the inquired algorithm performs 1D sampling on direction perpendicular to luminance gradient with varying range with following behavior:

reduce sample range by nearby sum with scale factor defined by

FXAA_REDUCE_MUL: reduce scale multiplier

and

FXAA_REDUCE_MIN : minimum reduce scale

after that, FXAA_SPAN_MAX will force clamp the sample range.

as the out come, the sampling vector (A, B ) will be characterize as (A', B') with min(A', B') < 1, and if |A - B| < scale factor, max(A', B') will also be less then 1, that == no blurring.