SherbyNet - /page/wikibook/entry.mu

Micron Document

<!DOCTYPE html>

<title>Floating-point arithmetic</title>

</head>

Floating-point arithmetic

</h1>

</header>

/* start https://en.wikipedia.org/ */

.mw-parser-output .hatnote{font-style:italic}.mw-parser-output div.hatnote{padding-left:1.6em;margin-bottom:0.5em}.mw-parser-output .hatnote i{font-style:normal}.mw-parser-output .hatnote+link+.hatnote{margin-top:-0.5em}@media print{body.ns-0 .mw-parser-output .hatnote{display:none!important}}

/* end https://en.wikipedia.org/ */

</style><div role="note" class="hatnote navigation-not-searchable">"Floating point" redirects here. For other uses, see <a href="Floating_point_(disambiguation)" class="mw-disambig" title="Floating point (disambiguation)">Floating point (disambiguation)</a>.</div>

/* start https://en.wikipedia.org/ */

.mw-parser-output .hlist dl,.mw-parser-output .hlist ol,.mw-parser-output .hlist ul{margin:0;padding:0}.mw-parser-output .hlist dd,.mw-parser-output .hlist dt,.mw-parser-output .hlist li{margin:0;display:inline}.mw-parser-output .hlist.inline,.mw-parser-output .hlist.inline dl,.mw-parser-output .hlist.inline ol,.mw-parser-output .hlist.inline ul,.mw-parser-output .hlist dl dl,.mw-parser-output .hlist dl ol,.mw-parser-output .hlist dl ul,.mw-parser-output .hlist ol dl,.mw-parser-output .hlist ol ol,.mw-parser-output .hlist ol ul,.mw-parser-output .hlist ul dl,.mw-parser-output .hlist ul ol,.mw-parser-output .hlist ul ul{display:inline}.mw-parser-output .hlist .mw-empty-li{display:none}.mw-parser-output .hlist dt::after{content:": "}.mw-parser-output .hlist dd::after,.mw-parser-output .hlist li::after{content:" · ";font-weight:bold}.mw-parser-output .hlist dd:last-child::after,.mw-parser-output .hlist dt:last-child::after,.mw-parser-output .hlist li:last-child::after{content:none}.mw-parser-output .hlist dd dd:first-child::before,.mw-parser-output .hlist dd dt:first-child::before,.mw-parser-output .hlist dd li:first-child::before,.mw-parser-output .hlist dt dd:first-child::before,.mw-parser-output .hlist dt dt:first-child::before,.mw-parser-output .hlist dt li:first-child::before,.mw-parser-output .hlist li dd:first-child::before,.mw-parser-output .hlist li dt:first-child::before,.mw-parser-output .hlist li li:first-child::before{content:" (";font-weight:normal}.mw-parser-output .hlist dd dd:last-child::after,.mw-parser-output .hlist dd dt:last-child::after,.mw-parser-output .hlist dd li:last-child::after,.mw-parser-output .hlist dt dd:last-child::after,.mw-parser-output .hlist dt dt:last-child::after,.mw-parser-output .hlist dt li:last-child::after,.mw-parser-output .hlist li dd:last-child::after,.mw-parser-output .hlist li dt:last-child::after,.mw-parser-output .hlist li li:last-child::after{content:")";font-weight:normal}.mw-parser-output .hlist ol{counter-reset:listitem}.mw-parser-output .hlist ol>li{counter-increment:listitem}.mw-parser-output .hlist ol>li::before{content:" "counter(listitem)"\a0 "}.mw-parser-output .hlist dd ol>li:first-child::before,.mw-parser-output .hlist dt ol>li:first-child::before,.mw-parser-output .hlist li ol>li:first-child::before{content:" ("counter(listitem)"\a0 "}

/* end https://en.wikipedia.org/ */

</style><style data-mw-deduplicate="TemplateStyles:r1126788409">

/* start https://en.wikipedia.org/ */

.mw-parser-output .plainlist ol,.mw-parser-output .plainlist ul{line-height:inherit;list-style:none;margin:0;padding:0}.mw-parser-output .plainlist ol li,.mw-parser-output .plainlist ul li{margin-bottom:0}

/* end https://en.wikipedia.org/ */

</style><style data-mw-deduplicate="TemplateStyles:r1246091330">

/* start https://en.wikipedia.org/ */

.mw-parser-output .sidebar{width:22em;float:right;clear:right;margin:0.5em 0 1em 1em;background:var(--background-color-neutral-subtle,#f8f9fa);border:1px solid var(--border-color-base,#a2a9b1);padding:0.2em;text-align:center;line-height:1.4em;font-size:88%;border-collapse:collapse;display:table}body.skin-minerva .mw-parser-output .sidebar{display:table!important;float:right!important;margin:0.5em 0 1em 1em!important}.mw-parser-output .sidebar-subgroup{width:100%;margin:0;border-spacing:0}.mw-parser-output .sidebar-left{float:left;clear:left;margin:0.5em 1em 1em 0}.mw-parser-output .sidebar-none{float:none;clear:both;margin:0.5em 1em 1em 0}.mw-parser-output .sidebar-outer-title{padding:0 0.4em 0.2em;font-size:125%;line-height:1.2em;font-weight:bold}.mw-parser-output .sidebar-top-image{padding:0.4em}.mw-parser-output .sidebar-top-caption,.mw-parser-output .sidebar-pretitle-with-top-image,.mw-parser-output .sidebar-caption{padding:0.2em 0.4em 0;line-height:1.2em}.mw-parser-output .sidebar-pretitle{padding:0.4em 0.4em 0;line-height:1.2em}.mw-parser-output .sidebar-title,.mw-parser-output .sidebar-title-with-pretitle{padding:0.2em 0.8em;font-size:145%;line-height:1.2em}.mw-parser-output .sidebar-title-with-pretitle{padding:0.1em 0.4em}.mw-parser-output .sidebar-image{padding:0.2em 0.4em 0.4em}.mw-parser-output .sidebar-heading{padding:0.1em 0.4em}.mw-parser-output .sidebar-content{padding:0 0.5em 0.4em}.mw-parser-output .sidebar-content-with-subgroup{padding:0.1em 0.4em 0.2em}.mw-parser-output .sidebar-above,.mw-parser-output .sidebar-below{padding:0.3em 0.8em;font-weight:bold}.mw-parser-output .sidebar-collapse .sidebar-above,.mw-parser-output .sidebar-collapse .sidebar-below{border-top:1px solid #aaa;border-bottom:1px solid #aaa}.mw-parser-output .sidebar-navbar{text-align:right;font-size:115%;padding:0 0.4em 0.4em}.mw-parser-output .sidebar-list-title{padding:0 0.4em;text-align:left;font-weight:bold;line-height:1.6em;font-size:105%}.mw-parser-output .sidebar-list-title-c{padding:0 0.4em;text-align:center;margin:0 3.3em}@media(max-width:640px){body.mediawiki .mw-parser-output .sidebar{width:100%!important;clear:both;float:none!important;margin-left:0!important;margin-right:0!important}}body.skin--responsive .mw-parser-output .sidebar a>img{max-width:none!important}@media screen{html.skin-theme-clientpref-night .mw-parser-output .sidebar:not(.notheme) .sidebar-list-title,html.skin-theme-clientpref-night .mw-parser-output .sidebar:not(.notheme) .sidebar-title-with-pretitle{background:transparent!important}html.skin-theme-clientpref-night .mw-parser-output .sidebar:not(.notheme) .sidebar-title-with-pretitle a{color:var(--color-progressive)!important}}@media screen and (prefers-color-scheme:dark){html.skin-theme-clientpref-os .mw-parser-output .sidebar:not(.notheme) .sidebar-list-title,html.skin-theme-clientpref-os .mw-parser-output .sidebar:not(.notheme) .sidebar-title-with-pretitle{background:transparent!important}html.skin-theme-clientpref-os .mw-parser-output .sidebar:not(.notheme) .sidebar-title-with-pretitle a{color:var(--color-progressive)!important}}@media print{body.ns-0 .mw-parser-output .sidebar{display:none!important}}

/* end https://en.wikipedia.org/ */

</style><table class="sidebar nomobile nowraplinks plainlist"><tbody><tr><th class="sidebar-title"> <a href="Computer_number_format" title="Computer number format">formats</a></th></tr><tr><th class="sidebar-heading">

<ul><li>16-bit: <a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half</a> (binary16)</li>

<li>32-bit: <a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single</a> (binary32), <a href="Decimal32_floating-point_format" title="Decimal32 floating-point format">decimal32</a></li>

<li>64-bit: <a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double</a> (binary64), <a href="Decimal64_floating-point_format" title="Decimal64 floating-point format">decimal64</a></li>

<li>128-bit: <a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">Quadruple</a> (binary128), <a href="Decimal128_floating-point_format" title="Decimal128 floating-point format">decimal128</a></li>

<li>256-bit: <a href="Octuple-precision_floating-point_format" title="Octuple-precision floating-point format">Octuple</a> (binary256)</li>

<li><a href="Extended_precision" title="Extended precision">Extended precision</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

Other</th></tr><tr><td class="sidebar-content">

<ul><li><a href="Minifloat" title="Minifloat">Minifloat</a></li>

<li><a href="Bfloat16_floating-point_format" title="Bfloat16 floating-point format">bfloat16</a></li>

<li><a href="TensorFloat-32" title="TensorFloat-32">TensorFloat-32</a></li>

<li><a href="Microsoft_Binary_Format" title="Microsoft Binary Format">Microsoft Binary Format</a></li>

<li><a href="IBM_hexadecimal_floating-point" title="IBM hexadecimal floating-point">IBM hexadecimal floating-point</a></li>

<li><a href="Power_Management_Bus#Linear11_Floating-Point_Format" title="Power Management Bus">PMBus Linear-11</a></li>

<li><a href="G.711" title="G.711">G.711 8-bit floats</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

Alternatives</th></tr><tr><td class="sidebar-content">

<ul><li><a href="Arbitrary-precision_arithmetic" title="Arbitrary-precision arithmetic">Arbitrary precision</a></li>

<li><a href="Block_floating_point" title="Block floating point">Block floating point</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

<a href="Tapered_floating_point" title="Tapered floating point">Tapered floating point</a></th></tr><tr><td class="sidebar-content">

<ul><li><a href="Unum_(number_format)" title="Unum (number format)">Posit</a></li></ul></td>

</tr><tr><td class="sidebar-navbar"><style data-mw-deduplicate="TemplateStyles:r1239400231">

/* start https://en.wikipedia.org/ */

.mw-parser-output .navbar{display:inline;font-size:88%;font-weight:normal}.mw-parser-output .navbar-collapse{float:left;text-align:left}.mw-parser-output .navbar-boxtext{word-spacing:0}.mw-parser-output .navbar ul{display:inline-block;white-space:nowrap;line-height:inherit}.mw-parser-output .navbar-brackets::before{margin-right:-0.125em;content:"[ "}.mw-parser-output .navbar-brackets::after{margin-left:-0.125em;content:" ]"}.mw-parser-output .navbar li{word-spacing:-0.125em}.mw-parser-output .navbar a>span,.mw-parser-output .navbar a>abbr{text-decoration:inherit}.mw-parser-output .navbar-mini abbr{font-variant:small-caps;border-bottom:none;text-decoration:none;cursor:inherit}.mw-parser-output .navbar-ct-full{font-size:114%;margin:0 7em}.mw-parser-output .navbar-ct-mini{font-size:114%;margin:0 4em}html.skin-theme-clientpref-night .mw-parser-output .navbar li a abbr{color:var(--color-base)!important}@media(prefers-color-scheme:dark){html.skin-theme-clientpref-os .mw-parser-output .navbar li a abbr{color:var(--color-base)!important}}@media print{.mw-parser-output .navbar{display:none!important}}

/* end https://en.wikipedia.org/ */

</style></td></tr></tbody></table>

In <a href="Computing" title="Computing">computing</a>, floating-point arithmetic (FP) is <a href="Arithmetic" title="Arithmetic">arithmetic</a> on subsets of <a href="Real_number" title="Real number">real numbers</a> formed by a <a href="Significand" title="Significand">significand</a> (a <a href="Sign_(mathematics)" title="Sign (mathematics)">signed</a> sequence of a fixed number of digits in some <a href="Radix" title="Radix">base</a>) multiplied by an <a href="Integer_power" class="mw-redirect" title="Integer power">integer power</a> of that base.

Numbers of this form are called floating-point numbers.<a href="#cite_note-Muller_2010-1">[1]</a>: 3 <a href="#cite_note-sterbenz1974fpcomp-2">[2]</a>: 10

For example, the number 2469/200 is a floating-point number in base ten with five digits:

</mrow>

</munder>

</mrow>

<mtext>significand</mtext>

</mrow>

</munder>

</munder>

</mrow>

</mrow>

</munder>

<mover>

<mover>

<msup>

</mrow>

</mrow>

</msup>

</mover>

</mrow>

<mtext>exponent</mtext>

</mrow>

</mover>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 2469/200=12.345=\!\underbrace {12345} _{\text{significand}}\!\times \!\underbrace {10} _{\text{base}}\!\!\!\!\!\!\!\overbrace {{}^{-3}} ^{\text{exponent}}}</annotation>

</semantics>

</math>

However, 7716/625 = 12.3456 is not a floating-point number in base ten with five digits—it needs six digits.

The nearest floating-point number with only five digits is 12.346.

And 1/3 = 0.3333… is not a floating-point number in base ten with any finite number of digits.

In practice, most floating-point systems use <a href="Binary_number" title="Binary number">base two</a>, though base ten (<a href="Decimal_floating_point" title="Decimal floating point">decimal floating point</a>) is also common.

Floating-point arithmetic operations, such as addition and division, approximate the corresponding real number arithmetic operations by <a href="Rounding" title="Rounding">rounding</a> any result that is not a floating-point number itself to a nearby floating-point number.<a href="#cite_note-Muller_2010-1">[1]</a>: 22 <a href="#cite_note-sterbenz1974fpcomp-2">[2]</a>: 10

For example, in a floating-point arithmetic with five base-ten digits, the sum 12.345 + 1.0001 = 13.3451 might be rounded to 13.345.

The term floating point refers to the fact that the number's <a href="Radix_point" class="mw-redirect" title="Radix point">radix point</a> can "float" anywhere to the left, right, or between the <a href="Significant_digits" class="mw-redirect" title="Significant digits">significant digits</a> of the number. This position is indicated by the exponent, so floating point can be considered a form of <a href="Scientific_notation" title="Scientific notation">scientific notation</a>.

A floating-point system can be used to represent, with a fixed number of digits, numbers of very different <a href="Orders_of_magnitude_(numbers)" title="Orders of magnitude (numbers)">orders of magnitude</a> — such as the number of meters <a href="Orders_of_magnitude_(length)#100_zettametres" title="Orders of magnitude (length)">between galaxies</a> or <a href="Orders_of_magnitude_(length)#10_femtometres" title="Orders of magnitude (length)">between protons in an atom</a>. For this reason, floating-point arithmetic is often used to allow very small and very large real numbers that require fast processing times. The result of this <a href="Dynamic_range" title="Dynamic range">dynamic range</a> is that the numbers that can be represented are not uniformly spaced; the difference between two consecutive representable numbers varies with their exponent.<a href="#cite_note-Smith_1997-3">[3]</a>

Over the years, a variety of floating-point representations have been used in computers. In 1985, the <a href="IEEE_754" title="IEEE 754">IEEE 754</a> Standard for Floating-Point Arithmetic was established, and since the 1990s, the most commonly encountered representations are those defined by the IEEE.

The speed of floating-point operations, commonly measured in terms of <a href="FLOPS" class="mw-redirect" title="FLOPS">FLOPS</a>, is an important characteristic of a <a href="Computer_system" class="mw-redirect" title="Computer system">computer system</a>, especially for applications that involve intensive mathematical calculations.

Floating-point numbers can be computed using software implementations (softfloat) or hardware implementations (hardfloat). <a href="Floating-point_unit" title="Floating-point unit">Floating-point units</a> (FPUs, colloquially math <a href="Coprocessor" title="Coprocessor">coprocessors</a>) are specially designed to carry out operations on floating-point numbers and are part of most computer systems. When FPUs are not available, software implementations can be used instead.

<div class="mw-heading mw-heading2"><h2 id="Overview">Overview</h2></div>

<div class="mw-heading mw-heading3"><h3 id="Floating-point_numbers">Floating-point numbers</h3></div>

A <a href="Number_representation" class="mw-redirect" title="Number representation">number representation</a> specifies some way of encoding a number, usually as a string of digits.

There are several mechanisms by which strings of digits can represent numbers. In standard mathematical notation, the digit string can be of any length, and the location of the <a href="Radix_point" class="mw-redirect" title="Radix point">radix point</a> is indicated by placing an explicit <a href="Decimal_separator" title="Decimal separator">"point" character</a> (dot or comma) there. If the radix point is not specified, then the string implicitly represents an <a href="Integer" title="Integer">integer</a> and the unstated radix point would be off the right-hand end of the string, next to the least significant digit. In <a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">fixed-point</a> systems, a position in the string is specified for the radix point. So a fixed-point scheme might use a string of 8 decimal digits with the decimal point in the middle, whereby "00012345" would represent 0001.2345.

In <a href="Scientific_notation" title="Scientific notation">scientific notation</a>, the given number is scaled by a <a href="Power_of_10" title="Power of 10">power of 10</a>, so that it lies within a specific range—typically between 1 and 10, with the radix point appearing immediately after the first digit. As a power of ten, the scaling factor is then indicated separately at the end of the number. For example, the orbital period of <a href="Jupiter" title="Jupiter">Jupiter</a>'s moon <a href="Io_(moon)" title="Io (moon)">Io</a> is 152,853.5047 seconds, a value that would be represented in standard-form scientific notation as 1.528535047×105 seconds.

Floating-point representation is similar in concept to scientific notation. Logically, a floating-point number consists of:

<ul><li>A signed (meaning positive or negative) digit string of a given length in a given <a href="Radix" title="Radix">radix</a> (or base). This digit string is referred to as the <a href="Significand" title="Significand">significand</a>, mantissa, or coefficient.<a href="#cite_note-NB_Significand-4">[nb 1]</a> The length of the significand determines the precision to which numbers can be represented. The radix point position is assumed always to be somewhere within the significand—often just after or just before the most significant digit, or to the right of the rightmost (least significant) digit. This article generally follows the convention that the radix point is set just after the most significant (leftmost) digit.</li>

<li>A signed integer <a href="Exponent" class="mw-redirect" title="Exponent">exponent</a> (also referred to as the characteristic, or scale),<a href="#cite_note-NB_Exponent-5">[nb 2]</a> which modifies the magnitude of the number.</li></ul>

To derive the value of the floating-point number, the significand is multiplied by the base raised to the power of the exponent, equivalent to shifting the radix point from its implied position by a number of places equal to the value of the exponent—to the right if the exponent is positive or to the left if the exponent is negative.

Using base-10 (the familiar <a href="Decimal_representation" title="Decimal representation">decimal</a> notation) as an example, the number 152,853.5047, which has ten decimal digits of precision, is represented as the significand 1,528,535,047 together with 5 as the exponent. To determine the actual value, a decimal point is placed after the first digit of the significand and the result is multiplied by 105 to give 1.528535047×105, or 152,853.5047. In storing such a number, the base (10) need not be stored, since it will be the same for the entire range of supported numbers, and can thus be inferred.

Symbolically, this final value is:

<mfrac>

<msup>

</mrow>

</msup>

</mfrac>

</mrow>

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle {\frac {s}{b^{\,p-1}}}\times b^{e},}</annotation>

</semantics>

</math>

where s is the significand (ignoring any implied decimal point), p is the precision (the number of digits in the significand), b is the base (in our example, this is the number ten), and e is the exponent.

Historically, several number bases have been used for representing floating-point numbers, with base two (<a href="Binary_numeral_system" class="mw-redirect" title="Binary numeral system">binary</a>) being the most common, followed by base ten (<a href="Decimal_floating_point" title="Decimal floating point">decimal floating point</a>), and other less common varieties, such as base sixteen (<a href="Hexadecimal_floating_point" title="Hexadecimal floating point">hexadecimal floating point</a><a href="#cite_note-Zehendner_2008-6">[4]</a><a href="#cite_note-Beebe_2017-7">[5]</a><a href="#cite_note-NB_9-8">[nb 3]</a>), base eight (octal floating point<a href="#cite_note-Muller_2010-1">[1]</a><a href="#cite_note-Beebe_2017-7">[5]</a><a href="#cite_note-Savard_2018-9">[6]</a><a href="#cite_note-Zehendner_2008-6">[4]</a><a href="#cite_note-NB_8-10">[nb 4]</a>), base four (quaternary floating point<a href="#cite_note-Parkinson_2000-11">[7]</a><a href="#cite_note-Beebe_2017-7">[5]</a><a href="#cite_note-NB_11-12">[nb 5]</a>), base three (<a href="Balanced_ternary_floating_point" class="mw-redirect" title="Balanced ternary floating point">balanced ternary floating point</a><a href="#cite_note-Muller_2010-1">[1]</a>) and even base 256<a href="#cite_note-Beebe_2017-7">[5]</a><a href="#cite_note-NB_12-13">[nb 6]</a> and base 65,536.<a href="#cite_note-Lazarus_1956-14">[8]</a><a href="#cite_note-NB_10-15">[nb 7]</a>

A floating-point number is a <a href="Rational_number" title="Rational number">rational number</a>, because it can be represented as one integer divided by another; for example 1.45×103 is (145/100)×1000 or 145,000/100. The base determines the fractions that can be represented; for instance, 1/5 cannot be represented exactly as a floating-point number using a binary base, but 1/5 can be represented exactly using a decimal base (0.2, or 2×10−1). However, 1/3 cannot be represented exactly by either binary (0.010101...) or decimal (0.333...), but in <a href="Ternary_numeral_system" title="Ternary numeral system">base 3</a>, it is trivial (0.1 or 1×3−1) . The occasions on which infinite expansions occur <a href="Positional_notation#Infinite_representations" title="Positional notation">depend on the base and its prime factors</a>.

The way in which the significand (including its sign) and exponent are stored in a computer is implementation-dependent. The common IEEE formats are described in detail later and elsewhere, but as an example, in the binary single-precision (32-bit) floating-point representation, <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle p=24}">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle p=24}</annotation>

</semantics>

</math><img src="./e7a1f78962b615188b35552e3a5c12c49edd7192.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; margin-left: -0.089ex; width:6.682ex; height:2.509ex;" alt="{\displaystyle p=24}" loading="lazy">, and so the significand is a string of 24 <a href="Bit" title="Bit">bits</a>. For instance, the number <a href="Pi" title="Pi">π</a>'s first 33 bits are:

</munder>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 11001001\ 00001111\ 1101101{\underline {0}}\ 10100010\ 0.}</annotation>

</semantics>

</math>

In this binary expansion, let us denote the positions from 0 (leftmost bit, or most significant bit) to 32 (rightmost bit). The 24-bit significand will stop at position 23, shown as the underlined bit 0 above. The next bit, at position 24, is called the round bit or rounding bit. It is used to round the 33-bit approximation to the nearest 24-bit number (there are <a href="Rounding#Tie-breaking" title="Rounding">specific rules for halfway values</a>, which is not the case here). This bit, which is 1 in this example, is added to the integer formed by the leftmost 24 bits, yielding:

</munder>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 11001001\ 00001111\ 1101101{\underline {1}}.}</annotation>

</semantics>

</math>

When this is stored in memory using the IEEE 754 encoding, this becomes the <a href="Significand" title="Significand">significand</a> s. The significand is assumed to have a binary point to the right of the leftmost bit. So, the binary representation of π is calculated from left-to-right as follows:

<mtr>

<mtd>

</mrow>

</mrow>

</mrow>

</munderover>

<msub>

</mrow>

</mrow>

</msub>

<msup>

</mrow>

</msup>

</mrow>

<msup>

</mrow>

</msup>

</mtd>

</mtr>

<mtr>

<mtd>

</mrow>

<mrow>

<mrow>

<msup>

</mrow>

</msup>

<msup>

</mrow>

</msup>

<msup>

</mrow>

</msup>

<msup>

</mrow>

</msup>

<msup>

</mrow>

</msup>

</mrow>

</mrow>

<msup>

</mrow>

</msup>

</mtd>

</mtr>

<mtr>

<mtd>

</mrow>

</mtd>

</mtr>

<mtr>

<mtd>

</mrow>

</mtd>

</mtr>

</mtable>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle {\begin{aligned}&{\biggl (}\sum _{n=0}^{p-1}{\text{bit}}_{n}\times 2^{-n}{\biggr )}\times 2^{e}\\&\qquad {}=\left(1\times 2^{-0}+1\times 2^{-1}+0\times 2^{-2}+0\times 2^{-3}+\cdots +1\times 2^{-23}\right)\times 2^{1}\\[2mu]&\qquad {}\approx 1.57079637\times 2\\[3mu]&\qquad {}\approx 3.1415927\end{aligned}}}</annotation>

</semantics>

</math>

where p is the precision (24 in this example), n is the position of the bit of the significand from the left (starting at 0 and finishing at 23 here) and e is the exponent (1 in this example).

