Merge changes between r20391 and r20863 from /fsf/trunk.

git-svn-id: svn://svn.eglibc.org/trunk@20864 7b3dc134-2b1b-0410-93df-9e9f96275f8d

46 files changed, 2679 insertions, 429 deletions

diff --git a/libc/sysdeps/generic/_G_config.h b/libc/sysdeps/generic/_G_config.h
index 6f756829e..fc83565b9 100644
--- a/libc/sysdeps/generic/_G_config.h
+++ b/libc/sysdeps/generic/_G_config.h
@@ -18,7 +18,6 @@
 # define __need_wint_t
 #endif
 #include <wchar.h>
-#define _G_size_t	size_t
 typedef struct
 {
   __off_t __pos;
@@ -29,13 +28,7 @@ typedef struct
   __off64_t __pos;
   __mbstate_t __state;
 } _G_fpos64_t;
-#define _G_ssize_t	__ssize_t
-#define _G_off_t	__off_t
 #define _G_off64_t	__off_t
-#define	_G_pid_t	__pid_t
-#define	_G_uid_t	__uid_t
-#define _G_wchar_t	wchar_t
-#define _G_wint_t	wint_t
 #define _G_stat64	stat
 #if defined _LIBC || defined _GLIBCPP_USE_WCHAR_T
 # include <gconv.h>
@@ -50,25 +43,11 @@ typedef union
 } _G_iconv_t;
 #endif
 
-typedef int _G_int16_t __attribute__ ((__mode__ (__HI__)));
-typedef int _G_int32_t __attribute__ ((__mode__ (__SI__)));
-typedef unsigned int _G_uint16_t __attribute__ ((__mode__ (__HI__)));
-typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
-
-#define _G_HAVE_BOOL 1
-
 
 /* These library features are always available in the GNU C library.  */
-#define _G_HAVE_ATEXIT 1
-#define _G_HAVE_SYS_WAIT 1
-#define _G_NEED_STDARG_H 1
 #define _G_va_list __gnuc_va_list
 
-#define _G_HAVE_PRINTF_FP 1
 #define _G_HAVE_MMAP 1
-#define _G_HAVE_LONG_DOUBLE_IO 1
-#define _G_HAVE_IO_FILE_OPEN 1
-#define _G_HAVE_IO_GETLINE_INFO 1
 
 #define _G_IO_IO_FILE_VERSION 0x20001
 
@@ -77,9 +56,4 @@ typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
 
 #define _G_BUFSIZ 8192
 
-/* These are the vtbl details for ELF.  */
-#define _G_NAMES_HAVE_UNDERSCORE 0
-#define _G_VTABLE_LABEL_PREFIX "_vt."
-#define _G_VTABLE_LABEL_HAS_LENGTH 1
-
 #endif	/* _G_config.h */
diff --git a/libc/sysdeps/generic/get-rounding-mode.h b/libc/sysdeps/generic/get-rounding-mode.h
new file mode 100644
index 000000000..0ecaddd7b
--- /dev/null
+++ b/libc/sysdeps/generic/get-rounding-mode.h
@@ -0,0 +1,124 @@
+/* Determine floating-point rounding mode within libc.  Generic version.
+   Copyright (C) 2012 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 _GET_ROUNDING_MODE_H
+#define _GET_ROUNDING_MODE_H	1
+
+#include <fpu_control.h>
+
+/* Define values for FE_* modes not defined for this architecture.  */
+#ifdef FE_DOWNWARD
+# define ORIG_FE_DOWNWARD FE_DOWNWARD
+#else
+# define ORIG_FE_DOWNWARD 0
+#endif
+#ifdef FE_TONEAREST
+# define ORIG_FE_TONEAREST FE_TONEAREST
+#else
+# define ORIG_FE_TONEAREST 0
+#endif
+#ifdef FE_TOWARDZERO
+# define ORIG_FE_TOWARDZERO FE_TOWARDZERO
+#else
+# define ORIG_FE_TOWARDZERO 0
+#endif
+#ifdef FE_UPWARD
+# define ORIG_FE_UPWARD FE_UPWARD
+#else
+# define ORIG_FE_UPWARD 0
+#endif
+#define FE_CONSTRUCT_DISTINCT_VALUE(X, Y, Z) \
+  ((((X) & 1) | ((Y) & 2) | ((Z) & 4)) ^ 7)
+#ifndef FE_DOWNWARD
+# define FE_DOWNWARD FE_CONSTRUCT_DISTINCT_VALUE (ORIG_FE_TONEAREST,	\
+						  ORIG_FE_TOWARDZERO,	\
+						  ORIG_FE_UPWARD)
+#endif
+#ifndef FE_TONEAREST
+# define FE_TONEAREST FE_CONSTRUCT_DISTINCT_VALUE (FE_DOWNWARD,		\
+						   ORIG_FE_TOWARDZERO,	\
+						   ORIG_FE_UPWARD)
+#endif
+#ifndef FE_TOWARDZERO
+# define FE_TOWARDZERO FE_CONSTRUCT_DISTINCT_VALUE (FE_DOWNWARD,	\
+						    FE_TONEAREST,	\
+						    ORIG_FE_UPWARD)
+#endif
+#ifndef FE_UPWARD
+# define FE_UPWARD FE_CONSTRUCT_DISTINCT_VALUE (FE_DOWNWARD,	\
+						FE_TONEAREST,	\
+						FE_TOWARDZERO)
+#endif
+
+/* Return the floating-point rounding mode.  */
+
+static inline int
+get_rounding_mode (void)
+{
+#if (defined _FPU_RC_DOWN			\
+     || defined _FPU_RC_NEAREST			\
+     || defined _FPU_RC_ZERO			\
+     || defined _FPU_RC_UP)
+  fpu_control_t fc;
+  const fpu_control_t mask = (0
+# ifdef _FPU_RC_DOWN
+			      | _FPU_RC_DOWN
+# endif
+# ifdef _FPU_RC_NEAREST
+			      | _FPU_RC_NEAREST
+# endif
+# ifdef _FPU_RC_ZERO
+			      | _FPU_RC_ZERO
+# endif
+# ifdef _FPU_RC_UP
+			      | _FPU_RC_UP
+# endif
+			      );
+
+  _FPU_GETCW (fc);
+  switch (fc & mask)
+    {
+# ifdef _FPU_RC_DOWN
+    case _FPU_RC_DOWN:
+      return FE_DOWNWARD;
+# endif
+
+# ifdef _FPU_RC_NEAREST
+    case _FPU_RC_NEAREST:
+      return FE_TONEAREST;
+# endif
+
+# ifdef _FPU_RC_ZERO
+    case _FPU_RC_ZERO:
+      return FE_TOWARDZERO;
+# endif
+
+# ifdef _FPU_RC_UP
+    case _FPU_RC_UP:
+      return FE_UPWARD;
+# endif
+
+    default:
+      abort ();
+    }
+#else
+  return FE_TONEAREST;
+#endif
+}
+
+#endif /* get-rounding-mode.h */
diff --git a/libc/sysdeps/generic/malloc-sysdep.h b/libc/sysdeps/generic/malloc-sysdep.h
new file mode 100644
index 000000000..bbc48c04f
--- /dev/null
+++ b/libc/sysdeps/generic/malloc-sysdep.h
@@ -0,0 +1,25 @@
+/* System-specific malloc support functions.  Generic version.
+   Copyright (C) 2012 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/>.  */
+
+/* Force an unmap when the heap shrinks in a secure exec.  This ensures that
+   the old data pages immediately cease to be accessible.  */
+static inline bool
+check_may_shrink_heap (void)
+{
+  return __libc_enable_secure;
+}
diff --git a/libc/sysdeps/gnu/_G_config.h b/libc/sysdeps/gnu/_G_config.h
index 9d6e23bb4..4d3ae36c4 100644
--- a/libc/sysdeps/gnu/_G_config.h
+++ b/libc/sysdeps/gnu/_G_config.h
@@ -18,7 +18,6 @@
 # define __need_wint_t
 #endif
 #include <wchar.h>
-#define _G_size_t	size_t
 typedef struct
 {
   __off_t __pos;
@@ -29,13 +28,7 @@ typedef struct
   __off64_t __pos;
   __mbstate_t __state;
 } _G_fpos64_t;
-#define _G_ssize_t	__ssize_t
-#define _G_off_t	__off_t
 #define _G_off64_t	__off64_t
-#define	_G_pid_t	__pid_t
-#define	_G_uid_t	__uid_t
-#define _G_wchar_t	wchar_t
-#define _G_wint_t	wint_t
 #define _G_stat64	stat64
 #if defined _LIBC || defined _GLIBCPP_USE_WCHAR_T
 # include <gconv.h>
@@ -50,26 +43,12 @@ typedef union
 } _G_iconv_t;
 #endif
 
-typedef int _G_int16_t __attribute__ ((__mode__ (__HI__)));
-typedef int _G_int32_t __attribute__ ((__mode__ (__SI__)));
-typedef unsigned int _G_uint16_t __attribute__ ((__mode__ (__HI__)));
-typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
-
-#define _G_HAVE_BOOL 1
-
 
 /* These library features are always available in the GNU C library.  */
-#define _G_HAVE_ATEXIT 1
-#define _G_HAVE_SYS_WAIT 1
-#define _G_NEED_STDARG_H 1
 #define _G_va_list __gnuc_va_list
 
-#define _G_HAVE_PRINTF_FP 1
 #define _G_HAVE_MMAP 1
 #define _G_HAVE_MREMAP 1
-#define _G_HAVE_LONG_DOUBLE_IO 1
-#define _G_HAVE_IO_FILE_OPEN 1
-#define _G_HAVE_IO_GETLINE_INFO 1
 
 #define _G_IO_IO_FILE_VERSION 0x20001
 
@@ -83,11 +62,4 @@ typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
 
 #define _G_BUFSIZ 8192
 
-/* These are the vtbl details for ELF.  */
-#define _G_NAMES_HAVE_UNDERSCORE 0
-#define _G_VTABLE_LABEL_HAS_LENGTH 1
-#define _G_USING_THUNKS	1
-#define _G_VTABLE_LABEL_PREFIX "__vt_"
-#define _G_VTABLE_LABEL_PREFIX_ID __vt_
-
 #endif	/* _G_config.h */
diff --git a/libc/sysdeps/i386/fpu/libm-test-ulps b/libc/sysdeps/i386/fpu/libm-test-ulps
index abd98a295..d35635ce0 100644
--- a/libc/sysdeps/i386/fpu/libm-test-ulps
+++ b/libc/sysdeps/i386/fpu/libm-test-ulps
@@ -2398,6 +2398,10 @@ ldouble: 1
 Test "sin (-0x1p65) == 0.047183876212354673805106149805700013943218":
 float: 1
 ifloat: 1
+Test "sin (0x1.fffff8p+127) == 4.85786063130487339701113680434728152037092e-02":
+float: 1
+Test "sin (0x1p+127) == 6.23385512955870240370428801097126489001833e-01":
+float: 1
 Test "sin (0x1p65) == -0.047183876212354673805106149805700013943218":
 float: 1
 ifloat: 1
@@ -2556,6 +2560,13 @@ ifloat: 1
 Test "sincos (-0x1p65, &sin_res, &cos_res) puts 0.047183876212354673805106149805700013943218 in sin_res":
 float: 1
 ifloat: 1
+Test "sincos (0x1.fffff8p+127, &sin_res, &cos_res) puts 4.85786063130487339701113680434728152037092e-02 in sin_res":
+float: 1
+Test "sincos (0x1p+127, &sin_res, &cos_res) puts 6.23385512955870240370428801097126489001833e-01 in sin_res":
+float: 1
+Test "sincos (0x1p+50, &sin_res, &cos_res) puts 8.68095904660550604334592502063501320395739e-01 in cos_res":
+float: 1
+ifloat: 1
 Test "sincos (0x1p65, &sin_res, &cos_res) puts -0.047183876212354673805106149805700013943218 in sin_res":
 float: 1
 ifloat: 1
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/Makefile b/libc/sysdeps/i386/i686/fpu/multiarch/Makefile
index 1de37fe4a..aa28f7284 100644
--- a/libc/sysdeps/i386/i686/fpu/multiarch/Makefile
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/Makefile
@@ -1,3 +1,3 @@
 ifeq ($(subdir),math)
-libm-sysdep_routines += e_expf-sse2 e_expf-ia32
+libm-sysdep_routines += e_expf-sse2 e_expf-ia32 s_sinf-sse2 s_cosf-sse2
 endif
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/e_expf.c b/libc/sysdeps/i386/i686/fpu/multiarch/e_expf.c
index 65858a311..1966400d1 100644
--- a/libc/sysdeps/i386/i686/fpu/multiarch/e_expf.c
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/e_expf.c
@@ -1,3 +1,21 @@
+/* Multiple versions of expf
+   Copyright (C) 2012 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 <init-arch.h>
 
