IRJET- Single Precision Floating Point Arithmetic using VHDL Coding

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2458
Single Precision Floating Point Arithmetic using Vhdl Coding
Som Sharma1, Ankit Trivedi2
1Student, Department of Electrical Engineering, Axis Institute of Technology&Management, Kanpur,
UttarPradesh, India
2Asst. Prof., Department of Electronic Engineering, Axis Institute of Technology&Management, Kanpur,
UttarPradesh, India
---------------------------------------------------------------------------***---------------------------------------------------------------------------
Abstract: FPGA stands for Field Programmable Gate Array
are semiconductor devices that are based on a matrix
structure of configurable logic blocks (CLBs) connected via
programmable interconnects. FPGAcanbereprogrammedto
desired application or functionality requirementsafter
manufacturing. Floating Point Arithmetic (FPA) is
arithmetic using formulaicrepresentationof realnumbers as
an approximation soas to support a trade-off between range
and precision. For this reason, Floating point computation is
often found in systemsthatincludesverysmallandverylarge
real numbers, and can done fast processing,The IEEEsetsan
Standard named as IEEE 754 for floating point in single
precision and double precision in 1985.Xilinx ISE 14.2
software, weareusingforcreatingFPGAprogrammingbythe
help of VHDL(Very High Speed Integrated Circuit Hardware
Description Language).
KEYWORDS: ASIC, FPGA, IEEE754, VHDL, Xilinx ISE.
1. INTRODUCTION:-
Floating point operation is the most frequent operation and
that is almost used for half of the scientific operation,
computer, and technology. It is a most fundamental
component of math coprocessor, DSP processors, and
embedded arithmetic processors, based. Floating-point
operation is a costly operation in terms of hardware and
timing as it needs different types of building blocks with
variable latency. In floating point operation require
implementations, latency is the overall performance
bottleneck. A major of work has been implemented to
improve the overall latency of floating point operation.
Various algorithms and design approaches have been
developed by the Very Large Scale Integrated (VLSI) circuit
community. Field Programmable Gate Array is made up of
silicon chips with unconnected logic blocks and these logic
blocks can be defined and redefined by user at any time.
Field Programmable Gate Array is increasingly being used
for applications which have high numerical stability and
accuracy. With less time to market and low cost, Field
Programmable Gate Array is becoming a more attractive
solution comparedtoApplicationSpecific IntegratedCircuits
(ASIC). In now time Field Programmable Gate Array are
mostly used in low volume applications that cannot afford
silicon fabricationordesigns whichrequirefrequentchanges
or upgrades in system. Devices afford with millions of gates
and frequencies reaching up to 300 MHz are becomingmore
suitable forfloating-point arithmeticreliantapplicationsand
data processing units. These components require high
numerical stability and accuracy and hence are floating-
point
1.1 Related Work:-
One of the first competitive floating-point operation
implementation is done by L. Louca, T. Cook, and W. Johnson
[8] in 1996. Single precision floating-point addition was
implemented for Altera FPGA device. The primary challenge
was to fit the design in the chip area while having reasonable
timing convergence. The main motive of their
implementation was to achieve IEEE standard accuracy with
reasonable performance consideration. This is claimed to be
the first IEEE single precision floating-point adder
implementation onaFPGA,beforethis, implementation with
only 18-bit word length was present [8].The majority of the
algorithms execute in FPGAs used to be fixed point. Floating
point operations are useful for calculation involving large
dynamic range, but they require significantlymoreresources
than integer operation With the present trend in system
supplies and existing FPGAs,floating-point implementations
are becoming more common and designers are increasingly
taking advantage of FPGAs as a platform for floating-point
implementations. The rapid advance in Field Programmable
Gate Array technology makes such devices ever more
attractive for implementing floating-point arithmetic.
Evaluate to Application Specific Integrated Circuits, FPGAs
offer compact developmenttimeandcosts.additionally,their
flexibility enables field upgrade and adjustment.
The IEEE 754 single precision format[4] is as shownbelow.It
divides in three parts are as follows:-
sign: 1 bit broad and used to denote the sign of the number i.e.
0 point to positive number and 1 represent negative number.
Exponent: 8 bit broad and signed exponent in excess 127
representation.This fieldrepresentsbothpositiveandnegative
exponents.
Mantissa: 23 bit wide and fractional component.
S 8 bit Exponent-E 23bit fraction –F

