forked from Github_Repos/cvw
640 lines
24 KiB
HTML
640 lines
24 KiB
HTML
|
|
<HTML>
|
|
|
|
<HEAD>
|
|
<TITLE>Berkeley TestFloat Source Documentation</TITLE>
|
|
</HEAD>
|
|
|
|
<BODY>
|
|
|
|
<H1>Berkeley TestFloat Release 3e: Source Documentation</H1>
|
|
|
|
<P>
|
|
John R. Hauser<BR>
|
|
2018 January 20<BR>
|
|
</P>
|
|
|
|
|
|
<H2>Contents</H2>
|
|
|
|
<BLOCKQUOTE>
|
|
<TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0>
|
|
<COL WIDTH=25>
|
|
<COL WIDTH=*>
|
|
<TR><TD COLSPAN=2>1. Introduction</TD></TR>
|
|
<TR><TD COLSPAN=2>2. Limitations</TD></TR>
|
|
<TR><TD COLSPAN=2>3. Acknowledgments and License</TD></TR>
|
|
<TR><TD COLSPAN=2>4. TestFloat Package Directory Structure</TD></TR>
|
|
<TR><TD COLSPAN=2>5. Dependence on Berkeley SoftFloat</TD></TR>
|
|
<TR><TD COLSPAN=2>6. Issues for Porting TestFloat to a New Target</TD></TR>
|
|
<TR>
|
|
<TD></TD>
|
|
<TD>6.1. Standard Headers <CODE><stdbool.h></CODE> and
|
|
<CODE><stdint.h></CODE></TD>
|
|
</TR>
|
|
<TR><TD></TD><TD>6.2. Standard Header <CODE><fenv.h></CODE></TD></TR>
|
|
<TR><TD></TD><TD>6.3. Macros for Build Options</TD></TR>
|
|
<TR><TD></TD><TD>6.4. Specializing the <CODE>testfloat</CODE> Program</TD></TR>
|
|
<TR><TD></TD><TD>6.5. Improving the Random Number Functions</TD></TR>
|
|
<TR><TD COLSPAN=2>7. Contact Information</TD></TR>
|
|
</TABLE>
|
|
</BLOCKQUOTE>
|
|
|
|
|
|
<H2>1. Introduction</H2>
|
|
|
|
<P>
|
|
This document gives information needed for compiling and/or porting Berkeley
|
|
TestFloat, a small collection of programs for testing that an implementation of
|
|
binary floating-point conforms to the IEEE Standard for Floating-Point
|
|
Arithmetic.
|
|
For basic documentation about TestFloat refer to
|
|
<A HREF="TestFloat-general.html"><NOBR><CODE>TestFloat-general.html</CODE></NOBR></A>.
|
|
</P>
|
|
|
|
<P>
|
|
The source code for TestFloat is intended to be relatively machine-independent.
|
|
Most programs in the TestFloat package should be compilable with any
|
|
ISO-Standard C compiler that also supports <NOBR>64-bit</NOBR> integers.
|
|
If the all-in-one <CODE>testfloat</CODE> program will be used to test a new
|
|
floating-point implementation, additional effort will likely be required to
|
|
retarget that program to invoke the new floating-point operations.
|
|
TestFloat has been successfully compiled with the GNU C Compiler
|
|
(<CODE>gcc</CODE>) for several platforms.
|
|
</P>
|
|
|
|
<P>
|
|
<NOBR>Release 3</NOBR> of TestFloat was a complete rewrite relative to
|
|
<NOBR>Release 2c</NOBR> or earlier.
|
|
The current version of TestFloat is <NOBR>Release 3e</NOBR>.
|
|
</P>
|
|
|
|
<P>
|
|
TestFloat depends on Berkeley SoftFloat, which is a software implementation of
|
|
binary floating-point that conforms to the IEEE Standard for Floating-Point
|
|
Arithmetic.
|
|
SoftFloat is not included with the TestFloat sources.
|
|
It can be obtained from the Web page
|
|
<A HREF="http://www.jhauser.us/arithmetic/SoftFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/SoftFloat.html</CODE></NOBR></A>.
