ee808e4 | Balint Joo | 23 October 2018, 16:23:11 UTC | regressions added for seoprec 1flavor, 2 flavor and hasenbusch. Original testing restores. All tests (except eigcg which was not built) pass. All new monomials conserve energy. Now we need to do a bit of validation, but modulo that this is ready to merge back into devel. | 23 October 2018, 16:23:11 UTC |
8b6b1a1 | Balint Joo | 22 October 2018, 21:19:35 UTC | Added regression for seoprec logdet monomial and fixed a bug when checkoing dt stepsize scaling | 22 October 2018, 21:19:35 UTC |
b9e15ec | Balint Joo | 22 October 2018, 19:57:37 UTC | Replaced these candidate test outputs as the monomial template uses a slightly different tag name. EvenOddPrec ..blahblah .. Monomial -> EOPrec... blah blah .. MonomialT This caused structural validation issues for XMLDIFF for these 4 cases. Verified manually that the DeltaH's are unchanged | 22 October 2018, 19:57:37 UTC |
8de6b99 | Balint Joo | 22 October 2018, 19:01:18 UTC | Added one flavor Seoprec monomial as well | 22 October 2018, 19:01:18 UTC |
7ddf55e | Balint Joo | 22 October 2018, 17:35:28 UTC | Refactored the EvenOdd Two Flavor and Two Flavor ratios, so they take the FermAct and LinOp as template params. Preserved previous type aliases using 'using'. This made it easy to add 'Symm' versions by simply changing the template params. Added Sym EO Two Flavor ConstDet monomial and Sym EO ConstDet Ratio Conv Conv monomial -- duplicated preivous const-det and changed types, derivations. This could maybe be tidyed up using templates again, the onlt thing is the registrations -- maybe come back to these later. Still to do the one flavor case | 22 October 2018, 17:35:28 UTC |
2fe5d45 | Balint Joo | 19 October 2018, 23:58:38 UTC | Fixed bug that stopped computation of TrLog on odd sites. This was sneaky: basically every call to ldaginv zeroed tr_log_diag so when we computed on the odd subset, the previous even subset was zeroed. Fixed by zeroing only the checkerboard we compute on. t_symm_prec will catch this. Added tests for logDetEvenEvenLinOp, logDetOddOddLinOp() Added test for TrLog forces Both the above use translation invariance. | 19 October 2018, 23:58:38 UTC |
288d2b3 | Balint Joo | 19 October 2018, 17:29:18 UTC | SEOPREC Matrix Derivative now passes test SEOPREC Qprop, MdagM solver and Multi-Syssolver now passes tests TODO: S_logdet forces, S_lodget energies, SEOPREC monomials - two flavor - logdet - hasenbusch - rational TODO: Add twisted Mass if specified. | 19 October 2018, 17:29:18 UTC |
249f30a | Balint Joo | 18 October 2018, 21:06:09 UTC | Working on SEOPREC Structure. Still to add multi-shift test to t_symm_prec. Still to add Forces test to t_symm_prec | 18 October 2018, 21:06:09 UTC |
66fd0c5 | Balint Joo | 09 October 2018, 16:59:41 UTC | Changed these to actually use the asymmetry parameter. | 09 October 2018, 16:59:41 UTC |
90f4726 | Balint Joo | 08 October 2018, 20:22:11 UTC | Fixed merge conflict | 08 October 2018, 20:22:11 UTC |
a60255e | Balint Joo | 05 October 2018, 20:56:24 UTC | Merge branch 'devel' of github.com:jeffersonlab/chroma into devel | 05 October 2018, 20:56:24 UTC |
4d38505 | Balint Joo | 05 October 2018, 20:56:06 UTC | Moved to MG_proto_eo not using a wrapper. Call it with FermAct CLOVER (not Unpreconditioned) | 05 October 2018, 20:56:06 UTC |
