forked from etmc/tmLQCD
-
Notifications
You must be signed in to change notification settings - Fork 2
/
NEWS
403 lines (315 loc) · 16.3 KB
/
NEWS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
08.09.2011: added code LapH_ev for the computation of the eigensystem
for the Laplacian Heaviside method (0905.2160 [hep-lat]).
In order to use it configure with --disable-halfspinor --enable-laph (serial),
and also with --enable-indexindepgeom --enable-tsplitpar --with-mpidimension=XYZ
--with-lemondir=${lemondir} (parallel).
Compile with make LapH_ev.
23.08.2010: indexindependentgeom (see below) extended to naive
communications (implemented with MPI_Sendrecv) and halfspinor.
16.04.2010: mixed precision CG implemented for invert and HMC
determination of the Ptilde degree optimised
new input parameter for NDPOLY monomial: MaxPtildeDegree
16.04.2010: as online measurements are available CORRELATORS, PIONNORM, POLYAKOV
syntax
BeginMeasurement TYPE
option = value
EndMeasurement
March.2010: parallel IO implemented and tested to work.
new operator syntax in input file
Begin Operator TYPE
option=value
EndOperator
10.02.2010: New hopping_test.c routing to test the parallel version of
an Hopping_Matrix
10.02.2010: when compiling with --enable-tsplitpar, the communications in the
Hopping Matrix are splitted and performed after each timeslice.
This allows the overlap of communications and computations.
tsplitpar needs indexindepgeom, SSE2/3, gaugecopy
10.02.2010: when compiling with --enable-indexindepgeom, the xchange routines do not
rely of absolute addressing anymore, but on addresses gI_?_?_?_? which are
defined based on the function Index.
This enables the parallelization in X, XY, XYZ directions
which is obtained --with-mpidimension=XYZ ...
indexindepgeom is still not compatible with blocking-mpi of halfspinors
13.11.2009: parallel IO for gauge fields tested and working.
parallel reading of Checksum still buggy, but not essential
09.11.2009: MMS successfully checked. Little code for checking
in contractions/check_propagators/check_mms_props.cc
04.06.2009: bug in eigensolver routine for ND case fixed
parallel IO included using lemon library
Schroedinger functional boundary conditions for
gauge case included
14.05.2009: multiple mass solver (CG) for twisted mass
implemented, input is solverflag = cgmms
the number of additional masses has to be specified with
CGMMSNoExtraMasses = N
and the extra masses have to be given line by line in a file called
"extra_masses.input"
The CGMMS solver writes for every mass (extra + the original one) one
propagator storing the result of (Q^\dagger Q)^(-1) \gamma_5 \phi
where \phi is the source.
Hence the result for +\mu and -\mu can be extracted at any later
stage, respectively.
14.05.2009: in the propagator files we now store additional info, namely
the xlf-info message is copied from the original gauge field, if
existing, as well as the scidac-checksum and the ildg-data-lfn
messages. This should ease the mapping from a propagator to the corresponding
gauge configuration file.
01.08.2008: online measurements implemented for
PP and PA correlators. Input variables are
PerformOnlineMeasurements = yes|no and
OnlineMeasuremntsFreq = n
Result will be written to a file called
onlinemeas.trajno
tested to work scalar and parallel
30.05.2008: support now for three formats for source and propagator, for compatibility:
cmi, GWC, ETMC. The latter is the standard and recommended.
07.12.2007: inversion for the flavour split doublet implemented and tested
new spinor field IO format implemented and tested
scidac checksums implemented
19.10.2007: hmc runs now also without even/odd preconditioning.
Set UseEvenOdd=no in the input file (thanks to Jan Volkholz).
05.09.2007: invert suppports now also inversions using D_psi. This is
triggered by UseEvenOdd (default = yes) as input parameter.
04.09.2007: D_psi in three versions tested, also parallel
versions. BG/L version of deriv_Sb still not checked.
31.08.2007: D_psi for full spinor field (no even/odd) implemented and tested
for SSE2 and normal version. new deriv_Sb implemented and tested.
BG/L versions still need testing.
08.08.2007: all integration schemes available for the PHMC
new input paramter
TimeScaleHeavyDoublet
which must be set to an integeter >=0 specifying the timescale
on which to integrate the heavy doublet on. Note that 0 is
the smallest possible timescale, which is the smallest
timescale available for pseudo fermion fields. (the gauge
field is one scale below)
not completely tested yet.
15.04.2007: eigenvalue computation for squared one flavour operator
implemented, few new input parameters connected to this,
see documentation.
Preconditioned CG added and tested with eigenvalues
estimates. However, CG with exact eigenvector subspace
projected out is faster.
Sloppy precision now also available for CG.
03.04.2007: Addition of 2 input parameters:
SplittedPropagator and SourceLocation
New function in start.c to use the SourceLocation
when its value is different from 0.
The same modification was done in the GWC code.
