Multi-operand Decimal Adder Trees for FPGAs

نویسندگان

  • Álvaro Vázquez
  • Florent de Dinechin
چکیده

The research and development of hardware designs for decimal arithmetic is currently going under an intense activity. For most part, the methods proposed to implement fixed and floating point operations are intended for ASIC designs. Thus, a direct mapping or adaptation of these techniques into a FPGA could be far from an optimal solution. Only a few studies have considered new methods more suitable for FPGA implementations. A basic operation that has not received enough attention in this context is multi-operand BCD addition. For example, it is of interest for low latency implementations of decimal fixed and floating point multipliers and decimal fused multiply-add units. We have explored the most representative proposals for multi-operand BCD addition and found that the resultant implementations in FPGAs are still very inefficient in terms of both area and latency when compared to their binary counterparts. In this paper we present a new method for fast and efficient implementation of multi-operand BCD addition in current FPGA devices. In particular, our proposal maps quite well into the slice structure of the Xilinx Virtex-5/Virtex-6 families and it is highly pipelineable. The synthesis results for a Virtex-6 device indicate that our implementations halve the area and latency of previous proposals, presenting area and delay figures close to those of optimal binary adder trees. Key-words: Decimal arithmetic, IEEE 754-2008, BCD multi-operand addition, carry-ripple adder, pipelined adder trees, FPGA implementation ∗ INRIA, LIP (UMR 5668 CNRS ENS de Lyon INRIA UCBL), Université de Lyon, [email protected] † ENS de Lyon, LIP (UMR 5668 CNRS ENS de Lyon INRIA UCBL), Université de Lyon, [email protected] in ria -0 05 26 32 7, v er si on 1 14 O ct 2 01 0 Addition décimale multi-opérande sur FPGA Résumé : La recherche sur l’implantation en matériel de l’arithmétique décimale est actuellement très active, la plupart des travaux portant sur des opérateurs pour les processeurs, en virgule fixe ou flottante. Mais les techniques développées pour un circuit intégré n’aboutissent pas forcément à une implémentation optimale dans un FPGA. Il n’y a que peu d’études ciblant explicitement les FPGA. Cet article s’intéresse dans ce contexte, à l’addition BCD multiopérande, au cœur de multiplieurs et de multiplieurs-accumulateurs à faible latence. Nous étudions les architectures proposées pour cette opération décimale, et nous observons que, sur FPGA, leur performance (surface et latence) est très inférieure à celle des opérations binaire à précision comparable. Nous présentons donc dans cet article une nouvelle technique d’addition BCD multi-opérandes qui s’avère plus efficace que les propositions précédentes sur les FPGA actuels. Elle s’adapte particulièrement bien à la structure fine des FPGA Xilinx Virtex5/Virtex-6, et se prête bien au pipeline. Les résultats de synthèse montrent que notre implémentation divise par deux la surface et la latence par rapport aux propositions précédentes, les ramenant à des valeurs comparables à celles des meilleurs additionneurs multi-opérandes binaires. Mots-clés : Arithmétique décimale, IEEE 754-2008, addition BCD multiopérande, additionneur à propagation de retenue, arbres d’addition pipelinés, implémentation sur FPGA in ria -0 05 26 32 7, v er si on 1 14 O ct 2 01 0 Multi-operand Decimal Adder Trees for FPGAs 3

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Fast Signed-Digit Multi-operand Decimal Adders

Decimal arithmetic is desirable for high precision requirements of many financial, industrial and scientific applications. Furthermore, hardware support for decimal arithmetic has gained momentum with IEEE 7542008, which standardized decimal floating-point. This paper presents a new architecture for two operand and multi-operand signed-digit decimal addition. Signed-digit architectures are adva...

متن کامل

Efficient High Speed Compression Trees on Xilinx FPGAs

Compressor trees are efficient circuits to realize multi-operand addition with fast carry-save arithmetic. They can be found in various arithmetic applications like multiplication, squaring and the evaluation of polynomials with application to function approximation. Finding good elementary compressing elements on FPGAs is a non-trivial task as an efficient mapping to look-up tables and carry-c...

متن کامل

Delay Redundant Multi Operand Adder Using Cyclic Combinational Method

In this paper, we propose the efficient implementation of multioperand redundant compressor trees in modern FPGAs by using their fast carry resources. Our approaches strongly reduce delay and they generally present no area overhead compared to a CPA tree. Moreover, they could be defined at a high level based on an array of standard CPAs. As a consequence, they are compatible with any FPGA famil...

متن کامل

From a Redundant Number System Adder to a Novel 4-2 Compressor

We present a novel 4-2 compressor for designing multi-operand addition circuits. The design is based on a current mode circuit from which the equivalent voltage mode circuit is derived. This circuit can be used to build very fast multi-operand addition structures. We provide a gate level cost and speed comparison between our design and existing 4-2 compressors. Also, we make an analytical compa...

متن کامل

A fully redundant decimal adder and its application in parallel decimal multipliers

Decimal hardware arithmetic units have recently regained popularity, as there is now a high demand for high performance decimal arithmetic. We propose a novel method for carry-free addition of decimal numbers, where each equally weighted decimal digit pair of the two operands is partitioned into two weighted bit-sets. The arithmetic values of these bit-sets are evaluated, in parallel, for fast ...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2010