8ebb33b | Balint Joo | 05 October 2018, 03:23:39 UTC | Subspace reinit fixes - Thanks Kate! | 05 October 2018, 03:23:39 UTC |
939dc2e | Balint Joo | 01 October 2018, 17:56:21 UTC | Added <reunit> flag to <Cfg> for SZINQIO. If working in double prec and reading a single prec Cfg this should be set to true if one plans on using gauge compression. | 01 October 2018, 17:56:21 UTC |
73f780f | Balint Joo | 18 September 2018, 20:17:01 UTC | Moved the creation of the outer solver object out of the operator() into the constructor. This avoids adding timers twice (tho we should fix that in MG_Proto) which fixes the problems of aggregate timings for now. | 18 September 2018, 20:17:01 UTC |
29c49d2 | Balint Joo | 17 September 2018, 23:13:08 UTC | Added timer reporting for MGProto | 17 September 2018, 23:13:08 UTC |
a910822 | Balint Joo | 05 September 2018, 00:47:38 UTC | reverted this | 05 September 2018, 00:47:38 UTC |
cac965c | Balint Joo | 04 September 2018, 22:01:41 UTC | Some tweak to the helpers.h file. The Commit Diff will tell me what it is :) | 04 September 2018, 22:01:41 UTC |
8348557 | Balint Joo | 25 August 2018, 00:53:47 UTC | Added bindings for EO MG Proto and tested | 25 August 2018, 00:53:47 UTC |
16703f3 | Balint Joo | 24 August 2018, 21:57:45 UTC | Helpers updated to allow for EO subspace creation, deletion | 24 August 2018, 21:57:45 UTC |
539b11b | Balint Joo | 02 August 2018, 02:08:32 UTC | Exposed a bunch of parameters as arrays (coarse_solver type, setup solver type, smoother related parameters) Added optional Precision parameter to the Chronology Exposed subspace solver | 02 August 2018, 02:08:32 UTC |
89ba8e3 | Balint Joo | 23 July 2018, 13:08:45 UTC | Fixed a minor bug preventing mult-shift from working properly | 23 July 2018, 13:08:45 UTC |
7178390 | Frank Winter | 22 July 2018, 15:53:26 UTC | Removes debugging output | 22 July 2018, 15:53:26 UTC |
8321024 | Frank Winter | 22 July 2018, 15:50:38 UTC | Update the QUDA/QDP-JIT device interface These changes modify the QUDA/QDP-JIT device interface to reflect the recent changes to the API of the qdp-jit memory cache. | 22 July 2018, 15:50:38 UTC |
993a416 | Frank Winter | 20 July 2018, 15:43:58 UTC | Merge That was only the deleted chroma_config_internal.in. | 20 July 2018, 15:43:58 UTC |
249cc70 | Frank Winter | 20 July 2018, 15:39:17 UTC | Updating the JIT kernels for the new API. | 20 July 2018, 15:39:17 UTC |
bc8b0e9 | Balint Joo | 16 July 2018, 20:41:15 UTC | Kate's CA_GCR Patches | 16 July 2018, 20:41:15 UTC |
8ff4448 | Balint Joo | 16 July 2018, 04:18:13 UTC | This code compiled and ran with 11 regression failures. (None in t_leapfrog) | 16 July 2018, 04:18:13 UTC |
11bf662 | Balint Joo | 14 July 2018, 04:05:59 UTC | removed some more extra ; | 14 July 2018, 04:05:59 UTC |
83e0b8a | Balint Joo | 14 July 2018, 02:43:11 UTC | My script messed these up so I fixed it by hand. | 14 July 2018, 02:43:11 UTC |
e3a8ad0 | Balint Joo | 13 July 2018, 19:46:14 UTC | Removed extra l using script | 13 July 2018, 19:46:14 UTC |
0523015 | Balint Joo | 13 July 2018, 13:07:51 UTC | Merge branch 'devel' into silence_warnings | 13 July 2018, 13:07:51 UTC |
032070b | Balint Joo | 13 July 2018, 13:06:26 UTC | Silencing warnings one at a time | 13 July 2018, 13:06:26 UTC |
4a4884c | Frank Winter | 12 July 2018, 19:02:01 UTC | Bugfix for qdp-jit constraction kernel Now builds in double precision fine | 12 July 2018, 19:02:01 UTC |
