Re: [PD] Fwd: Variable number of objects?
Le 2011-10-01 à 04:14:00, Ingo a écrit : I wonder what kind of ears it takes to listen to something so complex... rather, what kind of brain lobes it takes. It takes a regular pair of ears - one on the left side and one on the right side! per head ? how many heads do you have ? e.g. quadraphonic : http://upload.wikimedia.org/wikipedia/en/7/72/Mark_Wing-Davey_as_Zaphod_Beeblebrox.jpg __ | Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Actually, I do have a twin brother. I could almost compete with the guy on the picture. Since I am using 3 stereo outs I could maybe top that with around something like six ears. I'll have to see how quickly I can clone myself! I'm not sure if Pd supports cloning of the listener with the current version? Maybe with an abstraction like: [dac~] | \ [ear~ $1] [ear~ $2] Then do some dynamic patching? Ingo -Ursprüngliche Nachricht- Von: Mathieu Bouchard [mailto:ma...@artengine.ca] Gesendet: Montag, 3. Oktober 2011 19:38 An: Ingo Cc: 'Pd List' Betreff: Re: AW: [PD] Fwd: Variable number of objects? Le 2011-10-01 à 04:14:00, Ingo a écrit : I wonder what kind of ears it takes to listen to something so complex... rather, what kind of brain lobes it takes. It takes a regular pair of ears - one on the left side and one on the right side! per head ? how many heads do you have ? e.g. quadraphonic : http://upload.wikimedia.org/wikipedia/en/7/72/Mark_Wing- Davey_as_Zaphod_Beeblebrox.jpg __ | Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Ok, it looks like I was misunderstanding the way how the [send] / [receive] is working. But then I am still wondering why I got a lot of performance boost after replacing the [send] / [receive] with wired connections? Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von IOhannes m zmölnig Gesendet: Samstag, 1. Oktober 2011 18:18 An: pd-list@iem.at Betreff: Re: [PD] Fwd: Variable number of objects? -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/01/2011 04:48 AM, Ingo wrote: Every [receive] will have to check if any [send] message is meant to be for this particular [receive]. It will have to check if the header of the [send] matches the header of the [receive] until the first character doesn't match anymore. Then it will abort checking and the next [receive] will start checking, and so on ... I can't see how this can be done without taxing the cpu. this is not how send/receive work in Pd. in general, Pd's messaging system works in a push manner, where data is pushed from one object to the next, rather than a pull manner, where an object requests a message from the previous one. therefore, [receive] need not care which [send]s are attached to it. then, [send] need not search for attached [receive]s either, because the send-symbol will maintain a linked list of all attached receivers. going through the linked list for dispatching a message is quite fast. gfdstm IOhannes -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAk6HPTEACgkQkX2Xpv6ydvQ8bQCfStNUi4fxyCOe2ZK3uvHtN7BG p+oAoNqIIRG/oaeeD7Qjoi2mmgkNXcZV =Chc9 -END PGP SIGNATURE- ___ 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
Re: [PD] Fwd: Variable number of objects?
