| 1 |
|
|---|
| 2 | <HTML>
|
|---|
| 3 |
|
|---|
| 4 | <HEAD>
|
|---|
| 5 | <TITLE>testsoftfloat</TITLE>
|
|---|
| 6 | </HEAD>
|
|---|
| 7 |
|
|---|
| 8 | <BODY>
|
|---|
| 9 |
|
|---|
| 10 | <H1>Berkeley TestFloat Release 3e: <CODE>testsoftfloat</CODE></H1>
|
|---|
| 11 |
|
|---|
| 12 | <P>
|
|---|
| 13 | John R. Hauser<BR>
|
|---|
| 14 | 2018 January 20<BR>
|
|---|
| 15 | </P>
|
|---|
| 16 |
|
|---|
| 17 |
|
|---|
| 18 | <H2>Overview</H2>
|
|---|
| 19 |
|
|---|
| 20 | <P>
|
|---|
| 21 | The <CODE>testsoftfloat</CODE> program tests that a build of the Berkeley
|
|---|
| 22 | SoftFloat library conforms to the IEEE Standard for Binary Floating-Point
|
|---|
| 23 | Arithmetic as expected.
|
|---|
| 24 | Program <CODE>testsoftfloat</CODE> is part of the Berkeley TestFloat package, a
|
|---|
| 25 | small collection of programs for performing such tests.
|
|---|
| 26 | For general information about TestFloat, as well as for basics about the
|
|---|
| 27 | operation of <CODE>testsoftfloat</CODE> and how to interpret its output, see
|
|---|
| 28 | file
|
|---|
| 29 | <A HREF="TestFloat-general.html"><NOBR><CODE>TestFloat-general.html</CODE></NOBR></A>.
|
|---|
| 30 | </P>
|
|---|
| 31 |
|
|---|
| 32 | <P>
|
|---|
| 33 | Note that, even if there are no bugs in the source code for SoftFloat (not
|
|---|
| 34 | guaranteed), a build of SoftFloat might still fail due to an issue with the
|
|---|
| 35 | build process, such as an incompatible compiler option or a compiler bug.
|
|---|
| 36 | </P>
|
|---|
| 37 |
|
|---|
| 38 | <P>
|
|---|
| 39 | The <CODE>testsoftfloat</CODE> program will ordinarily test a function for all
|
|---|
| 40 | five rounding modes defined by the IEEE Floating-Point Standard, one after the
|
|---|
| 41 | other, plus possibly a sixth mode, <I>round to odd</I> (depending on the
|
|---|
| 42 | options selected when <CODE>testsoftfloat</CODE> was compiled).
|
|---|
| 43 | If an operation is not supposed to require rounding, it will by default be
|
|---|
| 44 | tested only with the rounding mode set to <CODE>near_even</CODE>
|
|---|
| 45 | (nearest/even).
|
|---|
| 46 | In the same way, if an operation is affected by the way in which underflow
|
|---|
| 47 | tininess is detected, <CODE>testsoftfloat</CODE> tests the function with
|
|---|
| 48 | tininess detected both before rounding and after rounding.
|
|---|
| 49 | For <NOBR>80-bit</NOBR> double-extended-precision operations affected by
|
|---|
| 50 | rounding precision control, <CODE>testsoftfloat</CODE> also tests the function
|
|---|
| 51 | for all three rounding precision modes, one after the other.
|
|---|
| 52 | Testing can be limited to a single rounding mode, a single tininess mode,
|
|---|
| 53 | and/or a single rounding precision with appropriate command-line options.