f092fcb | Frank Winter | 11 July 2018, 19:34:22 UTC | Removes debugging output | 11 July 2018, 19:34:22 UTC |
be923ca | Frank Winter | 11 July 2018, 19:32:30 UTC | PROP_AND_MATELEM_DISTILLATION (edit) Adds fused innerProduct of outer subtypes with qdp types over an array of those. Fusion happens on two levels: 1) the contraction of the source vectors with the solution vector and 2) the first part of the device summation. | 11 July 2018, 19:32:30 UTC |
49f4fe3 | Balint Joo | 11 July 2018, 18:35:15 UTC | Added -Wl,-rpath=QUDA_LIBPATH onto LDFLAGS for QUDA | 11 July 2018, 18:35:15 UTC |
28692ee | Balint Joo | 09 July 2018, 16:27:14 UTC | Replaced some t_leapfrog and hmc XMLs to account for the fact that force monitors now output F_dt_per_direction and F_dt_max All regressions now pass except eigcg, which I haven't built with Intel Par Studio 19 beta | 09 July 2018, 16:27:14 UTC |
b015409 | Frank Winter | 02 July 2018, 21:04:27 UTC | PROP_AND_MATELEM_DISTILLATION (edit) Adds parameter zero_colorvecs which puts measurement into a timing mode - no solver calls, use zero solution vectors | 02 July 2018, 21:04:27 UTC |
ab9a0d0 | Frank Winter | 28 June 2018, 18:12:29 UTC | .. but do not release propmat distillation2 for qdp-jit | 28 June 2018, 18:12:29 UTC |
bd81984 | Frank Winter | 20 June 2018, 17:52:47 UTC | Release distillation unsmeared hadron node for qdp-jit/nvptx compilation | 20 June 2018, 17:52:47 UTC |
b5ef777 | Frank Winter | 14 June 2018, 22:12:10 UTC | Release distillation prop_and_matelem for qdp-jit/nvptx compilation | 14 June 2018, 22:12:10 UTC |
f2d9113 | Frank Winter | 08 June 2018, 19:18:59 UTC | Slightly alters the Inverter Vector container When storing the Ritz eigenpairs, eigenvalues were stored passing lattice subset along. Now the Vector container allows to store things without a subset where it makes sense, like for eigenvalues. It is more natural and as a positive side effect removes the need for funny things like outer subscalars. | 08 June 2018, 19:18:59 UTC |
f48113f | Balint Joo | 25 May 2018, 18:11:50 UTC | Added the option of not checking the QUDA soluion to save on Host Amdahl's law. This is disabled by default. | 25 May 2018, 18:11:50 UTC |
cc18d25 | Balint Joo | 25 May 2018, 18:08:19 UTC | Merge branch 'solver_exception' into devel | 25 May 2018, 18:08:19 UTC |
12c1c6f | Balint Joo | 25 May 2018, 18:07:06 UTC | Merge branch 'devel' of github.com:jeffersonlab/chroma into devel | 25 May 2018, 18:07:06 UTC |
4fbb2ef | Balint Joo | 25 May 2018, 18:06:37 UTC | Added code to catch specifically QUDA MG Exceptions raised by nonconvergence and re-throw them. THey will be dealt with higher up. | 25 May 2018, 18:06:37 UTC |
62265f8 | Balint Joo | 25 May 2018, 17:53:47 UTC | Added erase subspace for QPHIX based MG_PROTO Multigrid Fixed interface issue that got incorrect results when MG_Proto added compression and antiperiodic BCs were note dealt with properly | 25 May 2018, 17:53:47 UTC |
8be35df | Balint Joo | 17 May 2018, 14:02:41 UTC | Set default pipeline depth = 0 for non MG solvers | 17 May 2018, 14:02:41 UTC |
88e5574 | Balint Joo | 04 May 2018, 18:45:03 UTC | Made transfer urls protocol agnostic (again?) | 04 May 2018, 18:45:03 UTC |
98d8421 | Balint Joo | 04 May 2018, 18:40:55 UTC | Updated WilsonMG | 04 May 2018, 18:40:55 UTC |
