A whole separate set of registers
Integers live in r3, r4, and friends. Floating-point values don't, they get a bank all to themselves, f1 through f13. Your first float argument lands in f1, the second in f2, and whatever you return heads back out in f1, exactly how r3 carries an int result. More float args keep climbing, f3, f4, and so on, the same way extra integer args walk up r5, r6. So the third float a function takes shows up in f3.
A float is single-precision, 32 bits wide, and PowerPC gives it its own arithmetic, the ones whose mnemonics end in s. For addition that's fadds:
fadds f1, f1, f2 # f1 = a + b, single precision
blr
Same three-operand layout you know from add. Destination first, then the two sources.
Your task
Write add_f, returning the sum of two f32s.