[Csnd] Floats 'n' Doubles
Date | 2011-03-18 07:11 |
From | Jim Aikin |
Subject | [Csnd] Floats 'n' Doubles |
As part of a writing project (about which, more details will surface as time goes on), I'd like to document precisely what's going on with respect to installing the floats vs. doubles versions of Csound. Looking at the directory on SourceForge, it appears that the Windows build of 5.13 is available only in the doubles version. But what I've installed on my Mac is the floats build of 5.13, and I see no other option for an installer for MacOS. My first question is, is this a significant difference? Not only with respect to the functionality of Csound itself, but for example with respect to the compatibility of the Csound install with various front ends. I seem to recall that at some point a few months ago, QuteCsound would only run with one of them, but I don't remember the details, and that may be old information by now. Second, is this the way Csound is intended to be distributed from now on, with the Mac version differing in word size from the Windows version, or is it just an artifact based on how much (or rather, how little) time folks have had to build installers? My impression from some recent discussions is that the floats version may run a little faster, which might be significant in CPU-intensive real-time applications. But I recall also seeing one or two messages suggesting that such differences are minor. The CPU load may not be the only significant difference between the floats and doubles versions; some opcodes may be written in such a way as to fail under one version or the other . Appendix H in the manual states that 64-bit doubles precision can occasionally sound cleaner -- but Appendix H is written in such a way as to imply that users will have access to both the 32-bit and 64-bit versions, which doesn't seem to be the case at present. Appendix H also states that lpanal and cvanal are not compatible with Csound64. Is this incompatibility likely to be rectified in the future, or is it likely to remain? For purposes of this writing project, I would like to be able to present the picture in an authoritative way. Any insights into this topic would be much appreciated. Thanks! --Jim Aikin -- View this message in context: http://csound.1045644.n5.nabble.com/Floats-n-Doubles-tp3943283p3943283.html Sent from the Csound - General mailing list archive at Nabble.com. Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-18 08:56 |
From | Oeyvind Brandtsegg |
Subject | Re: [Csnd] Floats 'n' Doubles |
my 2 cents on this, The doubles version sounds better, and in some cases the extra precision is actually needed to get what you ask for. An example would be bitreduction, where I've experienced quantization errors with the floats version. I can probably dig out an example if needed. I am eager to get the most optimized realtime performance, this is very important, but still I would be willing to sacrifice the slight performance gain of the floats version in favor of high precision audio processing. The floats version may have to be supportet for some time, for compatibility reasons, other than that I would say we could go to doubles only. ... hoping not to start a war over this :-) best Oeyvind 2011/3/18 Jim Aikin |
Date | 2011-03-18 09:09 |
From | Victor Lazzarini |
Subject | Re: [Csnd] Floats 'n' Doubles |
The MacOS version comes with doubles and floats; which one you use, will depend on the frontend: the csound command uses floats; csound64 uses doubles; third-party frontends will use one or the other (blue uses doubles; qutecsound comes for floats, but there might be a doubles version also available; winXound can use either csound or csound64, you set it in the preferences; Cecilia most likely will use doubles, as the python module installed by default is the doubles one) Victor On 18 Mar 2011, at 07:11, Jim Aikin wrote: > But what I've installed on my Mac is > the floats build of 5.13, and I see no other option for an installer > for > MacOS. Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-18 09:18 |
From | peiman khosravi |
Subject | Re: [Csnd] Floats 'n' Doubles |
I have noticed that on snow-leopard with a 64bit machine using the double version does not really affect the performance speed much so a few days ago I uninstalled floats and am only using doubles version from now on. It does make a difference to sound quality (e.g. the ATS opcodes sound much better with the doubles version). On 18 March 2011 09:09, Victor Lazzarini |
Date | 2011-03-18 09:30 |
From | Victor Lazzarini |
Subject | Re: [Csnd] Floats 'n' Doubles |