|
|---|
| 54 | </P>
|
|---|
| 55 |
|
|---|
| 56 |
|
|---|
| 57 | <H2>Command Syntax</H2>
|
|---|
| 58 |
|
|---|
| 59 | <P>
|
|---|
| 60 | The <CODE>testsoftfloat</CODE> program is executed as a command with this
|
|---|
| 61 | syntax:
|
|---|
| 62 | <BLOCKQUOTE>
|
|---|
| 63 | <PRE>
|
|---|
| 64 | testsoftfloat [<<I>option</I>>...] <<I>function</I>>
|
|---|
| 65 | </PRE>
|
|---|
| 66 | </BLOCKQUOTE>
|
|---|
| 67 | Square brackets (<CODE>[ ]</CODE>) denote optional arguments,
|
|---|
| 68 | <CODE><<I>option</I>></CODE> is a supported option, and
|
|---|
| 69 | <CODE><<I>function</I>></CODE> is the name of either a testable function
|
|---|
| 70 | or a function set.
|
|---|
| 71 | The available options and function sets are documented below.
|
|---|
| 72 | If <CODE>testsoftfloat</CODE> is executed without any arguments, a summary of
|
|---|
| 73 | usage is written.
|
|---|
| 74 | </P>
|
|---|
| 75 |
|
|---|
| 76 |
|
|---|
| 77 | <H2>Options</H2>
|
|---|
| 78 |
|
|---|
| 79 | <P>
|
|---|
| 80 | The <CODE>testsoftfloat</CODE> program accepts several command options.
|
|---|
| 81 | If mutually contradictory options are given, the last one has priority.
|
|---|
| 82 | </P>
|
|---|
| 83 |
|
|---|
| 84 | <H3><CODE>-help</CODE></H3>
|
|---|
| 85 |
|
|---|
| 86 | <P>
|
|---|
| 87 | The <CODE>-help</CODE> option causes a summary of program usage to be written,
|
|---|
| 88 | after which the program exits.
|
|---|
| 89 | </P>
|
|---|
| 90 |
|
|---|
| 91 | <H3><CODE>-seed <<I>num</I>></CODE></H3>
|
|---|
| 92 |
|
|---|
| 93 | <P>
|
|---|
| 94 | The <CODE>-seed</CODE> option sets the seed for the pseudo-random number
|
|---|
| 95 | generator used for generating test cases.
|
|---|
| 96 | The argument to <CODE>-seed</CODE> is a nonnegative integer.
|
|---|
| 97 | Executing the same <CODE>testsoftfloat</CODE> program with the same arguments
|
|---|
| 98 | (including the same pseudo-random number seed) should always perform the same
|
|---|
| 99 | sequence of tests, whereas changing the pseudo-random number seed should result
|
|---|
| 100 | in a different sequence of tests.
|
|---|
| 101 | The default seed number <NOBR>is 1</NOBR>.
|
|---|
| 102 | </P>
|
|---|
| 103 |
|
|---|
| 104 | <H3><CODE>-level <<I>num</I>></CODE></H3>
|
|---|
| 105 |
|
|---|
| 106 | <P>
|
|---|
| 107 | The <CODE>-level</CODE> option sets the level of testing.
|
|---|
| 108 | The argument to <CODE>-level</CODE> can be either 1 <NOBR>or 2</NOBR>.
|
|---|
| 109 | The default is <NOBR>level 1</NOBR>.
|
|---|
| 110 | Level 2 performs many more tests than <NOBR>level 1</NOBR> and thus can reveal
|
|---|
| 111 | bugs not found by <NOBR>level 1</NOBR>.
|
|---|
| 112 | </P>
|
|---|
| 113 |
|
|---|
| 114 | <H3><CODE>-errors <<I>num</I>></CODE></H3>
|
|---|
| 115 |
|
|---|
| 116 | <P>
|
|---|
| 117 | The <CODE>-errors</CODE> option instructs <CODE>testsoftfloat</CODE> to report
|
|---|
| 118 | no more than the specified number of errors for any combination of function,
|
|---|
| 119 | rounding mode, etc.
|
|---|
| 120 | The argument to <CODE>-errors</CODE> must be a nonnegative decimal integer.
|
|---|
| 121 | Once the specified number of error reports has been generated,
|
|---|
| 122 | <CODE>testsoftfloat</CODE> ends the current test and begins the next one, if
|
|---|
| 123 | any.
