https://github.com/JeffersonLab/chroma
Tip revision: 0682c39bf2c0b1223f4e83d965084020ae9dfdb2 authored by Balint Joo on 17 January 2019, 16:30:45 UTC
Added #ifdef QUDA_DEBUG guards around the diagnostic stuff
Added #ifdef QUDA_DEBUG guards around the diagnostic stuff
Tip revision: 0682c39
NEWS
Nov 7, 2017 - Balint Joo
Fixed issue of CERN IO
Nov 4, 2017 - Balint Joo
Added extra interfacing options for current quda feature/multigrid
Added Interface for MG-Proto code with QPhiX outer layers.
JIT-dslash/clover etc interfaces now require QDP-JIT/NVPTX with LLVM-4.0 (F. Winter)
Various fixes (e.g. gcc-6.3.0 readyness)
Known issues:
- some regressions fail with Intel 17 compilers
- This version tested with gcc-6.3.0 and passed regressions (apart from eigcg, which was not built
and some 1 flavor tests which would need the GMP interface code, also not built).
- QDP-JIT does not yet support OSubLattice so some measurements relying on this are commented out
- Some of the programs in tests/ do not link with QUDA -- these are not involved in regression tests
probably could be fixed with rejigging/reordering/duplicating -l flags
May 11, 2017 - Balint Joo
Merged in branch feature/hmc_mg for QUDA HMC & Multigrid
Merged in branch feature/mg_proto for new KNL MG code
Fixed issue of isnan, isinf, etc with old ICCv16 update 3
Checked basic regressions: failed with ICCv17 update 2, worked with ICCv16 update 3, gcc-5.3, gcc-6.2
Jun 24, 2016 - Balint Joo
Merged in branch feature/quda_multigrid
Fixed NEF operator parity to track fix in Quda
Checked with QUDA devel (commit bc696cf85917c813497352353d22a09c5ce02966
)
+ MG worked, NEF worked
Reran and checked regressions on CPU
(Not Checked: hmc -- will redo this post Titan build)
Sep 9, 2015 - Balint Joo
Added gtest testing: enable with --enable-gtest
Added Basic FGMRES-DR solver. Uses any LinOpSystemSolver as a preconditioner (used MR in testing)
Needs LAPACK (--enable-lapack) to work correctly. Currently tested with MKL.
Modified qdp_lapack submodule appropriately. There are some unit tests in
mainprogs/tests/t_inv_fgmresdr which need --enable-gtest to build and a double
precision build to run correctly. SP will still build, but will hit precision limits
on assertions. Solver should(?) work in pure FGMRES mode (no deflation) if LAPACK is
not enabled. XML example is included inline in mainprogs/tests/fgmres_dr_tests.cc.
All testing so far done on unit testing. Now ready for larger volumes/lighter quarks.
Jun 10, 2015 - Balint Joo
Updated Multi-Grid interface and Wilson MG
Feb 6, 2015 - Balint Joo
Pulled in QPhiX Mods for BG/Q
Updated std:: namespace issues
Updated CPP-Wilson Dslash with threading patches from Jacques Bloch, Regensburg University
Fixed CLANG compilation issue in the qdp-lapack submodule
Sept 11, 2014 - Balint Joo
QPhiX library for Xeon and Xeon Phi (re) integrated.
See https://http://jeffersonlab.github.io/qphix
Aug 12, 2014 - Balint Joo
MDWF solver can now do NEF type (Moebius solves) as a LinOpSysSolverArray
QUDA Mobius solver integrated for props as a LinOpSysSolverArray
(tested with QUDA, quda-0.7 branch commit ID: b79f883658f780d587bc1e6f3e494d6ed937b506 )
July 4, 2012 - Robert Edwards
Version 3.42.0
Imported changes from devel branch (new propagator support).
Oct 29, 2010 - Robert Edwards
Version 3.40.0
Overhauled usage of MapObject classes to match new QDP++ interface. The
MapObjectDisk now supports multiple closure/openings/insertions.
Oct 09, 2010 - Robert Edwards
Version 3.39.0
Pulled out the MapObject classes, made them more general, and put them in QDP.
Cleaned out some old regression tests.
Jan 07, 2010 - Robert Edwards
Version 3.38.0
New FILEDB interface down in QDP++. Changed Chroma's SerialDBKey/Data to match
accordingly.
Jan 07, 2010 - Balint Joo
Version 3.38.0
Merged 3.37.2 devel branch onto master. Fixed regression checks.
Ready to go into stable
Jan 01, 2010 - Robert Edwards
Version 3.37.2
Made a combined prop_colorvec and prop_matelem_colorvec that saves
prop dbs and combines in the peram construction. Can avoid using
lookups for solution vectors or storing all the solutions in memory/disk.
Dec 31, 2009 - Robert Edwards
Version 3.37.1
Moved old inline_prop_matelem_colorvec to a low-memory version.
New inline_prop_matelem_colorvec reorganizes the loops to cut-down
on lookups.
Dec 27, 2009 - Robert Edwards
Version 3.37.0
Balint wrote a nifty MapObjectDisk and MapObjectMemory that function like
databases. A use is to hold lattice objects. New factory system for the
beasties. Removed uses of SubsetVectors in favor of MapObject's. Cleaned
out some old routines. Lots of other cleanups. Rearranged more the MapObject
task system.
Sep 14, 2009 - Robert Edwards
Version 3.36.1
Switched back to old inline genprop code. Changed inline colorvec codes
to check for DB existence. If it does not exist, insert the metadata,
else do not reinsert. Lots of improvements in laph codes.
Aug 25, 2009 - Robert Edwards
Version 3.36.0
The start of a hybrid stochastic-distillation method (called LAPH -
horrible name) by Colin and John. Changed how genprops are contracted
in an effort to speed it up. Balint did lots of XLC cleanups.
July 14, 2009 - Robert Edwards
Version 3.35.0
Balint did all this in 3.34.1, but the latter was never actually
tagged it. John Bulava added new measurement routines related to
distilled dilution.
7/8/2009 - Balint
Version 3.34.1
Added IBiCGStab for BiCGStab with 1 Global Reduction per iteration
Added Reliable IBiCGStab (Clover only) to allow multi-precision use.
Added Regressions, tweaked output.
05/202009 - Balint Joo
Added Multi-Prec BiCGStab Solver (LinOp & MdagM) (Clover Only)
Added Internally Single Precision CG Solver (MdagM only,Clover Only)
04/21/2009 - Balint Joo
Version 3.34.0
Added regression support for multiprec clover in propagator and t_leapfrog
Integrated cpp-wilson-dslash for multiprec dslash support. (Experimental)
03/04/2009 - Robert Edwards
Version 3.33.0
Changed around eigcg support for MdagM and M. New block colorvec support
by Kostas. New sloppy precision sse-dslash support by Balint. Disco eigcg
by Kostas and Christopher.
01/26/2009 - Robert Edwards
Version 3.32.0
Delta tau version of min-norm-2 integrators. New stochastic
insertion for 3-pt corr. Added static prop colorvec. Added grid
sources. IO for eigcg vecs. Disconnected routines. Added static_light_cont
and heavy_light_cont.
11/11/2008 - Robert Edwards
Version 3.31.1
Fixed the "E" tensor operator in util/ferm/etensor. Reimplemented
deriv. seqsources to reuse basic routines. Checked against old deriv.
seqsources.
11/11/2008 - Robert Edwards
Version 3.31.0
New stochastic generalized prop (3-pt) by Kostas. Added twisted-mass to
clover. Put DB's into projection use. Fixed up colorvec generalized prop.
