2 dnl Copyright (C) 2011-2012 Free Software Foundation, Inc.
3 dnl This file is free software; the Free Software Foundation
4 dnl gives unlimited permission to copy and/or distribute it,
5 dnl with or without modifications, as long as this notice is preserved.
7 AC_DEFUN([gl_FUNC_FMAL],
9 AC_REQUIRE([gl_MATH_H_DEFAULTS])
10 AC_REQUIRE([gl_LONG_DOUBLE_VS_DOUBLE])
12 dnl Persuade glibc <math.h> to declare fmal().
13 AC_REQUIRE([gl_USE_SYSTEM_EXTENSIONS])
15 dnl Determine FMAL_LIBM.
16 gl_MATHFUNC([fmal], [long double], [(long double, long double, long double)])
17 if test $gl_cv_func_fmal_no_libm = yes \
18 || test $gl_cv_func_fmal_in_libm = yes; then
20 case "$gl_cv_func_fmal_works" in
21 *no) REPLACE_FMAL=1 ;;
26 if test $HAVE_FMAL = 0 || test $REPLACE_FMAL = 1; then
27 dnl Find libraries needed to link lib/fmal.c.
28 if test $HAVE_SAME_LONG_DOUBLE_AS_DOUBLE = 1; then
29 AC_REQUIRE([gl_FUNC_FMA])
32 AC_REQUIRE([gl_FUNC_FREXPL])
33 AC_REQUIRE([gl_FUNC_LDEXPL])
34 AC_REQUIRE([gl_FUNC_FEGETROUND])
36 dnl Append $FREXPL_LIBM to FMAL_LIBM, avoiding gratuitous duplicates.
37 case " $FMAL_LIBM " in
38 *" $FREXPL_LIBM "*) ;;
39 *) FMAL_LIBM="$FMAL_LIBM $FREXPL_LIBM" ;;
41 dnl Append $LDEXPL_LIBM to FMAL_LIBM, avoiding gratuitous duplicates.
42 case " $FMAL_LIBM " in
43 *" $LDEXPL_LIBM "*) ;;
44 *) FMAL_LIBM="$FMAL_LIBM $LDEXPL_LIBM" ;;
46 dnl Append $FEGETROUND_LIBM to FMAL_LIBM, avoiding gratuitous duplicates.
47 case " $FMAL_LIBM " in
48 *" $FEGETROUND_LIBM "*) ;;
49 *) FMAL_LIBM="$FMAL_LIBM $FEGETROUND_LIBM" ;;
56 dnl Test whether fmal() has any of the 15 known bugs of glibc 2.11.3 on x86_64.
57 AC_DEFUN([gl_FUNC_FMAL_WORKS],
59 AC_REQUIRE([AC_PROG_CC])
60 AC_REQUIRE([AC_CANONICAL_HOST]) dnl for cross-compiles
61 AC_REQUIRE([gl_FUNC_LDEXPL])
63 LIBS="$LIBS $FMAL_LIBM $LDEXPL_LIBM"
64 AC_CACHE_CHECK([whether fmal works], [gl_cv_func_fmal_works],
70 /* Override the values of <float.h>, like done in float.in.h. */
71 #if defined __i386__ && (defined __BEOS__ || defined __OpenBSD__)
73 # define LDBL_MANT_DIG 64
75 # define LDBL_MIN_EXP (-16381)
77 # define LDBL_MAX_EXP 16384
79 #if defined __i386__ && defined __FreeBSD__
81 # define LDBL_MANT_DIG 64
83 # define LDBL_MIN_EXP (-16381)
85 # define LDBL_MAX_EXP 16384
87 #if (defined _ARCH_PPC || defined _POWER) && defined _AIX && (LDBL_MANT_DIG == 106) && defined __GNUC__
89 # define LDBL_MIN_EXP DBL_MIN_EXP
91 #if defined __sgi && (LDBL_MANT_DIG >= 106)
93 # define LDBL_MANT_DIG 106
96 # define LDBL_MIN_EXP DBL_MIN_EXP
99 long double p0 = 0.0L;
102 int failed_tests = 0;
103 /* This test fails on glibc 2.11 powerpc. */
105 volatile long double x = 1.5L; /* 3 * 2^-1 */
106 volatile long double y = x;
107 volatile long double z = ldexpl (1.0L, LDBL_MANT_DIG + 1); /* 2^65 */
108 /* x * y + z with infinite precision: 2^65 + 9 * 2^-2.