|
|
</P>
|
|
|
|
|
|
<H2>2. Limitations</H2>
|
|
|
|
<P>
|
|
TestFloat assumes the computer has an addressable byte size of either 8 or
|
|
<NOBR>16 bits</NOBR>.
|
|
(Nearly all computers in use today have <NOBR>8-bit</NOBR> bytes.)
|
|
</P>
|
|
|
|
<P>
|
|
TestFloat is written entirely <NOBR>in C</NOBR>.
|
|
The C compiler used must conform at a minimum to the 1989 ANSI standard for the
|
|
C language (same as the 1990 ISO standard) and must in addition support basic
|
|
arithmetic on <NOBR>64-bit</NOBR> integers.
|
|
Earlier releases of TestFloat were capable of testing <NOBR>32-bit</NOBR>
|
|
single-precision and <NOBR>64-bit</NOBR> double-precision floating-point
|
|
without requiring compiler support for <NOBR>64-bit</NOBR> integers, but this
|
|
option is not supported starting with <NOBR>Release 3</NOBR>.
|
|
Since 1999, ISO standards for C have mandated compiler support for
|
|
<NOBR>64-bit</NOBR> integers.
|
|
A compiler conforming to the 1999 C Standard or later is recommended but not
|
|
strictly required.
|
|
</P>
|
|
|
|
<P>
|
|
<NOBR>C Standard</NOBR> header files <CODE><stdbool.h></CODE> and
|
|
<CODE><stdint.h></CODE> are required for defining standard Boolean and
|
|
integer types.
|
|
If these headers are not supplied with the C compiler, minimal substitutes must
|
|
be provided.
|
|
TestFloat’s dependence on these headers is detailed later in
|
|
<NOBR>section 6.1</NOBR>, <I>Standard Headers <CODE><stdbool.h></CODE>
|
|
and <CODE><stdint.h></CODE></I>.
|
|
</P>
|
|
|
|
|
|
<H2>3. Acknowledgments and License</H2>
|
|
|
|
<P>
|
|
The TestFloat package was written by me, <NOBR>John R.</NOBR> Hauser.
|
|
<NOBR>Release 3</NOBR> of TestFloat was a completely new implementation
|
|
supplanting earlier releases.
|
|
The project to create <NOBR>Release 3</NOBR> (now <NOBR>through 3e</NOBR>) was
|
|
done in the employ of the University of California, Berkeley, within the
|
|
Department of Electrical Engineering and Computer Sciences, first for the
|
|
Parallel Computing Laboratory (Par Lab) and then for the ASPIRE Lab.
|
|
The work was officially overseen by Prof. Krste Asanovic, with funding provided
|
|
by these sources:
|
|
<BLOCKQUOTE>
|
|
<TABLE>
|
|
<COL>
|
|
<COL WIDTH=10>
|
|
<COL>
|
|
<TR>
|
|
<TD VALIGN=TOP><NOBR>Par Lab:</NOBR></TD>
|
|
<TD></TD>
|
|
<TD>
|
|
Microsoft (Award #024263), Intel (Award #024894), and U.C. Discovery
|
|
(Award #DIG07-10227), with additional support from Par Lab affiliates Nokia,
|
|
NVIDIA, Oracle, and Samsung.
|
|
</TD>
|
|
</TR>
|
|
<TR>
|
|
<TD VALIGN=TOP><NOBR>ASPIRE Lab:</NOBR></TD>
|
|
<TD></TD>
|
|
<TD>
|
|
DARPA PERFECT program (Award #HR0011-12-2-0016), with additional support from
|
|
ASPIRE industrial sponsor Intel and ASPIRE affiliates Google, Nokia, NVIDIA,
|
|
Oracle, and Samsung.
|
|
</TD>
|
|
</TR>
|
|
</TABLE>
|
|
</BLOCKQUOTE>
|
|
</P>
|
|
|
|
<P>
|
|
The following applies to the whole of TestFloat <NOBR>Release 3e</NOBR> as well
|
|
as to each source file individually.
