RE: DDraw Blt vs BltFast
> > On Tue, Mar 12, 2002 at 09:30:05AM +0100, Ralf > Habacker wrote: > > I have done some analysing work with this and with the > > cygwin daemon (cygserver transport classes) > there may be a > > way in the future to implement unix domain sockets with > > named pipes which speed up unix domain sockets up to 250 > > MB/s, as I have measured with a quick an dirty sample > > Fine. But how do you implement them on 9x/Me? > If 9x/Me does not support named pipes, why not using the currrent tcp/ip code for this ? Ralf
Re: DDraw Blt vs BltFast
Do you have to ? You could allow more functional platform in the source tree to have this new functionality while leaving the DOS based Windows at the current support. Yes - this will complicat matters and will probably need some kind of abstraction layer in the source to 'hide' the difference for most developer, but the gain for the more advanced platforms is certainly worth it >On Tue, Mar 12, 2002 at 09:30:05AM +0100, Ralf Habacker wrote: >> I have done some analysing work with this and with the >> cygwin daemon (cygserver transport classes) there may be a >> way in the future to implement unix domain sockets with >> named pipes which speed up unix domain sockets up to 250 >> MB/s, as I have measured with a quick an dirty sample > >Fine. But how do you implement them on 9x/Me? > >Corinna Med venlig hilsen / Regards Franz Wolfhagen
Re: DDraw Blt vs BltFast
On Wed, Mar 13, 2002 at 03:14:53PM +1100, Robert Collins wrote: >>I'm glad it sounds logical. Cygwin already has lots of code that is NT >>specific, e.g., CYGWIN=ntsec. > >I agree. I got the impression from Corinna's email that what Ralf was >suggesting needed a 9x equivalent to be seriously considered. I'm not >100% sure why I got that impression though Ok. The horse is starting to decompose. I'm not going to whack it anymore. This is all actually off-topic for this mailing list anyway. I assume that Suhaib and Harold are just being too polite to inform us of that fact. cgf
RE: DDraw Blt vs BltFast
> -Original Message- > From: Christopher Faylor [mailto:[EMAIL PROTECTED]] > Sent: Wednesday, March 13, 2002 12:52 PM > To: [EMAIL PROTECTED] > Subject: Re: DDraw Blt vs BltFast > > > On Tue, Mar 12, 2002 at 06:06:16PM -0500, Harold Hunt wrote: > >> > Fine. But how do you implement them on 9x/Me? > >>Perhaps the same way I implemented FIFO's, but easier as > the semantics > >>seem less complex to me. > >> > >>Anyway, sure we don't need to have bot NT and 9x all at the same > >>time... > > > >Sounds logical to me. There is no reason to limit the > performance on > >Windows NT/2000/XP just because there would need to be a seperate > >routine for Windows 95/98/Me. > > I'm glad it sounds logical. Cygwin already has lots of code > that is NT specific, e.g., CYGWIN=ntsec. I agree. I got the impression from Corinna's email that what Ralf was suggesting needed a 9x equivalent to be seriously considered. I'm not 100% sure why I got that impression though Rob
Re: DDraw Blt vs BltFast
On Tue, Mar 12, 2002 at 06:06:16PM -0500, Harold Hunt wrote: >> > Fine. But how do you implement them on 9x/Me? >>Perhaps the same way I implemented FIFO's, but easier as the semantics >>seem less complex to me. >> >>Anyway, sure we don't need to have bot NT and 9x all at the same >>time... > >Sounds logical to me. There is no reason to limit the performance on >Windows NT/2000/XP just because there would need to be a seperate >routine for Windows 95/98/Me. I'm glad it sounds logical. Cygwin already has lots of code that is NT specific, e.g., CYGWIN=ntsec. cgf
RE: DDraw Blt vs BltFast
> > Fine. But how do you implement them on 9x/Me? > > Perhaps the same way I implemented FIFO's, but easier as the semantics > seem less complex to me. > > Anyway, sure we don't need to have bot NT and 9x all at the same time... Sounds logical to me. There is no reason to limit the performance on Windows NT/2000/XP just because there would need to be a seperate routine for Windows 95/98/Me. Harold
RE: DDraw Blt vs BltFast