New ILU style time-prec for clover. New readers for eigvecs.
08/26/2008 - Robert Edwards
Version 3.30.4
Changed format of colorvec dbs. Now use fixed record dbs.
08/26/2008 - Robert Edwards
Version 3.30.3
A working version of the colorvec chain.
08/02/2008:
Version 3.30.2
Balint did some rework of the comms in the sse wilson dslash.
Started adding some Berkeley DB flags, but these will all change.
07/21/2008:
Version 3.30.1
Kostas added some static-light routines from Will Detmold.
05/30/2008:
Version 3.30.0
Major overhaul of monomial structure which are now of the form
pseudoferm term tag
A=M^dag*M chi*A_1^-1*chi two_flavor
chi*M_2*A_1^-1*M_2^dag*chi two_flavor_ratio_conv_conv
chi*R(A_2)*A_1^-1*R(A_2)*chi two_flavor_ratio_conv_rat
chi*R(A_1)*chi one_flavor_rat
chi*M_2*R(A_1)*M_2^dag*chi one_flavor_ratio_rat_conv
chi*R(A_2)*R(A_1)*R(M_2)*chi one_flavor_ratio_rat_rat
This structure exists for both 4D and 5D variants. Not all are
implemented at the level of the derivatives, but can be. The 5D
monomials no longer have an explicit reference to the
Pauli-Villars (PV) linOpPV. That linOp is now added as an
appropriate separate monomial or the M_2 term. This is handled
simply by giving a DWF-like fermact with a quark mass of 1, or it
might require a kind of PV fermact in the case of some of the
more esoteric 5D actions.
05/22/2008:
Version 3.29.2
Precision related tweaks in some HMC monomials. Some new variants of qqqNucNuc
measurements and hadron_contract measurements.
05/06/2008:
Version 3.29.0
-- Support for MDWF Moebius DWF Library cleaned up and ready to go.
05/05/2008
Maintenance:
-- Added support for interfacing to MDWF Moebius DWF library from MIT.
Currently only the Shamir variant of DWF is used and only in the
propagator code.
Currently we are having difficulties with mixed precision so
I am enabling the double precision only.
05/02/2008
Version 3.28.5 - Maintenance release:
some changes here and there, but most importantly,
first version of the code to attempt to read write usqcd propagators
THIS VERSION NOW NEEDS qdp++ CVS Version qdp1-26-0 to build
Known problems:
FIXME: scalar-7n: asqtad candidate RUN_FAILS
sink_smearing_s.candidate RUN_FAILS
spectrum_s.candidate.xml RUN_FAILS
spectrum_s_HISQ.candidate.xml RUN_FAILS
Looking at the core file: segfault in 0x00000000005efd7b in Chroma::QDPStaggeredDslash::apply ()
01/21/2008 -- Robert Edwards
Version 3.28.1 - Cleaned out old intel_sse_dslash code. Removed all
stand-alone stout_smear routines. Changed task version to go through
stout_utils package. Some new versions of 2+2 gaugeacts. Some more
control over eigbounds.
01/21/2008 -- Robert Edwards
Version 3.28.0 - Overhauled plaq_gaugeact and lwldslash classes to now
use a general form of anistropy coeffients - on all plaqs or links. Now
have support for 2+2 anisotropy. New time-prec. ILU Clover hmc. New NPR
code. New HISQ code. Some new measurements for hadron_contract for deltas.
11/29/2007 -- Robert Edwards
Version 3.27.1 - Bug fix in stout_smear. This appeared in the link_smearing
but not in the fermion action smearing (the fermstate codes).
11/12/2007 -- Robert Edwards
Version 3.27.0 - First 3d SSE dslash. New SLIC-like and SLRC fermion actions.
Some new tasks. Working Eigenvalue accelerated CG code. Changed prop readers
to allow an optional Config_info.
10/26/2007 -- Balint Joo
Version 3.26.0 - New Dslash written in intrinsics with unit test suite and
complete QMT harness (as opposed to just 32bit parscalar).
--enable-sse-wilson-dslash now enables new dslash. However this may be
slow on 32 bit systems and pentia (rather than opterons). So the old GCC
Inline assembly dslash can be enabled instead with the --enable-sse-wilson-dslash-old
configure switch. Both new and old dslashes cannot be in the same build.
Its one or t'other. Code passed regressions on 7n and compiled and passed
all regressions but the mres one in parscalar-single with the Intel compiler
and the new dslash.
10/24/2007 -- Robert Edwards
Version 3.25.0 - New qdp-lapack package included in other_libs. New
system-solver routine "eigcg" that uses Kostas' eigenvector acceleration
technique. Renamed the eoprec logdet monomial to agree with other naming
schemes.
09/12/2007 -- Balint Joo
Version 3.24.3 - Lots of tydiups and fixes especially to SF related code.
5D solvers work with GCC-4. Additions for 7n related regression testing.
CG-DWF problems hopefully fixed in the chroma-config script file.
08/23/2007 -- Robert Edwards
Version 3.24.2 - Cleaned up some SF Dirichlet code. Mainly, this version
marks a QDP++ version that has QIO properly working on mesh machines.
08/14/2007 -- Robert Edwards
Version 3.24.1 - Some rearrangement in "hadron_contract" routines. Now
those regressions work in parallel. This code requires at least
QDP++ version 1.24.3 or greater.
07/17/2007 -- Balint Joo
Version 3.24.0 - Loads of I/O related stuff from Robert. Balint had
Clover related improvements including support for bagel_clover
on BG/Ls, speed improvements to clover force etc through
'cleanup'. Negative Logdets hopefully gone (but not forgotten) as we
moved to L^\dag D L inversion of clover term rather than Cholesky. The
former can cope with an indefinite matrix (as long as it is
invertible) the latter can't.
04/19/2007 -- Robert Edwards
Version 3.23.0 - Moved around some monomial files in anticipation of big reorg
of Remez and InvertParam groups. New output from hadspec.
05/03/2007 -- Balint Joo
Version 3.22.7 - BG/L fixes for BAGEL dslash op (refcounting etc)
New CG-DWF option --enable-cg-dwf-lowmem for a lower memory CG-DWF
New CG option --enable-cg-solver-restart for a single CG restart
RsdCG and MaxCG can be independently speficied by <RsdCGRestart>
and <MaxCGRestart> in the inv Param file. This works by default
for CG-DWF solves and the tags are respected in other CG solvers if
--enable-cg-solver-restart is enabled. Otherwise these XML tags
are ignored.
27/02/2007 -- Balint Joo
Version 3.22.6 - Cleanded out subsets. Regressions pass but needs recent
QDP++ (1.22.0 at least). Merged in Robert's recent changes. Added Schur Preconditioned Space,
Central Time preconditioned Wilson Operator.
22/02/2007 -- Balint Joo
Version 3.22.5 - Stamp before making subset related changes
18/02/2007 -- Balint Joo
Added ILU Preconditioned space and Central Preconditioned Time Wilson And Clover
Actions. Amazingly these actually work.
15/02/2007 -- Balint Joo
First cut Unprec Space Central Prec Time Wilson Fermions. Complete with regression tests
8/02/2007 -- Balint Joo
Bug fix for linking: Removed enum_wvfkind files to attic. Unincluded
them from enum_io.h
12/10/2006 -- Balint Joo
Version 3.22.1. Bug fix in how MD chronological predictor is reset.
12/10/2006 -- Balint Joo
Version 3.22.0. Overhaul of the MD integrator codes. Now recursive. Removed
old code.
12/10/2006 -- Robert Edwards
Version 3.21.1. Mostly housecleaning. Removed many obsolete source construction
and quark smearing codes and enums. Added a Klein-Gordon operator.