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/01/2011 04:48 AM, Ingo wrote: Every [receive] will have to check if any [send] message is meant to be for this particular [receive]. It will have to check if the header of the [send] matches the header of the [receive] until the first character doesn't match anymore. Then it will abort checking and the next [receive] will start checking, and so on ... I can't see how this can be done without taxing the cpu. this is not how send/receive work in Pd. in general, Pd's messaging system works in a push manner, where data is pushed from one object to the next, rather than a pull manner, where an object requests a message from the previous one. therefore, [receive] need not care which [send]s are attached to it. then, [send] need not search for attached [receive]s either, because the send-symbol will maintain a linked list of all attached receivers. going through the linked list for dispatching a message is quite fast. gfdstm IOhannes -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAk6HPTEACgkQkX2Xpv6ydvQ8bQCfStNUi4fxyCOe2ZK3uvHtN7BG p+oAoNqIIRG/oaeeD7Qjoi2mmgkNXcZV =Chc9 -END PGP SIGNATURE- ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
I actually do switch off everything possible with a spigot but the [receive]s do still need to check if the [send] message is meant to be for them or not. So once you get too many [receive] objects while sending a lot it CAN slow down the patch quite a bit. But unfortunately that only starts showing up once the patch is so big that it takes forever to replace all of the [receive] objects with wired connections. So for now I'm trying to use wires wherever possible to address data only to the object that needs the data rather than having ten thousands of objects checking hundreds of times per second if the data is meant to be for them or not. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Freitag, 30. September 2011 05:04 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? I see... What I do is put a spigot right after all receives and the spigot is controlled by the same messages that control switch~: [r foo] | [spigot 0] | ... In this way you'll at least stop using all lines and metros and the like. I am not entirely sure that having a receive immediately connected to a [spigot 0] has any computational expense, but if I'm measuring things correctly they don't. So no need to shut off receives, just send them to a closed gate best, J On Thu, Sep 29, 2011 at 10:30 PM, Ingo i...@miamiwave.com wrote: Why would you have control messages going if switch~ is off? Because the voice gets assigned to a certain midi channel when it receives a [noteon] and has to keep receiving all midi controllers on that channel until the envelope release has finished. Then the next voice will play on that same midi channel. There is nothing that cuts off the control inlets or [send/receive]s automatically because the audio gets switched off. So when you play 16 notes in a row all 16 voices are set up to receive control data on that particular midi channel. Everything in the control domain, like LFOs, [metro]s and [line]s keep running and keep calculating pitch, volume, filter offsets and so on ... Turning off the [receive]s would be very complicated if not impossible which is why they need to be avoided wherever it can be done. But all of the wired inlets can be cut off manually together with the [switch~] and turned back on when the next note will play that voice. On top of it all 500 parameters need to be updated to the current state of the external control input and the current preset data when played anew. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Donnerstag, 29. September 2011 19:56 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Thu, Sep 29, 2011 at 1:41 PM, Ingo i...@miamiwave.com wrote: One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. Why would you have control messages going if switch~ is off? J Ingo - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19
Re: [PD] Fwd: Variable number of objects?
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 09/30/2011 05:03 AM, Jaime Oliver wrote: I see... What I do is put a spigot right after all receives and the spigot is controlled by the same messages that control switch~: [r foo] | [spigot 0] | ... In this way you'll at least stop using all lines and metros and the like. I am not entirely sure that having a receive immediately connected to a [spigot 0] has any computational expense, it sure has. nevertheless, i usually can be neglected, so i would suggest the same to save some CPU power down the patch. [line] can be become quite expensive when used massively, and a switched-off [line~] that is fed with new values can also make problems (at least it did a while ago; dunno whether this has been fixed) fgmasdr IOhannes -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAk6FqfoACgkQkX2Xpv6ydvRN6wCfXCWQAedpQOf6CPnpgf99enft jFAAn36dDJYbyTO7ZfIjHYQ40gLqsPB7 =Tu2Z -END PGP SIGNATURE- ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Le 2011-09-29 à 10:17:00, Jonathan Wilkes a écrit :
Since you have a large patch, I'd be curious to know how much memory is
taken up by having the switched off voices just sitting there.
It's not easy to know how much memory is taken by a bunch of small
mallocs. If you use getbytes() you have some idea of what's going on, but
it doesn't count the malloc accounting info that only malloc()/free()
usually know about, and it doesn't count the overallocation that malloc()
might do in order to make its life easier and make gradual realloc()
faster. Then it also doesn't count the overallocation of the memory zones
so that brk() or mmap() or mremap() doesn't have to be called as often,
which means that when you look at the process size in a process monitor
(ps or top or other) the difference you might see will not be accurate at
all... even if you manage to get it in bytes 4k-blocks instead of
megabytes.
So, if you really want to know the memory usage for a certain part of pd,
make sure that everything uses getbytes() instead of malloc()/new/etc, and
then try to find a formula that can tell you how much extra you should
count, and call it from getbytes() and freebytes() in order to keep
accurate statistics.