94dd5bd | Balint Joo | 04 May 2018, 18:12:28 UTC | Updated SSE Wilson Dslash | 04 May 2018, 18:12:28 UTC |
9aeca60 | Balint Joo | 04 May 2018, 18:09:49 UTC | Updated qdp-lapack | 04 May 2018, 18:09:49 UTC |
a110792 | Balint Joo | 04 May 2018, 18:05:56 UTC | Updated cpp_wilson_dslash | 04 May 2018, 18:05:56 UTC |
a1022bb | Balint Joo | 04 May 2018, 17:43:59 UTC | Updated License to 2018 Jefferson Lab License | 04 May 2018, 17:43:59 UTC |
4ec5d71 | Balint Joo | 04 May 2018, 17:43:59 UTC | Removing submodule CG-DWF from seoprec | 04 May 2018, 17:43:59 UTC |
b35ec03 | Balint Joo | 04 May 2018, 17:40:45 UTC | Updated License to 2018 Jefferson Lab License | 04 May 2018, 17:40:45 UTC |
b06ee0a | Balint Joo | 04 May 2018, 17:40:45 UTC | Removing submodule CG-DWF from devel | 04 May 2018, 17:40:45 UTC |
ec0ef45 | Balint Joo | 27 April 2018, 17:55:51 UTC | Exposed SmootherHaloPrecision | 27 April 2018, 17:55:51 UTC |
64ba1bd | Robert Edwards | 26 April 2018, 13:41:29 UTC | Merge branch 'devel' into seoprec | 26 April 2018, 13:41:29 UTC |
9c8f2a9 | Balint Joo | 28 March 2018, 12:47:32 UTC | UPdated QDP-Lapack for COPYING | 28 March 2018, 12:47:32 UTC |
ff4de46 | Balint Joo | 27 March 2018, 20:43:21 UTC | UPdated copying in WilsonMG | 27 March 2018, 20:43:21 UTC |
6c92c81 | Balint Joo | 27 March 2018, 20:40:50 UTC | Updated License In SSE Wilson Dslash | 27 March 2018, 20:40:50 UTC |
72a47bd | Robert Edwards | 27 February 2018, 23:16:52 UTC | Merge pull request #31 from JeffersonLab/solver_exception Solver exception | 27 February 2018, 23:16:52 UTC |
debf260 | Balint Joo | 25 February 2018, 17:17:21 UTC | Added exception Handling feature to HMC and QUDA multigrid | 25 February 2018, 17:17:21 UTC |
def1baa | Balint Joo | 23 February 2018, 17:56:13 UTC | Added Exception Handling logic to abs_hmc.h | 23 February 2018, 17:56:13 UTC |
bd68203 | Balint Joo | 23 February 2018, 16:46:13 UTC | Added MG Solver Exception class | 23 February 2018, 16:46:13 UTC |
a7e2d41 | Balint Joo | 15 February 2018, 14:29:02 UTC | A few places where I forgot to add the Pipeline parameter | 15 February 2018, 14:29:02 UTC |
8e887bc | Balint Joo | 14 February 2018, 20:35:04 UTC | Added Pipeline member to params for Clover, NEF, Wilson. Added setting of quda_inv_param.pipeline for MdagM shifted CG (Clover,Wilson), Linop Solvers (Clover, NEF, Wilson, Clover MG, Wilson MG) MdagM solvers (Clover, Clover MG, Wilson (there is no MdagM wilson MG)) | 14 February 2018, 20:35:04 UTC |
7d22d54 | Balint Joo | 08 February 2018, 13:35:32 UTC | Updated force monitoring to also print out F_max * aniso-step-size factor | 08 February 2018, 13:35:32 UTC |
930fecc | Balint Joo | 03 February 2018, 04:42:49 UTC | Added GPU Free memory checker diagnostic to QUDA MdagM solver | 03 February 2018, 04:42:49 UTC |
f78b2b0 | Balint Joo | 31 January 2018, 01:08:09 UTC | Updated Authors file | 31 January 2018, 01:08:09 UTC |
f9d58dc | Balint Joo | 31 January 2018, 01:02:07 UTC | Various fixes: - Patch from Yi-Bo Yang for Linop solver to use MGUtils functions and allow reuse of MG space in LinOp solver - Patch to MdagM solver to use zero guess on re-solve - Patch to hmc.cc to run user measurements even on always accepted (WarmUp) trajecotries. | 31 January 2018, 01:02:07 UTC |
