtmann
7c86d2085b
WilsonMG: Some minor cleanup 2018-02-01 12:24:16 +01:00
Daniel Richtmann
9292be0b69
WilsonMG: Add check for Mdiag + Σ Mdir == M
Need to test my implementations of CoarsenedMatrix::Mdiag &
CoarsenedMatrix::Mdir.
2018-01-31 14:03:30 +01:00
Guido Cossu
f0fcdf75b5
Update README.md 2018-01-30 12:44:20 +01:00
Guido Cossu
53bffb83d4
Updating README with new SKL target 2018-01-30 12:42:36 +01:00
Daniel Richtmann
10141f90c9
WilsonMG: Rename test file 2018-01-30 10:25:09 +01:00
Guido Cossu
cd44e851f1 Fixing compilation error in FundtoHirep 2018-01-30 06:04:30 +01:00
Daniel Richtmann
a414430817
Merge remote-tracking branch 'upstream/develop' into feature/ddalphaamg 2018-01-29 18:32:31 +01:00
Daniel Richtmann
f20728baa9
WilsonMG: Some further steps towards a three level method
Currently this is very "manual" as we are still testing stuff. Will refactor
and make it an algorithm once everything works.

What currently does work:

  - All tests in MultiGridPreconditioner::runChecks for the first coarse grid
  - The tests for the intergrid operators going from the first to the second
    coarse grid
    - (1 - P R) v   == 0
    - (1 - R P) v_c == 0
  - A full solve with VPGCR and a two-level MG preconditioner

What hinders the rest of the tests from passing with a three-level method is the
absence of implementations of CoarsenedMatrix::Mdir and CoarsenedMatrix::Mdiag.
2018-01-29 18:29:49 +01:00
Daniel Richtmann
d2e68c4355
WilsonMG: Perform some minor cleanup 2018-01-29 18:07:10 +01:00
Daniel Richtmann
1cb745c8dc
Add function to return full type as std::string
Also works for nested templates. I find it useful for debugging.
Possible usage:

std::cout << "getTypename<AType>() = " << getTypename<Atype>() << std::endl;
std::cout << "getTypename<decltype(AnInstance)>() = " << getTypename<decltype(AnInstance)>() << std::endl;
2018-01-29 17:39:19 +01:00
Daniel Richtmann
faf4278019
Use 2 passes of GS in coarse operator construction 2018-01-29 17:21:42 +01:00
Daniel Richtmann
194e4b94bb
Make MG checking function work level-wise 2018-01-29 17:18:20 +01:00
Daniel Richtmann
bfc1411c1f
Use more iterations in subspace creation 2018-01-29 17:11:29 +01:00
Daniel Richtmann
161637e573
Turn on orthogonality checking temporarily 2018-01-29 17:10:05 +01:00
Guido Cossu
fb24e3a7d2 Adding utilities for perf profiling 2018-01-29 11:11:45 +01:00
Guido Cossu
655a69259a Added support for GCC compilation for Skylake AVX512 2018-01-28 17:02:46 +01:00
Guido Cossu
507c4e9efc Correcting an missing semicolumn in avx512 2018-01-27 10:59:55 +01:00
Guido Cossu
f8a5194c70 Merge branch 'develop' of https://github.com/paboyle/Grid into develop 2018-01-25 13:46:37 +01:00
Guido Cossu
cff3bae155 Adding support for general Nc in the benchmark outputs 2018-01-25 13:46:31 +01:00