E.g. I recall that will the Perl Allocator that was being used by Perl in
the late 90's, you had to add sizeof(void*) to your byte request, and then
round to the next power of two. From this info you can then write :
size_t real_size (size_t n) {return 1highest_bit(n+sizeof(void *));}
highest_bit is a function that finds the position of the highest «1» bit
of an int. I have such a function in gridflow.h :
#define Z(N) if ((xN)(((typeof(x))1N)-1)) { x=N; i+=N; }
static int highest_bit(uint32 x) {int i=0; Z(16)Z(8)Z(4)Z(2)Z(1)return i;}
#undef Z
__
| Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___
Pd-list@iem.at mailing list
UNSUBSCRIBE and account-management -
http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Le 2011-09-29 à 19:41:00, Ingo a écrit : One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. I wonder what kind of ears it takes to listen to something so complex... rather, what kind of brain lobes it takes. __ | Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Le 2011-09-30 à 12:01:00, Mathieu Bouchard a écrit :
size_t real_size (size_t n) {return 1highest_bit(n+sizeof(void *));}
erratum :
size_t real_size (size_t n) {return 2highest_bit(n+sizeof(void *)-1);}
because the first version I wrote gets the PREVIOUS-or-equal power, which
is nonsense. Doubling that gives the NEXT power, but not next-or-equal, so
I subtract 1 to compensate for that.
BTW, I haven't looked at modern glibc mallocs. I suspect it might do
things similar to the Perl Allocator nowadays, though I have not checked,
and it's probably not documented (?). Chances are that it's a different
formula, anyway.
__
| Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___
Pd-list@iem.at mailing list
UNSUBSCRIBE and account-management -
http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
If you're going to wire them why not just create specific send messages? Give every abstraction an index and send only to that one: [r $1-foo] | etc J On Sep 30, 2011, at 4:13 AM, Ingo i...@miamiwave.com wrote: I actually do switch off everything possible with a spigot but the [receive]s do still need to check if the [send] message is meant to be for them or not. So once you get too many [receive] objects while sending a lot it CAN slow down the patch quite a bit. But unfortunately that only starts showing up once the patch is so big that it takes forever to replace all of the [receive] objects with wired connections. So for now I'm trying to use wires wherever possible to address data only to the object that needs the data rather than having ten thousands of objects checking hundreds of times per second if the data is meant to be for them or not. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Freitag, 30. September 2011 05:04 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? I see... What I do is put a spigot right after all receives and the spigot is controlled by the same messages that control switch~: [r foo] | [spigot 0] | ... In this way you'll at least stop using all lines and metros and the like. I am not entirely sure that having a receive immediately connected to a [spigot 0] has any computational expense, but if I'm measuring things correctly they don't. So no need to shut off receives, just send them to a closed gate best, J On Thu, Sep 29, 2011 at 10:30 PM, Ingo i...@miamiwave.com wrote: Why would you have control messages going if switch~ is off? Because the voice gets assigned to a certain midi channel when it receives a [noteon] and has to keep receiving all midi controllers on that channel until the envelope release has finished. Then the next voice will play on that same midi channel. There is nothing that cuts off the control inlets or [send/receive]s automatically because the audio gets switched off. So when you play 16 notes in a row all 16 voices are set up to receive control data on that particular midi channel. Everything in the control domain, like LFOs, [metro]s and [line]s keep running and keep calculating pitch, volume, filter offsets and so on ... Turning off the [receive]s would be very complicated if not impossible which is why they need to be avoided wherever it can be done. But all of the wired inlets can be cut off manually together with the [switch~] and turned back on when the next note will play that voice. On top of it all 500 parameters need to be updated to the current state of the external control input and the current preset data when played anew. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Donnerstag, 29. September 2011 19:56 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Thu, Sep 29, 2011 at 1:41 PM, Ingo i...@miamiwave.com wrote: One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. Why would you have control messages going if switch~ is off? J Ingo - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd
Re: [PD] Fwd: Variable number of objects?