 extern double __ieee754_expf_sse2 (double);
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf-sse2.S b/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf-sse2.S
new file mode 100644
index 000000000..405c6ea4c
--- /dev/null
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf-sse2.S
@@ -0,0 +1,553 @@
+/* Optimized with sse2 version of cosf
+   Copyright (C) 2012 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 <sysdep.h>
+#define __need_Emath
+#include <bits/errno.h>
+
+/* Short algorithm description:
+ *
+ *  1) if |x| == 0: return 1.0-|x|.
+ *  2) if |x| <  2^-27: return 1.0-|x|.
+ *  3) if |x| <  2^-5 : return 1.0+x^2*DP_COS2_0+x^5*DP_COS2_1.
+ *  4) if |x| <   Pi/4: return 1.0+x^2*(C0+x^2*(C1+x^2*(C2+x^2*(C3+x^2*C4)))).
+ *  5) if |x| < 9*Pi/4:
+ *      5.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0x0e, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      5.2) Reconstruction:
+ *          s = (-1.0)^((n>>2)&1)
+ *          if(n&2 != 0) {
+ *              using cos(t) polynomial for |t|<Pi/4, result is
+ *              s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4))))).
+ *          } else {
+ *              using sin(t) polynomial for |t|<Pi/4, result is
+ *              s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4))))).
+ *          }
+ *  6) if |x| < 2^23, large args:
+ *      6.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      6.2) Reconstruction same as (5.2).
+ *  7) if |x| >= 2^23, very large args:
+ *      7.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      7.2) Reconstruction same as (5.2).
+ *  8) if x is Inf, return x-x, and set errno=EDOM.
+ *  9) if x is NaN, return x-x.
+ *
+ * Special cases:
+ *  cos(+-0) = 1 not raising inexact,
+ *  cos(subnormal) raises inexact,
+ *  cos(min_normalized) raises inexact,
+ *  cos(normalized) raises inexact,
+ *  cos(Inf) = NaN, raises invalid, sets errno to EDOM,
+ *  cos(NaN) = NaN.
+ */
+
+#ifdef	PIC
+# define MO1(symbol)			L(symbol)##@GOTOFF(%ebx)
+# define MO2(symbol,reg2,_scale)	L(symbol)##@GOTOFF(%ebx,reg2,_scale)
+# define CFI_PUSH(REG)	cfi_adjust_cfa_offset(4); cfi_rel_offset(REG,0)
+# define CFI_POP(REG)	cfi_adjust_cfa_offset(-4); cfi_restore(REG)
+# define PUSH(REG)			pushl REG; CFI_PUSH(REG)
+# define POP(REG)			popl REG; CFI_POP(REG)
+# define ENTRANCE			PUSH(%ebx); LOAD_PIC_REG(bx)
+# define RETURN				POP(%ebx); ret; CFI_PUSH(%ebx)
+# define ARG_X				8(%esp)
+#else
+# define MO1(symbol)			L(symbol)
+# define MO2(symbol,reg2,_scale)	L(symbol)(,reg2,_scale)
+# define ENTRANCE
+# define RETURN				ret
+# define ARG_X				4(%esp)
+#endif
+
+	.text
+ENTRY(__cosf_sse2)
+	/* Input: single precision x on stack at address ARG_X */
+
+	ENTRANCE
+	movl	ARG_X, %eax		/* Bits of x */
+	cvtss2sd ARG_X, %xmm0		/* DP x */
+	andl	$0x7fffffff, %eax	/* |x| */
+
+	cmpl	$0x3f490fdb, %eax	/* |x|<Pi/4?  */
+	jb	L(arg_less_pio4)
+
+	/* Here if |x|>=Pi/4 */
+	movd	%eax, %xmm3		/* SP |x| */
+	andpd	MO1(DP_ABS_MASK),%xmm0	/* DP |x| */
+	movss	MO1(SP_INVPIO4), %xmm2	/* SP 1/(Pi/4) */
+
+	cmpl	$0x40e231d6, %eax	/* |x|<9*Pi/4?  */
+	jae	L(large_args)
+
+	/* Here if Pi/4<=|x|<9*Pi/4 */
+	mulss	%xmm3, %xmm2		/* SP |x|/(Pi/4) */
+	cvttss2si %xmm2, %eax		/* k, number of Pi/4 in x */
+	addl	$1, %eax		/* k+1 */
+	movl	$0x0e, %edx
+	andl	%eax, %edx		/* j = (k+1)&0x0e */
+	addl	$2, %eax		/* n */
+	subsd	MO2(PIO4J,%edx,8), %xmm0 /* t = |x| - j * Pi/4 */
+
+L(reconstruction):
+	/* Input: %eax=n, %xmm0=t */
+	testl	$2, %eax		/* n&2 != 0?  */
+	jz	L(sin_poly)
+
+/*L(cos_poly):*/
+	/* Here if cos(x) calculated using cos(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))))
+	 */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	MO1(DP_C4), %xmm4	/* C4 */
+	mulsd	%xmm0, %xmm4		/* z*C4 */
+	movsd	MO1(DP_C3), %xmm3	/* C3 */
+	mulsd	%xmm0, %xmm3		/* z*C3 */
+	addsd	MO1(DP_C2), %xmm4	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm4		/* z*(C2+z*C4) */
+	lea	-8(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	addsd	MO1(DP_C1), %xmm3	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm3		/* z*(C1+z*C3) */
+	addsd	MO1(DP_C0), %xmm4	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm4		/* y*(C0+z*(C2+z*C4)) */
+
+	addsd	%xmm4, %xmm3		/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	/* 1.0+y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	MO1(DP_ONES), %xmm3
+
+	mulsd	MO2(DP_ONES,%eax,8), %xmm3 /* DP result */
+	movsd	%xmm3, 0(%esp)		/* Move result from sse...  */
+	fldl	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	8(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(sin_poly):
+	/* Here if cos(x) calculated using sin(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))))
+	 */
+
+	movaps	%xmm0, %xmm4		/* t */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	MO1(DP_S4), %xmm2	/* S4 */
+	mulsd	%xmm0, %xmm2		/* z*S4 */
+	movsd	MO1(DP_S3), %xmm3	/* S3 */
+	mulsd	%xmm0, %xmm3		/* z*S3 */
+	lea	-8(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	addsd	MO1(DP_S2), %xmm2	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm2		/* z*(S2+z*S4) */
+	addsd	MO1(DP_S1), %xmm3	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm3		/* z*(S1+z*S3) */
+	addsd	MO1(DP_S0), %xmm2	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm2		/* y*(S0+z*(S2+z*S4)) */
+	/* t*s, where s = sign(x) * (-1.0)^((n>>2)&1) */
+	mulsd	MO2(DP_ONES,%eax,8), %xmm4
+	addsd	%xmm2, %xmm3		/* y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	/* t*s*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	mulsd	%xmm4, %xmm3
+	/* t*s*(1.0+y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	movsd	%xmm3, 0(%esp)		/* Move result from sse...   */
+	fldl	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	8(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(large_args):
+	/* Here if |x|>=9*Pi/4 */
+	cmpl	$0x7f800000, %eax	/* x is Inf or NaN?  */
+	jae	L(arg_inf_or_nan)
+
+	/* Here if finite |x|>=9*Pi/4 */
+	cmpl	$0x4b000000, %eax	/* |x|<2^23?  */
+	jae	L(very_large_args)
+
+	/* Here if 9*Pi/4<=|x|<2^23 */
+	movsd	MO1(DP_INVPIO4), %xmm1	/* 1/(Pi/4) */
+	mulsd	%xmm0, %xmm1		/* |x|/(Pi/4) */
+	cvttsd2si %xmm1, %eax		/* k=trunc(|x|/(Pi/4)) */
+	addl	$1, %eax		/* k+1 */
+	movl	%eax, %edx
+	andl	$0xfffffffe, %edx	/* j=(k+1)&0xfffffffe */
+	cvtsi2sdl %edx, %xmm4		/* DP j */
+	movsd	MO1(DP_PIO4HI), %xmm2	/* -PIO4HI = high part of -Pi/4 */
+	mulsd	%xmm4, %xmm2		/* -j*PIO4HI */
+	movsd	MO1(DP_PIO4LO), %xmm3	/* -PIO4LO = low part of -Pi/4 */
+	addsd	%xmm2, %xmm0		/* |x| - j*PIO4HI */
+	addl	$2, %eax		/* n */
+	mulsd	%xmm3, %xmm4		/* j*PIO4LO */
+	addsd	%xmm4, %xmm0		/* t = |x| - j*PIO4HI - j*PIO4LO */
+	jmp	L(reconstruction)
+
+	.p2align	4
+L(very_large_args):
+	/* Here if finite |x|>=2^23 */
+
+	/* bitpos = (ix>>23) - BIAS_32 + 59; */
+	shrl	$23, %eax		/* eb = biased exponent of x */
+	/* bitpos = eb - 0x7f + 59, where 0x7f is exponent bias */
+	subl	$68, %eax
+	movl	$28, %ecx		/* %cl=28 */
+	movl	%eax, %edx		/* bitpos copy */
+
+	/* j = bitpos/28; */
+	div	%cl			/* j in register %al=%ax/%cl */
+	movapd	%xmm0, %xmm3		/* |x| */
+	/* clear unneeded remainder from %ah */
+	andl	$0xff, %eax
+
+	imull	$28, %eax, %ecx		/* j*28 */
+	movsd	MO1(DP_HI_MASK), %xmm4	/* DP_HI_MASK */
+	movapd	%xmm0, %xmm5		/* |x| */
+	mulsd	-2*8+MO2(_FPI,%eax,8), %xmm3	/* tmp3 = FPI[j-2]*|x| */
+	movapd	%xmm0, %xmm1		/* |x| */
+	mulsd	-1*8+MO2(_FPI,%eax,8), %xmm5	/* tmp2 = FPI[j-1]*|x| */
+	mulsd	0*8+MO2(_FPI,%eax,8), %xmm0	/* tmp0 = FPI[j]*|x| */
+	addl	$19, %ecx		/* j*28+19 */
+	mulsd	1*8+MO2(_FPI,%eax,8), %xmm1	/* tmp1 = FPI[j+1]*|x| */
+	cmpl	%ecx, %edx		/* bitpos>=j*28+19?  */
+	jl	L(very_large_skip1)
+
+	/* Here if bitpos>=j*28+19 */
+	andpd	%xmm3, %xmm4		/* HI(tmp3) */
+	subsd	%xmm4, %xmm3		/* tmp3 = tmp3 - HI(tmp3) */
+L(very_large_skip1):
+
+	movsd	MO1(DP_2POW52), %xmm6
+	movapd	%xmm5, %xmm2		/* tmp2 copy */
+	addsd	%xmm3, %xmm5		/* tmp5 = tmp3 + tmp2 */
+	movl	$1, %edx
+	addsd	%xmm5, %xmm6		/* tmp6 = tmp5 + 2^52 */
+	movsd	8+MO1(DP_2POW52), %xmm4
+	movd	%xmm6, %eax		/* k = I64_LO(tmp6); */
+	addsd	%xmm6, %xmm4		/* tmp4 = tmp6 - 2^52 */
+	comisd	%xmm5, %xmm4		/* tmp4 > tmp5?  */
+	jbe	L(very_large_skip2)
+
+	/* Here if tmp4 > tmp5 */
+	subl	$1, %eax		/* k-- */
+	addsd	8+MO1(DP_ONES), %xmm4	/* tmp4 -= 1.0 */
+L(very_large_skip2):
+
+	andl	%eax, %edx		/* k&1 */
+	subsd	%xmm4, %xmm3		/* tmp3 -= tmp4 */
+	addsd	MO2(DP_ZERONE,%edx,8), %xmm3 /* t  = DP_ZERONE[k&1] + tmp3 */
+	addsd	%xmm2, %xmm3		/* t += tmp2 */
+	addsd	%xmm3, %xmm0		/* t += tmp0 */
+	addl	$3, %eax		/* n=k+3 */
+	addsd	%xmm1, %xmm0		/* t += tmp1 */
+	mulsd	MO1(DP_PIO4), %xmm0	/* t *= PI04 */
+
+	jmp	L(reconstruction)	/* end of very_large_args peth */
+
+	.p2align	4
+L(arg_less_pio4):
+	/* Here if |x|<Pi/4 */
+	cmpl	$0x3d000000, %eax	/* |x|<2^-5?  */
+	jl	L(arg_less_2pn5)
+
+	/* Here if 2^-5<=|x|<Pi/4 */
+	mulsd	%xmm0, %xmm0		/* y=x^2 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=x^4 */
+	movsd	MO1(DP_C4), %xmm3	/* C4 */
+	mulsd	%xmm0, %xmm3		/* z*C4 */
+	movsd	MO1(DP_C3), %xmm5	/* C3 */
+	mulsd	%xmm0, %xmm5		/* z*C3 */
+	addsd	MO1(DP_C2), %xmm3	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm3		/* z*(C2+z*C4) */
+	addsd	MO1(DP_C1), %xmm5	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm5		/* z*(C1+z*C3) */
+	addsd	MO1(DP_C0), %xmm3	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm3		/* y*(C0+z*(C2+z*C4)) */
+	addsd	%xmm5, %xmm3		/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	/* 1.0 + y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	MO1(DP_ONES), %xmm3
+	cvtsd2ss %xmm3, %xmm3		/* SP result */
+
+L(epilogue):
+	lea	-4(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	movss	%xmm3, 0(%esp)		/* Move result from sse...  */
+	flds	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	4(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(arg_less_2pn5):
+	/* Here if |x|<2^-5 */
+	cmpl	$0x32000000, %eax	/* |x|<2^-27?  */
+	jl	L(arg_less_2pn27)
+
+	/* Here if 2^-27<=|x|<2^-5 */
+	mulsd	%xmm0, %xmm0		/* DP x^2 */
+	movsd	MO1(DP_COS2_1), %xmm3	/* DP DP_COS2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_COS2_1 */
+	addsd	MO1(DP_COS2_0), %xmm3	/* DP DP_COS2_0+x^2*DP_COS2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_COS2_0+x^4*DP_COS2_1 */
+	/* DP 1.0+x^2*DP_COS2_0+x^4*DP_COS2_1 */
+	addsd	MO1(DP_ONES), %xmm3
+	cvtsd2ss %xmm3, %xmm3		/* SP result */
+	jmp	L(epilogue)
+
+	.p2align	4
+L(arg_less_2pn27):
+	/* Here if |x|<2^-27 */
+	movss	ARG_X, %xmm0		/* x */
+	andps	MO1(SP_ABS_MASK),%xmm0	/* |x| */
+	movss	MO1(SP_ONE), %xmm3	/* 1.0 */
+	subss	%xmm0, %xmm3		/* result is 1.0-|x| */
+	jmp	L(epilogue)
+
+	.p2align	4
+L(arg_inf_or_nan):
+	/* Here if |x| is Inf or NAN */
+	jne	L(skip_errno_setting)	/* in case of x is NaN */
+
+	/* Here if x is Inf. Set errno to EDOM.  */
+	call	JUMPTARGET(__errno_location)
+	movl	$EDOM, (%eax)
+
+	.p2align	4
+L(skip_errno_setting):
+	/* Here if |x| is Inf or NAN. Continued.  */
+	movss	ARG_X, %xmm3		/* load x */
+	subss	%xmm3, %xmm3		/* Result is NaN */
+	jmp	L(epilogue)
+END(__cosf_sse2)
+
+	.section .rodata, "a"
+	.p2align 3
+L(PIO4J): /* Table of j*Pi/4, for j=0,1,..,10 */
+	.long	0x00000000,0x00000000
+	.long	0x54442d18,0x3fe921fb
+	.long	0x54442d18,0x3ff921fb
+	.long	0x7f3321d2,0x4002d97c
+	.long	0x54442d18,0x400921fb
+	.long	0x2955385e,0x400f6a7a
+	.long	0x7f3321d2,0x4012d97c
+	.long	0xe9bba775,0x4015fdbb
+	.long	0x54442d18,0x401921fb
+	.long	0xbeccb2bb,0x401c463a
+	.long	0x2955385e,0x401f6a7a
+	.type L(PIO4J), @object
+	ASM_SIZE_DIRECTIVE(L(PIO4J))
+
+	.p2align 3
+L(_FPI): /* 4/Pi broken into sum of positive DP values */
+	.long	0x00000000,0x00000000
+	.long	0x6c000000,0x3ff45f30
+	.long	0x2a000000,0x3e3c9c88
+	.long	0xa8000000,0x3c54fe13
+	.long	0xd0000000,0x3aaf47d4
+	.long	0x6c000000,0x38fbb81b
+	.long	0xe0000000,0x3714acc9
+	.long	0x7c000000,0x3560e410
+	.long	0x56000000,0x33bca2c7
+	.long	0xac000000,0x31fbd778
+	.long	0xe0000000,0x300b7246
+	.long	0xe8000000,0x2e5d2126
+	.long	0x48000000,0x2c970032
+	.long	0xe8000000,0x2ad77504
+	.long	0xe0000000,0x290921cf
+	.long	0xb0000000,0x274deb1c
+	.long	0xe0000000,0x25829a73
+	.long	0xbe000000,0x23fd1046
+	.long	0x10000000,0x2224baed
+	.long	0x8e000000,0x20709d33
+	.long	0x80000000,0x1e535a2f
+	.long	0x64000000,0x1cef904e
+	.long	0x30000000,0x1b0d6398
+	.long	0x24000000,0x1964ce7d
+	.long	0x16000000,0x17b908bf
+	.type L(_FPI), @object
+	ASM_SIZE_DIRECTIVE(L(_FPI))
+
+/* Coefficients of polynomial
+ for cos(x)~=1.0+x^2*DP_COS2_0+x^4*DP_COS2_1, |x|<2^-5.  */
+	.p2align 3
+L(DP_COS2_0):
+	.long	0xff5cc6fd,0xbfdfffff
+	.type L(DP_COS2_0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_COS2_0))
+
+	.p2align 3
+L(DP_COS2_1):
+	.long	0xb178dac5,0x3fa55514
+	.type L(DP_COS2_1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_COS2_1))
+
+	.p2align 3
+L(DP_ZERONE):
+	.long	0x00000000,0x00000000	/* 0.0 */
+	.long	0x00000000,0xbff00000	/* 1.0 */
+	.type L(DP_ZERONE),@object
+	ASM_SIZE_DIRECTIVE(L(DP_ZERONE))
+
+	.p2align 3
+L(DP_ONES):
+	.long	0x00000000,0x3ff00000	/* +1.0 */
+	.long	0x00000000,0xbff00000	/* -1.0 */
+	.type L(DP_ONES), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ONES))
+
+/* Coefficients of polynomial
+ for sin(t)~=t+t^3*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_S3):
+	.long	0x64e6b5b4,0x3ec71d72
+	.type L(DP_S3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S3))
+
+	.p2align 3
+L(DP_S1):
+	.long	0x10c2688b,0x3f811111
+	.type L(DP_S1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S1))
+
+	.p2align 3
+L(DP_S4):
+	.long	0x1674b58a,0xbe5a947e
+	.type L(DP_S4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S4))
+
+	.p2align 3
+L(DP_S2):
+	.long	0x8b4bd1f9,0xbf2a019f
+	.type L(DP_S2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S2))
+
+	.p2align 3
+L(DP_S0):
+	.long	0x55551cd9,0xbfc55555
+	.type L(DP_S0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S0))
+
+/* Coefficients of polynomial
+ for cos(t)~=1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_C3):
+	.long	0x9ac43cc0,0x3efa00eb
+	.type L(DP_C3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C3))
+
+	.p2align 3
+L(DP_C1):
+	.long	0x545c50c7,0x3fa55555
+	.type L(DP_C1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C1))
+
+	.p2align 3
+L(DP_C4):
+	.long	0xdd8844d7,0xbe923c97
+	.type L(DP_C4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C4))
+
+	.p2align 3
+L(DP_C2):
+	.long	0x348b6874,0xbf56c16b
+	.type L(DP_C2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C2))
+
+	.p2align 3
+L(DP_C0):
+	.long	0xfffe98ae,0xbfdfffff
+	.type L(DP_C0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C0))
+
+	.p2align 3
+L(DP_PIO4):
+	.long	0x54442d18,0x3fe921fb	/* Pi/4 */
+	.type L(DP_PIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4))
+
+	.p2align 3
+L(DP_2POW52):
+	.long	0x00000000,0x43300000	/* +2^52 */
+	.long	0x00000000,0xc3300000	/* -2^52 */
+	.type L(DP_2POW52), @object
+	ASM_SIZE_DIRECTIVE(L(DP_2POW52))
+
+	.p2align 3
+L(DP_INVPIO4):
+	.long	0x6dc9c883,0x3ff45f30	/* 4/Pi */
+	.type L(DP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_INVPIO4))
+
+	.p2align 3
+L(DP_PIO4HI):
+	.long	0x54000000,0xbfe921fb	/* High part of Pi/4 */
+	.type L(DP_PIO4HI), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4HI))
+
+	.p2align 3
+L(DP_PIO4LO):
+	.long	0x11A62633,0xbe010b46	/* Low part of Pi/4 */
+	.type L(DP_PIO4LO), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4LO))
+
+	.p2align 2
+L(SP_INVPIO4):
+	.long	0x3fa2f983		/* 4/Pi */
+	.type L(SP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(SP_INVPIO4))
+
+	.p2align 4
+L(DP_ABS_MASK): /* Mask for getting DP absolute value */
+	.long	0xffffffff,0x7fffffff
+	.long	0xffffffff,0x7fffffff
+	.type L(DP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ABS_MASK))
+
+	.p2align 3
+L(DP_HI_MASK): /* Mask for getting high 21 bits of DP value */
+	.long	0x00000000,0xffffffff
+	.type L(DP_HI_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_HI_MASK))
+
+	.p2align 4
+L(SP_ABS_MASK): /* Mask for getting SP absolute value */
+	.long	0x7fffffff,0x7fffffff
+	.long	0x7fffffff,0x7fffffff
+	.type L(SP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(SP_ABS_MASK))
+
+	.p2align 2
+L(SP_ONE):
+	.long	0x3f800000		/* 1.0 */
+	.type L(SP_ONE), @object
+	ASM_SIZE_DIRECTIVE(L(SP_ONE))
+
+weak_alias (__cosf, cosf)
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf.c b/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf.c
new file mode 100644
index 000000000..f07c3e510
--- /dev/null
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/s_cosf.c
@@ -0,0 +1,29 @@
+/* Multiple versions of cosf
+   Copyright (C) 2012 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 <init-arch.h>
+
+extern float __cosf_sse2 (float);
+extern float __cosf_ia32 (float);
+float __cosf (float);
+
+libm_ifunc (__cosf, HAS_SSE2 ? __cosf_sse2 : __cosf_ia32);
+weak_alias (__cosf, cosf);
+
+#define COSF __cosf_ia32
+#include <sysdeps/ieee754/flt-32/s_cosf.c>
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf-sse2.S b/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf-sse2.S
new file mode 100644
index 000000000..49d59b597
--- /dev/null
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf-sse2.S
@@ -0,0 +1,566 @@
+/* Optimized with sse2 version of sinf
+   Copyright (C) 2012 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 <sysdep.h>
+#define __need_Emath
+#include <bits/errno.h>
+
+/* Short algorithm description:
+ *
+ *  1) if |x| == 0: return x.
