IEEE International Conference on Advances in Engineering & Technology Research (ICAETR-2014),

August 01-02,2014, Dr. Virendra Swarup Group of Institutions, Unnao, India

FPGA Implementation of Addition/Subtraction Module for Double

Precision Floating Point Numbers Using Verilog

Sonali M. Rane Prof. Mrs. Trupti Wagh Dr. Mrs. P. Malathi

Department of E&TC Engineering Department of E&TC Engineering HOD, Dept. of E&TC Engineering
DYPCOE, Akurdi DYPCOE, Akurdi DYPCOE, Akurdi
Pune, India Pune, India Pune, India
[email protected] [email protected] [email protected]

Abstract- The floating point operations are critical to

implement on FPGAs due to their complexity of their algorithms. Ellponen Fraction.F
Sign
Thus, many scientific problems require floating point arithmetic s 1111 bil �521 bit
with high levels of accuracy in their calculations. Therefore, in
this paper the proposed work is explored FPGA implementation
of Addition/Subtraction for IEEE double precision floating point
I�! 1 1 I52! I 1
.... ·· ..·· ..... ..·· ·..... ···· ·..···· ...· .. ..· . .··· .. · ... ·· ... · ..·.. 1 1101
numbers. This kind of unit can be extremely useful in the FPGA
implementation of complex systems that benefits from the Fig. 1: Double precisionjloating point format.
parallelism of the FPGA device. The design is in Verilog
Hardware description language (HDL) and implemented on of a major concern. As a result, this work is primarily aimed
FPGA. The verilog code first simulated with isim and synthesized for improved implementation of double-precision floating­
on Xilinx ISE14.1i. point addition/subtraction, on FPGA platform. The mantissa of
double-precision floating-point numbers is 53 bits in length. In
Keywords-FPGA, Floating Point, Verilog, Double precision,
this work, an approach for the addition/ subtraction of double­
AdderlSubtractor, 1EEE754.
precision floating-point numbers has been proposed which
INTRODUCTION allows to using less amount of addition! subtraction, achieving
excellent performance at a relatively low cost of hardware
Floating point numbers are widely adopted in many
resources. The real numbers represented in binary format are
applications due their dynamic representation capability.
known as floating point numbers. Based on IEEE-754
Numbers representated in floating point format can represent
standard, floating point formats are classified into binary and
a much larger range than possible using the same number of
decimal interchange formats. Floating point operations are
bits in fixed point format. This makes a floating point a natural
very important in DSP applications. Figure 1 shows the
choice for scientific and signal processing computations. The
IEEE- 754 double precision binary format representation. Sign
advantage of floating-point representation over integer and
(S) is represented with one bit, exponent (E) with 11 bit and
fixed-point representation is that, it can shore up a much
Mantissa with fifty two bits. For a number is said to be a
wider series of values.
normalized number, if must consist of 'one' in the MSB of the
Now a days, new recent advances in FPGA technology
significant and exponent is greater than zero and smaller than
made it good choice for giving high performance in many
1023. Double precision format as shown in figure 1.
computationally scientific and Engineering applications.
Double precision floating point number is calculated as
Recent FPGAs have large amounts of programmable logics,
memory and often come with a large set of high speed IP cores
Value= (-Iss)* (IFF)*2(EE-I023) (I)
to implement scientific functions. The FPGAs are becoming
aggressive to general purpose processors in wide series of
The next segment provides a high level overview of the
applications, like communications, Image -processing,
related work, the way our system is implemented followed by
scientific computing, and cryptography.
the results obtained and conclusion.
FPUs Floating-point arithmetic is widely used in many
areas, particularly in scientific computation, signal processing RELATED WORKS
(like digital filters, image processing FFT, etc.) and; numerical
A lot of work has been carried in the field Floating point
processing .The IEEE defmes the standard for single-precision
implementation on FPGA. Since past two decades a lot of
and double precision formats. Hardware accomplishment of
work has been done, related high performance implementation
arithmetic operations for IEEE floating-point standard
of floating point adder Subtractor unit.
becomes a important part of almost all processors.
Yogin G. Oklobdzija (1994) implemented 32 bit and 64
The application area is always looking for high­
bit leading zero detector (LZD) circuit using CMOS and ECL
performance and area efficient implementation of floating­
Technology. In this the ECL implementation of the 64 bit
point arithmetic operation and, thus, an efficient
LZD circuit simulated to perform in under 200ps for nominal
implementation of floating point adder/subtraction module is