01.04.2007: many changes to PHMC
new input parameters:
PhmcNoFlavours (2+1+1 or 1+1)
PhmcComputeOnlyEVs (yes or no: compute EVs only and exit then)
PhmcStildeMax (flaot, upper bound for appr. interval)
PhmcStildeMin (float, lower bound for appr. interval)
PhmcDegreeOfP (int, degree of P, the less precise polynomial)
PhmcRecEVInterval (int, recompute EV's every n trajectories)
parallel Eigenvalue computation works now, also on BG/L
PHMC fuer 1+1 flavours exists and works
18.01.2007: towards version 4.0
merged with branch phmc. hmc_tm and phmc_tm
are both compiling and running. phmc needs
lapack.
phmc has so far only on integration scheme
available. The PHMC code is tested against a
code of I. Montway.
Credits to T. Chiarappa for the PHMC, see
doc directory for more details.
Tested so far only on PC's
15.01.2007: stout smearing for invert implemented by Craig McNeile
input parameters are: UseStoutSmearing, StoutRho and
StoutNoIterations (hopefully self explaining)
Tested against Chroma-3.17.0
stouting for hmc not yet implemented
12.12.2006: run time call for dram window on BG/L implemented (--with-bgldram)
persistent MPI for halfspinor available (--with-persistentmpi)
non-blocking MPI calls also for gauge fields implemented
now generally available with --with-nonblockingmpi for all platforms
The BG/L performance is now close to 18% peak.
Improved performance for opteron CPU's
02.09.2006: New Dirac operator implemented and working
On BG/L this brings a 20% improvement
it can be switched on with --enable-newdiraop
The exchange routines are now also with shmem API
available. Usable with --enable-shmem. this
feature is not yet completely tested.
15.04.2006: Various new input parameters, see doc/input.tex.
removing of hmc.reread now also works on BG/L.
The file conf.save is now almost always save: a new
configuration is first stored in .conf.temp and then moved
to conf.save. There might be still a problem in case the
job chrashes when .nstore_counter is written...
28.03.2006: chronological solver guess now independent of lapack
the current filename is now saved in .nstore_counter
last_configuration and last_state are therefore obsolet.
random number state now saved in lime format in the gauge file
rlxd_state files are not any longer produced, but will still be
read.
14.02.2006: --enable-gaugcopy will work now also without
--enable-eogeom . But it requires two additional
copies of the gauge fields instead of one.
A reasonably well tuned BG/L Dirac operator available
about 10% of peak scaling up to 2048 processors.
13.02.2006: four dimensional parallelisation more or less tested
12.02.2006: But in all but four dimensional parallelisation
fixed.
10.02.2006: package will compile only with lime >= 1.2.3!
10.02.2006: Build in separate directory possible now.
09.02.2006: Bug in io.c fixed
08.02.2006: Bug in write ILDG configs fixed
07.02.2006: In case of 4-dim. parallelisation the product
T*LX*LY _must_ be even, otherwise the field exchange
in z-direction does not work.
07.02.2006: 3 and 4 dimensional parallelisation implemented
working both so far _only_ with --disable-eogeom
(maybe configure then also --disable-gaugecopy)
New input parameter NrZProcs
local LZ _must_ be even
everything is not yet finally tested.
06.02.2006: Added the possibility for fixed volume at compiletime
--with-fixedvolume
please edit fixed_volume.h accordingly
not at all tested tested!
30.01.2006: New input parameter NrYProcs
29.01.2006: running and partially tested on the BGL in Juelich
configure with
./configure --host=powerpc64-bgl-linux-gnu --without-lapack CC=/opt/ibmcmp/vac/7.0/bin/blrts_xlc
and other options. Also lime needs to be configured
with ./configure CC=/opt/ibmcmp/vac/7.0/bin/blrts_xlc
please use lime-1.2.3 _at least_! Earlier version might not
work.
30.01.2006: at least for BGL lime needs to be configured with
./configure CC=/opt/ibmcmp/vac/7.0/bin/blrts_xlc --enable-largefile
and (remember) lime-1.2.3.
25.01.2006: Trying to setup a versioning system:
last digit -> bug fixes
second digit odd -> developement Version
second digit even -> stable release
25.01.2006: changed --disalbe-lapack to --without-lapack ...
05.01.2006: By setting the history parameters for the CSG to zero one gets
now a zero spinor as trial guess for the solvers.
Moreover, by specifying --disable-lapack to configure it is
possible to compile without the need of the external libs lapack
and blas and the fortran-lib.
14.11.2005: For the flavour non-degenerate eigenvalues computation, a new
structure, called bispinor, has been introduced. Consequently,
a serial and a parallel new Jacoby-Davidson routine working
with bispinors has been implemented, as well as two new solvers,
bicgstab_complex and cg_her, respectively. A bunch of linear
algebra files have been adapted to work with bispinors. All these
files are distinguished by the suffix "_bi" (e.g: "file_bi.c").
14.11.2005: New global parameters (g_mubar, g_epsbar) have been introduced.
These are the mass parameters needed in the eigenvalues
computation of the flavour non-degenerate case Dirac operator.