It can be required that the most significant digit of the significand of a non-zero number be non-zero (except when the corresponding exponent would be smaller than the minimum one). This process is called normalization. For binary formats (which uses only the digits 0 and 1), this non-zero digit is necessarily 1. Therefore, it does not need to be represented in memory, allowing the format to have one more bit of precision. This rule is variously called the leading bit convention, the implicit bit convention, the hidden bit convention,<a href="#cite_note-Muller_2010-1">[1]</a> or the assumed bit convention.

<div class="mw-heading mw-heading3"><h3 id="Alternatives_to_floating-point_numbers">Alternatives to floating-point numbers</h3></div>

The floating-point representation is by far the most common way of representing in computers an approximation to real numbers. However, there are alternatives:

<ul><li><a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">Fixed-point</a> representation uses integer hardware operations controlled by a software implementation of a specific convention about the location of the binary or decimal point, for example, 6 bits or digits from the right. The hardware to manipulate these representations is less costly than floating point, and it can be used to perform normal integer operations, too. Binary fixed point is usually used in special-purpose applications on embedded processors that can only do integer arithmetic, but decimal fixed point is common in commercial applications.</li>

<li><a href="Logarithmic_number_system" title="Logarithmic number system">Logarithmic number systems</a> (LNSs) represent a real number by the logarithm of its absolute value and a sign bit. The value distribution is similar to floating point, but the value-to-representation curve (i.e., the graph of the logarithm function) is smooth (except at 0). Conversely to floating-point arithmetic, in a logarithmic number system multiplication, division and exponentiation are simple to implement, but addition and subtraction are complex. The (<a href="Symmetric_level-index_arithmetic" title="Symmetric level-index arithmetic">symmetric</a>) <a href="Level-index_arithmetic" class="mw-redirect" title="Level-index arithmetic">level-index arithmetic</a> (LI and SLI) of Charles Clenshaw, <a href="Frank_William_John_Olver" class="mw-redirect" title="Frank William John Olver">Frank Olver</a> and Peter Turner is a scheme based on a <a href="Generalized_logarithm" class="mw-redirect" title="Generalized logarithm">generalized logarithm</a> representation.</li>

<li><a href="Tapered_floating-point_representation" class="mw-redirect" title="Tapered floating-point representation">Tapered floating-point representation</a>, used in <a href="Unum_(number_format)" title="Unum (number format)">Unum</a>.</li>

<li>Some simple rational numbers (e.g., 1/3 and 1/10) cannot be represented exactly in binary floating point, no matter what the precision is. Using a different radix allows one to represent some of them (e.g., 1/10 in decimal floating point), but the possibilities remain limited. Software packages that perform <a href="Fraction" title="Fraction">rational arithmetic</a> represent numbers as fractions with integral numerator and denominator, and can therefore represent any rational number exactly. Such packages generally need to use "<a href="Bignum" class="mw-redirect" title="Bignum">bignum</a>" arithmetic for the individual integers.</li>

<li><a href="Interval_arithmetic" title="Interval arithmetic">Interval arithmetic</a> allows one to represent numbers as intervals and obtain guaranteed bounds on results. It is generally based on other arithmetics, in particular floating point.</li>

<li><a href="Computer_algebra_system" title="Computer algebra system">Computer algebra systems</a> such as <a href="Mathematica" class="mw-redirect" title="Mathematica">Mathematica</a>, <a href="Maxima_(software)" title="Maxima (software)">Maxima</a>, and <a href="Maple_(software)" title="Maple (software)">Maple</a> can often handle irrational numbers like <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \pi }">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \pi }</annotation>

</semantics>

</math><img src="./9be4ba0bb8df3af72e90a0535fabcc17431e540a.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:1.332ex; height:1.676ex;" alt="{\displaystyle \pi }" loading="lazy"> or <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle {\sqrt {3}}}">

<msqrt>

</msqrt>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle {\sqrt {3}}}</annotation>

</semantics>

</math><img src="./3b19c09494138b5082459afac7f9a8d99c546fcd.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:3.098ex; height:2.843ex;" alt="{\displaystyle {\sqrt {3}}}" loading="lazy"> in a completely "formal" way (<a href="Symbolic_computation" class="mw-redirect" title="Symbolic computation">symbolic computation</a>), without dealing with a specific encoding of the significand. Such a program can evaluate expressions like "<math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \sin(3\pi )}">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \sin(3\pi )}</annotation>

</semantics>

</math><img src="./7d239f0e06c5d96ed18c3add618e47f27b3534b3.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.838ex; width:7.159ex; height:2.843ex;" alt="{\displaystyle \sin(3\pi )}" loading="lazy">" exactly, because it is programmed to process the underlying mathematics directly, instead of using approximate values for each intermediate calculation.</li></ul>

<div class="mw-heading mw-heading2"><h2 id="History">History</h2></div>

<div role="note" class="hatnote navigation-not-searchable">See also: <a href="IEEE_754#History" title="IEEE 754">IEEE 754 § History</a></div>

In 1914, the Spanish engineer <a href="Leonardo_Torres_Quevedo" title="Leonardo Torres Quevedo">Leonardo Torres Quevedo</a> published Essays on Automatics,<a href="#cite_note-16">[9]</a> where he designed a special-purpose electromechanical calculator based on <a href="Charles_Babbage" title="Charles Babbage">Charles Babbage</a>'s <a href="Analytical_engine" title="Analytical engine">analytical engine</a> and described a way to store floating-point numbers in a consistent manner. He stated that numbers will be stored in exponential format as n × 10<math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle ^{m}}">

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle ^{m}}</annotation>

</semantics>

</math><img src="./a8c4d272ab903e44501a5ac0aff0c8f2b0fcf611.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.171ex; width:1.675ex; height:2.176ex;" alt="{\displaystyle ^{m}}" loading="lazy">, and offered three rules by which consistent manipulation of floating-point numbers by machines could be implemented. For Torres, "n will always be the same number of <a href="Numerical_digit" title="Numerical digit">digits</a> (e.g. six), the first digit of n will be of order of tenths, the second of hundredths, etc, and one will write each quantity in the form: n; m." The format he proposed shows the need for a fixed-sized significand as is presently used for floating-point data, fixing the location of the decimal point in the significand so that each representation was unique, and how to format such numbers by specifying a syntax to be used that could be entered through a <a href="Typewriter" title="Typewriter">typewriter</a>, as was the case of his <a href="Leonardo_Torres_y_Quevedo" class="mw-redirect" title="Leonardo Torres y Quevedo">Electromechanical Arithmometer</a> in 1920.<a href="#cite_note-17">[10]</a><a href="#cite_note-FOOTNOTERandell19826,_11–13-18">[11]</a><a href="#cite_note-19">[12]</a>

In 1938, <a href="Konrad_Zuse" title="Konrad Zuse">Konrad Zuse</a> of Berlin completed the <a href="Z1_(computer)" title="Z1 (computer)">Z1</a>, the first binary, programmable <a href="Mechanical_computer" title="Mechanical computer">mechanical computer</a>;<a href="#cite_note-Rojas_1997-20">[13]</a> it uses a 24-bit binary floating-point number representation with a 7-bit signed exponent, a 17-bit significand (including one implicit bit), and a sign bit.<a href="#cite_note-Rojas_2014-21">[14]</a> The more reliable <a href="Relay" title="Relay">relay</a>-based <a href="Z3_(computer)" title="Z3 (computer)">Z3</a>, completed in 1941, has representations for both positive and negative infinities; in particular, it implements defined operations with infinity, such as <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle ^{1}/_{\infty }=0}">

<msup>

</mrow>

</msup>

<msub>

</mrow>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle ^{1}/_{\infty }=0}</annotation>

</semantics>

</math><img src="./23ef4d558c2fedd08aaaa744e83c7d36dd85cce6.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -1.005ex; width:8.353ex; height:3.343ex;" alt="{\displaystyle ^{1}/_{\infty }=0}" loading="lazy">, and it stops on undefined operations, such as <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle 0\times \infty }">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 0\times \infty }</annotation>

</semantics>

</math><img src="./6c2c67d872e7859a5b51d652639651d1e1384df0.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:6.327ex; height:2.176ex;" alt="{\displaystyle 0\times \infty }" loading="lazy">.

Zuse also proposed, but did not complete, carefully rounded floating-point arithmetic that includes <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \pm \infty }">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \pm \infty }</annotation>

</semantics>

</math><img src="./c586ae37f8efec026b8a4ea3f6a5253576c2c4e6.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:4.132ex; height:2.176ex;" alt="{\displaystyle \pm \infty }" loading="lazy"> and NaN representations, anticipating features of the IEEE Standard by four decades.<a href="#cite_note-Kahan_1997_JVNL-22">[15]</a> In contrast, <a href="John_von_Neumann" title="John von Neumann">von Neumann</a> recommended against floating-point numbers for the 1951 <a href="IAS_machine" title="IAS machine">IAS machine</a>, arguing that fixed-point arithmetic is preferable.<a href="#cite_note-Kahan_1997_JVNL-22">[15]</a>

The first commercial computer with floating-point hardware was Zuse's <a href="Z4_(computer)" title="Z4 (computer)">Z4</a> computer, designed in 1942–1945. In 1946, Bell Laboratories introduced the <a href="Model_V" title="Model V">Model V</a>, which implemented <a href="Decimal_floating_point" title="Decimal floating point">decimal floating-point numbers</a>.<a href="#cite_note-Randell_1982_2-23">[16]</a>

The <a href="Pilot_ACE" title="Pilot ACE">Pilot ACE</a> has binary floating-point arithmetic, and it became operational in 1950 at <a href="National_Physical_Laboratory%2C_UK" class="mw-redirect" title="National Physical Laboratory, UK">National Physical Laboratory, UK</a>. Thirty-three were later sold commercially as the <a href="English_Electric_DEUCE" title="English Electric DEUCE">English Electric DEUCE</a>. The arithmetic is actually implemented in software, but with a one megahertz clock rate, the speed of floating-point and fixed-point operations in this machine were initially faster than those of many competing computers.

The mass-produced <a href="IBM_704" title="IBM 704">IBM 704</a> followed in 1954; it introduced the use of a <a href="Exponent_bias" title="Exponent bias">biased exponent</a>. For many decades after that, floating-point hardware was typically an optional feature, and computers that had it were said to be "scientific computers", or to have "<a href="Scientific_computation" class="mw-redirect" title="Scientific computation">scientific computation</a>" (SC) capability (see also <a href="Extensions_for_Scientific_Computation" class="mw-redirect" title="Extensions for Scientific Computation">Extensions for Scientific Computation</a> (XSC)). It was not until the launch of the Intel i486 in 1989 that general-purpose personal computers had floating-point capability in hardware as a standard feature.

The <a href="UNIVAC_1100/2200_series" title="UNIVAC 1100/2200 series">UNIVAC 1100/2200 series</a>, introduced in 1962, supported two floating-point representations:

<ul><li>Single precision: 36 bits, organized as a 1-bit sign, an 8-bit exponent, and a 27-bit significand.</li>

<li>Double precision: 72 bits, organized as a 1-bit sign, an 11-bit exponent, and a 60-bit significand.</li></ul>

The <a href="IBM_7094" class="mw-redirect" title="IBM 7094">IBM 7094</a>, also introduced in 1962, supported single-precision and double-precision representations, but with no relation to the UNIVAC's representations. Indeed, in 1964, IBM introduced <a href="IBM_hexadecimal_floating-point" title="IBM hexadecimal floating-point">hexadecimal floating-point representations</a> in its <a href="System/360" class="mw-redirect" title="System/360">System/360</a> mainframes; these same representations are still available for use in modern <a href="Z/Architecture" title="Z/Architecture">z/Architecture</a> systems. In 1998, IBM implemented IEEE-compatible binary floating-point arithmetic in its mainframes; in 2005, IBM also added IEEE-compatible decimal floating-point arithmetic.

Initially, computers used many different representations for floating-point numbers. The lack of standardization at the mainframe level was an ongoing problem by the early 1970s for those writing and maintaining higher-level source code; these manufacturer floating-point standards differed in the word sizes, the representations, and the rounding behavior and general accuracy of operations. Floating-point compatibility across multiple computing systems was in desperate need of standardization by the early 1980s, leading to the creation of the <a href="IEEE_754" title="IEEE 754">IEEE 754</a> standard once the 32-bit (or 64-bit) <a href="Word_(computer_architecture)" title="Word (computer architecture)">word</a> had become commonplace. This standard was significantly based on a proposal from Intel, which was designing the <a href="Intel_8087" title="Intel 8087">i8087</a> numerical coprocessor; Motorola, which was designing the <a href="68000" class="mw-redirect" title="68000">68000</a> around the same time, gave significant input as well.

In 1989, mathematician and computer scientist <a href="William_Kahan" title="William Kahan">William Kahan</a> was honored with the <a href="Turing_Award" title="Turing Award">Turing Award</a> for being the primary architect behind this proposal; he was aided by his student Jerome Coonen and a visiting professor, <a href="Harold_S._Stone" title="Harold S. Stone">Harold Stone</a>.<a href="#cite_note-Severance_1998-24">[17]</a>

Among the x86 (more specifically i8087) innovations are these:

<ul><li>A precisely specified floating-point representation at the bit-string level, so that all compliant computers interpret bit patterns the same way. This makes it possible to accurately and efficiently transfer floating-point numbers from one computer to another (after accounting for <a href="Endianness" title="Endianness">endianness</a>).</li>

<li>A precisely specified behavior for the arithmetic operations: A result is required to be produced as if infinitely precise arithmetic were used to yield a value that is then rounded according to specific rules. This means that a compliant computer program would always produce the same result when given a particular input, thus mitigating the almost mystical reputation that floating-point computation had developed for its hitherto seemingly non-deterministic behavior.</li>

<li>The ability of <a href="IEEE_754#Exception_handling" title="IEEE 754">exceptional conditions</a> (overflow, <a href="Division_by_zero" title="Division by zero">divide by zero</a>, etc.) to propagate through a computation in a benign manner and then be handled by the software in a controlled fashion.</li></ul>

These features would be inherited into IEEE 754-1985 (with the exception of the encoding of special values and exceptions), though the extended internal precision of x87 means it requires explicit rounding of exact results directly to the destination precision in order to match standard IEEE 754 results.<a href="#cite_note-Goldberg_1991-25">[18]</a> However, the behavior may not be the same as a rounding to the destination format due to a possible wider exponent range of the extended format.

<div class="mw-heading mw-heading2"><h2 id="Range_of_floating-point_numbers">Range of floating-point numbers</h2></div>

A floating-point number consists of two <a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">fixed-point</a> components, whose range depends exclusively on the number of bits or digits in their representation. Whereas components linearly depend on their range, the floating-point range linearly depends on the significand range and exponentially on the range of exponent component, which attaches outstandingly wider range to the number.

On a typical computer system, a <a href="Double-precision_floating-point_format" title="Double-precision floating-point format">double-precision</a> (64-bit) binary floating-point number has a coefficient of 53 bits (including 1 implied bit), an exponent of 11 bits, and 1 sign bit. Since 210 = 1024, the complete range of the positive normal floating-point numbers in this format is from 2−1022 ≈ 2 × 10−308 to approximately 21024 ≈ 2 × 10308.

The number of normal floating-point numbers in a system (B, P, L, U) where

<ul><li>B is the base of the system,</li>

<li>P is the precision of the significand (in base B),</li>

<li>L is the smallest exponent of the system,</li>

<li>U is the largest exponent of the system,</li></ul>

is <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle 2\left(B-1\right)\left(B^{P-1}\right)\left(U-L+1\right)}">

<mrow>

<mrow>

</mrow>

</mrow>

<mrow>

<msup>

</mrow>

</msup>

</mrow>

<mrow>

<mrow>

</mrow>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 2\left(B-1\right)\left(B^{P-1}\right)\left(U-L+1\right)}</annotation>

</semantics>

</math><img src="./50a5e9a91a7338cf13ed00dfe9042b7aa0d8a79b.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -1.005ex; width:29.379ex; height:3.343ex;" alt="{\displaystyle 2\left(B-1\right)\left(B^{P-1}\right)\left(U-L+1\right)}" loading="lazy">.

There is a smallest positive normal floating-point number,

<dl><dd>Underflow level = UFL = <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle B^{L}}">

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle B^{L}}</annotation>

</semantics>

</math><img src="./03c80b97f6b23ac47189791d0c01da71ab982531.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:3.116ex; height:2.676ex;" alt="{\displaystyle B^{L}}" loading="lazy">,</dd></dl>

which has a 1 as the leading digit and 0 for the remaining digits of the significand, and the smallest possible value for the exponent.

There is a largest floating-point number,

<dl><dd>Overflow level = OFL = <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \left(1-B^{-P}\right)\left(B^{U+1}\right)}">

<mrow>

<mrow>

<msup>

</mrow>

</msup>

</mrow>

</mrow>

<mrow>

<msup>

</mrow>

</msup>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \left(1-B^{-P}\right)\left(B^{U+1}\right)}</annotation>

</semantics>

</math><img src="./89ef88d856c29d99a33aa71976c5827a32abf281.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -1.005ex; width:18.516ex; height:3.343ex;" alt="{\displaystyle \left(1-B^{-P}\right)\left(B^{U+1}\right)}" loading="lazy">,</dd></dl>

which has B − 1 as the value for each digit of the significand and the largest possible value for the exponent.

In addition, there are representable values strictly between −UFL and UFL. Namely, <a href="Signed_zero" title="Signed zero">positive and negative zeros</a>, as well as <a href="Subnormal_number" title="Subnormal number">subnormal numbers</a>.

<div class="mw-heading mw-heading2"><h2 id="IEEE_754:_floating_point_in_modern_computers">IEEE 754: floating point in modern computers </h2></div>

<div role="note" class="hatnote navigation-not-searchable">Main article: <a href="IEEE_754" title="IEEE 754">IEEE 754</a></div>

<table class="sidebar nomobile nowraplinks plainlist"><tbody><tr><th class="sidebar-title"> <a href="Computer_number_format" title="Computer number format">formats</a></th></tr><tr><th class="sidebar-heading">

<ul><li>16-bit: <a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half</a> (binary16)</li>

<li>256-bit: <a href="Octuple-precision_floating-point_format" title="Octuple-precision floating-point format">Octuple</a> (binary256)</li>

<li><a href="Extended_precision" title="Extended precision">Extended precision</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

Other</th></tr><tr><td class="sidebar-content">

<ul><li><a href="Minifloat" title="Minifloat">Minifloat</a></li>

<li><a href="Bfloat16_floating-point_format" title="Bfloat16 floating-point format">bfloat16</a></li>

<li><a href="TensorFloat-32" title="TensorFloat-32">TensorFloat-32</a></li>

<li><a href="Microsoft_Binary_Format" title="Microsoft Binary Format">Microsoft Binary Format</a></li>

<li><a href="IBM_hexadecimal_floating-point" title="IBM hexadecimal floating-point">IBM hexadecimal floating-point</a></li>

<li><a href="Power_Management_Bus#Linear11_Floating-Point_Format" title="Power Management Bus">PMBus Linear-11</a></li>

<li><a href="G.711" title="G.711">G.711 8-bit floats</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

Alternatives</th></tr><tr><td class="sidebar-content">

<ul><li><a href="Arbitrary-precision_arithmetic" title="Arbitrary-precision arithmetic">Arbitrary precision</a></li>

<li><a href="Block_floating_point" title="Block floating point">Block floating point</a></li></ul></td>

</tr><tr><th class="sidebar-heading">

<a href="Tapered_floating_point" title="Tapered floating point">Tapered floating point</a></th></tr><tr><td class="sidebar-content">

<ul><li><a href="Unum_(number_format)" title="Unum (number format)">Posit</a></li></ul></td>

</tr><tr><td class="sidebar-navbar"></td></tr></tbody></table>

The <a href="Institute_of_Electrical_and_Electronics_Engineers" title="Institute of Electrical and Electronics Engineers">IEEE</a> standardized the computer representation for binary floating-point numbers in <a href="IEEE_754" title="IEEE 754">IEEE 754</a> (a.k.a. IEC 60559) in 1985. This first standard is followed by almost all modern machines. It was <a href="IEEE_754-2008_revision" title="IEEE 754-2008 revision">revised in 2008</a>. IBM mainframes support <a href="IBM_hexadecimal_floating_point" class="mw-redirect" title="IBM hexadecimal floating point">IBM's own hexadecimal floating point format</a> and IEEE 754-2008 <a href="Decimal_floating_point" title="Decimal floating point">decimal floating point</a> in addition to the IEEE 754 binary format. The <a href="Cray_T90" title="Cray T90">Cray T90</a> series had an IEEE version, but the <a href="Cray_SV1" title="Cray SV1">SV1</a> still uses Cray floating-point format.

The standard provides for many closely related formats, differing in only a few details. Five of these formats are called basic formats, and others are termed extended precision formats and extendable precision format. Three formats are especially widely used in computer hardware and languages:

<ul><li><a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single precision</a> (binary32), usually used to represent the "float" <a href="C_data_types#Basic_types" title="C data types">type in the C language</a> family. This is a binary format that occupies 32 bits (4 bytes) and its significand has a precision of 24 bits (about 7 decimal digits).</li>

<li><a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double precision</a> (binary64), usually used to represent the "double" <a href="C_data_types#Basic_types" title="C data types">type in the C language</a> family. This is a binary format that occupies 64 bits (8 bytes) and its significand has a precision of 53 bits (about 16 decimal digits).</li>

<li><a href="Extended_precision" title="Extended precision">Double extended</a>, also ambiguously called "extended precision" format. This is a binary format that occupies at least 79 bits (80 if the hidden/implicit bit rule is not used) and its significand has a precision of at least 64 bits (about 19 decimal digits). The <a href="C99" title="C99">C99</a> and <a href="C11_(C_standard_revision)" title="C11 (C standard revision)">C11</a> standards of the C language family, in their annex F ("IEC 60559 floating-point arithmetic"), recommend such an extended format to be provided as "<a href="Long_double" title="Long double">long double</a>".<a href="#cite_note-C99-26">[19]</a> A format satisfying the minimal requirements (64-bit significand precision, 15-bit exponent, thus fitting on 80 bits) is provided by the <a href="X86" title="X86">x86</a> architecture. Often on such processors, this format can be used with "long double", though extended precision is not available with MSVC.<a href="#cite_note-MSVC-27">[20]</a> For <a href="Data_structure_alignment" title="Data structure alignment">alignment</a> purposes, many tools store this 80-bit value in a 96-bit or 128-bit space.<a href="#cite_note-GCC-28">[21]</a><a href="#cite_note-float_128-29">[22]</a> On other processors, "long double" may stand for a larger format, such as quadruple precision,<a href="#cite_note-ARM_2013_AArch64-30">[23]</a> or just double precision, if any form of extended precision is not available.<a href="#cite_note-ARM_2013_Compiler-31">[24]</a></li></ul>

Increasing the precision of the floating-point representation generally reduces the amount of accumulated <a href="Round-off_error" title="Round-off error">round-off error</a> caused by intermediate calculations.<a href="#cite_note-Kahan_2004-32">[25]</a>

Other IEEE formats include:

<ul><li><a href="Decimal64_floating-point_format" title="Decimal64 floating-point format">Decimal64</a> and <a href="Decimal128_floating-point_format" title="Decimal128 floating-point format">decimal128</a> floating-point formats. These formats (especially decimal128) are pervasive in financial transactions because, along with the <a href="Decimal32_floating-point_format" title="Decimal32 floating-point format">decimal32</a> format, they allow correct decimal rounding.</li>

<li><a href="Quadruple-precision_floating-point_format#IEEE_754_quadruple-precision_binary_floating-point_format:_binary128" title="Quadruple-precision floating-point format">Quadruple precision</a> (binary128). This is a binary format that occupies 128 bits (16 bytes) and its significand has a precision of 113 bits (about 34 decimal digits).</li>

<li><a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half precision</a>, also called binary16, a 16-bit floating-point value. It is being used in the NVIDIA <a href="Cg_(programming_language)" title="Cg (programming language)">Cg</a> graphics language, and in the openEXR standard (where it actually predates the introduction in the IEEE 754 standard).<a href="#cite_note-OpenEXR-33">[26]</a><a href="#cite_note-OpenEXR-half-34">[27]</a></li></ul>

Any integer with absolute value less than 224 can be exactly represented in the single-precision format, and any integer with absolute value less than 253 can be exactly represented in the double-precision format. Furthermore, a wide range of powers of 2 times such a number can be represented. These properties are sometimes used for purely integer data, to get 53-bit integers on platforms that have double-precision floats but only 32-bit integers.

