From maroberts@dial.pipex.com Tue, 07 Dec 1999 14:15:20 +0000 Date: Tue, 07 Dec 1999 14:15:20 +0000 From: Mark A Roberts maroberts@dial.pipex.com Subject: [Livid-dev] Sound & other problems I've finally got a system which can compile and run NIST: Hardware: Gigabyte ATX motherboard, K6-2/450 CPU, 128 MB RAM, Voodoo Banshee, SB PCI128 Software: egcs 2.95 nist [yesterdays CVS version] compiled as i586, O6 Problems: The sound is terrible, it seems to arrive in bursts with pauses in between - any ideas why ? Is there a clean Redhat 6.1 compatible RPM of pgcc-2.95 & pgcc-c++ anywhere around ? Regards Mark Roberts
From alan@lxorguk.ukuu.org.uk Tue, 7 Dec 1999 14:18:17 +0000 (GMT) Date: Tue, 7 Dec 1999 14:18:17 +0000 (GMT) From: Alan Cox alan@lxorguk.ukuu.org.uk Subject: [Livid-dev] Sound & other problems > Problems: > The sound is terrible, it seems to arrive in bursts with pauses in > between - any ideas why ? Your machine isnt fast enough for the current code.
From maroberts@dial.pipex.com Tue, 07 Dec 1999 14:43:59 +0000 Date: Tue, 07 Dec 1999 14:43:59 +0000 From: Mark A Roberts maroberts@dial.pipex.com Subject: [Livid-dev] Sound & other problems Alan Cox wrote: > > > Problems: > > The sound is terrible, it seems to arrive in bursts with pauses in > > between - any ideas why ? > > Your machine isnt fast enough for the current code. > Is this just a problem with current nist algorithms/ software design, or am I going to have to go shopping for a mega-machine ? Incidentally I did take a very fast browse through some of the AC3 software last night and have a few novice queries Apologies if any of these are silly, as I said it was a VERY fast look. Where can I find AC3 specifications ? In particular, I'd like to know a) if there's any real reason why a lot of floating point appears to be used - would integer calculations be possible and would there be any benefit in doing them ? b) does every single bit of the sound go though the CRC algorithm ? 44000 odd iterations of this per second must give the CPU some pause for thought.
From aholtzma@ess4.engr.UVic.CA Tue, 7 Dec 1999 07:55:58 -0800 Date: Tue, 7 Dec 1999 07:55:58 -0800 From: Aaron Holtzman aholtzma@ess4.engr.UVic.CA Subject: [Livid-dev] Sound & other problems It would seem that Mark A Roberts (maroberts@dial.pipex.com) said: > Is this just a problem with current nist algorithms/ software design, or > am I going to have to go shopping for a mega-machine ? NIST is a kluge. End of story. > Where can I find AC3 specifications ? There is a link on my ac3 page http://ess.engr.uvic.ca/~aholtzma/ac3 > In particular, I'd like to know > a) if there's any real reason why a lot of floating point appears to be > used - would integer calculations be possible and would there be any > benefit in doing them ? If you're into writing 0.24 fixed point fft's and mixers, then by all means try it out. I doubt it would be much faster on cpus with floating point that doesn't suck. > b) does every single bit of the sound go though the CRC algorithm ? > 44000 odd iterations of this per second must give the CPU some pause for > thought. It's not that big of a deal, only around 6% in the call profile. Also, I have a version that works on 8 bits at a time in my development tree. cheers, aaron
From alan@lxorguk.ukuu.org.uk Tue, 7 Dec 1999 15:35:58 +0000 (GMT) Date: Tue, 7 Dec 1999 15:35:58 +0000 (GMT) From: Alan Cox alan@lxorguk.ukuu.org.uk Subject: [Livid-dev] Sound & other problems > > > The sound is terrible, it seems to arrive in bursts with pauses in > > > between - any ideas why ? > > > > Your machine isnt fast enough for the current code. > > > > Is this just a problem with current nist algorithms/ software design, or > am I going to have to go shopping for a mega-machine ? NIST is slower than ideal. The X support for doing hardware scaling isnt there until XFree 3.9.x (to be 4.0 once stable). The combination is a killer. With an AMD K7/500, a G200 video card using hardware scaling hacks (mga_vid) I get between 18 and 25 fps. The best windows players would probaly do that on a PII/233. So yes we have work to do Alan