about summary refs log tree commit diff
path: root/sysdeps/ieee754
diff options
context:
space:
mode:
authorJoseph Myers <joseph@codesourcery.com>2016-12-21 01:48:27 +0000
committerJoseph Myers <joseph@codesourcery.com>2016-12-21 01:48:27 +0000
commit41c67149b94676347b0068a902058f130e3e88e6 (patch)
treeb4f50fdb48fd83914ad570b581188b74ab90e984 /sysdeps/ieee754
parent60c07b058f9fe34e2ffd7e02b20657db6f5f68d4 (diff)
downloadglibc-41c67149b94676347b0068a902058f130e3e88e6.tar.gz
glibc-41c67149b94676347b0068a902058f130e3e88e6.tar.xz
glibc-41c67149b94676347b0068a902058f130e3e88e6.zip
Add roundeven, roundevenf, roundevenl.
TS 18661-1 defines roundeven functions that round a floating-point
number to the nearest integer, in that floating-point type, with ties
rounding to even (whereas the round functions round ties away from
zero).  As with other such functions, they raise no exceptions apart
from "invalid" for signaling NaNs.  There was a previous user request
for this functionality in glibc in
<https://sourceware.org/ml/libc-help/2015-02/msg00005.html>.

This patch implements these functions for glibc.  The implementations
use integer bit-manipulation (or roundeven on the high and low parts,
in the IBM long double case).  It's possible that there may be faster
approaches on some architectures (in particular, on AArch64 the frintn
instruction should do exactly what's required); I'll leave it to
architecture maintainers or others interested to implement such
architecture-specific versions if desired.  (Where architectures have
instructions to round to nearest integer in the current rounding mode,
implementations saving and restoring the rounding mode - and dealing
with exceptions if those instructions generate "inexact" - are also
possible, though their performance depends on the cost of manipulating
exceptions / rounding mode state.)

Tested for x86_64, x86, mips64 and powerpc.

	* math/bits/mathcalls.h [__GLIBC_USE (IEC_60559_BFP_EXT)]
	(roundeven): New declaration.
	* math/tgmath.h [__GLIBC_USE (IEC_60559_BFP_EXT)] (roundeven): New
	macro.
	* math/Versions (roundeven): New libm symbol at version
	GLIBC_2.25.
	(roundevenf): Likewise.
	(roundevenl): Likewise.
	* math/Makefile (libm-calls): Add s_roundevenF.
	* math/libm-test.inc (roundeven_test_data): New array.
	(roundeven_test): New function.
	(main): Call roundeven_test.
	* math/test-tgmath.c (NCALLS): Increase to 134.
	(F(compile_test)): Call roundeven.
	(F(roundeven)): New function.
	* manual/arith.texi (Rounding Functions): Document roundeven,
	roundevenf and roundevenl.
	* manual/libm-err-tab.pl (@all_functions): Add roundeven.
	* include/math.h (roundeven): Use libm_hidden_proto.
	* sysdeps/ieee754/dbl-64/s_roundeven.c: New file.
	* sysdeps/ieee754/dbl-64/wordsize-64/s_roundeven.c: Likewise.
	* sysdeps/ieee754/flt-32/s_roundevenf.c: Likewise.
	* sysdeps/ieee754/ldbl-128/s_roundevenl.c: Likewise.
	* sysdeps/ieee754/ldbl-128ibm/s_roundevenl.c: Likewise.
	* sysdeps/ieee754/ldbl-96/s_roundevenl.c: Likewise.
	* sysdeps/ieee754/ldbl-opt/Makefile (libnldbl-calls): Add
	roundeven.
	(CFLAGS-nldbl-roundeven.c): New variable.
	* sysdeps/ieee754/ldbl-opt/nldbl-roundeven.c: New file.
	* sysdeps/nacl/libm.abilist: Update.
	* sysdeps/unix/sysv/linux/aarch64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/alpha/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/arm/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/hppa/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/i386/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/ia64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/m68k/coldfire/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/m68k/m680x0/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/microblaze/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/mips/mips32/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/mips/mips64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/nios2/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libm.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libm.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc64/libm-le.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/powerpc/powerpc64/libm.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/s390/s390-32/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/s390/s390-64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/sh/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/sparc/sparc32/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/sparc/sparc64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/libm.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/libm.abilist:
	Likewise.
