Octuple-precision floating-point format: Difference between revisions

Content deleted Content added
Undid revision 700870010 by Ansh666 (talk) English please, see your talk page. Consistency is important.
m Reverted edit by 2A01:CB16:201A:C683:0:67:3AB1:9A01 (talk) to last version by Vincent Lefèvre
 
(49 intermediate revisions by 34 users not shown)
Line 1:
{{short description|256-bit computer number format}}
<!-- Please do not remove or change this AfD message until the issue is settled -->
{{use dmy dates|date=December 2022|cs1-dates=y}}
{{Article for deletion/dated|page=Octuple-precision floating-point format|timestamp=20150623012026|year=2015|month=June|day=23|substed=yes|help=off}}
{{use list-defined references|date=December 2022}}
<!-- Once discussion is closed, please place on talk page: {{Old AfD multi|page=Octuple-precision floating-point format|date=23 June 2015|result='''keep'''}} -->
<!-- End of AfD message, feel free to edit beyond this point -->
In [[computing]], '''octuple precision''' is a binary [[floating-point]]-based [[computer number format]] that occupies 32 [[byte]]s (256 [[bit]]s) in computer memory. This 256-[[bit]] octuple precision is for applications requiring results in higher than [[quadruple precision]]. This format is rarely (if ever) used and very few things support it.
{{Floating-point}}
{{Computer architecture bit widths}}
In [[computing]], '''octuple precision''' is a binary [[floating-point]]-based [[computer number format]] that occupies 32 [[byte]]s (256 [[bit]]s) in computer memory. This 256-[[bit]] octuple precision is for applications requiring results in higher than [[quadruple precision]]. This format is rarely (if ever) used and very few things support it.
 
The range greatly exceeds what is needed to describe all known physical limitations within the observable universe or precisions better than [[Planck units]].
== IEEE 754 octuple-precision binary floating-point format: binary256 ==
 
== IEEE 754 octuple-precision binary floating-point format: binary256 ==
In its 2008 revision, the [[IEEE 754]] standard specifies a '''binary256''' format among the ''interchange formats'' (it is not a basic format), as having:
* [[Sign bit]]: 1 bit
Line 18 ⟶ 19:
The bits are laid out as follows:
 
[[File:Octuple persisionprecision visual demontrationdemonstration.pngsvg|1000px|Layout of octuple -precision floating -point format]]
 
=== Exponent encoding ===
 
The octuple-precision binary floating-point exponent is encoded using an [[offset binary]] representation, with the zero offset being 262143; also known as exponent bias in the IEEE&nbsp;754 standard.
 
Line 28:
* [[Exponent bias]] = 3FFFF<sub>16</sub> = 262143
 
Thus, as defined by the offset binary representation, in order to get the true exponent the offset of 16383262143 has to be subtracted from the stored exponent.
 
The stored exponents 00000<sub>16</sub> and 7FFFF<sub>16</sub> are interpreted specially.
 
{| class="wikitable" style="text-align: center;"
|-
! Exponent !! Significand zero !! Significand non-zero !! Equation
|-
| 00000<sub>16</sub> || [[0 (number)|0]], [[−0]] || [[subnormal numbers]] || <math>(-1−1)^{\text{<sup>signbit}}</sup> \times× 2^{-262142}<sup>−262142</sup> \times× 0.\text{significandbits}_2<sub>2</mathsub>
|-
| 00001<sub>16</sub>, ..., 7FFFE<sub>16</sub> ||colspan=2| normalized value || <math>(-1−1)^{\text{<sup>signbit}}</sup> \times× 2^{{\text{exponentbits}_2}<sup>exponent -bits<sub>2</sub></sup> 262143} \times× 1.\text{significandbits}_2<sub>2</mathsub>
|-
| 7FFFF<sub>16</sub> || ±[[infinity|∞]] || [[NaN]] (quiet, signallingsignaling)
|}
 
Line 47 ⟶ 48:
 
=== Octuple-precision examples ===
 
These examples are given in bit ''representation'', in [[hexadecimal]],
of the floating-point value. This includes the sign, (biased) exponent, and significand.
 
