Welcome, Guest |
You have to register before you can post on our site.
|
Latest Threads |
Concern about RBE3
Forum: MYSTRAN
Last Post: Admin
01-25-2024, 11:12 PM
» Replies: 2
» Views: 972
|
Abaqus/CCX to MYSTRAN/Nas...
Forum: MYSTRAN
Last Post: Admin
01-25-2024, 01:52 PM
» Replies: 0
» Views: 588
|
15.1.0 Update
Forum: MYSTRAN
Last Post: borges
12-19-2023, 02:50 PM
» Replies: 0
» Views: 708
|
15.0 Released - Major Upd...
Forum: MYSTRAN
Last Post: borges
12-19-2023, 02:46 PM
» Replies: 1
» Views: 943
|
New User facing issues
Forum: MYSTRAN
Last Post: Admin
12-12-2023, 06:11 AM
» Replies: 1
» Views: 923
|
Femap model hard crash
Forum: MYSTRAN
Last Post: borges
12-02-2023, 08:41 PM
» Replies: 0
» Views: 733
|
Converting MSC Nastran In...
Forum: ComLab and NASTRAN-95
Last Post: ceanwang
03-28-2023, 09:47 PM
» Replies: 1
» Views: 2,710
|
ComLab (an improved NASTR...
Forum: ComLab and NASTRAN-95
Last Post: Admin
12-13-2022, 04:55 AM
» Replies: 2
» Views: 6,772
|
Windows Complie Issue
Forum: MYSTRAN
Last Post: realbabilu
11-18-2022, 10:23 AM
» Replies: 4
» Views: 5,271
|
Mystran and Intel One API...
Forum: MYSTRAN
Last Post: Admin
11-17-2022, 02:48 AM
» Replies: 7
» Views: 9,619
|
|
|
Converter |
Posted by: Admin - 05-24-2020, 10:10 PM - Forum: Pre/Post Processors
- Replies (4)
|
|
One idea is to create a CaculiX to MYSTRAN converter. This may be useful since some pre-processors support CalculiX. While it is better if the if the preprocessor adds native support for MYSTRAn, this might be a more practical solution and is more versatile (automatically support multiple platforms). pyNastran can be used a post once Steve adds OP2 output to MYSTRAN.
|
|
|
PaStiX Sparse Solver |
Posted by: Admin - 05-19-2020, 04:41 AM - Forum: MYSTRAN
- No Replies
|
|
A few months ago, Guido Dhondt (developer of CalculiX), mentioned that he was looking into the PaStiX sparse solver. This solver seems to be a great option since it is modern, still being developed, and is open source. I recently contacted him about it and here are his comments (he said it is OK to post his comments here). I also contacted the authors of PaStiX some time ago and they said they are willing to help us incorporate it.
---
I had an IT-student to work on PaStiX for his master thesis. He finished two months ago. He took the standard version of PaStiX, however, he also did some improvements in PaStiX specifically for CalculiX. We had a test suite of 7 examples, between 700,000 and 5,000,000 equations. The results:
- the solver alone was a factor 4 to 13 faster than Pardiso
- the complete CalculiX-calculation was a factor 3 to 8 faster than with Pardiso.
PaStiX will be available in the next CalculiX-release. We are still working on some special cases. Since we changed PaStiX and it is not sure that these changes will be included by the originators of PaStiX we may have to include the modified version on our homepage.
There is one caveat: these accelarations only become true if you have a really good graphical card, let's say a card costing about 8,000 Euro. I have a GeForce GTX 1060 6GB Graphical card in my private computer, with this GPU CalculiX is slower with PaStiX (if you use the GPU).
---
I think high on the list of things to do is to add the PaStix Solver to MYSTRAN.
|
|
|
Compile 10.31 with Code::Blocks |
Posted by: Admin - 05-17-2020, 12:06 AM - Forum: MYSTRAN
- No Replies
|
|
EDIT: See this thread for the most current: https://www.mystran.com/forums/showthread.php?tid=2
I will create detailed instructions later, but for now, this is one option that seems to be very close.
1. You can use Code::Blocks in the same way as Visual Studio. I also tried CBFortran, but it seems to be 32-bit and was much slower when reading in the source files.
http://www.codeblocks.org/downloads/26
Download "codeblocks-20.03mingw-setup.exe". This will add the Fortran compilers in the CodeBlocks install directory.
2. Download this modified 10.31 source code, which has some minor modifications from 10.31.
https://www.mystran.com/misc/Source.zip
For reference, here are the modifications that were made (you don't need to do this if you download from the above link)
- Delete the BANDIT directory
- In GET_COMMAND_LINE.f90, comment out INTRINSIC GETCL in line 46
- In GET_COMMAND_LINE.f90, change "CALL GETCL ( FILNAM )" to "CALL GET_COMMAND_ARGUMENT ( 1, FILNAM )"
- In MYSTRAN.f90, line 128, change OPEN(SC1,CARRIAGECONTROL='FORTRAN') to OPEN(SC1)
- In MYSTRAN.f90, for the line that says : "PERCENT_CHANGE(LOAD_ISTEP,NL_ITER_NUM) = 1.0D2*(UG_NORM(LOAD_ISTEP,NL_ITER_NUM) - UG_NORM(LOAD_ISTEP,NL_ITER_NUM-1)) &" delete a space before the "&" character.