The standard specifies some special values, and their representation: positive <a href="Infinity" title="Infinity">infinity</a> (+∞), negative infinity (−∞), a <a href="Negative_zero" class="mw-redirect" title="Negative zero">negative zero</a> (−0) distinct from ordinary ("positive") zero, and "not a number" values (<a href="NaN" title="NaN">NaNs</a>).

Comparison of floating-point numbers, as defined by the IEEE standard, is a bit different from usual integer comparison. Negative and positive zero compare equal, and every NaN compares unequal to every value, including itself. All finite floating-point numbers are strictly smaller than +∞ and strictly greater than −∞, and they are ordered in the same way as their values (in the set of real numbers).

<div class="mw-heading mw-heading3"><h3 id="Internal_representation">Internal representation</h3></div>

Floating-point numbers are typically packed into a computer datum as the sign bit, the exponent field, and a field for the significand, from left to right. For the <a href="IEEE_754" title="IEEE 754">IEEE 754</a> binary formats (basic and extended) that have extant hardware implementations, they are apportioned as follows:

<th rowspan="2">Format

</th>

<th colspan="4">Bits for the encoding

</th>

</td>

<th rowspan="2">Exponent bias

</th>

<th rowspan="2">Bits precision

</th>

<th rowspan="2">Number of decimal digits

</th></tr>

<tr>

<th>Sign

</th>

<th>Exponent

</th>

<th>Significand

</th>

<th>Total

</th></tr>

<tr>

<td><a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half</a> (binary16)

</td>

<td>1

</td>

<td>5

</td>

<td>10

</td>

<td>16

</td>

<td>15

</td>

<td>11

</td>

<td>~3.3

</td></tr>

<tr>

<td><a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single</a> (binary32)

</td>

<td>1

</td>

<td>8

</td>

<td>23

</td>

<td>32

</td>

<td>127

</td>

<td>24

</td>

<td>~7.2

</td></tr>

<tr>

<td><a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double</a> (binary64)

</td>

<td>1

</td>

<td>11

</td>

<td>52

</td>

<td>64

</td>

<td>1023

</td>

<td>53

</td>

<td>~15.9

</td></tr>

<tr>

<td><a href="Extended_precision#x86_extended-precision_format" title="Extended precision">x86 extended</a>

</td>

<td>1

</td>

<td>15

</td>

<td>64

</td>

<td>80

</td>

<td>16383

</td>

<td>64

</td>

<td>~19.2

</td></tr>

<tr>

<td><a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">Quadruple</a> (binary128)

</td>

<td>1

</td>

<td>15

</td>

<td>112

</td>

<td>128

</td>

<td>16383

</td>

<td>113

</td>

<td>~34.0

</td></tr>

<tr>

<td><a href="Octuple-precision_floating-point_format" title="Octuple-precision floating-point format">Octuple</a> (binary256)

</td>

<td>1

</td>

<td>19

</td>

<td>236

</td>

<td>256

</td>

<td>262143

</td>

<td>237

</td>

<td>~71.3

</td></tr></tbody></table>

While the exponent can be positive or negative, in binary formats it is stored as an unsigned number that has a fixed "bias" added to it. Values of all 0s in this field are reserved for the zeros and <a href="Subnormal_number" title="Subnormal number">subnormal numbers</a>; values of all 1s are reserved for the infinities and NaNs. The exponent range for normal numbers is [−126, 127] for single precision, [−1022, 1023] for double, or [−16382, 16383] for quad. Normal numbers exclude subnormal values, zeros, infinities, and NaNs.

In the IEEE binary interchange formats the leading bit of a normalized significand is not actually stored in the computer datum, since it is always 1. It is called the "hidden" or "implicit" bit. Because of this, the single-precision format actually has a significand with 24 bits of precision, the double-precision format has 53, quad has 113, and octuple has 237.

For example, it was shown above that π, rounded to 24 bits of precision, has:

<ul><li>sign = 0 ; e = 1 ; s = 110010010000111111011011 (including the hidden bit)</li></ul>

The sum of the exponent bias (127) and the exponent (1) is 128, so this is represented in the single-precision format as

<ul><li>0 10000000 10010010000111111011011 (excluding the hidden bit) = 40490FDB<a href="#cite_note-IEEE-754_Analysis-35">[28]</a> as a <a href="Hexadecimal" title="Hexadecimal">hexadecimal</a> number.</li></ul>

An example of a layout for <a href="Single-precision_floating-point_format" title="Single-precision floating-point format">32-bit floating point</a> is

and the <a href="Double-precision_floating-point_format" title="Double-precision floating-point format">64-bit ("double")</a> layout is similar.

<div class="mw-heading mw-heading2"><h2 id="Other_notable_floating-point_formats">Other notable floating-point formats</h2></div>

In addition to the widely used <a href="IEEE_754" title="IEEE 754">IEEE 754</a> standard formats, other floating-point formats are used, or have been used, in certain domain-specific areas.

<ul><li>The <a href="Microsoft_Binary_Format" title="Microsoft Binary Format">Microsoft Binary Format (MBF)</a> was developed for the Microsoft BASIC language products, including Microsoft's first ever product the <a href="Altair_BASIC" title="Altair BASIC">Altair BASIC</a> (1975), <a href="TRS-80" title="TRS-80">TRS-80 LEVEL II</a>, <a href="CP/M" title="CP/M">CP/M</a>'s <a href="MBASIC" title="MBASIC">MBASIC</a>, <a href="IBM_PC_5150" class="mw-redirect" title="IBM PC 5150">IBM PC 5150</a>'s <a href="BASICA" class="mw-redirect" title="BASICA">BASICA</a>, <a href="MS-DOS" title="MS-DOS">MS-DOS</a>'s <a href="GW-BASIC" title="GW-BASIC">GW-BASIC</a> and <a href="QuickBASIC" title="QuickBASIC">QuickBASIC</a> prior to version 4.00. QuickBASIC version 4.00 and 4.50 switched to the IEEE 754-1985 format but can revert to the MBF format using the /MBF command option. MBF was designed and developed on a simulated <a href="Intel_8080" title="Intel 8080">Intel 8080</a> by <a href="Monte_Davidoff" title="Monte Davidoff">Monte Davidoff</a>, a dormmate of <a href="Bill_Gates" title="Bill Gates">Bill Gates</a>, during spring of 1975 for the <a href="MITS_Altair_8800" class="mw-redirect" title="MITS Altair 8800">MITS Altair 8800</a>. The initial release of July 1975 supported a single-precision (32 bits) format due to cost of the <a href="MITS_Altair_8800" class="mw-redirect" title="MITS Altair 8800">MITS Altair 8800</a> 4-kilobytes memory. In December 1975, the 8-kilobytes version added a double-precision (64 bits) format. A single-precision (40 bits) variant format was adopted for other CPU's, notably the <a href="MOS_6502" class="mw-redirect" title="MOS 6502">MOS 6502</a> (<a href="Apple_II" title="Apple II">Apple II</a>, <a href="Commodore_PET" title="Commodore PET">Commodore PET</a>, <a href="Atari" title="Atari">Atari</a>), <a href="Motorola_6800" title="Motorola 6800">Motorola 6800</a> (MITS Altair 680) and <a href="Motorola_6809" title="Motorola 6809">Motorola 6809</a> (<a href="TRS-80_Color_Computer" title="TRS-80 Color Computer">TRS-80 Color Computer</a>). All Microsoft language products from 1975 through 1987 used the <a href="Microsoft_Binary_Format" title="Microsoft Binary Format">Microsoft Binary Format</a> until Microsoft adopted the IEEE 754 standard format in all its products starting in 1988 to their current releases. MBF consists of the MBF single-precision format (32 bits, "6-digit BASIC"),<a href="#cite_note-Borland_1994_MBF-36">[29]</a><a href="#cite_note-Steil_2008_6502-37">[30]</a> the MBF extended-precision format (40 bits, "9-digit BASIC"),<a href="#cite_note-Steil_2008_6502-37">[30]</a> and the MBF double-precision format (64 bits);<a href="#cite_note-Borland_1994_MBF-36">[29]</a><a href="#cite_note-Microsoft_2006_KB35826-38">[31]</a> each of them is represented with an 8-bit exponent, followed by a sign bit, followed by a significand of respectively 23, 31, and 55 bits.</li>

<li>The <a href="Bfloat16_floating-point_format" title="Bfloat16 floating-point format">bfloat16 format</a> requires the same amount of memory (16 bits) as the <a href="Half-precision_floating-point_format" title="Half-precision floating-point format">IEEE 754 half-precision format</a>, but allocates 8 bits to the exponent instead of 5, thus providing the same range as a <a href="Single-precision_floating-point_format" title="Single-precision floating-point format">IEEE 754 single-precision</a> number. The tradeoff is a reduced precision, as the trailing significand field is reduced from 10 to 7 bits. This format is mainly used in the training of <a href="Machine_learning" title="Machine learning">machine learning</a> models, where range is more valuable than precision. Many machine learning accelerators provide hardware support for this format.</li>

<li>The TensorFloat-32<a href="#cite_note-Kharya_2020-39">[32]</a> format combines the 8 bits of exponent of the bfloat16 with the 10 bits of trailing significand field of half-precision formats, resulting in a size of 19 bits. This format was introduced by <a href="Nvidia" title="Nvidia">Nvidia</a>, which provides hardware support for it in the Tensor Cores of its <a href="Graphics_processing_unit" title="Graphics processing unit">GPUs</a> based on the Nvidia Ampere architecture. The drawback of this format is its size, which is not a power of 2. However, according to Nvidia, this format should only be used internally by hardware to speed up computations, while inputs and outputs should be stored in the 32-bit single-precision IEEE 754 format.<a href="#cite_note-Kharya_2020-39">[32]</a></li>

<li>The <a href="Hopper_(microarchitecture)" title="Hopper (microarchitecture)">Hopper</a> architecture GPUs provide two FP8 formats: one with the same numerical range as half-precision (E5M2) and one with higher precision, but less range (E4M3).<a href="#cite_note-NVIDIA_Hopper-40">[33]</a><a href="#cite_note-Micikevicius_2022-41">[34]</a></li>

<li>The <a href="Blackwell_(microarchitecture)" title="Blackwell (microarchitecture)">Blackwell</a> GPU architecture includes support for FP6 (E3M2 and E2M3) and FP4 (E2M1) formats. FP4 is the smallest floating-point format which allows for all IEEE 754 principles (see <a href="Minifloat" title="Minifloat">minifloat</a>).</li></ul>

<caption>Comparison of common floating-point formats

</caption>

<th>Type

</th>

<th>Sign

</th>

<th>Exponent

</th>

<th>Significand

</th>

<th>Total bits

</th></tr>

<tr>

<td>FP4

</td>

<td>1

</td>

<td>2

</td>

<td>1

</td>

<td>4

</td></tr>

<tr>

<td>FP6 (E2M3)

</td>

<td>1

</td>

<td>2

</td>

<td>3

</td>

<td>6

</td></tr>

<tr>

<td>FP6 (E3M2)

</td>

<td>1

</td>

<td>3

</td>

<td>2

</td>

<td>6

</td></tr>

<tr>

<td>FP8 (E4M3)

</td>

<td>1

</td>

<td>4

</td>

<td>3

</td>

<td>8

</td></tr>

<tr>

<td>FP8 (E5M2)

</td>

<td>1

</td>

<td>5

</td>

<td>2

</td>

<td>8

</td></tr>

<tr>

<td><a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half-precision</a>

</td>

<td>1

</td>

<td>5

</td>

<td>10

</td>

<td>16

</td></tr>

<tr>

<td><a href="Bfloat16_floating-point_format" title="Bfloat16 floating-point format">bfloat16</a>

</td>

<td>1

</td>

<td>8

</td>

<td>7

</td>

<td>16

</td></tr>

<tr>

<td><a href="TensorFloat-32" title="TensorFloat-32">TensorFloat-32</a>

</td>

<td>1

</td>

<td>8

</td>

<td>10

</td>

<td>19

</td></tr>

<tr>

<td><a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single-precision</a>

</td>

<td>1

</td>

<td>8

</td>

<td>23

</td>

<td>32

</td></tr>

<tr>

<td><a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double-precision</a>

</td>

<td>1

</td>

<td>11

</td>

<td>52

</td>

<td>64

</td></tr>

<tr>

<td><a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">Quadruple-precision</a>

</td>

<td>1

</td>

<td>15

</td>

<td>112

</td>

<td>128

</td></tr>

<tr>

<td><a href="Octuple-precision_floating-point_format" title="Octuple-precision floating-point format">Octuple-precision</a>

</td>

<td>1

</td>

<td>19

</td>

<td>236

</td>

<td>256

</td></tr></tbody></table>

<div class="mw-heading mw-heading2"><h2 id="Representable_numbers,_conversion_and_rounding">Representable numbers, conversion and rounding </h2></div>

By their nature, all numbers expressed in floating-point format are <a href="Rational_number" title="Rational number">rational numbers</a> with a terminating expansion in the relevant base (for example, a terminating decimal expansion in base-10, or a terminating binary expansion in base-2). Irrational numbers, such as <a href="Pi" title="Pi">π</a> or <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\textstyle {\sqrt {2}}}">

<msqrt>

</msqrt>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\textstyle {\sqrt {2}}}</annotation>

</semantics>

</math><img src="./5094a5b1e2f42490aa4de2c7a4b7235a27f1b73f.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:3.098ex; height:3.009ex;" alt="{\textstyle {\sqrt {2}}}" loading="lazy">, or non-terminating rational numbers, must be approximated. The number of digits (or bits) of precision also limits the set of rational numbers that can be represented exactly. For example, the decimal number 123456789 cannot be exactly represented if only eight decimal digits of precision are available (it would be rounded to one of the two straddling representable values, 12345678 × 101 or 12345679 × 101), the same applies to <a href="Repeating_decimal" title="Repeating decimal">non-terminating digits</a> (.5 to be rounded to either .55555555 or .55555556).

When a number is represented in some format (such as a character string) which is not a native floating-point representation supported in a computer implementation, then it will require a conversion before it can be used in that implementation. If the number can be represented exactly in the floating-point format then the conversion is exact. If there is not an exact representation then the conversion requires a choice of which floating-point number to use to represent the original value. The representation chosen will have a different value from the original, and the value thus adjusted is called the rounded value.

Whether or not a rational number has a terminating expansion depends on the base. For example, in base-10 the number 1/2 has a terminating expansion (0.5) while the number 1/3 does not (0.333...). In base-2 only rationals with denominators that are powers of 2 (such as 1/2 or 3/16) are terminating. Any rational with a denominator that has a prime factor other than 2 will have an infinite binary expansion. This means that numbers that appear to be short and exact when written in decimal format may need to be approximated when converted to binary floating-point. For example, the decimal number 0.1 is not representable in binary floating-point of any finite precision; the exact binary representation would have a "1100" sequence continuing endlessly:

where, as previously, s is the significand and e is the exponent.

When rounded to 24 bits this becomes

which is actually 0.100000001490116119384765625 in decimal.

As a further example, the real number <a href="Pi" title="Pi">π</a>, represented in binary as an infinite sequence of bits is

but is

when approximated by <a href="Rounding" title="Rounding">rounding</a> to a precision of 24 bits.

In binary single-precision floating-point, this is represented as s = 1.10010010000111111011011 with e = 1.

This has a decimal value of

whereas a more accurate approximation of the true value of π is

The result of rounding differs from the true value by about 0.03 parts per million, and matches the decimal representation of π in the first 7 digits. The difference is the <a href="Discretization_error" title="Discretization error">discretization error</a> and is limited by the <a href="Machine_epsilon" title="Machine epsilon">machine epsilon</a>.

The arithmetical difference between two consecutive representable floating-point numbers which have the same exponent is called a <a href="Unit_in_the_last_place" title="Unit in the last place">unit in the last place</a> (ULP). For example, if there is no representable number lying between the representable numbers 1.45A70C2216 and 1.45A70C2416, the ULP is 2×16−8, or 2−31. For numbers with a base-2 exponent part of 0, i.e. numbers with an absolute value higher than or equal to 1 but lower than 2, an ULP is exactly 2−23 or about 10−7 in single precision, and exactly 2−53 or about 10−16 in double precision. The mandated behavior of IEEE-compliant hardware is that the result be within one-half of a ULP.

<div class="mw-heading mw-heading3"><h3 id="Rounding_modes">Rounding modes</h3></div>

Rounding is used when the exact result of a floating-point operation (or a conversion to floating-point format) would need more digits than there are digits in the significand. IEEE 754 requires correct rounding: that is, the rounded result is as if infinitely precise arithmetic was used to compute the value and then rounded (although in implementation only three extra bits are needed to ensure this). There are several different <a href="Rounding" title="Rounding">rounding</a> schemes (or rounding modes). Historically, <a href="Truncation" title="Truncation">truncation</a> was the typical approach. Since the introduction of IEEE 754, the default method (<a href="Rounding" title="Rounding">round to nearest, ties to even</a>, sometimes called Banker's Rounding) is more commonly used. This method rounds the ideal (infinitely precise) result of an arithmetic operation to the nearest representable value, and gives that representation as the result.<a href="#cite_note-NB_1-42">[nb 8]</a> In the case of a tie, the value that would make the significand end in an even digit is chosen. The IEEE 754 standard requires the same rounding to be applied to all fundamental algebraic operations, including square root and conversions, when there is a numeric (non-NaN) result. It means that the results of IEEE 754 operations are completely determined in all bits of the result, except for the representation of NaNs. ("Library" functions such as cosine and log are not mandated.)

Alternative rounding options are also available. IEEE 754 specifies the following rounding modes:

<ul><li>round to nearest, where ties round to the nearest even digit in the required position (the default and by far the most common mode)</li>

<li>round to nearest, where ties round away from zero (optional for binary floating-point and commonly used in decimal)</li>

<li>round up (toward +∞; negative results thus round toward zero)</li>

<li>round down (toward −∞; negative results thus round away from zero)</li>

<li>round toward zero (truncation; it is similar to the common behavior of float-to-integer conversions, which convert −3.9 to −3 and 3.9 to 3)</li></ul>

Alternative modes are useful when the amount of error being introduced must be bounded. Applications that require a bounded error are multi-precision floating-point, and <a href="Interval_arithmetic" title="Interval arithmetic">interval arithmetic</a>.

The alternative rounding modes are also useful in diagnosing numerical instability: if the results of a subroutine vary substantially between rounding to + and − infinity then it is likely numerically unstable and affected by round-off error.<a href="#cite_note-Kahan_2006_Mindless-43">[35]</a>

<div class="mw-heading mw-heading3"><h3 id="Binary-to-decimal_conversion_with_minimal_number_of_digits">Binary-to-decimal conversion with minimal number of digits</h3></div>

Converting a double-precision binary floating-point number to a decimal string is a common operation, but an algorithm producing results that are both accurate and minimal did not appear in print until 1990, with Steele and White's Dragon4. Some of the improvements since then include:

<ul><li>David M. Gay's dtoa.c, a practical open-source implementation of many ideas in Dragon4.<a href="#cite_note-Gay_1990-44">[36]</a></li>

<li>Grisu3, with a 4× speedup as it removes the use of <a href="Bignum" class="mw-redirect" title="Bignum">bignums</a>. Must be used with a fallback, as it fails for ~0.5% of cases.<a href="#cite_note-Loitsch_2010-45">[37]</a></li>

<li>Errol3, an always-succeeding algorithm similar to, but slower than, Grisu3. Apparently not as good as an early-terminating Grisu with fallback.<a href="#cite_note-mazong-46">[38]</a></li>

<li>Ryū, an always-succeeding algorithm that is faster and simpler than Grisu3.<a href="#cite_note-Adams_2018-47">[39]</a></li>

<li>Schubfach, an always-succeeding algorithm that is based on a similar idea to Ryū, developed almost simultaneously and independently.<a href="#cite_note-Giulietti-48">[40]</a> Performs better than Ryū and Grisu3 in certain benchmarks.<a href="#cite_note-abolz-49">[41]</a></li></ul>

Many modern language runtimes use Grisu3 with a Dragon4 fallback.<a href="#cite_note-double_conversion_2020-50">[42]</a>

<div class="mw-heading mw-heading3"><h3 id="Decimal-to-binary_conversion">Decimal-to-binary conversion</h3></div>

The problem of parsing a decimal string into a binary FP representation is complex, with an accurate parser not appearing until Clinger's 1990 work (implemented in dtoa.c).<a href="#cite_note-Gay_1990-44">[36]</a> Further work has likewise progressed in the direction of faster parsing.<a href="#cite_note-Lemire_2021-51">[43]</a>

<div class="mw-heading mw-heading2"><h2 id="Floating-point_operations">Floating-point operations</h2></div>

For ease of presentation and understanding, decimal <a href="Radix" title="Radix">radix</a> with 7 digit precision will be used in the examples, as in the IEEE 754 decimal32 format. The fundamental principles are the same in any <a href="Radix" title="Radix">radix</a> or precision, except that normalization is optional (it does not affect the numerical value of the result). Here, s denotes the significand and e denotes the exponent.

