This is a library of single-precision floating-point routines for the eZ80.

You'll need fasmg to compile these routines, which are located in the /src folder. Routines are named __fpxxxx. For example, __fpmul refers to the floating-point multuplication routine.

Honestly, I still don't understand jacobly's test environment very well, so I can't add suggestions about that. However, float routines should follow some basic rules:

• The first operand is in AUBC
• The second operand is in EUHL
• Output is in AUBC
• If a routine doesn't require an argument, it must be preserved. For example, __fpln only takes one argument, so EUHL should be preserved.
• Names of float routines should start with __fp and end with a short name indicating what the routine is. For example, __fpneg negates the input float.

We need to optimize division and square roots! Currently, division is using Newton-Raphson which is overkill on such small precision, and square roots are directly using Newton's method. Square roots should first evaluate at least 8 bits and then implement a very optimized version of Newton's method for square roots (using an auxilliary recurrence).

These two routines also directly affect the performance of logarithms, inverse trig, and inverse hyperbolics.

We also need:

• sine
• cosine
• tangent
• exponential
  • hyperbolic sine
  • hyperbolic cosine
  • hyperbolic tangent
• 2^x
• 10^x
• x^y
• log base 2
• log base 10
• log base y
• Conversion to and from strings