Re: Making Lemonade (Floating-point format changes)

On 5/21/2024 5:12 PM, MitchAlsup1 wrote:

BGB wrote:
 

>
Currently, FCMPEQ will be false on NaN, whereas FCMPGT ignores NaN's.

 

I guess possible could be to add an FCMPGE instruction, and then make both FCMPGT and FCMPGE be false on NaN (where the LT and LE cases can be
>
handled by flipping the arguments, so would still be false on NaN).

 

So, as-is:
   if(!(a==a))
   {
     //NaN
   }

 

But, as-is:
   if(a>0)
   {
     //May still potentially get here with NaN
   }

 Not when the compare is true to IEEE 754. When there is a floating
point
compare and one of the operands is NaN, none of the 6 standard
comparison
forms are true and control transfers to the else-clause.
 In your top example the ! (not) causes NaNs to go to the then-clause
 In your bottom example, no NaN is allowed into the then-clause.

Yeah, but as noted, was describing existing behavior in my case, not the standard behavior.
But, thinking about it, I went and added a "FCMPGE Rm, Rn" instruction to my ISA spec, and to the instruction decoder and emulator (the logic for FCMPGE had already existed, so this was fairly easy).
Then added a special case so now both FCMPGT and FCMPGE will be false if the inputs are NaN (to potentially allow for the standard behavior here). Still need to update my compiler to use the new instruction.