This topic aims to collect each changes commited in the beta version in provision of the future release.

If you are interested by the beta release, please contact the jmmc user support to get access to the download page.

Note

There was a naming convention problem concerning beta packages V3.1b1 to V3.1b4 . Some have been distributed (between jully 2010 and March 2011) and correspond to some previous versions of the V3.0.2. Impacted Users have been warned and wrong numbered tags have been removed from the source code repository. Apologize for inconvenience -- Guillaume

V3.0.4 (or whatever it will be)

Here comes a new amdlib version aimed at speedup!:

1. the reduced Chi2 of the fringe fit is kept and stored in the raw visibilities OI-FITS files. To be complete, we should add in a next release the possibility to select visibilities according to this chi2 in amdlibPerform[Frame]Selection.
2. contrary to a well enforced useage, tagged values are no more blanked when it is possible. I should have set up a switch to permit this, but it would have been too complicated. So only the flags will really tell if a value is deemed wrong or not, which is what the OI-FITS norm expects, in fact.
3. The amdlibComputeOiData program has now a fast (switch -Z0) and a fussy (switch -Z2) mode. The "fussy" mode is the same as before: inversion of the full covariance matrix previous to complex coherent flux extraction, iteration on piston fit until high precision is reached. The fast mode considers that the covariance matrix is diagonal and the piston is evaluated with less precision. For most cases (especially when the S/N is not bad), the fast mode is enough. And it is fast, especially since the matrix routines have been optimized also. The -Z1 switch is reserved for an intermediate, "clever" mode based on S/N, that I have not developed yet.
4. The full O3 optimization with inlining etc is set by default in the Makefile. Moreover, the really slow parts of code have been parallelized with OpenMp pragma. The speed of amdlibComputeOiData (which is the only program really time-consuming in amdlib) scales now roughly with the number of cores, and is now limited by the cfitsio reading and writing of files. With gcc and modern computers this should be transparent, but one has to test it on several platforms.
5. All these changes have been reported in the Yorick interface. To set the speed switch to fast, use amdlibSetPreference,"fastMode",0.

Additionnally, this version closes two "tickets" we had:

1. Ticket 93: the VISDATA AND VISERR columns in OI_VIS (which are not OI-FITS compliant but ESO compliant) now store the Complex Coherent Fluxes & error (not the complex visibilities). One may use these values (on the RAW OI-FITS obviously) to add, combine and average the coherent fluxes as it likes.
2. Ticket 92: In case of selection averaging (amdlibPerformFrameSelection), the phase closure is retained as soon as two baselines are OK (in older versions (amdlib2 and amdlib3.0.1, etc) the phase closure would be deemed valid if the 3 baselines were retained by the selection criterium, giving really too few values for something which is essentially robust).

V3.0.3 *to be released on 27Jun2011*

• 'catalogsDir' preference is now properly defined for the users who are using the binary packages
• appendOiData now properly tags appended 'OIDATA_RAW' files as OIDATA_RAW.
• amdlibSearchAllStarDiameters now use the UD in H-band. It uses the estimation from only one catalog, the one with the best confidence (in order: Merand, Borde, JSDC).
• slight changes in Piston estimation (should be more robust).
• a memory leak in amdlibComputePiston2T (typo follwing the change above) has been corrected (10 May 2011).
• packaging script have been updated following cvs->svn migration
• yorick code version is checked against its C programs complement (a warning is displayed if versions differ)
• amdlib command get one new -e option to display one help message to configure emacs environment
• beta testers will get amdlib packages numbered as SVN_DDMMYY. this should increase tracability and help developpers and beta testers