The single- precisionfloating-pointnumberiscalculatedas(-1)
S × 1.F × 2(E-127
1.2 Proposed Algorithm:-
Description of the Proposed Algorithm:
Fig 1 shows the flowchart of standard floating point adder
algorithm. Let s1; e1; f1 and s2; e2; f2 be the signs,
exponents, and significant of two input floating –point
operands, N1 and N2, respectively. A description of the
standard floating point adder algorithm is as follows.
Fig. 1: Flowchart of standard floating point adder
Algorithm
1. The two operands, N1 and N2 are read in and
compared for demoralization and infinity. If
numbers are demoralized, set the implicit bit to
0 otherwise it is set to 1. At this instant, the
fraction part is extended to 24 bits.
2. The two exponents, e1 and e2 are compared
using 8-bit subtraction. If e1 is less than e2, N1
and N2 are swapped i.e. previous f2 will now be
referred to as f1 and vice versa.
3. The minor fraction, f2 is shifted right by the
absolute difference result ofthe two exponents’
subtraction. Now both the numbers have the
sameexponent.
4. The 2 signs are used to see whether the
operation is a subtraction (-) or anaddition (+).
5. If the process is a subtraction, the bits of the f2
areinverted.
6. At this instant the two fractions areaddedusing
a 2’s complement adder.
7. If the outcomesum is a negative number, it has to
beinverted and a 1 has to beadded totheresult.
8. The outcome is then passed through a leading
one detector orleading zero counter. This is the
first step in the normalization step.
9. Using the outcome from the leading one
detector, the result is then shifted left to be
normalized. In some cases, 1-bit right shift is
needed.
10. The outcome is then rounded towards nearest
even, the default roundingmode.
11. If the take out from the rounding adder is 1, the
result is left shifted byone.
12. Using the outcome from the leading one
detector, the exponent is adjusted. The sign is
calculate and after overflow and underflow
check, the result isregistered.

1.3 Block Diagram Of Single Precision Floating Point
Adder:-
Fig 2: Block Diagram of Single Precision floating-point
Adder
Fig 2 shows the block diagram of single precision floating
point adder.Num_A , Num_B ,CLK ,GND and VCC are inputsof
the block diagram of single precision floating point adder.
Output SUM is calculated by performing addition of Num_A
and Num_B using floating point addition algorithm shown in
Fig 1. It illustrate the main hardware modules necessary for
floating point addition. The different modules are Exponent
Difference Module, Swap Multiplexer, Shifter, Barrel Shifter ,
FractionComponentEffectOperation,2’sComplementAdder
, Inverter , Normalizer , Leading One Detector , Left Shift
Barrel Shifter, Exponent Sum and Mantissa Sum.
Exp_a and Exp_b are the exponents of inputs Num_A and
Num_B resp. Exp_a and Exp_b can be positive or negative
numbers. The output shift_amt is given to the shifter and
barrel shifter block forfurthercalculation.Sign_disgiventwo
the Swap Mux. Swap_Mux assigns greater Mantisa to
Mantisa_grt and Lesser value of Mantisa to Mantisa_less and
greater exponent is calculated and assigned to Exp_grt.
Exp_grt output isgiventothenormalizerblock,Mantisa_grtis
given to the Fraction Component Effect Operation Block and
Mantisa_less is given to the Barrel shifter Block respectively.
The shifter is used to shift the significant of the smaller
operand by the absolute exponent difference. Shifter is used
to shift the data bits. The Output Shift_amt of Exponent
Difference Block is given as input to the Shifter block which
gives output shifted_amtisfurthergiventotheinverterblock.
The inverter is used to invert the data bits of shifted amount.
The normalizer block gives us normalized result. Later than
the addition, the next step is to normalize the result. The
primary step is to identify the leading or first one in the
result. This outcome is used to shift left the adder result by
the number of zeros in front of the leading one. In order to
perform this operation, particular hardware, called Leading
One Detector (LOD) or Leading Zero Counter (LZC), has tobe
implemented. Exponent sum and mantissa sum blocks are
used to calculate the exponent and mantissa of output SUM
and sign bit SOP is initially considered as 1.
1.4 Floating point Multiplication:-
The figure 3 shows the flowchart of multiplicationalgorithm
of multiplication is demonstrated by flowchart. In1 and in2
are two numbers sign1, expo1, S1 and sign2, expo2, S2 are
sign bit, exponent and significant of in1 and in respectively.
Steps for multiplication are as follows.
1:- calculate sign bit. sign_f= sign1 XOR sign2, sign_fissignof
final result.
2:- add the exponents and subtract 127 to make adjustment
in exponent (expo1+127+expo2+127)-127.
3:- Multiply the significant. S=S1*S2.
4:- check for overflow and underflow and special flag. When
the value of biased exponent is less than 1 it shows
occurrence of underflow, if exponent is greater than 254
then it shows overflow of floating point operation.
5:- Take the first 23 bits of ‘S’ and from left side and discard
remaining bits.
6:- Arrange the results in 32 bit format.1sign bitfollowedby
eight bits exponent followed by 23 bits
Mantissa/significand.
Calculation of sign bit:- when both number have same sign
the sign of result is positive else sign will be negativethesign
in calculated by XORing both sign bits of inputs.
The multiplication result is 48 bits. If 47th bit is ‘1’then right
shifts the result and adds ‘1’ in exponent to normalize the
product. 46th to 23th bits are actual significant product.
Exponent addition is done by unsigned 8 bit adder and to
bias properly subtract 127 from the addition result for that
purpose unsigned 8 bit subtractor is used. Inanyofthecases
either addition of exponent in the beginning or while
adjusting result the exponent must in the range 1 to 254.
When overflow occur the result of multiplication goes to
±Infinity (+ or – sign is determined by the sign of two input