|
|---|
| 124 | The default is <NOBR><CODE>-errors</CODE> <CODE>20</CODE></NOBR>.
|
|---|
| 125 | </P>
|
|---|
| 126 |
|
|---|
| 127 | <P>
|
|---|
| 128 | Against intuition, <NOBR><CODE>-errors</CODE> <CODE>0</CODE></NOBR> causes
|
|---|
| 129 | <CODE>testsoftfloat</CODE> to report every error it finds.
|
|---|
| 130 | </P>
|
|---|
| 131 |
|
|---|
| 132 | <H3><CODE>-errorstop</CODE></H3>
|
|---|
| 133 |
|
|---|
| 134 | <P>
|
|---|
| 135 | The <CODE>-errorstop</CODE> option causes the program to exit after the first
|
|---|
| 136 | function for which any errors are reported.
|
|---|
| 137 | </P>
|
|---|
| 138 |
|
|---|
| 139 | <H3><CODE>-forever</CODE></H3>
|
|---|
| 140 |
|
|---|
| 141 | <P>
|
|---|
| 142 | The <CODE>-forever</CODE> option causes a single function to be repeatedly
|
|---|
| 143 | tested.
|
|---|
| 144 | Only one rounding mode and/or rounding precision can be tested in a single
|
|---|
| 145 | execution.
|
|---|
| 146 | If not specified, the rounding mode defaults to nearest/even.
|
|---|
| 147 | For <NOBR>80-bit</NOBR> double-extended-precision functions, the rounding
|
|---|
| 148 | precision defaults to full double-extended precision.
|
|---|
| 149 | The testing level is set to 2 by this option.
|
|---|
| 150 | </P>
|
|---|
| 151 |
|
|---|
| 152 | <H3><CODE>-precision32, -precision64, -precision80</CODE></H3>
|
|---|
| 153 |
|
|---|
| 154 | <P>
|
|---|
| 155 | For <NOBR>80-bit</NOBR> double-extended-precision funcions affected by
|
|---|
| 156 | rounding precision control, the <CODE>-precision32</CODE> option restricts
|
|---|
| 157 | testing to only the cases in which the rounding precision is
|
|---|
| 158 | <NOBR>32 bits</NOBR>, equivalent to <NOBR>32-bit</NOBR> single-precision.
|
|---|
| 159 | The other rounding precision choices are not tested.
|
|---|
| 160 | Likewise, <CODE>-precision64</CODE> fixes the rounding precision to
|
|---|
| 161 | <NOBR>64 bits</NOBR>, equivalent to <NOBR>64-bit</NOBR> double-precision;
|
|---|
| 162 | and <CODE>-precision80</CODE> fixes the rounding precision to the full
|
|---|
| 163 | <NOBR>80 bits</NOBR> of the double-extended-precision format.
|
|---|
| 164 | All these options are ignored for operations not affected by rounding precision
|
|---|
| 165 | control.
|
|---|
| 166 | </P>
|
|---|
| 167 |
|
|---|
| 168 | <H3><CODE>-rnear_even, -rnear_maxMag, -rminMag, -rmin, -rmax, -rodd</CODE></H3>
|
|---|
| 169 |
|
|---|
| 170 | <P>
|
|---|
| 171 | The <CODE>-rnear_even</CODE> option restricts testing to only the cases in
|
|---|
| 172 | which the rounding mode is nearest/even.
|
|---|
| 173 | The other rounding mode choices are not tested.
|
|---|
| 174 | Likewise, <CODE>-rnear_maxMag</CODE> forces rounding to nearest/maximum
|
|---|
| 175 | magnitude (nearest-away), <CODE>-rminMag</CODE> forces rounding to minimum
|
|---|
| 176 | magnitude (toward zero), <CODE>-rmin</CODE> forces rounding to minimum (down,
|
|---|
| 177 | toward negative infinity), <CODE>-rmax</CODE> forces rounding to maximum (up,
|
|---|
| 178 | toward positive infinity), and <CODE>-rodd</CODE>, if supported, forces
|
|---|
| 179 | rounding to odd.
