1README for libm-test math test suite
2====================================
3
4The libm-test math test suite tests a number of function points of
5math functions in the GNU C library.  The following sections contain a
6brief overview.  Please note that the test drivers and the Python
7script "gen-libm-test.py" have some options.  A full list of options
8is available with --help (for the test drivers) and -h for
9"gen-libm-test.py".
10
11
12What is tested?
13===============
14The tests just evaluate the functions at specified points and compare
15the results with precomputed values and the requirements of the ISO
16C99 standard.
17
18Besides testing the special values mandated by IEEE 754 (infinities,
19NaNs and minus zero), some more or less random values are tested.
20
21Files that are part of libm-test
22================================
23
24The main files are "libm-test-<func>.inc".  They are independent of
25the target platform and the specific real floating type and format and
26contain placeholder test "templates" for math functions defined in
27libm.  These files, along with generated files named
28"auto-libm-test-out-<func>", are preprocessed by the Python script
29"gen-libm-test.py" to expand the templates and produce a set of test
30cases for each math function that are specific to the target platform
31but still independent of the real floating type.  The results of the
32processing are "libm-test-<func>.c" and a file "libm-test-ulps.h" with
33platform specific deltas by which the actual math function results may
34deviate from the expected results and still be considered correct.
35
36The test drivers "test-double-<func>.c", "test-float-<func>.c", and
37"test-ldouble-<func>.c", generated by the Makefile, test the normal
38double, float and long double implementation of libm.  Each driver
39selects the desired real floating type to exercise the math functions
40to test with (float, double, or long double) by defining a small set
41of macros just before including the generic "libm-test.c" file. Each
42driver is compiled into a single executable test program with the
43corresponding name.
44
45As mentioned above, the "gen-libm-test.py" script looks for a file
46named "libm-test-ulps" in the platform specific sysdep directory (or
47its fpu or nofpu subdirectory) and for each variant (real floating
48type and rounding mode) of every tested function reads from it the
49maximum difference expressed as Units of Least Precision (ULP) the
50actual result of the function may deviate from the expected result
51before it's considered incorrect.
52
53The "auto-libm-test-out-<func>" files contain sets of test cases to
54exercise, the conditions under which to exercise each, and the
55expected results.  The files are generated by the
56"gen-auto-libm-tests" program from the "auto-libm-test-in" file.  See
57the comments in gen-auto-libm-tests.c for details about the content
58and format of the -in and -out files.
59
60How can I generate "libm-test-ulps"?
61====================================
62
63To automatically generate a new "libm-test-ulps" run "make regen-ulps".
64This generates the file "math/NewUlps" in the build directory.  The file
65contains the sorted results of all the tests.  You can use the "NewUlps"
66file as the machine's updated "libm-test-ulps" file.  Copy "NewUlps" to
67"libm-test-ulps" in the appropriate machine sysdep directory.  Verify
68the changes, post your patch, and check it in after review.
69
70To manually generate a new "libm-test-ulps" file, first remove "ULPs"
71file in the current directory, then you can execute for example:
72    ./testrun.sh math/test-double -u --ignore-max-ulp=yes
73This generates a file "ULPs" with all double ULPs in it, ignoring any
74previously calculated ULPs, and running with the newly built dynamic
75loader and math library (assumes you didn't install your build).  Now
76generate the ULPs for all other formats, the tests will be appending the
77data to the "ULPs" file.  As final step run "gen-libm-test.py" with the
78file as input and ask to generate a pretty printed output in the file
79"NewUlps":
80  gen-libm-test.py -u ULPs -n NewUlps
81Copy "NewUlps" to "libm-test-ulps" in the appropriate machine sysdep
82directory.
83
84Note that the test drivers have an option "-u" to output an unsorted
85list of all epsilons that the functions have.  The output can be read
86in directly but it's better to pretty print it first.
87"gen-libm-test.py" has an option to generate a pretty-printed and
88sorted new ULPs file from the output of the test drivers.
89
90Contents of libm-test-ulps
91==========================
92
93Since libm-test-ulps can be generated automatically, just a few notes.
94The file contains lines for maximal errors of single functions, like:
95
96Function "yn":
97double: 6
98
99The keywords are float, double, and ldouble.
100
101Adding tests to libm-test-<func>.inc
102====================================
103
104The tests are evaluated by a set of special test macros.  The macros
105start with "TEST_" followed by a specification the input values, an
106underscore and a specification of the output values.  As an example,
107the test macro for a function with input of type FLOAT (FLOAT is
108either float, double, long double) and output of type FLOAT is
109"TEST_f_f".  The macro's parameter are the name of the function, the
110input parameter, output parameter and optionally one exception
111parameter.
112
113The accepted parameter types are:
114- "f" for FLOAT
115- "j" for long double.
116- "a" for ARG_FLOAT, the argument type for narrowing functions.
117- "b" for boolean - just tests if the output parameter evaluates to 0
118  or 1 (only for output).
119- "c" for complex.  This parameter needs two values, first the real,
120  then the imaginary part.
121- "i" for int.
122- "l" for long int.
123- "L" for long long int.
124- "u" for unsigned int.
125- "M" for intmax_t.
126- "U" for uintmax_t.
127- "p" for an argument (described in the previous character) passed
128  through a pointer rather than directly.
129- "F" for the address of a FLOAT (only as input parameter)
130- "I" for the address of an int (only as input parameter)
131- "1" for an additional output (either output through a pointer passed
132  as an argument, or to a global variable such as signgam).
133
134How to read the test output
135===========================
136
137Running each test on its own at the default level of verbosity will
138print on stdout a line describing the implementation of math functions
139exercised by the test (float, double, or long double).  This is then
140followed by the details of test failures (if any).  The output concludes
141by a summary listing the number of test cases exercised and the number
142of test failures uncovered.
143
144For each test failure (and for each test case at higher levels of
145verbosity), the output contains the name of the function under test
146and its arguments or conditions that triggered the failure.  Note
147that the name of the function in the output need not correspond
148exactly to the name of the math function actually invoked. For example,
149the output will refer to the "acos" function even if the actual function
150under test is acosf (for the float version) or acosl (for the long
151double version).  Also note that the function arguments may be shown
152in either the decimal or the  hexadecimal floating point format which
153may or may not correspond to the format used in the auto-libm-test-in
154file. Besides the name of the function, for each test failure the
155output contains the actual and expected results and the difference
156between the two, printed in both the decimal and hexadecimal
157floating point format, and the ULP and maximum ULP for the test
158case.
159