12/04/2006 -- Robert Edwards
Version 3.21.0. Overhauled sequential source construction. Now have loads
of working derivative sources. New regression tests that cross check the
sequential propagator via the tie-back test and compares to a 2-pt at
the sink. Standardized propagator headers - now have fixed structs and
not the floating xml stuff.
11/25/2006 -- Robert Edwards
Version 3.20.0. New component integrator codes for HMC. New staggered
chroma task chain of MAKE_SOURCE, PROPAGATOR and SINK_SMEARING .
Fixes in NPR task codes.
11/07/2006 -- Robert Edwards
Version 3.19.1. NPR code compiles and linked, but not fully tested.
Small fix in AVP DWF-cg wrapper to correct cg-counts.
10/31/2006 -- Robert Edwards
Version 3.19.0. Split apart the fermact.h file into lots of
separate bits to make it easier to manage. Split up the linearop.h
file more as well. Split up and rearranged the tprec_linop.h files for
the anticipated new time-preconditioning. Added some new functionality
here. For neatness sake, moved and renamed all the "prec_*" files
(which are specific to 4d even-odd preconditioning) to "eoprec_*"
filenames. Start of some new inline measurements not yet turned on.
10/31/2006 -- Robert Edwards
Version 3.18.0. New inline measurements Qll by K. Orginos.
10/14/2006 -- Robert Edwards
Version 3.17.0. Changes in BuildingBlocks (D. Renner, K. Orginos)
to support time reversal and translation of source location. Removed
inline_spectrum and inline_spectrumOct - superceded by inline_hadspec.
Reset inline_propagator version to 9. Had to regenerate lots of
regression tests. Removed old style restart support - now done down
inside the system solvers.
10/11/2006 -- Robert Edwards
Version 3.16.1. Small fix in hadron_seqsource to make gcc4 happy.
10/10/2006 -- Robert Edwards
Version 3.16.0. Change in seqsource construction. Now pass in full
propagator headers as well as props. For baryons, now compensate
for time-ordering of source and sink baryons as well as the
spurious phase from a quark source not at the spatial origin.
10/10/2006 -- Robert Edwards
Version 3.15.1. Small changes in plaq+adj gauge action. Other small
regression test cleanups.
09/21/2006 -- Robert Edwards
Version 3.15.0. Added new ferm and gauge create state inlines.
Changed some inlines to now use the ferm or gauge state, or at
least the param group.
09/21/2006 -- Robert Edwards
Version 3.14.7. Added a registration in inline_wilslp
09/21/2006 -- Robert Edwards
Version 3.14.6. Major housecleaning. Changed all the uses of the
registration method to now be **runtime**. The "registerAll()"
function is exposed that will register the thingy. There is a
private "registered" flag that is initially false and then set
to true upon registration done at runtime.
09/18/2006 -- Robert Edwards
Version 3.14.5. House cleaning. Split off all gauge states
and ferm states to their own gaugestates and fermstates
directories. Changed all the paths.
09/14/2006 -- Robert Edwards
Version 3.14.0. New chroma/other_libs/cg-dwf package now
included. The DWF QpropT routine reworked to use the new package
(double prec. SSE). Changed hmc to split output into XMLLOG
and XMLDAT.
09/09/2006 -- Robert Edwards
Version 3.12.0. First apparently working version of stout force
code for HMC. Rearrangements in stouting for gauge and fermion
actions.
08/28/2006 -- Robert Edwards
Version 3.11.2. Small but important edge case in stout_fermstate
fixed. Bug fix in SftMom handling in test of zero_offset. Small
fix in inline_hadspec allowing optional displacement type
field. Retooled all important update routines, stouts, fermacts,
etc. to use START_CODE/END_CODE.
08/24/2006 -- Robert Edwards
Version 3.11.1. Fixed regression metric in mesons from v3.11.0 .
Rearranged mesplq and added measurements of all plaquette planes.
Regenerated all regression output since they depend on the
plaquette output. Fully converted the block-ing code along with
fuzglue, polycor and gluecor.
08/19/2006 -- George T. Fleming
Version 3.11.0. Changed the interface of the slow Fourier
transform (SftMom) class to allow for any lattice point to be
chosen as the spatial origin. Previously, this meant that the
origin was always implicitly (0,0,0,0), which lead to various
phase problems. One example is 2pt correlation functions between
a point source not at the origin and a sink at non-zero momentum:
the overall phase depends upon the location of the source.
Using the new interface, the phase can be made independent of
the location of the source by choosing the origin of the Fourier
transform to be the location of the source. Several things remain
to be fixed: (1) sequential sources at non-zero sink momentum,
(2) building blocks with link paths which cross boundaries, (3)
baryonic sequential sources where t_source > t_sink, i.e. the
boundary is between the source and sink.
08/17/2006 -- Balint Joo
Version 3.10.4. Fixed stout fermstate to support smear_dir.
New form of purgaug input/output.
08/15/2006 -- Balint Joo
Version 3.10.3. Stout Fermstate rides. Reinterpret nontrivialP()
flag to mean underlying gauge boundaries are trivial in FermBC-s
only (ie still old meaning in GaugeBC-s for now). Stout Fermstate
tested for SF, periodic, antiperiodic, and Dirichlet fermion
BCs. Twisted Ferm BC_s are not expected to work.
07/27/2006 -- Balint Joo
Version 3.10.2. Patches to SSE DWF CG to support double-prec sums.
Added more optimized plaq+two_plaq gauge monomial. Added profiling
to inline_eigbnds in 5D case.
07/21/2006 -- Balint Joo
Version 3.10.1. Added two_plaq monomials.
07/11/2006 -- Robert Edwards
Version 3.9.0. Changed interface to the inline WILSLP and MESONSPEC.
Turned on HADRON_SPECTRUM (hadspec). This is a replacement for SPECTRUM.
07/11/2006 -- Robert Edwards
Version 3.8.2. More small fixes in wilslp (Wilson loops).
07/06/2006 -- Robert Edwards
Version 3.8.1. Fixes in wilslp (Wilson loops).
07/04/2006 -- Robert Edwards
Version 3.8.0. Changed FermionAction API to produce system
solver classes. No longer have a fixed InvertParam
set. Removed old inverter enum and support. Have default creation
for inverters to solve M*psi=chi, MdagM*psi=chi, and multi-shift
support. Some older files still have explicit calls to CG solver.
Converted usage of source/source_type, fermact, sink/sink_type
to using GroupXML_t. Converted all monomials to use GroupXML_t as
well.
07/03/2006 -- Robert Edwards
Version 3.7.4. 5D Hasenbusch monomial support.
06/18/2006 -- Robert Edwards
Version 3.7.3. Bug fix in xml output in task MESONSPEC. Some code cleanups to make
various compilers happy.
06/11/2006 -- Robert Edwards
Version 3.7.2. Fixed param path in inline_link_smear.
06/11/2006 -- Robert Edwards
Version 3.7.1. Bug fixes. Removed all use of linkSmear - it is problematic
with the current qdp++.
06/11/2006 -- Robert Edwards
Version 3.7.0. Change in quarkProp and SystemSolver interface.
QuarkProp now takes a number of tries (restarts) at calling the qprop routine.
The SystemSolver routines now return a structure of iteration counts and
residual. Changed all regression tests to use this new propagator option
(version 10).
06/08/2006 -- Robert Edwards
Version 3.6.2. Improved memory usage with inline_mesonspec. Some
fixes/improvements in polynomial linops. New parallel regression runners.
