This one's weird: it doesn't affect the actual generation of ASM in
the function, but rather it affects the ASM of sinf and cosf.
You see, sinf and cosf are static - they're built right into the EXE.
Since the previous code used cosf and sinf, they were embedded into
the EXE, and InitTriangleTable would call them directly.
However, this isn't what the original EXE does: instead,
InitTriangleTable calls an intermediary function, that in turn calls
the real cosf and sinf. Turns out this strange code generation is
caused by calling cos and sin instead of cosf and sinf, but still
using float parameters.
It's really starting to look more and more like Pixel really did write
in C90... just look at the declarations I have to shove at the start of
the functions.
