A float divide that isn't a divide
Here's a divide the hardware never performs. Hand MWCC a power-of-two constant divisor and it skips fdivs. Why? The reciprocal of a power of two is exactly representable, so the compiler stores that number and multiplies. One fmuls, done. Watch y / 8.0f:
lfs f0, ... # load the reciprocal constant 0.125f from the pool
fmuls f1,f1,f0 # y * 0.125f — no fdivs in sight
blr
The constant rides in a read-only float pool. lfs fetches it, fmuls applies it, and the division is gone. Exactness is the whole game. Powers of two qualify; almost nothing else does. Ask for y / 3.0f and the divide comes back as fdivs, since 1/3 can't be written exactly in binary.
Spotting lfs then fmuls in the target? Work back to the power-of-two divisor behind that reciprocal. Write it as a plain division in C and trust the compiler to make the constant. Typing the reciprocal by hand defeats the point.
Your task
Write quarter taking an f32 x so it compiles to an lfs + fmuls pattern.