+ *  2) if |x| <  2^-27: return x-x*DP_SMALL, raise underflow only when needed.
+ *  3) if |x| <  2^-5 : return x+x^3*DP_SIN2_0+x^5*DP_SIN2_1.
+ *  4) if |x| <   Pi/4: return x+x^3*(S0+x^2*(S1+x^2*(S2+x^2*(S3+x^2*S4)))).
+ *  5) if |x| < 9*Pi/4:
+ *      5.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0x0e, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      5.2) Reconstruction:
+ *          s = sign(x) * (-1.0)^((n>>2)&1)
+ *          if(n&2 != 0) {
+ *              using cos(t) polynomial for |t|<Pi/4, result is
+ *              s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4))))).
+ *          } else {
+ *              using sin(t) polynomial for |t|<Pi/4, result is
+ *              s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4))))).
+ *          }
+ *  6) if |x| < 2^23, large args:
+ *      6.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      6.2) Reconstruction same as (5.2).
+ *  7) if |x| >= 2^23, very large args:
+ *      7.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      7.2) Reconstruction same as (5.2).
+ *  8) if x is Inf, return x-x, and set errno=EDOM.
+ *  9) if x is NaN, return x-x.
+ *
+ * Special cases:
+ *  sin(+-0) = +-0 not raising inexact/underflow,
+ *  sin(subnormal) raises inexact/underflow,
+ *  sin(min_normalized) raises inexact/underflow,
+ *  sin(normalized) raises inexact,
+ *  sin(Inf) = NaN, raises invalid, sets errno to EDOM,
+ *  sin(NaN) = NaN.
+ */
+
+#ifdef	PIC
+# define MO1(symbol)			L(symbol)##@GOTOFF(%ebx)
+# define MO2(symbol,reg2,_scale)	L(symbol)##@GOTOFF(%ebx,reg2,_scale)
+# define CFI_PUSH(REG)	cfi_adjust_cfa_offset(4); cfi_rel_offset(REG,0)
+# define CFI_POP(REG)	cfi_adjust_cfa_offset(-4); cfi_restore(REG)
+# define PUSH(REG)			pushl REG; CFI_PUSH(REG)
+# define POP(REG)			popl REG; CFI_POP(REG)
+# define ENTRANCE			PUSH(%ebx); LOAD_PIC_REG(bx)
+# define RETURN				POP(%ebx); ret; CFI_PUSH(%ebx)
+# define ARG_X				8(%esp)
+#else
+# define MO1(symbol)			L(symbol)
+# define MO2(symbol,reg2,_scale)	L(symbol)(,reg2,_scale)
+# define ENTRANCE
+# define RETURN				ret
+# define ARG_X				4(%esp)
+#endif
+
+	.text
+ENTRY(__sinf_sse2)
+	/* Input: single precision x on stack at address ARG_X */
+
+	ENTRANCE
+	movl	ARG_X, %eax		/* Bits of x */
+	cvtss2sd ARG_X, %xmm0		/* DP x */
+	andl	$0x7fffffff, %eax	/* |x| */
+
+	cmpl	$0x3f490fdb, %eax	/* |x|<Pi/4?  */
+	jb	L(arg_less_pio4)
+
+	/* Here if |x|>=Pi/4 */
+	movd	%eax, %xmm3		/* SP |x| */
+	andpd	MO1(DP_ABS_MASK),%xmm0	/* DP |x| */
+	movss	MO1(SP_INVPIO4), %xmm2	/* SP 1/(Pi/4) */
+
+	cmpl	$0x40e231d6, %eax	/* |x|<9*Pi/4?  */
+	jae	L(large_args)
+
+	/* Here if Pi/4<=|x|<9*Pi/4 */
+	mulss	%xmm3, %xmm2		/* SP |x|/(Pi/4) */
+	movl	ARG_X, %ecx		/* Load x */
+	cvttss2si %xmm2, %eax		/* k, number of Pi/4 in x */
+	shrl	$31, %ecx		/* sign of x */
+	addl	$1, %eax		/* k+1 */
+	movl	$0x0e, %edx
+	andl	%eax, %edx		/* j = (k+1)&0x0e */
+	subsd	MO2(PIO4J,%edx,8), %xmm0 /* t = |x| - j * Pi/4 */
+
+L(reconstruction):
+	/* Input: %eax=n, %xmm0=t, %ecx=sign(x) */
+	testl	$2, %eax		/* n&2 != 0?  */
+	jz	L(sin_poly)
+
+/*L(cos_poly):*/
+	/* Here if sin(x) calculated using cos(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))))
+	 */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	MO1(DP_C4), %xmm4	/* C4 */
+	mulsd	%xmm0, %xmm4		/* z*C4 */
+	xorl	%eax, %ecx		/* (-1.0)^((n>>2)&1) XOR sign(x) */
+	movsd	MO1(DP_C3), %xmm3	/* C3 */
+	mulsd	%xmm0, %xmm3		/* z*C3 */
+	addsd	MO1(DP_C2), %xmm4	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm4		/* z*(C2+z*C4) */
+	lea	-8(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	addsd	MO1(DP_C1), %xmm3	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm3		/* z*(C1+z*C3) */
+	addsd	MO1(DP_C0), %xmm4	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm4		/* y*(C0+z*(C2+z*C4)) */
+
+	addsd	%xmm4, %xmm3		/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	/* 1.0+y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	MO1(DP_ONES), %xmm3
+
+	mulsd	MO2(DP_ONES,%ecx,8), %xmm3 /* DP result */
+	movsd	%xmm3, 0(%esp)		/* Move result from sse...  */
+	fldl	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	8(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(sin_poly):
+	/* Here if sin(x) calculated using sin(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))))
+	 */
+
+	movaps	%xmm0, %xmm4		/* t */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	xorl	%eax, %ecx		/* (-1.0)^((n>>2)&1) XOR sign(x) */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	MO1(DP_S4), %xmm2	/* S4 */
+	mulsd	%xmm0, %xmm2		/* z*S4 */
+	movsd	MO1(DP_S3), %xmm3	/* S3 */
+	mulsd	%xmm0, %xmm3		/* z*S3 */
+	lea	-8(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	addsd	MO1(DP_S2), %xmm2	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm2		/* z*(S2+z*S4) */
+	addsd	MO1(DP_S1), %xmm3	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm3		/* z*(S1+z*S3) */
+	addsd	MO1(DP_S0), %xmm2	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm2		/* y*(S0+z*(S2+z*S4)) */
+	/* t*s, where s = sign(x) * (-1.0)^((n>>2)&1) */
+	mulsd	MO2(DP_ONES,%ecx,8), %xmm4
+	addsd	%xmm2, %xmm3		/* y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	/* t*s*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	mulsd	%xmm4, %xmm3
+	/* t*s*(1.0+y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	movsd	%xmm3, 0(%esp)		/* Move result from sse...  */
+	fldl	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	8(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(large_args):
+	/* Here if |x|>=9*Pi/4 */
+	cmpl	$0x7f800000, %eax	/* x is Inf or NaN?  */
+	jae	L(arg_inf_or_nan)
+
+	/* Here if finite |x|>=9*Pi/4 */
+	cmpl	$0x4b000000, %eax	/* |x|<2^23?  */
+	jae	L(very_large_args)
+
+	/* Here if 9*Pi/4<=|x|<2^23 */
+	movsd	MO1(DP_INVPIO4), %xmm1	/* 1/(Pi/4) */
+	mulsd	%xmm0, %xmm1		/* |x|/(Pi/4) */
+	cvttsd2si %xmm1, %eax		/* k=trunc(|x|/(Pi/4)) */
+	addl	$1, %eax		/* k+1 */
+	movl	%eax, %edx
+	andl	$0xfffffffe, %edx	/* j=(k+1)&0xfffffffe */
+	cvtsi2sdl %edx, %xmm4		/* DP j */
+	movl	ARG_X, %ecx		/* Load x */
+	movsd	MO1(DP_PIO4HI), %xmm2	/* -PIO4HI = high part of -Pi/4 */
+	shrl	$31, %ecx		/* sign bit of x */
+	mulsd	%xmm4, %xmm2		/* -j*PIO4HI */
+	movsd	MO1(DP_PIO4LO), %xmm3	/* -PIO4LO = low part of -Pi/4 */
+	addsd	%xmm2, %xmm0		/* |x| - j*PIO4HI */
+	mulsd	%xmm3, %xmm4		/* j*PIO4LO */
+	addsd	%xmm4, %xmm0		/* t = |x| - j*PIO4HI - j*PIO4LO */
+	jmp	L(reconstruction)
+
+	.p2align	4
+L(very_large_args):
+	/* Here if finite |x|>=2^23 */
+
+	/* bitpos = (ix>>23) - BIAS_32 + 59; */
+	shrl	$23, %eax		/* eb = biased exponent of x */
+	/* bitpos = eb - 0x7f + 59, where 0x7f is exponent bias */
+	subl	$68, %eax
+	movl	$28, %ecx		/* %cl=28 */
+	movl	%eax, %edx		/* bitpos copy */
+
+	/* j = bitpos/28; */
+	div	%cl			/* j in register %al=%ax/%cl */
+	movapd	%xmm0, %xmm3		/* |x| */
+	/* clear unneeded remainder from %ah */
+	andl	$0xff, %eax
+
+	imull	$28, %eax, %ecx		/* j*28 */
+	movsd	MO1(DP_HI_MASK), %xmm4	/* DP_HI_MASK */
+	movapd	%xmm0, %xmm5		/* |x| */
+	mulsd	-2*8+MO2(_FPI,%eax,8), %xmm3	/* tmp3 = FPI[j-2]*|x| */
+	movapd	%xmm0, %xmm1		/* |x| */
+	mulsd	-1*8+MO2(_FPI,%eax,8), %xmm5	/* tmp2 = FPI[j-1]*|x| */
+	mulsd	0*8+MO2(_FPI,%eax,8), %xmm0	/* tmp0 = FPI[j]*|x| */
+	addl	$19, %ecx		/* j*28+19 */
+	mulsd	1*8+MO2(_FPI,%eax,8), %xmm1	/* tmp1 = FPI[j+1]*|x| */
+	cmpl	%ecx, %edx		/* bitpos>=j*28+19?   */
+	jl	L(very_large_skip1)
+
+	/* Here if bitpos>=j*28+19 */
+	andpd	%xmm3, %xmm4		/* HI(tmp3) */
+	subsd	%xmm4, %xmm3		/* tmp3 = tmp3 - HI(tmp3) */
+L(very_large_skip1):
+
+	movsd	MO1(DP_2POW52), %xmm6
+	movapd	%xmm5, %xmm2		/* tmp2 copy */
+	addsd	%xmm3, %xmm5		/* tmp5 = tmp3 + tmp2 */
+	movl	$1, %edx
+	addsd	%xmm5, %xmm6		/* tmp6 = tmp5 + 2^52 */
+	movsd	8+MO1(DP_2POW52), %xmm4
+	movd	%xmm6, %eax		/* k = I64_LO(tmp6); */
+	addsd	%xmm6, %xmm4		/* tmp4 = tmp6 - 2^52 */
+	movl	ARG_X, %ecx		/* Load x */
+	comisd	%xmm5, %xmm4		/* tmp4 > tmp5?  */
+	jbe	L(very_large_skip2)
+
+	/* Here if tmp4 > tmp5 */
+	subl	$1, %eax		/* k-- */
+	addsd	8+MO1(DP_ONES), %xmm4	/* tmp4 -= 1.0 */
+L(very_large_skip2):
+
+	andl	%eax, %edx		/* k&1 */
+	subsd	%xmm4, %xmm3		/* tmp3 -= tmp4 */
+	addsd	MO2(DP_ZERONE,%edx,8), %xmm3 /* t  = DP_ZERONE[k&1] + tmp3 */
+	addsd	%xmm2, %xmm3		/* t += tmp2 */
+	shrl	$31, %ecx		/* sign of x */
+	addsd	%xmm3, %xmm0		/* t += tmp0 */
+	addl	$1, %eax		/* n=k+1 */
+	addsd	%xmm1, %xmm0		/* t += tmp1 */
+	mulsd	MO1(DP_PIO4), %xmm0	/* t *= PI04 */
+
+	jmp	L(reconstruction)	/* end of very_large_args peth */
+
+	.p2align	4
+L(arg_less_pio4):
+	/* Here if |x|<Pi/4 */
+	cmpl	$0x3d000000, %eax	/* |x|<2^-5?  */
+	jl	L(arg_less_2pn5)
+
+	/* Here if 2^-5<=|x|<Pi/4 */
+	movaps	%xmm0, %xmm3		/* x */
+	mulsd	%xmm0, %xmm0		/* y=x^2 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=x^4 */
+	movsd	MO1(DP_S4), %xmm4	/* S4 */
+	mulsd	%xmm0, %xmm4		/* z*S4 */
+	movsd	MO1(DP_S3), %xmm5	/* S3 */
+	mulsd	%xmm0, %xmm5		/* z*S3 */
+	addsd	MO1(DP_S2), %xmm4	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm4		/* z*(S2+z*S4) */
+	addsd	MO1(DP_S1), %xmm5	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm5		/* z*(S1+z*S3) */
+	addsd	MO1(DP_S0), %xmm4	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm4		/* y*(S0+z*(S2+z*S4)) */
+	mulsd	%xmm3, %xmm5		/* x*z*(S1+z*S3) */
+	mulsd	%xmm3, %xmm4		/* x*y*(S0+z*(S2+z*S4)) */
+	/* x*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm5, %xmm4
+	/* x + x*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	cvtsd2ss %xmm3, %xmm3		/* SP result */
+
+L(epilogue):
+	lea	-4(%esp), %esp		/* Borrow 4 bytes of stack frame */
+	movss	%xmm3, 0(%esp)		/* Move result from sse...  */
+	flds	0(%esp)			/* ...to FPU.  */
+	/* Return back 4 bytes of stack frame */
+	lea	4(%esp), %esp
+	RETURN
+
+	.p2align	4
+L(arg_less_2pn5):
+	/* Here if |x|<2^-5 */
+	cmpl	$0x32000000, %eax	/* |x|<2^-27?  */
+	jl	L(arg_less_2pn27)
+
+	/* Here if 2^-27<=|x|<2^-5 */
+	movaps	%xmm0, %xmm1		/* DP x */
+	mulsd	%xmm0, %xmm0		/* DP x^2 */
+	movsd	MO1(DP_SIN2_1), %xmm3	/* DP DP_SIN2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_SIN2_1 */
+	addsd	MO1(DP_SIN2_0), %xmm3	/* DP DP_SIN2_0+x^2*DP_SIN2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_SIN2_0+x^4*DP_SIN2_1 */
+	mulsd	%xmm1, %xmm3		/* DP x^3*DP_SIN2_0+x^5*DP_SIN2_1 */
+	addsd	%xmm1, %xmm3		/* DP x+x^3*DP_SIN2_0+x^5*DP_SIN2_1 */
+	cvtsd2ss %xmm3, %xmm3		/* SP result */
+	jmp	L(epilogue)
+
+	.p2align	4
+L(arg_less_2pn27):
+	movss	ARG_X, %xmm3		/* SP x */
+	cmpl	$0, %eax		/* x=0?  */
+	je	L(epilogue)		/* in case x=0 return sin(+-0)==+-0 */
+	/* Here if |x|<2^-27 */
+	/*
+	 * Special cases here:
+	 *  sin(subnormal) raises inexact/underflow
+	 *  sin(min_normalized) raises inexact/underflow
+	 *  sin(normalized) raises inexact
+	 */
+	movaps	%xmm0, %xmm3		/* Copy of DP x */
+	mulsd	MO1(DP_SMALL), %xmm0	/* x*DP_SMALL */
+	subsd	%xmm0, %xmm3		/* Result is x-x*DP_SMALL */
+	cvtsd2ss %xmm3, %xmm3		/* Result converted to SP */
+	jmp	L(epilogue)
+
+	.p2align	4
+L(arg_inf_or_nan):
+	/* Here if |x| is Inf or NAN */
+	jne	L(skip_errno_setting)	/* in case of x is NaN */
+
+	/* Here if x is Inf. Set errno to EDOM.  */
+	call	JUMPTARGET(__errno_location)
+	movl	$EDOM, (%eax)
+
+	.p2align	4
+L(skip_errno_setting):
+	/* Here if |x| is Inf or NAN. Continued.  */
+	movss	ARG_X, %xmm3		/* load x */
+	subss	%xmm3, %xmm3		/* Result is NaN */
+	jmp	L(epilogue)
+END(__sinf_sse2)
+
+	.section .rodata, "a"
+	.p2align 3
+L(PIO4J): /* Table of j*Pi/4, for j=0,1,..,10 */
+	.long	0x00000000,0x00000000
+	.long	0x54442d18,0x3fe921fb
+	.long	0x54442d18,0x3ff921fb
+	.long	0x7f3321d2,0x4002d97c
+	.long	0x54442d18,0x400921fb
+	.long	0x2955385e,0x400f6a7a
+	.long	0x7f3321d2,0x4012d97c
+	.long	0xe9bba775,0x4015fdbb
+	.long	0x54442d18,0x401921fb
+	.long	0xbeccb2bb,0x401c463a
+	.long	0x2955385e,0x401f6a7a
+	.type L(PIO4J), @object
+	ASM_SIZE_DIRECTIVE(L(PIO4J))
+
+	.p2align 3
+L(_FPI): /* 4/Pi broken into sum of positive DP values */
+	.long	0x00000000,0x00000000
+	.long	0x6c000000,0x3ff45f30
+	.long	0x2a000000,0x3e3c9c88
+	.long	0xa8000000,0x3c54fe13
+	.long	0xd0000000,0x3aaf47d4
+	.long	0x6c000000,0x38fbb81b
+	.long	0xe0000000,0x3714acc9
+	.long	0x7c000000,0x3560e410
+	.long	0x56000000,0x33bca2c7
+	.long	0xac000000,0x31fbd778
+	.long	0xe0000000,0x300b7246
+	.long	0xe8000000,0x2e5d2126
+	.long	0x48000000,0x2c970032
+	.long	0xe8000000,0x2ad77504
+	.long	0xe0000000,0x290921cf
+	.long	0xb0000000,0x274deb1c
+	.long	0xe0000000,0x25829a73
+	.long	0xbe000000,0x23fd1046
+	.long	0x10000000,0x2224baed
+	.long	0x8e000000,0x20709d33
+	.long	0x80000000,0x1e535a2f
+	.long	0x64000000,0x1cef904e
+	.long	0x30000000,0x1b0d6398
+	.long	0x24000000,0x1964ce7d
+	.long	0x16000000,0x17b908bf
+	.type L(_FPI), @object
+	ASM_SIZE_DIRECTIVE(L(_FPI))
+
+/* Coefficients of polynomial
+   for sin(x)~=x+x^3*DP_SIN2_0+x^5*DP_SIN2_1, |x|<2^-5.  */
+	.p2align 3
+L(DP_SIN2_0):
+	.long	0x5543d49d,0xbfc55555
+	.type L(DP_SIN2_0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SIN2_0))
+
+	.p2align 3
+L(DP_SIN2_1):
+	.long	0x75cec8c5,0x3f8110f4
+	.type L(DP_SIN2_1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SIN2_1))
+
+	.p2align 3
+L(DP_ZERONE):
+	.long	0x00000000,0x00000000	/* 0.0 */
+	.long	0x00000000,0xbff00000	/* 1.0 */
+	.type L(DP_ZERONE), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ZERONE))
+
+	.p2align 3
+L(DP_ONES):
+	.long	0x00000000,0x3ff00000	/* +1.0 */
+	.long	0x00000000,0xbff00000	/* -1.0 */
+	.type L(DP_ONES), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ONES))
+
+/* Coefficients of polynomial
+   for sin(t)~=t+t^3*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_S3):
+	.long	0x64e6b5b4,0x3ec71d72
+	.type L(DP_S3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S3))
+
+	.p2align 3
+L(DP_S1):
+	.long	0x10c2688b,0x3f811111
+	.type L(DP_S1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S1))
+
+	.p2align 3
+L(DP_S4):
+	.long	0x1674b58a,0xbe5a947e
+	.type L(DP_S4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S4))
+
+	.p2align 3
+L(DP_S2):
+	.long	0x8b4bd1f9,0xbf2a019f
+	.type L(DP_S2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S2))
+
+	.p2align 3
+L(DP_S0):
+	.long	0x55551cd9,0xbfc55555
+	.type L(DP_S0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S0))
+
+	.p2align 3
+L(DP_SMALL):
+	.long	0x00000000,0x3cd00000	/* 2^(-50) */
+	.type L(DP_SMALL), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SMALL))
+
+/* Coefficients of polynomial
+   for cos(t)~=1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_C3):
+	.long	0x9ac43cc0,0x3efa00eb
+	.type L(DP_C3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C3))
+
+	.p2align 3
+L(DP_C1):
+	.long	0x545c50c7,0x3fa55555
+	.type L(DP_C1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C1))
+
+	.p2align 3
+L(DP_C4):
+	.long	0xdd8844d7,0xbe923c97
+	.type L(DP_C4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C4))
+
+	.p2align 3
+L(DP_C2):
+	.long	0x348b6874,0xbf56c16b
+	.type L(DP_C2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C2))
+
+	.p2align 3
+L(DP_C0):
+	.long	0xfffe98ae,0xbfdfffff
+	.type L(DP_C0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C0))
+
+	.p2align 3
+L(DP_PIO4):
+	.long	0x54442d18,0x3fe921fb	/* Pi/4 */
+	.type L(DP_PIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4))
+
+	.p2align 3
+L(DP_2POW52):
+	.long	0x00000000,0x43300000	/* +2^52 */
+	.long	0x00000000,0xc3300000	/* -2^52 */
+	.type L(DP_2POW52), @object
+	ASM_SIZE_DIRECTIVE(L(DP_2POW52))
+
+	.p2align 3
+L(DP_INVPIO4):
+	.long	0x6dc9c883,0x3ff45f30	/* 4/Pi */
+	.type L(DP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_INVPIO4))
+
+	.p2align 3
+L(DP_PIO4HI):
+	.long	0x54000000,0xbfe921fb	/* High part of Pi/4 */
+	.type L(DP_PIO4HI), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4HI))
+
+	.p2align 3
+L(DP_PIO4LO):
+	.long	0x11A62633,0xbe010b46	/* Low part of Pi/4 */
+	.type L(DP_PIO4LO), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4LO))
+
+	.p2align 2
+L(SP_INVPIO4):
+	.long	0x3fa2f983		/* 4/Pi */
+	.type L(SP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(SP_INVPIO4))
+
+	.p2align 4
+L(DP_ABS_MASK): /* Mask for getting DP absolute value */
+	.long	0xffffffff,0x7fffffff
+	.long	0xffffffff,0x7fffffff
+	.type L(DP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ABS_MASK))
+
+	.p2align 3
+L(DP_HI_MASK): /* Mask for getting high 21 bits of DP value */
+	.long	0x00000000,0xffffffff
+	.type L(DP_HI_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_HI_MASK))
+
+weak_alias (__sinf, sinf)
diff --git a/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf.c b/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf.c
new file mode 100644
index 000000000..80b2fd687
--- /dev/null
+++ b/libc/sysdeps/i386/i686/fpu/multiarch/s_sinf.c
@@ -0,0 +1,28 @@
+/* Multiple versions of sinf
+   Copyright (C) 2012 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 <init-arch.h>
+
+extern float __sinf_sse2 (float);
+extern float __sinf_ia32 (float);
+float __sinf (float);
+
+libm_ifunc (__sinf, HAS_SSE2 ? __sinf_sse2 : __sinf_ia32);
+weak_alias (__sinf, sinf);
+#define SINF __sinf_ia32
+#include <sysdeps/ieee754/flt-32/s_sinf.c>
diff --git a/libc/sysdeps/ieee754/flt-32/s_cosf.c b/libc/sysdeps/ieee754/flt-32/s_cosf.c
index f4bd81904..864ab27a4 100644
--- a/libc/sysdeps/ieee754/flt-32/s_cosf.c
+++ b/libc/sysdeps/ieee754/flt-32/s_cosf.c
@@ -23,7 +23,13 @@ static char rcsid[] = "$NetBSD: s_cosf.c,v 1.4 1995/05/10 20:47:03 jtc Exp $";
 
