[PATCH v3] Provide a C++ version of issignaling that does not use __MATH_TG

[PATCH v3] Provide a C++ version of issignaling that does not use __MATH_TG

[Gabriel F. T. Gomes]

Changes since v2:

  - Added inline keyword to function definitions.
  - Replace __HAVE_FLOAT128 with __HAVE_DISTINCT_FLOAT128 in the test case.
  - Fixed various white-space errors.

Changes since v1:

  - Added test case.

-- 8< --
The macro __MATH_TG contains the logic to select between long double and
_Float128, when these types are ABI-distinct.  This logic relies on
__builtin_types_compatible_p, which is not available in C++ mode.

On the other hand, C++ function overloading provides the means to
distinguish between the floating-point types.  The overloading
resolution will match the correct parameter regardless of type
qualifiers, i.e.: const and volatile.

Tested for powerpc64le, s390x, and x86_64.

	* math/math.h [defined __cplusplus] (issignaling): Provide a C++
	definition for issignaling that does not rely on __MATH_TG,
	since __MATH_TG uses __builtin_types_compatible_p, which is only
	available in C mode.
	(CFLAGS-test-math-issignaling.cc): New variable.
	* math/Makefile [CXX] (tests): Add test-math-issignaling.
	* math/test-math-issignaling.cc: New test for C++ implementation
	of type-generic issignaling.
	* sysdeps/powerpc/powerpc64le/Makefile [subdir == math]
	(CXXFLAGS-test-math-issignaling.cc): Add -mfloat128 to the build
	options of test-math-issignaling on powerpc64le.
---
 math/Makefile                        |   3 +-
 math/math.h                          |  19 +++++-
 math/test-math-issignaling.cc        | 113 +++++++++++++++++++++++++++++++++++
 sysdeps/powerpc/powerpc64le/Makefile |   1 +
 4 files changed, 134 insertions(+), 2 deletions(-)
 create mode 100644 math/test-math-issignaling.cc

diff --git a/math/Makefile b/math/Makefile
index e09b0c0545..0130fcf38b 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -203,7 +203,7 @@ tests-static = test-fpucw-static test-fpucw-ieee-static \
 	       test-signgam-ullong-static test-signgam-ullong-init-static
 
 ifneq (,$(CXX))
-tests += test-math-isinff test-math-iszero
+tests += test-math-isinff test-math-iszero test-math-issignaling
 endif
 
 ifneq (no,$(PERL))
@@ -350,6 +350,7 @@ CFLAGS-test-signgam-ullong-init-static.c = -std=c99
 
 CFLAGS-test-math-isinff.cc = -std=gnu++11
 CFLAGS-test-math-iszero.cc = -std=gnu++11
+CFLAGS-test-math-issignaling.cc = -std=gnu++11
 
 CFLAGS-test-iszero-excess-precision.c = -fexcess-precision=standard
 CFLAGS-test-iseqsig-excess-precision.c = -fexcess-precision=standard
diff --git a/math/math.h b/math/math.h
index dea8dbe1ae..add86af724 100644
--- a/math/math.h
+++ b/math/math.h
@@ -474,7 +474,24 @@ enum
 # include <bits/iscanonical.h>
 
 /* Return nonzero value if X is a signaling NaN.  */
-# define issignaling(x) __MATH_TG ((x), __issignaling, (x))
+# ifndef __cplusplus
+#  define issignaling(x) __MATH_TG ((x), __issignaling, (x))
+# else
+   /* In C++ mode, __MATH_TG cannot be used, because it relies on
+      __builtin_types_compatible_p, which is a C-only builtin.  On the
+      other hand, overloading provides the means to distinguish between
+      the floating-point types.  The overloading resolution will match
+      the correct parameter (regardless of type qualifiers (i.e.: const
+      and volatile).  */
+extern "C++" {
+inline int issignaling (float __val) { return __issignalingf (__val); }
+inline int issignaling (double __val) { return __issignaling (__val); }
+inline int issignaling (long double __val) { return __issignalingl (__val); }
+#  if __HAVE_DISTINCT_FLOAT128
+inline int issignaling (_Float128 __val) { return __issignalingf128 (__val); }
+#  endif
+} /* extern C++ */
+# endif
 
