Floating-Point Trap - IBM 7090 Instruction-Reference

5.3.13 Floating Point Trap (Figure 5.3-74)
To simplify the programming associated with floating-point overflow or underflow, a
unique system of spill indication is used. When an underflow or overflow occurs in
either the accumulator or MQ during a floating-point operation, an instruction trap is
taken to location 00108, after ending operation on the executer instruction. In addition,
the location of the floating-point instruction plus one is stored in the address portion of
location 0000, and a spill identification code is stored in the decrement portion of loca-
tion 0000.
The 7090 floating-point trap mode trigger (Systems 2.10.71.1) is reset on, putting the
CPU in a floating-point trap mode status normally. To leave this condition, a LFTM
(leave floating-trap mode) instruction must be executed. The mode of operation may
then be re-entered by resetting the CPU or executing an EFTM (enter floating trap mode)
instruction.
The spill
identific~ion
code is produced as follows:
1. Position
14
is set to one if overflow or underflow occurs during a floating-
point divide operation.
2. Position 15 is set to one if overflow occurs in either the accumulator or MQ.
3. Position 16 is set to one if overflow or underflow occurs in the accumulator.
4. Position 17 is set to one if overflow or underflow occurs in the MQ.
This results in the spill identification codes for the various conditions as follows:
Floating-Point
Positions
Octal
Operation
Accumulator MQ
14 15 16 17
Code
Add, Subtract
Underflow 0 0
0 1 01
Multiply,
Underflow Underflow 0 0 1 1
03
Round
Overflow
0 1 1 0 06
Overflow
Overflow 0 1 1 1
07
Divide
Underflow 1
0 0 1 11
Underflow
1 0 1 0
12
Underflow
Underflow
1 0 1 1 13
Overflow 1 1
0
1
15
The code stored in the decrement may be loaded into an index register and an indexed
transfer instruction may be used to lead to the proper corrective routine.
Underflow and overflow of the floating-point characteristic are detected by examining
positions P and Q of the adders and the accumulator. Overflow is identified by a bit in
(P) only. Underflow is identified by bits in (P) and (Q). MQ overflow or underflow must
be recognized as the MQ characteristic is being computed in the adders. ACC overflow
to underflow may be recognized at any time after the final characteristic has been
assigned.
141