Email or username:

Password:

Forgot your password?
Top-level
Gemini6Ice

@b0rk I DIDNT REALIZE THERE WAS AKWAYS A LEADING 1! I’ll Google but if you have an eli5 as to WHY it was done this way I’m curious!

5 comments
Gemini6Ice

@b0rk like, I’d always thought that a significand of “10” was directly multiplied by the exponentiated base, not “1.000….0010”

simonf

@gemini6ice @b0rk there isn’t when you get into the denorms (exponent = 0)

Gemini6Ice

@simonf huh seems like it would be harder to build in the first place by having to have a single off-case.

simonf

@gemini6ice Ha. Indeed IIRC some early processors handled the special denorm cases as a trap and might have been in software or microcode or something. Either way, not fast.

Poor me, OTOH, who was doing very very early driver for certain 3D hardware and needed to convert floats to a custom semi-float format, thought he’d be clever and scale vals into the denorm range. Couldn’t work out why it was sooooo sloooowww

simonf

@gemini6ice Suffice to say I changed the mapping code to NOT use denorms, and we designed the second generation of the graphics chip to read floats directly.

Go Up