 static const float one=1.0;
 
-float __cosf(float x)
+#ifndef COSF
+# define COSF_FUNC __cosf
+#else
+# define COSF_FUNC COSF
+#endif
+
+float COSF_FUNC(float x)
 {
 	float y[2],z=0.0;
 	int32_t n,ix;
@@ -53,4 +59,7 @@ float __cosf(float x)
 	    }
 	}
 }
+
+#ifndef COSF
 weak_alias (__cosf, cosf)
+#endif
diff --git a/libc/sysdeps/ieee754/flt-32/s_sinf.c b/libc/sysdeps/ieee754/flt-32/s_sinf.c
index 02fa29f23..916e34557 100644
--- a/libc/sysdeps/ieee754/flt-32/s_sinf.c
+++ b/libc/sysdeps/ieee754/flt-32/s_sinf.c
@@ -21,7 +21,13 @@ static char rcsid[] = "$NetBSD: s_sinf.c,v 1.4 1995/05/10 20:48:16 jtc Exp $";
 #include <math.h>
 #include <math_private.h>
 
-float __sinf(float x)
+#ifndef SINF
+# define SINF_FUNC __sinf
+#else
+# define SINF_FUNC SINF
+#endif
+
+float SINF_FUNC(float x)
 {
 	float y[2],z=0.0;
 	int32_t n, ix;
@@ -51,4 +57,7 @@ float __sinf(float x)
 	    }
 	}
 }
+
+#ifndef SINF
 weak_alias (__sinf, sinf)
+#endif
diff --git a/libc/sysdeps/mach/bits/libc-lock.h b/libc/sysdeps/mach/bits/libc-lock.h
index 110e25e96..94877e0ea 100644
--- a/libc/sysdeps/mach/bits/libc-lock.h
+++ b/libc/sysdeps/mach/bits/libc-lock.h
@@ -137,7 +137,6 @@ void *__libc_getspecific (__libc_key_t key);
 #define __libc_lock_lock_recursive __libc_lock_lock
 
 #define __rtld_lock_define_initialized_recursive __libc_lock_define_initialized
-#define __rtld_lock_init_recursive __libc_lock_init
 #define __rtld_lock_fini_recursive __libc_lock_fini
 #define __rtld_lock_trylock_recursive __libc_lock_trylock
 #define __rtld_lock_unlock_recursive __libc_lock_unlock
diff --git a/libc/sysdeps/mach/hurd/_G_config.h b/libc/sysdeps/mach/hurd/_G_config.h
index 9ecd4aad3..b3a7db72a 100644
--- a/libc/sysdeps/mach/hurd/_G_config.h
+++ b/libc/sysdeps/mach/hurd/_G_config.h
@@ -18,7 +18,6 @@
 # define __need_wint_t
 #endif
 #include <wchar.h>
-#define _G_size_t	size_t
 typedef struct
 {
   __off_t __pos;
@@ -29,13 +28,7 @@ typedef struct
   __off64_t __pos;
   __mbstate_t __state;
 } _G_fpos64_t;
-#define _G_ssize_t	__ssize_t
-#define _G_off_t	__off_t
 #define _G_off64_t	__off64_t
-#define	_G_pid_t	__pid_t
-#define	_G_uid_t	__uid_t
-#define _G_wchar_t	wchar_t
-#define _G_wint_t	wint_t
 #define _G_stat64	stat64
 #if defined _LIBC || defined _GLIBCPP_USE_WCHAR_T
 # include <gconv.h>
@@ -50,25 +43,11 @@ typedef union
 } _G_iconv_t;
 #endif
 
-typedef int _G_int16_t __attribute__ ((__mode__ (__HI__)));
-typedef int _G_int32_t __attribute__ ((__mode__ (__SI__)));
-typedef unsigned int _G_uint16_t __attribute__ ((__mode__ (__HI__)));
-typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
-
-#define _G_HAVE_BOOL 1
-
 
 /* These library features are always available in the GNU C library.  */
-#define _G_HAVE_ATEXIT 1
-#define _G_HAVE_SYS_WAIT 1
-#define _G_NEED_STDARG_H 1
 #define _G_va_list __gnuc_va_list
 
-#define _G_HAVE_PRINTF_FP 1
 #define _G_HAVE_MMAP 1
-#define _G_HAVE_LONG_DOUBLE_IO 1
-#define _G_HAVE_IO_FILE_OPEN 1
-#define _G_HAVE_IO_GETLINE_INFO 1
 
 #define _G_IO_IO_FILE_VERSION 0x20001
 
@@ -82,11 +61,4 @@ typedef unsigned int _G_uint32_t __attribute__ ((__mode__ (__SI__)));
 
 #define _G_BUFSIZ 8192
 
-/* These are the vtbl details for ELF.  */
-#define _G_NAMES_HAVE_UNDERSCORE 0
-#define _G_VTABLE_LABEL_HAS_LENGTH 1
-#define _G_USING_THUNKS	1
-#define _G_VTABLE_LABEL_PREFIX "__vt_"
-#define _G_VTABLE_LABEL_PREFIX_ID __vt_
-
 #endif	/* _G_config.h */
diff --git a/libc/sysdeps/mach/hurd/bits/libc-lock.h b/libc/sysdeps/mach/hurd/bits/libc-lock.h
index d18a10c45..e8cb71bfd 100644
--- a/libc/sysdeps/mach/hurd/bits/libc-lock.h
+++ b/libc/sysdeps/mach/hurd/bits/libc-lock.h
@@ -119,8 +119,6 @@ typedef struct __libc_lock_recursive_opaque__ __libc_lock_recursive_t;
   })
 
 
-#define __rtld_lock_init_recursive(NAME) \
-  __libc_lock_init_recursive (NAME)
 #define __rtld_lock_initialize(NAME) \
   (void) ((NAME) = (__rtld_lock_recursive_t) _RTLD_LOCK_RECURSIVE_INITIALIZER)
 #define __rtld_lock_trylock_recursive(NAME) \
diff --git a/libc/sysdeps/s390/fpu/get-rounding-mode.h b/libc/sysdeps/s390/fpu/get-rounding-mode.h
new file mode 100644
index 000000000..3e427eca2
--- /dev/null
+++ b/libc/sysdeps/s390/fpu/get-rounding-mode.h
@@ -0,0 +1,37 @@
+/* Determine floating-point rounding mode within libc.  S/390 version.
+   Copyright (C) 2012 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 _S390_GET_ROUNDING_MODE_H
+#define _S390_GET_ROUNDING_MODE_H	1
+
+#include <fenv.h>
+#include <fenv_libc.h>
+#include <fpu_control.h>
+
+/* Return the floating-point rounding mode.  */
+
+static inline int
+get_rounding_mode (void)
+{
+  fpu_control_t fc;
+
+  _FPU_GETCW (fc);
+  return fc & FPC_RM_MASK;
+}
+
+#endif /* get-rounding-mode.h */
diff --git a/libc/sysdeps/s390/s390-32/multiarch/Makefile b/libc/sysdeps/s390/s390-32/multiarch/Makefile
index 5b0cd49c7..9baeecda6 100644
--- a/libc/sysdeps/s390/s390-32/multiarch/Makefile
+++ b/libc/sysdeps/s390/s390-32/multiarch/Makefile
@@ -1,9 +1,3 @@
-ASFLAGS-.o += -Wa,-mzarch
-ASFLAGS-.os += -Wa,-mzarch
-ASFLAGS-.op += -Wa,-mzarch
-ASFLAGS-.og += -Wa,-mzarch
-ASFLAGS-.ob += -Wa,-mzarch
-ASFLAGS-.oS += -Wa,-mzarch
 ifeq ($(subdir),string)
 sysdep_routines += ifunc-resolve memset memcpy memcmp
 endif
diff --git a/libc/sysdeps/s390/s390-32/multiarch/ifunc-resolve.c b/libc/sysdeps/s390/s390-32/multiarch/ifunc-resolve.c
index fd6e36706..f1bf8a3dc 100644
--- a/libc/sysdeps/s390/s390-32/multiarch/ifunc-resolve.c
+++ b/libc/sysdeps/s390/s390-32/multiarch/ifunc-resolve.c
@@ -32,14 +32,17 @@
        ".globl __GI_" #FUNC "\n\t"					\
        ".set   __GI_" #FUNC "," #FUNC "\n");				\
 									\
-  extern void *FUNC##_z10;						\
-  extern void *FUNC##_z196;						\
-  extern void *FUNC##_g5;						\
+  /* Make the declarations of the optimized functions hidden in order
+     to prevent GOT slots being generated for them. */			\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_z196;	\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_z10;	\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_g5;		\
 									\
   void *resolve_##FUNC (unsigned long int dl_hwcap)			\
   {									\
     if ((dl_hwcap & HWCAP_S390_STFLE)					\
-	&& (dl_hwcap & HWCAP_S390_HIGH_GPRS)) /* Implies zarch */	\
+	&& (dl_hwcap & HWCAP_S390_ZARCH)				\
+	&& (dl_hwcap & HWCAP_S390_HIGH_GPRS))				\
       {									\
 	/* We want just 1 double word to be returned.  */		\
 	register unsigned long reg0 asm("0") = 0;			\
diff --git a/libc/sysdeps/s390/s390-32/multiarch/memcmp.S b/libc/sysdeps/s390/s390-32/multiarch/memcmp.S
index cd8ea86b0..9ca88911c 100644
--- a/libc/sysdeps/s390/s390-32/multiarch/memcmp.S
+++ b/libc/sysdeps/s390/s390-32/multiarch/memcmp.S
@@ -31,6 +31,7 @@
 
 ENTRY(memcmp_z196)
 	.machine "z196"
+	.machinemode "zarch_nohighgprs"
         ltr     %r4,%r4
         je      .L_Z196_4
         ahi     %r4,-1
@@ -64,6 +65,7 @@ END(memcmp_z196)
 
 ENTRY(memcmp_z10)
 	.machine "z10"
+	.machinemode "zarch_nohighgprs"
         ltr     %r4,%r4
         je      .L_Z10_4
         ahi     %r4,-1
diff --git a/libc/sysdeps/s390/s390-32/multiarch/memcpy.S b/libc/sysdeps/s390/s390-32/multiarch/memcpy.S
index 40f4acfbb..7b85602aa 100644
--- a/libc/sysdeps/s390/s390-32/multiarch/memcpy.S
+++ b/libc/sysdeps/s390/s390-32/multiarch/memcpy.S
@@ -31,6 +31,7 @@
 
 ENTRY(memcpy_z196)
 	.machine "z196"
+	.machinemode "zarch_nohighgprs"
         llgfr   %r4,%r4
         ltgr    %r4,%r4
         je      .L_Z196_4
@@ -61,6 +62,7 @@ END(memcpy_z196)
 