	* sysdeps/unix/sysv/linux/tile/tilepro/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/x32/libm.abilist: Likewise.
Diffstat (limited to 'sysdeps/ieee754')
-rw-r--r--sysdeps/ieee754/dbl-64/s_roundeven.c106
-rw-r--r--sysdeps/ieee754/dbl-64/wordsize-64/s_roundeven.c72
-rw-r--r--sysdeps/ieee754/flt-32/s_roundevenf.c68
-rw-r--r--sysdeps/ieee754/ldbl-128/s_roundevenl.c102
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_roundevenl.c69
-rw-r--r--sysdeps/ieee754/ldbl-96/s_roundevenl.c124
-rw-r--r--sysdeps/ieee754/ldbl-opt/Makefile4
-rw-r--r--sysdeps/ieee754/ldbl-opt/nldbl-roundeven.c26
8 files changed, 570 insertions, 1 deletions
diff --git a/sysdeps/ieee754/dbl-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/s_roundeven.c
new file mode 100644
index 0000000000..27e15e35ad
--- /dev/null
+++ b/sysdeps/ieee754/dbl-64/s_roundeven.c
@@ -0,0 +1,106 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   dbl-64 version.
+   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 <math.h>
+#include <math_private.h>
+#include <stdint.h>
+
+#define BIAS 0x3ff
+#define MANT_DIG 53
+#define MAX_EXP (2 * BIAS + 1)
+
+double
+roundeven (double x)
+{
+  uint32_t hx, lx, uhx;
+  EXTRACT_WORDS (hx, lx, x);
+  uhx = hx & 0x7fffffff;
+  int exponent = uhx >> (MANT_DIG - 1 - 32);
+  if (exponent >= BIAS + MANT_DIG - 1)
+    {
+      /* Integer, infinity or NaN.  */
+      if (exponent == MAX_EXP)
+	/* Infinity or NaN; quiet signaling NaNs.  */
+	return x + x;
+      else
+	return x;
+    }
+  else if (exponent >= BIAS + MANT_DIG - 32)
+    {
+      /* Not necessarily an integer; integer bit is in low word.
+	 Locate the bits with exponents 0 and -1.  */
+      int int_pos = (BIAS + MANT_DIG - 1) - exponent;
+      int half_pos = int_pos - 1;
+      uint32_t half_bit = 1U << half_pos;
+      uint32_t int_bit = 1U << int_pos;
+      if ((lx & (int_bit | (half_bit - 1))) != 0)
+	{
+	  /* Carry into the exponent works correctly.  No need to test
+	     whether HALF_BIT is set.  */
+	  lx += half_bit;
+	  hx += lx < half_bit;
+	}
+      lx &= ~(int_bit - 1);
+    }
+  else if (exponent == BIAS + MANT_DIG - 33)
+    {
+      /* Not necessarily an integer; integer bit is bottom of high
+	 word, half bit is top of low word.  */
+      if (((hx & 1) | (lx & 0x7fffffff)) != 0)
+	{
+	  lx += 0x80000000;
+	  hx += lx < 0x80000000;
+	}
+      lx = 0;
+    }
+  else if (exponent >= BIAS)
+    {
+      /* At least 1; not necessarily an integer, integer bit and half
+	 bit are in the high word.  Locate the bits with exponents 0
+	 and -1 (when the unbiased exponent is 0, the bit with
+	 exponent 0 is implicit, but as the bias is odd it is OK to
+	 take it from the low bit of the exponent).  */
+      int int_pos = (BIAS + MANT_DIG - 33) - exponent;
+      int half_pos = int_pos - 1;
+      uint32_t half_bit = 1U << half_pos;
+      uint32_t int_bit = 1U << int_pos;
+      if (((hx & (int_bit | (half_bit - 1))) | lx) != 0)
+	hx += half_bit;
+      hx &= ~(int_bit - 1);
+      lx = 0;
+    }
+  else if (exponent == BIAS - 1 && (uhx > 0x3fe00000 || lx != 0))
+    {
+      /* Interval (0.5, 1).  */
+      hx = (hx & 0x80000000) | 0x3ff00000;
+      lx = 0;
+    }
+  else
+    {
+      /* Rounds to 0.  */
+      hx &= 0x80000000;
+      lx = 0;
+    }
+  INSERT_WORDS (x, hx, lx);
+  return x;
+}
+hidden_def (roundeven)
+#ifdef NO_LONG_DOUBLE
+weak_alias (roundeven, roundevenl)
+#endif
diff --git a/sysdeps/ieee754/dbl-64/wordsize-64/s_roundeven.c b/sysdeps/ieee754/dbl-64/wordsize-64/s_roundeven.c
new file mode 100644
index 0000000000..1fa3ef319d
--- /dev/null
+++ b/sysdeps/ieee754/dbl-64/wordsize-64/s_roundeven.c
@@ -0,0 +1,72 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   dbl-64/wordsize-64 version.