Le 2011-09-30 à 12:32:00, Jaime Oliver a écrit : If you're going to wire them why not just create specific send messages? Yes, that's much faster. But keep in mind that making large amounts of symbols makes gensym become gradually slower, which slows down interpreting $1-foo in all messageboxes, for example. The problem appears gradually as you have several thousand symbols at once, but I have not benchmarked to see how much it changes the balance of things. I think that send/receive will be almost always faster than anything else. [route] is of intermediate speed, faster than [spigot] by a constant ratio for an average of many different lookups. [route] is faster than [spigot] for high indices (right-side outlets), and even faster for lower indices (left-side outlets). [s]/[r] is fast like one of the smallest indices of [route]. __ | Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
I wonder what kind of ears it takes to listen to something so complex... rather, what kind of brain lobes it takes. It takes a regular pair of ears - one on the left side and one on the right side! Ingo ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
If you're going to wire them why not just create specific send messages? Give every abstraction an index and send only to that one: [r $1-foo] | etc J Every [receive] will have to check if any [send] message is meant to be for this particular [receive]. It will have to check if the header of the [send] matches the header of the [receive] until the first character doesn't match anymore. Then it will abort checking and the next [receive] will start checking, and so on ... I can't see how this can be done without taxing the cpu. Having several hundred of messages being sent per second going to several hundred [receives] multiplied by the voice number will add up to many millions of these checks per second. After a certain amount of objects and input data this definitely takes too much time for realtime low latency playing when using high voice counts. It may not affect anything until the number of [send/receive] exceeds a certain number. Getting rid of the [send/receive]s at this point takes a ridiculous amount of time (I'm still not done after several months but getting much better results already). Using dollar arguments only adds a number to the [receive] and doesn't keep it from having to do the checking. That's the reason why I try to avoid [send/receive] objects wherever realtime playing is involved. I still use them, but only for non realtime editing purposes. But there is still a tendency for audio dropouts. Ingo On Sep 30, 2011, at 4:13 AM, Ingo i...@miamiwave.com wrote: I actually do switch off everything possible with a spigot but the [receive]s do still need to check if the [send] message is meant to be for them or not. So once you get too many [receive] objects while sending a lot it CAN slow down the patch quite a bit. But unfortunately that only starts showing up once the patch is so big that it takes forever to replace all of the [receive] objects with wired connections. So for now I'm trying to use wires wherever possible to address data only to the object that needs the data rather than having ten thousands of objects checking hundreds of times per second if the data is meant to be for them or not. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Freitag, 30. September 2011 05:04 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? I see... What I do is put a spigot right after all receives and the spigot is controlled by the same messages that control switch~: [r foo] | [spigot 0] | ... In this way you'll at least stop using all lines and metros and the like. I am not entirely sure that having a receive immediately connected to a [spigot 0] has any computational expense, but if I'm measuring things correctly they don't. So no need to shut off receives, just send them to a closed gate best, J On Thu, Sep 29, 2011 at 10:30 PM, Ingo i...@miamiwave.com wrote: Why would you have control messages going if switch~ is off? Because the voice gets assigned to a certain midi channel when it receives a [noteon] and has to keep receiving all midi controllers on that channel until the envelope release has finished. Then the next voice will play on that same midi channel. There is nothing that cuts off the control inlets or [send/receive]s automatically because the audio gets switched off. So when you play 16 notes in a row all 16 voices are set up to receive control data on that particular midi channel. Everything in the control domain, like LFOs, [metro]s and [line]s keep running and keep calculating pitch, volume, filter offsets and so on ... Turning off the [receive]s would be very complicated if not impossible which is why they need to be avoided wherever it can be done. But all of the wired inlets can be cut off manually together with the [switch~] and turned back on when the next note will play that voice. On top of it all 500 parameters need to be updated to the current state of the external control input and the current preset data when played anew. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Donnerstag, 29. September 2011 19:56 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Thu, Sep 29, 2011 at 1:41 PM, Ingo i...@miamiwave.com wrote: One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. Why would you have control messages going if switch~ is off? J Ingo - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com
Re: [PD] Fwd: Variable number of objects?
