public inbox for libc-ports@sourceware.org
 help / color / mirror / Atom feed
From: Thomas Schwinge <thomas@codesourcery.com>
To: <libc-alpha@sourceware.org>, <libc-ports@sourceware.org>
Subject: [attention machine maintainers] [PATCH] <math.h> issignaling
Date: Fri, 22 Mar 2013 11:18:00 -0000	[thread overview]
Message-ID: <87r4j73cqo.fsf@schwinge.name> (raw)
In-Reply-To: <8762148l9v.fsf@schwinge.name>

[-- 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 --]

       reply	other threads:[~2013-03-22 11:18 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <8762148l9v.fsf@schwinge.name>
2013-03-22 11:18 ` Thomas Schwinge [this message]
2013-03-24 21:30   ` David Miller
2013-03-25  6:52     ` Thomas Schwinge
2013-03-25 18:39       ` David Miller
2013-03-25 15:54   ` Carlos O'Donell
2013-03-26 17:34   ` Joseph S. Myers
2013-03-28 12:16     ` Thomas Schwinge
2013-04-03 12:17       ` Thomas Schwinge

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=87r4j73cqo.fsf@schwinge.name \
    --to=thomas@codesourcery.com \
    --cc=libc-alpha@sourceware.org \
    --cc=libc-ports@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).