Although 'absolute' and 'relative' tolerances are sufficient certainly for most unit testing situations in signal processing, I feel that what's lacking is equality testing to a definable level of precision (i.e. in the mantissa).
To this end, I've added support 'precision' as a new ToleranceType. I've added support for the string ''precision'' in the obvious way to ParseFloatAssertInputs, and to CompareFloats.m, I've added to the switch block:
the following new case:
coreCompareFcn = @(A, B) magFcn(chop(A,params.Tolerance) - chop(B,params.Tolerance)) <= eps(max(magFcn(A), magFcn(B)));