05/26/2006 -- Robert Edwards
Version 3.6.1. Improved inline_mesonspec. Small bug fix down in
64bit SSE wilson dslash (intel_sse_wilson_dslash).
05/24/2006 -- Robert Edwards
Version 3.6.0. New format for inline_mesonspec. Added spin insertions.
Overhauled inline_stoch_baryon. Removed quark_smear_first/last from
shell source/sink construction/smearing. Now print true residuals
in some inversion routines.
05/19/2006 -- Robert Edwards
Version 3.5.1. Added no-quark-smearing.
05/17/2006 -- Balint Joo
Version 3.4.2. Added Support for configurable testcase runners.
Added mpirun_rsh runner. This runner is currently hardwired to
run the tests on qcd6n277 and qcd6n278
05/11/2006 -- Robert Edwards
Version 3.4.1. First possibly working version of stoch_baryon task.
05/10/2006 -- Robert Edwards
Version 3.4.0. Moved stoch_meson to an inline measurement. Removed stoch_meson.cc
mainprog. Made a new stoch_baryon inline measurement - not functional yet.
Fixed 5D fermacts to actually call their desired quarkProp routines. They
were missing the "const" qualifier at the end of the decl (thus new funcs
not seen by virtual funcs).
05/09/2006 -- Robert Edwards
Version 3.3.0. Upgraded SHELL_SOURCE,SHELL_SMEARING and SHELL_SINK
inline tasks to now take an order of the smearing versus displacements.
04/26/2006 -- Robert Edwards
Version 3.2.1. Removed SNARF task - was superceded. Added NERSC reader
task. Cleanups and generlization of sfpcac support.
04/18/2006 -- Robert Edwards
Version 3.2.0. Switched purgaug to using the gauge action construction.
Still somewhat limited, though. Putting legit aniso support in more of the
gauge actions - still more is needed.
04/14/2006 -- Robert Edwards
Version 3.1.1. Small changes. configure.ac now writes an empty run command for
make xcheck.
04/11/2006 -- Robert Edwards
Version 3.1.0. Big change on chroma,hmc,purgaug input - one must now always
specify the input gauge field on tasks that need it. The default reader
has been removed. Added more Schroedinger functional measurements.
04/04/2006 -- Robert Edwards
Version 3.0.1. Minor compilation fixes.
04/03/2006 -- Robert Edwards
Version 3.0.0. Major overhaul of fermion and gauge action interface. Basically,
all fermacts and gaugeacts now carry out <T,P,Q> template parameters. These are
the fermion type, the "P" - conjugate momenta, and "Q" - generalized coordinates
in the sense of Hamilton's equations. The fermbc's have been rationalized to never
be over multi1d<T>. The "createState" within the FermionAction is now fixed meaning
the "u" fields are now from the coordinate type. There are now "ConnectState" that
derive into FermState<T,P,Q> and GaugeState<P,Q>.
03/21/2006 -- Robert Edwards
Version 2.15.3. Bug fix, or rather new addition to funcmap. Changed all
IO funcmaps to disambiguate them and not pile onto the same funcmap!!!
03/21/2006 -- Robert Edwards
Version 2.15.2. Bug fixes in new derivative displacements. Changed around
output regression tests. Some cleanups in IO.
03/21/2006 -- Robert Edwards
Version 2.15.1. Removed use of overlap_state from various 5d fermacts
since it is not used. Added more regression tests. Cleaned out some
old input files.
03/20/2006 -- Robert Edwards
Version 2.15.0. Major change of inline suite. There is no longer a gauge field
passed into the measurements. Instead, tasks must read a gauge field. By default,
they all will try to read a "default_gauge_field" buffer placed there by the
main program. With this change, any task can smear a gauge field and hand
that to the next task for use instead of having to write the config, exiting,
and restarting.
03/16/2006 -- Robert Edwards
Version 2.14.0. Major overhaul in GaugeBC's and FermBC's . New initial
support for Schroedinger functional BC's. The GaugeBC's seem to work
under HMC (at least energy is conserved). Carried over BC's known under
SZIN. However, the FermBC's don't conserve energy yet. This is understood -
the LinOP.deriv() functions do not know the BC's hence can't enforce
them - only important to HMC. This needs more work.
03/06/2006 -- Robert Edwards
Version 2.13.4. Backed out changes to clover with external field. It's
breaking something but passes regression test.
03/06/2006 -- Robert Edwards
Version 2.13.3. Changes to mesonspec code.
03/02/2006 -- Robert Edwards
Version 2.13.2. Added new partial wall source. More regression tests. Added
flop counts on clover. Start of new simplistic (minimal) meson and future
baryon spectroscopy.
02/27/2006 -- Robert Edwards
Version 2.13.1. Some mods in supporting staggered spectrum as an inline.
Some changes by Alistair Hart via Craig McNeile to fuzzed wilson loops.
New HMC integrator support by Carsten Urbach.
02/25/2006 -- Robert Edwards
Version 2.13.0. Reorg of the fermbcs and gaugebcs. Now putting in
Schroedinger functional.
02/25/2006 -- Robert Edwards
Version 2.12.1. Small cleanups before reorg of fermbc and gaugebc.
New monomials for clover.
02/23/2006 -- Balint Joo & Thomas Streuer
Version 2.12.0. MultiMass overlap and eigenvector projection.
Inline Ritz measurement. Some XML magic needed in writing the
multiple props. For this reason we now need qdp1-20-1.
02/21/2006 -- Robert Edwards
Version 2.11.0. Yet another overhaul. This time, the idea of "displacement"
in make_source and sink_smear has been generalized to another factory generated
object. Now have lots of derivative things that are "displacements" of a source
or sink. More regression tests. New staggered spectrum regression tests and code.
02/16/2006 -- Robert Edwards
Version 2.10.1. Removed chroma_config.h from chromabase.h and instead only
include it in routines that require it. Thus, recompilations are drastically
lowered upon changing versions. Added timing to SSE DWF CG.
02/16/2006 -- Robert Edwards
Version 2.10.0. Changed naming scheme for Monomials. The fermact is no longer
in the run-time monomial name.
02/16/2006 -- Robert Edwards
Version 2.9.2. Preconditioned clover monomials and logdet monomials now
working. More regression tests for t_leapfrog.
02/15/2006 -- Robert Edwards
Version 2.9.1. New clover bits. New sequential sources.
02/10/2006 -- Robert Edwards
Version 2.9.0. Overhauls of fermion action classes and monomials. More to
go. Overhauled sequential source and morphed into an object factory. More
seqsources coming. Added start of polynomial preconditioning monomials.
01/03/2006 -- Robert Edwards
Version 2.8.2. Fixed clover builds and turned back on.
01/03/2006 -- Balint Joo
Version 2.8.1. Removed clover support temporarily till build problems
sorted. Added Hasenbusch 4d monomial support.
01/01/2006 -- Robert Edwards
Version 2.8.0. New clover fermion support. Added altivec dwf support.
12/26/2005 -- Robert Edwards
Version 2.7.4. Some fixes in barhqhq - used by spectrumOct.
12/25/2005 -- Robert Edwards
Version 2.7.3. Updated some regression tests. Added a few more.
12/25/2005 -- Robert Edwards
Version 2.7.2. Updated baryon spectrum codes and seqsources to abstract away
spin matrices and functionality into small functions.
12/20/2005 -- Robert Edwards
Version 2.7.1. Small changes for regression support. Fixed little things.
12/18/2005 -- Robert Edwards
Version 2.7.0. Overhauled how aniso is done - pushed it down into the
Wilson-esque dslashes. New clover support. Most importantly, start of new
rigourous regression test system.