<div class="mw-heading mw-heading3"><h3 id="Addition_and_subtraction">Addition and subtraction</h3></div>

A simple method to add floating-point numbers is to first represent them with the same exponent. In the example below, the second number (with the smaller exponent) is shifted right by three digits, and one then proceeds with the usual addition method:

<pre> 123456.7 = 1.234567 × 10^5

101.7654 = 1.017654 × 10^2 = 0.001017654 × 10^5

</pre>

<pre> Hence:

123456.7 + 101.7654 = (1.234567 × 10^5) + (1.017654 × 10^2)

= (1.234567 × 10^5) + (0.001017654 × 10^5)

= (1.234567 + 0.001017654) × 10^5

= 1.235584654 × 10^5

</pre>

In detail:

<pre> e=5; s=1.234567 (123456.7)

+ e=2; s=1.017654 (101.7654)

</pre>

<pre> e=5; s=1.234567

+ e=5; s=0.001017654 (after shifting)

--------------------

e=5; s=1.235584654 (true sum: 123558.4654)

</pre>

This is the true result, the exact sum of the operands. It will be rounded to seven digits and then normalized if necessary. The final result is

<pre> e=5; s=1.235585 (final sum: 123558.5)

</pre>

The lowest three digits of the second operand (654) are essentially lost. This is <a href="Round-off_error" title="Round-off error">round-off error</a>. In extreme cases, the sum of two non-zero numbers may be equal to one of them:

<pre> e=5; s=1.234567

+ e=−3; s=9.876543

</pre>

<pre> e=5; s=1.234567

+ e=5; s=0.00000009876543 (after shifting)

----------------------

e=5; s=1.23456709876543 (true sum)

e=5; s=1.234567 (after rounding and normalization)

</pre>

In the above conceptual examples it would appear that a large number of extra digits would need to be provided by the adder to ensure correct rounding; however, for binary addition or subtraction using careful implementation techniques only a guard bit, a rounding bit and one extra sticky bit need to be carried beyond the precision of the operands.<a href="#cite_note-Goldberg_1991-25">[18]</a><a href="#cite_note-Patterson-Hennessy_2014-52">[44]</a>: 218–220

Another problem of loss of significance occurs when approximations to two nearly equal numbers are subtracted. In the following example e = 5; s = 1.234571 and e = 5; s = 1.234567 are approximations to the rationals 123457.1467 and 123456.659.

<pre> e=5; s=1.234571

− e=5; s=1.234567

----------------

e=5; s=0.000004

e=−1; s=4.000000 (after rounding and normalization)

</pre>

The floating-point difference is computed exactly because the numbers are close—the <a href="Sterbenz_lemma" title="Sterbenz lemma">Sterbenz lemma</a> guarantees this, even in case of underflow when <a href="Gradual_underflow" class="mw-redirect" title="Gradual underflow">gradual underflow</a> is supported. Despite this, the difference of the original numbers is e = −1; s = 4.877000, which differs more than 20% from the difference e = −1; s = 4.000000 of the approximations. In extreme cases, all significant digits of precision can be lost.<a href="#cite_note-Goldberg_1991-25">[18]</a><a href="#cite_note-Sierra_1962-53">[45]</a> This <a href="Catastrophic_cancellation" title="Catastrophic cancellation">cancellation</a> illustrates the danger in assuming that all of the digits of a computed result are meaningful. Dealing with the consequences of these errors is a topic in <a href="Numerical_analysis" title="Numerical analysis">numerical analysis</a>; see also <a href="#Accuracy_problems">Accuracy problems</a>.

<div class="mw-heading mw-heading3"><h3 id="Multiplication_and_division">Multiplication and division</h3></div>

To multiply, the significands are multiplied while the exponents are added, and the result is rounded and normalized.

<pre> e=3; s=4.734612

× e=5; s=5.417242

-----------------------

e=8; s=25.648538980104 (true product)

e=8; s=25.64854 (after rounding)

e=9; s=2.564854 (after normalization)

</pre>

Similarly, division is accomplished by subtracting the divisor's exponent from the dividend's exponent, and dividing the dividend's significand by the divisor's significand.

There are no cancellation or absorption problems with multiplication or division, though small errors may accumulate as operations are performed in succession.<a href="#cite_note-Goldberg_1991-25">[18]</a> In practice, the way these operations are carried out in digital logic can be quite complex (see <a href="Booth's_multiplication_algorithm" title="Booth's multiplication algorithm">Booth's multiplication algorithm</a> and <a href="Division_algorithm" title="Division algorithm">Division algorithm</a>).<a href="#cite_note-NB_2-54">[nb 9]</a>

<div class="mw-heading mw-heading3"><h3 id="Literal_syntax">Literal syntax</h3></div>

Literals for floating-point numbers depend on languages. They typically use <code>e</code> or <code>E</code> to denote <a href="Scientific_notation" title="Scientific notation">scientific notation</a>. The <a href="C_(programming_language)" title="C (programming language)">C programming language</a> and the <a href="IEEE_754" title="IEEE 754">IEEE 754</a> standard also define a <a href="IEEE_754#Hexadecimal_literals" title="IEEE 754">hexadecimal literal syntax</a> with a base-2 exponent instead of 10. In languages like <a href="C_(programming_language)" title="C (programming language)">C</a>, when the decimal exponent is omitted, a decimal point is needed to differentiate them from integers. Other languages do not have an integer type (such as <a href="JavaScript" title="JavaScript">JavaScript</a>), or allow overloading of numeric types (such as <a href="Haskell_(programming_language)" class="mw-redirect" title="Haskell (programming language)">Haskell</a>). In these cases, digit strings such as <code>123</code> may also be floating-point literals.

Examples of floating-point literals are:

<li><code>0x1.fffffep+127</code> in C and IEEE 754</li></ul>

<div class="mw-heading mw-heading2"><h2 id="Dealing_with_exceptional_cases">Dealing with exceptional cases </h2></div>

<div role="note" class="hatnote navigation-not-searchable">Further information: <a href="IEEE_754#Exception_handling" title="IEEE 754">IEEE 754 § Exception handling</a></div>

Floating-point computation in a computer can run into three kinds of problems:

<ul><li>An operation can be mathematically undefined, such as ∞/∞, or <a href="Division_by_zero" title="Division by zero">division by zero</a>.</li>

<li>An operation can be legal in principle, but not supported by the specific format, for example, calculating the <a href="Square_root" title="Square root">square root</a> of −1 or the inverse sine of 2 (both of which result in <a href="Complex_number" title="Complex number">complex numbers</a>).</li>

<li>An operation can be legal in principle, but the result can be impossible to represent in the specified format, because the exponent is too large or too small to encode in the exponent field. Such an event is called an overflow (exponent too large), <a href="Arithmetic_underflow" title="Arithmetic underflow">underflow</a> (exponent too small) or <a href="Subnormal_number" title="Subnormal number">denormalization</a> (precision loss).</li></ul>

Prior to the IEEE standard, such conditions usually caused the program to terminate, or triggered some kind of <a href="Trap_(computing)" class="mw-redirect" title="Trap (computing)">trap</a> that the programmer might be able to catch. How this worked was system-dependent, meaning that floating-point programs were not <a href="Porting" title="Porting">portable</a>. (The term "exception" as used in IEEE 754 is a general term meaning an exceptional condition, which is not necessarily an error, and is a different usage to that typically defined in programming languages such as a C++ or Java, in which an "<a href="Exception_handling" title="Exception handling">exception</a>" is an alternative flow of control, closer to what is termed a "trap" in IEEE 754 terminology.)

Here, the required default method of handling exceptions according to IEEE 754 is discussed (the IEEE 754 optional trapping and other "alternate exception handling" modes are not discussed). Arithmetic exceptions are (by default) required to be recorded in "sticky" status flag bits. That they are "sticky" means that they are not reset by the next (arithmetic) operation, but stay set until explicitly reset. The use of "sticky" flags thus allows for testing of exceptional conditions to be delayed until after a full floating-point expression or subroutine: without them exceptional conditions that could not be otherwise ignored would require explicit testing immediately after every floating-point operation. By default, an operation always returns a result according to specification without interrupting computation. For instance, 1/0 returns +∞, while also setting the divide-by-zero flag bit (this default of ∞ is designed to often return a finite result when used in subsequent operations and so be safely ignored).

The original IEEE 754 standard, however, failed to recommend operations to handle such sets of arithmetic exception flag bits. So while these were implemented in hardware, initially programming language implementations typically did not provide a means to access them (apart from assembler). Over time some programming language standards (e.g., <a href="C99" title="C99">C99</a>/C11 and Fortran) have been updated to specify methods to access and change status flag bits. The 2008 version of the IEEE 754 standard now specifies a few operations for accessing and handling the arithmetic flag bits. The programming model is based on a single thread of execution and use of them by multiple threads has to be handled by a <a href="Concurrency_(computer_science)" title="Concurrency (computer science)">means</a> outside of the standard (e.g. <a href="C11_(C_standard_revision)" title="C11 (C standard revision)">C11</a> specifies that the flags have <a href="Thread-local_storage" title="Thread-local storage">thread-local storage</a>).

IEEE 754 specifies five arithmetic exceptions that are to be recorded in the status flags ("sticky bits"):

<ul><li>inexact, set if the rounded (and returned) value is different from the mathematically exact result of the operation.</li>

<li>underflow, set if the rounded value is tiny (as specified in IEEE 754) and inexact (or maybe limited to if it has denormalization loss, as per the 1985 version of IEEE 754), returning a subnormal value including the zeros.</li>

<li>overflow, set if the absolute value of the rounded value is too large to be represented. An infinity or maximal finite value is returned, depending on which rounding is used.</li>

<li>divide-by-zero, set if the result is infinite given finite operands, returning an infinity, either +∞ or −∞.</li>

<li>invalid, set if a finite or infinite result cannot be returned e.g. sqrt(−1) or 0/0, returning a quiet NaN.</li></ul>

The default return value for each of the exceptions is designed to give the correct result in the majority of cases such that the exceptions can be ignored in the majority of codes. inexact returns a correctly rounded result, and underflow returns a value less than or equal to the smallest positive normal number in magnitude and can almost always be ignored.<a href="#cite_note-Kahan_1997_Status-55">[46]</a> divide-by-zero returns infinity exactly, which will typically then divide a finite number and so give zero, or else will give an invalid exception subsequently if not, and so can also typically be ignored. For example, the effective resistance of n resistors in parallel (see fig. 1) is given by <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle R_{\text{tot}}=1/(1/R_{1}+1/R_{2}+\cdots +1/R_{n})}">

<msub>

</mrow>

</msub>

</mrow>

</mrow>

<msub>

</mrow>

</msub>

</mrow>

<msub>

</mrow>

</msub>

</mrow>

<msub>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle R_{\text{tot}}=1/(1/R_{1}+1/R_{2}+\cdots +1/R_{n})}</annotation>

</semantics>

</math><img src="./cd57a135e4c4097f3aac012da434c472b1fbe90b.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.838ex; width:38.168ex; height:2.843ex;" alt="{\displaystyle R_{\text{tot}}=1/(1/R_{1}+1/R_{2}+\cdots +1/R_{n})}" loading="lazy">. If a short-circuit develops with <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle R_{1}}">

<msub>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle R_{1}}</annotation>

</semantics>

</math><img src="./c1d63c96f59d98589d923c4f0b04222feaa7283e.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:2.818ex; height:2.509ex;" alt="{\displaystyle R_{1}}" loading="lazy"> set to 0, <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle 1/R_{1}}">

</mrow>

<msub>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 1/R_{1}}</annotation>

</semantics>

</math><img src="./e6e4aaeaef2ceb43e4b429f07d7bfd798d3db831.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.838ex; width:5.143ex; height:2.843ex;" alt="{\displaystyle 1/R_{1}}" loading="lazy"> will return +infinity which will give a final <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle R_{tot}}">

<msub>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle R_{tot}}</annotation>

</semantics>

</math><img src="./5db9ad2846f85f109dcfaaa60106f69304ff2377.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:3.981ex; height:2.509ex;" alt="{\displaystyle R_{tot}}" loading="lazy"> of 0, as expected<a href="#cite_note-Intel-56">[47]</a> (see the continued fraction example of <a href="Floating_point" class="mw-redirect" title="Floating point">IEEE 754 design rationale</a> for another example).

Overflow and invalid exceptions can typically not be ignored, but do not necessarily represent errors: for example, a <a href="Zero_of_a_function" title="Zero of a function">root-finding</a> routine, as part of its normal operation, may evaluate a passed-in function at values outside of its domain, returning NaN and an invalid exception flag to be ignored until finding a useful start point.<a href="#cite_note-Kahan_1997_Status-55">[46]</a>

<div class="mw-heading mw-heading2"><h2 id="Accuracy_problems">Accuracy problems</h2></div>

The fact that floating-point numbers cannot accurately represent all real numbers, and that floating-point operations cannot accurately represent true arithmetic operations, leads to many surprising situations. This is related to the finite <a href="Precision_(computer_science)" title="Precision (computer science)">precision</a> with which computers generally represent numbers.

For example, the decimal numbers 0.1 and 0.01 cannot be represented exactly as binary floating-point numbers. In the IEEE 754 binary32 format with its 24-bit significand, the result of attempting to square the approximation to 0.1 is neither 0.01 nor the representable number closest to it. The decimal number 0.1 is represented in binary as <var style="padding-right: 1px;">e</var> = −4; <var style="padding-right: 1px;">s</var> = 110011001100110011001101, which is

/* start https://en.wikipedia.org/ */

.mw-parser-output .block-indent{padding-left:3em;padding-right:0;overflow:hidden}

/* end https://en.wikipedia.org/ */

</style><div class="block-indent">0.100000001490116119384765625 exactly.</div>

Squaring this number gives

<div class="block-indent">0.010000000298023226097399174250313080847263336181640625 exactly.</div>

Squaring it with rounding to the 24-bit precision gives

<div class="block-indent">0.010000000707805156707763671875 exactly.</div>

But the representable number closest to 0.01 is

<div class="block-indent">0.009999999776482582092285156250 exactly.</div>

Also, the non-representability of π (and π/2) means that an attempted computation of tan(π/2) will not yield a result of infinity, nor will it even overflow in the usual floating-point formats (assuming an accurate implementation of tan). It is simply not possible for standard floating-point hardware to attempt to compute tan(π/2), because π/2 cannot be represented exactly. This computation in C:

<div class="mw-highlight mw-highlight-lang-c mw-content-ltr" dir="ltr"><pre>/* Enough digits to be sure we get the correct approximation. */

double pi = 3.1415926535897932384626433832795;

double z = tan(pi/2.0);

</pre></div>

will give a result of 16331239353195370.0. In single precision (using the <code>tanf</code> function), the result will be −22877332.0.

By the same token, an attempted computation of sin(π) will not yield zero. The result will be (approximately) 0.1225×10−15 in double precision, or −0.8742×10−7 in single precision.<a href="#cite_note-NB_3-57">[nb 10]</a>

While floating-point addition and multiplication are both <a href="Commutative" class="mw-redirect" title="Commutative">commutative</a> (<var style="padding-right: 1px;">a</var> + <var style="padding-right: 1px;">b</var> = <var style="padding-right: 1px;">b</var> + <var style="padding-right: 1px;">a</var> and <var style="padding-right: 1px;">a</var> × <var style="padding-right: 1px;">b</var> = <var style="padding-right: 1px;">b</var> × <var style="padding-right: 1px;">a</var>), they are not necessarily <a href="Associative_property" title="Associative property">associative</a>. That is, (<var style="padding-right: 1px;">a</var> + <var style="padding-right: 1px;">b</var>) + <var style="padding-right: 1px;">c</var> is not necessarily equal to <var style="padding-right: 1px;">a</var> + (<var style="padding-right: 1px;">b</var> + <var style="padding-right: 1px;">c</var>). Using 7-digit significand decimal arithmetic:

<pre> a = 1234.567, b = 45.67834, c = 0.0004

</pre>

<pre> (a + b) + c:

1234.567 (a)

+ 45.67834 (b)

____________

1280.24534 rounds to 1280.245

</pre>

<pre> 1280.245 (a + b)

+ 0.0004 (c)

____________

1280.2454 rounds to 1280.245 ← (a + b) + c

</pre>

<pre> a + (b + c):

45.67834 (b)

+ 0.0004 (c)

____________

45.67874

</pre>

<pre> 1234.567 (a)

+ 45.67874 (b + c)

____________

1280.24574 rounds to 1280.246 ← a + (b + c)

</pre>

They are also not necessarily <a href="Distributive_property" title="Distributive property">distributive</a>. That is, (<var style="padding-right: 1px;">a</var> + <var style="padding-right: 1px;">b</var>) × <var style="padding-right: 1px;">c</var> may not be the same as <var style="padding-right: 1px;">a</var> × <var style="padding-right: 1px;">c</var> + <var style="padding-right: 1px;">b</var> × <var style="padding-right: 1px;">c</var>:

<pre> 1234.567 × 3.333333 = 4115.223

1.234567 × 3.333333 = 4.115223

4115.223 + 4.115223 = 4119.338

but

1234.567 + 1.234567 = 1235.802

1235.802 × 3.333333 = 4119.340

</pre>

In addition to loss of significance, inability to represent numbers such as π and 0.1 exactly, and other slight inaccuracies, the following phenomena may occur:

<div><ul><li><a href="Catastrophic_cancellation" title="Catastrophic cancellation">Cancellation</a>: subtraction of nearly equal operands may cause extreme loss of accuracy.<a href="#cite_note-Harris-58">[48]</a><a href="#cite_note-Sierra_1962-53">[45]</a> When we subtract two almost equal numbers we set the most significant digits to zero, leaving ourselves with just the insignificant, and most erroneous, digits.<a href="#cite_note-Muller_2010-1">[1]</a>: 124 For example, when determining a <a href="Derivative" title="Derivative">derivative</a> of a function the following formula is used:

<mfrac>

<mrow>

</mrow>

</mfrac>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle Q(h)={\frac {f(a+h)-f(a)}{h}}.}</annotation>

</semantics>

</math>

Intuitively one would want an <var style="padding-right: 1px;">h</var> very close to zero; however, when using floating-point operations, the smallest number will not give the best approximation of a derivative. As <var style="padding-right: 1px;">h</var> grows smaller, the difference between <var style="padding-right: 1px;">f</var>(<var style="padding-right: 1px;">a</var> + <var style="padding-right: 1px;">h</var>) and <var style="padding-right: 1px;">f</var>(<var style="padding-right: 1px;">a</var>) grows smaller, cancelling out the most significant and least erroneous digits and making the most erroneous digits more important. As a result the smallest number of <var style="padding-right: 1px;">h</var> possible will give a more erroneous approximation of a derivative than a somewhat larger number. This is perhaps the most common and serious accuracy problem.</li><li>Conversions to integer are not intuitive: converting (63.0/9.0) to integer yields 7, but converting (0.63/0.09) may yield 6. This is because conversions generally truncate rather than round. <a href="Floor_and_ceiling_functions" title="Floor and ceiling functions">Floor and ceiling functions</a> may produce answers which are off by one from the intuitively expected value.</li><li>Limited exponent range: results might overflow yielding infinity, or underflow yielding a <a href="Subnormal_number" title="Subnormal number">subnormal number</a> or zero. In these cases precision will be lost.</li><li>Testing for <a href="Division_by_zero#Computer_arithmetic" title="Division by zero">safe division</a> is problematic: Checking that the divisor is not zero does not guarantee that a division will not overflow.</li><li>Testing for equality is problematic. Two computational sequences that are mathematically equal may well produce different floating-point values.<a href="#cite_note-Barker-59">[49]</a></li></ul></div>

<div class="mw-heading mw-heading3"><h3 id="Incidents">Incidents</h3></div>

<ul><li>On 25 February 1991, a <a href="Loss_of_significance" class="mw-redirect" title="Loss of significance">loss of significance</a> in a <a href="MIM-104_Patriot" title="MIM-104 Patriot">MIM-104 Patriot</a> missile battery <a href="MIM-104_Patriot#Failure_at_Dhahran" title="MIM-104 Patriot">prevented it from intercepting</a> an incoming <a href="Al_Hussein_(missile)" class="mw-redirect" title="Al Hussein (missile)">Scud</a> missile in <a href="Dhahran" title="Dhahran">Dhahran</a>, <a href="Saudi_Arabia" title="Saudi Arabia">Saudi Arabia</a>, contributing to the death of 28 soldiers from the U.S. Army's <a href="14th_Quartermaster_Detachment" title="14th Quartermaster Detachment">14th Quartermaster Detachment</a>.<a href="#cite_note-GAO_report_IMTEC_92-26-60">[50]</a> The weapons control computer counted time in an integer number of tenths of a second since boot. For conversion to a floating-point number of seconds in velocity and position calculations, the software originally multiplied this number by a 24-bit <a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">fixed-point</a> binary approximation to 0.1, specifically <math display="block" xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle 0.00011001100110011001100_{2}=0.1\times (1-2^{-20}).}">

<msub>

</mrow>

</msub>

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle 0.00011001100110011001100_{2}=0.1\times (1-2^{-20}).}</annotation>

</semantics>

</math> Some parts of the software were later adapted to use a more accurate conversion to floating-point, but some parts were not updated and still used the 24-bit approximation.<a href="#cite_note-Skeel-61">[51]</a> These parts of the software drifted from one another by about 3.43 milliseconds per hour. After 20 hours, the discrepancy of about 68.7 ms was enough for the radar tracking system to lose track of Scuds; the control system in the Dhahran missile battery had been running for about 100 hours when it failed to track and intercept an incoming Scud.<a href="#cite_note-GAO_report_IMTEC_92-26-60">[50]</a> The failure to intercept arose not from using floating point specifically, but from subtracting two different approximations to unit conversion with different errors when representing time, so the unit conversion error in the difference did not cancel out but rather grew indefinitely with uptime.<a href="#cite_note-Skeel-61">[51]</a></li>

<li><a href="Salami_slicing_tactics#Financial_schemes" title="Salami slicing tactics">Salami slicing</a> is the practice of removing the 'invisible' part of a transaction into a separate account.</li></ul>

<div class="mw-heading mw-heading3"><h3 id="Machine_precision_and_backward_error_analysis">Machine precision and backward error analysis</h3></div>

Machine precision is a quantity that characterizes the accuracy of a floating-point system, and is used in <a href="Error_analysis_(mathematics)#Error_analysis_in_numerical_modeling" title="Error analysis (mathematics)">backward error analysis</a> of floating-point algorithms. It is also known as unit roundoff or <a href="Machine_epsilon" title="Machine epsilon">machine epsilon</a>. Usually denoted <var style="padding-right: 1px;">Ε</var>mach, its value depends on the particular rounding being used.

With rounding to zero,

<msub>

</mrow>

</mrow>

</msub>

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \mathrm {E} _{\text{mach}}=B^{1-P},\,}</annotation>

</semantics>

</math>

whereas rounding to nearest,

<msub>

</mrow>

</mrow>

</msub>

<mfrac>

</mfrac>

</mstyle>

</mrow>

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \mathrm {E} _{\text{mach}}={\tfrac {1}{2}}B^{1-P},}</annotation>

</semantics>

</math>

where B is the base of the system and P is the precision of the significand (in base B).

