hi,
not a real profiling, just a simple measurement of a graph_getrect() execution time. If there is a graph in a patch, this routine is called once for every mouse movement message from the gui.
The figures below are for a test patch with nothing but a single 100000-element (~2.5s) array in it, measured on an 866Mhz linux box (with -rt flag):
0.35-test10: avg. 49 mili(sic)sec, peak 71ms
0.35-test4: avg. 1.9 micro(sic)sec, peak 3us
This gives ratio of 25789. For a 100-element array this is 57us vs 1.5us, giving the ratio of 38.
The (apparent, not measured) variance is pretty small in both cases. I did not measured graph_getrect() in test9, because of its recursive nature.
Krzysztof
Krzysztof Czaja wrote: ...
the primary suspect is the graph_getrect() routine...
...
Miha Tomšič wrote: ...
The test10 is even worse: the slugish mouse motion and the output "static" is present while passing any pd patch.