12/18/2005 -- Robert Edwards
Version 2.6.1. Updated many regression tests to now use a new WEAK_FIELD
instead of using the SZIN chroma/tests/test_purgaug.cfg1 config.
12/14/2005 -- Robert Edwards
Version 2.6.0. Changed FermionAction (fermact.h) interface to use a
quarkSpinType instead of a nonRelProp. Now, there can be FULL,UPPER,LOWER
quark prop calculations.
12/08/2005 -- Robert Edwards
Version 2.5.5. Small cleanups to make gcc4.0 happy.
11/30/2005 -- Robert Edwards
Version 2.5.4. Renamed PropSource_t and PropSink_t to PropSourceConst_t and
PropSinkSmear_t. Added new PropSourceSmear_t . Updates of stochastic codes.
11/30/2005 -- Robert Edwards
Version 2.5.3. Minutae.
11/20/2005 -- Robert Edwards
Version 2.5.2. Bug fixes. Hyp_smear3d didn't set the time link correctly.
Fixed stout smearing when called from quark smearing - otherwise it was
fine. Added LINK_SMEAR inline to replace all the separate inlines.
11/20/2005 -- Robert Edwards
Version 2.5.1. Some library routines require qdp-1.18.X . Rearranged diluted
source construction.
11/20/2005 -- Robert Edwards
Version 2.5.0. Some new main programs require qdp-1.18.X . The features are
new binary ops on Seeds.
11/16/2005 -- Robert Edwards
Version 2.4.3. Changed names in smearing and source/sink constructions.
11/08/2005 -- Robert Edwards
Version 2.4.2. Some new sources.
11/08/2005 -- Robert Edwards
Version 2.4.1. Small changes to source and sink smearing to get it to compile
on different platforms.
11/08/2005 -- Robert Edwards
Version 2.4.0. Overhauled source mechanisms. Now have classes and factories
for quark source construction, quark source smearing, quark sink smearing,
quark smearing generically, and link smearing. MAKE_SOURCE and SINK_SMEAR
now use these source_construction, and sink_smearing, resp.
11/08/2005 -- Robert Edwards
Version 2.3.6. Last changes and additions before move to new source
mechanism.
11/04/2005 -- Robert Edwards
Version 2.3.5. Version using new blas routines suitable for twisted-mass ferms
from qdp. Some new IO minutae.
11/02/2005 -- Robert Edwards
Version 2.3.4. Tiny changes.
10/25/2005 -- Robert Edwards
Version 2.3.3. Some housecleaning. Moved some files from hadron to sources.
Added flop counts for 4D CGNE. Removed some spurious "const" from linearop.h .
10/25/2005 -- Robert Edwards
Version 2.3.2. Last tweaks before moving around files.
10/24/2005 -- Robert Edwards
Version 2.3.1. Commented out some unused inline stuff.
10/24/2005 -- Robert Edwards
Version 2.3.0. New noisy sources in inline_make_source and
10/24/2005 -- Robert Edwards
Version 2.2.3. Added PARTFILE to qdp_volfmt string-to-enum map. Added call of
fermbc aggregator in fermacts.
10/18/2005 -- Robert Edwards
Version 2.2.2. Minor tweaks.
10/12/2005 -- Robert Edwards
Version 2.2.1. Tightened some QDP file checks. Added updating of xml
from gauge_startup in cases where no files read.
09/29/2005 -- Robert Edwards
Version 2.2.0. Stout support added to HMC.
09/29/2005 -- Robert Edwards
Version 2.1.0. Building blocks moved to version 3.
09/26/2005 -- Robert Edwards
Version 2.0.4. Added flavor label to inline_building_blocks.
09/26/2005 -- Robert Edwards
Version 2.0.3. Some cleanups in the inline IO codes.
09/26/2005 -- Robert Edwards
Version 2.0.2. Commented problematic conversion codes in QIO inline reader/writers
for now.
09/26/2005 -- Robert Edwards
Version 2.0.1. Fixed paths in chroma/tests/chroma . Fixed up seqsource
conventions.
09/25/2005 -- Robert Edwards
Version 2.0.beta. MAJOR UPGRADE! Many main progs vanish. New emphasis on inline
measurements. New regression tests in chroma/tests/chroma/* that mirror the
chroma/lib/meas/inline/* directories.
09/23/2005 -- Robert Edwards
Version 1.24.1. More updates of IO inlines.
09/23/2005 -- Robert Edwards
Version 1.24.0. Lots of new staggered measurements. New IO inline routines.
09/13/2005 -- Robert Edwards
Version 1.23.11. Fixes to lib/Makefile.am to solve install build problems.
09/06/2005 -- Robert Edwards
Version 1.23.10. New staggered meson spectrum code.
08/31/2005 -- Robert Edwards
Version 1.23.9. Added qqbar and mescomp support.
08/27/2005 -- Robert Edwards
Version 1.23.8. Added v4 back to prop reads. New staggered spectrum
measurements.
08/23/2005 -- Robert Edwards
Version 1.23.7. Added timings to inline measurements around IO routines.
08/23/2005 -- Robert Edwards
Version 1.23.6. Cosmetic changes before some production work. Corresponds
to using QDP++ 1.15.2 .
08/18/2005 -- Robert Edwards
Version 1.23.5. Added hypsmear4d and smearing stuff to inline measurements.
07/29/2005 -- Robert Edwards
Version 1.23.4. Added new info to building-blocks footer - new version 2.
07/20/2005 -- Robert Edwards
Version 1.23.3. Fixes to barseqsrc routines to fix malfeature/conventions of
quarkContract12.
07/19/2005 -- Robert Edwards
Version 1.23.2. New inline multi-mass propagator routine. Chroma.cc
now takes RNG seed.
07/13/2005 -- Robert Edwards
Version 1.23.1. New flopcounts in minvcg codes.
06/29/2005 -- Robert Edwards
Version 1.23.0. Major changes by Balint Joo in bagel lib, new flop counts,
new memory hints from QDP++. Most of these aimed at QCDOC.
06/14/2005 -- Balint Joo
Version 1.22.3. Fixed various things in the build system.
GMP is now enabled with --with-gmp=<DIR>. DIR is optional.
If not given one can use env variables GMP_CXXFLAGS, GMP_LDFLAGS
and GMP_LIBS. If these environment variables are empty, GMP_LIBS
will be defined as -lgmp assuming that the GMP is in a standard
location. Not specifying --with-gmp disables building the GMP
remez code
The BAGEL WILSON DSLASH is now enabled with --with-bagel-wilson-dslash=DIR. DIR is optional. If not given it falls back to environment variables
BAGEL_WILSON_DSLASH_CXXFLAGS, BAGEL_WILSON_DSLASH_LDFLAGS, BAGEL_WILSON_DSLASH_LIBS. Not giving the --with-bagel-wilson-dslash option disables the
use of the bagel wilson dslash
05/28/2005 -- Robert Edwards
Version 1.22.2. Moved dwf_linop_base to dwflike_linop_base. Made
the 1-flav Monomials also support 1 & 3 as a special case of more
generic fractional flavor.
05/27/2005 -- Balint Joo
Version 1.22.1. Cleaned up prec_ovlap_contfrac5d_linop codes.
05/27/2005 -- Balint Joo
Version 1.22.0. Added new --enable-opt-cfz-linop option. Now
have some optimized version of prec_ovlap_contfrac5d_linop.
05/27/2005 -- Balint Joo
Version 1.21.6. Changed integrators to use num-steps instead of
step-size.
05/19/2005 -- Robert Edwards
Version 1.21.5. Mods in CFZ codes. New HMC integrators.