This is important since it bounds the <a href="Relative_error" class="mw-redirect" title="Relative error">relative error</a> in representing any non-zero real number <var style="padding-right: 1px;">x</var> within the normalized range of a floating-point system:

<mrow>

<mfrac>

<mrow>

</mrow>

</mfrac>

</mrow>

</mrow>

<msub>

</mrow>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \left|{\frac {\operatorname {fl} (x)-x}{x}}\right|\leq \mathrm {E} _{\text{mach}}.}</annotation>

</semantics>

</math>

Backward error analysis, the theory of which was developed and popularized by <a href="James_H._Wilkinson" title="James H. Wilkinson">James H. Wilkinson</a>, can be used to establish that an algorithm implementing a numerical function is numerically stable.<a href="#cite_note-RalstonReilly2003-62">[52]</a> The basic approach is to show that although the calculated result, due to roundoff errors, will not be exactly correct, it is the exact solution to a nearby problem with slightly perturbed input data. If the perturbation required is small, on the order of the uncertainty in the input data, then the results are in some sense as accurate as the data "deserves". The algorithm is then defined as <a href="Numerical_stability#Forward,_backward,_and_mixed_stability" title="Numerical stability">backward stable</a>. Stability is a measure of the sensitivity to rounding errors of a given numerical procedure; by contrast, the <a href="Condition_number" title="Condition number">condition number</a> of a function for a given problem indicates the inherent sensitivity of the function to small perturbations in its input and is independent of the implementation used to solve the problem.<a href="#cite_note-Einarsson_2005-63">[53]</a>

As a trivial example, consider a simple expression giving the inner product of (length two) vectors <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle x}">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle x}</annotation>

</semantics>

</math><img src="./87f9e315fd7e2ba406057a97300593c4802b53e4.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:1.33ex; height:1.676ex;" alt="{\displaystyle x}" loading="lazy"> and <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle y}">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle y}</annotation>

</semantics>

</math><img src="./b8a6208ec717213d4317e666f1ae872e00620a0d.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:1.155ex; height:2.009ex;" alt="{\displaystyle y}" loading="lazy">, then

<mtr>

<mtd>

</mtd>

<mtd>

</mrow>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mrow>

</mtd>

<mtd>

<mtext> where </mtext>

</mrow>

<mtext> indicates correctly rounded floating-point arithmetic</mtext>

</mrow>

</mtd>

</mtr>

<mtr>

<mtd>

</mrow>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mrow>

</mtd>

<mtd>

<mtext> where </mtext>

</mrow>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</mrow>

</msub>

<mtext> from above</mtext>

</mrow>

</mtd>

</mtr>

<mtr>

<mtd>

</mrow>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mrow>

<msub>

</mrow>

</msub>

</mtd>

</mtr>

<mtr>

<mtd>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mtd>

</mtr>

</mtable>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle {\begin{aligned}\operatorname {fl} (x\cdot y)&=\operatorname {fl} {\big (}\operatorname {fl} (x_{1}\cdot y_{1})+\operatorname {fl} (x_{2}\cdot y_{2}){\big )},&&{\text{ where }}\operatorname {fl} (){\text{ indicates correctly rounded floating-point arithmetic}}\\&=\operatorname {fl} {\big (}(x_{1}\cdot y_{1})(1+\delta _{1})+(x_{2}\cdot y_{2})(1+\delta _{2}){\big )},&&{\text{ where }}\delta _{n}\leq \mathrm {E} _{\text{mach}},{\text{ from above}}\\&={\big (}(x_{1}\cdot y_{1})(1+\delta _{1})+(x_{2}\cdot y_{2})(1+\delta _{2}){\big )}(1+\delta _{3})\\&=(x_{1}\cdot y_{1})(1+\delta _{1})(1+\delta _{3})+(x_{2}\cdot y_{2})(1+\delta _{2})(1+\delta _{3}),\end{aligned}}}</annotation>

</semantics>

</math>

and so

<mover>

</mover>

</mrow>

<mover>

</mover>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \operatorname {fl} (x\cdot y)={\hat {x}}\cdot {\hat {y}},}</annotation>

</semantics>

</math>

where

<mtr>

<mtd>

<msub>

<mover>

</mover>

</mrow>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

<mover>

</mover>

</mrow>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mtd>

</mtr>

<mtr>

<mtd>

<msub>

<mover>

</mover>

</mrow>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

<mover>

</mover>

</mrow>

</mrow>

</msub>

</mtd>

<mtd>

<msub>

</mrow>

</msub>

<msub>

</mrow>

</msub>

</mtd>

</mtr>

</mtable>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle {\begin{aligned}{\hat {x}}_{1}&=x_{1}(1+\delta _{1});&{\hat {x}}_{2}&=x_{2}(1+\delta _{2});\\{\hat {y}}_{1}&=y_{1}(1+\delta _{3});&{\hat {y}}_{2}&=y_{2}(1+\delta _{3}),\\\end{aligned}}}</annotation>

</semantics>

</math>

where

<msub>

</mrow>

</msub>

<msub>

</mrow>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \delta _{n}\leq \mathrm {E} _{\text{mach}}}</annotation>

</semantics>

</math>

by definition, which is the sum of two slightly perturbed (on the order of Εmach) input data, and so is backward stable. For more realistic examples in <a href="Numerical_linear_algebra" title="Numerical linear algebra">numerical linear algebra</a>, see Higham 2002<a href="#cite_note-Higham_2002-64">[54]</a> and other references below.

<div class="mw-heading mw-heading3"><h3 id="Minimizing_the_effect_of_accuracy_problems">Minimizing the effect of accuracy problems</h3></div>

Although individual arithmetic operations of IEEE 754 are guaranteed accurate to within half a <a href="Unit_in_the_last_place" title="Unit in the last place">ULP</a>, more complicated formulae can suffer from larger errors for a variety of reasons. The loss of accuracy can be substantial if a problem or its data are <a href="Condition_number" title="Condition number">ill-conditioned</a>, meaning that the correct result is hypersensitive to tiny perturbations in its data. However, even functions that are well-conditioned can suffer from large loss of accuracy if an algorithm <a href="Numerical_stability" title="Numerical stability">numerically unstable</a> for that data is used: apparently equivalent formulations of expressions in a programming language can differ markedly in their numerical stability. One approach to remove the risk of such loss of accuracy is the design and analysis of numerically stable algorithms, which is an aim of the branch of mathematics known as <a href="Numerical_analysis" title="Numerical analysis">numerical analysis</a>. Another approach that can protect against the risk of numerical instabilities is the computation of intermediate (scratch) values in an algorithm at a higher precision than the final result requires,<a href="#cite_note-OliveiraStewart_2006-65">[55]</a> which can remove, or reduce by orders of magnitude,<a href="#cite_note-Kahan_2005_ARITH17-66">[56]</a> such risk: <a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">IEEE 754 quadruple precision</a> and <a href="Extended_precision" title="Extended precision">extended precision</a> are designed for this purpose when computing at double precision.<a href="#cite_note-Kahan_2011_Debug-67">[57]</a><a href="#cite_note-NB_4-68">[nb 11]</a>

For example, the following algorithm is a direct implementation to compute the function <var style="padding-right: 1px;">A</var>(<var style="padding-right: 1px;">x</var>) = (<var style="padding-right: 1px;">x</var>−1) / (exp(<var style="padding-right: 1px;">x</var>−1) − 1) which is well-conditioned at 1.0,<a href="#cite_note-NB_5-69">[nb 12]</a> however it can be shown to be numerically unstable and lose up to half the significant digits carried by the arithmetic when computed near 1.0.<a href="#cite_note-Kahan_2001_JavaHurt-70">[58]</a>

<div class="mw-highlight mw-highlight-lang-c mw-content-ltr mw-highlight-lines" dir="ltr"><pre>double A(double X)

{

double Y, Z; // [1]

Y = X - 1.0;

Z = exp(Y);

if (Z != 1.0)

Z = Y / (Z - 1.0); // [2]

return Z;

}

</pre></div>

If, however, intermediate computations are all performed in extended precision (e.g. by setting line [1] to <a href="C99" title="C99">C99</a> <code class="mw-highlight mw-highlight-lang-text mw-content-ltr" style="" dir="ltr">long double</code>), then up to full precision in the final double result can be maintained.<a href="#cite_note-NB_6-71">[nb 13]</a> Alternatively, a numerical analysis of the algorithm reveals that if the following non-obvious change to line [2] is made:

<div class="mw-highlight mw-highlight-lang-c mw-content-ltr" dir="ltr"><pre>Z = log(Z) / (Z - 1.0);

</pre></div>

then the algorithm becomes numerically stable and can compute to full double precision.

To maintain the properties of such carefully constructed numerically stable programs, careful handling by the <a href="Compiler" title="Compiler">compiler</a> is required. Certain "optimizations" that compilers might make (for example, reordering operations) can work against the goals of well-behaved software. There is some controversy about the failings of compilers and language designs in this area: C99 is an example of a language where such optimizations are carefully specified to maintain numerical precision. See the external references at the bottom of this article.

A detailed treatment of the techniques for writing high-quality floating-point software is beyond the scope of this article, and the reader is referred to,<a href="#cite_note-Higham_2002-64">[54]</a><a href="#cite_note-Kahan_2000_Marketing-72">[59]</a> and the other references at the bottom of this article. Kahan suggests several rules of thumb that can substantially decrease by orders of magnitude<a href="#cite_note-Kahan_2000_Marketing-72">[59]</a> the risk of numerical anomalies, in addition to, or in lieu of, a more careful numerical analysis. These include: as noted above, computing all expressions and intermediate results in the highest precision supported in hardware (a common rule of thumb is to carry twice the precision of the desired result, i.e. compute in double precision for a final single-precision result, or in double extended or quad precision for up to double-precision results<a href="#cite_note-Kahan_1981_WhyIEEE-73">[60]</a>); and rounding input data and results to only the precision required and supported by the input data (carrying excess precision in the final result beyond that required and supported by the input data can be misleading, increases storage cost and decreases speed, and the excess bits can affect convergence of numerical procedures:<a href="#cite_note-Kahan_2001_LN-74">[61]</a> notably, the first form of the iterative example given below converges correctly when using this rule of thumb). Brief descriptions of several additional issues and techniques follow.

As decimal fractions can often not be exactly represented in binary floating-point, such arithmetic is at its best when it is simply being used to measure real-world quantities over a wide range of scales (such as the orbital period of a moon around Saturn or the mass of a <a href="Proton" title="Proton">proton</a>), and at its worst when it is expected to model the interactions of quantities expressed as decimal strings that are expected to be exact.<a href="#cite_note-Kahan_2005_ARITH17-66">[56]</a><a href="#cite_note-Kahan_2000_Marketing-72">[59]</a> An example of the latter case is financial calculations. For this reason, financial software tends not to use a binary floating-point number representation.<a href="#cite_note-Speleotrove_2012-75">[62]</a> The "decimal" data type of the <a href="C_Sharp_(programming_language)" title="C Sharp (programming language)">C#</a> and <a href="Python_(programming_language)" title="Python (programming language)">Python</a> programming languages, and the decimal formats of the <a href="IEEE_754-2008" class="mw-redirect" title="IEEE 754-2008">IEEE 754-2008</a> standard, are designed to avoid the problems of binary floating-point representations when applied to human-entered exact decimal values, and make the arithmetic always behave as expected when numbers are printed in decimal.

Expectations from mathematics may not be realized in the field of floating-point computation. For example, it is known that <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle (x+y)(x-y)=x^{2}-y^{2}\,}">

<msup>

</mrow>

</msup>

<msup>

</mrow>

</msup>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle (x+y)(x-y)=x^{2}-y^{2}\,}</annotation>

</semantics>

</math><img src="./42f6b57467359a23ff92ad97626a078c836d5a3e.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.838ex; width:25.194ex; height:3.176ex;" alt="{\displaystyle (x+y)(x-y)=x^{2}-y^{2}\,}" loading="lazy">, and that <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \sin ^{2}{\theta }+\cos ^{2}{\theta }=1\,}">

<msup>

</mrow>

</msup>

</mrow>

<msup>

</mrow>

</msup>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \sin ^{2}{\theta }+\cos ^{2}{\theta }=1\,}</annotation>

</semantics>

</math><img src="./a9d6c234a50e2ecda8445a62785f9c27167a4d34.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.505ex; width:18.519ex; height:2.843ex;" alt="{\displaystyle \sin ^{2}{\theta }+\cos ^{2}{\theta }=1\,}" loading="lazy">, however these facts cannot be relied on when the quantities involved are the result of floating-point computation.

The use of the equality test (<code>if (x==y) ...</code>) requires care when dealing with floating-point numbers. Even simple expressions like <code>0.6/0.2-3==0</code> will, on most computers, fail to be true<a href="#cite_note-Christiansen_Perl-76">[63]</a> (in IEEE 754 double precision, for example, <code>0.6/0.2 - 3</code> is approximately equal to −4.44089209850063×10−16). Consequently, such tests are sometimes replaced with "fuzzy" comparisons (<code>if (abs(x-y) < epsilon) ...</code>, where epsilon is sufficiently small and tailored to the application, such as 1.0E−13). The wisdom of doing this varies greatly, and can require numerical analysis to bound epsilon.<a href="#cite_note-Higham_2002-64">[54]</a> Values derived from the primary data representation and their comparisons should be performed in a wider, extended, precision to minimize the risk of such inconsistencies due to round-off errors.<a href="#cite_note-Kahan_2000_Marketing-72">[59]</a> It is often better to organize the code in such a way that such tests are unnecessary. For example, in <a href="Computational_geometry" title="Computational geometry">computational geometry</a>, exact tests of whether a point lies off or on a line or plane defined by other points can be performed using adaptive precision or exact arithmetic methods.<a href="#cite_note-Shewchuk-77">[64]</a>

Small errors in floating-point arithmetic can grow when mathematical algorithms perform operations an enormous number of times. A few examples are <a href="Matrix_inversion" class="mw-redirect" title="Matrix inversion">matrix inversion</a>, <a href="Eigenvector" class="mw-redirect" title="Eigenvector">eigenvector</a> computation, and differential equation solving. These algorithms must be very carefully designed, using numerical approaches such as <a href="Iterative_refinement" title="Iterative refinement">iterative refinement</a>, if they are to work well.<a href="#cite_note-Kahan_1997_Cantilever-78">[65]</a>

Summation of a vector of floating-point values is a basic algorithm in <a href="Computational_science" title="Computational science">scientific computing</a>, and so an awareness of when loss of significance can occur is essential. For example, if one is adding a very large number of numbers, the individual addends are very small compared with the sum. This can lead to loss of significance. A typical addition would then be something like

<pre>3253.671

+ 3.141276

-----------

3256.812

</pre>

The low 3 digits of the addends are effectively lost. Suppose, for example, that one needs to add many numbers, all approximately equal to 3. After 1000 of them have been added, the running sum is about 3000; the lost digits are not regained. The <a href="Kahan_summation_algorithm" title="Kahan summation algorithm">Kahan summation algorithm</a> may be used to reduce the errors.<a href="#cite_note-Higham_2002-64">[54]</a>

Round-off error can affect the convergence and accuracy of iterative numerical procedures. As an example, <a href="Archimedes" title="Archimedes">Archimedes</a> approximated π by calculating the perimeters of polygons inscribing and circumscribing a circle, starting with hexagons, and successively doubling the number of sides. As noted above, computations may be rearranged in a way that is mathematically equivalent but less prone to error (<a href="Numerical_analysis" title="Numerical analysis">numerical analysis</a>). Two forms of the recurrence formula for the circumscribed polygon are:

<msub>

</mrow>

</msub>

<mfrac>

<msqrt>

</msqrt>

</mfrac>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\textstyle t_{0}={\frac {1}{\sqrt {3}}}}</annotation>

</semantics>

</math><img src="./a9a8d68112b0a6f1a41b69213bdc887a4f8bae7d.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -1.838ex; width:8.019ex; height:4.176ex;" alt="{\textstyle t_{0}={\frac {1}{\sqrt {3}}}}" loading="lazy"></li>

<li>First form: <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\textstyle t_{i+1}={\frac {{\sqrt {t_{i}^{2}+1}}-1}{t_{i}}}}">

<msub>

</mrow>

</msub>

<mfrac>

<mrow>

<msqrt>

</mrow>

</mrow>

</msubsup>

</msqrt>

</mrow>

</mrow>

<msub>

</mrow>

</msub>

</mfrac>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\textstyle t_{i+1}={\frac {{\sqrt {t_{i}^{2}+1}}-1}{t_{i}}}}</annotation>

</semantics>

</math><img src="./7b8e42962e073104445e5fb184833851ef480065.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -1.505ex; width:14.944ex; height:5.843ex;" alt="{\textstyle t_{i+1}={\frac {{\sqrt {t_{i}^{2}+1}}-1}{t_{i}}}}" loading="lazy"></li>

<li>Second form: <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\textstyle t_{i+1}={\frac {t_{i}}{{\sqrt {t_{i}^{2}+1}}+1}}}">

<msub>

</mrow>

</msub>

<mfrac>

<msub>

</mrow>

</msub>

<mrow>

<msqrt>

</mrow>

</mrow>

</msubsup>

</msqrt>

</mrow>

</mrow>

</mfrac>

</mrow>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\textstyle t_{i+1}={\frac {t_{i}}{{\sqrt {t_{i}^{2}+1}}+1}}}</annotation>

</semantics>

</math><img src="./171333bbf207a0be4a87530af334d931d755e13d.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -3.171ex; width:14.944ex; height:5.843ex;" alt="{\textstyle t_{i+1}={\frac {t_{i}}{{\sqrt {t_{i}^{2}+1}}+1}}}" loading="lazy"></li>

<msup>

</mrow>

</msup>

<msub>

</mrow>

</msub>

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle \pi \sim 6\times 2^{i}\times t_{i}}</annotation>

</semantics>

</math><img src="./c19a2c9acbd64b8495a8e1aa2fe018103c500d61.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:14.875ex; height:3.009ex;" alt="{\displaystyle \pi \sim 6\times 2^{i}\times t_{i}}" loading="lazy">, converging as <math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle i\rightarrow \infty }">

</mstyle>

</mrow>

<annotation encoding="application/x-tex">{\displaystyle i\rightarrow \infty }</annotation>

</semantics>

</math><img src="./9912b98084f3a5fa477e9fbe25597750d2c375fd.svg" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:6.74ex; height:2.176ex;" alt="{\displaystyle i\rightarrow \infty }" loading="lazy"></li></ul>

Here is a computation using IEEE "double" (a significand with 53 bits of precision) arithmetic:

<pre> i 6 × 2i × ti, first form 6 × 2i × ti, second form

---------------------------------------------------------

0 3.4641016151377543863 3.4641016151377543863

1 3.2153903091734710173 3.2153903091734723496

2 3.1596599420974940120 3.1596599420975006733

3 3.1460862151314012979 3.1460862151314352708

4 3.1427145996453136334 3.1427145996453689225

5 3.1418730499801259536 3.1418730499798241950

6 3.1416627470548084133 3.1416627470568494473

7 3.1416101765997805905 3.1416101766046906629

8 3.1415970343230776862 3.1415970343215275928

9 3.1415937488171150615 3.1415937487713536668

10 3.1415929278733740748 3.1415929273850979885

11 3.1415927256228504127 3.1415927220386148377

12 3.1415926717412858693 3.1415926707019992125

13 3.1415926189011456060 3.1415926578678454728

14 3.1415926717412858693 3.1415926546593073709

15 3.1415919358822321783 3.1415926538571730119

16 3.1415926717412858693 3.1415926536566394222

17 3.1415810075796233302 3.1415926536065061913

18 3.1415926717412858693 3.1415926535939728836

19 3.1414061547378810956 3.1415926535908393901

20 3.1405434924008406305 3.1415926535900560168

21 3.1400068646912273617 3.1415926535898608396

22 3.1349453756585929919 3.1415926535898122118

23 3.1400068646912273617 3.1415926535897995552

24 3.2245152435345525443 3.1415926535897968907

25 3.1415926535897962246

26 3.1415926535897962246

27 3.1415926535897962246

28 3.1415926535897962246

The true value is 3.14159265358979323846264338327...

</pre>

While the two forms of the recurrence formula are clearly mathematically equivalent,<a href="#cite_note-NB_7-79">[nb 14]</a> the first subtracts 1 from a number extremely close to 1, leading to an increasingly problematic loss of <a href="Significant_digit" class="mw-redirect" title="Significant digit">significant digits</a>. As the recurrence is applied repeatedly, the accuracy improves at first, but then it deteriorates. It never gets better than about 8 digits, even though 53-bit arithmetic should be capable of about 16 digits of precision. When the second form of the recurrence is used, the value converges to 15 digits of precision.

<div class="mw-heading mw-heading3"><h3 id=""Fast_math"_optimization">"Fast math" optimization</h3></div>

The aforementioned lack of <a href="Associative_property" title="Associative property">associativity</a> of floating-point operations in general means that <a href="Compilers" class="mw-redirect" title="Compilers">compilers</a> cannot as effectively reorder arithmetic expressions as they could with integer and fixed-point arithmetic, presenting a roadblock in optimizations such as <a href="Common_subexpression_elimination" title="Common subexpression elimination">common subexpression elimination</a> and auto-<a href="Single_instruction%2C_multiple_data" title="Single instruction, multiple data">vectorization</a>.<a href="#cite_note-Vectorizers-80">[66]</a> The "fast math" option on many compilers (ICC, GCC, Clang, MSVC...) turns on reassociation along with unsafe assumptions such as a lack of NaN and infinite numbers in IEEE 754. Some compilers also offer more granular options to only turn on reassociation. In either case, the programmer is exposed to many of the precision pitfalls mentioned above for the portion of the program using "fast" math.<a href="#cite_note-FPM-81">[67]</a>

In some compilers (GCC and Clang), turning on "fast" math may cause the program to <a href="Subnormal_number#Disabling_subnormal_floats_at_the_code_level" title="Subnormal number">disable subnormal floats</a> at startup, affecting the floating-point behavior of not only the generated code, but also any program using such code as a <a href="Library_(computing)" title="Library (computing)">library</a>.<a href="#cite_note-harmful-82">[68]</a>

In most <a href="Fortran" title="Fortran">Fortran</a> compilers, as allowed by the ISO/IEC 1539-1:2004 Fortran standard, reassociation is the default, with breakage largely prevented by the "protect parens" setting (also on by default). This setting stops the compiler from reassociating beyond the boundaries of parentheses.<a href="#cite_note-Gen-83">[69]</a> <a href="Intel_Fortran_Compiler" title="Intel Fortran Compiler">Intel Fortran Compiler</a> is a notable outlier.<a href="#cite_note-zheevd-84">[70]</a>

A common problem in "fast" math is that subexpressions may not be optimized identically from place to place, leading to unexpected differences. One interpretation of the issue is that "fast" math as implemented currently has a poorly defined semantics. One attempt at formalizing "fast" math optimizations is seen in Icing, a verified compiler.<a href="#cite_note-Becker-Darulova-Myreen-Tatlock_2019-85">[71]</a>

/* start https://en.wikipedia.org/ */

.mw-parser-output .div-col{margin-top:0.3em;column-width:30em}.mw-parser-output .div-col-small{font-size:90%}.mw-parser-output .div-col-rules{column-rule:1px solid #aaa}.mw-parser-output .div-col dl,.mw-parser-output .div-col ol,.mw-parser-output .div-col ul{margin-top:0}.mw-parser-output .div-col li,.mw-parser-output .div-col dd{page-break-inside:avoid;break-inside:avoid-column}

/* end https://en.wikipedia.org/ */

</style><div class="div-col" style="column-width: 20em;">

<ul><li><a href="Arbitrary-precision_arithmetic" title="Arbitrary-precision arithmetic">Arbitrary-precision arithmetic</a></li>

<li><a href="C99#IEEE_754_floating-point_support" title="C99">C99</a> for code examples demonstrating access and use of IEEE 754 features.</li>

<li><a href="Computable_number" title="Computable number">Computable number</a></li>

<li><a href="Coprocessor" title="Coprocessor">Coprocessor</a></li>

<li><a href="Decimal_floating_point" title="Decimal floating point">Decimal floating point</a></li>

<li><a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double-precision floating-point format</a></li>

<li><a href="Experimental_mathematics" title="Experimental mathematics">Experimental mathematics</a> – utilizes high precision floating-point computations</li>

<li><a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">Fixed-point arithmetic</a></li>

<li><a href="Floating-point_error_mitigation" title="Floating-point error mitigation">Floating-point error mitigation</a></li>

<li><a href="FLOPS" class="mw-redirect" title="FLOPS">FLOPS</a></li>

<li><a href="Gal's_accurate_tables" title="Gal's accurate tables">Gal's accurate tables</a></li>

<li><a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half-precision floating-point format</a></li>

<li><a href="IEEE_754" title="IEEE 754">IEEE 754</a> – Standard for Binary Floating-Point Arithmetic</li>

<li><a href="IBM_hexadecimal_floating-point" title="IBM hexadecimal floating-point">IBM Floating Point Architecture</a></li>

<li><a href="Kahan_summation_algorithm" title="Kahan summation algorithm">Kahan summation algorithm</a></li>

<li><a href="Microsoft_Binary_Format" title="Microsoft Binary Format">Microsoft Binary Format</a> (MBF)</li>

<li><a href="Minifloat" title="Minifloat">Minifloat</a></li>

<li><a href="Q_(number_format)" title="Q (number format)">Q (number format)</a> for constant resolution</li>

<li><a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">Quadruple-precision floating-point format</a> (including double-double)</li>

<li><a href="Significant_figures" title="Significant figures">Significant figures</a></li>

<li><a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single-precision floating-point format</a></li>

<li><a href="Standard_Apple_Numerics_Environment" title="Standard Apple Numerics Environment">Standard Apple Numerics Environment (SANE)</a></li></ul>

</div>

<div class="mw-heading mw-heading2"><h2 id="Notes">Notes</h2></div>

/* start https://en.wikipedia.org/ */

.mw-parser-output .reflist{margin-bottom:0.5em;list-style-type:decimal}@media screen{.mw-parser-output .reflist{font-size:90%}}.mw-parser-output .reflist .references{font-size:100%;margin-bottom:0;list-style-type:inherit}.mw-parser-output .reflist-columns-2{column-width:30em}.mw-parser-output .reflist-columns-3{column-width:25em}.mw-parser-output .reflist-columns{margin-top:0.3em}.mw-parser-output .reflist-columns ol{margin-top:0}.mw-parser-output .reflist-columns li{page-break-inside:avoid;break-inside:avoid-column}.mw-parser-output .reflist-upper-alpha{list-style-type:upper-alpha}.mw-parser-output .reflist-upper-roman{list-style-type:upper-roman}.mw-parser-output .reflist-lower-alpha{list-style-type:lower-alpha}.mw-parser-output .reflist-lower-greek{list-style-type:lower-greek}.mw-parser-output .reflist-lower-roman{list-style-type:lower-roman}

/* end https://en.wikipedia.org/ */

</style><div class="reflist">

<li id="cite_note-NB_Significand-4"><a href="#cite_ref-NB_Significand_4-0">^</a> The <a href="Significand" title="Significand">significand</a> of a floating-point number is also called mantissa by some authors—not to be confused with the <a href="Mantissa_(logarithm)" class="mw-redirect" title="Mantissa (logarithm)">mantissa</a> of a <a href="Logarithm" title="Logarithm">logarithm</a>. Somewhat vague, terms such as coefficient or argument are also used by some. The usage of the term fraction by some authors is potentially misleading as well. The term characteristic (as used e.g. by <a href="Control_Data_Corporation" title="Control Data Corporation">CDC</a>) is ambiguous, as it was historically also used to specify some form of <a href="#NB-Exponent">exponent</a> of floating-point numbers.