+   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 <math.h>
+#include <math_private.h>
+#include <stdint.h>
+
+#define BIAS 0x3ff
+#define MANT_DIG 53
+#define MAX_EXP (2 * BIAS + 1)
+
+double
+roundeven (double x)
+{
+  uint64_t ix, ux;
+  EXTRACT_WORDS64 (ix, x);
+  ux = ix & 0x7fffffffffffffffULL;
+  int exponent = ux >> (MANT_DIG - 1);
+  if (exponent >= BIAS + MANT_DIG - 1)
+    {
+      /* Integer, infinity or NaN.  */
+      if (exponent == MAX_EXP)
+	/* Infinity or NaN; quiet signaling NaNs.  */
+	return x + x;
+      else
+	return x;
+    }
+  else if (exponent >= BIAS)
+    {
+      /* At least 1; not necessarily an integer.  Locate the bits with
+	 exponents 0 and -1 (when the unbiased exponent is 0, the bit
+	 with exponent 0 is implicit, but as the bias is odd it is OK
+	 to take it from the low bit of the exponent).  */
+      int int_pos = (BIAS + MANT_DIG - 1) - exponent;
+      int half_pos = int_pos - 1;
+      uint64_t half_bit = 1ULL << half_pos;
+      uint64_t int_bit = 1ULL << int_pos;
+      if ((ix & (int_bit | (half_bit - 1))) != 0)
+	/* Carry into the exponent works correctly.  No need to test
+	   whether HALF_BIT is set.  */
+	ix += half_bit;
+      ix &= ~(int_bit - 1);
+    }
+  else if (exponent == BIAS - 1 && ux > 0x3fe0000000000000ULL)
+    /* Interval (0.5, 1).  */
+    ix = (ix & 0x8000000000000000ULL) | 0x3ff0000000000000ULL;
+  else
+    /* Rounds to 0.  */
+    ix &= 0x8000000000000000ULL;
+  INSERT_WORDS64 (x, ix);
+  return x;
+}
+hidden_def (roundeven)
+#ifdef NO_LONG_DOUBLE
+weak_alias (roundeven, roundevenl)
+#endif
diff --git a/sysdeps/ieee754/flt-32/s_roundevenf.c b/sysdeps/ieee754/flt-32/s_roundevenf.c
new file mode 100644
index 0000000000..db2e0487d6
--- /dev/null
+++ b/sysdeps/ieee754/flt-32/s_roundevenf.c
@@ -0,0 +1,68 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   flt-32 version.
