* [attention machine maintainers] [PATCH] <math.h> issignaling
[not found] <8762148l9v.fsf@schwinge.name>
@ 2013-03-22 11:18 ` Thomas Schwinge
2013-03-24 21:30 ` David Miller
` (2 more replies)
0 siblings, 3 replies; 8+ messages in thread
From: Thomas Schwinge @ 2013-03-22 11:18 UTC (permalink / raw)
To: libc-alpha, libc-ports
[-- Attachment #1: Type: text/plain, Size: 47853 bytes --]
Hi!
On Wed, 06 Mar 2013 18:56:28 +0100, I wrote:
> In IEEE 754-1985, »Annex A Recommended Functions and Predicates«, 10) a
> function to distinguish signaling from quiet NaNs is suggested. This is
> further elaborated in IEEE 754-2008, »5.7.2 General operations« as a
> isSignaling operation. Such a macro/function is not yet part of any
> published revision of the C programming language standard, but has been
> under discussion for some time alreay, see for example in
> <http://www.open-std.org/jtc1/sc22/wg14/www/docs/PostPortland2012.htm>'s
> N1664 (the latest of its kind that is publically available, as far as I
> know), »14.7 Inquiry macros [...] Suggested change to C11: [...] 7.12.3.7
> The issignaling macro«:
>
> | Synopsis
> | [1] #define __STDC_WANT_IEC_00000_EXT1__
> | #include <math.h>
> | int issignaling(real-floating x);
> |
> | Description
> | [2] The issignaling macro determines whether its argument value is a signaling NaN, without
> | raising a floating-point exception.
> |
> | Returns
> | [3] The issignaling macro returns a nonzero value if and only if its argument is a signaling NaN.
I have now pushed a new version of this patch in the
tschwinge/issignaling branch. Would machine maintainers please test this
branch/patch and report back? Any new testsuite failures are
interesting, and especially also any differences in math/*.out files.
Carlos, is my understanding correct (based on a Wikipedia article), that
HPPA is (as MIPS) one of the few HIGH_ORDER_BIT_IS_SET_FOR_SNAN (see the
patch) architectures?
My own testing:
32-bit x86: OK.
x86_64: OK.
PowerPC -m64:
math/basic-test.out:
Failure: double x = (double) (long double) sNaN, !issignaling
That is, this type cast -- which is a IEEE 754-2008 general-computational
convertFormat operation (IEEE 754-2008, 5.4.2) -- does not turn the sNaN
into a qNaN (whilst raising an INVALID exception; not checked here),
which is contrary to IEEE 754-2008 5.1 and 7.2. This I consider a
compiler issue (powerpc-linux-gnu-gcc (Sourcery CodeBench 2012.09-92)
4.7.2).
MIPS32 O32:
math/test-double.out, math/test-idouble.out:
Result:
is: sNaN
should be: qNaN
Seen for qNaN inputs with: pow, remainder.
MIPS -mabi=64:
math/test-double.out, math/test-idouble.out: as MIPS32 O32.
math/test-ldouble.out, math/test-ildoubl.out:
Result:
is: sNaN
should be: qNaN
Seen for different kinds of "invalid" input (out of range, +/-inf, NaN)
with: acos, asin, acosh, atanh, log, log10, log2, pow, sqrt, tgamma,
fmod, remainder.
MIPS -EL -msoft-float -mabi=64:
math/test-double.out, math/test-idouble.out: as MIPS32 O32.
math/test-ldouble.out, math/test-ildoubl.out: as MIPS -mabi=64.
The MIPS presumably are all consider bugs in the implementation due to
HIGH_ORDER_BIT_IS_SET_FOR_SNAN; to be worked on next.
diff --git NEWS NEWS
index f0965d6..0d70ead 100644
--- NEWS
+++ NEWS
@@ -26,6 +26,8 @@ Version 2.18
* Added a benchmark framework to track performance of functions in glibc.
+* New <math.h> macro named issignaling to check for a signaling NaN (sNaN).
+ This is currently GNU-specific.
\f
Version 2.17
diff --git Versions.def Versions.def
index 7c7d1f8..0462239 100644
--- Versions.def
+++ Versions.def
@@ -58,6 +58,7 @@ libm {
GLIBC_2.3.4
GLIBC_2.4
GLIBC_2.15
+ GLIBC_2.18
}
libnsl {
GLIBC_2.0
diff --git include/math.h include/math.h
index 2b040c6..cbbce59 100644
--- include/math.h
+++ include/math.h
@@ -23,11 +23,14 @@ hidden_proto (__isnanl)
libm_hidden_proto (__fpclassify)
libm_hidden_proto (__fpclassifyf)
+libm_hidden_proto (__issignaling)
+libm_hidden_proto (__issignalingf)
libm_hidden_proto (__exp)
libm_hidden_proto (__expf)
# ifndef __NO_LONG_DOUBLE_MATH
libm_hidden_proto (__fpclassifyl)
+libm_hidden_proto (__issignalingl)
libm_hidden_proto (__expl)
libm_hidden_proto (__expm1l)
# endif
diff --git manual/arith.texi manual/arith.texi
index faf25cc..0c0b53e 100644
--- manual/arith.texi
+++ manual/arith.texi
@@ -387,6 +387,13 @@ to
@end smallexample
@end deftypefn
+@comment math.h
+@comment GNU
+@deftypefn {Macro} int issignaling (@emph{float-type} @var{x})
+This macro returns a nonzero value if @var{x} is a signaling NaN
+(sNaN). It is a GNU extension.
