8d2d645a26
MPI: insert MPI engine into main engines inheritance scheme
...
new inheritance scheme:
[..] -> SSE_Compressed -> MPI -> Multithreading -> [..]
2011-03-10 10:39:07 +01:00
b395546483
MPI: some fixes in handling the MPI engine
2011-03-10 10:39:07 +01:00
3b402e976b
MPI: disable/rename processings if they are in conflict
2011-03-10 10:39:07 +01:00
27a4b9f482
MPI: synchronize time step + new MPI_Enabled flag
2011-03-10 10:39:07 +01:00
e2628740fc
new openEMS_FDTD_MPI for better MPI handling
2011-03-10 10:39:07 +01:00
9364ecd292
first working MPI implementation
...
todo:
- only/mainly rank 0 takes care of output
- rank 0 manages end-criteria, energy calculation
- field dumps split between ranks --> prevent file corruption
- a lot more...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2011-03-10 10:39:07 +01:00
895e6c635d
initial MPI support
2011-03-10 10:39:07 +01:00
b73004be48
new method to dump excitation signals
2011-03-10 10:34:40 +01:00
7df39bafcb
fix: missing operator construction
2011-02-08 11:42:59 +01:00
02f929aa53
fix: allow missing exciation...
2011-02-08 10:35:23 +01:00
2e2f75807e
Operator-Base: access stored discrete material
2011-01-31 12:22:21 +01:00
24bac9499f
allocate dynamic CSX and allow access from basic operator
2011-01-31 12:00:00 +01:00
10537d75ab
fill missing material storage in cylindrical multigrids
2011-01-25 10:27:51 +01:00
67d1307fdf
fixed the operator resetting & cleanup + a lot of memory leaks fixed
2011-01-24 11:11:45 +01:00
c0c66518c1
new dump type: total current density (rotH) field dump
2011-01-10 11:15:22 +01:00
c5a689b67c
fixed GetRawField handling
2011-01-10 11:14:23 +01:00
24a2e3007a
FDTD Engine Interface: update to GetRawField
2011-01-07 16:29:08 +01:00
2c3ebe5a7d
new dump type: electric current density (J = kappa * E)
2011-01-07 16:12:07 +01:00
ea496b6129
operator: new material data storage for post-processing purposes
2011-01-07 10:55:31 +01:00
3104335dce
replaced GetMeshDelta by GetEdgeLength & partially removed AdrOp
2010-12-08 16:55:27 +01:00
043ec7c1a1
processing constructor with Engine_Interface_Base only
...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-07 16:47:23 +01:00
57ead21ccc
real abstract base classes for common operator and engine-interface
2010-12-07 14:47:22 +01:00
9d3761b1d0
Init style: applied astyle.sh
...
astyle --style=allman --indent=tab --indent-cases {} \;
please apply this style prior to any future commit
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 15:31:11 +01:00
70ae86d63f
moved all FDTD extensions to folder FDTD/extensions
...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 15:30:47 +01:00
2f531b99c6
moved all processing classes to Common/
...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 10:44:25 +01:00
16263e8f9a
new Operator_Base class
...
- This class is meant as an abstract base for any time-domain and rect-grid numerical solver.
- All processing methods should only use this base class
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 10:00:08 +01:00
32cbdc5d0b
removed FDTD engine from (nearly) all processing classes
...
todo:
- remove FTDT operator from all processingX and replace by abstract-base-operator?
- remove FDTD engine from ProcessCurrent
- remove or cleanup ProcessEField + ProcessHField
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 10:00:08 +01:00
ab1119f468
new: Engine_Interface
...
This is a new common class designed as an interface between an engine
and the processing routines which should become a part of common as well.
todo:
- migrate all processings to use this interface only
- lots of testing...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-12-06 10:00:08 +01:00
2d12db5ea6
Merge remote branch 'seb/master'
2010-11-04 12:20:23 +01:00
ce56184681
multi-threading support for MUR-ABC added
2010-11-04 12:13:43 +01:00
5900e2726c
more logical boundary handling
2010-11-04 07:48:15 +01:00
6728cee054
minor bugfix: missing initialization
2010-10-28 21:50:50 +02:00
907ac3cff2
default debug flag and update/fix for cylindrical coords
2010-10-27 14:49:16 +02:00
d31b3ffbb2
updated debug dumps
...
The debug dump routines do not work correctly in cylinder coords (--nativeFieldDumps may help)!
2010-10-27 11:17:58 +02:00
4f8aa3faa8
fix in Operator_SSE_Compressed: cleanup old compression prior to an InitOperator()
2010-10-26 18:47:21 +02:00
f204bc2723
Mode matching probe: fixed an error in HField interpolation
...