That surprises me since the release csound is built for i386, not x86_64 (unless you are using your own custom build). However, reports are that doubles is indeed slightly faster on x86_64. Victor On 18 Mar 2011, at 09:18, peiman khosravi wrote: > I have noticed that on snow-leopard with a 64bit machine using the > double version does not really affect the performance speed much so a > few days ago I uninstalled floats and am only using doubles version > from now on. It does make a difference to sound quality (e.g. the ATS > opcodes sound much better with the doubles version). > > On 18 March 2011 09:09, Victor Lazzarini |
Date | 2011-03-18 09:39 |
From | Andres Cabrera |
Subject | Re: [Csnd] Floats 'n' Doubles |
Hi, The explanation for this is that if you use the API, you can only use either doubles or floats. So you can build QuteCsound (or blue, which is also an API user) to use one or the other, but there is currently no way to do both with a single binary. A program like WinXsound can use both because it calls the command line version of Csound, so it can be configured (as opposed to rebuilt) to run either, but it can't provide rich interaction through the API. I'm not sure how Michael builds QuteCsound for windows packaging. I think the latest version is built for doubles. Is that right? You can check this by running Csound inside QuteCsound (not on the command line), and one of the first lines should tell you if you are using doubles or floats. If you build QuteCsound from source, this information is in the help window. Cheers, Andres On Fri, Mar 18, 2011 at 9:09 AM, Victor Lazzarini |
Date | 2011-03-18 09:42 |
From | peiman khosravi |
Subject | Re: [Csnd] Floats 'n' Doubles |
No I am using the installer. I am not sure if it is faster but I can't notice any difference in performance (other than sound quality). On my older computer there is a huge difference and doubles is almost unusable in real-time in most cases. Perhaps it's related to some other parameter. Thanks Peiman On 18 March 2011 09:30, Victor Lazzarini |
Date | 2011-03-18 10:09 |
From | joachim heintz |
Subject | Re: [Csnd] Floats 'n' Doubles |
My impression is that Csound should move (in the next release?) to just distributing the doubles version. This would avoid some confusion in installing or in the combination of the csound engine with a frontend. I am wondering what might be problems doing it: - Some Python libs just as floats? - Problems with QuteCsound (I remember there was a problem with the scope widget running very slow with doubles - I don't know if this is still the case). - Others? joachim Am 18.03.2011 08:11, schrieb Jim Aikin: > As part of a writing project (about which, more details will surface as time > goes on), I'd like to document precisely what's going on with respect to > installing the floats vs. doubles versions of Csound. Looking at the > directory on SourceForge, it appears that the Windows build of 5.13 is > available only in the doubles version. But what I've installed on my Mac is > the floats build of 5.13, and I see no other option for an installer for > MacOS. > > My first question is, is this a significant difference? Not only with > respect to the functionality of Csound itself, but for example with respect > to the compatibility of the Csound install with various front ends. I seem > to recall that at some point a few months ago, QuteCsound would only run > with one of them, but I don't remember the details, and that may be old > information by now. > > Second, is this the way Csound is intended to be distributed from now on, > with the Mac version differing in word size from the Windows version, or is > it just an artifact based on how much (or rather, how little) time folks > have had to build installers? > > My impression from some recent discussions is that the floats version may > run a little faster, which might be significant in CPU-intensive real-time > applications. But I recall also seeing one or two messages suggesting that > such differences are minor. The CPU load may not be the only significant > difference between the floats and doubles versions; some opcodes may be > written in such a way as to fail under one version or the other . > > Appendix H in the manual states that 64-bit doubles precision can > occasionally sound cleaner -- but Appendix H is written in such a way as to > imply that users will have access to both the 32-bit and 64-bit versions, > which doesn't seem to be the case at present. Appendix H also states that > lpanal and cvanal are not compatible with Csound64. Is this incompatibility > likely to be rectified in the future, or is it likely to remain? > > For purposes of this writing project, I would like to be able to present the > picture in an authoritative way. Any insights into this topic would be much > appreciated. Thanks! > > --Jim Aikin > > -- > View this message in context: http://csound.1045644.n5.nabble.com/Floats-n-Doubles-tp3943283p3943283.html > Sent from the Csound - General mailing list archive at Nabble.com. > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" > > Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-18 11:52 |