+@end deftypefn
+
Another set of floating-point classification functions was provided by
BSD. @Theglibc{} also supports these functions; however, we
recommend that you use the ISO C99 macros in new code. Those are standard
diff --git manual/libm-err-tab.pl manual/libm-err-tab.pl
index 5aa2603..1564669 100755
--- manual/libm-err-tab.pl
+++ manual/libm-err-tab.pl
@@ -77,7 +77,9 @@ use vars qw (%results @all_floats %suffices @all_functions);
"remainder", "remquo", "rint", "round", "scalb", "scalbn", "scalbln",
"sin", "sincos", "sinh", "sqrt", "tan", "tanh", "tgamma",
"trunc", "y0", "y1", "yn" );
-# "fpclassify", "isfinite", "isnormal", "signbit" are not tabulated
+# fpclassify, isnormal, isfinite, isinf, isnan, issignaling, signbit,
+# isgreater, isgreaterequal, isless, islessequal, islessgreater, isunordered
+# are not tabulated.
if ($#ARGV == 0) {
$sources = $ARGV[0];
diff --git math/Makefile math/Makefile
index e216dfb..f396ba2 100644
--- math/Makefile
+++ math/Makefile
@@ -58,7 +58,7 @@ libm-calls = e_acos e_acosh e_asin e_atan2 e_atanh e_cosh e_exp e_fmod \
s_catan s_casin s_ccos s_csin s_ctan s_ctanh s_cacos \
s_casinh s_cacosh s_catanh s_csqrt s_cpow s_cproj s_clog10 \
s_fma s_lrint s_llrint s_lround s_llround e_exp10 w_log2 \
- s_isinf_ns $(calls:s_%=m_%) x2y2m1 k_casinh
+ s_isinf_ns s_issignaling $(calls:s_%=m_%) x2y2m1 k_casinh
include ../Makeconfig
diff --git math/Versions math/Versions
index 0988851..513ab14 100644
--- math/Versions
+++ math/Versions
@@ -198,4 +198,7 @@ libm {
__gamma_r_finite; __gammaf_r_finite; __gammal_r_finite;
__exp_finite; __expf_finite; __expl_finite;
}
+ GLIBC_2.18 {
+ __issignaling; __issignalingf; __issignalingl;
+ }
}
diff --git math/basic-test.c math/basic-test.c
index ffead2e..61df3a0 100644
--- math/basic-test.c
+++ math/basic-test.c
@@ -20,6 +20,9 @@
#include <float.h>
#include <stdio.h>
+#include <math-tests.h>
+
+
static int errors = 0;
@@ -39,6 +42,10 @@ NAME (void) \
/* Variables are declared volatile to forbid some compiler \
optimizations. */ \
volatile FLOAT Inf_var, qNaN_var, zero_var, one_var; \
+ /* A sNaN is only guaranteed to be representable in variables with */ \
+ /* static (or thread-local) storage duration. */ \
+ static volatile FLOAT sNaN_var = __builtin_nans ## SUFFIX (""); \
+ static volatile FLOAT minus_sNaN_var = -__builtin_nans ## SUFFIX (""); \
FLOAT x1, x2; \
\
zero_var = 0.0; \
@@ -49,6 +56,8 @@ NAME (void) \
(void) &zero_var; \
(void) &one_var; \
(void) &qNaN_var; \
+ (void) &sNaN_var; \
+ (void) &minus_sNaN_var; \
(void) &Inf_var; \
\
\
@@ -56,16 +65,41 @@ NAME (void) \
check (#FLOAT " isinf (-inf) == -1", isinf (-Inf_var) == -1); \
check (#FLOAT " !isinf (1)", !(isinf (one_var))); \
check (#FLOAT " !isinf (qNaN)", !(isinf (qNaN_var))); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " !isinf (sNaN)", !(isinf (sNaN_var))); \
\
check (#FLOAT " isnan (qNaN)", isnan (qNaN_var)); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " isnan (sNaN)", isnan (sNaN_var)); \
check (#FLOAT " isnan (-qNaN)", isnan (-qNaN_var)); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " isnan (-sNaN)", isnan (minus_sNaN_var)); \
check (#FLOAT " !isnan (1)", !(isnan (one_var))); \
check (#FLOAT " !isnan (inf)", !(isnan (Inf_var))); \
\
+ check (#FLOAT " !issignaling (qNaN)", !(issignaling (qNaN_var))); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " issignaling (sNaN)", issignaling (sNaN_var)); \
+ check (#FLOAT " !issignaling (-qNaN)", !(issignaling (-qNaN_var))); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " issignaling (-sNaN)", issignaling (minus_sNaN_var)); \
+ check (#FLOAT " !issignaling (1)", !(issignaling (one_var))); \
+ check (#FLOAT " !issignaling (inf)", !(issignaling (Inf_var))); \
+ \
check (#FLOAT " inf == inf", Inf_var == Inf_var); \
check (#FLOAT " -inf == -inf", -Inf_var == -Inf_var); \
check (#FLOAT " inf != -inf", Inf_var != -Inf_var); \
check (#FLOAT " qNaN != qNaN", qNaN_var != qNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " sNaN != sNaN", sNaN_var != sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " qNaN != sNaN", qNaN_var != sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " -sNaN != -sNaN", minus_sNaN_var != minus_sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " sNaN != -sNaN", sNaN_var != minus_sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " qNaN != -sNaN", qNaN_var != minus_sNaN_var); \
