> I think this had best be concluded by saying that a signed right shift
> does implement an operation that can be labelled "division", but that
> most people would be surprised by the result of (-1)/2 in this case. Where a
> particular application can accept rounding towards negative infinity, use
> of a right shift for signed division may be a fast solution.

You may prefer the round-toward-zero semantics; I for one have NEVER had a
situation in which they were desirable, and have frequently had cases where
I wished that the identity:

  ((a mod m)+b) mod m == (a+b) mod m

would hold for--most notably--all 0<=a<m, -m<b<m, but unfortunately the
round-to-zero semantics on many machines break this requiring me to use
constructs like:


  a=(a+m+b) mod m

rather than

  a=(a+b) mod m

which is simpler, faster, and more logical.  The only case I know of in
which round-toward-zero semantics are desirable are when converting a
number into a displayable representation; even there, the normal first
step is to take the absolute value of a number after (optionally) output-
ting a plus or minus sign; delaying any rounding until after the abs-val
operation will ensure consistent behavior.