The four VAX floating-point formats (32-bit F format, 64-bit D format, alternate 64-bit G format, 128-bit H format) all have three classes of floating-point data, encoded by a sign bit, an exponent field, and a significand field:

1. Normal floating-point numbers, encoded with a nonzero exponent field.
2. Reserved operands, encoded with zero exponent and sign bit set. Triggers reserved operand fault on input to floating-point operations, rather than propagating without a trap like a quiet NaN would. (IEEE 754 uses this encoding for negative zero or negative subnormals instead.)
3. Zero, encoded with zero exponent, sign bit clear, and zero significand field.
4. Dirty zero, encoded with zero exponent, sign bit clear, and nonzero significand field. Treated as zero on input to floating-point operations. (IEEE 754 uses this encoding for positive subnormals instead.)

On NetBSD/vax, the fpclassify() function has four possible return values for these cases:

• FP_NORMAL
• FP_ROP
• FP_ZERO
• FP_DIRTYZERO

FP_ROP is an example of a non-finite floating-point class other than infinity and NaN.

References:

• VAX Architecture Reference Manual, Digital Equipment Corporation, 1987, Chapter 1: “Basic Architecture”, pp. 7–9
• NetBSD math.h fragment for VAX defining FP_ROP and FP_DIRTYZERO
• NetBSD fpclassify() implementation for VAX D floating-point format
• NetBSD 10.1 fpclassify(3) man page