 ENTRY(memcpy_z10)
 	.machine "z10"
+	.machinemode "zarch_nohighgprs"
         llgfr   %r4,%r4
         cgije   %r4,0,.L_Z10_4
         aghi    %r4,-1
diff --git a/libc/sysdeps/s390/s390-32/multiarch/memset.S b/libc/sysdeps/s390/s390-32/multiarch/memset.S
index 5ce7b4de8..ba43d671c 100644
--- a/libc/sysdeps/s390/s390-32/multiarch/memset.S
+++ b/libc/sysdeps/s390/s390-32/multiarch/memset.S
@@ -31,6 +31,7 @@
 
 ENTRY(memset_z196)
 	.machine "z196"
+	.machinemode "zarch_nohighgprs"
         llgfr   %r4,%r4
         ltgr    %r4,%r4
         je      .L_Z196_4
@@ -62,6 +63,7 @@ END(memset_z196)
 
 ENTRY(memset_z10)
 	.machine "z10"
+	.machinemode "zarch_nohighgprs"
         llgfr   %r4,%r4
         cgije   %r4,0,.L_Z10_4
         stc     %r3,0(%r2)
diff --git a/libc/sysdeps/s390/s390-64/multiarch/ifunc-resolve.c b/libc/sysdeps/s390/s390-64/multiarch/ifunc-resolve.c
index f980bc939..5f5676490 100644
--- a/libc/sysdeps/s390/s390-64/multiarch/ifunc-resolve.c
+++ b/libc/sysdeps/s390/s390-64/multiarch/ifunc-resolve.c
@@ -32,9 +32,11 @@
        ".globl __GI_" #FUNC "\n\t"					\
        ".set   __GI_" #FUNC "," #FUNC "\n");				\
 									\
-  extern void *FUNC##_z10;						\
-  extern void *FUNC##_z196;						\
-  extern void *FUNC##_z900;						\
+  /* Make the declarations of the optimized functions hidden in order
+     to prevent GOT slots being generated for them. */			\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_z196;	\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_z10;	\
+  extern __attribute__((visibility("hidden"))) void *FUNC##_z900;	\
 									\
   void *resolve_##FUNC (unsigned long int dl_hwcap)			\
   {									\
diff --git a/libc/sysdeps/sparc/bits/hwcap.h b/libc/sysdeps/sparc/bits/hwcap.h
index d922d7b0e..74ca130bb 100644
--- a/libc/sysdeps/sparc/bits/hwcap.h
+++ b/libc/sysdeps/sparc/bits/hwcap.h
@@ -45,3 +45,6 @@
 #define HWCAP_SPARC_IMA		0x00400000
 #define HWCAP_SPARC_ASI_CACHE_SPARING \
 				0x00800000
+#define HWCAP_SPARC_PAUSE	0x01000000
+#define HWCAP_SPARC_CBCOND	0x02000000
+#define HWCAP_SPARC_CRYPTO	0x04000000
diff --git a/libc/sysdeps/sparc/dl-procinfo.c b/libc/sysdeps/sparc/dl-procinfo.c
index 6e107dd1d..8280e0179 100644
--- a/libc/sysdeps/sparc/dl-procinfo.c
+++ b/libc/sysdeps/sparc/dl-procinfo.c
@@ -46,13 +46,13 @@
 #if !defined PROCINFO_DECL && defined SHARED
   ._dl_sparc_cap_flags
 #else
-PROCINFO_CLASS const char _dl_sparc_cap_flags[24][11]
+PROCINFO_CLASS const char _dl_sparc_cap_flags[27][11]
 #endif
 #ifndef PROCINFO_DECL
   = { "flush", "stbar", "swap", "muldiv", "v9", "ultra3", "v9v", "v9v2",
       "mul32", "div32", "fsmuld", "v8plus", "popc", "vis", "vis2",
       "ASIBlkInit", "fmaf", "vis3", "hpc", "random", "trans", "fjfmau",
-      "ima", "cspare" }
+      "ima", "cspare", "pause", "cbcond", "crypto" }
 #endif
 #if !defined SHARED || defined PROCINFO_DECL
 ;
diff --git a/libc/sysdeps/sparc/dl-procinfo.h b/libc/sysdeps/sparc/dl-procinfo.h
index 8ceb83072..6ae876828 100644
--- a/libc/sysdeps/sparc/dl-procinfo.h
+++ b/libc/sysdeps/sparc/dl-procinfo.h
@@ -24,7 +24,7 @@
 #include <ldsodefs.h>
 #include <sysdep.h>
 
-#define _DL_HWCAP_COUNT 24
+#define _DL_HWCAP_COUNT 27
 
 static inline int
 __attribute__ ((unused))
diff --git a/libc/sysdeps/sparc/fpu/libm-test-ulps b/libc/sysdeps/sparc/fpu/libm-test-ulps
index 74af9d571..b7dc3b483 100644
--- a/libc/sysdeps/sparc/fpu/libm-test-ulps
+++ b/libc/sysdeps/sparc/fpu/libm-test-ulps
@@ -1187,6 +1187,12 @@ idouble: 1
 ifloat: 1
 
 # cos
+Test "cos (0x1p+120) == -9.25879022854837867303861764107414946730833e-01":
+float: 1
+ifloat: 1
+Test "cos (0x1p+127) == 7.81914638714960072263910298466369236613162e-01":
+float: 1
+ifloat: 1
 Test "cos (M_PI_6l * 2.0) == 0.5":
 double: 1
 idouble: 1
@@ -2490,6 +2496,12 @@ float: 1
 ifloat: 1
 
 # sincos
+Test "sincos (0x1p+120, &sin_res, &cos_res) puts -9.25879022854837867303861764107414946730833e-01 in cos_res":
+float: 1
+ifloat: 1
+Test "sincos (0x1p+127, &sin_res, &cos_res) puts 7.81914638714960072263910298466369236613162e-01 in cos_res":
+float: 1
+ifloat: 1
 Test "sincos (M_PI_6l*2.0, &sin_res, &cos_res) puts 0.5 in cos_res":
 double: 1
 idouble: 1
diff --git a/libc/sysdeps/sparc/sparc32/sparcv9/addmul_1.S b/libc/sysdeps/sparc/sparc32/sparcv9/addmul_1.S
new file mode 100644
index 000000000..563bfb1c0
--- /dev/null
+++ b/libc/sysdeps/sparc/sparc32/sparcv9/addmul_1.S
@@ -0,0 +1 @@
+#include <sparcv8/addmul_1.S>
diff --git a/libc/sysdeps/sparc/sparc32/sparcv9/mul_1.S b/libc/sysdeps/sparc/sparc32/sparcv9/mul_1.S
new file mode 100644
index 000000000..42284eada
--- /dev/null
+++ b/libc/sysdeps/sparc/sparc32/sparcv9/mul_1.S
@@ -0,0 +1 @@
+#include <sparcv8/mul_1.S>
diff --git a/libc/sysdeps/sparc/sparc32/sparcv9/submul_1.S b/libc/sysdeps/sparc/sparc32/sparcv9/submul_1.S
new file mode 100644
index 000000000..de69533f6
--- /dev/null
+++ b/libc/sysdeps/sparc/sparc32/sparcv9/submul_1.S
@@ -0,0 +1 @@
+#include <sparcv8/submul_1.S>
diff --git a/libc/sysdeps/unix/sysv/linux/clock_getcpuclockid.c b/libc/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
index 1cfdf3489..8700354a9 100644
--- a/libc/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
+++ b/libc/sysdeps/unix/sysv/linux/clock_getcpuclockid.c
@@ -29,60 +29,24 @@
 int
 clock_getcpuclockid (pid_t pid, clockid_t *clock_id)
 {
-#ifdef __NR_clock_getres
   /* The clockid_t value is a simple computation from the PID.
      But we do a clock_getres call to validate it.  */
 
   const clockid_t pidclock = MAKE_PROCESS_CPUCLOCK (pid, CPUCLOCK_SCHED);
 
-# if !(__ASSUME_POSIX_CPU_TIMERS > 0)
-  extern int __libc_missing_posix_cpu_timers attribute_hidden;
-  if (!__libc_missing_posix_cpu_timers)
-# endif
+  INTERNAL_SYSCALL_DECL (err);
+  int r = INTERNAL_SYSCALL (clock_getres, err, 2, pidclock, NULL);
+  if (!INTERNAL_SYSCALL_ERROR_P (r, err))
     {
-      INTERNAL_SYSCALL_DECL (err);
-      int r = INTERNAL_SYSCALL (clock_getres, err, 2, pidclock, NULL);
-      if (!INTERNAL_SYSCALL_ERROR_P (r, err))
-	{
-	  *clock_id = pidclock;
-	  return 0;
-	}
-
-      if (INTERNAL_SYSCALL_ERRNO (r, err) == EINVAL)
-	{
-# if !(__ASSUME_POSIX_CPU_TIMERS > 0)
-	  if (pidclock == MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED)
-	      || INTERNAL_SYSCALL_ERROR_P (INTERNAL_SYSCALL
-					   (clock_getres, err, 2,
-					    MAKE_PROCESS_CPUCLOCK
-					    (0, CPUCLOCK_SCHED), NULL),
-					   err))
-	    /* The kernel doesn't support these clocks at all.  */
-	    __libc_missing_posix_cpu_timers = 1;
-	  else
-# endif
-	    /* The clock_getres system call checked the PID for us.  */
-	    return ESRCH;
-	}
-      else
-	return INTERNAL_SYSCALL_ERRNO (r, err);
+      *clock_id = pidclock;
+      return 0;
     }
-#endif
-
-  /* We don't allow any process ID but our own.  */
-  if (pid != 0 && pid != getpid ())
-    return EPERM;
 
-#ifdef CLOCK_PROCESS_CPUTIME_ID
-  if (HAS_CPUCLOCK)
+  if (INTERNAL_SYSCALL_ERRNO (r, err) == EINVAL)
     {
-      /* Store the number.  */
-      *clock_id = CLOCK_PROCESS_CPUTIME_ID;
-
-      return 0;
+      /* The clock_getres system call checked the PID for us.  */
+      return ESRCH;
     }
-#endif
-
-  /* We don't have a timer for that.  */
-  return ENOENT;
+  else
+    return INTERNAL_SYSCALL_ERRNO (r, err);
 }
diff --git a/libc/sysdeps/unix/sysv/linux/clock_getres.c b/libc/sysdeps/unix/sysv/linux/clock_getres.c
index 2bada33b4..1e87dd9a8 100644
--- a/libc/sysdeps/unix/sysv/linux/clock_getres.c
+++ b/libc/sysdeps/unix/sysv/linux/clock_getres.c
@@ -46,89 +46,11 @@
   case CLOCK_MONOTONIC_COARSE:						      \
     SYSCALL_GETRES
 
-#ifdef __NR_clock_getres
 /* We handled the REALTIME clock here.  */
-# define HANDLED_REALTIME	1
-# define HANDLED_CPUTIME	1
+#define HANDLED_REALTIME	1
+#define HANDLED_CPUTIME		1
 
-# if __ASSUME_POSIX_CPU_TIMERS > 0
-
-#  define SYSDEP_GETRES_CPU SYSCALL_GETRES
-#  define SYSDEP_GETRES_CPUTIME	/* Default catches them too.  */
-
-# else
-
-extern int __libc_missing_posix_cpu_timers attribute_hidden;
-
-static int
-maybe_syscall_getres_cpu (clockid_t clock_id, struct timespec *res)
-{
-  int e = EINVAL;
-
-  if (!__libc_missing_posix_cpu_timers)
-    {
-      INTERNAL_SYSCALL_DECL (err);
-      int r = INTERNAL_VSYSCALL (clock_getres, err, 2, clock_id, res);
-      if (!INTERNAL_SYSCALL_ERROR_P (r, err))
-	return 0;
-
-      e = INTERNAL_SYSCALL_ERRNO (r, err);
-      if (e == EINVAL)
-	{
-	  /* Check whether the kernel supports CPU clocks at all.
-	     If not, record it for the future.  */
-	  r = INTERNAL_VSYSCALL (clock_getres, err, 2,
-				 MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-				 NULL);
-	  if (INTERNAL_SYSCALL_ERROR_P (r, err))
-	    __libc_missing_posix_cpu_timers = 1;
-	}
-    }
-
-  return e;
-}
-
-#  define SYSDEP_GETRES_CPU						      \
-  retval = maybe_syscall_getres_cpu (clock_id, res);			      \
-  if (retval == 0)							      \
-    break;								      \
-  if (retval != EINVAL || !__libc_missing_posix_cpu_timers)		      \
-    {									      \
-      __set_errno (retval);						      \
-      retval = -1;							      \
-      break;								      \
-    }									      \
-  retval = -1 /* Otherwise continue on to the HP_TIMING version.  */;
-
-static inline int
-maybe_syscall_getres_cputime (clockid_t clock_id, struct timespec *res)
-{
-  return maybe_syscall_getres_cpu
-    (clock_id == CLOCK_THREAD_CPUTIME_ID
-     ? MAKE_THREAD_CPUCLOCK (0, CPUCLOCK_SCHED)
-     : MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-     res);
-}
-
-#  define SYSDEP_GETRES_CPUTIME						      \
-    case CLOCK_PROCESS_CPUTIME_ID:					      \
-    case CLOCK_THREAD_CPUTIME_ID:					      \
-      retval = maybe_syscall_getres_cputime (clock_id, res);		      \
-      if (retval == 0)							      \
-	break;								      \
-      if (retval != EINVAL || !__libc_missing_posix_cpu_timers)		      \
-	{								      \
-	  __set_errno (retval);						      \
-	  retval = -1;							      \
-	  break;							      \
-	}								      \
-      retval = hp_timing_getres (res);					      \
-      break;
-#  if !HP_TIMING_AVAIL
-#   define hp_timing_getres(res) (__set_errno (EINVAL), -1)
-#  endif
-
-# endif
-#endif
+#define SYSDEP_GETRES_CPU SYSCALL_GETRES
+#define SYSDEP_GETRES_CPUTIME	/* Default catches them too.  */
 
 #include <sysdeps/posix/clock_getres.c>
diff --git a/libc/sysdeps/unix/sysv/linux/clock_gettime.c b/libc/sysdeps/unix/sysv/linux/clock_gettime.c
index e333d8d9e..05dfb9d09 100644
--- a/libc/sysdeps/unix/sysv/linux/clock_gettime.c
+++ b/libc/sysdeps/unix/sysv/linux/clock_gettime.c
@@ -49,99 +49,13 @@
     retval = SYSCALL_GETTIME (clock_id, tp);				      \
     break
 
-#ifdef __NR_clock_gettime
 /* We handled the REALTIME clock here.  */
-# define HANDLED_REALTIME	1
-# define HANDLED_CPUTIME	1
+#define HANDLED_REALTIME	1
+#define HANDLED_CPUTIME	1
 
-# if __ASSUME_POSIX_CPU_TIMERS > 0
-
-#  define SYSDEP_GETTIME_CPU(clock_id, tp) \
+#define SYSDEP_GETTIME_CPU(clock_id, tp) \
   retval = SYSCALL_GETTIME (clock_id, tp); \
   break
-#  define SYSDEP_GETTIME_CPUTIME	/* Default catches them too.  */
-
-# else
-
-int __libc_missing_posix_cpu_timers attribute_hidden;
-
-static int
-maybe_syscall_gettime_cpu (clockid_t clock_id, struct timespec *tp)
-{
-  int e = EINVAL;
-
-  if (!__libc_missing_posix_cpu_timers)
-    {
-      INTERNAL_SYSCALL_DECL (err);
-      int r = INTERNAL_GETTIME (clock_id, tp);
-      if (!INTERNAL_SYSCALL_ERROR_P (r, err))
-	return 0;
-
-      e = INTERNAL_SYSCALL_ERRNO (r, err);
-      if (e == EINVAL)
-	{
-# ifdef HAVE_CLOCK_GETRES_VSYSCALL
-	  /* Check whether the kernel supports CPU clocks at all.
-	     If not, record it for the future.  */
-	  r = INTERNAL_VSYSCALL (clock_getres, err, 2,
-				 MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-				 NULL);
-# else
-	  /* Check whether the kernel supports CPU clocks at all.
-	     If not, record it for the future.  */
-	  r = INTERNAL_SYSCALL (clock_getres, err, 2,
-				MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-				NULL);
-# endif
-	  if (INTERNAL_SYSCALL_ERROR_P (r, err))
-	    __libc_missing_posix_cpu_timers = 1;
-	}
-    }
-
-  return e;
-}
-
-#  define SYSDEP_GETTIME_CPU(clock_id, tp) \
-  retval = maybe_syscall_gettime_cpu (clock_id, tp);			      \
-  if (retval == 0)							      \
-    break;								      \
-  if (retval != EINVAL || !__libc_missing_posix_cpu_timers)		      \
-    {									      \
-      __set_errno (retval);						      \
-      retval = -1;							      \
-      break;								      \
-    }									      \
-  retval = -1 /* Otherwise continue on to the HP_TIMING version.  */;
-
-static inline int
-maybe_syscall_gettime_cputime (clockid_t clock_id, struct timespec *tp)
-{
-  return maybe_syscall_gettime_cpu
-    (clock_id == CLOCK_THREAD_CPUTIME_ID
-     ? MAKE_THREAD_CPUCLOCK (0, CPUCLOCK_SCHED)
-     : MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-     tp);
-}
-
-#  define SYSDEP_GETTIME_CPUTIME					      \
-    case CLOCK_PROCESS_CPUTIME_ID:					      \
-    case CLOCK_THREAD_CPUTIME_ID:					      \
-      retval = maybe_syscall_gettime_cputime (clock_id, tp);		      \
-      if (retval == 0)							      \
-	break;								      \
-      if (retval != EINVAL || !__libc_missing_posix_cpu_timers)		      \
-	{								      \
-	  __set_errno (retval);						      \
-	  retval = -1;							      \
-	  break;							      \
-	}								      \
-      retval = hp_timing_gettime (clock_id, tp);			      \
-      break
-#  if !HP_TIMING_AVAIL
-#   define hp_timing_gettime(clock_id, tp) (__set_errno (EINVAL), -1)
-#  endif
-
-# endif
-#endif
+#define SYSDEP_GETTIME_CPUTIME	/* Default catches them too.  */
 