+   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 <math.h>
+#include <math_private.h>
+#include <stdint.h>
+
+#define BIAS 0x7f
+#define MANT_DIG 24
+#define MAX_EXP (2 * BIAS + 1)
+
+float
+roundevenf (float x)
+{
+  uint32_t ix, ux;
+  GET_FLOAT_WORD (ix, x);
+  ux = ix & 0x7fffffff;
+  int exponent = ux >> (MANT_DIG - 1);
+  if (exponent >= BIAS + MANT_DIG - 1)
+    {
+      /* Integer, infinity or NaN.  */
+      if (exponent == MAX_EXP)
+	/* Infinity or NaN; quiet signaling NaNs.  */
+	return x + x;
+      else
+	return x;
+    }
+  else if (exponent >= BIAS)
+    {
+      /* At least 1; not necessarily an integer.  Locate the bits with
+	 exponents 0 and -1 (when the unbiased exponent is 0, the bit
+	 with exponent 0 is implicit, but as the bias is odd it is OK
+	 to take it from the low bit of the exponent).  */
+      int int_pos = (BIAS + MANT_DIG - 1) - exponent;
+      int half_pos = int_pos - 1;
+      uint32_t half_bit = 1U << half_pos;
+      uint32_t int_bit = 1U << int_pos;
+      if ((ix & (int_bit | (half_bit - 1))) != 0)
+	/* Carry into the exponent works correctly.  No need to test
+	   whether HALF_BIT is set.  */
+	ix += half_bit;
+      ix &= ~(int_bit - 1);
+    }
+  else if (exponent == BIAS - 1 && ux > 0x3f000000)
+    /* Interval (0.5, 1).  */
+    ix = (ix & 0x80000000) | 0x3f800000;
+  else
+    /* Rounds to 0.  */
+    ix &= 0x80000000;
+  SET_FLOAT_WORD (x, ix);
+  return x;
+}
diff --git a/sysdeps/ieee754/ldbl-128/s_roundevenl.c b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
new file mode 100644
index 0000000000..de042acf2d
--- /dev/null
+++ b/sysdeps/ieee754/ldbl-128/s_roundevenl.c
@@ -0,0 +1,102 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   ldbl-128 version.
+   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 <math.h>
+#include <math_private.h>
+#include <stdint.h>
+
+#define BIAS 0x3fff
+#define MANT_DIG 113
+#define MAX_EXP (2 * BIAS + 1)
+
+_Float128
+roundevenl (_Float128 x)
+{
+  uint64_t hx, lx, uhx;
+  GET_LDOUBLE_WORDS64 (hx, lx, x);
+  uhx = hx & 0x7fffffffffffffffULL;
+  int exponent = uhx >> (MANT_DIG - 1 - 64);
+  if (exponent >= BIAS + MANT_DIG - 1)
+    {
+      /* Integer, infinity or NaN.  */
+      if (exponent == MAX_EXP)
+	/* Infinity or NaN; quiet signaling NaNs.  */
+	return x + x;
+      else
+	return x;
+    }
+  else if (exponent >= BIAS + MANT_DIG - 64)
+    {
+      /* Not necessarily an integer; integer bit is in low word.
+	 Locate the bits with exponents 0 and -1.  */
+      int int_pos = (BIAS + MANT_DIG - 1) - exponent;
+      int half_pos = int_pos - 1;
+      uint64_t half_bit = 1ULL << half_pos;
+      uint64_t int_bit = 1ULL << int_pos;
+      if ((lx & (int_bit | (half_bit - 1))) != 0)
+	{
+	  /* Carry into the exponent works correctly.  No need to test
+	     whether HALF_BIT is set.  */
+	  lx += half_bit;
+	  hx += lx < half_bit;
+	}
+      lx &= ~(int_bit - 1);
+    }
+  else if (exponent == BIAS + MANT_DIG - 65)
+    {
+      /* Not necessarily an integer; integer bit is bottom of high
+	 word, half bit is top of low word.  */
+      if (((hx & 1) | (lx & 0x7fffffffffffffffULL)) != 0)
+	{
+	  lx += 0x8000000000000000ULL;
+	  hx += lx < 0x8000000000000000ULL;
+	}
+      lx = 0;
+    }
+  else if (exponent >= BIAS)
+    {
+      /* At least 1; not necessarily an integer, integer bit and half
+	 bit are in the high word.  Locate the bits with exponents 0
+	 and -1 (when the unbiased exponent is 0, the bit with
+	 exponent 0 is implicit, but as the bias is odd it is OK to
+	 take it from the low bit of the exponent).  */
+      int int_pos = (BIAS + MANT_DIG - 65) - exponent;
+      int half_pos = int_pos - 1;
+      uint64_t half_bit = 1ULL << half_pos;
+      uint64_t int_bit = 1ULL << int_pos;
+      if (((hx & (int_bit | (half_bit - 1))) | lx) != 0)
+	hx += half_bit;
+      hx &= ~(int_bit - 1);
+      lx = 0;
+    }
+  else if (exponent == BIAS - 1 && (uhx > 0x3ffe000000000000ULL || lx != 0))
+    {
+      /* Interval (0.5, 1).  */
+      hx = (hx & 0x8000000000000000ULL) | 0x3fff000000000000ULL;
+      lx = 0;
+    }
+  else
+    {
+      /* Rounds to 0.  */
+      hx &= 0x8000000000000000ULL;
+      lx = 0;
+    }
+  SET_LDOUBLE_WORDS64 (x, hx, lx);
+  return x;
+}
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_roundevenl.c b/sysdeps/ieee754/ldbl-128ibm/s_roundevenl.c
new file mode 100644
index 0000000000..cd91d3905c
--- /dev/null
+++ b/sysdeps/ieee754/ldbl-128ibm/s_roundevenl.c
@@ -0,0 +1,69 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   ldbl-128ibm version.
