American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....semantics. So, the rigorous definition of the program concrete semantics is mandatory for all formal methods. In practice, the concrete semantics is defined by: the ISO IEC 9899 standard for the C programming language [14] as well as the ANSI IEEE 754 standard for floating point arithmetic [2]; the compiler and machine implementation of these standards; the end user expectations. Each semantics is a refinement of the previous one where some non determinism is resolved. The number of lines of code (LOCs) is counted with the Unix TM command wc l after stripping comments out ....

....Arithmetic A major di#culty of the analysis of floating point arithmetic is the rounding errors, both in the analyzed semantics and in the analyzer itself. One has to consider that: transfer functions should model floating point arithmetic, that is to say (according to the IEEE standard [2]) infinite precision real arithmetic followed by a rounding phase; abstract operators should be themselves implemented using floating point arithmetic (for e#ciency, arbitrary precision floating point, rational, and algebraic arithmetics should be prohibited) In particular, special care has ....

American National Standards Institute, Inc. IEEE standard for binary floating-point arithmetic. Technical Report 754-1985, ANSI/IEEE, 1985. http://grouper.ieee.org/groups/754/.

....when we stick to floating point numbers, the mathematically correct result of an arithmetic operation will in general not be representable as a floating point number. Consequently, floating point arithmetic can always be just an approximation to the arithmetic on the reals. The IEEE Standard 754 [4] specifies such a floating point arithmetic in a way that the result of the floating point operation is closest to the mathematically correct result. Note that, in addition to overflow, we may now also experience underflow. The discrepancy between mathematical arithmetic operations and their ....

American National Standard Institute. IEEE standard for binary floating point arithmetic IEEE/ANSI 754-1985. Technical report, New York, 1985.

....to use two tables for u and t. The entries in u are simply the power of 2 such that 1 ur 2 2. The entries in t are u 3=2 . The only problem is to figure out which table entries to use for a given input value. The range reduction I use is based on the IEEE double precision number format. [6] Each number consists of 64 bits one sign bit, 11 exponent bits, and 52 fraction bits. In addition, there is an implicit 1 bit for normalized numbers. If I know the argument is positive, as it must be for the function I am interested in, I can extract the exponent by shifting the high order 32 ....

....input at the cost of one additional cycle. However, the output value is accurate for any floating point input. Very large input values produce denormalized results; very small input values produce floating point infinity as they should. If your machine supports the IEEE double extended format,[6] a format with at least 64 bits in the fraction that is usually reserved for data kept in the registers, you can get the last few bits right using a simple trick. Compute the array t in extended precision, but store it as two double precision numbers, t(1,i) and t(2,i) Then the final scaling ....

American National Standards Institute, Inc. IEEE Standard for Binary FloatingPoint Arithmetic. Technical Report ANSI/IEEE Std 754-

....and to launch the discussion towards a standard. 1 Introductory discussion We take the opportunity of the current discussion on the revision of the IEEE 754 Standard for Floating Point Arithmetic to discuss the possibility of standardizing (some of) the elementary functions. The IEEE 754 Standard [1] does not deal with these functions. This is due to several reasons, the most serious one being the Table Maker s Dilemma (TMD for the sake of brevity) which is the problem of providing correctly rounded transcendentals. To quote Kahan [4] Committing an approximation function to hardware is ....

....f(x) should be preserved 3 . Level 1 there is a domain (usually around zero) where the implemented function is correctly rounded. Outside this domain, the implementation should satisfy the criteria of Level 0. We suggest the domain should at least contain [ 2#, 2#] for sin, cos and tan; and [ 1, 1] for exp, cosh, sinh, 2 x and 10 x (other functions: to be discussed. Compromise involving facility of implementation and usefulness of requirement) 3 In practice this requirement is not a problem: function implementers will use these symmetries for simplifying their programs. 5 Level 2 ....

American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

....N D q 4 (1 2 8p 4 ) For instance, if p = 9, the error is bounded by 2 76 . This method has a quadratic convergence: at each step, the number of significant bits of the approximation to the quotient roughly doubles. Getting correctly rounded results, as required by the IEEE 754 standard [8], may seem less straightforward than with the digit recurrence methods. And yet, many studies performed during the past recent years [1, 9, 15] show that with some care this can be done easily, for division as well as for square root. See [1, 12] for more details. 3 1 4 5 6 2 3 7 8 9 10 11 17 16 ....

American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

....de tables, Arrondi correct, Virgule flottante. 1 Introduction In general, the result of an arithmetic operation on two floating point numbers is not exactly representable in the same floating point format: it must be rounded. In a floating point system that follows the IEEE 754 standard [2, 3, 6], the user can choose an active rounding mode from: rounding towards 1, rounding towards 1, rounding towards 0 and rounding to the nearest (with a special convention if x is exactly between two machine numbers) The IEEE 754 standard requires that the system should behave as if the result of an ....