11.11.2005: gcc-4.x does not use libg2c anymore. One has to link
against gfortran, which is done now.
26.10.2005: Added gauge file format conversion programs for
gwc -> ildg and ildg -> gwc
All solver have additional parameter now. It is now possible
to invert with realtive precision. Moreover, the propagator
(or source) format of Chris Michael can be read in.
The new input parameter is:
SourceFormat = cmi (otherwise gwc assumed)
other related input parameters are
ReadSource = yes
SolverPrecision = 1.e-10
SourceInputFilename = random_test
UseRelativePrecision = yes
11.08.2005: 2MN integrator implemented and tested. Two versions
available: velocity and position version (hep-lat/0505020).
Integrator=2MN or 2MNposition
29.06.2005: ILDG LIME file format introduced.
old file format deprecated. But it will be still
automatically detected and read in.
trajectory counter introduced which will now allow
to correctly keep the Nskip's between the confs.
this is as well as the plaquette value stored as
xlf-info record in the new LIME format.
10.03.2005: Precisions in the solver for force and Acceptance are input
parameter now for each mu parameter:
ForcePrecisionMu, ForcePrecisionMu2, ForcePrecisionMu3
AcceptancePrecisionMu,
AcceptancePrecisionMu2, AcceptancePrecisionMu3
ExtIntStepsMu0 is now called IntegrationStepsMu,
ExtIntStepsMu1 is now called IntegrationStepsMu2,
ExtIntStepsMu2 is now called IntegrationStepsMu3,
matchin the input names for the mu parameter.
The old one are still usable.
Added an input parameter DebugLevel to control the
debug output. Setting it to one will cause the program
to compute and print out the norms of the forces.
17.02.2005: Reversibility check implemented. Input parameters are
ReversibilityCheck = yes|no (default no)
ReversibilityCheckIntervall = 100 (default 100)
Precisions in the solver for force and Acceptance are input
parameter now:
ForcePrecision = float (default 1.e-7)
AcceptancePrecision = float (default 16.e-7)
One can choose to have relative precision:
UseRelativePrecision = yes|no (default no)
13.02.2005: Integration scheme with error cancellation implemented.
It is only implemented for the highest level and should
have errors in \delta\tau^5 only.
07.02.2005: LX,LY,LZ now possible as input parameter.
07.01.2005: Extended leap-frog and extended Sexton-Weingarten
integration schemes (multiple time scales) implemented
and tested.
17.12.2004: Possibility for rereading some parameters added.
If there is a file hmc.reread, it will be parsed
automatically and deleted afterwards. It is not possible
to change T, L, RGIC1 from zero to a non zero value, NrXProcs.
13.12.2004: Extended leapfrog integration scheme implemented
and tested.
23.11.2004: The lattice size must be now set in the input file.
Recompilation is only needed for one or two dimensional
parallelisation.
For invert there is a new input parameter:
ReadSource = yes|no
SourceInputFilename = filename
This let's you read in a generalised source for the
inversion. The real filename must be of the form
filename${massnumber}.is${is}ic${ic}.${nstore}
22.11.2004: DBW2 implemented for the serial code and the parallel
code with new and old geometry.
Input parameter BoundaryCond is supplemented by
BCAngleT, like in the GWC code. BoundaryCond is
deprecated now.
28.09.2004: input parameter added:
input parameter MaxSolverIterations
and SolverPrecision available.
history_hmc_tm file added with history of
written configurations and corresponding
Plaquette values and timestamp
20.08.2004: SSE3 Version of the most important macros added and
tested. SSE3 versus SSE2: 1.84 Gflops versus 1.64 Gflops.
(P4 3.20GHz prescott)
17.08.2004: 64 Bit Version of the code running and tested
Cache Optimisation for Opteron added.
New configure options:
--enable-opteron : Enables cache optimisation
for Opteron [default=no]
--enable-gaugecopy : Enables usage of a copy of
the gauge field [default=yes]
--enable-eogeom : Enables usage of EO geometry
also for the gauge fields [default=yes]
13.08.2004 New EO geometry for gauge fields implemented also
for the 2-dim parallelisation and tested.
04.05.2004: additional parallelisation in x-direction added
and tested. IO added as well and tested, apart
from write and read for spinor fields.
extended test functions written (hmc/test)
05.04.2004: Bug fix. serial version without MPI running now.
Bug was in geometry_eo.c
program for thermal cycles added.
11.03.2004: third pseudo fermion field added and tested.
09.03.2004: .nstore_counter file introduced to easily restart the
programm. Just set InitialStoreCounter = readin. Also a
sighandler was added to savely finish the program.
04.03.2004: Hasenbusch trick tested and working.
New version of the Hopping matrix for the IBM implemented
with special improvements. Even Odd ordering translated
also to the gauge fields and tested, but not yet default.
03.03.2004: second pseudo fermion (trick of Martin Hasenbusch)
implemented also for tmQCD
02.03.2004: Release 1.0.1