05/04/2005 -- Robert Edwards
Version 1.21.4. New z2 sources. New HMC monomial control in 1-flavor.
04/25/2005 -- Robert Edwards
Version 1.21.3. Added qpropqio for chroma.
04/25/2005 -- Robert Edwards
Version 1.21.2. BJ added OvExt code and factories. Small cleanups and
adding xml_file support to inline measurements.
04/17/2005 -- Robert Edwards
Version 1.21.1. Changed one-flavor HMC monomial codes to support
nth-rootery.
04/17/2005 -- Robert Edwards
Version 1.21.0. Moved most mainprogs/main/*.cc codes to be wrappers
over inline measurements. Inline codes significantly extended.
04/06/2005 -- Balint Joo
Version 1.20.6. Small changes before major update of mainprogs.
04/02/2005 -- Balint Joo
Version 1.20.5. Optimizations to the diag bits of the DWF,NEF variants
03/22/2005 -- Robert Edwards
Version 1.20.4. Added mainprogs/main/qpropqio.cc
03/21/2005 -- Robert Edwards
Version 1.20.3. No major changes, but an improvement in qdp++ to the
sumMulti to now use a QMP global sum on an array instead of individual
sums.
03/14/2005 -- Robert Edwards
Version 1.20.2. Fixed Building-Blocks stuff (ExampleBuildingBlocks)
to now do BB for each seqprop instead of all together. This
helps get around the max file-descriptor problem.
03/12/2005 -- Robert Edwards
Version 1.20.1. Changed ExampleBuildingBlocks to use an XML
interface.
03/06/2005 -- Robert Edwards
Version 1.20.0. Overhauled sequential source construction.
Eliminated "magic numbers" for sources. Now all string
driven. Added force measurements to HMC.
03/06/2005 -- Robert Edwards
Version 1.19.14. Bug fix in SzinGauge_t constructor. This was
introduced when the initHeader's were removed in favor of
constructors.
03/02/2005 -- Robert Edwards
Version 1.19.13. Bug fix in ZoloNef FermAct codes. Only showed
up in Zolo case. This bug was introduced 2/14/05 - now squashed.
03/02/2005 -- Robert Edwards
Version 1.19.12. Argh - bug fix. Resize's in linop array routines
caused problems. Updated command-line args to all main programs.
02/27/2005 -- Robert Edwards
Version 1.19.11. Small code simplifications.
02/27/2005 -- Robert Edwards
Version 1.19.10. 5D versions of Chrono predictor in place.
Changed input format of sink_smear and qqq_w to take arrays.
02/23/2005 -- Balint Joo
Version 1.19.9. Upgrade again of Chrono predictors. Seems to
work for 4D. Need 5D.
02/23/2005 -- Robert Edwards
Version 1.19.8. Updated interface for Chrono predictors. Start of
aniso support in GaugeActs. Split off some structs to separate
files to thin-out file dependencies.
02/21/2005 -- Robert Edwards
Version 1.19.7. Added aniso support for DWF and SSE DWF.
Changed all initHeader-s to appropriate default constructors
for the structs. Changed quarkProp routines to take SystemSolver
instead of FermAct.
02/21/2005 -- Robert Edwards
Version 1.19.6. Moved stock mainprogs to bin_PROGRAMS. They
are now installed.
02/16/2005 -- Robert Edwards
Version 1.19.5. Added hybrid meson stuff to spectrum_w.
02/13/2005 -- Robert Edwards
Version 1.19.4. Changed/fixed all ContFrac codes to have a +OverMass
sign convention. Changed Zolo fermacts to take ApproxMin/Max from
FermionAction group.
02/10/2005 -- Robert Edwards
Version 1.19.3. Fixed ritz code to work over subsets.
02/09/2005 -- Robert Edwards
Version 1.19.2. Small bug fixes. Split off hypsmear.cc
02/06/2005 -- Robert Edwards
Version 1.19.1. Added a bunch of aggregators to simplify
linkage stuff.
02/02/2005 -- Robert Edwards
Version 1.19.0. One flavor RHMC support now in place.
One currently needs a --enable-gmp flag on configure
to hook in GNU GMP.
02/02/2005 -- Robert Edwards
Version 1.18.9. Added missing files to lib/Makefile.am .
Upgraded Remez codes. New configure switch for GMP.
02/01/2005 -- Robert Edwards
Version 1.18.8. Cosmetic changes. Code sync.
01/31/2005 -- Robert Edwards
Version 1.18.7. Fixed output names in proginfo to produce valid XML !
01/31/2005 -- Robert Edwards
Version 1.18.6. Fixes in HT fermact codes.
01/27/2005 -- Robert Edwards
Version 1.18.5. Split abs_monomial into separate
two flavor and new one flavor rational support.
01/16/2005 -- Robert Edwards
Version 1.18.4. Mainprogs including HMC finally appears
stable after all the XML singleton instances, etc. More
Chrono predictors.
01/16/2005 -- Robert Edwards
Version 1.18.3. Added more 5D monomials. Added Pauli-Villars
for contfrac5d.
01/15/2005 -- Robert Edwards
Version 1.18.2. Fixed norms on new gaugeacts. Seem okay.
01/15/2005 -- Balint Joo
Version 1.18.1. Added missing include files to lib/Makefile.am
Fixed norms on new gaugeacts. Seem okay.
01/14/2005 -- Robert Edwards
Version 1.18.0. Major housecleaning. All lib routines now in
Chroma namespace. Removed all "using" directives in lib routines.
The QDP namespace is however, "used" in chromabase.h
01/14/2005 -- Robert Edwards
Version 1.17.1. Some code cleanups/namespaces. Logging in HMC.
Merged in Al Hart's changes.
01/13/2005 -- Balint Joo
Version 1.17.0. Major reorg of molecdyn dir. Files split and
rearranged. Basic HMC structure in place - still some polishing.
Gauge acts like plaq,rect,pg seem to all work.
01/12/2005 -- Robert Edwards
Version 1.16.1. Added new gauge actions (wrappers over plaq,rect,pg).
01/12/2005 -- Robert Edwards
Version 1.16.0. Rearranged actions/gauge directories.
01/12/2005 -- Robert Edwards
Version 1.15.5. More hardening of HMC. Some bug fixes in gaugeacts.
01/11/2005 -- Robert Edwards
Version 1.15.4. Added plaq,rect,pg gaugeacts. Made gauge monomial
class generic.
01/11/2005 -- Robert Edwards
Version 1.15.3. More hardeninng of HMC. Chrono prediction added.
Added ParWilson (Twisted Mass Wilson) support within HMC.
01/10/2005 -- Robert Edwards
Version 1.15.2. Changed HMC monomial routines to more generic
method.
01/10/2005 -- Robert Edwards
Version 1.15.1. Some more HMC suport. Small code cleanups
to make code pass -Wall.
01/07/2005 -- Robert Edwards
Version 1.15.0. Added SSE qpropT for DWF.
01/06/2005 -- Robert Edwards
Version 1.14.3. HMC support appears fixed.
01/04/2005 -- Robert Edwards
Version 1.14.2. Added linOpPV back into FermAct5D hierachy.
Rearranged and extended Monomial support which now has a 4D
and 5D branch. Abstracted derivative/refresh codes. Added
new monomials.
01/04/2005 -- Robert Edwards
Version 1.14.1. Small changes in derivative support.
01/01/2005 -- Robert Edwards
Version 1.14.0. Another major overhaul of FermionAction
classes. The concept of derivative is pushed up higher in
the tree. All fermacts now have some default derivative
support (possibly nothing more than Not Implemented message).