109 Lies between (2^63 + 0) * 2^2 and (2^63 + 1) * 2^2
110 and is closer to (2^63 + 1) * 2^2, therefore the rounding
111 must round up and produce (2^63 + 1) * 2^2. */
112 volatile long double expected = z + 4.0L;
113 volatile long double result = fmal (x, y, z);
114 if (result != expected)
117 /* This test fails on glibc 2.11 powerpc. */
119 volatile long double x = 1.25L; /* 2^0 + 2^-2 */
120 volatile long double y = - x;
121 volatile long double z = ldexpl (1.0L, LDBL_MANT_DIG + 1); /* 2^65 */
122 /* x * y + z with infinite precision: 2^65 - 2^0 - 2^-1 - 2^-4.
123 Lies between (2^64 - 1) * 2^1 and 2^64 * 2^1
124 and is closer to (2^64 - 1) * 2^1, therefore the rounding
125 must round down and produce (2^64 - 1) * 2^1. */
126 volatile long double expected = (ldexpl (1.0L, LDBL_MANT_DIG) - 1.0L) * 2.0L;
127 volatile long double result = fmal (x, y, z);
128 if (result != expected)
131 /* This test fails on glibc 2.11 x86,x86_64,powerpc, glibc 2.7 hppa,sparc,
134 volatile long double x = 1.0L + ldexpl (1.0L, 1 - LDBL_MANT_DIG); /* 2^0 + 2^-63 */
135 volatile long double y = x;
136 volatile long double z = 4.0L; /* 2^2 */
137 /* x * y + z with infinite precision: 2^2 + 2^0 + 2^-62 + 2^-126.
138 Lies between (2^63 + 2^61) * 2^-61 and (2^63 + 2^61 + 1) * 2^-61
139 and is closer to (2^63 + 2^61 + 1) * 2^-61, therefore the rounding
140 must round up and produce (2^63 + 2^61 + 1) * 2^-61. */
141 volatile long double expected = 4.0L + 1.0L + ldexpl (1.0L, 3 - LDBL_MANT_DIG);
142 volatile long double result = fmal (x, y, z);
143 if (result != expected)
146 /* This test fails on glibc 2.11 x86,x86_64,powerpc glibc 2.7 hppa,sparc,
149 volatile long double x = 1.0L + ldexpl (1.0L, 1 - LDBL_MANT_DIG); /* 2^0 + 2^-63 */
150 volatile long double y = - x;
151 volatile long double z = 8.0L; /* 2^3 */
152 /* x * y + z with infinite precision: 2^2 + 2^1 + 2^0 - 2^-62 - 2^-126.
153 Lies between (2^63 + 2^62 + 2^61 - 1) * 2^-61 and
154 (2^63 + 2^62 + 2^61) * 2^-61 and is closer to
155 (2^63 + 2^62 + 2^61 - 1) * 2^-61, therefore the rounding
156 must round down and produce (2^63 + 2^62 + 2^61 - 1) * 2^-61. */
157 volatile long double expected = 7.0L - ldexpl (1.0L, 3 - LDBL_MANT_DIG);
158 volatile long double result = fmal (x, y, z);
159 if (result != expected)
162 /* This test fails on glibc 2.11 powerpc. */
164 volatile long double x = 1.25L; /* 2^0 + 2^-2 */
165 volatile long double y = - 0.75L; /* - 2^0 + 2^-2 */
166 volatile long double z = ldexpl (1.0L, LDBL_MANT_DIG); /* 2^64 */
167 /* x * y + z with infinite precision: 2^64 - 2^0 + 2^-4.
168 Lies between (2^64 - 2^0) and 2^64 and is closer to (2^64 - 2^0),
169 therefore the rounding must round down and produce (2^64 - 2^0). */
170 volatile long double expected = ldexpl (1.0L, LDBL_MANT_DIG) - 1.0L;
171 volatile long double result = fmal (x, y, z);
172 if (result != expected)
175 if ((LDBL_MANT_DIG % 2) == 1)
177 /* These tests fail on glibc 2.7 hppa,sparc, OSF/1 5.1. */
179 volatile long double x = 1.0L + ldexpl (1.0L, - (LDBL_MANT_DIG + 1) / 2); /* 2^0 + 2^-27 */
180 volatile long double y = 1.0L - ldexpl (1.0L, - (LDBL_MANT_DIG + 1) / 2); /* 2^0 - 2^-27 */
181 volatile long double z = - ldexpl (1.0L, LDBL_MIN_EXP - LDBL_MANT_DIG); /* - 2^-1074 */
182 /* x * y + z with infinite precision: 2^0 - 2^-54 - 2^-1074.
183 Lies between (2^53 - 1) * 2^-53 and 2^53 * 2^-53 and is closer to
184 (2^53 - 1) * 2^-53, therefore the rounding must round down and
185 produce (2^53 - 1) * 2^-53. */
186 volatile long double expected = 1.0L - ldexpl (1.0L, - LDBL_MANT_DIG);
187 volatile long double result = fmal (x, y, z);
188 if (result != expected)
192 volatile long double x = 1.0L + ldexpl (1.0L, - (LDBL_MANT_DIG + 1) / 2); /* 2^0 + 2^-57 */
193 volatile long double y = x;
194 volatile long double z = ldexpl (1.0L, - LDBL_MANT_DIG); /* 2^-113 */
195 /* x * y + z with infinite precision: 2^0 + 2^-56 + 2^-113 + 2^-114.
