hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
thanks... the pain! the pain!
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
- compile with -mfpmath=sse -msse (the sse unit is much better handling
denormals)
- use devel_0_39. it includes code to switch on hardware DAZ (Denormals
Are Zero) on cpus supporting sse2
hope this helps ... t
Hmm, weren't PD_BADFLOAT and PD_BIGORSMALL defined already on your system? It's "idfef __i386__" which I assumed would be turned on for Celerons. What compiler are you using???
It's PD_BIGORSMALL that gets tested everywhere these days. But if you found a way to make denormals despite all the PD_BIGORSMALL fixes, I'd like to know where they're sneaking in.
I don't remember offhand if celerons have sse2 or not, and don't know if the -mfpmath=sse -msse flags work outside linux, but will at least try those on my system...
cheers Miller
On Tue, Aug 23, 2005 at 09:29:56AM +0200, Tim Blechmann wrote:
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
- compile with -mfpmath=sse -msse (the sse unit is much better handling
denormals)
- use devel_0_39. it includes code to switch on hardware DAZ (Denormals
Are Zero) on cpus supporting sse2
hope this helps ... t
-- mailto:TimBlechmann@gmx.de ICQ: 96771783 http://www.mokabar.tk
latest mp3: kMW.mp3 http://mattin.org/mp3.html
latest cd: Goh Lee Kwang & Tim Blechmann: Drone http://www.geocities.com/gohleekwangtimblechmannduo/
After one look at this planet any visitor from outer space would say "I want to see the manager." William S. Burroughs
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
On Tue, 2005-08-23 at 09:31 -0700, Miller Puckette wrote:
Hmm, weren't PD_BADFLOAT and PD_BIGORSMALL defined already on your system? It's "idfef __i386__" which I assumed would be turned on for Celerons. What compiler are you using???
gcc (GCC) 3.4.3 sorry - my c is a bit rusty (read: non-existant). i didn't know if it was being defined or not, so i added it to the makefile.
It's PD_BIGORSMALL that gets tested everywhere these days. But if you found a way to make denormals despite all the PD_BIGORSMALL fixes, I'd like to know where they're sneaking in.
so would i...
I don't remember offhand if celerons have sse2 or not, and don't know if the -mfpmath=sse -msse flags work outside linux, but will at least try those on my system...
they do have sse2.
cheers Miller
thanks!
On Tue, Aug 23, 2005 at 09:29:56AM +0200, Tim Blechmann wrote:
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
- compile with -mfpmath=sse -msse (the sse unit is much better handling
denormals)
- use devel_0_39. it includes code to switch on hardware DAZ (Denormals
Are Zero) on cpus supporting sse2
hope this helps ... t
-- mailto:TimBlechmann@gmx.de ICQ: 96771783 http://www.mokabar.tk
latest mp3: kMW.mp3 http://mattin.org/mp3.html
latest cd: Goh Lee Kwang & Tim Blechmann: Drone http://www.geocities.com/gohleekwangtimblechmannduo/
After one look at this planet any visitor from outer space would say "I want to see the manager." William S. Burroughs
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Hmm, weren't PD_BADFLOAT and PD_BIGORSMALL defined already on your system? It's "idfef __i386__" which I assumed would be turned on for Celerons. What compiler are you using???
gcc (GCC) 3.4.3 sorry - my c is a bit rusty (read: non-existant). i didn't know if it was being defined or not, so i added it to the makefile.
well, off course this will only work, if the externals you're using are denormal safe...
It's PD_BIGORSMALL that gets tested everywhere these days. But if you found a way to make denormals despite all the PD_BIGORSMALL fixes, I'd like to know where they're sneaking in.
so would i...
what about:
|osc~| | |*~ 1e-10| | |*~ 1e-10| | |*~ 1e-10| | |*~ 1e-10|
this example is rather academic, but it shows, that denormals can't completely be prevented (unless denormal bashing would be added everywhere in the dsp tree)
t
well, fwiw, adding the sse flags, an -o3 and removing -g gave me about a 33% inprovement in cpu usage. and denormals *seem* to be gone, but this was a brief test. will test more and let you know. thanks for the help!