978-1-4799-6393-5/14/$3I.00©2014IEEE

Authorized licensed use limited to: VIT University. Downloaded on February 23,2022 at 16:01:24 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Advances in Engineering & Technology Research (ICAETR-2014),
August 01-02,2014, Dr. Virendra Swarup Group of Institutions, Unnao, India
speed. This circuit has been implemented in 0.6fl CMOS performance of floating point operators in a library developed
technology. at Sandia National Labs.[8].
The algorithmic approach outperformed LS consistently
with improvement in speed ranging from 12%-56% and
improvements in layout area ranging from 14.5%-35%.The DESIGN DESCRIPTION

resulting LZD circuit has notable performance, which is In this section we will discuss about the hardware that we
essential while it is often a part of the critical path in the have used, block diagram of the proposed system.
floating point unit.[I].Nabeel Shirazi, Walters, and Peter The floating point module performs addition, subtraction.
Athanas (1995) implemented custom 16118 bit three stage Also it implements all four rounding modes-round to nearest
pipelined floating point multiplier that doesn't support even, round to zero, round to positive infinity, round to
rounding modes. By pipelining the each unit the sufficient negative infmity.
number of times in order to maximize speed and to minimize
area. They investigate the ways of implementing the floating
Block Diagram
point operators in the best combination of speed and area in The block diagram and the architectural schematic view of
order to minimize the area requirements small floating point double precision floating point adderlSubtractor is shown in
representation formats are used. Pipeline techniques are figures 2 and 3 respectively. Table 1&2 shows the Rounding
applied to produce a result every clock cycle.[2].Loucas Modes selected for various bit combinations of rmode and
Louca, Cook, and Johnson (1996)explored FPGA Operations selected for various bit combinations of Fp_op.
implementation of addition and multiplication for single Each of the blocks along with its brief description is
precision Floating point numbers that tradeoff the area and explained below:
speed for the accuracy.The design achieved 2.3 MFlops and
doesn't supports rounding modes. And 7 MFlops for addition.
[3].Barry Lee, and Neil Burgess (2002) implemented the
floating point unit using primitives of Xillinx vertex-2 FPGA.
The design achieved the operating frequency of IOOMHz. with
I n p ut
a latency of 4 clock cycles. They focused on using primitives
of a specific FPGA to produce components with low delay and
low area matrices.[4].Pavle Belanovic, and Miriam Leeser
(2002) implemented the reconfigurable floating point unit
using VHDL, which is mapped on to Xillinx
XCVI000FPGA.They developed a library of fully Fig.2: Block Diagram of Double Precision Floating point AdderlSubtractor.

parameterized hardware modules for format control,

arithmetic operation and conversion to and from any fixed aa sub
point format.[5].Ali Malik, and Seok-Bum Ko (2006)
implemented the floating point adder using leading one
predictor (LOP) algorithm instead of Leading one Detector
(LOD) algorithm. The key role of the LOP is to calculate the
leading number of zeros in the addition result, working in
parallel with the 2's complement adder. The design enable

�
implemented in Vertex2p FPGA. The improvement seen in
1lJil(530)
LOP design is the level of logic reduced by 23%with an added
C,a(63
expense of increasing the area by 38%.[6]. Dhiraj Sangwan,
and Mahesh K. Yadav investigated adderlsubtractor and
opb(530) I
IlIb(63:0
multiplication unit for floating point arithmetic using VHDL. eJjJonent)(t�O)
The floating point multiplication operation implemented using
sequential architecture based on booths Radix-4 recording
algorithm. For floating point addition, the sequential addition
aa sub
could have been complex so the combinational architecture
has been implemented.[7]. K. Scott Hemmert, and Keith Fig. 3: Black box view of Double Precision Floatingpoint AdderiSubtractor

Underwood (2006), implemented open sourse library of highly

Tablel Rounding modes selected for various bit combination
optimized floating point units for Xilinx -FPGAs. The units
are fully IEEE complaint .The double precision add and Bit combination Rounding mode

multiply achieved the operating frequency of 230 MHz using a 00 Round-nearest -even
10 stage adder pipeline and a 12 stage multiplier pipeline. The 01 Round-to- zero
area requirement is 571 slices for adder. The floating point 10 Round-up
modules are hand-mapped and placed using JHDL as a design 11 Round-down
entry language. This presentation details the size and the