\
/* \
the same tests but this time with NAN from <bits/nan.h> \
@@ -76,6 +110,11 @@ NAME (void) \
check (#FLOAT " !isinf (NAN)", !(isinf (NAN))); \
check (#FLOAT " !isinf (-NAN)", !(isinf (-NAN))); \
check (#FLOAT " NAN != NAN", NAN != NAN); \
+ check (#FLOAT " NAN != qNaN", NAN != qNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " NAN != sNaN", NAN != sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " NAN != -sNaN", NAN != minus_sNaN_var); \
\
/* \
And again with the value returned by the `nan' function. \
@@ -86,6 +125,12 @@ NAME (void) \
check (#FLOAT " !isinf (-nan (\"\"))", !(isinf (-nan ## SUFFIX ("")))); \
check (#FLOAT " nan (\"\") != nan (\"\")", \
nan ## SUFFIX ("") != nan ## SUFFIX ("")); \
+ check (#FLOAT " nan (\"\") != qNaN", nan ## SUFFIX ("") != qNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " nan (\"\") != sNaN", nan ## SUFFIX ("") != sNaN_var); \
+ if (SNAN_TESTS (FLOAT)) \
+ check (#FLOAT " nan (\"\") != -sNaN", \
+ nan ## SUFFIX ("") != minus_sNaN_var); \
\
/* test if EPSILON is ok */ \
x1 = 1.0; \
@@ -108,6 +153,9 @@ void \
NAME (void) \
{ \
volatile DOUBLE Inf_var, qNaN_var, zero_var, one_var; \
+ /* A sNaN is only guaranteed to be representable in variables with */ \
+ /* static (or thread-local) storage duration. */ \
+ static volatile DOUBLE sNaN_var = __builtin_nans ## SUFFIX (""); \
FLOAT x1, x2; \
\
zero_var = 0.0; \
@@ -116,10 +164,22 @@ NAME (void) \
Inf_var = one_var / zero_var; \
\
(void) &qNaN_var; \
+ (void) &sNaN_var; \
(void) &Inf_var; \
\
x1 = (FLOAT) qNaN_var; \
- check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") qNaN", isnan (x1) != 0); \
+ check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") qNaN, isnan", isnan (x1)); \
+ check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") qNaN, !issignaling", \
+ !issignaling (x1)); \
+ if (SNAN_TESTS (FLOAT)) \
+ { \
+ x1 = (FLOAT) sNaN_var; \
+ check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") sNaN, isnan", isnan (x1)); \
+ /* Upon type conversion, a sNaN is converted into a qNaN plus an */ \
+ /* INVALID exception (not checked here). */ \
+ check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") sNaN, !issignaling", \
+ !issignaling (x1)); \
+ } \
x2 = (FLOAT) Inf_var; \
check (" "#FLOAT" x = ("#FLOAT") ("#DOUBLE") Inf", isinf (x2) != 0); \
}
diff --git math/bits/mathcalls.h math/bits/mathcalls.h
index e5af507..870c54c 100644
--- math/bits/mathcalls.h
+++ math/bits/mathcalls.h
@@ -360,6 +360,12 @@ __MATHCALL (fma,, (_Mdouble_ __x, _Mdouble_ __y, _Mdouble_ __z));
__END_NAMESPACE_C99
#endif
+#ifdef __USE_GNU
+/* Test for signaling NaN. */
+__MATHDECL_1 (int, __issignaling,, (_Mdouble_ __value))
+ __attribute__ ((__const__));
+#endif
+
#if defined __USE_MISC || defined __USE_XOPEN_EXTENDED
/* Return X times (2 to the Nth power). */
__MATHCALL (scalb,, (_Mdouble_ __x, _Mdouble_ __n));
diff --git math/gen-libm-test.pl math/gen-libm-test.pl
index f50f1d9..aa60d9d 100755
--- math/gen-libm-test.pl
+++ math/gen-libm-test.pl
@@ -322,7 +322,7 @@ sub parse_args {
}
# Special handling for some macros:
$cline .= " (\"$str\", ";
- if ($args[0] =~ /fpclassify|isnormal|isfinite|isinf|isnan|signbit
+ if ($args[0] =~ /fpclassify|isnormal|isfinite|isinf|isnan|issignaling|signbit
|isgreater|isgreaterequal|isless|islessequal
|islessgreater|isunordered/x) {
$c_call = "$args[0] (";
diff --git math/libm-test.inc math/libm-test.inc
index 1b70c35..d5bc060 100644
--- math/libm-test.inc
+++ math/libm-test.inc
@@ -42,7 +42,7 @@
cbrt, ceil, copysign, cos, cosh, erf, erfc, exp, exp10, exp2, expm1,
fabs, fdim, finite, floor, fma, fmax, fmin, fmod, fpclassify,
frexp, gamma, hypot,
- ilogb, isfinite, isinf, isnan, isnormal,
+ ilogb, isfinite, isinf, isnan, isnormal, issignaling,
isless, islessequal, isgreater, isgreaterequal, islessgreater, isunordered,
j0, j1, jn,
ldexp, lgamma, log, log10, log1p, log2, logb,
@@ -87,9 +87,8 @@
aren't checked at the moment.