> -Original Message- > From: Corinna Vinschen [mailto:[EMAIL PROTECTED]] > Sent: Wednesday, March 13, 2002 5:06 AM > To: Cygwin-Xfree > Subject: Re: DDraw Blt vs BltFast > > > On Tue, Mar 12, 2002 at 09:30:05AM +0100, Ralf Habacker wrote: > > I have done some analysing work with this and with the > > cygwin daemon (cygserver transport classes) there may be a > way in the > > future to implement unix domain sockets with named pipes > which speed > > up unix domain sockets up to 250 MB/s, as I have measured > with a quick > > an dirty sample > > Fine. But how do you implement them on 9x/Me? Perhaps the same way I implemented FIFO's, but easier as the semantics seem less complex to me. Anyway, sure we don't need to have bot NT and 9x all at the same time... Rob
RE: DDraw Blt vs BltFast
> -Original Message- > From: Harold Hunt [mailto:[EMAIL PROTECTED]] > Sent: Wednesday, March 13, 2002 4:46 AM > To: cygx > Subject: RE: DDraw Blt vs BltFast > > > Ralf, > > Those are some very interesting results. I especially like > the ones where Cygwin is 10 to 1000 times slower than Linux > :) Oh well... I can't bitch anymore for fear of obligating > myself to contribute to Cygwin proper. > Lol! Seriously though, I'm quite excited by what Ralf has been working on, and the cygserver is in HEAD now, so cygwin 1.11 will have it, and Ralf's patches don't need to be against a development branch. Rob
RE: DDraw Blt vs BltFast
> Do you have even a tiny bit of information about these "certain > functions" that you'd like to impart? Or is this an even higher level > version of the "I don't see why they don't..." > > I'm really surprised to see this attitude. It's quite disappointing. Dick-head mode detected... aborting conversation. You should be surprised to see that attitude because you are completely misinterpreting what I have said. Just forget it. Harold
Re: DDraw Blt vs BltFast
On Tue, Mar 12, 2002 at 01:43:11PM -0500, Harold Hunt wrote: >>>Those are some very interesting results. I especially like the ones >>>where Cygwin is 10 to 1000 times slower than Linux :) Oh well... I >>>can't bitch anymore for fear of obligating myself to contribute to >>>Cygwin proper. >> >>Hopefully, it stupefyingly obvious why cygwin is slower than linux. > >Yes, but it is also stupefyingly obvious that with a lot of hard work >certain functions can be sped up, like the speed up done with certain >types of fork calls. Do you have even a tiny bit of information about these "certain functions" that you'd like to impart? Or is this an even higher level version of the "I don't see why they don't..." I'm really surprised to see this attitude. It's quite disappointing. cgf
RE: DDraw Blt vs BltFast
> >Those are some very interesting results. I especially like the ones > >where Cygwin is 10 to 1000 times slower than Linux :) Oh well... I > >can't bitch anymore for fear of obligating myself to contribute to > >Cygwin proper. > > Hopefully, it stupefyingly obvious why cygwin is slower than linux. Yes, but it is also stupefyingly obvious that with a lot of hard work certain functions can be sped up, like the speed up done with certain types of fork calls. Thus my statement that I had better shut up before someone suggests that I spend a lot of time trying to speed things up. Harold
Re: DDraw Blt vs BltFast
On Tue, Mar 12, 2002 at 12:46:01PM -0500, Harold Hunt wrote: >Those are some very interesting results. I especially like the ones >where Cygwin is 10 to 1000 times slower than Linux :) Oh well... I >can't bitch anymore for fear of obligating myself to contribute to >Cygwin proper. Hopefully, it stupefyingly obvious why cygwin is slower than linux. cgf
Re: DDraw Blt vs BltFast
On Tue, Mar 12, 2002 at 09:30:05AM +0100, Ralf Habacker wrote: > I have done some analysing work with this and with the > cygwin daemon (cygserver transport classes) there may be a > way in the future to implement unix domain sockets with > named pipes which speed up unix domain sockets up to 250 > MB/s, as I have measured with a quick an dirty sample Fine. But how do you implement them on 9x/Me? Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Developermailto:[EMAIL PROTECTED] Red Hat, Inc.