fc8ee8f | Balint Joo | 26 January 2018, 21:18:24 UTC | Updated Chroma License to Jefferson Lab, 3 part BSD style license | 26 January 2018, 21:18:24 UTC |
4e04f16 | Balint Joo | 25 January 2018, 19:53:34 UTC | Reorganized refresh strategy for failing MG solves in MdagM clover Multigrid solve Occasionally during HMC multigrid solves can fail (e.g. when going through some zeor mode). In this case, instead of refreshing we blow away the old subspace completely now and regenerate it. | 25 January 2018, 19:53:34 UTC |
b572bd6 | Robert Edwards | 19 January 2018, 20:24:17 UTC | Merge branch 'devel' into seoprec | 19 January 2018, 20:24:17 UTC |
3918d5e | Robert Edwards | 19 January 2018, 04:54:12 UTC | First running version of a new chain of symmetric even-odd precondtioned monomials. There is a new SEOPREC_CLOVER fermact. | 19 January 2018, 04:54:12 UTC |
33d924f | Balint Joo | 11 January 2018, 20:15:28 UTC | Merge branch 'quda-predictor' into devel | 11 January 2018, 20:15:28 UTC |
aa45f7e | Balint Joo | 11 January 2018, 20:14:31 UTC | QUDA Chrono rides! | 11 January 2018, 20:14:31 UTC |
5a52957 | Balint Joo | 11 January 2018, 16:05:56 UTC | Added in QUDA predictor wrapper. Picks 'channel ID's for QUDA's internal chronology. Needs now to fill out the required fields for QUDA to use chrono properly | 11 January 2018, 16:05:56 UTC |
472e17e | Kostas Orginos | 11 January 2018, 12:10:45 UTC | finished the wave source for SF | 11 January 2018, 12:10:45 UTC |
9211521 | Balint Joo | 10 January 2018, 21:09:58 UTC | Added basic files for QUDA Predictor | 10 January 2018, 21:09:58 UTC |
825ba03 | Balint Joo | 03 January 2018, 15:06:06 UTC | Following Kate's advice, turned off QUDA residual check in MdagM solvers (Since chroma checks it anyway) Also added stricter residuum checking in multishift solver. | 03 January 2018, 15:06:06 UTC |
4842d6a | Balint Joo | 27 December 2017, 21:48:29 UTC | Changed the gauge update in the FG term to be a regular leapQ rather than a full sub integration. -- That should perhaps be the appropriate thing to do with multilevel integrator, but will check on this first | 27 December 2017, 21:48:29 UTC |
dd3a25e | Balint Joo | 27 December 2017, 00:22:06 UTC | Added Boram Yoon to AUTHORS for his work on Force Gradient Added Kate Clark to AUTHORS for her help with debuggin memory issues on SummitDev, over and above all her hard work on on QUDA and algorithms. | 27 December 2017, 00:22:06 UTC |
2a7e9cd | Balint Joo | 26 December 2017, 23:57:50 UTC | Updated to version of inline_unsmeared_hadron node from Nov 25 version in branch master after noting this change was present in Boram Yoon's branch. This replacees version I had from Nov 23 from branch feature/unsmeared-node Hope this is the right version. Sending to Robert for confirmation. | 26 December 2017, 23:57:50 UTC |
dbb9cae | Balint Joo | 26 December 2017, 23:38:09 UTC | Merged inline_hadron_aggregate.cc resolving in favour of chroma devel | 26 December 2017, 23:38:09 UTC |
437bd14 | Balint Joo | 25 December 2017, 04:23:38 UTC | Modified to update code with new QUDA qudaUpdateMultigrid() routine. This reuses previous suspace and polishes it with max_iter set by max_iter_setup in the QudaMultigridParam. We introduce 3 new XML tags: <MaxIterSubspaceCreate> -- used to set up the subspace <MaxIterSubspaceRefresh> -- number of max iterations in a refresh (0 means just reconstruct operator) <RsdTargetSubspaceCreate> If no values are given we revert to defaults: RsdTargetSubspaceCreate = 5.0e-6 MaxIterSubspaceCreate = 500 MaxIterSubspaceRefresh = MaxIterSubspaceCreate | 25 December 2017, 04:23:38 UTC |
