=========== YAO package =========== Copyright (c) Francois Rigaut (frigaut) 2002-2011. Initial release June 2002. Developers: ------------ Francois Rigaut frigaut@gemini.edu | frigaut@gmail.com Marcos van Dam marcos@flatwavefronts.com Ralf Flicker + July 2009 Damien Gratadour damien.gratadour@obspm.fr Aurea Garcia-Rissmann agrissmann@gmail.com Synopsis: --------- Yao is a Monte-Carlo simulation tool for Adaptive optics (AO) systems, written as a yorick plugin. It uses a number of custom developed functions to simulate wavefront sensors (WFS), deformable mirrors (DM) and many other aspects of an AO loop. The core functions are written in C, hence are very fast. Help pages: ----------- http://frigaut.github.com/yao/index.html RELEASE NOTES: -------------- version 4.8.3, 2011feb11: - move all calls to fitsread and fitswrite to new yao_fitsread and yao_fitswrite - changed graphic style to display residual phase - implemented forks over WFSs (sim.svipc) - some changes to be compatible with xft - changes to test sim.svipc version 4.8.2, 2010dec17: - Added parallelization of WFS (i.e. one thread/WFS in a multi WFS system) - Tried implementing FFTW parallelization, but the gains were really marginal and often it would be more detrimental than good. Reverted, but left the code commented in Makefile, yao_fast.i and yao_fast.c - yao now looks for and includes yao.conf, if any (before was only in yao GUI) - changed all calls to fitsRead and fitsWrite to new function names yao_fitsread and yao_fitswrite. This way, I work around any conflict with fits.i function checks. Renamed newfits.i to yao_newfits.i - fixed a number of bugs in svipc implementation (mostly how it recovers from a previous svipc run within the same session). test-all.i now runs good with sim.svipc = 1, 5, with wfs.svipc>1, as well as sim.svipc=1 *and* wfs.svipc>1 within the same run. - added items in check_parameters() to check for the svipc bad settings - re-introduced sim.name in graphic and nudged some labels in graphics a bit - introduced some changes to make the graphic look better when using xftfonts. also resized controlscreen - better handling of window closing/opening when using svipc - push to get reference tip is now 0.005 instead of 0.05 (there's no noise) - fixed a bug that animate would not be ON for display in close loop when using svipc - switched the Makefile to -Wall - 2 new sim structure members: sim.svipc_wfs_nfork and sim.svipc_wfs_forknb - fixed a bug in pyramid wfs, hopefully once and for all (bad sinc() function used). - added test3.par (same as glao.par) - fix a bug in test5.par (phase screen referred in parfile may not have existed) - fixed issue with animate being ON and preventing further plot - fixed bug in which background calib was not passed to wfs.svipc child (since 4.7.1) - fixed nitegcycle for shwfs when no svipc - written test routine (examples/test-fr.i) to check correctness of main modes of svipc (wfs and sim). version 4.8.1, 2010dec06: - changed layout of graphic window to add residual phase - display of residual phase (Marcos) - nintegcycle in pyramid wfs (Marcos) - "extended" object in pyramid wfs (Marcos) - a couple of bugfix (Marcos and I) version 4.8.0, 2010nov09: - XFT related: * increased size of axis labels fonts on plsys(4) - newfits: * fixed error handling. Now error in place, instead of calling error handling function (exitInError). Allows much clearer debugging. - Pyramid: * merged pyramid_wfs() changes from marcos, upgraded my francois * changes in yao main routines to make re-introduced obsolete "pyramid" wfs type * upgraded GUI to include pyramid modulation amplitude control * added 4 new structure members to wfs structure - MISC: * added "--dpi ##" command line call option (for GUI call). Example: yao --dpi 75 sh6x6.par * added default animate() to iMat acquisition display. * cleanup of yao.py (dictionary autoconnect) + dpi stuff * shortened get_style() calls to viewport() call * gotten rid of get_processor_number() (use nprocs() function in svipc) * added a bunch of extern wfs, dm, ... in functions called by python to set values of parameters (how could it have worked before is a mystery). - BUG FIXES: * fixed a bug in which the graphic would get stuck (animate call was done without ensuring a plsys,1) version 4.7.3, 2010 oct12: - merged SH threshold methods #2 and #3, patches from Yann Clenet Credits for original idea of method #3 (compute centroid on N brightest pixels instead of using a flux threshold) go to