 #include <sysdeps/unix/clock_gettime.c>
diff --git a/libc/sysdeps/unix/sysv/linux/clock_settime.c b/libc/sysdeps/unix/sysv/linux/clock_settime.c
index f75ceca24..fde662177 100644
--- a/libc/sysdeps/unix/sysv/linux/clock_settime.c
+++ b/libc/sysdeps/unix/sysv/linux/clock_settime.c
@@ -31,38 +31,6 @@
 # include <bits/libc-vdso.h>
 #endif
 
-#if __ASSUME_POSIX_CPU_TIMERS <= 0 && defined __NR_clock_settime
-extern int __libc_missing_posix_cpu_timers attribute_hidden;
-
-static int
-maybe_syscall_settime_cpu (clockid_t clock_id, const struct timespec *tp)
-{
-  int e = EINVAL;
-
-  if (!__libc_missing_posix_cpu_timers)
-    {
-      INTERNAL_SYSCALL_DECL (err);
-      int r = INTERNAL_SYSCALL (clock_settime, err, 2, clock_id, tp);
-      if (!INTERNAL_SYSCALL_ERROR_P (r, err))
-	return 0;
-
-      e = INTERNAL_SYSCALL_ERRNO (r, err);
-      if (e == EINVAL)
-	{
-	  /* Check whether the kernel supports CPU clocks at all.
-	     If not, record it for the future.  */
-	  r = INTERNAL_VSYSCALL (clock_getres, err, 2,
-				 MAKE_PROCESS_CPUCLOCK (0, CPUCLOCK_SCHED),
-				 NULL);
-	  if (INTERNAL_SYSCALL_ERROR_P (r, err))
-	    __libc_missing_posix_cpu_timers = 1;
-	}
-    }
-
-  return e;
-}
-#endif
-
 
 /* The REALTIME clock is definitely supported in the kernel.  */
 #define SYSDEP_SETTIME \
@@ -70,27 +38,11 @@ maybe_syscall_settime_cpu (clockid_t clock_id, const struct timespec *tp)
     retval = INLINE_SYSCALL (clock_settime, 2, clock_id, tp);		      \
     break
 
-#ifdef __NR_clock_settime
 /* We handled the REALTIME clock here.  */
-# define HANDLED_REALTIME	1
-#endif
+#define HANDLED_REALTIME	1
 
-#if __ASSUME_POSIX_CPU_TIMERS > 0
-# define HANDLED_CPUTIME 1
-# define SYSDEP_SETTIME_CPU \
+#define HANDLED_CPUTIME 1
+#define SYSDEP_SETTIME_CPU \
   retval = INLINE_SYSCALL (clock_settime, 2, clock_id, tp)
-#elif defined __NR_clock_settime
-# define SYSDEP_SETTIME_CPU \
-  retval = maybe_syscall_settime_cpu (clock_id, tp);			      \
-  if (retval == 0)							      \
-    break;								      \
-  if (retval != EINVAL || !__libc_missing_posix_cpu_timers)		      \
-    {									      \
-      __set_errno (retval);						      \
-      retval = -1;							      \
-      break;								      \
-    }									      \
-  do { } while (0)
-#endif
 
 #include <sysdeps/unix/clock_settime.c>
diff --git a/libc/sysdeps/unix/sysv/linux/kernel-features.h b/libc/sysdeps/unix/sysv/linux/kernel-features.h
index 33fca5c4f..61bf32d1b 100644
--- a/libc/sysdeps/unix/sysv/linux/kernel-features.h
+++ b/libc/sysdeps/unix/sysv/linux/kernel-features.h
@@ -57,9 +57,6 @@
    MIPS n32).  */
 #define __ASSUME_GETDENTS64_SYSCALL	1
 
-/* Beginning with 2.6.12 the clock and timer supports CPU clocks.  */
-#define __ASSUME_POSIX_CPU_TIMERS	1
-
 /* The statfs64 syscalls are available in 2.5.74 (but not for alpha).  */
 #define __ASSUME_STATFS64	1
 
diff --git a/libc/sysdeps/unix/sysv/linux/malloc-sysdep.h b/libc/sysdeps/unix/sysv/linux/malloc-sysdep.h
new file mode 100644
index 000000000..f926aeaff
--- /dev/null
+++ b/libc/sysdeps/unix/sysv/linux/malloc-sysdep.h
@@ -0,0 +1,57 @@
+/* System-specific malloc support functions.  Linux version.
+   Copyright (C) 2012 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 <fcntl.h>
+#include <not-cancel.h>
+
+/* The Linux kernel overcommits address space by default and if there is not
+   enough memory available, it uses various parameters to decide the process to
+   kill.  It is however possible to disable or curb this overcommit behavior
+   by setting the proc sysctl vm.overcommit_memory to the value '2' and with
+   that, a process is only allowed to use the maximum of a pre-determined
+   fraction of the total address space.  In such a case, we want to make sure
+   that we are judicious with our heap usage as well, and explicitly give away
+   the freed top of the heap to reduce our commit charge.  See the proc(5) man
+   page to know more about overcommit behavior.
+
+   Other than that, we also force an unmap in a secure exec.  */
+static inline bool
+check_may_shrink_heap (void)
+{
+  static int may_shrink_heap = -1;
+
+  if (__builtin_expect (may_shrink_heap >= 0, 1))
+    return may_shrink_heap;
+
+  may_shrink_heap = __libc_enable_secure;
+
+  if (__builtin_expect (may_shrink_heap == 0, 1))
+    {
+      int fd = open_not_cancel_2 ("/proc/sys/vm/overcommit_memory",
+				  O_RDONLY | O_CLOEXEC);
+      if (fd >= 0)
+	{
+	  char val;
+	  ssize_t n = read_not_cancel (fd, &val, 1);
+	  may_shrink_heap = n > 0 && val == '2';
+	  close_not_cancel_no_status (fd);
+	}
+    }
+
+  return may_shrink_heap;
+}
diff --git a/libc/sysdeps/unix/sysv/linux/ttyname.c b/libc/sysdeps/unix/sysv/linux/ttyname.c
index 3dacd40c2..d3070c3ba 100644
--- a/libc/sysdeps/unix/sysv/linux/ttyname.c
+++ b/libc/sysdeps/unix/sysv/linux/ttyname.c
@@ -148,12 +148,6 @@ ttyname (int fd)
     }
 
   ssize_t len = __readlink (procname, ttyname_buf, buflen);
-  if (__builtin_expect (len == -1 && errno == ENOENT, 0))
-    {
-      __set_errno (EBADF);
-      return NULL;
-    }
-
   if (__builtin_expect (len != -1, 1))
     {
       if ((size_t) len >= buflen)
diff --git a/libc/sysdeps/unix/sysv/linux/ttyname_r.c b/libc/sysdeps/unix/sysv/linux/ttyname_r.c
index 2dd678134..211ed189b 100644
--- a/libc/sysdeps/unix/sysv/linux/ttyname_r.c
+++ b/libc/sysdeps/unix/sysv/linux/ttyname_r.c
@@ -128,12 +128,6 @@ __ttyname_r (int fd, char *buf, size_t buflen)
   *_fitoa_word (fd, __stpcpy (procname, "/proc/self/fd/"), 10, 0) = '\0';
 
   ssize_t ret = __readlink (procname, buf, buflen - 1);
-  if (__builtin_expect (ret == -1 && errno == ENOENT, 0))
-    {
-      __set_errno (EBADF);
-      return EBADF;
-    }
-
   if (__builtin_expect (ret == -1 && errno == ENAMETOOLONG, 0))
     {
       __set_errno (ERANGE);
diff --git a/libc/sysdeps/wordsize-64/Makefile b/libc/sysdeps/wordsize-64/Makefile
index 9903f51f9..2fa934751 100644
--- a/libc/sysdeps/wordsize-64/Makefile
+++ b/libc/sysdeps/wordsize-64/Makefile
@@ -1,6 +1,3 @@
 ifeq ($(subdir),misc)
 tests += tst-writev
-
-# Time enough for a large writev syscall to complete.
-tst-writev-ENV = TIMEOUTFACTOR="10"
 endif
diff --git a/libc/sysdeps/wordsize-64/tst-writev.c b/libc/sysdeps/wordsize-64/tst-writev.c
index 49a809cfa..0bc2f70b6 100644
--- a/libc/sysdeps/wordsize-64/tst-writev.c
+++ b/libc/sysdeps/wordsize-64/tst-writev.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2011 Free Software Foundation, Inc.
+/* Copyright (C) 2011-2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ryan S. Arnold <rsa@us.ibm.com>, 2011.
 
@@ -108,5 +108,8 @@ do_test (void)
   return 0;
 }
 
+/* Time enough for a large writev syscall to complete.  */
+#define TIMEOUT 20
+
 #define TEST_FUNCTION do_test ()
 #include "../test-skeleton.c"
diff --git a/libc/sysdeps/x86_64/Makefile b/libc/sysdeps/x86_64/Makefile
index d4525bc7c..0acb69a17 100644
--- a/libc/sysdeps/x86_64/Makefile
+++ b/libc/sysdeps/x86_64/Makefile
@@ -41,7 +41,7 @@ $(objpfx)tst-quad2pie: $(objpfx)tst-quadmod2pie.o
 tests: $(objpfx)tst-xmmymm.out
 $(objpfx)tst-xmmymm.out: ../sysdeps/x86_64/tst-xmmymm.sh $(objpfx)ld.so
 	@echo "Checking ld.so for SSE register use.  This will take a few seconds..."
-	$(SHELL) -e $< $(objpfx) > $@
+	$(SHELL) $< $(objpfx) > $@
 endif
 
 ifeq ($(subdir),csu)
diff --git a/libc/sysdeps/x86_64/dl-machine.h b/libc/sysdeps/x86_64/dl-machine.h
index 9c27ecffd..d2654aa04 100644
--- a/libc/sysdeps/x86_64/dl-machine.h
+++ b/libc/sysdeps/x86_64/dl-machine.h
@@ -41,13 +41,9 @@ elf_machine_matches_host (const ElfW(Ehdr) *ehdr)
 static inline ElfW(Addr) __attribute__ ((unused))
 elf_machine_dynamic (void)
 {
-  ElfW(Addr) addr;
-
-  /* This works because we have our GOT address available in the small PIC
-     model.  */
-  addr = (ElfW(Addr)) &_DYNAMIC;
-
-  return addr;
+  /* This produces an IP-relative reloc which is resolved at link time. */
+  extern const ElfW(Addr) _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
+  return _GLOBAL_OFFSET_TABLE_[0];
 }
 
 
@@ -55,31 +51,11 @@ elf_machine_dynamic (void)
 static inline ElfW(Addr) __attribute__ ((unused))
 elf_machine_load_address (void)
 {
-  ElfW(Addr) addr;
-
-  /* The easy way is just the same as on x86:
-       leaq _dl_start, %0
-       leaq _dl_start(%%rip), %1
-       subq %0, %1
-     but this does not work with binutils since we then have
-     a R_X86_64_32S relocation in a shared lib.
-
-     Instead we store the address of _dl_start in the data section
-     and compare it with the current value that we can get via
-     an RIP relative addressing mode.  Note that this is the address
-     of _dl_start before any relocation performed at runtime.  In case
-     the binary is prelinked the resulting "address" is actually a
-     load offset which is zero if the binary was loaded at the address
-     it is prelinked for.  */
-
-  asm ("lea _dl_start(%%rip), %0\n\t"
-       "sub 1f(%%rip), %0\n\t"
-       ".section\t.data.rel.ro\n"
-       "1:\t" ASM_ADDR " _dl_start\n\t"
-       ".previous\n\t"
-       : "=r" (addr) : : "cc");
-
-  return addr;
+  /* Compute the difference between the runtime address of _DYNAMIC as seen
+     by an IP-relative reference, and the link-time address found in the
+     special unrelocated first GOT entry.  */
+  extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
+  return (ElfW(Addr)) &_DYNAMIC - elf_machine_dynamic ();
 }
 
 /* Set up the loaded object described by L so its unrelocated PLT
diff --git a/libc/sysdeps/x86_64/fpu/libm-test-ulps b/libc/sysdeps/x86_64/fpu/libm-test-ulps
index 8a3822113..b02b9e6ba 100644
--- a/libc/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/libc/sysdeps/x86_64/fpu/libm-test-ulps
@@ -1208,6 +1208,8 @@ ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_downward (10) == -0.8390715290764524522588639478240648345199":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_downward (2) == -0.4161468365471423869975682295007621897660":
@@ -1239,6 +1241,8 @@ ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_downward (9) == -0.9111302618846769883682947111811653112463":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
@@ -1285,15 +1289,22 @@ ildouble: 1
 ldouble: 1
 
 # cos_upward
+Test "cos_upward (1) == 0.5403023058681397174009366074429766037323":
+float: 1
+ifloat: 1
 Test "cos_upward (10) == -0.8390715290764524522588639478240648345199":
 float: 1
 ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_upward (2) == -0.4161468365471423869975682295007621897660":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_upward (3) == -0.9899924966004454572715727947312613023937":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_upward (4) == -0.6536436208636119146391681830977503814241":
@@ -1311,6 +1322,8 @@ Test "cos_upward (7) == 0.7539022543433046381411975217191820122183":
 float: 1
 ifloat: 1
 Test "cos_upward (8) == -0.1455000338086135258688413818311946826093":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "cos_upward (9) == -0.9111302618846769883682947111811653112463":
@@ -2319,6 +2332,8 @@ Test "sin_downward (8) == 0.9893582466233817778081235982452886721164":
 ildouble: 1
 ldouble: 1
 Test "sin_downward (9) == 0.4121184852417565697562725663524351793439":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
@@ -2374,6 +2389,8 @@ Test "sin_upward (1) == 0.8414709848078965066525023216302989996226":
 float: 1
 ifloat: 1
 Test "sin_upward (10) == -0.5440211108893698134047476618513772816836":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "sin_upward (2) == 0.9092974268256816953960198659117448427023":
@@ -2382,6 +2399,8 @@ ifloat: 2
 ildouble: 1
 ldouble: 1
 Test "sin_upward (3) == 0.1411200080598672221007448028081102798469":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "sin_upward (4) == -0.7568024953079282513726390945118290941359":
@@ -2390,11 +2409,19 @@ ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "sin_upward (5) == -0.9589242746631384688931544061559939733525":
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "sin_upward (6) == -0.2794154981989258728115554466118947596280":
 ildouble: 1
 ldouble: 1
+Test "sin_upward (7) == 0.6569865987187890903969990915936351779369":
+float: 1
+ifloat: 1
+Test "sin_upward (8) == 0.9893582466233817778081235982452886721164":
+float: 1
+ifloat: 1
 Test "sin_upward (9) == 0.4121184852417565697562725663524351793439":
 float: 1
 ifloat: 1
@@ -2403,6 +2430,12 @@ ifloat: 1
 Test "sincos (0.80190127184058835, &sin_res, &cos_res) puts 0.69534156199418473 in cos_res":
 double: 1
 idouble: 1
+Test "sincos (0x1p+120, &sin_res, &cos_res) puts -9.25879022854837867303861764107414946730833e-01 in cos_res":
+float: 1
+ifloat: 1
+Test "sincos (0x1p+127, &sin_res, &cos_res) puts 7.81914638714960072263910298466369236613162e-01 in cos_res":
+float: 1
+ifloat: 1
 Test "sincos (M_PI_6l*2.0, &sin_res, &cos_res) puts 0.5 in cos_res":
 double: 1
 float: 1
@@ -3052,8 +3085,8 @@ ildouble: 1
 ldouble: 1
 
 Function: "cos_upward":
-float: 2
-ifloat: 2
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
@@ -3412,8 +3445,8 @@ ildouble: 1
 ldouble: 1
 