On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list #N canvas 609 0 565 676 10; #X text 193 513 pos left; #X text 251 513 pos top; #X obj 244 568 pack f f f; #X msg 117 554 selectall; #X msg 87 554 cut; #X msg 47 554 paste; #X obj 30 608 s reset_system_delay; #X obj 262 243 f; #X obj 294 243 + 1; #X obj 228 280 sel; #X obj 246 228 bng 15 250 50 0 empty empty empty 17 7 0 10 -262144 -1 -1; #X msg 228 307 0; #X obj 160 307 spigot; #X obj 262 435 t b f f; #X obj 272 462 * 20; #X msg 406 372 clear; #X obj 30 527 t b b b b; #X obj 193 124 f; #X text 305 513 argument (voice number); #X text 277 108 set number of voices; #X obj 298 16 loadbang; #X obj 247 110 nbx 2 14 1 24 0 1 empty empty empty 0 -8 0 10 -261682 -1 -1 4 256; #X msg 214 531 30; #X obj 272 482 + 20; #N canvas 268 529 331 383 voicecanvas 1; #X restore 26 18 pd voicecanvas; #X obj 406 392 s pd-voicecanvas; #X obj 273 48 bng 15 250 50 0 empty empty add_voices -52 7 1 9 -257985 -1 -1; #X obj 47 581 s pd-voicecanvas; #X obj 244 608 s pd-voicecanvas; #X msg 244 588 obj \$1 \$2 samplevoice \$3; #X obj 30 507 pipe 100; #X text 20 279 pipe can be set faster; #X text 242 623 send to window-name; #X obj 298 63 t b b b; #X obj 337 244 nbx 2 14 0 16 0 0 empty empty empty 0 -8 0 10 -261682 -1 -1 0 256; #X obj 262 295 +; #X obj 289 402 s pd-voicecanvas; #X msg 289 382 obj 30 10 inlet; #X obj 262 315 t f f; #X obj 289 355 t b b; #X obj 289 335 sel 1; #X msg 262 201 0; #X obj 193 144 t b f f; #X obj 160 280 pipe 100; #X text 339 333 clear subpatch and create inlet; #X text 339 343 when starting from 1st voice; #X msg 443 212 set \$1; #X obj 346 477 f; #X text 371 476 set automatic voice offset; #X obj 406 21 bng 15 250 50 0 empty empty reset 17 7 0 10 -258113 -1 -1; #X msg 406 39 0; #X text 367 242 automatic voice number
Re: [PD] Fwd: Variable number of objects?
I just added the [; pd dsp 0( when starting to creat voices to speed it up and eliminated the 17 voices limit of the patch. Maybe it's useful for somebody. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Ingo Gesendet: Donnerstag, 29. September 2011 11:33 An: 'Roman Haefeli'; 'Ludwig Maes' Cc: 'Pd List' Betreff: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list #N canvas 609 0 565 681 10; #X text 193 513 pos left; #X text 251 513 pos top; #X obj 244 568 pack f f f; #X msg 131 554 selectall; #X msg 101 554 cut; #X msg 61 554 paste; #X obj 45 608 s reset_system_delay; #X obj 260 223 f; #X obj 292 223 + 1; #X obj 228 280 sel; #X obj 244 208 bng 15 250 50 0 empty empty empty 17 7 0 10 -262144 -1 -1; #X msg 228 307 0; #X obj 160 307 spigot; #X obj 262 435 t b f f; #X obj 272 462 * 20; #X msg 406 372 clear; #X obj 174 124 f; #X text 305 513 argument (voice number); #X text 269 108 set number of voices; #X obj 308 16 loadbang; #X obj 238 110 nbx 2 14 1 24 0 1 empty empty empty 0 -8 0 10 -261682 -1 -1 4 256; #X msg 214 531 30; #X obj 272 482 + 20; #N canvas 268 529 331 383 voicecanvas 1; #X restore 26 18 pd voicecanvas; #X obj 406 392 s pd-voicecanvas; #X obj 283 48 bng 15 250 50 0 empty empty add_voices -52 7 1 9 -257985 -1 -1; #X obj 61 581 s pd-voicecanvas; #X obj 244 608 s pd-voicecanvas; #X msg 244 588 obj \$1 \$2 samplevoice \$3; #X obj 30 507 pipe 100; #X text 20 279 pipe can be set faster; #X text 242 623 send to window-name; #X obj 342 244 nbx 2 14 0 1e+037 0 0 empty empty empty 0 -8 0 10 -261682 -1 -1 0 256; #X obj 262 295 +; #X obj 289 402 s pd-voicecanvas; #X msg
Re: [PD] Fwd: Variable number of objects?
