testfloat_ver
John R. Hauser
2015 February 16
The testfloat_ver
program takes test-case results obtained from
exercising an implementation of floating-point arithmetic and verifies that
those results conform to the IEEE Standard for Binary Floating-Point
Arithmetic.
testfloat_ver
is part of the Berkeley TestFloat package, a small
collection of programs for performing such tests.
For general information about TestFloat, see file
TestFloat-general.html
A single execution of testfloat_ver
verifies results for only a
single floating-point operation and associated options.
The testfloat_ver
program must be repeatedly executed to verify
results for each operation to be tested.
The test cases to be verified are read by testfloat_ver
from
standard input.
This input will typically be piped from another program that, for each test
case, invokes the floating-point operation and writes out the results.
The format of testfloat_ver
’s input is raw hexadecimal text,
described in the section below titled Input Format.
For each test case given to it, testfloat_ver
examines the
computed results and reports any unexpected results as likely errors.
For more about the operation of testfloat_ver
and how to interpret
its output, refer to
TestFloat-general.html
The testfloat_ver
program is executed as a command with this
syntax:
Square brackets (testfloat_ver [<option>...] <function>
[ ]
) denote optional arguments,
<option>
is a supported option, and
<function>
is the name of a testable operation.
The available options are documented below.
The testable operation names are listed in
TestFloat-general.html
testfloat_ver
is executed without any arguments, a summary of
usage is written.
The testfloat_ver
program accepts several command options.
If mutually contradictory options are given, the last one has priority.
-help
The -help
option causes a summary of program usage to be written,
after which the program exits.
-errors <num>
The -errors
option instructs testfloat_ver
to report
no more than the specified number of errors.
The argument to -errors
must be a nonnegative decimal integer.
Once the specified number of error reports has been generated, the program
exits.
The default is -errors
20
Against intuition, -errors
0
testfloat_ver
to continue for any number of errors.
-checkNaNs
The -checkNaNs
option causes testfloat_ver
to verify
the bitwise correctness of NaN results.
In order for this option to be sensible, testfloat_ver
must have
been compiled so that its internal reference implementation of floating-point
(SoftFloat) generates the proper NaN results for the system being tested.
-precision32, -precision64, -precision80
When <function>
is an -precision32
option indicates that the rounding precision should
be -precision64
indicates that the rounding precision
should be -precision80
indicates that the rounding
precision should be the full -precision80
.
-rnear_even, -rnear_maxMag, -rminMag, -rmin, -rmax
When <function>
is an operation that requires rounding, the
-rnear_even
option indicates that rounding should be to
nearest/even, -rnear_maxMag
indicates rounding to nearest/maximum
magnitude (nearest-away), -rminMag
indicates rounding to minimum
magnitude (toward zero), -rmin
indicates rounding to minimum
(down, toward negative infinity), and -rmax
indicates rounding to
maximum (up, toward positive infinity).
These options are ignored for operations that are exact and thus do not round.
When rounding mode is relevant but not specified, the default assumption is
rounding to nearest/even, same as -rnear_even
.
-tininessbefore, -tininessafter
When <function>
is an operation that requires rounding, the
-tininessbefore
option indicates that tininess on underflow should
be detected before rounding, while -tininessafter
indicates that
tininess on underflow should be detected after rounding.
These options are ignored for operations that are exact and thus do not round.
When the method of tininess detection matters but is not specified, the default
assumption is that tininess should be detected before rounding, same as
-tininessbefore
.
-notexact, -exact
When <function>
is an operation that rounds to an integer
(either conversion to an integer type or a roundToInt
operation),
the -notexact
option indicates that the inexact exception
flag should never be raised, while -exact
indicates that the
inexact exception flag should be raised when the result is inexact.
For other operations, these options are ignored.
If neither option is specified, the default assumption is that the
inexact exception flag should not be raised when rounding to an integer,
same as -notexact
.
For a given <function>
argument, the input format expected
by testfloat_ver
is the same as the output generated by program
testfloat_gen
Input to testfloat_ver
is expected to be text, with each line
containing the data for one test case.
The number of input lines thus equals the number of test cases.
A single test case is organized as follows: first are the operands for the
operation, next is the result value obtained, and last is a number indicating
the exception flags that were raised.
These values are all expected to be provided as raw hexadecimal numbers
separated on the line by spaces.
For example, for the command
valid input could include these lines:testfloat_ver f64_add
On each line above, the first two hexadecimal numbers represent the3F90EB5825D6851E C3E0080080000000 C3E0080080000000 01 41E3C00000000000 C182024F8AE474A8 41E377F6C1D46E2D 01 7FD80FFFFFFFFFFF 7FEFFFFFFFFFFF80 7FF0000000000000 05 3FFFED6A25C534BE 3CA1000000020000 3FFFED6A25C534BF 01 ...
Note that, for floating-point values, the sign and exponent are at the
most-significant end of the number.
Thus, for the first number on the first line above, the leading hexadecimal
digits 3F9
are the sign and encoded exponent of the
Exception flags are encoded with one bit per flag as follows:
bit 0
inexact exception bit 1 underflow exception bit 2 overflow exception bit 3 infinite exception (“divide by zero”) bit 4 invalid exception