Cray CRAY-1 - 070 Ijx Floating Reciprocal Approximation of (Sj) to si

1--------------------------------------------------------------------
070ijx
Floating reciprocal approximation of (Sj) to 5;
,----------------------------------------------------- --------~------
This instruction is executed in t~e reciprocal approximation cnit.
The instruction forms an approximation to the reciprocal of the normalized
floating point quantity
ir, Sj and enters the result into Si. This
instruction occurs in the divide sequence to compute the quotient of
two floating point quantities as described in Section 3 under Floating
Point Arith~etic.
The reciprocal approximation instruction produces a result that is
I accurate to 30 bits. A second approximation ray be generated to
extend the accuracy to
47 bits using the reciprocal iteration instruction.
Hold issue conditions
034 - 037 in process
Exchange in process
Si or Sj reserved
174 in process; unit busy (VL) + 4 CPs
Execution time
Si ready 14 CPs
Instructior issue 1 CP
Special cases
2240004
An arithmetic error allows 17 CPs + 2 parcels to issue if the
f.p. error flag is set.
(Si) is meaningless if (Sj) is not normalized; the unit assumes
that bit 247 of (Sj) = 1; no test is made of this bit.
(Sj) = 0 produces a range error; the result is meaningless.
(Sj) = 0 if j = O.
4-44
E