|
|
</P>
|
|
|
|
<P>
|
|
Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018 The Regents of the
|
|
University of California.
|
|
All rights reserved.
|
|
</P>
|
|
|
|
<P>
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, are permitted provided that the following conditions are met:
|
|
<OL>
|
|
|
|
<LI>
|
|
<P>
|
|
Redistributions of source code must retain the above copyright notice, this
|
|
list of conditions, and the following disclaimer.
|
|
</P>
|
|
|
|
<LI>
|
|
<P>
|
|
Redistributions in binary form must reproduce the above copyright notice, this
|
|
list of conditions, and the following disclaimer in the documentation and/or
|
|
other materials provided with the distribution.
|
|
</P>
|
|
|
|
<LI>
|
|
<P>
|
|
Neither the name of the University nor the names of its contributors may be
|
|
used to endorse or promote products derived from this software without specific
|
|
prior written permission.
|
|
</P>
|
|
|
|
</OL>
|
|
</P>
|
|
|
|
<P>
|
|
THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS “AS IS”,
|
|
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
|
|
DISCLAIMED.
|
|
IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
|
|
INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
|
|
BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
|
|
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
|
|
OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
|
|
ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
</P>
|
|
|
|
|
|
<H2>4. TestFloat Package Directory Structure</H2>
|
|
|
|
<P>
|
|
Because TestFloat is targeted to multiple platforms, its source code is
|
|
slightly scattered between target-specific and target-independent directories
|
|
and files.
|
|
The supplied directory structure is as follows:
|
|
<BLOCKQUOTE>
|
|
<PRE>
|
|
doc
|
|
source
|
|
subj-C
|
|
build
|
|
template
|
|
Linux-386-GCC
|
|
Linux-386-SSE2-GCC
|
|
Linux-x86_64-GCC
|
|
Linux-ARM-VFPv2-GCC
|
|
Win32-MinGW
|
|
Win32-SSE2-MinGW
|
|
Win64-MinGW-w64
|
|
</PRE>
|
|
</BLOCKQUOTE>
|
|
The majority of the TestFloat sources are provided in the <CODE>source</CODE>
|
|
directory.
|
|
The <NOBR><CODE>subj-C</CODE></NOBR> subdirectory contains the sources that
|
|
configure the all-in-one <CODE>testfloat</CODE> program to test the C
|
|
compiler’s implementation of the standard C types <CODE>float</CODE>,
|
|
<CODE>double</CODE>, and possibly <CODE>long</CODE> <CODE>double</CODE>.
|
|
The ‘<CODE>subj</CODE>’ in <NOBR><CODE>subj-C</CODE></NOBR> is an
|
|
abbreviation of <I>subject</I>, referring to the floating-point that is the
|
|
subject of the test.
|
|
If <CODE>testfloat</CODE> is retargeted to test other floating-point
|
|
implementations, the corresponding source files would be expected to be in
|
|
other subdirectories alongside <NOBR><CODE>subj-C</CODE></NOBR>, with names of
|
|
the form <NOBR><CODE>subj-<<I>target</I>></CODE></NOBR>.
|
|
More about retargeting <CODE>testfloat</CODE> is found in
|
|
<NOBR>section 6.4</NOBR>, <I>Specializing the <CODE>testfloat</CODE>
|
|
Program</I>.
|
|
</P>
|
|
|
|
<P>
|
|
The <CODE>build</CODE> directory is intended to contain a subdirectory for each
|
|
target platform for which builds of the TestFloat programs may be created.
|
|
For each build target, the target’s subdirectory is where all derived
|
|
object files and the completed TestFloat executables are created.
|
|
The <CODE>template</CODE> subdirectory is not an actual build target but
|
|
contains sample files for creating new target directories.
|
|
</P>
|
|
|
|
<P>
|
|
Ignoring the <CODE>template</CODE> directory, the supplied target directories
|
|
are intended to follow a naming system of
|
|
<NOBR><CODE><<I>execution-environment</I>>-<<I>compiler</I>></CODE></NOBR>.
