Hey John,
I was reading the posts about posits ... pretty interesting.
But the Model T ROM doesn't actually use IEEE floating point format ...
It uses a format where the first byte contains the sign bit and 7-bit
exponent, followed by 3 or 7 bytes of BCD encoded data. For the
exponent, 40h represents "zero". In single precision, pi would be (all
hex):
41 31 41 53
and 0.05 would be
3F 50 00 00
The ROM does this so it can easily display and convert numbers and can
calculate floating point easily (since it has the DAA instruction).
Given the encoding in the posit number, it would likely result in RAM
savings for storing floating point numbers, but it is unclear to me if a
Model T implementation would actually result in ROM savings. It might
actually make the ROM bigger! :)
Ken
On 7/10/19 2:00 PM, John R. Hogerhuis wrote:
Something new under the sun?
Tangentially related, but I thought this was interesting. The Model T
ROM uses 32-bit single precision and 64-bit double precision floats.
A software posits ROM library could produce big speed and space
improvements.
https://www.nextplatform.com/2019/07/08/new-approach-could-sink-floating-point-computation/
https://www.johndcook.com/blog/2018/04/11/anatomy-of-a-posit-number/
-- John.