NaN values: There exist signalling and quiet NaNs. This implementation
- only uses quiet NaN as parameter but does not differentiate
- between the two kinds of NaNs as result. Where the sign of a NaN is
- significant, this is not tested.
+ only uses quiet NaN as parameter. Where the sign of a NaN is
+ significant, this is not tested. The payload of NaNs is not examined.
Inline functions: Inlining functions should give an improvement in
speed - but not in precission. The inlined functions return
@@ -263,6 +262,19 @@ set_max_error (FLOAT current, FLOAT *curr_max_error)
}
+/* Print a FLOAT. */
+static void
+print_float (FLOAT f)
+{
+ /* As printf doesn't differ between a sNaN and a qNaN, do this manually. */
+ if (issignaling (f))
+ printf ("sNaN\n");
+ else if (isnan (f))
+ printf ("qNaN\n");
+ else
+ printf ("% .20" PRINTF_EXPR " % .20" PRINTF_XEXPR "\n", f, f);
+}
+
/* Should the message print to screen? This depends on the verbose flag,
and the test status. */
static int
@@ -532,7 +544,11 @@ check_float_internal (const char *test_name, FLOAT computed, FLOAT expected,
FLOAT ulp = 0;
test_exceptions (test_name, exceptions);
- if (isnan (computed) && isnan (expected))
+ if (issignaling (computed) && issignaling (expected))
+ ok = 1;
+ else if (issignaling (computed) || issignaling (expected))
+ ok = 0;
+ else if (isnan (computed) && isnan (expected))
ok = 1;
else if (isinf (computed) && isinf (expected))
{
@@ -546,8 +562,9 @@ check_float_internal (const char *test_name, FLOAT computed, FLOAT expected,
else
ok = 1;
}
- /* Don't calc ulp for NaNs or infinities. */
- else if (isinf (computed) || isnan (computed) || isinf (expected) || isnan (expected))
+ /* Don't calculate ULPs for infinities or any kind of NaNs. */
+ else if (isinf (computed) || isnan (computed)
+ || isinf (expected) || isnan (expected))
ok = 0;
else
{
@@ -592,10 +609,10 @@ check_float_internal (const char *test_name, FLOAT computed, FLOAT expected,
printf ("Failure: ");
printf ("Test: %s\n", test_name);
printf ("Result:\n");
- printf (" is: % .20" PRINTF_EXPR " % .20" PRINTF_XEXPR "\n",
- computed, computed);
- printf (" should be: % .20" PRINTF_EXPR " % .20" PRINTF_XEXPR "\n",
- expected, expected);
+ printf (" is: ");
+ print_float (computed);
+ printf (" should be: ");
+ print_float (expected);
if (print_diff)
{
printf (" difference: % .20" PRINTF_EXPR " % .20" PRINTF_XEXPR
@@ -6440,6 +6457,22 @@ isnormal_test (void)
}
static void
+issignaling_test (void)
+{
+ START (issignaling);
+
+ TEST_f_b (issignaling, 0, 0);
+ TEST_f_b (issignaling, minus_zero, 0);
+ TEST_f_b (issignaling, 10, 0);
+ TEST_f_b (issignaling, min_subnorm_value, 0);
+ TEST_f_b (issignaling, plus_infty, 0);
+ TEST_f_b (issignaling, minus_infty, 0);
+ TEST_f_b (issignaling, qnan_value, 0);
+
+ END (issignaling);
+}
+
+static void
isunordered_test (void)
{
START (isunordered);
@@ -11075,6 +11108,7 @@ main (int argc, char **argv)
isinf_test ();
isnan_test ();
isnormal_test ();
+ issignaling_test ();
signbit_test ();
/* Trigonometric functions: */
diff --git math/math.h math/math.h
index 2f25c23..e3adf09 100644
--- math/math.h
+++ math/math.h
@@ -282,6 +282,20 @@ enum
#endif /* Use ISO C99. */
+#ifdef __USE_GNU
+/* Return nonzero value if X is a signaling NaN. */
+# ifdef __NO_LONG_DOUBLE_MATH
+# define issignaling(x) \
+ (sizeof (x) == sizeof (float) ? __issignalingf (x) : __issignaling (x))
+# else
+# define issignaling(x) \
+ (sizeof (x) == sizeof (float) \
+ ? __issignalingf (x) \
+ : sizeof (x) == sizeof (double) \
+ ? __issignaling (x) : __issignalingl (x))
+# endif
+#endif /* Use GNU. */
+
#ifdef __USE_MISC
/* Support for various different standard error handling behaviors. */
typedef enum
diff --git math/test-snan.c math/test-snan.c
index f185cbb..82f1dbe 100644
--- math/test-snan.c
+++ math/test-snan.c
@@ -1,4 +1,4 @@
-/* Test signaling NaNs in isnan, isinf, and similar functions.