+   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 <math.h>
+#include <math_private.h>
+
+long double
+roundevenl (long double x)
+{
+  double xh, xl, hi;
+
+  ldbl_unpack (x, &xh, &xl);
+
+  if (xh != 0 && isfinite (xh))
+    {
+      hi = roundeven (xh);
+      if (hi != xh)
+	{
+	  /* The high part is not an integer; the low part only
+	     affects the result if the high part is exactly half way
+	     between two integers and the low part is nonzero in the
+	     opposite direction to the rounding of the high part.  */
+	  double diff = hi - xh;
+	  if (fabs (diff) == 0.5)
+	    {
+	      if (xl < 0 && diff > 0)
+		xh = hi - 1;
+	      else if (xl > 0 && diff < 0)
+		xh = hi + 1;
+	      else
+		xh = hi;
+	    }
+	  else
+	    xh = hi;
+	  xl = 0;
+	}
+      else
+	{
+	  /* The high part is a nonzero integer.  Rounding the low
+	     part to nearest, ties round to even, is always correct,
+	     as a high part that is an odd integer together with a low
+	     part with magnitude 0.5 is not a valid long double.  */
+	  xl = roundeven (xl);
+	  xh = hi;
+	  ldbl_canonicalize_int (&xh, &xl);
+	}
+    }
+  else
+    /* Quiet signaling NaN arguments.  */
+    xh += xh;
+
+  return ldbl_pack (xh, xl);
+}
diff --git a/sysdeps/ieee754/ldbl-96/s_roundevenl.c b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
new file mode 100644
index 0000000000..f635b1f683
--- /dev/null
+++ b/sysdeps/ieee754/ldbl-96/s_roundevenl.c
@@ -0,0 +1,124 @@
+/* Round to nearest integer value, rounding halfway cases to even.
+   ldbl-96 version.
+   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 <math.h>
+#include <math_private.h>
+#include <stdint.h>
+
+#define BIAS 0x3fff
+#define MANT_DIG 64
+#define MAX_EXP (2 * BIAS + 1)
+
+long double
+roundevenl (long double x)
+{
+  uint16_t se;
+  uint32_t hx, lx;
+  GET_LDOUBLE_WORDS (se, hx, lx, x);
+  int exponent = se & 0x7fff;
+  if (exponent >= BIAS + MANT_DIG - 1)
+    {
+      /* Integer, infinity or NaN.  */
+      if (exponent == MAX_EXP)
+	/* Infinity or NaN; quiet signaling NaNs.  */
+	return x + x;
+      else
+	return x;
+    }
+  else if (exponent >= BIAS + MANT_DIG - 32)
+    {
+      /* Not necessarily an integer; integer bit is in low word.