|
|---|
| 180 | These options are ignored for operations that are exact and thus do not round.
|
|---|
| 181 | </P>
|
|---|
| 182 |
|
|---|
| 183 | <H3><CODE>-tininessbefore, -tininessafter</CODE></H3>
|
|---|
| 184 |
|
|---|
| 185 | <P>
|
|---|
| 186 | The <CODE>-tininessbefore</CODE> option restricts testing to only the cases in
|
|---|
| 187 | which tininess on underflow is detected before rounding.
|
|---|
| 188 | Likewise, <CODE>-tininessafter</CODE> restricts testing to only the cases in
|
|---|
| 189 | which tininess on underflow is detected after rounding.
|
|---|
| 190 | </P>
|
|---|
| 191 |
|
|---|
| 192 | <H3><CODE>-notexact, -exact</CODE></H3>
|
|---|
| 193 |
|
|---|
| 194 | <P>
|
|---|
| 195 | For functions that round to an integer (conversions to integer types and the
|
|---|
| 196 | <CODE>roundToInt</CODE> functions), the <CODE>-notexact</CODE> option restricts
|
|---|
| 197 | testing to only the cases for which the <CODE><I>exact</I></CODE> operand
|
|---|
| 198 | (specifying whether the <I>inexact</I> exception flag may be raised) is
|
|---|
| 199 | <CODE>false</CODE>.
|
|---|
| 200 | Likewise, the <CODE>-exact</CODE> option restricts testing to only the cases
|
|---|
| 201 | for which the <CODE><I>exact</I></CODE> operand is <CODE>true</CODE>.
|
|---|
| 202 | </P>
|
|---|
| 203 |
|
|---|
| 204 |
|
|---|
| 205 | <H2>Function Sets</H2>
|
|---|
| 206 |
|
|---|
| 207 | <P>
|
|---|
| 208 | Just as <CODE>testsoftfloat</CODE> can test a function for all five or six
|
|---|
| 209 | rounding modes in sequence, multiple functions can be tested with a single
|
|---|
| 210 | execution of <CODE>testsoftfloat</CODE>.
|
|---|
| 211 | Two sets are recognized: <CODE>-all1</CODE> and <CODE>-all2</CODE>.
|
|---|
| 212 | The set <CODE>-all1</CODE> is all one-operand operations, while
|
|---|
| 213 | <CODE>-all2</CODE> is all two-operand operations.
|
|---|
| 214 | A function set is used in place of a function name in the
|
|---|
| 215 | <CODE>testsoftfloat</CODE> command line, such as
|
|---|
| 216 | <BLOCKQUOTE>
|
|---|
| 217 | <PRE>
|
|---|
| 218 | testsoftfloat [<<I>option</I>>...] -all1
|
|---|
| 219 | </PRE>
|
|---|
| 220 | </BLOCKQUOTE>
|
|---|
| 221 | </P>
|
|---|
| 222 |
|
|---|
| 223 | <P>
|
|---|
| 224 | For the purpose of deciding the number of operands of an operation, any
|
|---|
| 225 | <CODE><I>roundingMode</I></CODE> and <CODE><I>exact</I></CODE> arguments are
|
|---|
| 226 | ignored.
|
|---|
| 227 | (Such arguments specify the rounding mode and whether the <I>inexact</I>
|
|---|
| 228 | exception flag may be raised, respectively.)
|
|---|
| 229 | Thus, functions that convert to integer type and the <CODE>roundToInt</CODE>
|
|---|
| 230 | functions are included in the set of one-operand operations tested by
|
|---|
| 231 | <CODE>-all1</CODE>.
|
|---|
| 232 | </P>
|
|---|
| 233 |
|
|---|
| 234 |
|
|---|
| 235 | </BODY>
|
|---|
| 236 |
|
|---|