RE: DDraw Blt vs BltFast
Ralf, Those are some very interesting results. I especially like the ones where Cygwin is 10 to 1000 times slower than Linux :) Oh well... I can't bitch anymore for fear of obligating myself to contribute to Cygwin proper. Harold > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED]]On Behalf Of Ralf Habacker > Sent: Tuesday, March 12, 2002 3:30 AM > To: Cygwin-Xfree > Subject: RE: DDraw Blt vs BltFast > > > > > > On Mon, 11 Mar 2002, Harold Hunt wrote: > > > > > Alexander, > > > > > > Because the following article: > > > > > > > > http://msdn.microsoft.com/library/default.asp?url= > > /library/en-us/wcegmm/htm/ > > > ddover_51.asp > > > > > > states that: > > > > > > "BltFast cannot be used on surfaces that have > > an attached clipper." > > > > Ok. > > > > > We use a clipper to account for any other > > Windows window overlaying our > > > window, which would cause a portion of our blit > > destiation to be > > > undrawable... which would mean that we wouldn't > > be allowed to use BltFast. > > > > > > Now... could we use BltFast in DirectDraw > > fullscreen modes? > > > > Not needed. The blitting does not slow down the > > server significantly. I > > think the cygwin socket layer slow much more. > > Do you have seen the lmbench results on the cygwin list > http://sources.redhat.com/ml/cygwin/2002-01/msg01719.html > This confirm your statement. > > *Local* Communication bandwidths in MB/s - bigger is better > --- > HostOS Pipe AFTCP File Mmap Bcopy > Bcopy Mem Mem > UNIX reread reread (libc) > (hand) read write > - - -- -- -- > -- - > BRAMSCHE CYGWIN_NT-5.0 93.9 17.6 40.7 335.0 477.7 145.0 > 134.0 477. 195.9 > BRAMSCHE Linux 2.2.18 343. 235. 64.4 177.7 238.5 71.5 > 61.4 238. 75.3 > > I have done some analysing work with this and with the > cygwin daemon (cygserver transport classes) there may be a > way in the future to implement unix domain sockets with > named pipes which speed up unix domain sockets up to 250 > MB/s, as I have measured with a quick an dirty sample > implementation. (Unfortunally currently I have no time to > work on) > > Another way seems to me in using local connection for the > xtrans layer, but there are two ioctl functions missing in > cygwin I_FDINSERT and I_RECVFD. > > I_FDINSERT > Creates a message from specified buffer(s), adds information > about another STREAM, and sends the message downstream. > > I_RECVFD > Retrieves the reference to an open file description from a > message written to a STREAMS-based pipe using the I_SENDFD > command, and allocates a new file descriptor in the calling > process that refers to this open file description. > > In any case there is some work to do. > > Regards > Ralf > >
RE: DDraw Blt vs BltFast
> > On Mon, 11 Mar 2002, Harold Hunt wrote: > > > Alexander, > > > > Because the following article: > > > > > http://msdn.microsoft.com/library/default.asp?url= > /library/en-us/wcegmm/htm/ > > ddover_51.asp > > > > states that: > > > > "BltFast cannot be used on surfaces that have > an attached clipper." > > Ok. > > > We use a clipper to account for any other > Windows window overlaying our > > window, which would cause a portion of our blit > destiation to be > > undrawable... which would mean that we wouldn't > be allowed to use BltFast. > > > > Now... could we use BltFast in DirectDraw > fullscreen modes? > > Not needed. The blitting does not slow down the > server significantly. I > think the cygwin socket layer slow much more. Do you have seen the lmbench results on the cygwin list http://sources.redhat.com/ml/cygwin/2002-01/msg01719.html This confirm your statement. *Local* Communication bandwidths in MB/s - bigger is better --- HostOS Pipe AFTCP File Mmap Bcopy Bcopy Mem Mem UNIX reread reread (libc) (hand) read write - - -- -- -- -- - BRAMSCHE CYGWIN_NT-5.0 93.9 17.6 40.7 335.0 477.7 145.0 134.0 477. 