American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

....b is the base of the floating point system; d 0 : d 1 d 2 : dm is the significand , 0 d i b Gamma 1, d 0 6= 0; the integer e is the exponent , satisfying E min e E max . The floating point format (1) is characterized by four integer parameters: b, m, E min and E max : The IEEE Standard [2] uses the terms single, double and extended to denote different floatingpoint formats and the corresponding precisions. Since we use more precisions than specified by the IEEE Standard and provide a simple method of precision control, we denote the precisions by powers of two: 1; 2; 4; 8; ....

American National Standards Institute, New York. IEEE Standard for Binary FloatingPoint Arithmetic, 1985. ANSI/IEEE Std. 754--1985.

....cycles for each method, assuming that an addition, a multiplication and a shift are performed in one clock cycle whereas a division is five times longer. Estimated times for three different kinds of processors are presented in appendix. An IEEE standard floating point number is stored as follows ([IEEE 754 (1985)] sign 1 bit biased exp. 8 bits mantissa 1 hidden bit 23 bits Figure 3: A single precision IEEE floating point number. sign 1 bit biased exp. mantissa 11 bits 1 hidden bit 52 bits Figure 4: A double precision IEEE floating point number. Relative error has to be less than 2 Gamma24 for the ....

American National Standards Institute and Institute of Electrical and Electronic Engineers, IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Standard 754-1985, New York, 1985.

....that can work with higher precision. ffl Getting a discrete world, or in other words, work in the computer only with discrete representations of it. These approaches must be careful with the growing or changing of the numbers that can go farther than was expected. The generated sequence is: se[1]=10.000000 se[2] 100.000000 se[3] 1000.000000 se[4] 10000.000000 se[5] 100000.000000 se[6] 1000000.000000 se[7] 10000000.000000 se[8] 100000000.000000 se[9] 1000000000.000000 se[10] 10000000000.000000 se[11] 99999997952.000000 se[12] 999999995904.000000 se[13] 9999999827968.000000 ....

....res=res se[i] printf( n First way: f n , res) res=se[LONGSEQ 1] for(j=LONGSEQ 2; j =0; j ) res=res se[j] printf( n Second way: f n , res) Figure 3: The source code that implements a finite growing sequence of numbers. The generated sequence is: se[0] 0. 0000000000000000001000 se[1]= 0.0000000000000000002500 se[2] 0.0000000000000000006250 se[3] 0.0000000000000000015625 se[4] 0.0000000000000000039062 se[5] 0.0000000000000000097656 se[6] 0.0000000000000000244141 se[7] 0.0000000000000000610351 se[8] 0.0000000000000001525879 se[9] 0.0000000000000003814697 se[10] ....

[Article contains additional citation context not shown here]

American National Standards Institute (ANSI), Institute of Electrical, and Electronic Engineers (IEEE). IEEE standard for binary floatingpoint arithmetic. Technical Report Std 754-1985, ANSI/IEEE Standard, New York, 1985.

.... of our approach by applying it to the design of the Floating Point Arithmetic and Logic Unit (FP ALU) for a DLX 32 bit RISC Microprocessor [10] This is a modular design based on the requirements for FP functionality as contained in the ANSI IEEE Standard for Binary Floating Point Arithmetic [2], and those which can be derived from standard understandings of computer arithmetic [16] The rest of the paper is organized as follows. Section 2 presents background to this work in the area of design process modeling, particularly the Design As Exploration information process model. Section 3 ....

American National Standards Institute and the IEEE Standards Board, New York, NY. IEEE Standard for Binary Floating-Point Arithmetic, 1985. ANSI/IEEE Std 754-1985.

....the implementation of roundoff in the FPAU. Our goal is to evaluate the performance of this isolated part of the design, and compare it to its requirements. Many of the requirements for roundoff performance are contained in the ANSI IEEE Standard for Binary Floating Point Arithmetic [2]. Other requirements can be derived from standard understandings of computer arithmetic, all of which we capture in the requirements model. Figure 4 shows a fragment of the requirements database, including some line item and RSL representations for requirements for floating point number formats. ....

American National Standards Institute and the IEEE Standards Board, New York, NY. IEEE Standard for Binary Floating-Point Arithmetic, 1985. ANSI/IEEE Std 754-1985.