196 Lies between (2^112 + 2^56) * 2^-112 and (2^112 + 2^56 + 1) * 2^-112
197 and is closer to (2^112 + 2^56 + 1) * 2^-112, therefore the rounding
198 must round up and produce (2^112 + 2^56 + 1) * 2^-112. */
199 volatile long double expected =
200 1.0L + ldexpl (1.0L, - (LDBL_MANT_DIG - 1) / 2) + ldexpl (1.0L, 1 - LDBL_MANT_DIG);
201 volatile long double result = fmal (x, y, z);
202 if (result != expected)
208 /* These tests fail on glibc 2.11 x86,x86_64,powerpc, mingw. */
210 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
211 volatile long double y = x;
212 volatile long double z = ldexpl (1.0L, LDBL_MIN_EXP - LDBL_MANT_DIG); /* 2^-16445 */
213 /* x * y + z with infinite precision: 2^0 + 2^-31 + 2^-64 + 2^-16445.
214 Lies between (2^63 + 2^32 + 0) * 2^-63 and (2^63 + 2^32 + 1) * 2^-63
215 and is closer to (2^63 + 2^32 + 1) * 2^-63, therefore the rounding
216 must round up and produce (2^63 + 2^32 + 1) * 2^-63. */
217 volatile long double expected =
218 1.0L + ldexpl (1.0L, 1 - LDBL_MANT_DIG / 2) + ldexpl (1.0L, 1 - LDBL_MANT_DIG);
219 volatile long double result = fmal (x, y, z);
220 if (result != expected)
224 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
225 volatile long double y = x;
226 volatile long double z = ldexpl (1.0L, - LDBL_MANT_DIG); /* 2^-64 */
227 /* x * y + z with infinite precision: 2^0 + 2^-31 + 2^-63.
228 Rounding must return this value unchanged. */
229 volatile long double expected = 1.0L + ldexpl (1.0L, 1 - LDBL_MANT_DIG / 2) + ldexpl (1.0L, 1 - LDBL_MANT_DIG);
230 volatile long double result = fmal (x, y, z);
231 if (result != expected)
235 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
236 volatile long double y = x;
237 volatile long double z = ldexpl (1.0L, 1 - LDBL_MANT_DIG); /* 2^-63 */
238 /* x * y + z with infinite precision: 2^0 + 2^-31 + 2^-63 + 2^-64.
239 Lies between (2^63 + 2^32 + 1) * 2^-63 and (2^63 + 2^32 + 2) * 2^-63
240 and is at the same distance from each. According to the round-to-even
241 rule, the rounding must round up and produce (2^63 + 2^32 + 2) * 2^-63. */
242 volatile long double expected = 1.0L + ldexpl (1.0L, -31) + ldexpl (1.0L, -62);
243 volatile long double result = fmal (x, y, z);
244 if (result != expected)
248 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
249 volatile long double y = x;
250 volatile long double z = ldexpl (1.0L, LDBL_MANT_DIG / 2 + 1); /* 2^33 */
251 /* x * y + z with infinite precision: 2^33 + 2^0 + 2^-31 + 2^-64.
252 Lies between (2^63 + 2^30) * 2^-30 and (2^63 + 2^30 + 1) * 2^-30
253 and is closer to (2^63 + 2^30 + 1) * 2^-30, therefore the rounding
254 must round up and produce (2^63 + 2^30 + 1) * 2^-30. */
255 volatile long double expected = z + 1.0L + ldexp (1.0L, 2 - LDBL_MANT_DIG / 2);
256 volatile long double result = fmal (x, y, z);
257 if (result != expected)
261 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
262 volatile long double y = x;
263 volatile long double z = - ldexpl (1.0, 1 - LDBL_MANT_DIG); /* - 2^-63 */
264 /* x * y + z with infinite precision: 2^0 + 2^-31 - 2^-64.
265 Lies between (2^63 + 2^32 - 1) * 2^-63 and (2^63 + 2^32) * 2^-63
266 and is at the same distance from each. According to the round-to-even
267 rule, the rounding must round up and produce (2^63 + 2^32) * 2^-63. */
268 volatile long double expected = 1.0L + ldexpl (1.0L, 1 - LDBL_MANT_DIG / 2);
269 volatile long double result = fmal (x, y, z);
270 if (result != expected)
274 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
275 volatile long double y = x;
276 volatile long double z = - 1.0L; /* - 2^0 */
277 /* x * y + z with infinite precision: 2^-31 + 2^-64.