 /* Return nonzero value if X is subnormal.  */
 # define issubnormal(x) (fpclassify (x) == FP_SUBNORMAL)
diff --git a/math/test-math-issignaling.cc b/math/test-math-issignaling.cc
new file mode 100644
index 0000000000..72b2a92ebe
--- /dev/null
+++ b/math/test-math-issignaling.cc
@@ -0,0 +1,113 @@
+/* Test for the C++ implementation of issignaling.
+   Copyright (C) 2017 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/>.  */
+
+#define _GNU_SOURCE 1
+#include <math.h>
+#include <stdio.h>
+
+#include <limits>
+
+/* There is no signaling_NaN for _Float128 in std::numeric_limits.
+   Include ieee754_float128.h and use the bitfields in the union
+   ieee854_float128.ieee_nan to build a signaling NaN.  */
+#if __HAVE_DISTINCT_FLOAT128
+# include <ieee754_float128.h>
+#endif
+
+static bool errors;
+
+static void
+check (int actual, int expected, const char *actual_expr, int line)
+{
+  if (actual != expected)
+    {
+      errors = true;
+      printf ("%s:%d: error: %s\n", __FILE__, line, actual_expr);
+      printf ("%s:%d:   expected: %d\n", __FILE__, line, expected);
+      printf ("%s:%d:   actual: %d\n", __FILE__, line, actual);
+    }
+}
+
+#define CHECK(actual, expected) \
+  check ((actual), (expected), #actual, __LINE__)
+
+template <class T>
+static void
+check_type ()
+{
+  typedef std::numeric_limits<T> limits;
+  CHECK (issignaling (T{0}), 0);
+  if (limits::has_infinity)
+    {
+      CHECK (issignaling (limits::infinity ()), 0);
+      CHECK (issignaling (-limits::infinity ()), 0);
+    }
+  if (limits::has_quiet_NaN)
+    CHECK (issignaling (limits::quiet_NaN ()), 0);
+  if (limits::has_signaling_NaN)
+    CHECK (issignaling (limits::signaling_NaN ()), 1);
+}
+
+#if __HAVE_DISTINCT_FLOAT128
+static void
+check_float128 ()
+{
+  ieee854_float128 q;
+
+  q.d = 0;
+  CHECK (issignaling (q.d), 0);
+
+  /* Infinity.  */
+  q.ieee.negative = 0;
+  q.ieee.exponent = 0x7FFF;
+  q.ieee.mantissa0 = 0x0000;
+  q.ieee.mantissa1 = 0x00000000;
+  q.ieee.mantissa2 = 0x00000000;
+  q.ieee.mantissa3 = 0x00000000;
+  CHECK (issignaling (q.d), 0);
+
+  /* Quite NaN.  */
+  q.ieee_nan.quiet_nan = 1;
+  q.ieee_nan.mantissa0 = 0x0000;
+  CHECK (issignaling (q.d), 0);
+
+  /* Still a quite NaN.  */
+  q.ieee_nan.quiet_nan = 1;
+  q.ieee_nan.mantissa0 = 0x4000;
+  CHECK (issignaling (q.d), 0);
+
+  /* Signaling NaN.  */
+  q.ieee_nan.quiet_nan = 0;
+  q.ieee_nan.mantissa0 = 0x4000;
+  CHECK (issignaling (q.d), 1);
+}
+#endif
+
+static int
+do_test (void)
+{
+  check_type<float> ();
+  check_type<double> ();
+  check_type<long double> ();
+#if __HAVE_DISTINCT_FLOAT128
+  check_float128 ();
+#endif
+  return errors;
+}
+
+#include <support/test-driver.c>
diff --git a/sysdeps/powerpc/powerpc64le/Makefile b/sysdeps/powerpc/powerpc64le/Makefile
index 77617b670a..19adbfa1c1 100644
--- a/sysdeps/powerpc/powerpc64le/Makefile
+++ b/sysdeps/powerpc/powerpc64le/Makefile
@@ -16,6 +16,7 @@ $(foreach suf,$(all-object-suffixes),%f128_r$(suf)): CFLAGS += -mfloat128
 $(foreach suf,$(all-object-suffixes),$(objpfx)test-float128%$(suf)): CFLAGS += -mfloat128
 $(foreach suf,$(all-object-suffixes),$(objpfx)test-ifloat128%$(suf)): CFLAGS += -mfloat128
 CFLAGS-libm-test-support-float128.c += -mfloat128
+CFLAGS-test-math-issignaling.cc += -mfloat128
 $(objpfx)test-float128% $(objpfx)test-ifloat128%: \
   gnulib-tests += $(f128-loader-link)
 endif
-- 
2.13.5

Re: [PATCH v3] Provide a C++ version of issignaling that does not use __MATH_TG

[Joseph Myers]

On Mon, 21 Aug 2017, Gabriel F. T. Gomes wrote:

> +  /* Quite NaN.  */

Quiet, not quite.