|
|
For the example targets,
|
|
<NOBR><CODE><<I>execution-environment</I>></CODE></NOBR> is
|
|
<NOBR><CODE>Linux-386</CODE></NOBR>, <NOBR><CODE>Linux-386-SSE2</CODE></NOBR>,
|
|
<NOBR><CODE>Linux-x86_64</CODE></NOBR>,
|
|
<NOBR><CODE>Linux-ARM-VFPv2</CODE></NOBR>, <CODE>Win32</CODE>,
|
|
<NOBR><CODE>Win32-SSE2</CODE></NOBR>, or <CODE>Win64</CODE>, and
|
|
<NOBR><CODE><<I>compiler</I>></CODE></NOBR> is <CODE>GCC</CODE>,
|
|
<CODE>MinGW</CODE>, or <NOBR><CODE>MinGW-w64</CODE></NOBR>.
|
|
</P>
|
|
|
|
<P>
|
|
All of the supplied target directories are merely examples that may or may not
|
|
be correct for compiling on any particular system.
|
|
There are currently no plans to include and maintain in the TestFloat package
|
|
the build files needed for a great many users’ compilation environments,
|
|
which can span a huge range of operating systems, compilers, and other tools.
|
|
</P>
|
|
|
|
<P>
|
|
As supplied, each target directory contains two files:
|
|
<BLOCKQUOTE>
|
|
<PRE>
|
|
Makefile
|
|
platform.h
|
|
</PRE>
|
|
</BLOCKQUOTE>
|
|
The provided <CODE>Makefile</CODE> is written for GNU <CODE>make</CODE>.
|
|
A build of TestFloat for the specific target is begun by executing the
|
|
<CODE>make</CODE> command with the target directory as the current directory.
|
|
A completely different build tool can be used if an appropriate
|
|
<CODE>Makefile</CODE> equivalent is created.
|
|
</P>
|
|
|
|
<P>
|
|
The <CODE>platform.h</CODE> header file exists to provide a location for
|
|
additional C declarations specific to the build target.
|
|
Every C source file of TestFloat contains a <CODE>#include</CODE> for
|
|
<CODE>platform.h</CODE>.
|
|
In many cases, the contents of <CODE>platform.h</CODE> can be as simple as one
|
|
or two lines of code.
|
|
If the target’s compiler or library has bugs or other shortcomings,
|
|
workarounds for these issues may be possible with target-specific declarations
|
|
in <CODE>platform.h</CODE>, without the need to modify the main TestFloat
|
|
sources.
|
|
</P>
|
|
|
|
<P>
|
|
It may not be necessary to build all of the TestFloat programs.
|
|
For testing a floating-point implementation, typically
|
|
<CODE>testfloat_gen</CODE> and <CODE>testfloat</CODE> will not both be used,
|
|
and <CODE>testfloat_ver</CODE> may not be needed either.
|
|
The Makefile (or equivalent) can be modified not to create unneeded programs.
|
|
This may be especially relevant for the all-in-one test program
|
|
<CODE>testfloat</CODE>, which might not build without special attention.
|
|
</P>
|
|
|
|
|
|
<H2>5. Dependence on Berkeley SoftFloat</H2>
|
|
|
|
<P>
|
|
In addition to the distributed sources, TestFloat depends on the existence of a
|
|
compatible Berkeley SoftFloat library and the corresponding header file
|
|
<CODE>softfloat.h</CODE>.
|
|
As mentioned earlier, SoftFloat is a separate package available at Web page
|
|
<A HREF="http://www.jhauser.us/arithmetic/SoftFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/SoftFloat.html</CODE></NOBR></A>.
|
|
The SoftFloat library must be compiled before the TestFloat programs can be
|
|
built.
|
|
In the example Makefiles, the locations of the SoftFloat header files and
|
|
pre-compiled library are specified by these macros:
|
|
<BLOCKQUOTE>
|
|
<DL>
|
|
<DT><CODE>SOFTFLOAT_INCLUDE_DIR</CODE>
|
|
<DD>
|
|
The path of the directory containing <CODE>softfloat.h</CODE>, as well as other
|
|
nonstandard header files referenced by <CODE>softfloat.h</CODE>, if any.
