[PD-dev] CodeWarrior compilation, measurements

Thomas Grill t.grill at gmx.net
Wed Aug 6 03:59:13 CEST 2003

there was a thread some time ago concerning compilation of PD for OSX 
with Codewarrior.
With some efforts i managed to do it (the key is to enable "Enums 
always int" !) and i did some profiling with it.
Surprisingly, the cpu load was practically identical between the gcc 
3.1 and the CW8 build for all the patches of the pd doc that i tried 
out... so, since ProjectBuilder is free, it seems that there is no 
apparent reason to engage in the somehow tiresome compilation with CW. 
(there is also the additional issue that in order to build externals 
with CW, pd has to be split up into a dylib part and a main app part)
Independently from that I also measured that a noticeable amount of cpu 
percentage is eaten up by the sys_getrealtime calls in the scheduler 
loop. Maybe there's a way to spare some of these calls? The other thing 
is that all the plus/minus/times dsp functions that are dominant for 
most patches could be easily accelerated using SIMD instructions 
(AltiVec on the PPC or SSE for Intel) since memory speed doesn't seem 
to be fully exploited here (signal vectors are in the cache). When i've 
got the time i'll transfer some of the flext SIMD code to the 
devel_0_37 branch.

best greetings,

More information about the Pd-dev mailing list