FLIPS status
From Theory of Measurements Wiki
Current version of FLIPS
Source: FLIPS-1.1
Revision history
Revision 132 [FLIPS-1.1]
Version 1.1.
- Small bug fixes
- Manual updated
- FLIPS now keeps record of zeros in the beginning and end of theory matrix rows, and does not rotate zeros unnecessarily. This leads to huge increase in performance if theory matrix is band matrix (or limited bandwidth matrix).
--Morispaa 12:02, 10 October 2007 (EEST)
Revision 115 [FLIPS-1.0.1]
Version 1.0.1. Few bugs in covariance calculation routines (found by Ilkka) fixed. Please update your FLIPS!
--Morispaa 03:46, 11 September 2007 (EDT)
Revision 108 tagged as FLIPS-1.0 [rev. 109]
Version 1.0 is out!
--Morispaa 07:59, 24 May 2007 (EDT)
Revisions 72-107
Too busy and/or lazy to explain these. Sorry!
--Morispaa 07:59, 24 May 2007 (EDT)
Revision 71
Fixed a bug and added two test programs.
--Morispaa 07:10, 8 January 2007 (CST)
Revisions 69-70 [FLIPS-0.9.9a]
Tagged Rev 68 as FLIPS-0.9.9a.
--Morispaa 01:35, 15 November 2006 (CST)
Revisions 67-68
Cleaning things up. FLIPS seems to be ready for final tests. Tag Rev 68 as FLIPS-0.9.9a!
Revisions 65-66 [trunk]
Manual added to folder doc/.
--Morispaa
Revision 64 [trunk]
Added some things for Intel's ifort.
--Morispaa
Revisions 61-62
Cleaning things up. Nothing special.
--Morispaa
Revision 61 [trunk]
Added Juha's new build system. Changed all file extensions to f90.
--Morispaa
Revision 60
Deleted old files. Nothing special.
--Morispaa
Revisions 58-59 [new-da]
Changed file extensions to f90.
--Morispaa
Revision 57
New branch for direct access I/O [new-da].
--Morispaa
Revision 56
Nothing special. Cleaning things up.
--Morispaa
Revision 55 (FLIPS-0.9.3)
Rev 54 tagged as FLIPS-0.9.3.
--Morispaa 10:26, 8 September 2006 (EEST)
Revision 54 (trunk)
Fixed a bug in the complex covariance calculation.
--Morispaa 10:18, 8 September 2006 (EEST)
Revision 53 (rot-branch)
New branch created for improving Givens rotation routines.
--Morispaa 10:23, 8 September 2006 (EEST)
Revision 45-52 (directaccess-branch)
Created a new branch for direct access I/O version. Works OK. Slow! Compiles on Intel Fortran.
--Morispaa 10:20, 8 September 2006 (EEST)
Revision 42 and 43 (FLIPS-0.9.2)
Changed the way the posteriori covariance matrix is calculated. Works little faster than before, and most importantly, works now also in Power5+.
Tagged as FLIPS-0.9.2.
--Morispaa 15:27, 14 June 2006 (EEST)
Revision 41
Code polishing. Nothing special.
--Morispaa 09:20, 13 June 2006 (EEST)
Revisions 39 and 40 (FLIPS-0.9.1)
Fixed a little bug and tagged rev 39 as FLIPS-0.9.1 (rev 40)
--Morispaa
Revision 38 (FLIPS-0.9)
Tagged trunk as FLIPS-0.9.
--Morispaa 11:39, 29 May 2006 (EEST)
Revisions 26 - 36 and 37
Please ignore revs 26-36. I just completely messed up the copying of newio-branch into the trunk (I blame SvnX!).
Revision 37 has the newio-branch moved into the trunk.
--Morispaa 11:36, 29 May 2006 (EEST)
Revisions 24 and 25 (newio)
Little fixes and I deleted the old doc files. New one is coming soon. This is now more or less ready to be merged into the trunk and to be tagged as 0.9. Besides the manual and some code polishing FLIPS is ready.
--130.231.89.87 11:20, 29 May 2006 (EEST)
Revision 23 (newio)
Fixed a little bug in the new I/O routines.
--Morispaa 12:46, 2 May 2006 (EEST)
Revision 22 (newio)
Added new I/O routine flips_get. Some little bugs fixed.
--130.231.89.87 10:22, 24 April 2006 (EEST)
Revision 21 (newio)
Started writting some new I/O routines. Routines ready, interfaces missing.
--Morispaa 14:56, 12 April 2006 (EEST)
Revision 19 (trunk), tagged as FLIPS-0.8.2 (rev 20)
Covariance bug fixed. Tagged as FLIPS-0.8.2.
--Morispaa 15:30, 11 April 2006 (EEST)
Revision 18 (newio-branch)
New makefiles. Some internal additions in the data types.
--Morispaa 14:16, 10 April 2006 (EEST)
Revision 17 (newio-branch)
New I/O routines ready and working. Performance decreases slightly (few procents) in memory storage side, but increases in the binary file storage side.
--Morispaa 14:14, 6 April 2006 (EEST)
Revision 16 (newio-branch)
New I/O subroutines added. Old I/O calls not yet replaced with the new ones.
--Morispaa 14:43, 5 April 2006 (EEST)
Revision 14, new branch (rev 15)
Revision 14 contains just little fixes. New branch 'newio' (revision 15) is an attempt to rationalize hopelessly incoherent I/O routines.
--Morispaa 11:43, 27 March 2006 (EEST)
Revision 12, tagged as FLIPS-0.8.1 (rev 13)
FLIPS-0.8.1 includes missing files for direct access. Module 'flips2.f95' contains direct access files and module 'flips.f95' has stream access files. Modify Makefile to use the version you want. The performance especially in covariance calculation is horrible with direct access (at least with Xlf).
--Morispaa 10:49, 24 March 2006 (EET)
Revisions 10 and 11, FLIPS-0.8
Committed revision 10. FLIPS_RESIZE ready. With flips_resize one can marginalize unknowns and/or add new unknowns.
Tagged as FLIPS-0.8 (rev 11) in tags. Some test programs added in 'tests'. Please download, test and comment!
--Morispaa 13:17, 22 March 2006 (EET)
Other stuff
BUG in full covariance matrix calculation! (FIXED in 0.8.2)
Just found a bug in calculation of full covariance matrix. Do not use it! Calculation of the diagonal (default) seems to be OK. I try to fix this asap.
--Morispaa 14:50, 11 April 2006 (EEST)
Quick'n'Dirty instructions for FLIPS_RESIZE:
subroutine flips_resize(nfob,ofob,newsize,remove,idnum,buffersize) type(flips_<s|d|c|z>) :: nfob type(flips_<s|d|c|z>) :: ofob integer, optional :: newsize logical, dimension(ofob%ncols), optional :: remove integer, optional :: idnum, buffersize
- nfob Uninitialized FLIPS data type (real, double, complex or double complex).
- ofob Old FLIPS data type (must be of the same type as nfob).
- newsize Integer, optional. Number of unknowns in nfob.
- remove Logical vector, dimension(ofob%ncols), optional. Mask vector for unknowns to be removed. If remove(i)==.TRUE., ith unknown will be marginalized.
- idnum Integer, optional. If given, nfob wii use binary files for storage with given ID number.
- buffersize Integer, optional. Rotation buffer size for nfob. If not given, default buffer size will be used (whatever that will be, currently 25).
Some rules
Optional arguments
- If only newsize is given:
- If newsize < ofob%ncols, ofob%ncols - newsize first unknowns will be marginalized.
- If newsize > ofob%ncols, newsize - ofob%ncols new unknowns will be added to the end.
- If newsize = ofob%ncols, nothing special happens. R and Y matrices of ofob will be copied to nfob.
- If only remove is given:
- All unknowns for which remove(i)==.TRUE. will first be rotated to the beginning and then marginalized. Number of unknowns in nfob will be ofob%ncols - count(remove).
- If both newsize and remove are given:
- First all unknowns for which remove(i)==.TRUE. will first be rotated to the beginning and then marginalized. After that newsize - (ofob%ncols - count(remove)) new unknowns are added to the end, so that the number of unknowns in nfob will be newsize.
Data types
- nfob must be uninitialized. If it is used before, it must be deallocated with flips_kill before calling flips_resize.
- ofob can be solved or unsolved. Note however that nfob will always be in unsolved state.
--Morispaa 10:20, 23 March 2006 (EET)