+/* Test signaling NaNs in issignaling, isnan, isinf, and similar functions.
Copyright (C) 2008-2013 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Andreas Jaeger <aj@suse.de>, 2005.
@@ -119,6 +119,48 @@ NAME (void) \
feenableexcept (FE_ALL_EXCEPT); \
if (sigsetjmp(sigfpe_buf, 0)) \
{ \
+ printf (#FLOAT " issignaling (qNaN) raised SIGFPE\n"); \
+ errors++; \
+ } else { \
+ check (#FLOAT " issignaling (qNaN)", !issignaling (qNaN_var)); \
+ } \
+ \
+ feclearexcept(FE_ALL_EXCEPT); \
+ feenableexcept (FE_ALL_EXCEPT); \
+ if (sigsetjmp(sigfpe_buf, 0)) \
+ { \
+ printf (#FLOAT " issignaling (-qNaN) raised SIGFPE\n"); \
+ errors++; \
+ } else { \
+ check (#FLOAT " issignaling (-qNaN)", !issignaling (-qNaN_var)); \
+ } \
+ \
+ feclearexcept(FE_ALL_EXCEPT); \
+ feenableexcept (FE_ALL_EXCEPT); \
+ if (sigsetjmp(sigfpe_buf, 0)) \
+ { \
+ printf (#FLOAT " issignaling (sNaN) raised SIGFPE\n"); \
+ errors++; \
+ } else { \
+ check (#FLOAT " issignaling (sNaN)", \
+ SNAN_TESTS (FLOAT) ? issignaling (sNaN_var) : 1); \
+ } \
+ \
+ feclearexcept(FE_ALL_EXCEPT); \
+ feenableexcept (FE_ALL_EXCEPT); \
+ if (sigsetjmp(sigfpe_buf, 0)) \
+ { \
+ printf (#FLOAT " issignaling (-sNaN) raised SIGFPE\n"); \
+ errors++; \
+ } else { \
+ check (#FLOAT " issignaling (-sNaN)", \
+ SNAN_TESTS (FLOAT) ? issignaling (minus_sNaN_var) : 1); \
+ } \
+ \
+ feclearexcept(FE_ALL_EXCEPT); \
+ feenableexcept (FE_ALL_EXCEPT); \
+ if (sigsetjmp(sigfpe_buf, 0)) \
+ { \
printf (#FLOAT " isnan (qNaN) raised SIGFPE\n"); \
errors++; \
} else { \
diff --git ports/sysdeps/hppa/math_private.h ports/sysdeps/hppa/math_private.h
new file mode 100644
index 0000000..03b0711
--- /dev/null
+++ ports/sysdeps/hppa/math_private.h
@@ -0,0 +1,27 @@
+/* Internal math stuff. HPPA version.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#ifndef _MATH_PRIVATE_H
+
+/* One of the few architectures where the meaning of the quiet/signaling bit is
+ inverse to IEEE 754-2008 (as well as common practice for IEEE 754-1985). */
+#define HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+
+#include_next <math_private.h>
+
+#endif
diff --git ports/sysdeps/mips/math_private.h ports/sysdeps/mips/math_private.h
new file mode 100644
index 0000000..f0ba4ee
--- /dev/null
+++ ports/sysdeps/mips/math_private.h
@@ -0,0 +1,27 @@
+/* Internal math stuff. MIPS version.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#ifndef _MATH_PRIVATE_H
+
+/* One of the few architectures where the meaning of the quiet/signaling bit is
+ inverse to IEEE 754-2008 (as well as common practice for IEEE 754-1985). */
+#define HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+
+#include_next <math_private.h>
+
+#endif
diff --git ports/sysdeps/unix/sysv/linux/aarch64/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/aarch64/nptl/libm.abilist
index 14b7edf..0f1cfc8 100644
--- ports/sysdeps/unix/sysv/linux/aarch64/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/aarch64/nptl/libm.abilist
@@ -395,3 +395,8 @@ GLIBC_2.17
yn F
ynf F
ynl F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
diff --git ports/sysdeps/unix/sysv/linux/alpha/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/alpha/nptl/libm.abilist
index 9c4170d..400a851 100644
--- ports/sysdeps/unix/sysv/linux/alpha/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/alpha/nptl/libm.abilist
@@ -391,6 +391,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
feclearexcept F
diff --git ports/sysdeps/unix/sysv/linux/arm/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/arm/nptl/libm.abilist
index cee3472..614e5eb 100644
--- ports/sysdeps/unix/sysv/linux/arm/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/arm/nptl/libm.abilist
@@ -54,6 +54,10 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
GLIBC_2.4
GLIBC_2.4 A
_LIB_VERSION D 0x4
diff --git ports/sysdeps/unix/sysv/linux/ia64/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/ia64/nptl/libm.abilist
index 1fb51a1..db8b279 100644
--- ports/sysdeps/unix/sysv/linux/ia64/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/ia64/nptl/libm.abilist
@@ -18,6 +18,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
_LIB_VERSION D 0x4
