Re: LW compiler optimizations

Double precision has a mantissa of 53 bits. Which is serious overkill. But if you have it, you might as well use it. I found that, back in the days of the 386 and 486, double precision was faster than single precision. The processors actually used 80-bit floating point internally.

24 bits of precision is really plenty of space. You can have 24 dB of headroom above nominal 0 dBVU, and still have 20 bits of mantissa for quiet dynamic range and soak-up bits at the bottom. If processing generates on average Sqrt(N) bits of noise for each stage of plugin processing, 8 guard bits at the bottom will soak up 64 plugins - totally unheard of in any studio that I know. And you still have 12 bits of quiet dynamic range.

While it is supposedly true that you can hear 96 dB of dynamic range, you can’t hear it all at once. At most, human hearing can accommodate about 6 bits of dynamic range at any one time. So 12 bits of quiet dynamic range is still overkill. That gives you an unmolested 72 dB of quiet dynamic range, below 0 dBVU, while still having headroom of 24 dB for loud percussive strikes in the music. 

So really, more than 24 bits is kind of a waste.

> On Jan 14, 2025, at 05:21, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote:
> 
> Yes, that’s what I meant by 24-bit floating point. The mantissa is 24 bits, but the whole word is 32-bit single precision floating point, like you stated.
> 
> 
> 
> 
>> On Jan 14, 2025, at 05:20, Yuri Davidovsky <[email protected]> wrote:
>> 
>> 
>> 
>>> On 14 Jan 2025, at 13:11, David McClain <[email protected]> wrote:
>>> 
>>> But the modern approach with WAV format has that upon reading the samples are immediately converted to 24-bit floating point values. That is a fiction that keeps things easy for everyone, despite the problems of denormals and loss of dynamic range. It also helps avoid intermediate clipping, where that decision is deferred to the very end when you go out to the DAC.
>> 
>> I was under impression that audio processing is all 32 bit standard nowadays, isn’t it? Meaning, you may be reading a 16 bit wav off the disk, but it all gets converted into 32 bit floats, at least that is what I’d do. Some VST plugins consider even double floats as the standard resolution in their processing nowadays, Ozone mastering plugin was doing that since inception, which is like since the late 90s.
>> 
> 
> 
> _______________________________________________
> Lisp Hug - the mailing list for LispWorks users
> [email protected]
> http://www.lispworks.com/support/lisp-hug.html


_______________________________________________
Lisp Hug - the mailing list for LispWorks users
[email protected]
http://www.lispworks.com/support/lisp-hug.html