978-1-4799-6393-5/14/$31.00©2014IEEE

Authorized licensed use limited to: VIT University. Downloaded on February 23,2022 at 16:01:24 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Advances in Engineering & Technology Research (ICAETR-2014),
August 01-02,2014, Dr. Virendra Swarup Group of Institutions, Unnao, India
Result = X+Y = (Xm x 2Xe) + (Ym x 2Ye)
Table2. Operation selecting mode. involves the following steps:
Bit combination Operation I) Align binary point:
000 Addition • Initial result exponent: the larger of Xe, Ye
Compute exponent difference: Ye - Xe
001 subtraction
•

• If Ye > Xe Right shift Xm that many positions to form Xm 2

Xe-Ye
A. ADDITION
• If Xe > Ye Right shift Ym that many positions to form Ym 2
This module fp-add, will responsible to perform operation
Ye-Xe
of addition. In the module fp_add., the input operands are
2) Compute sum of aligned m antissas:
separated into their mantissa and exponent components. Then
i.e Xm2 Xe-Ye + Ym or Xm + Xm2 Ye-Xe
the exponents are compared to check which operand is larger.
(3) If normalization of result is required, then a normalization
Then the larger operand goes into "mantissa_large" and
step follows:
"exponent_large" with the smaller operand populating
• Left shift result, decrement result exponent (e.g., if result is
"mantissa_small" and "exponent_small". The sign and
O.OOl xx . . . ) or
exponent of the output is the sign and exponent of the larger
• Right shift result, increment result exponent (e.g., if result is
operand. To fmd the mantissa of the output, the difference
10.lxx . . . )
between the exponents is found and the mantissa_small is
prolong until MSB of data is I (NOTE: Hidden bit in IEEE
right shifted equal to the "difference" amount. Finally, the
Standard)
addition is performed.
4) Check result exponent:
• If bigger than maximum exponent permitted return exponent
B. SUBTRACTION
overflow
Subtraction is performed in the module fp-sub.v.
• If lesser than minimum exponent permitted return exponent
Subtraction is similar to addition where the operands are
underflow
separated, compared and right shifted. Then subtraction is
5) If result mantissa is 0, may possibly require to set the
performed instead of addition. exponent to zero by a special step to return a proper zero.

C. ROUNDING 2. Basic floating point subtraction algorithm

The rounding operation is performed in the module fp­
round. There are four possible rounding modes. Round to Assuming that the operands are already in the IEEE 754
nearest, round to zero, round to positive infinity and round to format, performing floating point addition:
negative infinity. . If after addition extra carry is generated, Result = X-Y = (Xm x 2Xe) - (Ym x 2Ye)
there is a need of increase in exponent by one, along with right involves the following steps:
shifting of the addition result by one bit. And after subtraction I) Align binary point:
of mantissa, the MSB bit lost by the difference it required to • Initial result exponent: the larger of Xe, Ye
check and the mantissa shifted by that value. • Compute exponent difference: Ye - Xe
• If Ye > Xe Right shift Xm that many positions to form Xm 2
D. EXCEPTION Xe-Ye
The exception performs in the FP-exception module. In this • If Xe > Ye Right shift Ym that many positions to form Ym 2
module, the special cases are checked for exception conditions Ye-Xe
and if they are found, the appropriate output is created and the 2) Subtract the aligned mantissas:
individual output signals of overflow, underflow, exception, i.e Xm2 Xe-Ye - Ym or Xm - Xm2 Ye-Xe

inexact, and invalid are asserted . 3) If normalization of result is required, then a normalization
step follows:
• Left shift result, decrement result exponent (e.g., if result is
ALGORITHM O.OOl xx . . . ) or
Right shift result, increment result exponent (e.g., if result is
Floating point addition is one of the difficult units in the
•

10.lxx ...)
floating point arithmetic operations. Addition and the related
Continue until MSB of data is 1 (NOTE: Hidden bit in IEEE
operation of subtraction is the most basic arithmetic operation.
Standard)
The hardware implementation of this arithmetic for floating
4) Check result exponent:
point is a complicated operation due to the normalization
If larger than maximum exponent allowed return exponent
requirements.
•

overflow
• If smaller than minimum exponent allowed return exponent
1. Basic floating point Addition algorithm
underflow
5) If result mantissa is 0, may need to set the exponent to zero
Assuming that the operands are already in the IEEE 754
by a special step to return a proper zero.
format, performing floating point addition:

978-1-4799-6393-5/14/$31.00©2014IEEE

Authorized licensed use limited to: VIT University. Downloaded on February 23,2022 at 16:01:24 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Advances in Engineering & Technology Research (ICAETR-2014),
August 01-02,2014, Dr. Virendra Swarup Group of Institutions, Unnao, India