3.
- In Code::Blocks, start a new project. You don't need to select a compiler now.
- Under settings/compiler/Toolchain Executables, click the GNU Fortran Compiler and point to this directory, C:\Program Files\CodeBlocks\MinGW. Then, replace the first 3 EXE files with with gfortran.exe (or the 64 bit MinGW)
- Right click the project name in the tree and add all 10.31 directory recursively. You may also want to put the in a directory that has a short and simple path. In my case this took about 10 min so you may need to be patient.
- In Project/Build Options/Search Directories, select the INCLUDE directory location under both "Compiler" and "Linker" tabs. Also, check both the Debug and Release tabs. Otherwise, it will not find debug.h and stat.h
- Build/Rebuild
ISSUE:
At this point, Intel would normally compile, but I am not having success. At the end is says "Execution of 'X.. failed", where X is the compiler (Execution of 'mingw32-gfortran.exe...or Exectution of 'gfortran.exe..., etc). A google search has not helped to resolve it.
Upon closing CodeBlocks and then reopening it, I get a different error related to Pentium-II. That seems to be resolved by just clicking Settings/Compiler/Toolchain Executables and selecting the GNU Fortran compiler. But this is odd behavior!
For reference, here is the Makefile associated with this attempt. I changed the path to C:\A instead of C:\Source though in an attempt to shorten fix it (though that did not work).
https://www.mystran.com/misc/Makefile
And here is the build log for reference.
https://www.mystran.com/misc/M5_build_log.html
|
|
|
Compile 10.3 With Intel and gFortran Assistance |
Posted by: Admin - 05-13-2020, 11:12 PM - Forum: MYSTRAN
- Replies (1)
|
|
EDIT: See this thread for the most current: https://www.mystran.com/forums/showthread.php?tid=2
Using Visual Studio/Intel Fortran, here are the changes that can be made to the 10.31 on GitHub (they are minor). This may be useful if you are trying to compile with gFortran since similar changes may need to be made. This is a smaller list than in 10.2 and removes the MKL dependencies. It also removes BANDIT, which was causing a problem and is not necessary. We can restore BANDIT later.
1. OPTIONAL - Commented out reference to CDIR$ NEXTSCALAR on lines 4477 and 4479 of LAPACK_BLAS_AUX.f. This does not cause an error in Lahey or Intel so it seems optional.
2. Commented out INTRINSIC GETCL in line 46 of GET_COMMAND_LINE.f90
3. Commented out INTRINSIC GETENV in line 44 of GET_MYSTRAN_DIR.f90
4a (OPTION 1). In GET_COMMAND_LINE.f90, change "CALL GETCL ( FILNAM )" to "CALL GET_COMMAND_ARGUMENT ( 1, FILNAM )"
-------------------------------FOR REFERENCE (OLD)-------------------------------------------------------------
4b (OPTION 2). Remove GETCL.f90 (if it exists) and replace it with K2_GETCL.f90 (linked below)
https://www.mystran.com/docs/K2_GETCL.f90
|
|
|
More Notes from Harry Schaeffer |
Posted by: Admin - 05-13-2020, 06:49 PM - Forum: ComLab and NASTRAN-95
- Replies (1)
|
|
Notes from Harry Schaeffer:
1. The IFP should be thrown away. I suggest marrying pynastan code and sqlite. Steve has done a lot of work in writing code for parsing and validating bulk data. These data can then be written to sglite tables.
2. Write interface routines to replace the geom1, 2, etc files. The code that implements the bulk data items still expects to find the internal bulk data codes.
3. The MPL should be replaced so that it is easier add new modules.
4.The checkpoint/restart must be replaced with a database restart.
5. Several helper fortran modules such as a scratch file manager would be useful.
6. Rethink the startup code.
7. Remove and replace the eigen and solve routines.
|
|
|
Bandit Remove/Add |
Posted by: Admin - 05-13-2020, 05:08 PM - Forum: MYSTRAN
- No Replies
|
|
EDIT: BANDIT has been restored in Version 11.0
BANDIT is desired (although not mandatory) if we continue to have a "banded" solver. Although the LAPACK routines, which are optimized for banded matrices, will work on matrices that are not banded, it could result in excessive solution times. Once we get a sparse solver the only use of the banded solver will be as a backup in case the sparse solver fails.
Starting with 10.3, we decided to depreciate BANDIT because it was causing compiling errors. It can be added back later once those errors are resolved.
To add BANDIT back in, download the following ZIP file:
https://www.mystran.com/docs/BANDIT.zip
|
|
|
Mecway Pre/Post |
Posted by: Admin - 05-13-2020, 02:06 AM - Forum: Pre/Post Processors
- No Replies
|
|
Mecway is a fork of LISA (https://mecway.com/).
There is an internal solver and it also has support for CaclculiX. This program is still in development and it might be possible to add MYSTRAN support.
While this is not a free program, it is low cost and there is a free 1,000 node version. I like this program a lot, especially as you become more familiar with it. If you would like to see MYSTRAN support, you may want to mention this to the author.
|
|
|
|