|
|
<DT><CODE>SOFTFLOAT_H</CODE>
|
|
<DD>
|
|
A list of the full paths of all SoftFloat header files needed by SoftFloat
|
|
clients. This list must include <CODE>softfloat.h</CODE> and may also include
|
|
other header files referenced by <CODE>softfloat.h</CODE>, such as
|
|
<CODE>softfloat_types.h</CODE>.
|
|
This macro is used only to establish build dependencies between the SoftFloat
|
|
header files and TestFloat’s source files, in case the SoftFloat header
|
|
files are changed.
|
|
<DT><CODE>SOFTFLOAT_LIB</CODE>
|
|
<DD>
|
|
The full path of the compiled SoftFloat library (usually
|
|
<CODE>softfloat.a</CODE> or <CODE>libsoftfloat.a</CODE>).
|
|
</DL>
|
|
</BLOCKQUOTE>
|
|
</P>
|
|
|
|
|
|
<H2>6. Issues for Porting TestFloat to a New Target</H2>
|
|
|
|
<H3>6.1. Standard Headers <CODE><stdbool.h></CODE> and <CODE><stdint.h></CODE></H3>
|
|
|
|
<P>
|
|
The TestFloat sources make use of standard headers
|
|
<CODE><stdbool.h></CODE> and <CODE><stdint.h></CODE>, which have
|
|
been part of the ISO C Standard Library since 1999.
|
|
With any recent compiler, these standard headers are likely to be supported,
|
|
even if the compiler does not claim complete conformance to the latest ISO C
|
|
Standard.
|
|
For older or nonstandard compilers, substitutes for
|
|
<CODE><stdbool.h></CODE> and <CODE><stdint.h></CODE> may need to be
|
|
created.
|
|
TestFloat depends on these names from <CODE><stdbool.h></CODE>:
|
|
<BLOCKQUOTE>
|
|
<PRE>
|
|
bool
|
|
true
|
|
false
|
|
</PRE>
|
|
</BLOCKQUOTE>
|
|
and on these names from <CODE><stdint.h></CODE>:
|
|
<BLOCKQUOTE>
|
|
<PRE>
|
|
uint16_t
|
|
uint32_t
|
|
uint64_t
|
|
int32_t
|
|
int64_t
|
|
UINT64_C
|
|
INT64_C
|
|
uint_least8_t
|
|
uint_fast8_t
|
|
uint_fast16_t
|
|
uint_fast32_t
|
|
uint_fast64_t
|
|
int_fast8_t
|
|
int_fast16_t
|
|
int_fast32_t
|
|
int_fast64_t
|
|
</PRE>
|
|
</BLOCKQUOTE>
|
|
</P>
|
|
|
|
|
|
<H3>6.2. Standard Header <CODE><fenv.h></CODE></H3>
|
|
|
|
<P>
|
|
Because the supplied all-in-one <CODE>testfloat</CODE> program tests the
|
|
floating-point operations of the C language, it uses the facilities provided by
|
|
standard C header <CODE><fenv.h></CODE> to access the floating-point
|
|
environment of C, in particular to set the rounding mode and to access the
|
|
floating-point exception flags.
|
|
Like <CODE><stdbool.h></CODE> and <CODE><stdint.h></CODE>,
|
|
<CODE><fenv.h></CODE> has been part of the ISO C Standard Library since
|
|
1999, but older or nonstandard C compilers may not support it.
|
|
</P>
|
|
|
|
<P>
|
|
Some form of standard header <CODE><fenv.h></CODE> is needed only if the
|
|
<CODE>testfloat</CODE> program is wanted <EM>and</EM> the program will not be
|
|
retargeted to invoke a floating-point implementation in a way that bypasses the
|
|
standard C environment.
|
|
Typically, if <CODE>testfloat</CODE> is wanted, it will be retargeted to invoke
|
|
a new floating-point implementation directly, making
|
|
<CODE><fenv.h></CODE> irrelevant.
|
|
For more about retargeting <CODE>testfloat</CODE>, see <NOBR>section 6.4</NOBR>
|
|
below, <I>Specializing the <CODE>testfloat</CODE> Program</I>.
|
|
</P>
|
|
|
|
|
|
<H3>6.3. Macros for Build Options</H3>
|
|
|
|
<P>
|
|
The TestFloat source files are affected by several C preprocessor macros:
|
|
<BLOCKQUOTE>
|
|
<DL>
|
|
<DT><CODE>LITTLEENDIAN</CODE>
|
|
<DD>
|
|
Must be defined for little-endian machines;
|
|
must not be defined for big-endian machines.
|
|
<DT><CODE>INLINE</CODE>
|
|
<DD>
|
|
Can be defined to a sequence of tokens used to indicate that a C function
|
|
should be inlined.
|
|
If the compiler does not support the inlining of functions, this macro must not
|
|
be defined.
|
|
For compilers that conform to the C Standard’s rules for inline
|
|
functions, this macro can be defined as the single keyword <CODE>inline</CODE>.
|
|
For other compilers that follow a convention pre-dating the standardization of
|
|
<CODE>inline</CODE>, this macro may need to be defined to <CODE>extern</CODE>
|
|
<CODE>inline</CODE>.
|
|
<DT><CODE>THREAD_LOCAL</CODE>
|
|
<DD>
|
|
Can be defined to a sequence of tokens that, when appearing at the start of a
|
|
variable declaration, indicates to the C compiler that the variable is
|
|
<I>per-thread</I>, meaning that each execution thread gets its own separate
|
|
instance of the variable.
|
|
This macro is used in the supplied version of Berkeley SoftFloat’s header
|
|
<CODE>softfloat.h</CODE>, in the declarations of variables
|
|
<CODE>softfloat_roundingMode</CODE>, <CODE>softfloat_detectTininess</CODE>,
|
|
<CODE>extF80_roundingPrecision</CODE>, and
|
|
<CODE>softfloat_exceptionFlags</CODE>.
|
|
To use the supplied, unmodified header <CODE>softfloat.h</CODE>, this macro
|
|
must be defined (or not defined) the same as when the SoftFloat library was
|
|
built.
|
|
</DL>
|
|
<DL>
|
|
<DT><CODE>FLOAT16</CODE>
|
|
<DD>
|
|
Must be defined if the TestFloat programs are to support the
|
|
<NOBR>16-bit</NOBR> half-precision floating-point format.
|
|
<DT><CODE>FLOAT64</CODE>
|
|
<DD>
|
|
Must be defined if the TestFloat programs are to support the
|
|
<NOBR>64-bit</NOBR> double-precision floating-point format.
|
|
<DT><CODE>EXTFLOAT80</CODE>
|
|
<DD>
|
|
Must be defined if the TestFloat programs are to support the
|
|
<NOBR>80-bit</NOBR> double-extended-precision floating-point format.
|
|
<DT><CODE>FLOAT128</CODE>
|
|
<DD>
|
|
Must be defined if the TestFloat programs are to support the
|
|
<NOBR>128-bit</NOBR> quadruple-precision floating-point format.
|
|
<DT><CODE>FLOAT_ROUND_ODD</CODE>
|
|
<DD>
|
|
Must be defined if the TestFloat programs are to support rounding to odd
|
|
(jamming).
|
|
To be useful, this option also requires that the Berkeley SoftFloat library was
|
|
compiled with macro <CODE>SOFTFLOAT_ROUND_ODD</CODE> defined.
|
|
</DL>
|
|
</BLOCKQUOTE>
|
|
Following the usual custom <NOBR>for C</NOBR>, for all the macros except
|
|
<CODE>INLINE</CODE> and <CODE>THREAD_LOCAL</CODE>, the content of a
|
|
macro’s definition is irrelevant;
|
|
what matters is a macro’s effect on <CODE>#ifdef</CODE> directives.