278 Rounding must return this value unchanged. */
279 volatile long double expected = ldexpl (1.0L, 1 - LDBL_MANT_DIG / 2) + ldexpl (1.0L, - LDBL_MANT_DIG);
280 volatile long double result = fmal (x, y, z);
281 if (result != expected)
285 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
286 volatile long double y = - x;
287 volatile long double z = 2.0L; /* 2^1 */
288 /* x * y + z with infinite precision: 2^0 - 2^31 - 2^-64.
289 Rounding must return this value unchanged. */
290 volatile long double expected = 1.0L - ldexpl (1.0L, 1 - LDBL_MANT_DIG / 2) - ldexpl (1.0L, - LDBL_MANT_DIG);
291 volatile long double result = fmal (x, y, z);
292 if (result != expected)
296 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2); /* 2^0 + 2^-32 */
297 volatile long double y = - x;
298 volatile long double z = ldexpl (1.0L, LDBL_MANT_DIG / 2 + 2); /* 2^34 */
299 /* x * y + z with infinite precision: 2^34 - (2^0 + 2^-31 + 2^-64).
300 Lies between (2^64 - 2^30 - 1) * 2^-30 and (2^64 - 2^30) * 2^-30
301 and is closer to (2^64 - 2^30 - 1) * 2^-30, therefore the rounding
302 must round down and produce (2^64 - 2^30 - 1) * 2^-30. */
303 volatile long double expected = z - 1.0L - ldexpl (1.0L, 2 - LDBL_MANT_DIG / 2);
304 volatile long double result = fmal (x, y, z);
305 if (result != expected)
309 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2 - 1); /* 2^0 + 2^-33 */
310 volatile long double y = 1.0L - ldexpl (1.0L, - LDBL_MANT_DIG / 2 - 1); /* 2^0 - 2^-33 */
311 volatile long double z = - ldexpl (1.0L, - LDBL_MANT_DIG - 1); /* 2^-65 */
312 /* x * y + z with infinite precision: 2^0 - 2^-65 - 2^-66.
313 Lies between (2^64 - 1) * 2^-64 and 2^64 * 2^-64 and is closer to
314 (2^64 - 1) * 2^-64, therefore the rounding must round down and
315 produce (2^64 - 1) * 2^-64. */
316 volatile long double expected = 1.0L - ldexpl (1.0L, - LDBL_MANT_DIG);
317 volatile long double result = fmal (x, y, z);
318 if (result != expected)
322 volatile long double x = 1.0L + ldexpl (1.0L, - LDBL_MANT_DIG / 2 - 1); /* 2^0 + 2^-33 */
323 volatile long double y = 1.0L - ldexpl (1.0L, - LDBL_MANT_DIG / 2 - 1); /* 2^0 - 2^-33 */
324 volatile long double z = - 1.0L; /* 2^0 */
325 /* x * y + z with infinite precision: - 2^-66.
326 Rounding must return this value unchanged. */
327 volatile long double expected = - ldexpl (1.0L, - LDBL_MANT_DIG - 2);
328 volatile long double result = fmal (x, y, z);
329 if (result != expected)
333 /* This test fails on glibc 2.11 x86,x86_64,powerpc, glibc 2.7 hppa,sparc,
334 FreeBSD 6.4 x86, mingw. */
336 long double minus_inf = -1.0L / p0;
337 volatile long double x = ldexpl (1.0L, LDBL_MAX_EXP - 1);
338 volatile long double y = ldexpl (1.0L, LDBL_MAX_EXP - 1);
339 volatile long double z = minus_inf;
340 volatile long double result = fmal (x, y, z);
341 if (!(result == minus_inf))
344 /* This test fails on glibc 2.11 x86,x86_64,powerpc glibc 2.7 hppa,sparc,
345 MacOS X 10.5, FreeBSD 6.4 x86, OSF/1 5.1, mingw. */
347 volatile long double x = ldexpl (1.0L, LDBL_MAX_EXP - 1);
348 volatile long double y = 2.0L;
349 volatile long double z =
350 - ldexpl (ldexpl (1.0L, LDBL_MAX_EXP - 1) - ldexpl (1.0L, LDBL_MAX_EXP - LDBL_MANT_DIG - 1), 1);
351 volatile long double expected = ldexpl (1.0L, LDBL_MAX_EXP - LDBL_MANT_DIG);
352 volatile long double result = fmal (x, y, z);
353 if (result != expected)
358 [gl_cv_func_fmal_works=yes],
359 [gl_cv_func_fmal_works=no],
360 [dnl Guess no, even on glibc systems.
361 gl_cv_func_fmal_works="guessing no"
367 # Prerequisites of lib/fmal.c.
368 AC_DEFUN([gl_PREREQ_FMAL], [:])