 Function: "sin_upward":
-float: 2
-ifloat: 2
+float: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
diff --git a/libc/sysdeps/x86_64/fpu/s_cosf.S b/libc/sysdeps/x86_64/fpu/s_cosf.S
new file mode 100644
index 000000000..dc8c76a13
--- /dev/null
+++ b/libc/sysdeps/x86_64/fpu/s_cosf.S
@@ -0,0 +1,527 @@
+/* Optimized cosf function.
+   Copyright (C) 2012 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 <sysdep.h>
+#define __need_Emath
+#include <bits/errno.h>
+
+/* Short algorithm description:
+ *
+ *  1) if |x| == 0: return 1.0-|x|.
+ *  2) if |x| <  2^-27: return 1.0-|x|.
+ *  3) if |x| <  2^-5 : return 1.0+x^2*DP_COS2_0+x^5*DP_COS2_1.
+ *  4) if |x| <   Pi/4: return 1.0+x^2*(C0+x^2*(C1+x^2*(C2+x^2*(C3+x^2*C4)))).
+ *  5) if |x| < 9*Pi/4:
+ *      5.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0x0e, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      5.2) Reconstruction:
+ *          s = (-1.0)^((n>>2)&1)
+ *          if(n&2 != 0) {
+ *              using cos(t) polynomial for |t|<Pi/4, result is
+ *              s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4))))).
+ *          } else {
+ *              using sin(t) polynomial for |t|<Pi/4, result is
+ *              s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4))))).
+ *          }
+ *  6) if |x| < 2^23, large args:
+ *      6.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      6.2) Reconstruction same as (5.2).
+ *  7) if |x| >= 2^23, very large args:
+ *      7.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+3,
+ *           t=|x|-j*Pi/4.
+ *      7.2) Reconstruction same as (5.2).
+ *  8) if x is Inf, return x-x, and set errno=EDOM.
+ *  9) if x is NaN, return x-x.
+ *
+ * Special cases:
+ *  cos(+-0) = 1 not raising inexact,
+ *  cos(subnormal) raises inexact,
+ *  cos(min_normalized) raises inexact,
+ *  cos(normalized) raises inexact,
+ *  cos(Inf) = NaN, raises invalid, sets errno to EDOM,
+ *  cos(NaN) = NaN.
+ */
+
+	.text
+ENTRY(__cosf)
+	/* Input: single precision x in %xmm0 */
+
+	movd	%xmm0, %eax		/* Bits of x */
+	movaps	%xmm0, %xmm7		/* Copy of x */
+	cvtss2sd %xmm0, %xmm0		/* DP x */
+	movss	L(SP_ABS_MASK)(%rip), %xmm3
+	andl	$0x7fffffff, %eax	/* |x| */
+
+	cmpl	$0x3f490fdb, %eax	/* |x|<Pi/4?  */
+	jb	L(arg_less_pio4)
+
+	/* Here if |x|>=Pi/4 */
+	andps	%xmm7, %xmm3		/* SP |x| */
+	andpd	L(DP_ABS_MASK)(%rip), %xmm0	/* DP |x| */
+	movss	L(SP_INVPIO4)(%rip), %xmm2	/* SP 1/(Pi/4) */
+
+	cmpl	$0x40e231d6, %eax	/* |x|<9*Pi/4?  */
+	jae	L(large_args)
+
+	/* Here if Pi/4<=|x|<9*Pi/4 */
+	mulss	%xmm3, %xmm2		/* SP |x|/(Pi/4) */
+	cvttss2si %xmm2, %eax		/* k, number of Pi/4 in x */
+	lea	L(PIO4J)(%rip), %rsi
+	addl	$1, %eax		/* k+1 */
+	movl	$0x0e, %edx
+	andl	%eax, %edx		/* j = (k+1)&0x0e */
+	addl	$2, %eax		/* n */
+	subsd	(%rsi,%rdx,8), %xmm0	/* t = |x| - j * Pi/4 */
+
+L(reconstruction):
+	/* Input: %eax=n, %xmm0=t */
+	testl	$2, %eax		/* n&2 != 0?  */
+	jz	L(sin_poly)
+
+/*L(cos_poly):*/
+	/* Here if cos(x) calculated using cos(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))))
+	 */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	L(DP_C4)(%rip), %xmm4	/* C4 */
+	mulsd	%xmm0, %xmm4		/* z*C4 */
+	movsd	L(DP_C3)(%rip), %xmm3	/* C3 */
+	mulsd	%xmm0, %xmm3		/* z*C3 */
+	lea	L(DP_ONES)(%rip), %rsi
+	addsd	L(DP_C2)(%rip), %xmm4	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm4		/* z*(C2+z*C4) */
+	addsd	L(DP_C1)(%rip), %xmm3	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm3		/* z*(C1+z*C3) */
+	addsd	L(DP_C0)(%rip), %xmm4	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm4		/* y*(C0+z*(C2+z*C4)) */
+
+	addsd	%xmm4, %xmm3		/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	/* 1.0+y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	L(DP_ONES)(%rip), %xmm3
+
+	mulsd	(%rsi,%rax,8), %xmm3	/* DP result */
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(sin_poly):
+	/* Here if cos(x) calculated using sin(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))))
+	 */
+
+	movaps	%xmm0, %xmm4		/* t */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	L(DP_S4)(%rip), %xmm2	/* S4 */
+	mulsd	%xmm0, %xmm2		/* z*S4 */
+	movsd	L(DP_S3)(%rip), %xmm3	/* S3 */
+	mulsd	%xmm0, %xmm3		/* z*S3 */
+	lea	L(DP_ONES)(%rip), %rsi
+	addsd	L(DP_S2)(%rip), %xmm2	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm2		/* z*(S2+z*S4) */
+	addsd	L(DP_S1)(%rip), %xmm3	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm3		/* z*(S1+z*S3) */
+	addsd	L(DP_S0)(%rip), %xmm2	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm2		/* y*(S0+z*(S2+z*S4)) */
+	/* t*s, where s = sign(x) * (-1.0)^((n>>2)&1) */
+	mulsd	(%rsi,%rax,8), %xmm4
+	/* y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm2, %xmm3
+	/* t*s*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	mulsd	%xmm4, %xmm3
+	/* t*s*(1.0+y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(large_args):
+	/* Here if |x|>=9*Pi/4 */
+	cmpl	$0x7f800000, %eax	/* x is Inf or NaN?  */
+	jae	L(arg_inf_or_nan)
+
+	/* Here if finite |x|>=9*Pi/4 */
+	cmpl	$0x4b000000, %eax	/* |x|<2^23?  */
+	jae	L(very_large_args)
+
+	/* Here if 9*Pi/4<=|x|<2^23 */
+	movsd	L(DP_INVPIO4)(%rip), %xmm1 /* 1/(Pi/4) */
+	mulsd	%xmm0, %xmm1		/* |x|/(Pi/4) */
+	cvttsd2si %xmm1, %eax		/* k=trunc(|x|/(Pi/4)) */
+	addl	$1, %eax		/* k+1 */
+	movl	%eax, %edx
+	andl	$0xfffffffe, %edx	/* j=(k+1)&0xfffffffe */
+	cvtsi2sdl %edx, %xmm4		/* DP j */
+	movsd	L(DP_PIO4HI)(%rip), %xmm2 /* -PIO4HI = high part of -Pi/4 */
+	mulsd	%xmm4, %xmm2		/* -j*PIO4HI */
+	movsd	L(DP_PIO4LO)(%rip), %xmm3 /* -PIO4LO = low part of -Pi/4 */
+	addsd	%xmm2, %xmm0		/* |x| - j*PIO4HI */
+	addl	$2, %eax		/* n */
+	mulsd	%xmm3, %xmm4		/* j*PIO4LO */
+	addsd	%xmm4, %xmm0		/* t = |x| - j*PIO4HI - j*PIO4LO */
+	jmp	L(reconstruction)
+
+	.p2align	4
+L(very_large_args):
+	/* Here if finite |x|>=2^23 */
+
+	/* bitpos = (ix>>23) - BIAS_32 + 59; */
+	shrl	$23, %eax		/* eb = biased exponent of x */
+	/* bitpos = eb - 0x7f + 59, where 0x7f is exponent bias */
+	subl	$68, %eax
+	movl	$28, %ecx		/* %cl=28 */
+	movl	%eax, %edx		/* bitpos copy */
+
+	/* j = bitpos/28; */
+	div	%cl			/* j in register %al=%ax/%cl */
+	movapd	%xmm0, %xmm3		/* |x| */
+	/* clear unneeded remainder from %ah */
+	andl	$0xff, %eax
+
+	imull	$28, %eax, %ecx		/* j*28 */
+	lea	L(_FPI)(%rip), %rsi
+	movsd	L(DP_HI_MASK)(%rip), %xmm4 /* DP_HI_MASK */
+	movapd	%xmm0, %xmm5		/* |x| */
+	mulsd	-16(%rsi,%rax,8), %xmm3	/* tmp3 = FPI[j-2]*|x| */
+	movapd	%xmm0, %xmm1		/* |x| */
+	mulsd	-8(%rsi,%rax,8), %xmm5	/* tmp2 = FPI[j-1]*|x| */
+	mulsd	(%rsi,%rax,8), %xmm0	/* tmp0 = FPI[j]*|x| */
+	addl	$19, %ecx		/* j*28+19 */
+	mulsd	8(%rsi,%rax,8), %xmm1	/* tmp1 = FPI[j+1]*|x| */
+	cmpl	%ecx, %edx		/* bitpos>=j*28+19?  */
+	jl	L(very_large_skip1)
+
+	/* Here if bitpos>=j*28+19 */
+	andpd	%xmm3, %xmm4		/* HI(tmp3) */
+	subsd	%xmm4, %xmm3		/* tmp3 = tmp3 - HI(tmp3) */
+L(very_large_skip1):
+
+	movsd	L(DP_2POW52)(%rip), %xmm6
+	movapd	%xmm5, %xmm2		/* tmp2 copy */
+	addsd	%xmm3, %xmm5		/* tmp5 = tmp3 + tmp2 */
+	movl	$1, %edx
+	addsd	%xmm5, %xmm6		/* tmp6 = tmp5 + 2^52 */
+	movsd	8+L(DP_2POW52)(%rip), %xmm4
+	movd	%xmm6, %eax		/* k = I64_LO(tmp6); */
+	addsd	%xmm6, %xmm4		/* tmp4 = tmp6 - 2^52 */
+	comisd	%xmm5, %xmm4		/* tmp4 > tmp5?  */
+	jbe	L(very_large_skip2)
+
+	/* Here if tmp4 > tmp5 */
+	subl	$1, %eax		/* k-- */
+	addsd	8+L(DP_ONES)(%rip), %xmm4 /* tmp4 -= 1.0 */
+L(very_large_skip2):
+
+	andl	%eax, %edx		/* k&1 */
+	lea	L(DP_ZERONE)(%rip), %rsi
+	subsd	%xmm4, %xmm3		/* tmp3 -= tmp4 */
+	addsd	(%rsi,%rdx,8), %xmm3	/* t  = DP_ZERONE[k&1] + tmp3 */
+	addsd	%xmm2, %xmm3		/* t += tmp2 */
+	addsd	%xmm3, %xmm0		/* t += tmp0 */
+	addl	$3, %eax		/* n=k+3 */
+	addsd	%xmm1, %xmm0		/* t += tmp1 */
+	mulsd	L(DP_PIO4)(%rip), %xmm0	/* t *= PI04 */
+
+	jmp	L(reconstruction)	/* end of very_large_args peth */
+
+	.p2align	4
+L(arg_less_pio4):
+	/* Here if |x|<Pi/4 */
+	cmpl	$0x3d000000, %eax	/* |x|<2^-5?  */
+	jl	L(arg_less_2pn5)
+
+	/* Here if 2^-5<=|x|<Pi/4 */
+	mulsd	%xmm0, %xmm0		/* y=x^2 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=x^4 */
+	movsd	L(DP_C4)(%rip), %xmm3	/* C4 */
+	mulsd	%xmm0, %xmm3		/* z*C4 */
+	movsd	L(DP_C3)(%rip), %xmm5	/* C3 */
+	mulsd	%xmm0, %xmm5		/* z*C3 */
+	addsd	L(DP_C2)(%rip), %xmm3	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm3		/* z*(C2+z*C4) */
+	addsd	L(DP_C1)(%rip), %xmm5	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm5		/* z*(C1+z*C3) */
+	addsd	L(DP_C0)(%rip), %xmm3	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm3		/* y*(C0+z*(C2+z*C4)) */
+	/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	%xmm5, %xmm3
+	/* 1.0 + y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	L(DP_ONES)(%rip), %xmm3
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(arg_less_2pn5):
+	/* Here if |x|<2^-5 */
+	cmpl	$0x32000000, %eax	/* |x|<2^-27?  */
+	jl	L(arg_less_2pn27)
+
+	/* Here if 2^-27<=|x|<2^-5 */
+	mulsd	%xmm0, %xmm0		/* DP x^2 */
+	movsd	L(DP_COS2_1)(%rip), %xmm3 /* DP DP_COS2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_COS2_1 */
+	addsd	L(DP_COS2_0)(%rip), %xmm3 /* DP DP_COS2_0+x^2*DP_COS2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_COS2_0+x^4*DP_COS2_1 */
+	/* DP 1.0+x^2*DP_COS2_0+x^4*DP_COS2_1 */
+	addsd	L(DP_ONES)(%rip), %xmm3
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(arg_less_2pn27):
+	/* Here if |x|<2^-27 */
+	andps	L(SP_ABS_MASK)(%rip),%xmm7 /* |x| */
+	movss	L(SP_ONE)(%rip), %xmm0	/* 1.0 */
+	subss	%xmm7, %xmm0		/* result is 1.0-|x| */
+	ret
+
+	.p2align	4
+L(arg_inf_or_nan):
+	/* Here if |x| is Inf or NAN */
+	jne	L(skip_errno_setting)	/* in case of x is NaN */
+
+	/* Here if x is Inf. Set errno to EDOM.  */
+	call	JUMPTARGET(__errno_location)
+	movl	$EDOM, (%rax)
+
+	.p2align	4
+L(skip_errno_setting):
+	/* Here if |x| is Inf or NAN. Continued.  */
+	movaps	%xmm7, %xmm0		/* load x */
+	subss	%xmm0, %xmm0		/* Result is NaN */
+	ret
+END(__cosf)
+
+	.section .rodata, "a"
+	.p2align 3
+L(PIO4J): /* Table of j*Pi/4, for j=0,1,..,10 */
+	.long	0x00000000,0x00000000
+	.long	0x54442d18,0x3fe921fb
+	.long	0x54442d18,0x3ff921fb
+	.long	0x7f3321d2,0x4002d97c
+	.long	0x54442d18,0x400921fb
+	.long	0x2955385e,0x400f6a7a
+	.long	0x7f3321d2,0x4012d97c
+	.long	0xe9bba775,0x4015fdbb
+	.long	0x54442d18,0x401921fb
+	.long	0xbeccb2bb,0x401c463a
+	.long	0x2955385e,0x401f6a7a
+	.type L(PIO4J), @object
+	ASM_SIZE_DIRECTIVE(L(PIO4J))
+
+	.p2align 3
+L(_FPI): /* 4/Pi broken into sum of positive DP values */
+	.long	0x00000000,0x00000000
+	.long	0x6c000000,0x3ff45f30
+	.long	0x2a000000,0x3e3c9c88
+	.long	0xa8000000,0x3c54fe13
+	.long	0xd0000000,0x3aaf47d4
+	.long	0x6c000000,0x38fbb81b
+	.long	0xe0000000,0x3714acc9
+	.long	0x7c000000,0x3560e410
+	.long	0x56000000,0x33bca2c7
+	.long	0xac000000,0x31fbd778
+	.long	0xe0000000,0x300b7246
+	.long	0xe8000000,0x2e5d2126
+	.long	0x48000000,0x2c970032
+	.long	0xe8000000,0x2ad77504
+	.long	0xe0000000,0x290921cf
+	.long	0xb0000000,0x274deb1c
+	.long	0xe0000000,0x25829a73
+	.long	0xbe000000,0x23fd1046
+	.long	0x10000000,0x2224baed
+	.long	0x8e000000,0x20709d33
+	.long	0x80000000,0x1e535a2f
+	.long	0x64000000,0x1cef904e
+	.long	0x30000000,0x1b0d6398
+	.long	0x24000000,0x1964ce7d
+	.long	0x16000000,0x17b908bf
+	.type L(_FPI), @object
+	ASM_SIZE_DIRECTIVE(L(_FPI))
+
+/* Coefficients of polynomial
+   for cos(x)~=1.0+x^2*DP_COS2_0+x^4*DP_COS2_1, |x|<2^-5.  */
+	.p2align 3
+L(DP_COS2_0):
+	.long	0xff5cc6fd,0xbfdfffff
+	.type L(DP_COS2_0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_COS2_0))
+
+	.p2align 3
+L(DP_COS2_1):
+	.long	0xb178dac5,0x3fa55514
+	.type L(DP_COS2_1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_COS2_1))
+
+	.p2align 3
+L(DP_ZERONE):
+	.long	0x00000000,0x00000000	/* 0.0 */
+	.long	0x00000000,0xbff00000	/* 1.0 */
+	.type L(DP_ZERONE), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ZERONE))
+
+	.p2align 3
+L(DP_ONES):
+	.long	0x00000000,0x3ff00000	/* +1.0 */
+	.long	0x00000000,0xbff00000	/* -1.0 */
+	.type L(DP_ONES), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ONES))
+
+/* Coefficients of polynomial
+   for sin(t)~=t+t^3*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_S3):
+	.long	0x64e6b5b4,0x3ec71d72
+	.type L(DP_S3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S3))
+
+	.p2align 3
+L(DP_S1):
+	.long	0x10c2688b,0x3f811111
+	.type L(DP_S1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S1))
+
+	.p2align 3
+L(DP_S4):
+	.long	0x1674b58a,0xbe5a947e
+	.type L(DP_S4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S4))
+
+	.p2align 3
+L(DP_S2):
+	.long	0x8b4bd1f9,0xbf2a019f