From | Stéphane Rollandin |
Subject | Re: [Csnd] Floats 'n' Doubles |
> My impression is that Csound should move (in the next release?) to just > distributing the doubles version. This would avoid some confusion in > installing or in the combination of the csound engine with a frontend. +1 Stef Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-18 12:41 |
From | Michael Gogins |
Subject | Re: [Csnd] Floats 'n' Doubles |
I build both Csound and QuteCsound for double-precision samples only on Windows, for myself and for the Windows installers, and on Linux, for myself only. I do not use any of the utilities mentioned as possibly incompatible with 64 bit samples, so I don't know about that.\ I recommend, as I often have, that all builds on all platforms switch to 64 bit samples. Regards, Mike On Fri, Mar 18, 2011 at 5:39 AM, Andres Cabrera |
Date | 2011-03-18 13:04 |
From | Michael Gogins |
Subject | Re: [Csnd] Floats 'n' Doubles |
Thanks for the info about the ats opcodes, how do you know this? On Mar 18, 2011 5:19 AM, "peiman khosravi" <peimankhosravi@gmail.com> wrote:
> I have noticed that on snow-leopard with a 64bit machine using the > double version does not really affect the performance speed much so a > few days ago I uninstalled floats and am only using doubles version > from now on. It does make a difference to sound quality (e.g. the ATS > opcodes sound much better with the doubles version). > > On 18 March 2011 09:09, Victor Lazzarini <Victor.Lazzarini@nuim.ie> wrote: >> The MacOS version comes with doubles and floats; which one you use, will >> depend on the frontend: the csound command >> uses floats; csound64 uses doubles; third-party frontends will use one or >> the other (blue uses >> doubles; qutecsound comes for floats, but there might be a doubles version >> also available; winXound >> can use either csound or csound64, you set it in the preferences; Cecilia >> most likely will use doubles, as the >> python module installed by default is the doubles one) >> >> Victor >> On 18 Mar 2011, at 07:11, Jim Aikin wrote: >> >>> But what I've installed on my Mac is >>> the floats build of 5.13, and I see no other option for an installer for >>> MacOS. >> >> >> >> Send bugs reports to the Sourceforge bug tracker >> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >> Discussions of bugs and features can be posted here >> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe >> csound" >> >> > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" > |
Date | 2011-03-18 13:14 |
From | peiman khosravi |
Subject | Re: [Csnd] Floats 'n' Doubles |
Just from experience. I can't remember which one as I was using them a long time ago but one of the ATS opcodes used to output a rather glitchy sound (as if the amplitude interpolation was not smooth enough) with the float but not the doubles. Sorry I can't be more specific but I can try and dig out my old CSDs tomorrow. Best, Peiman On 18 March 2011 13:04, Michael Gogins |
Date | 2011-03-18 13:46 |
From | Francois PINOT |
Subject | Re: [Csnd] Floats 'n' Doubles |
+1 Francois 2011/3/18 Stéphane Rollandin <lecteur@zogotounga.net> My impression is that Csound should move (in the next release?) to just |
Date | 2011-03-18 15:37 |
From | Jim Aikin |
Subject | [Csnd] Re: Floats 'n' Doubles |
Thanks, everyone, for the informative replies. It appears I would be not too far off the mark in saying: 1) Csound is moving gradually toward being a doubles-only app, so in the future users probably won't need to worry about this factor. 2) Ultimately the version you'll install if you run an installer will depend on what the person who created the installer built. 3) Currently the Windows version (including QuteCsound) is doubles-only. The MacOS installer gives you both doubles and floats versions, and QuteCsound for Mac uses floats. 4) It's possible that a few things (such as those mentioned in Appendix H, and possibly the QuteCsound scope widget) will need to be updated in order to work in a doubles-only world. With respect to item 3, what about Linux? I don't have a Linux computer, so I have no easy way to look at this question. Do most Linux users build Csound themselves? Or are there Csound installers for the popular Linux versions? --JA -- View this message in context: http://csound.1045644.n5.nabble.com/Floats-n-Doubles-tp3943283p3967953.html Sent from the Csound - General mailing list archive at Nabble.com. Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-18 15:40 |