</li>

<li id="cite_note-NB_Exponent-5"><a href="#cite_ref-NB_Exponent_5-0">^</a> The <a href="Exponent" class="mw-redirect" title="Exponent">exponent</a> of a floating-point number is sometimes also referred to as scale. The term characteristic (for <a href="Biased_exponent" class="mw-redirect" title="Biased exponent">biased exponent</a>, exponent bias, or excess n representation) is ambiguous, as it was historically also used to specify the <a href="#NB-Significand">significand</a> of floating-point numbers.

</li>

<li id="cite_note-NB_9-8"><a href="#cite_ref-NB_9_8-0">^</a> <a href="Hexadecimal_floating-point" class="mw-redirect" title="Hexadecimal floating-point">Hexadecimal (base-16) floating-point</a> arithmetic is used in the <a href="IBM_System_360" class="mw-redirect" title="IBM System 360">IBM System 360</a> (1964) and <a href="IBM_System_370" class="mw-redirect" title="IBM System 370">370</a> (1970) as well as various newer IBM machines, in the <a href="RCA_Spectra_70" title="RCA Spectra 70">RCA Spectra 70</a> (1964), the Siemens 4004 (1965), 7.700 (1974), 7.800, 7.500 (1977) series mainframes and successors, the Unidata 7.000 series mainframes, the <a href="Manchester_MU5" class="mw-redirect" title="Manchester MU5">Manchester MU5</a> (1972), the <a href="Heterogeneous_Element_Processor" title="Heterogeneous Element Processor">HEP</a> (1982) computers, and in 360/370-compatible mainframe families made by Fujitsu, Amdahl and Hitachi. It is also used in the <a href="Illinois_ILLIAC_III" class="mw-redirect" title="Illinois ILLIAC III">Illinois ILLIAC III</a> (1966), <a href="Data_General_Eclipse_S/200" class="mw-redirect" title="Data General Eclipse S/200">Data General Eclipse S/200</a> (ca. 1974), <a href="Gould_Powernode_9080" class="mw-redirect" title="Gould Powernode 9080">Gould Powernode 9080</a> (1980s), <a href="Interdata_8/32" class="mw-redirect" title="Interdata 8/32">Interdata 8/32</a> (1970s), the SEL Systems 85 and 86 as well as the <a href="SDS_Sigma_5" class="mw-redirect" title="SDS Sigma 5">SDS Sigma 5</a> (1967), <a href="SDS_Sigma_7" class="mw-redirect" title="SDS Sigma 7">7</a> (1966) and <a href="Xerox_Sigma_9" title="Xerox Sigma 9">Xerox Sigma 9</a> (1970).

</li>

<li id="cite_note-NB_8-10"><a href="#cite_ref-NB_8_10-0">^</a> Octal (base-8) floating-point arithmetic is used in the <a href="Ferranti_Atlas" class="mw-redirect" title="Ferranti Atlas">Ferranti Atlas</a> (1962), <a href="Burroughs_B5500" class="mw-redirect" title="Burroughs B5500">Burroughs B5500</a> (1964), <a href="Burroughs_B5700" class="mw-redirect" title="Burroughs B5700">Burroughs B5700</a> (1971), <a href="Burroughs_B6700" class="mw-redirect" title="Burroughs B6700">Burroughs B6700</a> (1971) and <a href="Burroughs_B7700" class="mw-redirect" title="Burroughs B7700">Burroughs B7700</a> (1972) computers.

</li>

<li id="cite_note-NB_11-12"><a href="#cite_ref-NB_11_12-0">^</a> Quaternary (base-4) floating-point arithmetic is used in the <a href="Illinois_ILLIAC_II" class="mw-redirect" title="Illinois ILLIAC II">Illinois ILLIAC II</a> (1962) computer. It is also used in the Digital Field System DFS IV and V high-resolution site survey systems.

</li>

<li id="cite_note-NB_12-13"><a href="#cite_ref-NB_12_13-0">^</a> Base-256 floating-point arithmetic is used in the <a href="Rice_Institute_R1" class="mw-redirect" title="Rice Institute R1">Rice Institute R1</a> computer (since 1958).

</li>

<li id="cite_note-NB_10-15"><a href="#cite_ref-NB_10_15-0">^</a> Base-65536 floating-point arithmetic is used in the <a href="MANIAC_II" title="MANIAC II">MANIAC II</a> (1956) computer.

</li>

<li id="cite_note-NB_1-42"><a href="#cite_ref-NB_1_42-0">^</a> Computer hardware does not necessarily compute the exact value; it simply has to produce the equivalent rounded result as though it had computed the infinitely precise result.

</li>

<li id="cite_note-NB_2-54"><a href="#cite_ref-NB_2_54-0">^</a> The enormous complexity of modern <a href="Division_algorithm" title="Division algorithm">division algorithms</a> once led to a famous error. An early version of the <a href="Intel_Pentium" class="mw-redirect" title="Intel Pentium">Intel Pentium</a> chip was shipped with a <a href="FDIV" class="mw-redirect" title="FDIV">division instruction</a> that, on rare occasions, gave slightly incorrect results. Many computers had been shipped before the error was discovered. Until the defective computers were replaced, patched versions of compilers were developed that could avoid the failing cases. See <a href="Pentium_FDIV_bug" title="Pentium FDIV bug">Pentium FDIV bug</a>.

</li>

<li id="cite_note-NB_3-57"><a href="#cite_ref-NB_3_57-0">^</a> But an attempted computation of cos(π) yields −1 exactly. Since the derivative is nearly zero near π, the effect of the inaccuracy in the argument is far smaller than the spacing of the floating-point numbers around −1, and the rounded result is exact.

</li>

<li id="cite_note-NB_4-68"><a href="#cite_ref-NB_4_68-0">^</a> <a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">William Kahan</a> notes: "Except in extremely uncommon situations, extra-precise arithmetic generally attenuates risks due to roundoff at far less cost than the price of a competent error-analyst."

</li>

<li id="cite_note-NB_5-69"><a href="#cite_ref-NB_5_69-0">^</a> The <a href="Taylor_expansion" class="mw-redirect" title="Taylor expansion">Taylor expansion</a> of this function demonstrates that it is well-conditioned near 1: A(x) = 1 − (x−1)/2 + (x−1)^2/12 − (x−1)^4/720 + (x−1)^6/30240 − (x−1)^8/1209600 + ... for |x−1| < π.

</li>

<li id="cite_note-NB_6-71"><a href="#cite_ref-NB_6_71-0">^</a> If <a href="Long_double" title="Long double">long double</a> is <a href="IEEE_quad_precision" class="mw-redirect" title="IEEE quad precision">IEEE quad precision</a> then full double precision is retained; if long double is <a href="IEEE_double_extended_precision" class="mw-redirect" title="IEEE double extended precision">IEEE double extended precision</a> then additional, but not full precision is retained.

</li>

<li id="cite_note-NB_7-79"><a href="#cite_ref-NB_7_79-0">^</a> The equivalence of the two forms can be verified algebraically by noting that the <a href="Denominator" class="mw-redirect" title="Denominator">denominator</a> of the fraction in the second form is the <a href="Conjugate_(algebra)" class="mw-redirect" title="Conjugate (algebra)">conjugate</a> of the <a href="Numerator" class="mw-redirect" title="Numerator">numerator</a> of the first. By multiplying the top and bottom of the first expression by this conjugate, one obtains the second expression.

</li>

</ol></div></div>

<div class="mw-heading mw-heading2"><h2 id="References">References</h2></div>

/* start https://en.wikipedia.org/ */

.mw-parser-output cite.citation{font-style:inherit;word-wrap:break-word}.mw-parser-output .citation q{quotes:"\"""\"""'""'"}.mw-parser-output .citation:target{background-color:rgba(0,127,255,0.133)}.mw-parser-output .id-lock-free.id-lock-free a{background:url("./mw/Lock-green.svg")right 0.1em center/9px no-repeat}.mw-parser-output .id-lock-limited.id-lock-limited a,.mw-parser-output .id-lock-registration.id-lock-registration a{background:url("./mw/Lock-gray-alt-2.svg")right 0.1em center/9px no-repeat}.mw-parser-output .id-lock-subscription.id-lock-subscription a{background:url("./mw/Lock-red-alt-2.svg")right 0.1em center/9px no-repeat}.mw-parser-output .cs1-ws-icon a{background:url("./mw/Wikisource-logo.svg")right 0.1em center/12px no-repeat}body:not(.skin-timeless):not(.skin-minerva) .mw-parser-output .id-lock-free a,body:not(.skin-timeless):not(.skin-minerva) .mw-parser-output .id-lock-limited a,body:not(.skin-timeless):not(.skin-minerva) .mw-parser-output .id-lock-registration a,body:not(.skin-timeless):not(.skin-minerva) .mw-parser-output .id-lock-subscription a,body:not(.skin-timeless):not(.skin-minerva) .mw-parser-output .cs1-ws-icon a{background-size:contain;padding:0 1em 0 0}.mw-parser-output .cs1-code{color:inherit;background:inherit;border:none;padding:inherit}.mw-parser-output .cs1-hidden-error{display:none;color:var(--color-error,#d33)}.mw-parser-output .cs1-visible-error{color:var(--color-error,#d33)}.mw-parser-output .cs1-maint{display:none;color:#085;margin-left:0.3em}.mw-parser-output .cs1-kern-left{padding-left:0.2em}.mw-parser-output .cs1-kern-right{padding-right:0.2em}.mw-parser-output .citation .mw-selflink{font-weight:inherit}@media screen{.mw-parser-output .cs1-format{font-size:95%}html.skin-theme-clientpref-night .mw-parser-output .cs1-maint{color:#18911f}}@media screen and (prefers-color-scheme:dark){html.skin-theme-clientpref-os .mw-parser-output .cs1-maint{color:#18911f}}

/* end https://en.wikipedia.org/ */

</style><cite id="muller_et_al_pg_16" class="citation book cs1">Muller, Jean-Michel; Brisebarre, Nicolas; de Dinechin, Florent; Jeannerod, Claude-Pierre; Lefèvre, Vincent; Melquiond, Guillaume; <a href="Nathalie_Revol" title="Nathalie Revol">Revol, Nathalie</a>; Stehlé, Damien; Torres, Serge (2010). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=baFvrIOPvncC&pg=PA16">Handbook of Floating-Point Arithmetic</a> (1st ed.). <a href="Birkh%C3%A4user" title="Birkhäuser">Birkhäuser</a>. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2F978-0-8176-4705-6">10.1007/978-0-8176-4705-6</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-8176-4704-9</bdi>. <a href="LCCN_(identifier)" class="mw-redirect" title="LCCN (identifier)">LCCN</a> <a rel="nofollow" class="external text" href="https://lccn.loc.gov/2009939668">2009939668</a>.</cite>

</li>

<li id="cite_note-sterbenz1974fpcomp-2">^ <a href="#cite_ref-sterbenz1974fpcomp_2-0">a</a> <a href="#cite_ref-sterbenz1974fpcomp_2-1">b</a> <cite id="CITEREFSterbenz1974" class="citation book cs1">Sterbenz, Pat H. (1974). <a rel="nofollow" class="external text" href="https://archive.org/details/SterbenzFloatingPointComputation/mode/2up">Floating-Point Computation</a>. Englewood Cliffs, NJ, United States: Prentice-Hall. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>0-13-322495-3</bdi>.</cite>

</li>

<li id="cite_note-Smith_1997-3"><a href="#cite_ref-Smith_1997_3-0">^</a> <cite id="CITEREFSmith1997" class="citation book cs1">Smith, Steven W. (1997). <a rel="nofollow" class="external text" href="http://www.dspguide.com/ch28/4.htm">"Chapter 28, Fixed versus Floating Point"</a>. The Scientist and Engineer's Guide to Digital Signal Processing. California Technical Pub. p. 514. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-9660176-3-2</bdi>. Retrieved 2012-12-31.</cite>

</li>

<li id="cite_note-Zehendner_2008-6">^ <a href="#cite_ref-Zehendner_2008_6-0">a</a> <a href="#cite_ref-Zehendner_2008_6-1">b</a> <cite id="CITEREFZehendner2008" class="citation web cs1 cs1-prop-foreign-lang-source">Zehendner, Eberhard (Summer 2008). <a rel="nofollow" class="external text" href="https://users.fmi.uni-jena.de/~nez/rechnerarithmetik_5/folien/Rechnerarithmetik.2008.05.handout.pdf">"Rechnerarithmetik: Fest- und Gleitkommasysteme"</a> (PDF) (Lecture script) (in German). <a href="Friedrich-Schiller-Universit%C3%A4t_Jena" class="mw-redirect" title="Friedrich-Schiller-Universität Jena">Friedrich-Schiller-Universität Jena</a>. p. 2. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180807062449/https://users.fmi.uni-jena.de/~nez/rechnerarithmetik_5/folien/Rechnerarithmetik.2008.05.handout.pdf">Archived</a> (PDF) from the original on 2018-08-07. Retrieved 2018-08-07.</cite> <a rel="nofollow" class="external autonumber" href="https://web.archive.org/web/20180806175620/https://users.fmi.uni-jena.de/~nez/rechnerarithmetik_5/folien/Rechnerarithmetik.2008.komplett.pdf">[1]</a> (NB. This reference incorrectly gives the MANIAC II's floating point base as 256, whereas it actually is 65536.)

</li>

<li id="cite_note-Beebe_2017-7">^ <a href="#cite_ref-Beebe_2017_7-0">a</a> <a href="#cite_ref-Beebe_2017_7-1">b</a> <a href="#cite_ref-Beebe_2017_7-2">c</a> <a href="#cite_ref-Beebe_2017_7-3">d</a> <cite id="CITEREFBeebe2017" class="citation book cs1">Beebe, Nelson H. F. (2017-08-22). "Chapter H. Historical floating-point architectures". The Mathematical-Function Computation Handbook - Programming Using the MathCW Portable Software Library (1st ed.). Salt Lake City, UT, USA: <a href="Springer_International_Publishing_AG" class="mw-redirect" title="Springer International Publishing AG">Springer International Publishing AG</a>. p. 948. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2F978-3-319-64110-2">10.1007/978-3-319-64110-2</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-3-319-64109-6</bdi>. <a href="LCCN_(identifier)" class="mw-redirect" title="LCCN (identifier)">LCCN</a> <a rel="nofollow" class="external text" href="https://lccn.loc.gov/2017947446">2017947446</a>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:30244721">30244721</a>.</cite>

</li>

<li id="cite_note-Savard_2018-9"><a href="#cite_ref-Savard_2018_9-0">^</a> <cite id="CITEREFSavard2018" class="citation cs2">Savard, John J. G. (2018) [2007], <a rel="nofollow" class="external text" href="http://www.quadibloc.com/comp/cp020302.htm">"The Decimal Floating-Point Standard"</a>, quadibloc, <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180703002322/http://www.quadibloc.com/comp/cp020302.htm">archived</a> from the original on 2018-07-03, retrieved 2018-07-16</cite>

</li>

<li id="cite_note-Parkinson_2000-11"><a href="#cite_ref-Parkinson_2000_11-0">^</a> <cite id="CITEREFParkinson2000" class="citation book cs1">Parkinson, Roger (2000-12-07). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=Ocip5vpLD4wC&pg=PA24">"Chapter 2 - High resolution digital site survey systems - Chapter 2.1 - Digital field recording systems"</a>. High Resolution Site Surveys (1st ed.). <a href="CRC_Press" title="CRC Press">CRC Press</a>. p. 24. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-20318604-6</bdi>. Retrieved 2019-08-18. <q>[…] Systems such as the [Digital Field System] DFS IV and DFS V were quaternary floating-point systems and used gain steps of 12 dB. […]</q></cite> (256 pages)

</li>

<li id="cite_note-Lazarus_1956-14"><a href="#cite_ref-Lazarus_1956_14-0">^</a> <cite id="CITEREFLazarus1957" class="citation web cs1">Lazarus, Roger B. (1957-01-30) [1956-10-01]. <a rel="nofollow" class="external text" href="http://bitsavers.org/pdf/lanl/LA-2083_MANIAC_II_Oct56.pdf">"MANIAC II"</a> (PDF). Los Alamos, NM, USA: Los Alamos Scientific Laboratory of the University of California. p. 14. LA-2083. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180807200914/http://bitsavers.org/pdf/lanl/LA-2083_MANIAC_II_Oct56.pdf">Archived</a> (PDF) from the original on 2018-08-07. Retrieved 2018-08-07. <q>[…] the Maniac's floating base, which is 216 = 65,536. […] The Maniac's large base permits a considerable increase in the speed of floating point arithmetic. Although such a large base implies the possibility of as many as 15 lead zeros, the large word size of 48 bits guarantees adequate significance. […]</q></cite>

</li>

<li id="cite_note-16"><a href="#cite_ref-16">^</a> Torres Quevedo, Leonardo. <a rel="nofollow" class="external text" href="https://quickclick.es/rop/pdf/publico/1914/1914_tomoI_2043_01.pdf">Automática: Complemento de la Teoría de las Máquinas, (pdf)</a>, pp. 575–583, Revista de Obras Públicas, 19 November 1914.

</li>

<li id="cite_note-17"><a href="#cite_ref-17">^</a> Ronald T. Kneusel. <a rel="nofollow" class="external text" href="https://books.google.com/books?id=eq4ZDgAAQBAJ&dq=leonardo+torres+quevedo++electromechanical+machine+essays&pg=PA84">Numbers and Computers</a>, Springer, pp. 84–85, 2017. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-3319505084</bdi>

</li>

<li id="cite_note-FOOTNOTERandell19826,_11–13-18"><a href="#cite_ref-FOOTNOTERandell19826,_11–13_18-0">^</a> <a href="#CITEREFRandell1982">Randell 1982</a>, pp. 6, 11–13.

</li>

<li id="cite_note-19"><a href="#cite_ref-19">^</a> Randell, Brian. <a rel="nofollow" class="external text" href="https://dl.acm.org/doi/pdf/10.5555/1074100.1074334">Digital Computers, History of Origins, (pdf)</a>, p. 545, Digital Computers: Origins, Encyclopedia of Computer Science, January 2003.

</li>

<li id="cite_note-Rojas_1997-20"><a href="#cite_ref-Rojas_1997_20-0">^</a> <cite id="CITEREFRojas1997" class="citation journal cs1"><a href="Ra%C3%BAl_Rojas" title="Raúl Rojas">Rojas, Raúl</a> (April–June 1997). <a rel="nofollow" class="external text" href="http://ed-thelen.org/comp-hist/Zuse_Z1_and_Z3.pdf">"Konrad Zuse's Legacy: The Architecture of the Z1 and Z3"</a> (PDF). <a href="IEEE_Annals_of_the_History_of_Computing" title="IEEE Annals of the History of Computing">IEEE Annals of the History of Computing</a>. 19 (2): 5–16. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2F85.586067">10.1109/85.586067</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20220703082408/http://ed-thelen.org/comp-hist/Zuse_Z1_and_Z3.pdf">Archived</a> (PDF) from the original on 2022-07-03. Retrieved 2022-07-03.</cite> (12 pages)

</li>

<li id="cite_note-Rojas_2014-21"><a href="#cite_ref-Rojas_2014_21-0">^</a> <cite id="CITEREFRojas2014" class="citation arxiv cs1"><a href="Ra%C3%BAl_Rojas" title="Raúl Rojas">Rojas, Raúl</a> (2014-06-07). "The Z1: Architecture and Algorithms of Konrad Zuse's First Computer". <a href="ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<a rel="nofollow" class="external text" href="https://arxiv.org/abs/1406.1886">1406.1886</a> [<a rel="nofollow" class="external text" href="https://arxiv.org/archive/cs.AR">cs.AR</a>].</cite>

</li>

<li id="cite_note-Kahan_1997_JVNL-22">^ <a href="#cite_ref-Kahan_1997_JVNL_22-0">a</a> <a href="#cite_ref-Kahan_1997_JVNL_22-1">b</a> <cite id="CITEREFKahan1997" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (1997-07-15). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/SIAMjvnl.pdf">"The Baleful Effect of Computer Languages and Benchmarks upon Applied Mathematics, Physics and Chemistry. John von Neumann Lecture"</a> (PDF). p. 3. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20080905103125/http://www.cs.berkeley.edu/~wkahan/SIAMjvnl.pdf">Archived</a> (PDF) from the original on 2008-09-05.</cite>

</li>

<li id="cite_note-Randell_1982_2-23"><a href="#cite_ref-Randell_1982_2_23-0">^</a> <cite id="CITEREFRandell1982" class="citation book cs1"><a href="Brian_Randell" title="Brian Randell">Randell, Brian</a>, ed. (1982) [1973]. The Origins of Digital Computers: Selected Papers (3rd ed.). Berlin; New York: <a href="Springer-Verlag" class="mw-redirect" title="Springer-Verlag">Springer-Verlag</a>. p. 244. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-3-540-11319-5</bdi>.</cite>

</li>

<li id="cite_note-Severance_1998-24"><a href="#cite_ref-Severance_1998_24-0">^</a> <cite id="CITEREFSeverance1998" class="citation web cs1"><a href="Charles_Severance_(computer_scientist)" title="Charles Severance (computer scientist)">Severance, Charles</a> (1998-02-20). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/ieee754status/754story.html">"An Interview with the Old Man of Floating-Point"</a>.</cite>

</li>

<li id="cite_note-Goldberg_1991-25">^ <a href="#cite_ref-Goldberg_1991_25-0">a</a> <a href="#cite_ref-Goldberg_1991_25-1">b</a> <a href="#cite_ref-Goldberg_1991_25-2">c</a> <a href="#cite_ref-Goldberg_1991_25-3">d</a> <cite id="CITEREFGoldberg1991" class="citation journal cs1">Goldberg, David (March 1991). <a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F103162.103163">"What Every Computer Scientist Should Know About Floating-Point Arithmetic"</a>. <a href="ACM_Computing_Surveys" title="ACM Computing Surveys">ACM Computing Surveys</a>. 23 (1): 5–48. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F103162.103163">10.1145/103162.103163</a>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:222008826">222008826</a>.</cite> (With the addendum "Differences Among IEEE 754 Implementations": <a rel="nofollow" class="external autonumber" href="https://web.archive.org/web/20171011072644/http://www.cse.msu.edu/~cse320/Documents/FloatingPoint.pdf">[2]</a>, <a rel="nofollow" class="external autonumber" href="https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html">[3]</a>)

</li>

<li id="cite_note-C99-26"><a href="#cite_ref-C99_26-0">^</a> <cite class="citation book cs1">ISO/IEC 9899:1999 - Programming languages - C. Iso.org. §F.2, note 307. <q>"Extended" is IEC 60559's double-extended data format. Extended refers to both the common 80-bit and quadruple 128-bit IEC 60559 formats.</q></cite>

</li>

<li id="cite_note-MSVC-27"><a href="#cite_ref-MSVC_27-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://learn.microsoft.com/en-us/cpp/build/ieee-floating-point-representation">"IEEE Floating-Point Representation"</a>. 2021-08-03.</cite>

</li>

<li id="cite_note-GCC-28"><a href="#cite_ref-GCC_28-0">^</a> <a rel="nofollow" class="external text" href="https://gcc.gnu.org/onlinedocs/gcc/i386-and-x86-64-Options.html">Using the GNU Compiler Collection, i386 and x86-64 Options</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150116065447/http://gcc.gnu.org/onlinedocs/gcc/i386-and-x86-64-Options.html">Archived</a> 2015-01-16 at the <a href="Wayback_Machine" title="Wayback Machine">Wayback Machine</a>.