0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 <sub>16</sub> = +0
8000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 <sub>16</sub> = −0
 
7fff f000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 <sub>16</sub> = +infinity
ffff f000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 <sub>16</sub> = −infinity
 
0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0001<sub>16</sub>
= 2<sup>−262142</sup> × 2<sup>−236</sup> = 2<sup>−262378</sup>
≈ 2.24800708647703657297018614776265182597360918266100276294348974547709294462 × 10<sup>−78984</sup>
(smallest positive subnormal number)
 
0000 0fff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff<sub>16</sub>
= 2<sup>−262142</sup> × (1 − 2<sup>−236</sup>)
≈ 2.4824279514643497882993282229138717236776877060796468692709532979137875392 × 10<sup>−78913</sup>
(largest subnormal number)
 
0000 1000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000<sub>16</sub>
= 2<sup>−262142</sup>
≈ 2.48242795146434978829932822291387172367768770607964686927095329791378756168 × 10<sup>−78913</sup>
(smallest positive normal number)
 
7fff efff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff<sub>16</sub>
= 2<sup>262143</sup> × (2 − 2<sup>−236</sup>)
≈ 1.61132571748576047361957211845200501064402387454966951747637125049607182699 × 10<sup>78913</sup>
(largest normal number)
 
3fff efff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff ffff<sub>16</sub>
= 1 − 2<sup>−237</sup>
≈ 0.999999999999999999999999999999999999999999999999999999999999999999999995472
(largest number less than one)
 
3fff f000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000<sub>16</sub>
= 1 (one)
 
3fff f000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0001<sub>16</sub>
= 1 + 2<sup>−236</sup>
≈ 1.00000000000000000000000000000000000000000000000000000000000000000000000906
(smallest number larger than one)
 
By default, 1/3 rounds down like [[double precision]], because of the odd number of bits in the significand.
So the bits beyond the rounding point are <code>0101...</code> which is less than 1/2 of a [[unit in the last place]].
 
== Implementations ==
Octuple precision is rarely implemented since usage of it is extremely rare. [[Apple Inc.]] had an implementation of addition, subtraction and multiplication of octuple-precision numbers with a 224-bit [[two's complement]] significand and a 32-bit exponent.<ref>{{cite web | urlname=http://images.apple.com/ca/acg/pdf/oct3a.pdf | title=Octuple"Crandall-precision floating point on Apple G4 | author1=R. Crandall | author2=J. Papadopoulos | date=8 May 2002}}<Papadopoulos_2002"/ref> One can use general [[arbitrary-precision arithmetic]] libraries to obtain octuple (or higher) precision, but specialized octuple-precision implementations may achieve higher performance.
 
=== Hardware support ===
There is littleno to noknown hardware with native support for it.octuple Octuple-precision arithmetic is too impractical for most commercial uses of it, making implementation of it very rare (if any).
 
==Processing Statistics==
Since an octuple precision numeral takes up 32 bytes of storage, the requirements in-order to transport this piece of data are as follows.
* [[8-bit|8-bit architecture]] – 32 separate packets<sup>1</sup> of information (at least) in order to transport this across the main data bus
* [[16-bit|x16 architecture]] – 16 separate packets of information (at least) in order to transport this across the main data bus
* [[x86|x86 architecture]] – 8 separate packets of information (at least) in order to transport this across the main data bus
* [[x86-64|x64 architecture]] – 4 separate packets of information (at least) in order to transport this across the main data bus
<br />
<sup>1</sup>statistical extrapolation since it would take 1/8 of the entire memory just to store 1 binary256 numeral, making it completely impractical.
 
== See also ==
* [[IEEE 754]]
* [[IEEE 754-2008|IEEE Standard for Floating-Point Arithmetic (IEEE 754)]]
* [[ISO/IEC 10967]], Language-independent arithmetic
* [[Primitive data type]]
* [[Scientific notation]]
 
== References ==
{{reflist}}|refs=
<ref name="Crandall-Papadopoulos_2002">{{cite web |title=Octuple-precision floating point on Apple G4 (archived copy on web.archive.org) |author-first1=Richard E. |author-last1=Crandall |author-link1=Richard E. Crandall |author-first2=Jason S. |author-last2=Papadopoulos |date=2002-05-08 |url=http://images.apple.com/ca/acg/pdf/oct3a.pdf |url-status=unfit |archive-url=https://web.archive.org/web/20060728140052/http://images.apple.com/ca/acg/pdf/oct3a.pdf |archive-date=2006-07-28}} (8 pages)</ref>
}}
 
== Further reading ==
* {{cite book |author-first=Nelson H. F. |author-last=Beebe |title=The Mathematical-Function Computation Handbook - Programming Using the MathCW Portable Software Library |date=2017-08-22 |___location=Salt Lake City, UT, USA |publisher=[[Springer International Publishing AG]] |edition=1 |lccn=2017947446 |isbn=978-3-319-64109-6 |doi=10.1007/978-3-319-64110-2 }}
 
{{data types}}