� utizatiJo Sunrnary (estimated valles) I

SIMULATION RESULTS
UJ
The simulation results of double precision floating point
LogkUtillition Used Av.i�b� Utir�.tion
adder, Subtractor and the device utilization summary for
adderlSubtractor unit is shown in fig. 4, fig 5, and fig.6 IlJnberofSices 1922 768 250%

respectively. Snapshot is nothing but every moment of the tliJnberof �ce Fip Aops 2274 1536 148%
application while running. It gives the obvious elaborated of
Ibnberof 4 i1JJt l\JTs 2746 1536 178%
application. It will be useful for the new user to comprehend
for the future steps. IlmberofOOlled lOBs 206 124 166%
The double precision floating point adderlSubtractor,
tliJnberof GClKs 1 8 12%
supports the IEEE-754 binary interchange format. The
proposed design is done in Verilog.
Fig. 6: Device utilization summary.

CONCLUSION
The verilog code first simulated with isim and synthesized on
We have presented a Double Precision Floating Point
targeted device XC3s200a-5vq I DO. The adder Subtractor
adderlSubtractor unit Using Verilog. The actual logic design
module occupies an area of 412slices, 834 slices respectively.
of a floating point adder is not too difficult, and most of the
and its operating frequency is 418. I 56Mhz.where as in Xilinx
design time was spent trying to minimize the logic resources
core there are 1266 and operating frequency is 284Mhz. The
used. As a part of our plans, proper rounding will be added in
double precision adderlSubtractor Modules are compliant with
the implementations. The design is in Verilog Hardware
IEEE754 format and handle the various rounding conditions.
description language (HDL), and implemented on FPGA.
The designed arithmetic unit operates on 64-bit operands. It
can be designed for 64-quad double precision. It can be
extended to have more mathematical operations like
trigonometric, exponential and logarithmic functions.

REFERENCES

[1] V. Oklobdzija, "An arithmetic and Novel Design of a leading

zero detector circuit: comparison with logic synthesis", IEEE
Transactions on very large scale Integration (VLSI) systems,
Vol. 2, No. 1,(1994) March, pp.124-128.
[2] N. Shirazi, A. Walters and P. Athanas,'Quantitative Analysis of
floating Point Arithmetic on FPGA Based Custom Computing
Machines", Proceedings of IEEE symposium on FPGAs for
custom Computing Machines (FCCM'95),(1995),pp.155-162.
[3] L. Louca, T. A. Cook and W. H. Johnson, "Implementation of
IEEE Single Precision Floating Point Addition and
Multiplication on FPGAs", Proceedings of 83rd IEEE
Fig4:Simulation results of Addition operation Symposium on FPGAs for Custom Computing Machines
(FCCM"96), (1996), pp. 107-116.
[4] B. Lee and N. Burgess, "Parameterisable Floating-point
Operations on FPGA", Conference Record of the Thirty-Sixth
Asilomar Conference on Signals, Systems, and Computers,
(2002).
[5] P. Belanovic and M. Leeser, "A Library of Parameterized
Floating-Point Modules and Their Use", in 12th International
Conference on Field-Programmable Logic and Applications
(FPL-02). London, UK: Springer-Verlag, (2002) September, pp.
657-666.
[6] Malik and S. -B. Ko, "A Study on the Floating-Point Adder in
FPGAs", in Canadian Conference on Electrical and Computer
Engineering (CCECE-06j, (2006) May, pp. 86-89.
[7] D. Sangwan and M. K. Yadav, "Design and Implementation of
AdderiSubtractor and Multiplication Units for Floating-Point
Arithmetic", in International Journal of Electronics Engineering,
(2010), pp. 197-203.
[8] K. Hemmert and K. Underwood. "Open Source High
Fig. 5: Simulation results of Subtraction operation Performance Floating-Point Modules", in 14th Annual IEEE
Symposium on Field-Programmable Custom Computing
Machines (FCCM-06), (2006) April, pp. 349-350.

978-1-4799-6393-5/14/$31.00©2014IEEE

Authorized licensed use limited to: VIT University. Downloaded on February 23,2022 at 16:01:24 UTC from IEEE Xplore. Restrictions apply.
 IEEE International Conference on Advances in Engineering & Technology Research (lCAETR-2014),
August 01-02,2014, Dr. Virendra Swarup Group of Institutions, Unnao, India

978-1-4799-6393-5/14/$31.00©2014IEEE

Authorized licensed use limited to: VIT University. Downloaded on February 23,2022 at 16:01:24 UTC from IEEE Xplore. Restrictions apply.