Qprop and QpropT now produced via factory functions ala
LinearOperators. Now have a SystemSolver class - can be used
for all solvers.
01/01/2005 -- Robert Edwards
Version 1.13.1. HMC now runs, but still needs work.
12/29/2004 -- Robert Edwards
Version 1.13.0. Rearranged top-level FermionAction support.
Consolidated FermFactories.
12/29/2004 -- Robert Edwards
Version 1.12.1. HMC changes and support for gauge and
ferm Monomials.
12/23/2004 -- Robert Edwards
Version 1.12.0. Now have a FermBC factory/foundries. Removed
explicit FermBC args from FermAct factories and now the creation
of the FermBC is done within the construction of the FermAct.
Similarly, HMC will create FermActs as needed. The old "boundary"
within propagator.cc input is now shoved into the FermionAction
xml.
12/23/2004 -- Robert Edwards
Version 1.11.3. More HMC changes.
12/22/2004 -- Robert Edwards
Version 1.11.2. New HMC support. First versions of Monomials.
12/15/2004 -- Robert Edwards
Version 1.11.1. Fixed chroma_config (for libs). Moved other routines
to chroma namespace. Added more to EXTRA_DIST .
12/12/2004 -- Robert Edwards
Version 1.11.0. Major overhaul of fermact and linearop class
structure. Removed dsdu stuff - now in linops. Introduced
levels of fermacts (base or not) that have derivs. All low
level fermacts/linops have an additional template param for
type of conjugate momenta in derivative. Made all fermacts and
linops within chroma namespace.
12/09/2004 -- Robert Edwards
Version 1.10.2. Changed prec. DWF-base to have linOp4D only
use prec. 5D operators.
12/09/2004 -- Robert Edwards
Version 1.10.1. Small fixes/tweaks
12/09/2004 -- Robert Edwards
Version 1.10.0. Overhaul of DWF-like class structure.
12/06/2004 -- Robert Edwards
Version 1.9.2. Changes to staggered stuff. Start of xmldiff checks.
11/22/2004 -- Robert Edwards
Version 1.9.1. Removed last of all Namelist-isms.
11/22/2004 -- Robert Edwards
Version 1.9.0. Big merge of Wilson and Staggered builds - now all
the same. Cleaned up automake files so make distcheck should work.
11/20/2004 -- Robert Edwards
Version 1.8.2. More changes within Zolo Ferm support.
Changed staggered files to use LatticeStaggeredFermion
10/21/2004 -- Robert Edwards
Version 1.8.1. More updates. First start of Zolo in NEF.
10/18/2004 -- Robert Edwards
Version 1.8.0. First version. Changed to support QMP-2. New fermion action
initialization. New support for variable initialization.
09/28/2004 -- Balint
-- ZolotarevStateInfo has been renamed to OverlapStateInfo
( and the previous empty OverlapStateInfo base class has been removed)
-- I have added the state_info field to Propagator_Input_t
as a standard string. I had it in ChromaProp_t first but I
believe it should not be part of the <Param> </Param>
tagset but rather live at the same level as Cfg_t in the
top level applications. Look in propagator.cc to see how
I manage it. Essentially I do this (propagator.cc line 84):
// Read any auxiliary state information
// if it exists
if( inputtop.count("StateInfo") == 1 ) {
XMLReader xml_state_info(inputtop, "StateInfo");
std::ostringstream os;
xml_state_info.print(os);
input.stateInfo = os.str();
}
else {
// Write an empty <?xml ... ><StateInfo/> document if not
// found
XMLBufferWriter s_i_xml;
push(s_i_xml, "StateInfo");
pop(s_i_xml);
input.stateInfo = s_i_xml.str();
}
I have added dummy createState() functions to the various
fermacts that read from an XML containing StateInfo. Apart
from the overlap, these createState()-s just ignore
the XML, but this virtual func could be a good hook for you
if you add StateInfo to some ZolotarevCoeff-ed DWF fermacts.
In fact, if you just write these, they'll be picked up appropriately
by propagator...
A sample usage of state info is in
chroma/tests/propagator/propagator_zolo4d_v6.xml.in
-- As agreed, I have removed backward compatibility with
SZIN - type vectors in createOverlapState() -- now just
uses my evecs.
-- I have written factory functions for creating OverlapConnectState-s
in overlap_state.h overlap_state.cc. I originally wrote these
as templated funcs, but then I couldn't shove them into the .cc file
- they'd needed to have stayed in the .h file where the prototypes
are defined. So I did similar things to the Inverter functions.
I wrote out the "specialisations" in the .h file and wrapped a templated
version in the .cc. Your fermact's createState()-s can just dive on these
see ovlap_partfrac_4d_fermact_w.[h,cc]. The issue of 4D and 5D BC's is
hacked around as discussed:
i) The aux ferm act gets trivial BC-s. This doesn't matter as
we never use them. We always use the BC-s of the parent.
ii) the createOverlapState() function is passed a linOp which
the caller FermAct constructs. AuxMass is therefore no longer
passed
iii) the createOverlapState() function creates a SimpleConnectState
internally to call the passed linOp with, after applying
the modifyU() function of the passed FermBC
-- Zolotarev4DFermAct and Zolotarev5DFermActArray have been renamed:
as OverlapPartFrac4DFermAct
and
UnprecOverlapContFrac5DFermActArray
and put in ovlap_partfrac4d_fermact_w.[h,cc]
and unprec_ovlap_contfrac4d_fermact_array_w.[h,cc]
the zolotarev5d_linop_array_w.[h,cc] is still there but will
also be renamed real soon now.
-- I have removed the double mappings for backward compatibility
ZOLOTAREV_4D and ZOLOTAREV_5D are dead. Now you need to specify
something like:
<FermionAction>
<FermAct>OVERLAP_PARTIAL_FRACTION_4D</FermAct>
<IsChiral>true</IsChiral>
<ApproximationType>ZOLOTAREV</ApproximationType>
<Mass>0.04</Mass>
<RatPolyDeg>17</RatPolyDeg>
<InnerSolve>
<MaxCG>1000</MaxCG>
<RsdCG>1.0e-7</RsdCG>
<ReorthFreq>20</ReorthFreq>
</InnerSolve>
<AuxFermAct>
<FermAct>UNPRECONDITIONED_WILSON</FermAct>
<Mass>-1.4</Mass>
</AuxFermAct>
</FermionAction>
Other new FermActs:
"UNPRECONDITIONED_OVERLAP_CONTINUED_FRACTION_5D"
and soon "OVERLAP_CONTINUED_FRACTION_5D" (which will be preconditioned)
ApproximationType can be "ZOLOTAREV" or "TANH" for now.
Robert feel free to add your own here if you want.
io/enum_io/enum_coeffs_io.[h,cc]
io/enum_io/enum_fermacttype_io.[h,cc]
I have removed therefore the old XML input files from the repository
and added in the new ones.
-- Approximation type can now also be TANH with the following caveats:
i) rescaling: I am still not clear on this. I may not need
to do this at all or I may need to do it the same way
as done in the ZOLO case. I'll ask Tony.
ii) projection: I saw in your paper that in the Borici
(TANH) case, the EigValFunc is not signum but in fact
the polar approximation: \epsilon_{Ls/2}
-- Also note from the above XML, that as agreed, I pushed the IsChiral
to be a user supplied parameter. So that if we have a bad approximation
for Zolotarev we can set it to false. Likewise if we have a supremely
high order tanh approximation with projection so that we are effectively
chiral then you can set it to true.