Now also excluding the boundaries from the MM.
Updated waveguide example: Rect_Waveguide.m
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-10-26 08:30:35 +02:00
1586c76af6
Merge remote branch 'seb/master'
2010-10-20 10:48:18 +02:00
dcc9f86c8a
--debug-PEC improvements
2010-10-20 09:25:50 +02:00
4e91cb95dc
bugfix: --debug-PEC now works with all engines; operator now uses Get... and Set... functions
2010-10-20 07:26:16 +02:00
7dec28731c
changed CSXCAD API handling used primitives
2010-10-19 16:54:35 +02:00
b6362495d8
operator: fix in PEC curves
...
Operator::FindPath method is not optimal yet,
will find weird pathes in some occasions,
at least they will be closed now... hopefully :)
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-10-12 12:49:44 +02:00
16e0e2f3f2
bugfix: correct scaling of --debug-boxes
2010-10-11 21:05:02 +02:00
c1250ce34d
upml ext: save memory and speedup
...
Instead of using an additional voltage/current array mirroring the main volt/curr,
the upml extension now swaps volt with volt_flux and curr with curr_flux
before and after the main engine updates.
This saves memory and therefore also increases the upml speed.
2010-10-07 19:25:22 +02:00
5cc6438ab7
engine_sse_compressed: speed up
2010-10-07 14:39:54 +02:00
672f2a436a
upml extension: operator get functions and disabled pml in non-pml regions
2010-10-06 15:07:17 +02:00
043ef6ec4c
extensions: multithreading support added
...
every extension can either implement the single threaded methods e.g. DoPreVoltageUpdates()
or the multithreading aquivalent method e.g. DoPreVoltageUpdates(int threadID)
Engine_Ext_UPML class is an example for a multithreading extensions
2010-10-06 10:30:55 +02:00
1b2ac008e7
fix in upml operator
2010-10-06 10:04:19 +02:00
430a04e9e4
new/better method to assign jobs to working threads
2010-10-05 17:24:36 +02:00
ae281d3dea
upml extension: use inline engine access methods
...
todo: further speed optimization for this extension...
Signed-off-by: Thorsten Liebig <Thorsten.Liebig@gmx.de>
2010-10-05 11:51:12 +02:00
6c00a844d2
upml extension: added support for cylindrical multigrid meshes
2010-10-05 11:51:12 +02:00
66c79aabe6
mur-abc extension copy-constructor should be protected
2010-10-05 11:51:12 +02:00
684e864a75
upml extension: added support for cylindrical meshs
2010-10-05 11:51:12 +02:00
cc50b5bbef
ShowStat method added to upml extension
2010-10-05 11:51:12 +02:00
06bbec106f
new extension: upml
...
Operator extension implementing an uniaxial perfectly matched layer (upml)
This new pml implementation is going to replace the old split-field pml
2010-10-05 11:51:12 +02:00
887e07a394
define vtk dump data type
...
Using double as vtk data type will fix some errors/warnings in Paraview about a datasize mismatch.
2010-10-05 11:51:12 +02:00
007e8c72f3
doc: comments on extensions priority executing order
2010-10-03 20:15:20 +02:00
5e244bf7cb
fixed crashing "--debug-operator" option with sse-compressed engines
2010-10-03 20:15:20 +02:00
f6663eeccb
multithreading engine: removed separate excitation thread
...
the excitation is now handled by the first (ID=0) "conventional" thread
2010-10-02 18:06:38 +02:00
1c24877bf4
operator extensions, fixed missing deconstructor
2010-10-02 17:55:19 +02:00
fb3ccd36bf
engine: update handling extensions & multithreading priority handling fixed
2010-10-02 17:54:34 +02:00
af922a10bc
new extension priority system
2010-09-25 19:23:53 +02:00
832792eb8e
fixed multigrid interpolation, added closed_cylinder support
2010-09-24 12:01:38 +02:00
8870a20787
fixed extensions handling for cylindrical multigrid
2010-09-22 16:08:39 +02:00
8c95f21858
extensions check for CylindircalMultiGrid & allow Mur-ABC in radial direction
2010-09-22 09:58:45 +02:00
060b93d089
operator: removed redundant methods using integer positions
2010-09-17 15:34:39 +02:00
1e78e5c2d6
fix vtk-dump: vector fields consider mesh type
...