On Tue, 2005-08-23 at 23:26 +0200, Tim Blechmann wrote:
Hmm, weren't PD_BADFLOAT and PD_BIGORSMALL defined already on your system? It's "idfef __i386__" which I assumed would be turned on for Celerons. What compiler are you using???
gcc (GCC) 3.4.3 sorry - my c is a bit rusty (read: non-existant). i didn't know if it was being defined or not, so i added it to the makefile.
well, off course this will only work, if the externals you're using are denormal safe...
It's PD_BIGORSMALL that gets tested everywhere these days. But if you found a way to make denormals despite all the PD_BIGORSMALL fixes, I'd like to know where they're sneaking in.
so would i...
what about:
|osc~| | |*~ 1e-10| | |*~ 1e-10| | |*~ 1e-10| | |*~ 1e-10|
this example is rather academic, but it shows, that denormals can't completely be prevented (unless denormal bashing would be added everywhere in the dsp tree)
t
well, fwiw, adding the sse flags, an -o3 and removing -g gave me about a 33% inprovement in cpu usage. and denormals *seem* to be gone, but this was a brief test. will test more and let you know. thanks for the help!
well, this is no big surprise ... the sse unit is _far_ better handling denormals than the fpu is ... even without hardware DAZ
t
moin folks,
sorry to butt in, but as my current laptop is dying a slow and painful death (by eating its own RAM), I'm likely soon to be faced once again with one of the eternal questions: AMD or Intel?
I know I've had this conversation with Tim, but the last time I looked (homebrewed Knoppix derivate with a test patch using [plugin~ flanger] (Steve Harris's LADSPA flanger)), AMD's hardware hack simply blew every Intel processor out of the water when denormals cropped up -- can I assume that with a bit of work tweaking compilation flags, I can get similar performance out of an Intel processor these days?
marmosets, Bryan
Tim Blechmann wrote:
well, fwiw, adding the sse flags, an -o3 and removing -g gave me about a 33% inprovement in cpu usage. and denormals *seem* to be gone, but this was a brief test. will test more and let you know. thanks for the help!
well, this is no big surprise ... the sse unit is _far_ better handling denormals than the fpu is ... even without hardware DAZ
t
sorry to butt in, but as my current laptop is dying a slow and painful death (by eating its own RAM), I'm likely soon to be faced once again with one of the eternal questions: AMD or Intel?
well, i don't think the question is this easy ... you could ask "celeron or centrino or amd or ... ?", you could ask, "does my favourite software make full use of my hardware?", you could ask, "how does my distribution build binaries?" ...
running on pentium-m/gentoo(CFLAGS="-O3 -march=pentium-m -mmmx -msse -msse2 -mfpmath=sse")/devel_0_39(enabling hardware DAZ), i haven't had a denormal problem for ages ...
t
thanks - i'll give that a try.
On Tue, 2005-08-23 at 09:29 +0200, Tim Blechmann wrote:
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
- compile with -mfpmath=sse -msse (the sse unit is much better handling
denormals)
- use devel_0_39. it includes code to switch on hardware DAZ (Denormals
Are Zero) on cpus supporting sse2
hope this helps ... t
Hi
It's kind of a beginner question but I'll dare asking:
What is a denormal? (-:
Jé
shift8 a écrit :
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
thanks... the pain! the pain!
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
in some caculations for audio (feedback, etc) there can be the creation of really small or really large numbers. these numbers can cause cup usage spikes in some processors when they try to process them.
http://www.musicdsp.org/phpWiki/index.php/What%20is%20a%20denormal% 20number%3F
On Wed, 2005-08-24 at 02:03 +0200, Jerome Tuncer wrote:
Hi
It's kind of a beginner question but I'll dare asking:
What is a denormal? (-:
Jé
shift8 a écrit :
hi all - i'm having really bad denormal problems on a intel celeron 2.4ghz. pd is from cvs aug 8th. looked around and found PD_BADFLOAT and PD_BIGORSMALL. on a lark i defined them and rebuilt, and even though it solved it for a while, but no - i'm still having the same problem. looked around some more and found those only appear in m_pd.h... anyone know what i can do to fix this?
thanks... the pain! the pain!
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-
Bryan Jurish -
Jerome Tuncer -
Miller Puckette -
shift8 -
Tim Blechmann