On Sat, May 26, 2007 at 05:14:21PM -0400, Dave
Robillard wrote:
On Sat, 2007-05-26 at 12:25 -0700, Ken Restivo
wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
I've noticed the same strange behaviour in two completely different peices of
software/hardware, and have to wonder if it's a integer-size issue.
First of all, the AZR3 LV2 softsynth clips on my 64-bit machine. Really bad. Lars looked
at it, and gave me the action item to go digging through the source and test the sound at
various points in the processing chain, to try to isolate it further. Which I am still
going to do.
The horrible noise on 64-bit from AZR3 (for me) comes from the key
click. Turn the click control all the way down and it works fine.
I just discovered this a few hours ago. And came up with a hack to fix it... AZR3 now
works on my 64-bit machine, including the click!
I sent the patch below to larsl but he appears to be AFK right now.
Still not sure what's up with the Audiophile.
- -ken
- -------------
Index: plugins/azr3/voice_classes.cpp
===================================================================
RCS file: /sources/ll-plugins/ll-plugins/plugins/azr3/voice_classes.cpp,v
retrieving revision 1.7
diff -u -b -B -w -p -u -b -B -w -p -r1.7 voice_classes.cpp
- --- plugins/azr3/voice_classes.cpp 16 May 2007 10:10:05 -0000 1.7
+++ plugins/azr3/voice_classes.cpp 27 May 2007 00:50:23 -0000
@@ -229,7 +229,7 @@ float voice::clock()
static unsigned long randSeed = 22222;
float rand=0;
randSeed = (randSeed * 196314165) + 907633515;
- - rand=(float)randSeed/0xffffffff;
+ rand=(float)randSeed/0xfffffffffffffff;
clicklp.clock(click*rand*.3f);
noise=clicklp.bp()*clickvol*.7f;
output+=noise;
@@ -286,7 +286,7 @@ float voice::clock()
if(mattack>1)
mattack=1;
randSeed = (randSeed * 196314165) + 907633515;
- - rand=(float)randSeed/0xffffffff;
+ rand=(float)randSeed/0xfffffffffffffff;
clicklp.clock(click*rand*.3f);
noise=clicklp.bp();
noise*=clickvol;
This patch has been applied to CVS now (with the large hex numbers
replaced by std::numeric_limits<unsigned long>::max()). I think it
should work on all platforms but I still have no 64-bit machine to test
it with.
--ll