Hello pd-folks!
I have a problem someone might have a solution to. Same pd-patch giving wildly different results on 2 machines.
Using a section of several [vibraphone~] objects from the percolate-lib, PD eats all my cpu, pushing it above 90%, and destroying sound. The machine is a Dell 5150, with 2.8GHz cpu.
The same patch runs clean (stays around 20% cpu) on another machine here, with slower cpu! (somewhere around 1.4GHz)
The computer giving me aches is a Dell laptop, inspiron 5150, running fedora core 2, with planetccrma-low-latency kernel.
uname -a: Linux localhost 2.6.7-1.437.1.ll.rhfc2.ccrma #1 Thu Jun 17 10:45:42 PDT 2004 i686 i686 i386 GNU/Linux
Pd is:
CVS-version: Pd version 0.39 TEST 0 Planet-CCRMA-version: Pd version 0.37.1 devel
Tried with pd from planet-ccrma and latest cvs-version compiled on this machine. Both the percolate-lib coming with the ccrma-dist, and one i built from a src.rpm on this machine. Same result whatsoever.
Ive tried shutting of the 'cpuspeed' process to keep the latptop running at full speed all times, but it didnt do much with the problem. Pd very soon hits 97-98% and stays there. Only save is to run pd -r 22050, where the patch runs around 40-50%.
Anybody got a clue?
Thanks,
-anders
Using a section of several [vibraphone~] objects from the percolate-lib, PD eats all my cpu, pushing it above 90%, and destroying sound. The machine is a Dell 5150, with 2.8GHz cpu.
my first thought would be that it's a denormal issue ...
you could try to recompile percolate with -mfpmath=sse -msse in your cflags. and try to use the devel_ branch of pd, since it switches on the sse2 flag for treating denormals as zero (by hardware) ...
second, you can try to profile your patch to figure out, what is eating your cpu ... (switch~ of the vibraphone~s)
third, look at the percolate code and try to fix it ;-)
cheers ... tim
Maybe its useful to mention the cpu-types. The fast one - 2.8GHz giving sluggish performance - is a Pentium 4. The "slow" one running at ~1.4GHz is an Athlon.
-anders
"a" == andersvi andersvi@extern.uio.no writes:
a> Using a section of several [vibraphone~] objects from the a> percolate-lib, PD eats all my cpu, pushing it above 90%, and a> destroying sound. The machine is a Dell 5150, with 2.8GHz cpu. a> a> The same patch runs clean (stays around 20% cpu) on another a> machine here, with slower cpu! (somewhere around 1.4GHz) a> a> The computer giving me aches is a Dell laptop, inspiron 5150, a> running fedora core 2, with planetccrma-low-latency kernel.
Quoting andersvi@extern.uio.no:
Maybe its useful to mention the cpu-types. The fast one - 2.8GHz giving sluggish performance - is a Pentium 4. The "slow" one running at ~1.4GHz is an Athlon.
Yes, this is almost certainly a denormal problem. P4s are terrible for that. Tim B and I went around and around a few times on this issue. He gave some fantastic advice, the gist of which is that you have to recompile PD and *all* externals with some denormal-busting flags. Check the list for "denormal" and my and Tim's names, and you'll find the conversation. Incidentally, I never did get my thing fixed. My P4 was too early to support these flags. But yours is probably a lot newer, so there's hope.
good luck, d.
hello,
it may comes from a denormal problem. i think vibraphone~ is based on a physical model, so there is certainly a loop inside.
I may wrong, but you should have a look at denormal on this list archive.
hope that help cyrille
andersvi@extern.uio.no wrote:
Hello pd-folks!
I have a problem someone might have a solution to. Same pd-patch giving wildly different results on 2 machines.
Using a section of several [vibraphone~] objects from the percolate-lib, PD eats all my cpu, pushing it above 90%, and destroying sound. The machine is a Dell 5150, with 2.8GHz cpu.
The same patch runs clean (stays around 20% cpu) on another machine here, with slower cpu! (somewhere around 1.4GHz)
The computer giving me aches is a Dell laptop, inspiron 5150, running fedora core 2, with planetccrma-low-latency kernel.
uname -a: Linux localhost 2.6.7-1.437.1.ll.rhfc2.ccrma #1 Thu Jun 17 10:45:42 PDT 2004 i686 i686 i386 GNU/Linux
Pd is:
CVS-version: Pd version 0.39 TEST 0 Planet-CCRMA-version: Pd version 0.37.1 devel
Tried with pd from planet-ccrma and latest cvs-version compiled on this machine. Both the percolate-lib coming with the ccrma-dist, and one i built from a src.rpm on this machine. Same result whatsoever.
Ive tried shutting of the 'cpuspeed' process to keep the latptop running at full speed all times, but it didnt do much with the problem. Pd very soon hits 97-98% and stays there. Only save is to run pd -r 22050, where the patch runs around 40-50%.
Anybody got a clue?
Thanks,
-anders
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://iem.at/cgi-bin/mailman/listinfo/pd-list
"a" == andersvi andersvi@extern.uio.no writes:
a> Hello pd-folks! I have a problem someone might have a
a> solution to. Same pd-patch giving wildly different
a> results on 2 machines.
Problem fixed. Yes, it was the "denormal-numbers-on-P4"-problem
Recompiled pd and percolate-lib with
-mfpmath=sse -msse
included in CFLAGS got rid of the problem. (not sure whether acutally recompiling pd was needed here, guess it dont hurt either.)
Christmas bell-chimes playing at 20% cpu - sincere thanks to those who replied with suggestions.
-anders
-
andersvi＠extern.uio.no -
cyrille henry -
Derek Holzer -
Tim Blechmann