IBM 7090 Instruction-Reference page 86

During I time of the next instruction, the AC fraction is checked for zero by adding
one to the
l'
s complement of the fraction. If the fraction is zero, the addition will
cause a column 9 carry. For a zero AC fraction, the whole AC is set to zero and the
characteristic of the MQ is left at zero. If the AC fraction is not zero, there is no
column 9 carry and the MQ characteristic is set 2710 less than the AC characteristic.
During I time of the next instruction, the AC and MQ signs are set to the algebraic sign
of the pr oduct.
Unnormalized Floating Multiply UFM -0260 (Min I, E)
(Max I, E, 11L)
This instruction multiplies the floating-point number stored at the location indicated
by the address by the floating-point number in the MQ. This instruction operates the
same as FMP except that the result is not normalized or tested. The minus PR (S) pre-
vents FP normalizing gate (Systems 2. 10.24. 1). The PR (S) not being plus prevents the
reset of the AC during the following I cycle (Systems 2. 10.25.1). The minus PR (S)
prevents turning on the FP trigger if MQ
=
0, Systems 02.12.07. O.
Floating Round FRN +0760 ... 0011(1, L) Figure 5.3-25
This instruction examines the contents of MQ(9) and if MQ(9) contains a bit, a one
is added to the contents of the AC. A carry out of AC(9) increases the characteristic
by one, causes the fraction to be shifted right, and a one to be placed in AC(9).
Floating Divide or Halt FDH +0240 (Min I, E, L) Figure 5.3 -26
(Max I, E, 12L)
This instruction divides the floating-point number placed in the AC by the floating-
point number stored at location X. The result is a floating-point quotient in the MQ
and the floating-point remainder of the dividend in the AC. The sign of the MQ is the
algebraic sign of the quotient. The sign of the AC is the sign of the dividend. The
characteristic of the MQ is equal to the characteristic of the ACC minus the character-
istic of the SR, increased by 2008. The characteristic of the remainder is 2710 less
than the original dividend characteristic.
If two normalized numbers are used as divisor and dividend fractions, the dividend
fraction cannot be twice as large as the divisor fraction. Therefore, the 7090 is
designated to halt only when the dividend fraction is at least twice as large as the divisor
fraction. The divide check indicator is turned on to cause the halt. The contents of the
MQ will be normal zero and the contents of the ACC will remain tulchanged.
A zero dividend fraction causes the division to be skipped, but does not halt the
computer. A normal zero quotient, plus remainder, results. If the initial factors are
normal floating point numbers, the quotient is also normal.
As in any division using exponents, the fractions are divided and the exponent of the
divisor is subtracted from the exponent of the dividend. The result is the correct quo-
tient factor and exponent. Division of the fractions is done in the same manner as DVH,
by shifting left and attempting reduction. When reduction is successful, a bit is put
into MQ(35). Only 2710 reductions are attempted in FDH because the fraction is con-
tained in 27 10 positions.
FDH uses an
E
cycle and several L cycles. The L cycles are divided into three
types (first, second, and third step) through use of the tally counter.
85