first test of yao + svipc on my macbook pro

[brujo]

today i finally found some time to test the new yao version françois
sent us a week ago.
i was really eager to get the numbers on my brand new macbook pro 6,2
(i7, 8G, ubuntu 10.04, default yorick install).
i'm targeting the elt scale and here are the relevant parameters i
used:

//-------------------------------
sim.pupildiam = 1008;
atm.screen = &(Y_USER+"data/screen4096_"+["1","2","3","4"]
+".fits");
//-------------------------------
nwfs = 1; // number of WFSs
wfs(n).shnxsub = 84;
wfs(n).pixsize = 0.2;
wfs(n).npixels = 12;
wfs(n).svipc = 4;
//-------------------------------
dm(n).nxact = 85;
dm(n).pitch = 12;
//-------------------------------
mat.method = "mmse-sparse";
mat.sparse_MR = 25000;
mat.sparse_MN = 500000;
//-------------------------------
tel.diam = 42;
tel.cobs = 0.3;

with this config i'm getting almost 1 it / sec (0.89). the system
setup takes several tens of minutes though (dominated by the imat
computation). reducing the number of phase pixels (4x4 / subap) i'm
getting 8 to 9 it / sec !!!

there are some remaining small bugs that prevented me from a
completely orgasmic experience:
- the graphic window systematically pops out from the python gui when
i launch simulations using svipc ... haven't identified the problem
yet
- i'm not able to launch 2 different simulations one after the
other ... i need a yorick restart ... françois told us it's a known
bug

anyway, very impressive françois ! two thumbs up !

d.

[Francois Rigaut]

On Sun, Jul 18, 2010 at 10:02 AM, brujo <dgrat...@gmail.com> wrote:

- the graphic window systematically pops out from the python gui when
i launch simulations using svipc ... haven't identified the problem
yet

that's also a known "bug". I forgot to mention it because I am rarely using the GUI these days.

there are several ways to startup your processes with svipc, but one convenient way is just to fork the main process: this way, you get everything initialized and all children in sync. One problem of this approach is that if you have some plot window opened when you fork, you also fork the X poll stuff, and this is bad (leads to a crash). The solution is to close all plot windows, fork, and re-open them.

That's what's causing the GUI plot window to detach: I haven't coded the bit that would re-open it at the correct window id (the drawing area inside the GUI).

Will fix in next release, or send you a patch when I get around to fix the code.

Francois

[Francois Rigaut]

Nice numbers brujo ! Obviously your new laptop has some guts.

By the way, there is another bug related to window handling: because I have to close the window and re-open after the fork, and I don't pass the dpi, the window will be re-open with the default dpi (I believe it's 75 (use window_geometry() to get the dpi of an existing window).

F

On Sun, Jul 18, 2010 at 10:02 AM, brujo <dgrat...@gmail.com> wrote: