Bit-exact matches are nice but they do not remove floating-point drift (i.e. discrepancy between the calculated values and the true mathematical value), which is indeed unavoidable because the mathematical result of most operations you perform will not be exactly a floating-point representable value.
No comments yet.