E.Gendron. - added some safety for the choice of shthreshold when using shmethod=3 to avoid getting segfaults in yao_fast. version 4.7.2, 2010 oct12: - added smooth animate=1 for iMat recording/display - added ETA when taking iMat - added option to findfwhm() to save RAM requirements (help,findfwhm) - merged patch from Marcos van Dam: * implemented virtual DMs for tomography * implemented save of residual phase in go() (help,go) version 4.7.1, 2010sep16: - fixed nasty bug that was not switching noise off for background calib (SHWFS) Thanks Yann Clenet for noticing and pointing me to the right direction - make screen now can drop the imaginary part (no_ipart=1) to produce larger screens by saving some RAM. - Gotten rid of hdf5 call in yaodh (and thus of hdf5 dependency) - switched the default mode in test-all.i to non svipc Version 4.7.0, 2010aug14: - Merged a bunch of changes from Marcos van Dam. From his changelog: 1. There is a new version of soy.i. I found more bugs and since it had to be modified, I added new functions. FR: I merged these changes in soy 1.3 2. filtertilt set to 0 for interaction matrix generation. I need this to generate sparse matrices. 3. filtertilt for zernike and stackarray DMs. I need to filter tilt from DMs when doing tomography with LGSs. 4. WFS measurements set to 0 if not from same subsystem as DM. 5. minimum zernike (to exclude piston/tip/tilt if desired). I need this feature for tip-tilt tomography. 6. I remove the reporting of the number of corrected modes... do we really need this? it makes no sense for anything other than classical AO. 7. pupil (real = 0). We discussed this earlier. 8. Introduced virtual DMs needed for tomographic reconstruction 9. mmse and mmse-sparse works fine with MCAO 10. Have loop.method set to closed-loop, open-loop and pseudo open-loop. Open loop means WFS sees uncorrected WF, not corrected WF. 11. Tomographic reconstruction! - Fixed a broken re-opening of graphic window because of undefined dpi. Version 4.6, 2010jun17: - Merged Aurea Garcia-Rissmann additions for diskharmonic DMs Part of the code from Mark Milton (at the time at UofA). Used with authorization. - Merged parallelization code (svipc). Can lead to 3x gains on multicore machines. Tweak with sim.svipc = value (see yao_structures) - Gotten rid of redondant code in utils.c. poidev was in there, as well as other routines. These are already in imutil and imutil is definitely a dependency. This, combine with using the latest imutil v 0.5.6, solves a bug in which poidev would SIGFPE when using too brights GS in with CWFS mode. - solved a bug in which not specifying skymag would means skymag=0 and thus lots of photons in the sky in CWFS mode. Now skymag=0 (default) is interpreted as "no sky". - upgraded all the PROTOTYPES to new recommended syntax "type array" instead of "pointer". V 4.5.6 - changed "least-squares" -> "mmse" - changed "sparse" -> "mmse-sparse" - updated documentation with input from Marcos Van Dam // sorry for inconsistency. Below are cvs revision number of yao.i * Revision 1.16 2010/06/09 15:03:42 frigaut * - Merged changes of Marcos Van Dam: This implements new reconstructors * methods "least-squares" (in fact a MMSE-like) and "sparse" (same but * using sparse matrices, very fast). This adds a dependency on soy. * There's now a few more elements in the dm and mat structures * * - added thback and cleaned up indentation in yao_fast.c * * Revision 1.15 2010/04/15 18:11:24 frigaut * * - strlower -> strcase * * Revision 1.14 2010/04/15 02:36:53 frigaut * * * final commit to upgrade this repo to yao 4.5.1 * * Revision 1.13 2008/11/19 00:53:19 frigaut * - fixed memory leak in yao_fast.c (thanks Damien for reporting that) * - fixed comments in newfits.i * - upped version to 4.2.6 * * Revision 1.12 2008/05/12 18:23:48 frigaut * version change * * Revision 1.11 2008/05/11 14:03:56 frigaut * - implemented zernike wfs * - gotten rid of faulty round function in yao_util * * Revision 1.10 2007/12/27 09:06:28 frigaut * - bumped to version 4.2.3 * - corrected problem with glade path (python does not like ~, so * replaced by expansion) * * Revision 1.9 2007/12/20 13:35:55 frigaut * bumped to v4.2.2 * * Revision 1.8 2007/12/20 13:34:53 frigaut * - various bug fixes * - better handlng of default parfile path * - better handling of options menu (WFS and DM) * * Revision 1.7 2007/12/19 20:00:08 frigaut * - statusbar updated to indicates where the results are saved when done * - bumped to 4.2.1 * * Revision 1.6 