I made some more changes and added some help information to the voice creation patch so you can simple use a float to add voices and 0 to clear all voices. There are wired inlets for the voices now. Hope it's helpful for anybody! Ingo -Ursprüngliche Nachricht- Von: Ingo [mailto:i...@miamiwave.com] Gesendet: Donnerstag, 29. September 2011 12:02 An: 'Ingo'; 'Roman Haefeli'; 'Ludwig Maes' Cc: 'Pd List' Betreff: AW: [PD] Fwd: Variable number of objects? I just added the [; pd dsp 0( when starting to creat voices to speed it up and eliminated the 17 voices limit of the patch. Maybe it's useful for somebody. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Ingo Gesendet: Donnerstag, 29. September 2011 11:33 An: 'Roman Haefeli'; 'Ludwig Maes' Cc: 'Pd List' Betreff: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list #N canvas 607 0 578 744 10; #X obj 26 21 inlet; #N canvas 342 529 257 383 voicecanvas 1; #X restore 26 58 pd voicecanvas; #X text 254 559 pos left; #X text 312 559 pos top; #X obj 297 623 pack f f f; #X msg 112 562 selectall; #X msg 82 562 cut; #X obj 39 616 s reset_system_delay; #X obj 240 348 f; #X obj 272 348 + 1; #X obj 196 395 sel; #X obj 220 333 bng 15 250 50 0 empty empty empty 17 7 0 10 -262144 -1 -1; #X msg 196 422 0; #X obj 98 422 spigot; #X obj 315 576 * 20; #X msg 459 429 clear; #X obj 315 596 + 20; #X obj 459 449 s pd-voicecanvas; #X obj 82 589 s pd-voicecanvas; #X obj 297 663 s pd-voicecanvas; #X text 295 678 send
Re: [PD] Fwd: Variable number of objects?
What happens if you just have the maximum number of voices you think you'd ever need and [switch~] them on and off as needed? Since you have a large patch, I'd be curious to know how much memory is taken up by having the switched off voices just sitting there. -Jonathan - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ 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 ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
On Thu, Sep 29, 2011 at 1:17 PM, Jonathan Wilkes jancs...@yahoo.com wrote: What happens if you just have the maximum number of voices you think you'd ever need and [switch~] them on and off as needed? In my case it takes a bit long to load the patch, but if the voices are switched off then they don't hurt cpu usage at all. In Pd this is definitely the way to go unless you don't care about audio dropouts... J Since you have a large patch, I'd be curious to know how much memory is taken up by having the switched off voices just sitting there. -Jonathan - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ 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 ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list -- Jaime E Oliver LR www.jaimeoliver.pe 858 750 0924 (cel) 858 202 1522 (home) ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Fwd: Variable number of objects?
What happens if you just have the maximum number of voices you think you'd ever need and [switch~] them on and off as needed? Since you have a large patch, I'd be curious to know how much memory is taken up by having the switched off voices just sitting there. -Jonathan That's what I am doing anyway. Memory is not an issue. There is obviously no change in memory by simply switching the voices on or off. After I got most of the control stuff as well as a large number of the [receive] objects out of the way the patch doesn't need that much cpu time at all unless the voices are turned on and playing. Now that the switched off voices are not hardly doing anything anymore there is no more need to adjust the voice number as it was the case before I got rid of a whole bunch of [receive] objects and cut most of the control messages when the [switch~] object gets turned off. In certain cases I can limit the voice number with different [poly] objects. One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. Ingo - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said, that often you can gain quite some time while turning off DSP during dynamic instantiation. But I guess, this makes only a difference when performing several instantiations at the same time. When DSP is on, each cycle would cause a complete DSP graph recompilation, whereas when DSP is off, it's only recompiled once (after turning it on again). Roman ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list ___ Pd-list@iem.at mailing list
Re: [PD] Fwd: Variable number of objects?