|
|
</P>
|
|
|
|
<P>
|
|
It is recommended that any definition of macros <CODE>LITTLEENDIAN</CODE>,
|
|
<CODE>INLINE</CODE>, and <CODE>THREAD_LOCAL</CODE> be made in a build
|
|
target’s <CODE>platform.h</CODE> header file, because these macros are
|
|
expected to be determined inflexibly by the target machine and compiler.
|
|
The other five macros select build options, and hence might be better located
|
|
in the target’s Makefile (or its equivalent).
|
|
</P>
|
|
|
|
|
|
<H3>6.4. Specializing the <CODE>testfloat</CODE> Program</H3>
|
|
|
|
<P>
|
|
The supplied sources for the all-in-one <CODE>testfloat</CODE> program cause
|
|
<CODE>testfloat</CODE> to test the C compiler’s <CODE>float</CODE> and
|
|
<CODE>double</CODE> types for C operations <CODE>+</CODE>, <CODE>-</CODE>,
|
|
<CODE>*</CODE>, <CODE>/</CODE>, etc.
|
|
The supplied version is also capable of testing C type <CODE>long</CODE>
|
|
<CODE>double</CODE> if the sources are compiled with one of these macros
|
|
defined:
|
|
<BLOCKQUOTE>
|
|
<DL>
|
|
<DT><CODE>LONG_DOUBLE_IS_EXTFLOAT80</CODE>
|
|
<DD>
|
|
Indicates that type <CODE>long</CODE> <CODE>double</CODE> is
|
|
<NOBR>80-bit</NOBR> double-extended-precision floating-point.
|
|
<DT><CODE>LONG_DOUBLE_IS_FLOAT128</CODE>
|
|
<DD>
|
|
Indicates that type <CODE>long</CODE> <CODE>double</CODE> is
|
|
<NOBR>128-bit</NOBR> quadruple-precision floating-point.
|
|
</DL>
|
|
</BLOCKQUOTE>
|
|
By default, <CODE>testfloat</CODE> assumes that only the IEEE Standard’s
|
|
original four rounding modes (<CODE>near_even</CODE>, <CODE>minMag</CODE>,
|
|
<CODE>min</CODE>, and <CODE>max</CODE>) are supported by the floating-point
|
|
being tested.
|
|
For other rounding modes, additional macro can be defined:
|
|
<BLOCKQUOTE>
|
|
<DL>
|
|
<DT><CODE>SUBJFLOAT_ROUND_NEAR_MAXMAG</CODE>
|
|
<DD>
|
|
Indicates that the subject floating-point supports rounding mode
|
|
<CODE>near_maxMag</CODE> (nearest/away).
|
|
<DT><CODE>SUBJFLOAT_ROUND_ODD</CODE>
|
|
<DD>
|
|
Indicates that the subject floating-point supports rounding mode
|
|
<CODE>odd</CODE> (jamming).
|
|
</DL>
|
|
</BLOCKQUOTE>
|
|
</P>
|
|
|
|
<P>
|
|
To test a new and/or different implementation of floating-point,
|
|
<CODE>testfloat</CODE> must normally be retargeted to invoke this other
|
|
floating-point instead of C’s floating-point.
|
|
Two source files define the functions that <CODE>testfloat</CODE> uses to
|
|
invoke floating-point operations for testing:
|
|
<BLOCKQUOTE>
|
|
<PRE>
|
|
subjfloat_config.h
|
|
subjfloat.c
|
|
</PRE>
|
|
</BLOCKQUOTE>
|
|
For the default target of testing C’s floating-point, these files are
|
|
contained in directory <NOBR><CODE>source/subj-C</CODE></NOBR> as discussed
|
|
earlier.
|
|
For a different subject floating-point, it is recommended that appropriate
|
|
versions of <CODE>subjfloat_config.h</CODE> and <CODE>subjfloat.c</CODE> be
|
|
stored in a sibling <NOBR><CODE>subj-<<I>target</I>></CODE></NOBR>
|
|
directory, where <CODE><<I>target</I>></CODE> names the particular
|
|
target.
