Subject: Re: [linux-audio-dev] double <--> float conversions speed benchmark andanalysis
From: Erik Steffl (esteffl_AT_pbi.net)
Date: Tue Apr 04 2000 - 19:42:18 EEST
Benno Senoner wrote:
...
> This again proved that my multidatatype design is superiour to a
> design limited to float.
> Consider the fact that many VST plugins are now internally (except for inputs
> and outputs) working with double to get full 24bit precision. (synths, and other
> HQ stuff, since you can't pretend to deliver full 24bit precision while only
> doing 24bit internal computations)
> That means at least a double to float and float to
> double conversion between many plugins, which can eat a part of your CPU to do
> useless work.
in addition to that C math library uses double.
> With my design you can for example keep plugins using the "double"
> datatype in a sequential chain without any useless datatype ( float
> to double at every input and double to float at every output) conversions.
exactly, it was already pointed out that when you support multiple
data types you might actually end up doing less data conversions than if
only one data format is supported...
erik
This archive was generated by hypermail 2b28 : Tue Apr 04 2000 - 20:44:05 EEST