13820ba | Boram Yoon | 22 December 2017, 19:30:16 UTC | First implementation of force-gradient integrator. Implementation adopted the Taylor expansion trick in Yin and Mawhinney (arXiv:1111.5059). | 22 December 2017, 19:30:16 UTC |
46e77ea | Balint Joo | 20 December 2017, 17:34:36 UTC | Modified output and timers from chroma. Chroma will now also print out the internal plaquette time the measurement XML parsing time and the total measurement time | 20 December 2017, 17:34:36 UTC |
fa20413 | Balint Joo | 20 December 2017, 16:53:14 UTC | Bugfixes to avoid weird memory uninitialization errors. Thanks to Kate for tracking these with valgrind! - prec member of MULTIGRIDInnerParams was not read from XML and left initialized - loops over QUDA_MAX_DIM looped to dim=5 (because DWF). Changed to Nd - dereferencing inner params now uses const ref rather than a struct copy. | 20 December 2017, 16:53:14 UTC |
28dc247 | Balint Joo | 19 December 2017, 15:18:10 UTC | Added some time logging. So we can get a feeling of MD time, energy time, etc. | 19 December 2017, 15:18:10 UTC |
45f8dd7 | Balint Joo | 14 December 2017, 19:34:13 UTC | New readcern.cc from Colin | 14 December 2017, 19:34:13 UTC |
9b36d49 | Balint Joo | 14 December 2017, 19:26:05 UTC | Messed around with MRE Extrap predictor. Added in code to reduce the number of matrix multiplies, but this is not yet working right. Predictions are still expensive, and not as good as before. So I will leave the code in but use the old interfaces. I made the reporting a bit better in the QUDA solvers to allow easier extraction of predition times, solver times etc. | 14 December 2017, 19:26:05 UTC |
79493d3 | Balint Joo | 07 December 2017, 15:57:20 UTC | The inline unsmeared hadron node does not build with QDP-JIT due to lack of isnan, isinf, etc in QDP-JIT/LLVM. So for now I have disabled this measurement in QDP-JIT builds | 07 December 2017, 15:57:20 UTC |
69c1342 | Balint Joo | 04 December 2017, 20:14:35 UTC | Exposed setup location via XML tag SetupOnGPU *) tag has n_levels - 1 boolean values in a list (for whether setup should be done on CPU for that parent level) *) lowest level doesn't do setup (hence n_levels -1 ) *) If tag has 1 value it should be broadcast to all levels. | 04 December 2017, 20:14:35 UTC |
b968000 | Balint Joo | 04 December 2017, 15:05:04 UTC | Merge branch 'feature/unsmeared-node' into devel | 04 December 2017, 15:05:04 UTC |
c7d372e | Balint Joo | 29 November 2017, 16:08:29 UTC | Propagated QDP/JIT QUDA startup based on MPI Comm split from QDP-JIT experimental branch. This will only be called if the QDP_USE_COMM_SPLIT_INIT is defined in qdp_init.h So it should be safe for non QUDA/QDP-JIT builds hopefully. | 29 November 2017, 16:08:29 UTC |
27dbbc6 | Robert Edwards | 27 November 2017, 04:21:59 UTC | Merge branch 'feature/unsmeared-node' | 27 November 2017, 04:21:59 UTC |
c43042a | Robert Edwards | 25 November 2017, 17:03:43 UTC | Changed unsmeared meson nodes to have an explicit gamma5 on the left solution vectors. This could be handled in the insertion, but less confusing to have an explicit gamma5. | 25 November 2017, 17:03:43 UTC |