Why would you have control messages going if switch~ is off? Because the voice gets assigned to a certain midi channel when it receives a [noteon] and has to keep receiving all midi controllers on that channel until the envelope release has finished. Then the next voice will play on that same midi channel. There is nothing that cuts off the control inlets or [send/receive]s automatically because the audio gets switched off. So when you play 16 notes in a row all 16 voices are set up to receive control data on that particular midi channel. Everything in the control domain, like LFOs, [metro]s and [line]s keep running and keep calculating pitch, volume, filter offsets and so on ... Turning off the [receive]s would be very complicated if not impossible which is why they need to be avoided wherever it can be done. But all of the wired inlets can be cut off manually together with the [switch~] and turned back on when the next note will play that voice. On top of it all 500 parameters need to be updated to the current state of the external control input and the current preset data when played anew. Ingo -Ursprüngliche Nachricht- Von: Jaime Oliver [mailto:jaime.oliv...@gmail.com] Gesendet: Donnerstag, 29. September 2011 19:56 An: Ingo Cc: Jonathan Wilkes; Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Thu, Sep 29, 2011 at 1:41 PM, Ingo i...@miamiwave.com wrote: One shouldn't underestimate the cpu load when several hundreds of midi controllers per second are modulating hundreds of parameters and the get multiplied by 16 voices constantly because the control messages are still active all of the time. Why would you have control messages going if switch~ is off? J Ingo - Original Message - From: Ingo i...@miamiwave.com To: 'Roman Haefeli' reduz...@gmail.com; 'Ludwig Maes' ludwig.m...@gmail.com Cc: 'Pd List' pd-list@iem.at Sent: Thursday, September 29, 2011 5:33 AM Subject: Re: [PD] Fwd: Variable number of objects? Actually, I just tried again to simply copy a couple of voices and it only took a fraction of a second with a very short dropout - even with the dsp on. I have recently installed Natty instead of Lucid on a new machine. Maybe it has something to do with realloc that Mathieu mentioned. So it looks like dynamic patching of voices isn't that much of a problem here anymore. It still takes 7-8 seconds to create 16 voices in my case with the dsp off (12 with the dsp on) but that's still faster than restarting the patch. I would never have checked again if nobody would have mentioned it. I have attached a patch that I used for testing. These voices were receiving their input with [receive] so no connections were needed. If you are using wired inlets you will have to dynamically add the connections of course. I added a cut paste at the end because for some reasons the voices didn't get initialized correctly. Not sure if this is needed for other voice-abstractions. Ingo -Ursprüngliche Nachricht- Von: pd-list-boun...@iem.at [mailto:pd-list-boun...@iem.at] Im Auftrag von Roman Haefeli Gesendet: Donnerstag, 29. September 2011 08:36 An: Ludwig Maes Cc: Pd List Betreff: Re: [PD] Fwd: Variable number of objects? On Wed, 2011-09-28 at 19:29 +0200, Ludwig Maes wrote: -- Forwarded message -- From: Ludwig Maes ludwig.m...@gmail.com Date: 28 September 2011 19:29 Subject: Re: [PD] Variable number of objects? To: Ingo i...@miamiwave.com I actually meant more in general, also for non-~ signals (i.e. also control rate .pd patches). I referred to polysynth such that people would see more easily what I meant. Are there really no such primitives? That seems like quite a restriction... How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) also, how is it done even with the long delays? As I understand it, the whole DSP is recompiled whenever it is changed. So, if you have a very large patch (and Ingo seems to be an expert in very large patches) and you create another voice, it's easily possible to eat up quite some time. Also, it's probably much slower the first time, if the voice abstraction is built of many other abstractions, which need to be read from disk. But I guess you are right about the increase in CPU workload _after_ the DSP graph has been recompiled: A switch from 10 two 11 instances is expected to show only an increase of 10% in CPU usage. It's been said
Re: [PD] Fwd: Variable number of objects?
Le 2011-09-28 à 19:29:00, Ludwig Maes a écrit : How can that take 10 seconds?? I dont see what would cause such a huge overhead, i'd expect an increase in computations memory though (say from 10 voices to 11: 10% increase in cpu workload ram dedicated to these voices..., I fail to see what would necessitate a long initialization...) BTW, IIRC, PureData's DSP compiler just relies on realloc() being really fast all of the time. I don't know whether there's any version of libc that contains an older-style realloc() that could make turning DSP on really slow... I haven't seen one in a long time. If anyone can identify such a problem, I know how to fix it... it's not very hard. __ | Mathieu BOUCHARD - téléphone : +1.514.383.3801 - Montréal, QC___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