diff --git ports/sysdeps/unix/sysv/linux/m68k/coldfire/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/m68k/coldfire/nptl/libm.abilist
index cee3472..614e5eb 100644
--- ports/sysdeps/unix/sysv/linux/m68k/coldfire/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/m68k/coldfire/nptl/libm.abilist
@@ -54,6 +54,10 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
GLIBC_2.4
GLIBC_2.4 A
_LIB_VERSION D 0x4
diff --git ports/sysdeps/unix/sysv/linux/m68k/m680x0/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/m68k/m680x0/nptl/libm.abilist
index 35fc1e3..0beecb7 100644
--- ports/sysdeps/unix/sysv/linux/m68k/m680x0/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/m68k/m680x0/nptl/libm.abilist
@@ -396,6 +396,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
feclearexcept F
diff --git ports/sysdeps/unix/sysv/linux/mips/mips32/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/mips/mips32/nptl/libm.abilist
index defa1b0..5381246 100644
--- ports/sysdeps/unix/sysv/linux/mips/mips32/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/mips/mips32/nptl/libm.abilist
@@ -212,6 +212,10 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
GLIBC_2.2
GLIBC_2.2 A
__clog10 F
diff --git ports/sysdeps/unix/sysv/linux/mips/mips64/n32/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/mips/mips64/n32/nptl/libm.abilist
index 064eeba..bb39795 100644
--- ports/sysdeps/unix/sysv/linux/mips/mips64/n32/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/mips/mips64/n32/nptl/libm.abilist
@@ -239,6 +239,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
__clog10 F
diff --git ports/sysdeps/unix/sysv/linux/mips/mips64/n64/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/mips/mips64/n64/nptl/libm.abilist
index 064eeba..bb39795 100644
--- ports/sysdeps/unix/sysv/linux/mips/mips64/n64/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/mips/mips64/n64/nptl/libm.abilist
@@ -239,6 +239,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
__clog10 F
diff --git ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/nptl/libm.abilist
index ddc8576..9bd593c 100644
--- ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/nptl/libm.abilist
@@ -396,6 +396,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
feclearexcept F
diff --git ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/nptl/libm.abilist
index 601b505..cb0d1a4 100644
--- ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/nptl/libm.abilist
@@ -367,3 +367,7 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
diff --git ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/nptl/libm.abilist
index 601b505..cb0d1a4 100644
--- ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/nptl/libm.abilist
@@ -367,3 +367,7 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
diff --git ports/sysdeps/unix/sysv/linux/tile/tilepro/nptl/libm.abilist ports/sysdeps/unix/sysv/linux/tile/tilepro/nptl/libm.abilist
index 601b505..cb0d1a4 100644
--- ports/sysdeps/unix/sysv/linux/tile/tilepro/nptl/libm.abilist
+++ ports/sysdeps/unix/sysv/linux/tile/tilepro/nptl/libm.abilist
@@ -367,3 +367,7 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
diff --git sysdeps/ieee754/dbl-64/s_issignaling.c sysdeps/ieee754/dbl-64/s_issignaling.c
new file mode 100644
index 0000000..caa761e
--- /dev/null
+++ sysdeps/ieee754/dbl-64/s_issignaling.c
@@ -0,0 +1,46 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignaling
+int __issignaling (double x)
+{
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+ u_int32_t hxi;
+ GET_HIGH_WORD (hxi, x);
+ /* We only have to care about the high-order bit of x's significand, because
+ having it set (sNaN) already makes the significand different from that
+ used to designate infinity. */
+ return (hxi & 0x7ff80000) == 0x7ff80000;
+#else
+ u_int32_t hxi, lxi;
+ EXTRACT_WORDS (hxi, lxi, x);
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ hxi ^= 0x00080000;
+ /* If lxi != 0, then set any suitable bit of the significand in hxi. */
+ hxi |= (lxi | -lxi) >> 31;
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return (hxi & 0x7fffffff) > 0x7ff80000;
+#endif
+}
+libm_hidden_def (__issignaling)
diff --git sysdeps/ieee754/dbl-64/wordsize-64/s_issignaling.c sysdeps/ieee754/dbl-64/wordsize-64/s_issignaling.c