195.9 BRAMSCHE Linux 2.2.18 343. 235. 64.4 177.7 238.5 71.5 61.4 238. 75.3 I have done some analysing work with this and with the cygwin daemon (cygserver transport classes) there may be a way in the future to implement unix domain sockets with named pipes which speed up unix domain sockets up to 250 MB/s, as I have measured with a quick an dirty sample implementation. (Unfortunally currently I have no time to work on) Another way seems to me in using local connection for the xtrans layer, but there are two ioctl functions missing in cygwin I_FDINSERT and I_RECVFD. I_FDINSERT Creates a message from specified buffer(s), adds information about another STREAM, and sends the message downstream. I_RECVFD Retrieves the reference to an open file description from a message written to a STREAMS-based pipe using the I_SENDFD command, and allocates a new file descriptor in the calling process that refers to this open file description. In any case there is some work to do. Regards Ralf
RE: DDraw Blt vs BltFast
On Mon, 11 Mar 2002, Harold Hunt wrote: > Alexander, > > Because the following article: > > http://msdn.microsoft.com/library/default.asp?url=/library/en-us/wcegmm/htm/ > ddover_51.asp > > states that: > > "BltFast cannot be used on surfaces that have an attached clipper." Ok. > We use a clipper to account for any other Windows window overlaying our > window, which would cause a portion of our blit destiation to be > undrawable... which would mean that we wouldn't be allowed to use BltFast. > > Now... could we use BltFast in DirectDraw fullscreen modes? Not needed. The blitting does not slow down the server significantly. I think the cygwin socket layer slow much more. > But, you are welcome to try it... Hell no ;) It was just a thought. bye ago -- [EMAIL PROTECTED] http://www.gotti.org ICQ: 126018723 phone: +49 3725 349 80 80 mobile: +49 172 7854017 4. Chemnitzer Linux-Tag http://www.tu-chemnitz.de/linux/tag/lt4
RE: DDraw Blt vs BltFast
Alexander, Because the following article: http://msdn.microsoft.com/library/default.asp?url=/library/en-us/wcegmm/htm/ ddover_51.asp states that: "BltFast cannot be used on surfaces that have an attached clipper." We use a clipper to account for any other Windows window overlaying our window, which would cause a portion of our blit destiation to be undrawable... which would mean that we wouldn't be allowed to use BltFast. Now... could we use BltFast in DirectDraw fullscreen modes? Probably... but some coding would have to be done that prevents the clipper from being allocated, and it would only give a (slight) performance boost if we wrote a Windowed and FullScreen version of the shadow function (else we'd be doing the if clipper test that Blt does but BltFast does not do). I don't think that using BltFast would be worth the effort. I'm going to guess that you wouldn't see any measurable improvement, whatsoever. But, you are welcome to try it... Harold > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED]]On Behalf Of Alexander Gottwald > Sent: Monday, March 11, 2002 10:50 AM > To: [EMAIL PROTECTED] > Subject: DDraw Blt vs BltFast > > > Hi, > > Is there any reason why we use Blt instead of BltFast for copying the > shadow surface. I just searched the MSDN and found this article > http://msdn.microsoft.com/library/default.asp?url=/library/en-us/w cegmm/htm/ddover_52.asp bye ago -- [EMAIL PROTECTED] http://www.gotti.org ICQ: 126018723 phone: +49 3725 349 80 80 mobile: +49 172 7854017