[LAD] Floating point Denormals: C++ and Waf
linuxaudio at cryptomys.de
Fri Aug 3 07:40:27 UTC 2012
David Robillard-4 wrote:
> Branch per sample, eeck...
Yes - but as this reverb is inside the realtime emulation of an electronic
(vox continental) or electromechanical (hammond) organ where I do hundreds
of operations and a lot of branches to create each little sample these few
more branches don't make a difference.
OTOH my connie project should be a toolbox to check out different approaches
to solve one problem on different platforms. The first solution was to add
1e-20 - but hey this is only a quick hack and we're done.
As I wanted to learn something from struggling with dsp algos my next try
was the processor support for denormals - but this wasn't available in older
PIII without sse2 so I created a diy daz. The test showed no big impact on
cpu usage compared with the 1e-20 solution so I was happy. :)
View this message in context: http://old.nabble.com/Floating-point-Denormals%3A-C%2B%2B-and-Waf-tp34245224p34249634.html
Sent from the linux-audio-dev mailing list archive at Nabble.com.
More information about the Linux-audio-dev