2007/12/19 19:44:19 frigaut * - solved a number of bugs and inconsistencies between regular yao call and * GUI calls. * - fixed misregistration for curvature systems * - change: misregistration entry from the GUI is now in pupil diameter unit, * not in subaperture unit! * - changed default efd in c188-bench.par * * Revision 1.5 2007/12/19 15:45:32 frigaut * - implemented yao.conf which defines the YAO_SAVEPATH directory where * all temporary files and result files will be saved * - modified yao.i and aoutil.i to save in YAO_SAVEPATH * - bumped version to 4.2.0 * - slight changes to GUI (edit conf file) * version 4.1.1: 2007dec13: - upgrade/update of Makefile version 4.1: 2007dec12: - cleaned up all things for the debian release and a fresh pkg_mngr release - cleaned up the LICENSE issue (GPLv2) - suppressed warning message if fftw_wisdom.dat not found. In my experience, on modern hardware and for fftw3, this does not improve speed. One can still generate and use one if wanted (run init_fftw_wisdom) - produced man page (doc/yao.1) version 4.0: 2007jun03: - Added all GTK GUI. No further modifs in yao, hence yao can still be used safely without a gui (scripts, scripts). - migrated to a flat directory structure for yao. There is no more yao_fftw subdirectory - This is now meant to be installed in i0, so the path to the yao required functions has been changed from yao/file.i to file.i version 3.81, 2007jun02: added gui_message version 3.80, 2007apr19: now 64 bits safe ! version 3.75, 2007feb22: added angle offset for curvature wfs and dms 2006mar01: corrected bug. reset put command to 0, while it needed to zero dm._command v3.6 2005dec01 - renamed aosimul -> yao - reworked files and directory structure to accommodate new yorick plugin binary distribution scheme. - aoloop now -> aoloop + go structure to have prompt active while the loop is running. type "stop" to stopo the loop and "cont" to resume it. - simple principle demo of tyk control v3.4.0 2004sep10 - Makefiles completely reworked. Install more easy. See README in distribution directory (yao_veclib or yao_fftw) for more info. - added a printout of the yao versoin used in yao_fast.i's - modified to include yao and yao_fast.i at parse - This is the first version compatible with plugins. - modified tic and tac to accomodate multiple counters - now the "Time Left" functionality works again. v3.3.2 2004aug02 - fixed a few bugs: - a division by zero was causing a SIGFPE for odd number of subaperture shack hartmann systems in the angle/elongation calculations in shwfs_init (bug noted by Miska) - Added a check of indices overflow when determining the Y interpolation points in get_turb_phase_init v3.3 2004July28 - semi major changes in this version. Watch out, as some parameters meaning has changed (I know this is silly, but it's more logical now) - Implemented zenith dependance. Now one can change one parameter (gs.zenithangle) and everything change as needed (r0, lgs altitude and thickness of Na layer, atm layer altitude, LGS brightness) - Imlemented Rayleigh fratricide effect for multiple LGS systems! This was quite an endeavour. I had to modify the _shwfs routine to include several new lines and parameters relative to Rayleigh and calibration (a side effect is that we can now do and use calibration frame in this routine). To enable Rayleigh calculations set wfs.rayleighflag to 1 (it will *not* be calculated for NGS WFS as it is assumed that you can easily block the light from the laser wavelength). - The way to specify the photometry has also changed. Now, if you deal with a LGS, you specify a power and a return per watt, not anymore a magnitude (magnitude are still ok for a NGS). Also, everything -in term of zeropoints- is now specified at the entrance of the telescope: gs.zeropoint is now the number of photons for a zero mag star per sec per pupil at M1 (it used to be detected by the WFS). gs.lgsreturnperwatt is also in photons at M1. Instead, I have added a wfs.optthroughput parameter to specify the optical throughput of each WFS. Please set it, as it defaults to 1. - In summary, the following keywords were added in the parfile: - gs.lgsreturnperwatt - gs.zenithangle - wfs.laserpower - wfs.rayleighflag - wfs.optthroughput v3.2.3 2004jun07 - fixed a bug that yao_fftw could not write the file fftw_wisdom.dat in ~/Yorick. Made use of get_end("HOME") to get correct path. - fixed screen names in all example parfiles to 1,2,3,4 - changed several other minor inaccuracies v3.2.2 2004jun07 - Yao is now under CVS! - several cosmetic modifications (this NOTES file, ...) - changed convVE -> convol for compat with yao_fftw in aoutil.i v3.2.1 2004apr29 - implemented sky and dark current in WFS structure and in yao_fast.c, both veclib and fftw. checked by printing _fimage, (sum) and (rms). for a 4cell. the level is OK and the noise is OK. - implemented the same for the CWFS. Note that the dark current is per channel and per sample, i.e. for 2 reads (each extra-focal image). Per read, the dark current is taken as wfs.darkcurrent/2 v3.2.0 2004apr07 - reworked the distribution tree - implemented a faster computation of the DM shape for PZT DMs where only the relevant subarray is used in the summation. Instead of precomputing the Influence function (IF) on the whole pupil, I compute it on a fraction of it and store the indices. The branch is done in the new function comp_dm_shape. if dm.elt == 1, then we use the new and faster method. To date, dm.elt is set to 1 in check_parameters (aoutil.i) for any strackarray DM. This is just a patch for now as there is a problem with the computation of the anisoplanatism modes. v2.5.3 2004mar11 - found and fixed a bug in the determination of wfs._tiprefv and wfs._tiltrefv. - implemented dithering for LGS (using uplink, so only for LGS) v2.5.2 2004mar10 - added capability to have extrapolated actuators. I have done that with only a few lines of codes, which tends to prove that the code is easy to upgrade and relatively well structured (allright, subjective judgement!). v2.5.1 2004mar02 - fined tuned the spot elongation - changed wfs.imatkernel into wfs.kernel (serves also for LGS spot FWHM in aoloop) v2.5 2004feb13 - implemented the LGS spot elongation: - added a number of _entries in wfs structures (_kernels, _kerfftr, _kerffti, _x, _y) - modified shwfs_init accordingly - added the element wfs.LLTxy - added wfs.gsdepth v2.4.1 2004jan30 - Strehl, fwhm and e50 are now extern variables so one can access them after completion of aoloop. - added wfs.fracIllum = fraction of subaper. illuminated for sub to be valid - reworked configuration graphical display ( graphic_config() ). - added the possibility to interactively select the threshold for valid/not valid actuators (put dm.thresholdresp = -1 in parfile) v2.4 2004jan22 - written and implemented parameterCheck, to check parameters in aoread. - rather big change, at least in matter of compatibility: I have changed the normalization of the phase, IFs and phase screens. Previously, the problem was that I could not have different lambda per sensor, as the input phase screens are normalized at one single given dr0. Now the input dr0 is given at 0.5 microns, and everything is normalized in microns (IF, phase screens, phase). New keyword: atm.dr0at05mic, instead of atm.dr0 v2.3.2 2004jan21 - I got the MCAO LGS working nicely !! with the anisoplanatism modes as they are handled in MCAO, and implemented in a simple, non obstrusive way. Great. I had to think a long time about the best way to implement it, but now it's done. - worked out the output to the file parprefix+".res" - debugged quite a bit with different parfiles - implemented a new input "sim.name" that contains a comment string to put on all graphics. - reworked the format of some output (e.g. WFS parameters) v2.3.1 2004jan20 - implemented different subsystem, i.e. each dm and wfs is specified to belong to a given subsystem. The cMat for each subsystem is then computed separately. This allow a good separation of e.g. the TT NGS WFS and the HO LGS WFS in a LGS system. I have added the following keywords: - wfs.subsystem - dm.subsystem - mat.condition is now a pointer to a vector of condition numbers v2.3.0 2004jan18 - implemented ShWFS method 1, i.e. straight gradient average, method 2 being the full propagation calculations. - implemented bias and flat in ShWFS, method 2 (new keywords wfs.biasrmserror and wfs.flatrmserror) - implemented target multi-lambda PSF. Changed target.lambda, now a pointer, use to be a float. e.g. target.lambda = &([1.25,2.2]*1e-6) note 2004jan22: now lambda in microns - implemented hysteresis on DMs. Adapted routine from idl routine new parameter dm.hyst - passed thresholdresp from hardcoded to parameter, dm.thresholdresp, separate for each DM. - implemented central obstruction. v2.2.4 2004jan12 - implemented gaussian and poisson noise in shwfs and cwfs ! - implemented the correction of the uplink TT for LGS. v2.2.3 2004jan11 - added proper reference measurement