Make iseqsig handle excess precision [committed]

[Joseph Myers <joseph at codesourcery dot com>]

iseqsig, like other type-generic comparison macros, should behave like
a comparison operator in not removing excess range and precision from
its arguments (see C11 F.10.11).  This patch implements this by making
definitions of iseqsig appropriately conditional on
__FLT_EVAL_METHOD__ (including support for TS 18661-3 values of that
macro), with a corresponding testcase (that failed for 32-bit x86 in
the absence of the math.h changes) being added.  (Of course the
definitions may need reworking when float128 support is added, just as
with other type-generic macros.)

Tested for x86_64 and x86.  Committed.

2016-10-07  Joseph Myers  <joseph@codesourcery.com>

	* math/math.h [__GLIBC_USE (IEC_60559_BFP_EXT)] (iseqsig): Define
	conditional on value of [__FLT_EVAL_METHOD__].
	* math/test-iseqsig-excess-precision.c: New file.
	* math/Makefile (tests): Add test-iseqsig-excess-precision.

diff --git a/math/Makefile b/math/Makefile
index a9f0608..ca07f9d 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -156,7 +156,8 @@ tests = test-matherr test-fenv atest-exp atest-sincos atest-exp2 basic-test \
 	test-signgam-ullong-init test-nan-overflow test-nan-payload \
 	test-fexcept test-fexcept-traps test-fesetexcept \
 	test-fesetexcept-traps test-fetestexceptflag test-femode \
-	test-femode-traps test-iszero-excess-precision $(tests-static)
+	test-femode-traps test-iszero-excess-precision \
+	test-iseqsig-excess-precision $(tests-static)
 tests-static = test-fpucw-static test-fpucw-ieee-static \
 	       test-signgam-uchar-static test-signgam-uchar-init-static \
 	       test-signgam-uint-static test-signgam-uint-init-static \
@@ -267,6 +268,7 @@ CFLAGS-test-signgam-ullong-init-static.c = -std=c99
 CFLAGS-test-math-isinff.cc = -std=gnu++11
 
 CFLAGS-test-iszero-excess-precision.c = -fexcess-precision=standard
+CFLAGS-test-iseqsig-excess-precision.c = -fexcess-precision=standard
 
 # The -lieee module sets the _LIB_VERSION_ switch to IEEE mode
 # for error handling in the -lm functions.
diff --git a/math/math.h b/math/math.h
index 394cefa..880b4a0 100644
--- a/math/math.h
+++ b/math/math.h
@@ -535,17 +535,32 @@ extern int matherr (struct exception *__exc);
 /* Return X == Y but raising "invalid" and setting errno if X or Y is
    a NaN.  */
 # ifdef __NO_LONG_DOUBLE_MATH
-#  define iseqsig(x, y)				\
+#  if (__FLT_EVAL_METHOD__ == 1			\
+       || __FLT_EVAL_METHOD__ == 2		\
+       || __FLT_EVAL_METHOD__ > 32)
+#   define iseqsig(x, y) __iseqsig ((x), (y))
+#  else
+#   define iseqsig(x, y)			\
   (sizeof ((x) + (y)) == sizeof (float)		\
    ? __iseqsigf ((x), (y))			\
    : __iseqsig ((x), (y)))
+#  endif
 # else
-#  define iseqsig(x, y)				\
+#  if __FLT_EVAL_METHOD__ == 2 || __FLT_EVAL_METHOD__ > 64
+#   define iseqsig(x, y) __iseqsigl ((x), (y))
+#  elif __FLT_EVAL_METHOD__ == 1 || __FLT_EVAL_METHOD__ > 32
+#   define iseqsig(x, y)			\
+  (sizeof ((x) + (y)) <= sizeof (double)	\
+   ? __iseqsig ((x), (y))			\
+   : __iseqsigl ((x), (y)))
+#  else
+#   define iseqsig(x, y)			\
   (sizeof ((x) + (y)) == sizeof (float)		\
    ? __iseqsigf ((x), (y))			\
    : sizeof ((x) + (y)) == sizeof (double)	\
    ? __iseqsig ((x), (y))			\
    : __iseqsigl ((x), (y)))
+#  endif
 # endif
 #endif
 
diff --git a/math/test-iseqsig-excess-precision.c b/math/test-iseqsig-excess-precision.c
new file mode 100644
index 0000000..01dea1c
--- /dev/null
+++ b/math/test-iseqsig-excess-precision.c
@@ -0,0 +1,80 @@
+/* Test iseqsig with excess precision.
+   Copyright (C) 2016 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 <float.h>
+#include <math.h>
+#include <stdio.h>
+
+static int
+do_test (void)
+{
+  int result = 0;
+
+  if (FLT_EVAL_METHOD == 1 || FLT_EVAL_METHOD == 2 || FLT_EVAL_METHOD > 32)
+    {
+      /* Excess precision for float.  */
+      if (iseqsig (1.0f, 1.0f + (float) DBL_EPSILON))
+	{
+	  puts ("iseqsig removes excess precision float -> double");
+	  result = 1;
+	}
+      else
+	puts ("iseqsig preserves excess precision float -> double");
+      if (iseqsig (__builtin_inff (), FLT_MAX * FLT_MAX))
+	{
+	  puts ("iseqsig removes excess range float -> double");
+	  result = 1;
+	}
+      else
+	puts ("iseqsig preserves excess range float -> double");
+    }
+
+  if (FLT_EVAL_METHOD == 2 || FLT_EVAL_METHOD > 64)
+    {
+      /* Excess precision for float and double.  */
+      if (iseqsig (1.0f, 1.0f + (float) LDBL_EPSILON))
+	{
+	  puts ("iseqsig removes excess precision float -> long double");
+	  result = 1;
+	}
+      else
+	puts ("iseqsig preserves excess precision float -> long double");
+      if (iseqsig (1.0, 1.0 + (double) LDBL_EPSILON))
+	{
+	  puts ("iseqsig removes excess precision double -> long double");
+	  result = 1;
+	}
+      else
+	puts ("iseqsig preserves excess precision double -> long double");
+      if (LDBL_MAX_EXP >= 2 * DBL_MAX_EXP)
+	{
+	  if (iseqsig (__builtin_inf (), DBL_MAX * DBL_MAX))
+	    {
+	      puts ("iseqsig removes excess range double -> long double");
+	      result = 1;
+	    }
+	    else
+	      puts ("iseqsig preserves excess range double -> long double");
+	}
+    }
+
+  return result;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"

-- 
Joseph S. Myers
joseph@codesourcery.com