furthermore:
- new argument: --nativeFieldDumps to dump original FDTD field components for debugging purposes
- show global arguments on startup
2010-09-17 15:34:39 +02:00
482de93d3d
fix in operator_sse_compression gcc compatibility check
2010-09-17 15:34:39 +02:00
9dcf42f080
skip filling "et" and "ht" with 0 after excitation is done
2010-09-17 15:34:39 +02:00
b0a0d69faf
bugfix 32-bit SSE-compressed: aligned allocator for STL vector
2010-09-17 15:06:09 +02:00
d42d2f0108
dump_PEC: don't use depreciated MainOp
2010-09-17 12:08:49 +02:00
8cd01c9c0c
compatibility with gcc-4.1.2 (for ABI2 builds)
2010-09-17 11:07:52 +02:00
cc4ffd235f
dump_PEC: get the scaling right
2010-09-17 10:50:06 +02:00
ab701c4a7f
new allow multiple ProcessIntegral and ProcessModeMatch using this to dump mode purity additionally
2010-09-09 10:02:33 +02:00
e3c0805ad9
fix in number of threads calc for multigrid engine
2010-09-08 16:57:10 +02:00
8d657430c0
Merge branch with multigrid
...
Conflicts:
FDTD/operator.h
FDTD/operator_cylinder.cpp
2010-09-08 16:37:40 +02:00
20ade0f053
new: enable cascaded multi-grids... incl. an example
2010-09-08 16:07:28 +02:00
bd4794ecc4
NEW: cylindrical multigrid FDTD
...
this is a new multi grid approach for the cylindrical FDTD.
The FDTD domain will be split in two regions in radial direction.
The "inner" region will have half as many disc-lines in alpha direction and therefore allow for a much larger timestep which increases the simulation speed.
Todo:
- currently only a homogeneous disc is allowed in alpha direction
- some extensions have to be tested and prepared for this approach (e.g. pml)
- speed enhancement and more efficient memory usage
- lots and lots of testing...
2010-09-08 07:36:32 +02:00
a52cd4711a
MT operator: new separate calc start-stop lines method
2010-09-03 12:14:25 +02:00
fc2b60ba3e
Operator Extension clone method: allows to create a copy/clone of an existing extension
...
This will be necessary for the upcoming multi-grid approach...
2010-09-03 11:53:33 +02:00
db0f4ab3e0
MT-engine has access to MT-operator
2010-09-02 22:16:37 +02:00
93fcd44f42
missing gpl header in cylinder extension
2010-09-02 22:14:40 +02:00
12c26f834a
always show the optimal timestep
2010-09-02 22:12:03 +02:00
cabdf4a84a
clear extensions method allows saver reset of MT engine
2010-09-02 22:04:35 +02:00
f7360cbdc1
Dump material as a vector instead of averaged.
2010-09-02 15:35:57 +02:00
df3e7c0c12
major operator revision
...
- now the equivalent circuits are calculated by first calculating the averaged material properties
- this approach should also be save for the cylindrical FDTD
This needs some further testing, especially for the cylindrical operator!!
2010-09-02 15:35:13 +02:00
200832b423
operator: setup excitation from xml
2010-08-30 08:25:10 +02:00
594b38e345
operator: set a forced timestep
2010-08-30 08:25:10 +02:00
e081a9cf94
Engine API: GetVolt/GetCurr methodes spilt up to Get/Set
2010-08-30 08:25:10 +02:00
b75476cc04
apply clear inheritance for cylindrical coords operator
2010-08-30 08:25:10 +02:00
23a3f6fb9c
simplified memory management
2010-08-26 15:54:19 +02:00
cc51c2334d
indention changed
2010-08-26 15:47:04 +02:00
ea7047f7e7
feature: SSE operator compression will now always work at maximum compression
2010-08-26 14:50:23 +02:00
4d67bab7c7
fix in calculation of numThreads --> calculation moved to operator
2010-08-25 15:27:59 +02:00
6ad49b9cae
excitation: warning for unknown types fixed...
2010-08-23 22:52:08 +02:00
6fa9fef29f
fix operator: fixing a critical error in the mue-weighting
2010-08-23 22:15:52 +02:00
42e1968348
cylindrical op: cylinder extension should be always first to execute
2010-08-23 21:53:29 +02:00
35939584df
cleanup: renaming voltage excitation E_* to Volt_*
2010-08-16 13:28:19 +02:00
0319c35f55
count and show excitations per direction
2010-08-16 13:22:40 +02:00
2a9dc040dc
new process: mode matching integral parameter
...
todo: should use a special CSX property in the future
and have its own fundamental mode functions
2010-08-16 11:53:43 +02:00
de054658ef
avoid vector copy operation
2010-08-11 18:48:23 +02:00
Thorsten Liebig
Sebastian Held
Sebastian Held