subtraction - added tip-tilt filtering of WFS measurements v2.2.2 2004jan11 - fixed a rather major bug in the computation of the DM shapes. In my haste of implementing the get2dphase, I had used only one interpolation x coordinate vector (see get_turb_phase_init). I was using the vector precomputed for the screens, for the DM... I have fixed it by computing dmxposcub (and y). - Got back to using *dm._def instead of the def variable to stock the influence functions. This is because I can save much more space by allowing different n1 and n2 per mirror. v2.2 2003dec28 - coded _shwfs, shwfs routine in C, using veclib FFT call. - Got rid of wfs.shmethod in parfile - added wfs.npixels in parfile - changed wfs.subsize into wfs.pixsize in parfile - added the routine wfs_check_pixel_size - also, I have moved the influence function setup functions in aoutil.i to reduce the size of yao.i v2.0 2003dec12 - written and included "optimized" C routines in aoloop. - _fftVE to compute set of image from set of phases - _dmsum to compute mirror shape(s) from set of command - _get2dPhase to integrate a set of phase screens along a given direction and return an integrated phase. all in all, this has resulted in a rather large reduction of execution time in this routine (the most critical). When I started the optimization, I was at 10 iteration/sec for sh6.par, a 6x6 SH, 30 pixel pupil, 1 DM, 1 GS, 4 targets case. In the intermediate stage I am right now, with the 3D phase integral, the tagret FFTs and the DM shape computation optimized, I am now running at about 95 iterations per seconds on the G4. I am left with some optimization to do, namely the SH WFS and grouping the integrated phase and target image computation in one subroutine call. v1.2.5 2003Jun15 - added multi viewport display in aoloop v1.2.4 2003May8 - debugged and fixed the method 2 for SHWFS v1.2.3 2003mar16 - modified support of IFs to reduce RAM consumption. before: with tests/mcao.par (5wfs, 3dm, 16x16, 64x64 pupil), the ram (RSIZE) was 37M at the complexion of the IF calculation. v1.2.2 2003mar03 - added some graphics output at the end of aoinit - added some comment and general explanation of calling sequences below v1.2.1 Feb 2003. - Modified the selection of valid actuators to cope with altitude DMs. v1.2 February 2003. - a whole lot of changes in this version, which is basically made to handle MCAO, i.e. one or more DM, one or more WFS. Should be able to deal with Classical AO, altitude conjugated AO, ground conjugated wide field AO, MCAO. - The input parameter file has been changed, together with the code. Now each DM and each WFS has its own parameter set stored in a structure. The idea is that the structure is loaded into a working structure before one calls the WFS routine. feb 24, 00:35. This version seems now stable enough that I can freeze it and go on subversion for improvements - the basic layout of the software is now as follow: - in aoloop: - aoloop calls mult_wfs(iter). - for each WFSs, mult_wfs calls get_phase2d_from_dms(offsets,gsalt), which returns the integrated phase along the line of sight for the given WFS (offset,gsalt). The core shifting routine of get_phase2d_from_dms is interp2dmir. - for each WFS, multWFS calls get_turb_phase(iter,#star,starType (wfs|target)), which returns the integrated phase accross the phase screens for the given star and iteration #. The core shifting routine of get_turb_phase is interp2dturb. - mult_wfs then execute the given WFS routine on the total phase and returns the cummulative measurement vector. - aoloop then manages the WFS measurement vector history to pass the vector with correct delay to the matrix multiplication - multiplication by command matrix. - for each mirror, the mirror shape is constructed from the command vector and stored in "mircube(size,size,ndm)". - if certain conditions are met, the PSFs are computed at all needed points in the FoV, and statistics are accumulated. - To summarize: 1) aoloop 1.1) multWFS(iter,ok), for each sensor: 1.1.1) get_phase2d_from_dms(offsets,gsalt) 1.1.1.1) interp2dmir(shifts) 1.1.2) get_turb_phase(iter,gsalt,ok) 1.1.2.1) interp2dturb(shifts) 1.1.3) Get and accumulate sensor measurements 1.2) delay management 1.3) multiply by cMat 1.4) compute PSFs and accumulate statistics if ok to do so. v1.0.8 February 7, 2003 - upgraded all calls to fits_read and fits_write (from fitsRead and fitsWrite, to conform to fits2.i) - fixed one problem in make_pzt_dm (wrong pitch) - also IF back