....data model is shown in Figure 2. Here we introduce a part of the Floating Point Arithmetic and Logic Unit (FP ALU) for a DLX 32 bit RISC Microprocessor [6] The design is based on the requirements for FP functionality as contained in the ANSI IEEE Standard for Binary FloatingPoint Arithmetic [2], and those which can be derived from standard understandings of computer arithmetic [11] We discuss aspects of the FP ALU example in more detail in Sections IV and V. Architectural Design Entities Pre normalization ion A ite precision B d to infinite precision C Version Link Configuration Link ....

American National Standards Institute and the IEEE Standards Board, New York, NY. IEEE Standard for Binary Floating-Point Arithmetic, 1985. ANSI/IEEE Std 754-1985.

....other bases than 2 are very rare. Moreover, the algorithm can be formulated in a very neat and compact form for this special case. Hence we understand by machine independence that the algorithm is defined relative to an abstract computer arithmetic definition, e.g. the IEEE floating point format [IEEE85]. The independence concerns the architecture of the hardware platform. The definition of a standard binary computer arithmetic includes the size r of the mantissa, the exponent range, and the basic operators , Gamma , and = that compute a result with machine precision, i.e. with ....

....In order to make the algorithm available to a wide domain of users, a short and efficient implementation in C [Ker83] is offered by the author. If you like a copy of the source file send your request via e mail to kobbelt ira.uka.de. The underlying arithmetic is the IEEE standard double format [IEEE85]. If other arithmetics are used, only the few defines in the preamble of the code have to be changed which access the various data fields of the IEEE floating point numbers. The data structure holding the addends which have not yet found their partner for the addition without rounding error are ....

American National Standards Institute / Institute of Electrical and Electronical Engineers, IEEE Standard for Binary Floating-Point Arithmetic, ANSI/IEEE Std. 754-1985, New York, 1985

....the base of the floating point system; d 0 :d 1 d 2 . dm is the significand 1 , 0 d i b Gamma 1, d 0 6= 0; the integer e is the exponent, satisfying E min e E max . The floating point format (2. 1) is characterized by four integer parameters: b, m, E min and E max : The IEEE Standard [1] uses the terms single, double and extended to denote different floating point formats and the corresponding precisions. Since our purpose 1 Note that significand denotes the part of the floating point format (2.1) which contains (m 1) significant digits, d 0 ; d 1 ; dm . is to use more ....

....denote different floating point formats and the corresponding precisions. Since our purpose 1 Note that significand denotes the part of the floating point format (2.1) which contains (m 1) significant digits, d 0 ; d 1 ; dm . is to use more precisions than specified by the IEEE Standard [1] and to provide a simple method of precision control, we denote the precisions by the powers of two: 1; 2; 4; 8; Our implementation of variable precision arithmetic has the following properties: 1. the base of the floating point system is 2; 2. precision 1 is the same as normalized single ....

[Article contains additional citation context not shown here]

American National Standards Institute, New York. IEEE standard for binary floating-point arithmetic, 1985. ANSI/IEEE Std. 754--1985.

No context found.

American National Standards Institute, Institute of Electrical, and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

....path. Because the result of subtraction is positive, the adder and the LOP are simpler, resulting in a smaller critical path [10] We consider this LOP as our starting point. Moreover, we consider operands represented in signand magnitude since this is the representation used by the IEEE standard [1]. It is worth noticing that the scheme proposed in [10] is not applicable for the case in which the floating point adder is designed with two parallel paths, close and far, depending on the exponent difference. In this case, the 1 The LOP is also called LZA (Leading Zero Anticipator) add sub. ....

American National Standard Institute and Institute of Electrical and Electronic Engineers. IEEE Standard for binary Floating--Point Arithmetic. ANSI/IEEE Standard, std. 745--1985. (1985).

....explained only by a few examples. The LOP schemes presented previously consider the operands to be either in two s complement representation [10] or in sign and magnitude [3, 11] We discuss the case for sign and magnitude representation, since this is the representation used by the IEEE standard [1]. In Section 2 the structure of the LOP we propose is described. Then, in following sections, each part of the LOP is analyzed: the algorithm for the encoding of the leading position is described in detail in Section 3 and the concurrent position correction in Section 4. Finally, the LOP is ....

American National Standard Institute and Institute of Electrical and Electronic Engineers. IEEE Standard for binary Floating--Point Arithmetic. ANSI/IEEE Standard, std. 745-- 1985. (1985).

No context found.

American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floating-point arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

No context found.

American National Standards Institute and Institute of Electrical and Electronic Engineers. IEEE standard for binary floatingpoint arithmetic. ANSI/IEEE Standard, Std 754-1985, New York, 1985.

No context found.

American National Standards Institute & Institute of Electrical and Electronics Engineers, "IEEE Standard for Binary Floating Point Arithmetic," ANSI/IEEE Standard, Std 754-1985, New York, 1985.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC