Rounding increment is unclear when d equals to 0

[sun-jacobi]

In 3.8, we have a pair of functions roundoff_unsigned and roundoff_signed

roundoff_unsigned(v, d) = (unsigned(v) >> d) + r
roundoff_signed(v, d) = (signed(v) >> d) + r

and the Rounding increment r depends on the rounding mode ( i.e vxrm)

In the table, r is given by expressions like v[d-1].
But this definition is unclear when d equals 0.

[aswaterman]

Naturally, shifting by 0 should not result in any rounding. One fix is to define v[-1] to be 0, so that the existing expressions do the right thing. Another is to change the definition of r to be 0 when d is 0.

[sun-jacobi]

Naturally, shifting by 0 should not result in any rounding. One fix is to define v[-1] to be 0, so that the existing expressions do the right thing. Another is to change the definition of r to be 0 when d is 0.

Thanks for your response, there is actually also a similar case when vxrm[1:0] equals 01 and d equals 1
then the r would be v[0] & (v[-1:0]≠0 | v[1])

vxrm[1:0]	Abbreviation	Rounding Mode	Rounding increment, r
00	rnu	round-to-nearest-up (add +0.5 LSB)	v[d-1]
01	rne	round-to-nearest-even	v[d-1] & (v[d-2:0]≠0 | v[d])
10	rdn	round-down (truncate)	0
11	rod	round-to-odd (OR bits into LSB, aka "jam")	!v[d] & v[d-1:0]≠0

[aswaterman]

So we'd also need to say that v[-1:0] is zero by definition, I suppose.