Emulating PS2's Emotion Engine floats. Converts the single-precision floating points of PS2 to double-precision floating points, and perform double arithmetic.

PS2's floating points are not IEEE754 compliant:-

Therefore, some large floats in the PS2 are not emulatable in x86 due to Infinity/NaN propagation. By doing a single<->double conversion, the +128 exponent is freed up (1024 is used for Inf/NaNs in doubles).

1. Manually cast the float into a double (usual way of casting causes NaN propagation)
2. Perform the arithmetic
3. Manually cast the double back into a float (if EXP>255, return 0xFFFFFFFF; if EXP<=0, return 0x0)