+	 Locate the bits with exponents 0 and -1.  */
+      int int_pos = (BIAS + MANT_DIG - 1) - exponent;
+      int half_pos = int_pos - 1;
+      uint32_t half_bit = 1U << half_pos;
+      uint32_t int_bit = 1U << int_pos;
+      if ((lx & (int_bit | (half_bit - 1))) != 0)
+	{
+	  /* No need to test whether HALF_BIT is set.  */
+	  lx += half_bit;
+	  if (lx < half_bit)
+	    {
+	      hx++;
+	      if (hx == 0)
+		{
+		  hx = 0x80000000;
+		  se++;
+		}
+	    }
+	}
+      lx &= ~(int_bit - 1);
+    }
+  else if (exponent == BIAS + MANT_DIG - 33)
+    {
+      /* Not necessarily an integer; integer bit is bottom of high
+	 word, half bit is top of low word.  */
+      if (((hx & 1) | (lx & 0x7fffffff)) != 0)
+	{
+	  lx += 0x80000000;
+	  if (lx < 0x80000000)
+	    {
+	      hx++;
+	      if (hx == 0)
+		{
+		  hx = 0x80000000;
+		  se++;
+		}
+	    }
+	}
+      lx = 0;
+    }
+  else if (exponent >= BIAS)
+    {
+      /* At least 1; not necessarily an integer, integer bit and half
+	 bit are in the high word.  Locate the bits with exponents 0
+	 and -1.  */
+      int int_pos = (BIAS + MANT_DIG - 33) - exponent;
+      int half_pos = int_pos - 1;
+      uint32_t half_bit = 1U << half_pos;
+      uint32_t int_bit = 1U << int_pos;
+      if (((hx & (int_bit | (half_bit - 1))) | lx) != 0)
+	{
+	  hx += half_bit;
+	  if (hx < half_bit)
+	    {
+	      hx = 0x80000000;
+	      se++;
+	    }
+	}
+      hx &= ~(int_bit - 1);
+      lx = 0;
+    }
+  else if (exponent == BIAS - 1 && (hx > 0x80000000 || lx != 0))
+    {
+      /* Interval (0.5, 1).  */
+      se = (se & 0x8000) | 0x3fff;
+      hx = 0x80000000;
+      lx = 0;
+    }
+  else
+    {
+      /* Rounds to 0.  */
+      se &= 0x8000;
+      hx = 0;
+      lx = 0;
+    }
+  SET_LDOUBLE_WORDS (x, se, hx, lx);
+  return x;
+}
diff --git a/sysdeps/ieee754/ldbl-opt/Makefile b/sysdeps/ieee754/ldbl-opt/Makefile
index 99aa6c6e0b..006d0d59ee 100644
--- a/sysdeps/ieee754/ldbl-opt/Makefile
+++ b/sysdeps/ieee754/ldbl-opt/Makefile
@@ -43,7 +43,8 @@ libnldbl-calls = asprintf dprintf fprintf fscanf fwprintf fwscanf iovfscanf \
 		 isoc99_wscanf isoc99_fwscanf isoc99_swscanf \
 		 isoc99_vwscanf isoc99_vfwscanf isoc99_vswscanf \
 		 nextup nextdown totalorder totalordermag getpayload \
-		 canonicalize setpayload setpayloadsig llogb fmaxmag fminmag
+		 canonicalize setpayload setpayloadsig llogb fmaxmag fminmag \
+		 roundeven
 libnldbl-routines = $(libnldbl-calls:%=nldbl-%)
 libnldbl-inhibit-o = $(object-suffixes)
 libnldbl-static-only-routines = $(libnldbl-routines)
@@ -138,6 +139,7 @@ CFLAGS-nldbl-remainder.c = -fno-builtin-remainderl -fno-builtin-dreml
 CFLAGS-nldbl-remquo.c = -fno-builtin-remquol
 CFLAGS-nldbl-rint.c = -fno-builtin-rintl
 CFLAGS-nldbl-round.c = -fno-builtin-roundl
+CFLAGS-nldbl-roundeven.c = -fno-builtin-roundevenl
 CFLAGS-nldbl-scalb.c = -fno-builtin-scalbl
 CFLAGS-nldbl-scalbln.c = -fno-builtin-scalblnl
 CFLAGS-nldbl-scalbn.c = -fno-builtin-scalbnl
diff --git a/sysdeps/ieee754/ldbl-opt/nldbl-roundeven.c b/sysdeps/ieee754/ldbl-opt/nldbl-roundeven.c
new file mode 100644
index 0000000000..cd52e7376f
--- /dev/null
+++ b/sysdeps/ieee754/ldbl-opt/nldbl-roundeven.c
@@ -0,0 +1,26 @@
+/* Compatibility routine for IEEE double as long double for roundeven.
+   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 "nldbl-compat.h"
+
+double
+attribute_hidden
+roundevenl (double x)
+{
+  return roundeven (x);
+}