to log form (laserdot form) as sinc is no good. - SVdec broken in last yorick versions. Works with 1.5.13 - implemented debug printout and displays - This version is fully functional. v1.0.7 June 17, 2002, FR - added some parameters (LoopFrameDelay, LoopModalGainFile) - implemented modal gain optimization routine. v1.0.6 - Remove ao.Size from accesible structure. Set in aoinit. - Cleaned up par file - Trim ActIF at the end of aoinit (therefore for use in aoloop). This saves a lot of computation time in the mirror surface calculation. v1.0.5 First Release, June 11, 2002, FR BUGS: ----- - dm.elt is not compatible with MCAO systems, or more generally "aniso" DMs. This is because one has to project on modes and this is not possible with the localized version of dm shape computation (with dm.elt=1). TO DO: ------ - implement NCPAs - add a red light when actuators are saturating in the controlscreen (and the number of saturated actuators) ---------------------------------------------------------------------- YAO_FFTW: Note: updated 2007jun03 This directory includes all the files necessary for the production of the executable "yao" (Yorick Adaptive Optics). This package makes use of dedicated C routines to compute wavefront sensor images/signal, deformable mirror shape, etc... It uses fast FFT routines from FFTW. FFTW HAS TO BE INSTALLED ON YOUR SYSTEM (see below). //============================================================= INSTRUCTION FOR THE COMPILATION OF YAO_FFTW: 1. First Update the Makefile: yorick -batch make.i NOTE: if you don't have yorick in your path, or don't have a symlink to it, you can specify the path by hand: e.g.: ../../../Darwin-Power_Macintosh/bin/yorick -batch make.i or ../../../Linux-i686/bin/yorick -batch make.i 2. If you have installed FFTW3 locally (i.e. not system wide install, that is you used the -prefix="/some/path/in/your/home/directory/fftw3/" when compiling FFTW3), then you have to edit manually the Makefile and update the path to the libraries and include files. I have put a FFTW3_PATH variable that you need to update to reflect your installation, e.g.: FFTW3_PATH=/home/frigaut/fftw3 3. Update the path to the yorick executable (search for YORICK_EXE at the end of the Makefile). 4. Compile, test and install the plugin: make make check make install You may want to specify optimization flags when building the plugin. Here is the one I use: On a Apple G4: make plugin COPTIONS="-O3 -fomit-frame-pointer -mcpu=7450" On a Apple G5: make plugin COPTIONS="-O3 -fomit-frame-pointer -fstrict-aliasing -mcpu=970 -mtune=970 -mpowerpc-gpopt" for a Linux system: make COPTIONS="your favorite options, if any" 5. yao should be installed. Now you can start a regular yorick session and type #include "path_to_include_file/yao.i" e.g. #include "yao/yao.i" if the yao tree is installed in Y_SITE/contrib This will include all the necessary files and load the plugin you just built. //============================================================= HOW TO INSTALL FFTW3: Obviously, if you are on a debian or fedora type system, you can easily install fftw3 fro the repositories. If not: 1. Download fftw3 from http://www.fftw.org/ 2. gunzip and untar the package 3. configure and compile. You have to use special compilation flags to enable the use of the vectorial processor on your machine (if you have any). Also, you have to enable fftw3f, i.e. float operation (yao uses floats for FFTs). --enable-float enables float operations. If you don't have root privilege, you can install fftw3 locally. Use the --prefix to do that (see an example below) Each machine takes different flags. A good resource to find out which one you should use is the speed benchmark pages at fftw.org (http://www.fftw.org/speed/), i.e. http://www.fftw.org/speed/g5-2GHz/ for the G5. For an Dual Ahtlon, I used: ./configure --enable-float --enable-k7 CFLAGS="-O2 -fomit-frame-pointer -Wall -W -Wcast-qual -Wpointer-arith -Wcast-align -pedantic -malign-double -fstrict-aliasing -mpreferred-stack-boundary=4 -mcpu=pentiumpro" -prefix="/home/frigaut/fftw3/" The -prefix="/home/frigaut/fftw3/" points to the local installation directory. For a G5, I use: ./configure --enable-float --enable-altivec CFLAGS="-O3 -fomit-frame-pointer -fstrict-aliasing -mcpu=970 -mtune=970 -mpowerpc-gpopt" For a G4: ./configure --enable-float --enable-altivec CFLAGS="-O3 -fomit-frame-pointer -mcpu=7450 -faltivec" then: make make install //==============================================================