new file mode 100644
index 0000000..5de76d4
--- /dev/null
+++ sysdeps/ieee754/dbl-64/wordsize-64/s_issignaling.c
@@ -0,0 +1,43 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignaling
+int __issignaling (double x)
+{
+ u_int64_t xi;
+ EXTRACT_WORDS64 (xi, x);
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+# error untested
+ /* We only have to care about the high-order bit of x's significand, because
+ having it set (sNaN) already makes the significand different from that
+ used to designate infinity. */
+ return (xi & UINT64_C (0x7ff8000000000000)) == UINT64_C (0x7ff8000000000000);
+#else
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ xi ^= UINT64_C (0x0008000000000000);
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return (xi & UINT64_C (0x7fffffffffffffff)) > UINT64_C (0x7ff8000000000000);
+#endif
+}
+libm_hidden_def (__issignaling)
diff --git sysdeps/ieee754/flt-32/s_issignalingf.c sysdeps/ieee754/flt-32/s_issignalingf.c
new file mode 100644
index 0000000..2942002
--- /dev/null
+++ sysdeps/ieee754/flt-32/s_issignalingf.c
@@ -0,0 +1,42 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignalingf
+int __issignalingf (float x)
+{
+ u_int32_t xi;
+ GET_FLOAT_WORD (xi, x);
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+ /* We only have to care about the high-order bit of x's significand, because
+ having it set (sNaN) already makes the significand different from that
+ used to designate infinity. */
+ return (xi & 0x7fc00000) == 0x7fc00000;
+#else
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ xi ^= 0x00400000;
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return (xi & 0x7fffffff) > 0x7fc00000;
+#endif
+}
+libm_hidden_def (__issignalingf)
diff --git sysdeps/ieee754/ldbl-128/s_issignalingl.c sysdeps/ieee754/ldbl-128/s_issignalingl.c
new file mode 100644
index 0000000..873074d
--- /dev/null
+++ sysdeps/ieee754/ldbl-128/s_issignalingl.c
@@ -0,0 +1,46 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignalingl
+int __issignalingl (long double x)
+{
+ u_int64_t hxi, lxi __attribute__ ((unused));
+ GET_LDOUBLE_WORDS64 (hxi, lxi, x);
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+ /* We only have to care about the high-order bit of x's significand, because
+ having it set (sNaN) already makes the significand different from that
+ used to designate infinity. */
+ return ((hxi & UINT64_C (0x7fff800000000000))
+ == UINT64_C (0x7fff800000000000));
+#else
+# error untested
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ hxi ^= UINT64_C (0x0000800000000000);
+ /* If lxi != 0, then set any suitable bit of the significand in hxi. */
+ hxi |= (lxi | -lxi) >> 63;
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return (hxi & UINT64_C (0x7fffffffffffffff)) > UINT64_C (0x7fff800000000000);
+#endif
+}
+libm_hidden_def (__issignalingl)
diff --git sysdeps/ieee754/ldbl-128ibm/s_issignalingl.c sysdeps/ieee754/ldbl-128ibm/s_issignalingl.c
new file mode 100644
index 0000000..daead8c
--- /dev/null
+++ sysdeps/ieee754/ldbl-128ibm/s_issignalingl.c
@@ -0,0 +1,45 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignalingl
+int __issignalingl (long double x)
+{
+ u_int64_t xi;
+ /* For inspecting NaN status, we only have to look at the first of the pair
+ of IEEE 754 64-bit precision numbers. */
+ GET_LDOUBLE_MSW64 (xi, x);
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+# error untested
+ /* We only have to care about the high-order bit of x's significand, because
+ having it set (sNaN) already makes the significand different from that
+ used to designate infinity. */
+ return (xi & UINT64_C (0x7ff8000000000000)) == UINT64_C (0x7ff8000000000000);
+#else
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ xi ^= UINT64_C (0x0008000000000000);
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return (xi & UINT64_C (0x7fffffffffffffff)) > UINT64_C (0x7ff8000000000000);
+#endif
+}
+libm_hidden_def (__issignalingl)
diff --git sysdeps/ieee754/ldbl-96/s_issignalingl.c sysdeps/ieee754/ldbl-96/s_issignalingl.c
new file mode 100644
index 0000000..614e253
--- /dev/null
+++ sysdeps/ieee754/ldbl-96/s_issignalingl.c
@@ -0,0 +1,43 @@
+/* Test for signaling NaN.
+ Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_private.h>
+
+#undef __issignalingl
+int __issignalingl (long double x)
+{
+ u_int32_t exi, hxi, lxi;
+ GET_LDOUBLE_WORDS (exi, hxi, lxi, x);
+#ifdef HIGH_ORDER_BIT_IS_SET_FOR_SNAN
+# error not implemented
+#else
+ /* To keep the following comparison simple, toggle the quiet/signaling bit,
+ so that it is set for sNaNs. This is inverse to IEEE 754-2008 (as well as
+ common practice for IEEE 754-1985). */
+ hxi ^= 0x40000000;
+ /* If lxi != 0, then set any suitable bit of the significand in hxi. */
+ hxi |= (lxi | -lxi) >> 31;
+ /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and
+ later. */
+ /* We have to compare for greater (instead of greater or equal), because x's
+ significand being all-zero designates infinity not NaN. */
+ return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000);
+#endif
+}
+libm_hidden_def (__issignalingl)
diff --git sysdeps/unix/sysv/linux/i386/nptl/libm.abilist sysdeps/unix/sysv/linux/i386/nptl/libm.abilist
index 401a2c6..c185f0b 100644
--- sysdeps/unix/sysv/linux/i386/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/i386/nptl/libm.abilist
@@ -396,6 +396,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
__expl F
diff --git sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/nptl/libm.abilist sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/nptl/libm.abilist
index 620aff9..76a4ba3 100644
--- sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/nptl/libm.abilist
@@ -397,6 +397,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
feclearexcept F
diff --git sysdeps/unix/sysv/linux/powerpc/powerpc64/nptl/libm.abilist sysdeps/unix/sysv/linux/powerpc/powerpc64/nptl/libm.abilist
index 89422ab..d309a6f 100644
--- sysdeps/unix/sysv/linux/powerpc/powerpc64/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/powerpc/powerpc64/nptl/libm.abilist
@@ -81,6 +81,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.3
GLIBC_2.3 A
_LIB_VERSION D 0x4
diff --git sysdeps/unix/sysv/linux/s390/s390-32/nptl/libm.abilist sysdeps/unix/sysv/linux/s390/s390-32/nptl/libm.abilist
index ae8af50..f836c90 100644
--- sysdeps/unix/sysv/linux/s390/s390-32/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/s390/s390-32/nptl/libm.abilist
@@ -393,6 +393,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
fedisableexcept F
diff --git sysdeps/unix/sysv/linux/s390/s390-64/nptl/libm.abilist sysdeps/unix/sysv/linux/s390/s390-64/nptl/libm.abilist
index acf4d68..a0891ad 100644
--- sysdeps/unix/sysv/linux/s390/s390-64/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/s390/s390-64/nptl/libm.abilist
@@ -81,6 +81,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
_LIB_VERSION D 0x4
diff --git sysdeps/unix/sysv/linux/sh/nptl/libm.abilist sysdeps/unix/sysv/linux/sh/nptl/libm.abilist
index d851929..92821fd 100644
--- sysdeps/unix/sysv/linux/sh/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/sh/nptl/libm.abilist
@@ -54,6 +54,10 @@ GLIBC_2.15
__y1f_finite F
__yn_finite F
__ynf_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
GLIBC_2.2
GLIBC_2.2 A
_LIB_VERSION D 0x4
diff --git sysdeps/unix/sysv/linux/sparc/sparc32/nptl/libm.abilist sysdeps/unix/sysv/linux/sparc/sparc32/nptl/libm.abilist
index 4c6ec6b..9a1fcb1 100644
--- sysdeps/unix/sysv/linux/sparc/sparc32/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/sparc/sparc32/nptl/libm.abilist
@@ -388,6 +388,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
feclearexcept F
diff --git sysdeps/unix/sysv/linux/sparc/sparc64/nptl/libm.abilist sysdeps/unix/sysv/linux/sparc/sparc64/nptl/libm.abilist
index 22fb92a..2b41d34 100644
--- sysdeps/unix/sysv/linux/sparc/sparc64/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/sparc/sparc64/nptl/libm.abilist
@@ -81,6 +81,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2
GLIBC_2.2 A
_LIB_VERSION D 0x4
diff --git sysdeps/unix/sysv/linux/x86_64/64/nptl/libm.abilist sysdeps/unix/sysv/linux/x86_64/64/nptl/libm.abilist
index 7bf568c..2390934 100644
--- sysdeps/unix/sysv/linux/x86_64/64/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/x86_64/64/nptl/libm.abilist
@@ -81,6 +81,11 @@ GLIBC_2.15
__yn_finite F
__ynf_finite F
__ynl_finite F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
GLIBC_2.2.5
GLIBC_2.2.5 A
_LIB_VERSION D 0x4
diff --git sysdeps/unix/sysv/linux/x86_64/x32/nptl/libm.abilist sysdeps/unix/sysv/linux/x86_64/x32/nptl/libm.abilist
index 0f3ea4b..1825adb 100644
--- sysdeps/unix/sysv/linux/x86_64/x32/nptl/libm.abilist
+++ sysdeps/unix/sysv/linux/x86_64/x32/nptl/libm.abilist
@@ -395,3 +395,8 @@ GLIBC_2.16
yn F
ynf F
ynl F
+GLIBC_2.18
+ GLIBC_2.18 A
+ __issignaling F
+ __issignalingf F
+ __issignalingl F
Grüße,
Thomas
[-- Attachment #2: Type: application/pgp-signature, Size: 489 bytes --]
^ permalink raw reply [flat|nested] 8+ messages in thread