|
|
</P>
|
|
|
|
<P>
|
|
Header file <CODE>subjfloat_config.h</CODE> defines a macro of the form
|
|
<CODE>SUBJ_*</CODE> for each subject function supported.
|
|
For example, if function <CODE>subj_f32_add</CODE> exists to perform
|
|
<NOBR>32-bit</NOBR> floating-point addition, then
|
|
<CODE>subjfloat_config.h</CODE> should have a definition for macro
|
|
<CODE>SUBJ_F32_ADD</CODE>.
|
|
The actual function <CODE>subj_f32_add</CODE> is expected to be defined in
|
|
<CODE>subjfloat.c</CODE>, along with all other subject functions.
|
|
A common header file, <CODE>subjfloat.h</CODE>, (not target-specific) provides
|
|
prototype declarations for all possible subject functions that
|
|
<CODE>testfloat</CODE> may be compiled to test, whether actually existing or
|
|
not.
|
|
(There is no penalty for the header to declare prototypes of nonexistent
|
|
functions that are never called.)
|
|
For a specific build of <CODE>testfloat</CODE>, the <CODE>-list</CODE> option
|
|
will list all subject functions that the <CODE>testfloat</CODE> program is able
|
|
to invoke and thus test.
|
|
</P>
|
|
|
|
<P>
|
|
In the source code as supplied, macros <CODE>LONG_DOUBLE_IS_EXTFLOAT80</CODE>
|
|
and <CODE>LONG_DOUBLE_IS_FLOAT128</CODE> affect only the target-specific source
|
|
files in <NOBR><CODE>source/subj-C</CODE></NOBR>, so these macros can be
|
|
ignored for any other subject floating-point that does not depend on them.
|
|
On the other hand, macros <CODE>SUBJFLOAT_ROUND_NEAR_MAXMAG</CODE> and
|
|
<CODE>SUBJFLOAT_ROUND_ODD</CODE> always determine whether the
|
|
<CODE>testfloat</CODE> program attempts to test rounding modes
|
|
<CODE>near_maxMag</CODE> and <CODE>odd</CODE>, regardless of the subject
|
|
floating-point.
|
|
</P>
|
|
|
|
|
|
<H3>6.5. Improving the Random Number Functions</H3>
|
|
|
|
<P>
|
|
If you are serious about using TestFloat for testing floating-point, you should
|
|
consider replacing the random number functions in <CODE>random.c</CODE>.
|
|
The supplied random number functions are built on top of the standard C
|
|
<CODE>rand</CODE> function.
|
|
Because function <CODE>rand</CODE> is rather poor on some systems, the
|
|
functions in <CODE>random.c</CODE> assume very little about the quality of
|
|
<CODE>rand</CODE>.
|
|
As a result, <CODE>rand</CODE> is called more frequently than it might need to
|
|
be, shortening the time before random number sequences repeat, and possibly
|
|
wasting time as well.
|
|
If <CODE>rand</CODE> is better on a given target platform, or if another,
|
|
better random number generator is available (such as <CODE>rand48</CODE> on
|
|
UNIX-derived systems), TestFloat can be improved by overriding the given
|
|
<CODE>random.c</CODE> with a target-specific one.
|
|
</P>
|
|
|
|
<P>
|
|
Rather than modifying the supplied file <CODE>random.c</CODE>, it is
|
|
recommended instead that a new, alternate file be created and the
|
|
target’s Makefile be modified to refer to that alternate file in place of
|
|
<CODE>random.c</CODE>.
|
|
</P>
|
|
|
|
|
|
<H2>7. Contact Information</H2>
|
|
|
|
<P>
|
|
At the time of this writing, the most up-to-date information about TestFloat
|
|
and the latest release can be found at the Web page
|
|
<A HREF="http://www.jhauser.us/arithmetic/TestFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/TestFloat.html</CODE></NOBR></A>.
|
|
</P>
|
|
|
|
|
|
</BODY>
|
|
|