From: quadibloc@ca.invalid   
      
   On Thu, 05 Feb 2026 01:57:22 +0000, MitchAlsup wrote:   
   > John Savard  posted:   
      
   >> I had looked into unusual memory architectures to allow a computer to   
   >> be designed which had single-precision floats that were 36 bits long,   
   >> so that it would be possible more often to avoid recourse to double   
   >> precision, and which had double-precision floats that were 60 bits   
   >> long, also a multiple of 12, because it wasn't necessary to have all   
   >> the precision of 64-bit floats.   
      
   > Any machine on sale today (selling at les 100,000 machines/year)   
   > provide 36-bit or 60-bit or 72-bit FP ?!?   
      
   Not that I know of. Of course, there's Univac, which still sells machines   
   supporting their old 36-bit architecture.   
      
   > If you want to build a 12-bit=base machine, go ahead--just don't expect   
   > much takeup.   
      
   That's indeed the problem, so I tried to address the problem.   
      
   >> So how could I achieve my original goals while avoiding awkwardness?   
      
   > Avoid non 8^n design points altogether.   
      
   That, unfortunately, couldn't achieve my original goals.   
      
   >> Well, I came up with this:   
      
   >> Have floating-point formats that are either 36 bits long or 72 bits   
   >> long.   
   >   
   > Ok, better than the above, 12^n -> {12, 24, 48, 96}   
   > WOOPS no 36, 60 or 72 !!!   
   > ............................6^n -> {6, 12, 24, 48, 96} still does not   
   > work.   
      
   The idea is now there's a 9-bit byte, and everything is build around that   
   9-bit byte. Although 9 is not a power of two, all other lengths are 9   
   times a power of two, so binary addressing of these bytes and two-byte and   
   four-byte and eight-byte quantities remains just as simple as on a pure   
   2^n machine.   
      
   Since 2^n machines with *bit addressing* are just about as rare as 36-bit   
   and 60-bit machines... now my proposal is "just as good".   
      
   I _still_ don't _really_ expect much takeup, even though my floats have   
   sizes that seem to match the precisions those engaged in scientific   
   computing were fond of.   
      
   >> One of the 72-bit formats has the same significand (or mantissa) length   
   >> as the 48-bit floats in my idealized computer. But no bits are wasted;   
   >> instead, the exponent field is just enlarged.   
   >   
   > 72-bit FP (ala IEEE754 rules) is arguably better than Posits.   
      
   At least one bit of positivity.   
      
   >> The other 72-bit format has a significand   
      
   > ?? fraction ??   
      
   A floating-point number usually has three parts; a sign, an exponent   
   (which includes its own sign) and...   
      
   a coefficient or mantissa or fraction... which is now referred to, in the   
   IEEE standard, as a "significand", so I guess we have to get use to the   
   new official name for it.   
      
   John Savard   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)