From | Rory Walsh |
Subject | Re: [Csnd] Re: Floats 'n' Doubles |
You can apt-get for Csound for debian based Linux distro, I think it's currently at 5.12? I'd say that most Linux users build Csound themselves however? On 18 March 2011 15:37, Jim Aikin |
Date | 2011-03-18 16:31 |
From | Andres Cabrera |
Subject | Re: [Csnd] Re: Floats 'n' Doubles |
Hi, I think the Debian version is doubles and the Fedora version is floats. I would think that many people build their own Csound, but building Csound can be tricky for many Linux users, so I would think many stick to the available packages for their distribution. Cheers, Andres On Fri, Mar 18, 2011 at 3:40 PM, Rory Walsh |
Date | 2011-03-18 16:46 |
From | joachim heintz |
Subject | Re: [Csnd] Re: Floats 'n' Doubles |
hi andrés - is it still (or has it ever been) a problem to use qutecsound-doubles because of the graph widget? best - joachim Am 18.03.2011 17:31, schrieb Andres Cabrera: > Hi, > > I think the Debian version is doubles and the Fedora version is > floats. I would think that many people build their own Csound, but > building Csound can be tricky for many Linux users, so I would think > many stick to the available packages for their distribution. > > Cheers, > Andres > > On Fri, Mar 18, 2011 at 3:40 PM, Rory Walsh |
Date | 2011-03-18 17:43 |
From | Victor Lazzarini |
Subject | Re: [Csnd] Re: Floats 'n' Doubles |
Given that floats and doubles are perfectly fine together, I do not think I am going to be dropping the float framework anytime soon. I do not want people to update their csound installation only to see that some frontend they like does not work anymore. However, what I might do for next version is link the csound command to the doubles framework and provide another command, csound32 to use the floats framework. Note that by default the installed modules for Python and Java are now linked to the doubles framework. So the move, as far as OSX is concerned, is to make doubles the default, but still provide access to floats. Note that I am trying to do this as transparently as possible for the Mac user. S/he does not need to know any of this, if s/he does not want to. Victor On 18 Mar 2011, at 15:37, Jim Aikin wrote: > Thanks, everyone, for the informative replies. It appears I would be > not too > far off the mark in saying: > > 1) Csound is moving gradually toward being a doubles-only app, so in > the > future users probably won't need to worry about this factor. > > 2) Ultimately the version you'll install if you run an installer > will depend > on what the person who created the installer built. > > 3) Currently the Windows version (including QuteCsound) is doubles- > only. The > MacOS installer gives you both doubles and floats versions, and > QuteCsound > for Mac uses floats. > > 4) It's possible that a few things (such as those mentioned in > Appendix H, > and possibly the QuteCsound scope widget) will need to be updated in > order > to work in a doubles-only world. > > With respect to item 3, what about Linux? I don't have a Linux > computer, so > I have no easy way to look at this question. Do most Linux users build > Csound themselves? Or are there Csound installers for the popular > Linux > versions? > > --JA > > -- > View this message in context: http://csound.1045644.n5.nabble.com/Floats-n-Doubles-tp3943283p3967953.html > Sent from the Csound - General mailing list archive at Nabble.com. > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body > "unsubscribe csound" > Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |
Date | 2011-03-19 10:05 |
From | joachim heintz |
Subject | Re: [Csnd] Re: Floats 'n' Doubles |
> However, what I might do for next version is link the csound command to > the doubles > framework and provide another command, csound32 to use the floats > framework. I think this is an excellent idea. Perhaps there can be an advice about how users who need to revert to the floats version can achieve this (how to call csound32 instead of csound). joachim Send bugs reports to the Sourceforge bug tracker https://sourceforge.net/tracker/?group_id=81968&atid=564599 Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" |