> +  /* Still a quite NaN.  */

Likewise.

OK with those changes.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: [PATCH v3] Provide a C++ version of issignaling that does not use __MATH_TG

[Joseph Myers]

This patch breaks building tests for long double = double configurations.  
__issignalingl does not exist in that case; the long double version of 
issignaling has to call __issignaling if __NO_LONG_DOUBLE_MATH is defined.

-- 
Joseph S. Myers
joseph@codesourcery.com

[PATCH] Fix the C++ version of issignaling when __NO_LONG_DOUBLE_MATH is defined

[Gabriel F. T. Gomes]

On Wed, 23 Aug 2017 00:41:23 +0000
Joseph Myers <joseph@codesourcery.com> wrote:

>This patch breaks building tests for long double = double configurations.
>__issignalingl does not exist in that case; the long double version of
>issignaling has to call __issignaling if __NO_LONG_DOUBLE_MATH is defined.

The following patch should fix this, however I do not have access to a
system where __NO_LONG_DOUBLE_MATH is defined.  I'm building this with
build-many-glibcs.py, but that will take some time.

OK for master?

-- 8< --
When __NO_LONG_DOUBLE_MATH is defined, __issignalingl is not available,
thus issignaling with long double argument should call __issignaling,
instead.

Tested for powerpc64le.

	* math/math.h [defined __cplusplus] (issignaling): In the long
	double case, call __issignalingl only if __NO_LONG_DOUBLE_MATH
	is not defined.  Call __issignaling, otherwise.
---
 math/math.h | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/math/math.h b/math/math.h
index b5d6c43fcf..5acbe88906 100644
--- a/math/math.h
+++ b/math/math.h
@@ -486,7 +486,15 @@ enum
 extern "C++" {
 inline int issignaling (float __val) { return __issignalingf (__val); }
 inline int issignaling (double __val) { return __issignaling (__val); }
-inline int issignaling (long double __val) { return __issignalingl (__val); }
+inline int
+issignaling (long double __val)
+{
+#  ifdef __NO_LONG_DOUBLE_MATH
+  return __issignaling (__val);
+#  else
+  return __issignalingl (__val);
+#  endif
+}
 #  if __HAVE_DISTINCT_FLOAT128
 inline int issignaling (_Float128 __val) { return __issignalingf128 (__val); }
 #  endif
-- 
2.13.5

Re: [PATCH] Fix the C++ version of issignaling when __NO_LONG_DOUBLE_MATH is defined

[Joseph Myers]

On Wed, 23 Aug 2017, Gabriel F. T. Gomes wrote:

> On Wed, 23 Aug 2017 00:41:23 +0000
> Joseph Myers <joseph@codesourcery.com> wrote:
> 
> >This patch breaks building tests for long double = double configurations.
> >__issignalingl does not exist in that case; the long double version of
> >issignaling has to call __issignaling if __NO_LONG_DOUBLE_MATH is defined.
> 
> The following patch should fix this, however I do not have access to a
> system where __NO_LONG_DOUBLE_MATH is defined.  I'm building this with
> build-many-glibcs.py, but that will take some time.
> 
> OK for master?

OK.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: [PATCH] Fix the C++ version of issignaling when __NO_LONG_DOUBLE_MATH is defined

[Gabriel F. T. Gomes]

On Thu, 24 Aug 2017 12:16:40 +0000
Joseph Myers <joseph@codesourcery.com> wrote:

>On Wed, 23 Aug 2017, Gabriel F. T. Gomes wrote:
>
>> The following patch should fix this, however I do not have access to a
>> system where __NO_LONG_DOUBLE_MATH is defined.  I'm building this with
>> build-many-glibcs.py, but that will take some time.
>> 
>> OK for master?  
>
>OK.

Thanks.  Pushed as 3d7b66f66cb2.

Running this with build-many-glibcs.py fixed the problem for the
mipsel-linux-gnu-soft configuration (which failed without the patch).

The other configurations are still running.  If other errors show up, I'll
come back to this.