Floating Point Arithmetic

FLOATING POINT
ARITHMETIC
Intro:
Let (XM , XE ) be the floating point
representation of number, X. [X= XM * BXE]
B- base of X.
Assumptions:
XM is an binary fraction; nM is the length
of XM
XE is an binary integer in excess 2nE-1
code; nE is the length of XE
FLOATING POINT
ARITHMETIC
Basic Operations:
1. Multiplication & Division:
-. Add/subtract the exponents. (Mul/div)
-. Multiply/divide the mantissa part.
-. Fixed point operation.
2. Addition & Subtraction:
-. Compare the exponent(Subtract).
-. Shift the smaller exponent having
number right by the difference
between the exponents.
FLOATING POINT
ARITHMETIC
Cntd..
-Now the exponent of both the numbers
become the same.
-Add/ subtract the mantissa.(Fixed point
process)
-Normalize the final result by shifting the
mantissa left to eliminate the leading
zeroes & in parallel correct the exponent
for shifting.
XY = (XM 2XE-YE YM)* 2E
(XE YE )
FLOATING POINT
ARITHMETIC
DIFFICULTIES:
1. While doing multiplication operation, we
need to add exponents which produces doubly
biased result. (Eg: 128+128=256(Doubly
Biased)-127=129(Correct One))
2. If X=0 and it is multiplied with a non zero
number; the mantissa part equals zero, but
the exponent will not be in all-zero form.
(Eg: 0 X 1.1* 22 produces
exponent=0+129=129should be made zero)
FLOATING POINT
ARITHMETIC
DIFFICULTIES:
3. Overflow & Underflow:
-Mantissa overflow can be overcome by shifting the
mantissa & alter the exponent.
-Exponent overflow- indicated by error signal.
-The floating point result overflowed or underflowedmay be represented in denormalized form.
4. Guard bit attachment:
-To preserve accuracy by rounding off (adding 1 to
xo and truncating)
-If mantissa is right shifted; the bits shifted right can
be retained as guard bits.
FLOATING POINT
ARITHMETIC
FLOATING POINT UNITS:
FLOATING POINT
ARITHMETIC
-FPU is implemented by two loosely

connected fixed point data path circuits.
Mantissa Unit:
-perform all basic operation; structure of
basic sequential ALU can be used.
Exponent Unit:
-perform Addition, Subtraction.
-Exponents are put E1 & E2 which are
connected to adder/subtractor which is
connected to the counter.
FLOATING POINT
ARITHMETIC
Cntd..
-Exponent Comparison is done by E1-E2; Larger
exponent is determined from the sign of E.
-Shifting of mantissa is controlled by E; Each
decrement results in one shift.
-Exponent of final result is also placed in E.
-Thus the floating point unit can be constructed
from the fixed point unit; and thus both units
can be implemented in same IC.
-But in recent years, they construct separate
units to enable parallel execution.
FLOATING POINT
ARITHMETIC
Improvements in Design:
-Test should be performed for underflow
and overflow; if occurs ERROR flag should
set.
-Test for zero should be carried out; if the
inputs are equal while subtracting,
mantissa becomes zero but the exponent
will be non-zero and it must be set to zero.
-If one operand is zero while adding; then
make the non-zero operand as the result.
FLOATING POINT
ARITHMETIC
Improvements in Design:
-If both the operands are zero while
adding; then make any one of the
operand as the result.
-If the difference between exponents is
very large while addition (E>nm); then E
number shifts can be replaced by nm
number of shifts.
FLOATING POINT
ARITHMETIC
Coprocessors:
FLOATING POINT
ARITHMETIC
Coprocessors:
-Separate instruction set computer that is
closely coupled to CPU whose instructions
& registers are extensions of CPUs.
-Instruction of coprocessor- fetched by
CPU, decoded by CPU and coprocessor,
executed by coprocessor.
-Coprocessors- used as memory
management system & graphic
controlling devices.
FLOATING POINT
ARITHMETIC
Cntd..
- Coprocessors require separate control
logic; it is a slave unit to main processor.
- If coprocessor instructions are
encountered; communication between
processor and coprocessor starts.
- Even though if coprocessor is not present;
the coprocessor instructions take the
program to the memory location for software
routine where the desired coprocessor
instructions are stored.
FLOATING POINT
ARITHMETIC
Cntd..
- CPU generated interruption of normal
program flow is called as coprocessor trap.
-Thus the coprocessor approach gives h/w or
s/w support without altering the CPU
program.
-Coprocessor instructions- three fieldsopcode, address of coprocessor, type of
operation executed by coprocessor.
-CPU & coprocessor can do their operation
parallely.
FLOATING POINT
ARITHMETIC
Cntd..
- Drawback: coprocessor dont know the
contents of registers; (i.e) When the
coprocessor needs to load or store
operands, it will ask CPU for addressing &
data info.

Floating Point Arithmetic

Uploaded by

Floating Point Arithmetic

Uploaded by

FLOATING POINT

FLOATING POINT UNITS:

-FPU is implemented by two loosely

You might also like