</li>

<li id="cite_note-float_128-29"><a href="#cite_ref-float_128_29-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://stackoverflow.com/questions/13516476">"long double (GCC specific) and __float128"</a>. StackOverflow.</cite>

</li>

<li id="cite_note-ARM_2013_AArch64-30"><a href="#cite_ref-ARM_2013_AArch64_30-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0055b/IHI0055B_aapcs64.pdf">"Procedure Call Standard for the ARM 64-bit Architecture (AArch64)"</a> (PDF). 2013-05-22. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20130731181404/http://infocenter.arm.com/help/topic/com.arm.doc.ihi0055b/IHI0055B_aapcs64.pdf">Archived</a> (PDF) from the original on 2013-07-31. Retrieved 2019-09-22.</cite>

</li>

<li id="cite_note-ARM_2013_Compiler-31"><a href="#cite_ref-ARM_2013_Compiler_31-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://infocenter.arm.com/help/topic/com.arm.doc.dui0491i/DUI0491I_arm_compiler_reference.pdf">"ARM Compiler toolchain Compiler Reference, Version 5.03"</a> (PDF). 2013. Section 6.3 Basic data types. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150627210618/http://infocenter.arm.com/help/topic/com.arm.doc.dui0491i/DUI0491I_arm_compiler_reference.pdf">Archived</a> (PDF) from the original on 2015-06-27. Retrieved 2019-11-08.</cite>

</li>

<li id="cite_note-Kahan_2004-32"><a href="#cite_ref-Kahan_2004_32-0">^</a> <cite id="CITEREFKahan2004" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2004-11-20). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/Qdrtcs.pdf">"On the Cost of Floating-Point Computation Without Extra-Precise Arithmetic"</a> (PDF). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20060525111157/http://www.cs.berkeley.edu/~wkahan/Qdrtcs.pdf">Archived</a> (PDF) from the original on 2006-05-25. Retrieved 2012-02-19.</cite>

</li>

<li id="cite_note-OpenEXR-33"><a href="#cite_ref-OpenEXR_33-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://web.archive.org/web/20130508221152/http://www.openexr.com/about.html">"openEXR"</a>. openEXR. Archived from <a rel="nofollow" class="external text" href="http://www.openexr.com/about.html">the original</a> on 2013-05-08. Retrieved 2012-04-25. <q>Since the IEEE-754 floating-point specification does not define a 16-bit format, ILM created the "half" format. Half values have 1 sign bit, 5 exponent bits, and 10 mantissa bits.</q></cite>

</li>

<li id="cite_note-OpenEXR-half-34"><a href="#cite_ref-OpenEXR-half_34-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://openexr.com/en/latest/TechnicalIntroduction.html#the-half-data-type">"Technical Introduction to OpenEXR – The half Data Type"</a>. openEXR. Retrieved 2024-04-16.</cite>

</li>

<li id="cite_note-IEEE-754_Analysis-35"><a href="#cite_ref-IEEE-754_Analysis_35-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://christophervickery.com/IEEE-754/">"IEEE-754 Analysis"</a>. Retrieved 2024-08-29.</cite>

</li>

<li id="cite_note-Borland_1994_MBF-36">^ <a href="#cite_ref-Borland_1994_MBF_36-0">a</a> <a href="#cite_ref-Borland_1994_MBF_36-1">b</a> <cite id="CITEREFBorland_staff1998" class="citation web cs1">Borland staff (1998-07-02) [1994-03-10]. <a rel="nofollow" class="external text" href="https://community.embarcadero.com/index.php/article/technical-articles/162-programming/14799-converting-between-microsoft-binary-and-ieee-forma">"Converting between Microsoft Binary and IEEE formats"</a>. Technical Information Database (TI1431C.txt). <a href="Embarcadero_USA" class="mw-redirect" title="Embarcadero USA">Embarcadero USA</a> / <a href="Inprise" class="mw-redirect" title="Inprise">Inprise</a> (originally: <a href="Borland" title="Borland">Borland</a>). ID 1400. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190220230417/https://community.embarcadero.com/index.php/article/technical-articles/162-programming/14799-converting-between-microsoft-binary-and-ieee-forma">Archived</a> from the original on 2019-02-20. Retrieved 2016-05-30. <q>[…] _fmsbintoieee(float *src4, float *dest4) […] MS Binary Format […] byte order => m3 | m2 | m1 | exponent […] m1 is <a href="Most_significant_byte" class="mw-redirect" title="Most significant byte">most significant byte</a> => sbbb|bbbb […] m3 is the <a href="Least_significant_byte" class="mw-redirect" title="Least significant byte">least significant byte</a> […] m = mantissa byte […] s = sign bit […] b = bit […] MBF is bias 128 and IEEE is bias 127. […] MBF places the <a href="Decimal_point" class="mw-redirect" title="Decimal point">decimal point</a> before the <a href="Assumed_bit" class="mw-redirect" title="Assumed bit">assumed bit</a>, while IEEE places the decimal point after the assumed bit. […] ieee_exp = msbin[3] - 2; /* actually, msbin[3]-1-128+127 */ […] _dmsbintoieee(double *src8, double *dest8) […] MS Binary Format […] byte order => m7 | m6 | m5 | m4 | m3 | m2 | m1 | exponent […] m1 is most significant byte => smmm|mmmm […] m7 is the least significant byte […] MBF is bias 128 and IEEE is bias 1023. […] MBF places the decimal point before the assumed bit, while IEEE places the decimal point after the assumed bit. […] ieee_exp = msbin[7] - 128 - 1 + 1023; […]</q></cite>

</li>

<li id="cite_note-Steil_2008_6502-37">^ <a href="#cite_ref-Steil_2008_6502_37-0">a</a> <a href="#cite_ref-Steil_2008_6502_37-1">b</a> <cite id="CITEREFSteil2008" class="citation web cs1">Steil, Michael (2008-10-20). <a rel="nofollow" class="external text" href="http://www.pagetable.com/?p=46">"Create your own Version of Microsoft BASIC for 6502"</a>. pagetable.com. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160530092603/http://www.pagetable.com/?p=46">Archived</a> from the original on 2016-05-30. Retrieved 2016-05-30.</cite>

</li>

<li id="cite_note-Microsoft_2006_KB35826-38"><a href="#cite_ref-Microsoft_2006_KB35826_38-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.betaarchive.com/wiki/index.php/Microsoft_KB_Archive/35826#IEEE_vs._Microsoft_Binary_Format.3B_Rounding_Issues_.28Complete.29">"IEEE vs. Microsoft Binary Format; Rounding Issues (Complete)"</a>. Microsoft Support. <a href="Microsoft" title="Microsoft">Microsoft</a>. 2006-11-21. Article ID KB35826, Q35826. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200828130651/https://www.betaarchive.com/wiki/index.php/Microsoft_KB_Archive/35826">Archived</a> from the original on 2020-08-28. Retrieved 2010-02-24.</cite>

</li>

<li id="cite_note-Kharya_2020-39">^ <a href="#cite_ref-Kharya_2020_39-0">a</a> <a href="#cite_ref-Kharya_2020_39-1">b</a> <cite id="CITEREFKharya2020" class="citation web cs1">Kharya, Paresh (2020-05-14). <a rel="nofollow" class="external text" href="https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/">"TensorFloat-32 in the A100 GPU Accelerates AI Training, HPC up to 20x"</a>. Retrieved 2020-05-16.</cite>

</li>

<li id="cite_note-NVIDIA_Hopper-40"><a href="#cite_ref-NVIDIA_Hopper_40-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://developer.nvidia.com/blog/nvidia-hopper-architecture-in-depth/">"NVIDIA Hopper Architecture In-Depth"</a>. 2022-03-22.</cite>

</li>

<li id="cite_note-Micikevicius_2022-41"><a href="#cite_ref-Micikevicius_2022_41-0">^</a> <cite id="CITEREFMicikeviciusStosicBurgessCornea2022" class="citation arxiv cs1">Micikevicius, Paulius; Stosic, Dusan; Burgess, Neil; Cornea, Marius; Dubey, Pradeep; Grisenthwaite, Richard; Ha, Sangwon; Heinecke, Alexander; Judd, Patrick; Kamalu, John; Mellempudi, Naveen; Oberman, Stuart; Shoeybi, Mohammad; Siu, Michael; Wu, Hao (2022-09-12). "FP8 Formats for Deep Learning". <a href="ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<a rel="nofollow" class="external text" href="https://arxiv.org/abs/2209.05433">2209.05433</a> [<a rel="nofollow" class="external text" href="https://arxiv.org/archive/cs.LG">cs.LG</a>].</cite>

</li>

<li id="cite_note-Kahan_2006_Mindless-43"><a href="#cite_ref-Kahan_2006_Mindless_43-0">^</a> <cite id="CITEREFKahan2006" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2006-01-11). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/Mindless.pdf">"How Futile are Mindless Assessments of Roundoff in Floating-Point Computation?"</a> (PDF). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20041221020332/http://www.cs.berkeley.edu/~wkahan/Mindless.pdf">Archived</a> (PDF) from the original on 2004-12-21.</cite>

</li>

<li id="cite_note-Gay_1990-44">^ <a href="#cite_ref-Gay_1990_44-0">a</a> <a href="#cite_ref-Gay_1990_44-1">b</a> <cite id="CITEREFGay1990" class="citation techreport cs1">Gay, David M. (1990). Correctly Rounded Binary-Decimal and Decimal-Binary Conversions (Technical report). NUMERICAL ANALYSIS MANUSCRIPT 90-10, AT&T BELL LABORATORIES. <a href="CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.31.4049">10.1.1.31.4049</a>.</cite> (<a rel="nofollow" class="external text" href="http://www.netlib.org/fp/dtoa.c">dtoa.c in netlab</a>)

</li>

<li id="cite_note-Loitsch_2010-45"><a href="#cite_ref-Loitsch_2010_45-0">^</a> <cite id="CITEREFLoitsch2010" class="citation conference cs1">Loitsch, Florian (2010). <a rel="nofollow" class="external text" href="https://www.cs.tufts.edu/~nr/cs257/archive/florian-loitsch/printf.pdf">"Printing floating-point numbers quickly and accurately with integers"</a> (PDF). Proceedings of the 31st ACM SIGPLAN Conference on Programming Language Design and Implementation. PLDI '10: ACM SIGPLAN Conference on Programming Language Design and Implementation. pp. 233–243. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F1806596.1806623">10.1145/1806596.1806623</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-1-45030019-3</bdi>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:910409">910409</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20140729005717/http://www.cs.tufts.edu/~nr/cs257/archive/florian-loitsch/printf.pdf">Archived</a> (PDF) from the original on 2014-07-29.</cite>

</li>

<li id="cite_note-mazong-46"><a href="#cite_ref-mazong_46-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://github.com/dotnet/coreclr/pull/14646">"Added Grisu3 algorithm support for double.ToString(). by mazong1123 · Pull Request #14646 · dotnet/coreclr"</a>. GitHub.</cite>

</li>

<li id="cite_note-Adams_2018-47"><a href="#cite_ref-Adams_2018_47-0">^</a> <cite id="CITEREFAdams2018" class="citation journal cs1">Adams, Ulf (2018-12-02). <a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F3296979.3192369">"Ryū: fast float-to-string conversion"</a>. ACM SIGPLAN Notices. 53 (4): 270–282. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F3296979.3192369">10.1145/3296979.3192369</a>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:218472153">218472153</a>.</cite>

</li>

<li id="cite_note-Giulietti-48"><a href="#cite_ref-Giulietti_48-0">^</a> <cite id="CITEREFGiulietti" class="citation web cs1">Giulietti, Rafaello. <a rel="nofollow" class="external text" href="https://drive.google.com/file/d/1IEeATSVnEE6TkrHlCYNY2GjaraBjOT4f">"The Schubfach way to render doubles"</a>.</cite>

</li>

<li id="cite_note-abolz-49"><a href="#cite_ref-abolz_49-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://github.com/abolz/Drachennest">"abolz/Drachennest"</a>. <a href="GitHub" title="GitHub">GitHub</a>. 2022-11-10.</cite>

</li>

<li id="cite_note-double_conversion_2020-50"><a href="#cite_ref-double_conversion_2020_50-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://github.com/google/double-conversion">"google/double-conversion"</a>. <a href="GitHub" title="GitHub">GitHub</a>. 2020-09-21.</cite>

</li>

<li id="cite_note-Lemire_2021-51"><a href="#cite_ref-Lemire_2021_51-0">^</a> <cite id="CITEREFLemire2021" class="citation journal cs1">Lemire, Daniel (2021-03-22). "Number parsing at a gigabyte per second". Software: Practice and Experience. 51 (8): 1700–1727. <a href="ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<a rel="nofollow" class="external text" href="https://arxiv.org/abs/2101.11408">2101.11408</a>. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1002%2Fspe.2984">10.1002/spe.2984</a>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:231718830">231718830</a>.</cite>

</li>

<li id="cite_note-Patterson-Hennessy_2014-52"><a href="#cite_ref-Patterson-Hennessy_2014_52-0">^</a> <cite id="CITEREFPattersonHennessy2014" class="citation book cs1">Patterson, David A.; Hennessy, John L. (2014). Computer Organization and Design, The Hardware/Software Interface. The Morgan Kaufmann series in computer architecture and design (5th ed.). Waltham, Massachusetts, USA: Elsevier. p. 793. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-9-86605267-5</bdi>.</cite>

</li>

/* start https://en.wikipedia.org/ */

.mw-parser-output .citation{word-wrap:break-word}.mw-parser-output .citation:target{background-color:rgba(0,127,255,0.133)}

/* end https://en.wikipedia.org/ */

</style><a rel="nofollow" class="external text" href="https://worldwide.espacenet.com/textdoc?DB=EPODOC&IDX=US3037701A">US patent 3037701A</a>, Huberto M Sierra, "Floating decimal point arithmetic control means for calculator", issued 1962-06-05<span class="Z3988" title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Apatent&rft.number=3037701A&rft.cc=US&rft.title=Floating+decimal+point+arithmetic+control+means+for+calculator&rft.inventor=Huberto+M+Sierra&rft.date=1962-06-05">

</li>

<li id="cite_note-Kahan_1997_Status-55">^ <a href="#cite_ref-Kahan_1997_Status_55-0">a</a> <a href="#cite_ref-Kahan_1997_Status_55-1">b</a> <cite id="CITEREFKahan1997" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (1997-10-01). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/ieee754status/IEEE754.PDF">"Lecture Notes on the Status of IEEE Standard 754 for Binary Floating-Point Arithmetic"</a> (PDF). p. 9. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20020622093102/http://www.cs.berkeley.edu/~wkahan/ieee754status/IEEE754.PDF">Archived</a> (PDF) from the original on 2002-06-22.</cite>

</li>

<li id="cite_note-Intel-56"><a href="#cite_ref-Intel_56-0">^</a> <cite class="citation book cs1"><a rel="nofollow" class="external text" href="http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html">"D.3.2.1"</a>. Intel 64 and IA-32 Architectures Software Developers' Manuals. Vol. 1.</cite>

</li>

<li id="cite_note-Harris-58"><a href="#cite_ref-Harris_58-0">^</a> <cite id="CITEREFHarris2010" class="citation journal cs1">Harris, Richard (October 2010). <a rel="nofollow" class="external text" href="http://accu.org/index.php/journals/1702">"You're Going To Have To Think!"</a>. <a href="Overload_(magazine)" title="Overload (magazine)">Overload</a> (99): 5–10. <a href="ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://search.worldcat.org/issn/1354-3172">1354-3172</a>. Retrieved 2011-09-24. <q>Far more worrying is cancellation error which can yield catastrophic loss of precision.</q></cite> <a rel="nofollow" class="external autonumber" href="http://accu.org/var/uploads/journals/overload99.pdf">[4]</a>

</li>

<li id="cite_note-Barker-59"><a href="#cite_ref-Barker_59-0">^</a> Christopher Barker: <a rel="nofollow" class="external text" href="https://www.python.org/dev/peps/pep-0485/">PEP 485 -- A Function for testing approximate equality</a>

</li>

<li id="cite_note-GAO_report_IMTEC_92-26-60">^ <a href="#cite_ref-GAO_report_IMTEC_92-26_60-0">a</a> <a href="#cite_ref-GAO_report_IMTEC_92-26_60-1">b</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://www.gao.gov/products/IMTEC-92-26">"Patriot missile defense, Software problem led to system failure at Dharhan, Saudi Arabia"</a>. <a href="US_Government_Accounting_Office" class="mw-redirect" title="US Government Accounting Office">US Government Accounting Office</a>. GAO report IMTEC 92-26.</cite>

</li>

<li id="cite_note-Skeel-61">^ <a href="#cite_ref-Skeel_61-0">a</a> <a href="#cite_ref-Skeel_61-1">b</a> <cite id="CITEREFSkeel1992" class="citation cs2">Skeel, Robert (July 1992), <a rel="nofollow" class="external text" href="https://www-users.cse.umn.edu/~arnold/disasters/Patriot-dharan-skeel-siam.pdf">"Roundoff Error and the Patriot Missile"</a> (PDF), SIAM News, 25 (4): 11, retrieved 2024-11-15</cite>

</li>

<li id="cite_note-RalstonReilly2003-62"><a href="#cite_ref-RalstonReilly2003_62-0">^</a> <cite id="CITEREFWilkinson2003" class="citation book cs1"><a href="James_Hardy_Wilkinson" class="mw-redirect" title="James Hardy Wilkinson">Wilkinson, James Hardy</a> (2003-09-08). "Error Analysis". In Ralston, Anthony; Reilly, Edwin D.; Hemmendinger, David (eds.). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=OLRwQgAACAAJ">Encyclopedia of Computer Science</a>. <a href="Wiley_(publisher)" title="Wiley (publisher)">Wiley</a>. pp. 669–674. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-470-86412-8</bdi>. Retrieved 2013-05-14.</cite>

</li>

<li id="cite_note-Einarsson_2005-63"><a href="#cite_ref-Einarsson_2005_63-0">^</a> <cite id="CITEREFEinarsson2005" class="citation book cs1">Einarsson, Bo (2005). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=sh4orx_qB_QC&pg=PA50">Accuracy and reliability in scientific computing</a>. <a href="Society_for_Industrial_and_Applied_Mathematics" title="Society for Industrial and Applied Mathematics">Society for Industrial and Applied Mathematics</a> (SIAM). pp. 50–. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-89871-815-7</bdi>. Retrieved 2013-05-14.</cite>

</li>

<li id="cite_note-Higham_2002-64">^ <a href="#cite_ref-Higham_2002_64-0">a</a> <a href="#cite_ref-Higham_2002_64-1">b</a> <a href="#cite_ref-Higham_2002_64-2">c</a> <a href="#cite_ref-Higham_2002_64-3">d</a> <cite id="CITEREFHigham2002" class="citation book cs1"><a href="Nicholas_Higham" title="Nicholas Higham">Higham, Nicholas John</a> (2002). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=epilvM5MMxwC">Accuracy and Stability of Numerical Algorithms</a> (2nd ed.). <a href="Society_for_Industrial_and_Applied_Mathematics" title="Society for Industrial and Applied Mathematics">Society for Industrial and Applied Mathematics</a> (SIAM). pp. 27–28, 110–123, 493. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-89871-521-7</bdi>. 0-89871-355-2.</cite>

</li>

<li id="cite_note-OliveiraStewart_2006-65"><a href="#cite_ref-OliveiraStewart_2006_65-0">^</a> <cite id="CITEREFOliveiraStewart2006" class="citation book cs1">Oliveira, Suely; Stewart, David E. (2006-09-07). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=E6a8oZOS8noC&pg=PA10">Writing Scientific Software: A Guide to Good Style</a>. <a href="Cambridge_University_Press" title="Cambridge University Press">Cambridge University Press</a>. pp. 10–. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-1-139-45862-7</bdi>.</cite>

</li>

<li id="cite_note-Kahan_2005_ARITH17-66">^ <a href="#cite_ref-Kahan_2005_ARITH17_66-0">a</a> <a href="#cite_ref-Kahan_2005_ARITH17_66-1">b</a> <cite id="CITEREFKahan2005" class="citation conference cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2005-07-15). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/ARITH_17.pdf">Floating-Point Arithmetic Besieged by "Business Decisions"</a> (PDF). IEEE-sponsored <a href="ARITH_17" class="mw-redirect" title="ARITH 17">ARITH 17</a>, Symposium on Computer Arithmetic (Keynote Address). pp. 6, 18. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20060317103619/http://www.cs.berkeley.edu/~wkahan/ARITH_17.pdf">Archived</a> (PDF) from the original on 2006-03-17. Retrieved 2013-05-23.</cite> (NB. Kahan estimates that the incidence of excessively inaccurate results near singularities is reduced by a factor of approx. 1/2000 using the 11 extra bits of precision of <a href="Extended_precision" title="Extended precision">double extended</a>.)