-- fermact_paramio.h fermact_paramio.cc zolotarev4d_fermact_paramio.h
zolotarev4d_fermact_paramio.cc are GONE (removed from CVS)
zolotarev5d_fermact_paramio.[h,cc] are still around but
will be gone real soon now.
08/30/2004 -- Robert Edwards
Version 1.7.4. Really, these are the last changes before 1.8
08/30/2004 -- Robert Edwards
Version 1.7.3. Last changes before 1.8
08/20/2004 -- Robert Edwards
Version 1.7.2. Changed wall-FF codes to allow wall-source.
08/20/2004 -- Robert Edwards
Version 1.7.1. Some more hmc support.
08/01/2004 -- Robert Edwards
Version 1.7.0. Added SSE_DWF_CG support.
07/27/2004 -- Robert Edwards
Version 1.6.2. Changed START/END_CODE() to not have first arg.
07/27/2004 -- Robert Edwards
Version 1.6.1. Changed sink_smear2_w.{h,cc} to fermion indep.
sink_smear2.{h,cc}. Other small lib/Makefile.am fixes.
07/26/2004 -- Robert Edwards
Version 1.6.0. New HMC support. New profiling support.
Changed mem. layout to remove PDWVector layer - only matters
for SSE routines. Removed completely all uses, decls, files
with LatticeDWFermion.
07/13/2004 -- Robert Edwards
Version 1.5.20. Final version before minor release of QDP breaks old
usage of LatticeDWFermion.
06/08/2004 -- Robert Edwards
Version 1.5.19. Modifed Wallformfac mainprog to write binary.
06/08/2004 -- Robert Edwards
Version 1.5.18. Wallformfac codes debugged to some
extent. Added new delta operators in baryons_w.cc .
Time for major tests.
05/28/2004 -- Robert Edwards
Version 1.5.17. Changed Coulomb gfixing to use gauge transf matrices.
Put relaxed solvers in for overlap and into things like quarkProp4
05/27/2004 -- Robert Edwards
Version 1.5.16. Fixed SciDAC gauge IO to always read/write in
single prec. Fixed initialization in multi-mass quark prop.
05/25/2004 -- Robert Edwards
Version 1.5.15. Fixed KYU qprop readers to correct basis.
Completed insertion of GMRES and relaxed codes. Also inner
"preconditioned" versions.
05/19/2004 -- Robert Edwards
Version 1.5.14. Relaxed solvers for overlap appear to work. Now have
BiCG codes, SUMR, CG, etc. and multi-mass variants. Fixed KYU qprop
reader.
05/13/2004 -- Robert Edwards
Version 1.5.13. QQQ program now can output in Dirac or Degrand-Rossi
basis. Made basis rotation more public.
05/13/2004 -- Balint Joo
Version 1.5.12. Working multi-mass SUMR code.
05/12/2004 -- Robert Edwards
Version 1.5.11. Tidying in the sequential source/prop reader/writers.
Working SUMR codes. Overlap under development to use new tricks.
05/04/2004 -- Balint Joo
Version 1.5.10. Overlap double pass appears okay.
05/02/2004 -- Robert Edwards
Version 1.5.9. Qproptransf codes in place. Now have completed
conversion to new seqsource/propagator qprop format.
04/29/2004 -- Robert Edwards
Version 1.5.8. Wall-formfactor code appears to work and agree
with bar3ptfn code with a wall sink.
04/29/2004 -- Balint Joo
Version 1.5.7. Changed auto-isms to allow correct installation. Fixed
chroma-config to be useful. Now have a chroma_config.h and internal.h.
04/27/2004 -- Robert Edwards
Version 1.5.6. Last changes (mostly overlap) before splitup of main codes.
04/23/2004 -- Robert Edwards
Version 1.5.5. More support for overlap.
04/22/2004 -- Robert Edwards
Version 1.5.4. BUG FIX in baryon_w.cc . The proton_{4,5,6} were converted
incorrectly from szin.
Cleanups in gauge initialization. Overlap support now in propagator codes.
04/21/2004 -- Robert Edwards
Version 1.5.3. New version of propagator input. New wall-sink
formfactor codes. New multi-prop. support.
04/05/2004 -- Robert Edwards
Version 1.5.2. Start of support for SZINQIO.
04/01/2004 -- Robert Edwards
Version 1.5.1. Added non-relativistic support to seqprop.
02/22/2004 -- Robert Edwards
Version 1.5.0. Major change in propagator model in main programs. ONLY
support SciDAC format of propagators. Now, the "Param" group on input is the
header of some propagator. There are XML and binary parts of propagators
(SciDAC format). The XML part is used in subsequent calculations, thus removing
potentially erroneous use of input params in calcs.
All test files changed to match
02/22/2004 -- Robert Edwards
Version 1.4.4. Some new programs added. Fixed make_source/propagator/seqprop/spectrum
chain of tests to be consistent and give reasonable output. Cross-checked against
new (un-committed) XML driven input/output model.
01/27/2004 -- Robert Edwards
Version 1.4.3. Overhauled SSE Wilson dslash support. New index tables.
01/26/2004 -- Robert Edwards
Version 1.4.2. More eigenvalue stuff. Tagging pre-release of the
new SSE Wilson dslash support.
01/07/2004 -- Robert Edwards
Version 1.4.1. Functioning Zolotarev code by Balint Joo.
Also, added WILSON and STAGGERED build control.
01/01/2004 -- Robert Edwards
Version 1.4.0. Removed all proxies. Switched to only use Handles
(proxies were just glorified handles). Changed fermact factory
functions to require a Handle<const ConnectState> instead of a
pointer to a connect state. Added Fermion and Gauge BC support.
Now a new virtual getFermBC in FermAct. Now, the semantics is
the FermAct is responsible for applying the Fermionic BC. All
the fermacts changed to pass a FermBC which is then applied
to the U fields in createState. This is supplied as a default
in createState.
First tentative code for pure gauge heatbath and dyn. ferm. HMC.
12/31/2003 -- Robert Edwards
Version 1.3.2. More tweaks in fermacts. Added first versions
of heatbath and HMC (not working yet).
This is a pre-release before major changes to come. The next
version changes substantially the fermacts - now gauge and
fermion BC are included in the proper way.
12/06/2003 -- Robert Edwards
Version 1.3.1. Added various gauge fixing routines.
12/02/2003 -- Robert Edwards
Version 1.3.0. Another overhaul of FermionAction support. Now
create and pass around gauge connection state.
12/01/2003 -- Robert Edwards
Version 1.2.1. Minor tweaks. Pre-release for 1.3.0 .
11/23/2003 -- Robert Edwards
Version 1.2.0. Overhauled FermionAction and LinearOperator. Added
new even-odd preconditioned support.
11/20/2003 -- Robert Edwards
Version 1.1.4. Switched LinearOperator and FermionAction interfaces to
pass what was return type as instead argument to operator() & apply (linop)
and dsdu (fermact).
11/20/2003 -- Robert Edwards
Version 1.1.3. Small tweaks.
11/16/2003 -- Robert Edwards
Version 1.1.2. Added extended-overlap action and linop.
11/14/2003 -- Robert Edwards
Version 1.1.1. Rearranged fermact files and made more base classes.
Merged separate dsdu and qprop files into corresponding fermact files.
11/14/2003 -- Robert Edwards
Version 1.1.0. Added array versions of fermacts. Now have functioning
unprecondition DWF operator.
10/09/2003 -- Robert Edwards
Version 1.0.1. Changed all cout/cerr/QDP_info to use QDPIO::cout/cerr.
10/07/2003 -- Robert Edwards
Version 1.0.0. First major release. Supports sse extensions. Several
bits in production. More to come.