numbers). When underflow occur it makes underflow flag
high and the result goes to ±0 (+ or – signedisdeterminedby
the sign of two input numbers).
FIG 3:-Flowchart of standard floating point multiplication
1.5 Floating point division:-
The figure 4 shows the flowchart of division, algorithm for
division is explained through this flowchart. In1 and in2 are
two numbers sign1, expo1, S1 and sign2, expo2, S2 are sign
bit, exponent and significant of in1 and in2 respectively. It is
assumed that in1 andin2 are in normalized form.
Steps for floating point division are as follows.
1:- calculate sign bit. sign_f= sign1 XOR sign2, sign_fissignof
final result.
2:- divide the significand S1 byS2fordivisionbinarydivision
method is used.
• Expand divisor and dividend to double of their size
- Expanded divisor = divisor (24 bits MSB) zeroes (24 bits
LSB)
- Expanded dividend = zeroes (24 bits MSB) dividend (32
bits, LSB)
• For each step, determine if divisor is smaller than dividend
- Subtract divisor from dividend look at sign
- If result is greater than or equal to‘0’:dividend/divisor>=1,
mark quotient as “1”.
- If result is negative: divisor larger than dividend; make in
quotient as “0”
• Shift quotient left and divisor right to cover next power of
two.
3:- Subtract the expo2 from expo1.
4:- check for overflow and underflow flags.
5: assemble the result in 32 bits format
FIG 4:-Flowchart of standard floating point division

2. EXPERIMENTAL RESULTS:-
Simulation result of addition of two 32 bit floating point
number. The round mode is 00 and the operation mode
is 000
Simulation result of subtraction of two 32 bit floating
point number. The round mode is 10 and the operation
mode is 001

Simulation result of multiplication of two 32 bit floating
point number. The round mode is 01 and the operation
mode is 010
Simulation result of division of two 32 bit floating point
number. The round mode is 00 and the operation mode
is 011
3. Conclusion and Future Work:-
This FPU has action as addition, subtraction
multiplication, and division. Addition, subtraction and
division have been implemented by conventional
technique. The design of FPU has done using VHDL on
Xilinx VIVADO 2014.4 and has implemented on Virtex-7
xc7v585tffg1157-3.We have proved that this FPU have
requisite less memory but still we have a huge amount of
work that can be set on this FPU to further make up the
efficiency of the FPU by using other Vedic sutra.
A single precision floating-point adder is implemented in
this paper. The main contribution of our work is to
implement and analyze floating-point addition algorithms
and hardware modules were implemented using VHDL
and is Synthesized using Xilinx ISE14.2 Suite. The results
are obtained using ISim (VHDL/Verilog) Simulator. In
order to expand our paper further some of the works can
be proposed in order to accommodate any exponent and
mantissa length. This will gives more versatility while
choosing the design criteria. The design can also be
pipelined further for different number of pipeline stages
to give even more adaptability and flexibility.
4. References:-
1) Ronald Vincent, Ms.Anju.S.L “Decimal Floating
Point Format Based on Commonly Used Precision
For Embedded System Applications.” International
Conference on Microelectronics, Communication
and Renewable Energy (ICMiCR-2013).
2) Somsubhra Ghosh, Prarthana Bhattacharyya and
Arka Dutta “FPGA Based Implementation of a
Double Precision IEEE Floating-Point Adder”,
Proceedings of7'h International Conference on
Intelligent Systems and Control (ISCO 2013).
3) Maarten Boersma, Michael Kr¨oner, Christophe
Layer, Petra Leber, Silvia M. M¨uller, KerstinSchelm
“The POWER7 Binary Floating- Point Unit”, 2011
20th IEEE Symposium on Computer Arithmetic.
4) Reshma Cherian, Nisha Thomas, Y. Shyju
“Implementation of Binary to Floating Point
Converter using HDL”P-461-P464.
5) Anand Mehta, C. B. Bidhul, Sajeevan Joseph,
Jayakrishnan. P “ Implementation of Single
Precision Floating Point Multiplier using Karatsuba
Algorithm”,2013International ConferenceonGreen
Computing, Communication and Conservation of
Energy (ICGCE).

IRJET- Single Precision Floating Point Arithmetic using VHDL Coding

Recommended

More Related Content

What's hot (19)

Similar to IRJET- Single Precision Floating Point Arithmetic using VHDL Coding (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Single Precision Floating Point Arithmetic using VHDL Coding