</li>

<li id="cite_note-Kahan_2011_Debug-67"><a href="#cite_ref-Kahan_2011_Debug_67-0">^</a> <cite id="CITEREFKahan2011" class="citation conference cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2011-08-03). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/Boulder.pdf">Desperately Needed Remedies for the Undebuggability of Large Floating-Point Computations in Science and Engineering</a> (PDF). IFIP/SIAM/NIST Working Conference on Uncertainty Quantification in Scientific Computing, Boulder, CO. p. 33. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20130620140729/http://www.eecs.berkeley.edu/~wkahan/Boulder.pdf">Archived</a> (PDF) from the original on 2013-06-20.</cite>

</li>

<li id="cite_note-Kahan_2001_JavaHurt-70"><a href="#cite_ref-Kahan_2001_JavaHurt_70-0">^</a> <cite id="CITEREFKahanDarcy2001" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a>; Darcy, Joseph (2001) [1998-03-01]. <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/JAVAhurt.pdf">"How Java's floating-point hurts everyone everywhere"</a> (PDF). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20000816043653/http://www.cs.berkeley.edu/~wkahan/JAVAhurt.pdf">Archived</a> (PDF) from the original on 2000-08-16. Retrieved 2003-09-05.</cite>

</li>

<li id="cite_note-Kahan_2000_Marketing-72">^ <a href="#cite_ref-Kahan_2000_Marketing_72-0">a</a> <a href="#cite_ref-Kahan_2000_Marketing_72-1">b</a> <a href="#cite_ref-Kahan_2000_Marketing_72-2">c</a> <a href="#cite_ref-Kahan_2000_Marketing_72-3">d</a> <cite id="CITEREFKahan2000" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2000-08-27). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/MktgMath.pdf">"Marketing versus Mathematics"</a> (PDF). pp. 15, 35, 47. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20030815150333/http://www.cs.berkeley.edu/~wkahan/MktgMath.pdf">Archived</a> (PDF) from the original on 2003-08-15.</cite>

</li>

<li id="cite_note-Kahan_1981_WhyIEEE-73"><a href="#cite_ref-Kahan_1981_WhyIEEE_73-0">^</a> <cite id="CITEREFKahan1981" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (1981-02-12). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/ieee754status/why-ieee.pdf">"Why do we need a floating-point arithmetic standard?"</a> (PDF). p. 26. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20041204070746/http://www.cs.berkeley.edu/~wkahan/ieee754status/why-ieee.pdf">Archived</a> (PDF) from the original on 2004-12-04.</cite>

</li>

<li id="cite_note-Kahan_2001_LN-74"><a href="#cite_ref-Kahan_2001_LN_74-0">^</a> <cite id="CITEREFKahan2001" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a> (2001-06-04). Bindel, David (ed.). <a rel="nofollow" class="external text" href="http://www.cims.nyu.edu/~dbindel/class/cs279/notes-06-04.pdf">"Lecture notes of System Support for Scientific Computation"</a> (PDF). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20130517181356/http://www.cims.nyu.edu/~dbindel/class/cs279/notes-06-04.pdf">Archived</a> (PDF) from the original on 2013-05-17.</cite>

</li>

<li id="cite_note-Speleotrove_2012-75"><a href="#cite_ref-Speleotrove_2012_75-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://speleotrove.com/decimal/">"General Decimal Arithmetic"</a>. Speleotrove.com. Retrieved 2012-04-25.</cite>

</li>

<li id="cite_note-Christiansen_Perl-76"><a href="#cite_ref-Christiansen_Perl_76-0">^</a> <cite id="CITEREFChristiansenTorkington2006" class="citation web cs1">Christiansen, Tom; Torkington, Nathan; et al. (2006). <a rel="nofollow" class="external text" href="https://perldoc.perl.org/5.8.8/perlfaq4#Why-is-int()-broken?">"perlfaq4 / Why is int() broken?"</a>. perldoc.perl.org. Retrieved 2011-01-11.</cite>

</li>

<li id="cite_note-Shewchuk-77"><a href="#cite_ref-Shewchuk_77-0">^</a> <cite id="CITEREFShewchuk1997" class="citation journal cs1">Shewchuk, Jonathan Richard (1997). <a rel="nofollow" class="external text" href="https://doi.org/10.1007%2FPL00009321">"Adaptive Precision Floating-Point Arithmetic and Fast Robust Geometric Predicates"</a>. <a href="Discrete_%26_Computational_Geometry" title="Discrete & Computational Geometry">Discrete & Computational Geometry</a>. 18 (3): 305–363. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2FPL00009321">10.1007/PL00009321</a>.</cite>

</li>

<li id="cite_note-Kahan_1997_Cantilever-78"><a href="#cite_ref-Kahan_1997_Cantilever_78-0">^</a> <cite id="CITEREFKahanIvory1997" class="citation web cs1"><a href="William_Morton_Kahan" class="mw-redirect" title="William Morton Kahan">Kahan, William Morton</a>; Ivory, Melody Y. (1997-07-03). <a rel="nofollow" class="external text" href="https://people.eecs.berkeley.edu/~wkahan/Cantilever.pdf">"Roundoff Degrades an Idealized Cantilever"</a> (PDF). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20031205191038/http://www.cs.berkeley.edu/~wkahan/Cantilever.pdf">Archived</a> (PDF) from the original on 2003-12-05.</cite>

</li>

<li id="cite_note-Vectorizers-80"><a href="#cite_ref-Vectorizers_80-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://llvm.org/docs/Vectorizers.html">"Auto-Vectorization in LLVM"</a>. LLVM 13 documentation. <q>We support floating point reduction operations when -ffast-math is used.</q></cite>

</li>

<li id="cite_note-FPM-81"><a href="#cite_ref-FPM_81-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://gcc.gnu.org/wiki/FloatingPointMath">"FloatingPointMath"</a>. GCC Wiki.</cite>

</li>

<li id="cite_note-harmful-82"><a href="#cite_ref-harmful_82-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55522">"55522 – -funsafe-math-optimizations is unexpectedly harmful, especially w/ -shared"</a>. gcc.gnu.org.</cite>

</li>

<li id="cite_note-Gen-83"><a href="#cite_ref-Gen_83-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html">"Code Gen Options (The GNU Fortran Compiler)"</a>. gcc.gnu.org.</cite>

</li>

<li id="cite_note-zheevd-84"><a href="#cite_ref-zheevd_84-0">^</a> <cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://github.com/Reference-LAPACK/lapack/issues/43">"Bug in zheevd · Issue #43 · Reference-LAPACK/lapack"</a>. GitHub.</cite>

</li>

<li id="cite_note-Becker-Darulova-Myreen-Tatlock_2019-85"><a href="#cite_ref-Becker-Darulova-Myreen-Tatlock_2019_85-0">^</a> <cite id="CITEREFBeckerDarulovaMyreenTatlock2019" class="citation conference cs1">Becker, Heiko; Darulova, Eva; Myreen, Magnus O.; Tatlock, Zachary (2019). Icing: Supporting Fast-Math Style Optimizations in a Verified Compiler. CAV 2019: Computer Aided Verification. Vol. 11562. pp. 155–173. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2F978-3-030-25543-5_10">10.1007/978-3-030-25543-5_10</a>.</cite>

</li>

</ol></div>

<div class="mw-heading mw-heading2"><h2 id="Further_reading">Further reading</h2></div>

<ul><li><cite id="CITEREFWilkinson1963" class="citation book cs1"><a href="James_Hardy_Wilkinson" class="mw-redirect" title="James Hardy Wilkinson">Wilkinson, James Hardy</a> (1963). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=yFogU9Ot-qsC">Rounding Errors in Algebraic Processes</a> (1st ed.). Englewood Cliffs, New Jersey, USA: <a href="Prentice-Hall%2C_Inc." class="mw-redirect" title="Prentice-Hall, Inc.">Prentice-Hall, Inc.</a> <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>9780486679990</bdi>. <a href="MR_(identifier)" class="mw-redirect" title="MR (identifier)">MR</a> <a rel="nofollow" class="external text" href="https://mathscinet.ams.org/mathscinet-getitem?mr=0161456">0161456</a>.</cite> <code class="cs1-code">{{cite book}}</code>: ISBN / Date incompatibility (help) (NB. Classic influential treatises on floating-point arithmetic.)</li>

<li><cite id="CITEREFWilkinson1965" class="citation book cs1"><a href="James_Hardy_Wilkinson" class="mw-redirect" title="James Hardy Wilkinson">Wilkinson, James Hardy</a> (1965). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=N98IAQAAIAAJ&q=editions:ISBN0198534183">The Algebraic Eigenvalue Problem</a>. Monographs on Numerical Analysis (1st ed.). <a href="Oxford_University_Press" title="Oxford University Press">Oxford University Press</a> / <a href="Clarendon_Press" class="mw-redirect" title="Clarendon Press">Clarendon Press</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>9780198534037</bdi>. Retrieved 2016-02-11.</cite></li>

<li><cite id="CITEREFSterbenz1974" class="citation book cs1">Sterbenz, Pat H. (1974). Floating-Point Computation. Prentice-Hall Series in Automatic Computation (1st ed.). Englewood Cliffs, New Jersey, USA: <a href="Prentice_Hall" title="Prentice Hall">Prentice Hall</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-13-322495-5</bdi>.</cite></li>

<li><cite id="CITEREFGolubvan_Loan1986" class="citation book cs1">Golub, Gene F.; van Loan, Charles F. (1986). Matrix Computations (3rd ed.). <a href="Johns_Hopkins_University_Press" title="Johns Hopkins University Press">Johns Hopkins University Press</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-8018-5413-2</bdi>.</cite></li>

<li><cite id="CITEREFPressTeukolskyVetterlingFlannery2007" class="citation book cs1"><a href="William_Henry_Press" class="mw-redirect" title="William Henry Press">Press, William Henry</a>; <a href="Saul_A._Teukolsky" class="mw-redirect" title="Saul A. Teukolsky">Teukolsky, Saul A.</a>; <a href="William_T._Vetterling" class="mw-redirect" title="William T. Vetterling">Vetterling, William T.</a>; <a href="Brian_P._Flannery" title="Brian P. Flannery">Flannery, Brian P.</a> (2007) [1986]. <a href="Numerical_Recipes" title="Numerical Recipes">Numerical Recipes - The Art of Scientific Computing</a> (3rd ed.). <a href="Cambridge_University_Press" title="Cambridge University Press">Cambridge University Press</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-521-88407-5</bdi>.</cite> (NB. Edition with source code CD-ROM.)</li>

<li><cite id="CITEREFKnuth1997" class="citation book cs1"><a href="Donald_Ervin_Knuth" class="mw-redirect" title="Donald Ervin Knuth">Knuth, Donald Ervin</a> (1997). "Section 4.2: Floating-Point Arithmetic". <a href="The_Art_of_Computer_Programming" title="The Art of Computer Programming">The Art of Computer Programming</a>, Vol. 2: Seminumerical Algorithms (3rd ed.). <a href="Addison-Wesley" title="Addison-Wesley">Addison-Wesley</a>. pp. 214–264. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-201-89684-8</bdi>.</cite></li>

<li><cite id="CITEREFBlaauwBrooks,_Jr.1997" class="citation book cs1"><a href="Gerrit_Anne_Blaauw" class="mw-redirect" title="Gerrit Anne Blaauw">Blaauw, Gerrit Anne</a>; <a href="Frederick_Phillips_Brooks%2C_Jr." class="mw-redirect" title="Frederick Phillips Brooks, Jr.">Brooks, Jr., Frederick Phillips</a> (1997). Computer Architecture: Concepts and Evolution (1st ed.). <a href="Addison-Wesley" title="Addison-Wesley">Addison-Wesley</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>0-201-10557-8</bdi>.</cite> (1213 pages) (NB. This is a single-volume edition. This work was also available in a two-volume version.)</li>

<li><cite id="CITEREFKornerupMatula2010" class="citation book cs1">Kornerup, Peter; Matula, David W. (2010). Finite Precision Number Systems and Arithmetic. <a href="Cambridge_University_Press" title="Cambridge University Press">Cambridge University Press</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-0-521-76135-2</bdi>.</cite></li>

<li><cite id="CITEREFSavard2018" class="citation cs2">Savard, John J. G. (2018) [2005], <a rel="nofollow" class="external text" href="http://www.quadibloc.com/comp/cp0201.htm">"Floating-Point Formats"</a>, quadibloc, <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180703001709/http://www.quadibloc.com/comp/cp0201.htm">archived</a> from the original on 2018-07-03, retrieved 2018-07-16</cite></li>

<li><cite id="CITEREFMullerBruniede_DinechinJeannerod2018" class="citation book cs1">Muller, Jean-Michel; Brunie, Nicolas; de Dinechin, Florent; Jeannerod, Claude-Pierre; Joldes, Mioara; Lefèvre, Vincent; Melquiond, Guillaume; <a href="Nathalie_Revol" title="Nathalie Revol">Revol, Nathalie</a>; Torres, Serge (2018) [2010]. <a rel="nofollow" class="external text" href="https://books.google.com/books?id=h3ZZDwAAQBAJ">Handbook of Floating-Point Arithmetic</a> (2nd ed.). <a href="Birkh%C3%A4user" title="Birkhäuser">Birkhäuser</a>. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2F978-3-319-76526-6">10.1007/978-3-319-76526-6</a>. <a href="ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <bdi>978-3-319-76525-9</bdi>. <a href="LCCN_(identifier)" class="mw-redirect" title="LCCN (identifier)">LCCN</a> <a rel="nofollow" class="external text" href="https://lccn.loc.gov/2018935254">2018935254</a>.</cite></li></ul>

<div class="mw-heading mw-heading2"><h2 id="External_links">External links</h2></div>

<ul><li><cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://www.mrob.com/pub/math/floatformats.html">"Survey of Floating-Point Formats"</a>.</cite> (NB. This page gives a very brief summary of floating-point formats that have been used over the years.)</li>

<li><cite id="CITEREFMonniaux2008" class="citation journal cs1">Monniaux, David (May 2008). <a rel="nofollow" class="external text" href="https://hal.science/hal-00128124/en/">"The pitfalls of verifying floating-point computations"</a>. ACM Transactions on Programming Languages and Systems. 30 (3). <a href="Association_for_Computing_Machinery" title="Association for Computing Machinery">Association for Computing Machinery</a> (ACM) Transactions on programming languages and systems (TOPLAS): 1–41. <a href="ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<a rel="nofollow" class="external text" href="https://arxiv.org/abs/cs/0701192">cs/0701192</a>. <a href="Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F1353445.1353446">10.1145/1353445.1353446</a>. <a href="S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:218578808">218578808</a>.</cite> (NB. A compendium of non-intuitive behaviors of floating point on popular architectures, with implications for program verification and testing.)</li>

<li><a rel="nofollow" class="external text" href="https://opencores.org/">OpenCores</a>. (NB. This website contains open source floating-point IP cores for the implementation of floating-point operators in FPGA or ASIC devices. The project double_fpu contains verilog source code of a double-precision floating-point unit. The project fpuvhdl contains vhdl source code of a single-precision floating-point unit.)</li>

<li><cite id="CITEREFFleegal2004" class="citation web cs1">Fleegal, Eric (2004). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170706020455/http://msdn.microsoft.com/en-us/library/aa289157(v=vs.71).aspx">"Microsoft Visual C++ Floating-Point Optimization"</a>. <a href="Microsoft_Developer_Network" title="Microsoft Developer Network">Microsoft Developer Network</a>. Archived from <a rel="nofollow" class="external text" href="http://msdn.microsoft.com/en-us/library/aa289157(v=vs.71).aspx">the original</a> on 2017-07-06.</cite></li></ul>

/* start https://en.wikipedia.org/ */

.mw-parser-output .navbox{box-sizing:border-box;border:1px solid #a2a9b1;width:100%;clear:both;font-size:88%;text-align:center;padding:1px;margin:1em auto 0}.mw-parser-output .navbox .navbox{margin-top:0}.mw-parser-output .navbox+.navbox,.mw-parser-output .navbox+.navbox-styles+.navbox{margin-top:-1px}.mw-parser-output .navbox-inner,.mw-parser-output .navbox-subgroup{width:100%}.mw-parser-output .navbox-group,.mw-parser-output .navbox-title,.mw-parser-output .navbox-abovebelow{padding:0.25em 1em;line-height:1.5em;text-align:center}.mw-parser-output .navbox-group{white-space:nowrap;text-align:right}.mw-parser-output .navbox,.mw-parser-output .navbox-subgroup{background-color:#fdfdfd}.mw-parser-output .navbox-list{line-height:1.5em;border-color:#fdfdfd}.mw-parser-output .navbox-list-with-group{text-align:left;border-left-width:2px;border-left-style:solid}.mw-parser-output tr+tr>.navbox-abovebelow,.mw-parser-output tr+tr>.navbox-group,.mw-parser-output tr+tr>.navbox-image,.mw-parser-output tr+tr>.navbox-list{border-top:2px solid #fdfdfd}.mw-parser-output .navbox-title{background-color:#ccf}.mw-parser-output .navbox-abovebelow,.mw-parser-output .navbox-group,.mw-parser-output .navbox-subgroup .navbox-title{background-color:#ddf}.mw-parser-output .navbox-subgroup .navbox-group,.mw-parser-output .navbox-subgroup .navbox-abovebelow{background-color:#e6e6ff}.mw-parser-output .navbox-even{background-color:#f7f7f7}.mw-parser-output .navbox-odd{background-color:transparent}.mw-parser-output .navbox .hlist td dl,.mw-parser-output .navbox .hlist td ol,.mw-parser-output .navbox .hlist td ul,.mw-parser-output .navbox td.hlist dl,.mw-parser-output .navbox td.hlist ol,.mw-parser-output .navbox td.hlist ul{padding:0.125em 0}.mw-parser-output .navbox .navbar{display:block;font-size:100%}.mw-parser-output .navbox-title .navbar{float:left;text-align:left;margin-right:0.5em}body.skin--responsive .mw-parser-output .navbox-image img{max-width:none!important}@media print{body.ns-0 .mw-parser-output .navbox{display:none!important}}

/* end https://en.wikipedia.org/ */

</style></div><div role="navigation" class="navbox" aria-labelledby="Data_types177" style="padding:3px"><table class="nowraplinks hlist mw-collapsible autocollapse navbox-inner" style="border-spacing:0;background:transparent;color:inherit"><tbody><tr><th scope="col" class="navbox-title" colspan="2"><div id="Data_types177" style="font-size:114%;margin:0 4em"><a href="Data_type" title="Data type">Data types</a></div></th></tr><tr><th scope="row" class="navbox-group" style="width:1%"><a href="Units_of_information" title="Units of information">Uninterpreted</a></th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">

<li><a href="Ternary_numeral_system#Tryte" title="Ternary numeral system">Tryte</a></li>

<li><a href="Bit_array" title="Bit array">Bit array</a></li></ul>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Numeric</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Arbitrary-precision_arithmetic" title="Arbitrary-precision arithmetic">Arbitrary-precision or bignum</a></li>

<li><a href="Complex_data_type" title="Complex data type">Complex</a></li>

<li><a href="Decimal_data_type" title="Decimal data type">Decimal</a></li>

<li><a href="Fixed-point_arithmetic" title="Fixed-point arithmetic">Fixed point</a></li>

<li><a href="Block_floating_point" title="Block floating point">Block floating point</a></li>

<li>

<ul><li>Reduced precision

<ul><li><a href="Minifloat" title="Minifloat">Minifloat</a></li>

<li><a href="Half-precision_floating-point_format" title="Half-precision floating-point format">Half precision</a></li>

<li><a href="Bfloat16_floating-point_format" title="Bfloat16 floating-point format">bfloat16</a></li></ul></li>

<li><a href="Single-precision_floating-point_format" title="Single-precision floating-point format">Single precision</a></li>

<li><a href="Double-precision_floating-point_format" title="Double-precision floating-point format">Double precision</a></li>

<li><a href="Quadruple-precision_floating-point_format" title="Quadruple-precision floating-point format">Quadruple precision</a></li>

<li><a href="Octuple-precision_floating-point_format" title="Octuple-precision floating-point format">Octuple precision</a></li>

<li><a href="Extended_precision" title="Extended precision">Extended precision</a>

<ul><li><a href="Long_double" title="Long double">Long double</a></li></ul></li></ul></li>

<li><a href="Integer_(computer_science)" title="Integer (computer science)">Integer</a>

<ul><li><a href="Signedness" title="Signedness">signedness</a></li></ul></li>

<li><a href="Interval_arithmetic#Implementations" title="Interval arithmetic">Interval</a></li>

<li><a href="Rational_data_type" title="Rational data type">Rational</a></li></ul>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%"><a href="Pointer_(computer_programming)" title="Pointer (computer programming)">Pointer</a></th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Memory_address" title="Memory address">Address</a>

<ul><li><a href="Physical_address" title="Physical address">physical</a></li>

<li><a href="Virtual_address_space" title="Virtual address space">virtual</a></li></ul></li>

<li><a href="Reference_(computer_science)" title="Reference (computer science)">Reference</a></li></ul>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%"><a href="Plain_text" title="Plain text">Text</a></th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Character_(computing)" title="Character (computing)">Character</a></li>

<li><a href="String_(computer_science)" title="String (computer science)">String</a>

<ul><li><a href="Null-terminated_string" title="Null-terminated string">null-terminated</a></li></ul></li></ul>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%"><a href="Composite_data_type" title="Composite data type">Composite</a></th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Algebraic_data_type" title="Algebraic data type">Algebraic data type</a>

<ul><li><a href="Generalized_algebraic_data_type" title="Generalized algebraic data type">generalized</a></li></ul></li>

<li><a href="Array_(data_type)" title="Array (data type)">Array</a></li>

<li><a href="Associative_array" title="Associative array">Associative array</a></li>

<li><a href="Class_(computer_programming)" title="Class (computer programming)">Class</a></li>

<li><a href="Dependent_type" title="Dependent type">Dependent</a></li>

<li><a href="Intuitionistic_type_theory#Equality_type" title="Intuitionistic type theory">Equality</a></li>

<li><a href="Inductive_type" title="Inductive type">Inductive</a></li>

<li><a href="Intersection_type" title="Intersection type">Intersection</a></li>

<li><a href="Object_(computer_science)" title="Object (computer science)">Object</a>

<ul><li><a href="Metaobject" title="Metaobject">metaobject</a></li></ul></li>

<li><a href="Option_type" title="Option type">Option type</a></li>

<li><a href="Product_type" title="Product type">Product</a></li>

<li><a href="Record_(computer_science)" title="Record (computer science)">Record or Struct</a></li>

<li><a href="Refinement_type" title="Refinement type">Refinement</a></li>

<li><a href="Union_type" title="Union type">Union</a>

<ul><li><a href="Tagged_union" title="Tagged union">tagged</a></li></ul></li></ul>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Other</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Boolean_data_type" title="Boolean data type">Boolean</a></li>

<li><a href="Bottom_type" title="Bottom type">Bottom type</a></li>

<li><a href="Container_(abstract_data_type)" title="Container (abstract data type)">Collection</a></li>

<li><a href="Enumerated_type" title="Enumerated type">Enumerated type</a></li>

<li><a href="Exception_handling" title="Exception handling">Exception</a></li>

<li><a href="Function_type" title="Function type">Function type</a></li>

<li><a href="Opaque_data_type" title="Opaque data type">Opaque data type</a></li>

<li><a href="Recursive_data_type" title="Recursive data type">Recursive data type</a></li>

<li><a href="Semaphore_(programming)" title="Semaphore (programming)">Semaphore</a></li>

<li><a href="Stream_(computing)" title="Stream (computing)">Stream</a></li>

<li><a href="Strongly_typed_identifier" title="Strongly typed identifier">Strongly typed identifier</a></li>

<li><a href="Type_class" title="Type class">Type class</a></li>

<li><a href="Empty_type" title="Empty type">Empty type</a></li>

</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Related topics</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">

<ul><li><a href="Abstract_data_type" title="Abstract data type">Abstract data type</a></li>

<li><a href="Boxing_(computer_programming)" title="Boxing (computer programming)">Boxing</a></li>

<li><a href="Data_structure" title="Data structure">Data structure</a></li>

<li><a href="Generic_programming" title="Generic programming">Generic</a></li>

<ul><li><a href="Metaclass" title="Metaclass">metaclass</a></li></ul></li>

<li><a href="Parametric_polymorphism" title="Parametric polymorphism">Parametric polymorphism</a></li>

<li><a href="Primitive_data_type" title="Primitive data type">Primitive data type</a></li>

<li><a href="Interface_(object-oriented_programming)" title="Interface (object-oriented programming)">Interface</a></li>

<li><a href="Subtyping" title="Subtyping">Subtyping</a></li>

<li><a href="Type_constructor" title="Type constructor">Type constructor</a></li>

<li><a href="Type_conversion" title="Type conversion">Type conversion</a></li>

<li><a href="Type_system" title="Type system">Type system</a></li>

<li><a href="Type_theory" title="Type theory">Type theory</a></li>

<li><a href="Variable_(computer_science)" title="Variable (computer science)">Variable</a></li></ul>

</div></td></tr></tbody></table></div></div><div><div class="zim-footer">

This article is issued from <a class="external text" title="Last edited on 2025-08-07" href="https://en.wikipedia.org/wiki/?title=Floating-point_arithmetic&oldid=1304687672">Wikipedia</a>. The text is available under <a class="external text" href="https://creativecommons.org/licenses/by-sa/4.0/deed.en">Creative Commons Attribution-Share Alike 4.0</a> unless otherwise noted. Additional terms may apply for the media files.

</div>

</div></div>

</div>

</main>

</div>

</body></html>