+	.type L(DP_S2),@object
+	ASM_SIZE_DIRECTIVE(L(DP_S2))
+
+	.p2align 3
+L(DP_S0):
+	.long	0x55551cd9,0xbfc55555
+	.type L(DP_S0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S0))
+
+/* Coefficients of polynomial
+   for cos(t)~=1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_C3):
+	.long	0x9ac43cc0,0x3efa00eb
+	.type L(DP_C3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C3))
+
+	.p2align 3
+L(DP_C1):
+	.long	0x545c50c7,0x3fa55555
+	.type L(DP_C1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C1))
+
+	.p2align 3
+L(DP_C4):
+	.long	0xdd8844d7,0xbe923c97
+	.type L(DP_C4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C4))
+
+	.p2align 3
+L(DP_C2):
+	.long	0x348b6874,0xbf56c16b
+	.type L(DP_C2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C2))
+
+	.p2align 3
+L(DP_C0):
+	.long	0xfffe98ae,0xbfdfffff
+	.type L(DP_C0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C0))
+
+	.p2align 3
+L(DP_PIO4):
+	.long	0x54442d18,0x3fe921fb	/* Pi/4 */
+	.type L(DP_PIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4))
+
+	.p2align 3
+L(DP_2POW52):
+	.long	0x00000000,0x43300000	/* +2^52 */
+	.long	0x00000000,0xc3300000	/* -2^52 */
+	.type L(DP_2POW52), @object
+	ASM_SIZE_DIRECTIVE(L(DP_2POW52))
+
+	.p2align 3
+L(DP_INVPIO4):
+	.long	0x6dc9c883,0x3ff45f30	/* 4/Pi */
+	.type L(DP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_INVPIO4))
+
+	.p2align 3
+L(DP_PIO4HI):
+	.long	0x54000000,0xbfe921fb	/* High part of Pi/4 */
+	.type L(DP_PIO4HI), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4HI))
+
+	.p2align 3
+L(DP_PIO4LO):
+	.long	0x11A62633,0xbe010b46	/* Low part of Pi/4 */
+	.type L(DP_PIO4LO), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4LO))
+
+	.p2align 2
+L(SP_INVPIO4):
+	.long	0x3fa2f983		/* 4/Pi */
+	.type L(SP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(SP_INVPIO4))
+
+	.p2align 4
+L(DP_ABS_MASK): /* Mask for getting DP absolute value */
+	.long	0xffffffff,0x7fffffff
+	.long	0xffffffff,0x7fffffff
+	.type L(DP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ABS_MASK))
+
+	.p2align 3
+L(DP_HI_MASK): /* Mask for getting high 21 bits of DP value */
+	.long	0x00000000,0xffffffff
+	.type L(DP_HI_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_HI_MASK))
+
+	.p2align 4
+L(SP_ABS_MASK): /* Mask for getting SP absolute value */
+	.long	0x7fffffff,0x7fffffff
+	.long	0x7fffffff,0x7fffffff
+	.type L(SP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(SP_ABS_MASK))
+
+	.p2align 2
+L(SP_ONE):
+	.long	0x3f800000		/* 1.0 */
+	.type L(SP_ONE), @object
+	ASM_SIZE_DIRECTIVE(L(SP_ONE))
+
+weak_alias(__cosf, cosf)
diff --git a/libc/sysdeps/x86_64/fpu/s_sinf.S b/libc/sysdeps/x86_64/fpu/s_sinf.S
new file mode 100644
index 000000000..9a6c87ff7
--- /dev/null
+++ b/libc/sysdeps/x86_64/fpu/s_sinf.S
@@ -0,0 +1,553 @@
+/* Optimized sinf function.
+   Copyright (C) 2012 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 <sysdep.h>
+#define __need_Emath
+#include <bits/errno.h>
+
+/* Short algorithm description:
+ *
+ *  1) if |x| == 0: return x.
+ *  2) if |x| <  2^-27: return x-x*DP_SMALL, raise underflow only when needed.
+ *  3) if |x| <  2^-5 : return x+x^3*DP_SIN2_0+x^5*DP_SIN2_1.
+ *  4) if |x| <   Pi/4: return x+x^3*(S0+x^2*(S1+x^2*(S2+x^2*(S3+x^2*S4)))).
+ *  5) if |x| < 9*Pi/4:
+ *      5.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0x0e, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      5.2) Reconstruction:
+ *          s = sign(x) * (-1.0)^((n>>2)&1)
+ *          if(n&2 != 0) {
+ *              using cos(t) polynomial for |t|<Pi/4, result is
+ *              s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4))))).
+ *          } else {
+ *              using sin(t) polynomial for |t|<Pi/4, result is
+ *              s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4))))).
+ *          }
+ *  6) if |x| < 2^23, large args:
+ *      6.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      6.2) Reconstruction same as (5.2).
+ *  7) if |x| >= 2^23, very large args:
+ *      7.1) Range reduction: k=trunc(|x|/(Pi/4)), j=(k+1)&0xfffffffe, n=k+1,
+ *           t=|x|-j*Pi/4.
+ *      7.2) Reconstruction same as (5.2).
+ *  8) if x is Inf, return x-x, and set errno=EDOM.
+ *  9) if x is NaN, return x-x.
+ *
+ * Special cases:
+ *  sin(+-0) = +-0 not raising inexact/underflow,
+ *  sin(subnormal) raises inexact/underflow,
+ *  sin(min_normalized) raises inexact/underflow,
+ *  sin(normalized) raises inexact,
+ *  sin(Inf) = NaN, raises invalid, sets errno to EDOM,
+ *  sin(NaN) = NaN.
+ */
+
+	.text
+ENTRY(__sinf)
+	/* Input: single precision x in %xmm0 */
+
+	movd	%xmm0, %eax		/* Bits of x */
+	movaps	%xmm0, %xmm7		/* Copy of x */
+	cvtss2sd %xmm0, %xmm0		/* DP x */
+	movss	L(SP_ABS_MASK)(%rip), %xmm3
+	movl	%eax, %edi		/* Copy of x bits */
+	andl	$0x7fffffff, %eax	/* |x| */
+
+	cmpl	$0x3f490fdb, %eax	/* |x|<Pi/4?  */
+	jb	L(arg_less_pio4)
+
+	/* Here if |x|>=Pi/4 */
+	andps	%xmm7, %xmm3		/* SP |x| */
+	andpd	L(DP_ABS_MASK)(%rip),%xmm0 /* DP |x| */
+	movss	L(SP_INVPIO4)(%rip), %xmm2 /* SP 1/(Pi/4) */
+
+	cmpl	$0x40e231d6, %eax	/* |x|<9*Pi/4?  */
+	jae	L(large_args)
+
+	/* Here if Pi/4<=|x|<9*Pi/4 */
+	mulss	%xmm3, %xmm2		/* SP |x|/(Pi/4) */
+	movl	%edi, %ecx		/* Load x */
+	cvttss2si %xmm2, %eax		/* k, number of Pi/4 in x */
+	lea	L(PIO4J)(%rip), %rsi
+	shrl	$31, %ecx		/* sign of x */
+	addl	$1, %eax		/* k+1 */
+	movl	$0x0e, %edx
+	andl	%eax, %edx		/* j = (k+1)&0x0e */
+	subsd	(%rsi,%rdx,8), %xmm0	/* t = |x| - j * Pi/4 */
+
+L(reconstruction):
+	/* Input: %eax=n, %xmm0=t, %ecx=sign(x) */
+	testl	$2, %eax		/* n&2 != 0?  */
+	jz	L(sin_poly)
+
+/*L(cos_poly):*/
+	/* Here if sin(x) calculated using cos(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s     * (1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))))
+	 */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	L(DP_C4)(%rip), %xmm4	/* C4 */
+	mulsd	%xmm0, %xmm4		/* z*C4 */
+	xorl	%eax, %ecx		/* (-1.0)^((n>>2)&1) XOR sign(x) */
+	movsd	L(DP_C3)(%rip), %xmm3	/* C3 */
+	mulsd	%xmm0, %xmm3		/* z*C3 */
+	lea	L(DP_ONES)(%rip), %rsi
+	addsd	L(DP_C2)(%rip), %xmm4	/* C2+z*C4 */
+	mulsd	%xmm0, %xmm4		/* z*(C2+z*C4) */
+	addsd	L(DP_C1)(%rip), %xmm3	/* C1+z*C3 */
+	mulsd	%xmm0, %xmm3		/* z*(C1+z*C3) */
+	addsd	L(DP_C0)(%rip), %xmm4	/* C0+z*(C2+z*C4) */
+	mulsd	%xmm1, %xmm4		/* y*(C0+z*(C2+z*C4)) */
+
+	/* y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	%xmm4, %xmm3
+	/* 1.0+y*(C0+y*(C1+y*(C2+y*(C3+y*C4)))) */
+	addsd	L(DP_ONES)(%rip), %xmm3
+
+	mulsd	(%rsi,%rcx,8), %xmm3	/* DP result */
+	cvtsd2ss %xmm3, %xmm0 		/* SP result */
+	ret
+
+	.p2align	4
+L(sin_poly):
+	/* Here if sin(x) calculated using sin(t) polynomial for |t|<Pi/4:
+	 * y = t*t; z = y*y;
+	 * s = sign(x) * (-1.0)^((n>>2)&1)
+	 * result = s * t * (1.0+t^2*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))))
+	 */
+
+	movaps	%xmm0, %xmm4		/* t */
+	shrl	$2, %eax		/* n>>2 */
+	mulsd	%xmm0, %xmm0		/* y=t^2 */
+	andl	$1, %eax		/* (n>>2)&1 */
+	movaps	%xmm0, %xmm1		/* y */
+	xorl	%eax, %ecx		/* (-1.0)^((n>>2)&1) XOR sign(x) */
+	mulsd	%xmm0, %xmm0		/* z=t^4 */
+
+	movsd	L(DP_S4)(%rip), %xmm2	/* S4 */
+	mulsd	%xmm0, %xmm2		/* z*S4 */
+	movsd	L(DP_S3)(%rip), %xmm3	/* S3 */
+	mulsd	%xmm0, %xmm3		/* z*S3 */
+	lea	L(DP_ONES)(%rip), %rsi
+	addsd	L(DP_S2)(%rip), %xmm2	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm2		/* z*(S2+z*S4) */
+	addsd	L(DP_S1)(%rip), %xmm3	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm3		/* z*(S1+z*S3) */
+	addsd	L(DP_S0)(%rip), %xmm2	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm2		/* y*(S0+z*(S2+z*S4)) */
+	/* t*s, where s = sign(x) * (-1.0)^((n>>2)&1) */
+	mulsd	(%rsi,%rcx,8), %xmm4
+	/* y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm2, %xmm3
+	/* t*s*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	mulsd	%xmm4, %xmm3
+	/* t*s*(1.0+y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	cvtsd2ss %xmm3, %xmm0 		/* SP result */
+	ret
+
+	.p2align	4
+L(large_args):
+	/* Here if |x|>=9*Pi/4 */
+	cmpl	$0x7f800000, %eax	/* x is Inf or NaN?  */
+	jae	L(arg_inf_or_nan)
+
+	/* Here if finite |x|>=9*Pi/4 */
+	cmpl	$0x4b000000, %eax	/* |x|<2^23?  */
+	jae	L(very_large_args)
+
+	/* Here if 9*Pi/4<=|x|<2^23 */
+	movsd	L(DP_INVPIO4)(%rip), %xmm1 /* 1/(Pi/4) */
+	mulsd	%xmm0, %xmm1		/* |x|/(Pi/4) */
+	cvttsd2si %xmm1, %eax		/* k=trunc(|x|/(Pi/4)) */
+	addl	$1, %eax		/* k+1 */
+	movl	%eax, %edx
+	andl	$0xfffffffe, %edx	/* j=(k+1)&0xfffffffe */
+	cvtsi2sdl %edx, %xmm4		/* DP j */
+	movl	%edi, %ecx		/* Load x */
+	movsd	L(DP_PIO4HI)(%rip), %xmm2 /* -PIO4HI = high part of -Pi/4 */
+	shrl	$31, %ecx		/* sign bit of x */
+	mulsd	%xmm4, %xmm2		/* -j*PIO4HI */
+	movsd	L(DP_PIO4LO)(%rip), %xmm3 /* -PIO4LO = low part of -Pi/4 */
+	addsd	%xmm2, %xmm0		/* |x| - j*PIO4HI */
+	mulsd	%xmm3, %xmm4		/* j*PIO4LO */
+	addsd	%xmm4, %xmm0		/* t = |x| - j*PIO4HI - j*PIO4LO */
+	jmp	L(reconstruction)
+
+	.p2align	4
+L(very_large_args):
+	/* Here if finite |x|>=2^23 */
+
+	/* bitpos = (ix>>23) - BIAS_32 + 59; */
+	shrl	$23, %eax		/* eb = biased exponent of x */
+	/* bitpos = eb - 0x7f + 59, where 0x7f is exponent bias */
+	subl	$68, %eax
+	movl	$28, %ecx		/* %cl=28 */
+	movl	%eax, %edx		/* bitpos copy */
+
+	/* j = bitpos/28; */
+	div	%cl			/* j in register %al=%ax/%cl */
+	movapd	%xmm0, %xmm3		/* |x| */
+	/* clear unneeded remainder from %ah */
+	andl	$0xff, %eax
+
+	imull	$28, %eax, %ecx		/* j*28 */
+	lea	L(_FPI)(%rip), %rsi
+	movsd	L(DP_HI_MASK)(%rip), %xmm4 /* DP_HI_MASK */
+	movapd	%xmm0, %xmm5		/* |x| */
+	mulsd	-16(%rsi,%rax,8), %xmm3	/* tmp3 = FPI[j-2]*|x| */
+	movapd	%xmm0, %xmm1		/* |x| */
+	mulsd	-8(%rsi,%rax,8), %xmm5	/* tmp2 = FPI[j-1]*|x| */
+	mulsd	(%rsi,%rax,8), %xmm0	/* tmp0 = FPI[j]*|x| */
+	addl	$19, %ecx		/* j*28+19 */
+	mulsd	8(%rsi,%rax,8), %xmm1	/* tmp1 = FPI[j+1]*|x| */
+	cmpl	%ecx, %edx		/* bitpos>=j*28+19?  */
+	jl	L(very_large_skip1)
+
+	/* Here if bitpos>=j*28+19 */
+	andpd	%xmm3, %xmm4		/* HI(tmp3) */
+	subsd	%xmm4, %xmm3		/* tmp3 = tmp3 - HI(tmp3) */
+L(very_large_skip1):
+
+	movsd	L(DP_2POW52)(%rip), %xmm6
+	movapd	%xmm5, %xmm2		/* tmp2 copy */
+	addsd	%xmm3, %xmm5		/* tmp5 = tmp3 + tmp2 */
+	movl	$1, %edx
+	addsd	%xmm5, %xmm6		/* tmp6 = tmp5 + 2^52 */
+	movsd	8+L(DP_2POW52)(%rip), %xmm4
+	movd	%xmm6, %eax		/* k = I64_LO(tmp6); */
+	addsd	%xmm6, %xmm4		/* tmp4 = tmp6 - 2^52 */
+	movl	%edi, %ecx		/* Load x */
+	comisd	%xmm5, %xmm4		/* tmp4 > tmp5?  */
+	jbe	L(very_large_skip2)
+
+	/* Here if tmp4 > tmp5 */
+	subl	$1, %eax		/* k-- */
+	addsd	8+L(DP_ONES)(%rip), %xmm4 /* tmp4 -= 1.0 */
+L(very_large_skip2):
+
+	andl	%eax, %edx		/* k&1 */
+	lea	L(DP_ZERONE)(%rip), %rsi
+	subsd	%xmm4, %xmm3		/* tmp3 -= tmp4 */
+	addsd	(%rsi,%rdx,8), %xmm3	/* t  = DP_ZERONE[k&1] + tmp3 */
+	addsd	%xmm2, %xmm3		/* t += tmp2 */
+	shrl	$31, %ecx		/* sign of x */
+	addsd	%xmm3, %xmm0		/* t += tmp0 */
+	addl	$1, %eax		/* n=k+1 */
+	addsd	%xmm1, %xmm0		/* t += tmp1 */
+	mulsd	L(DP_PIO4)(%rip), %xmm0	/* t *= PI04 */
+
+	jmp	L(reconstruction)	/* end of very_large_args peth */
+
+	.p2align	4
+L(arg_less_pio4):
+	/* Here if |x|<Pi/4 */
+	cmpl	$0x3d000000, %eax	/* |x|<2^-5?  */
+	jl	L(arg_less_2pn5)
+
+	/* Here if 2^-5<=|x|<Pi/4 */
+	movaps	%xmm0, %xmm3		/* x */
+	mulsd	%xmm0, %xmm0		/* y=x^2 */
+	movaps	%xmm0, %xmm1		/* y */
+	mulsd	%xmm0, %xmm0		/* z=x^4 */
+	movsd	L(DP_S4)(%rip), %xmm4	/* S4 */
+	mulsd	%xmm0, %xmm4		/* z*S4 */
+	movsd	L(DP_S3)(%rip), %xmm5	/* S3 */
+	mulsd	%xmm0, %xmm5		/* z*S3 */
+	addsd	L(DP_S2)(%rip), %xmm4	/* S2+z*S4 */
+	mulsd	%xmm0, %xmm4		/* z*(S2+z*S4) */
+	addsd	L(DP_S1)(%rip), %xmm5	/* S1+z*S3 */
+	mulsd	%xmm0, %xmm5		/* z*(S1+z*S3) */
+	addsd	L(DP_S0)(%rip), %xmm4	/* S0+z*(S2+z*S4) */
+	mulsd	%xmm1, %xmm4		/* y*(S0+z*(S2+z*S4)) */
+	mulsd	%xmm3, %xmm5		/* x*z*(S1+z*S3) */
+	mulsd	%xmm3, %xmm4		/* x*y*(S0+z*(S2+z*S4)) */
+	/* x*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm5, %xmm4
+	/* x + x*y*(S0+y*(S1+y*(S2+y*(S3+y*S4)))) */
+	addsd	%xmm4, %xmm3
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(arg_less_2pn5):
+	/* Here if |x|<2^-5 */
+	cmpl	$0x32000000, %eax	/* |x|<2^-27?  */
+	jl	L(arg_less_2pn27)
+
+	/* Here if 2^-27<=|x|<2^-5 */
+	movaps	%xmm0, %xmm1		/* DP x */
+	mulsd	%xmm0, %xmm0		/* DP x^2 */
+	movsd	L(DP_SIN2_1)(%rip), %xmm3 /* DP DP_SIN2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_SIN2_1 */
+	addsd	L(DP_SIN2_0)(%rip), %xmm3 /* DP DP_SIN2_0+x^2*DP_SIN2_1 */
+	mulsd	%xmm0, %xmm3		/* DP x^2*DP_SIN2_0+x^4*DP_SIN2_1 */
+	mulsd	%xmm1, %xmm3		/* DP x^3*DP_SIN2_0+x^5*DP_SIN2_1 */
+	addsd	%xmm1, %xmm3		/* DP x+x^3*DP_SIN2_0+x^5*DP_SIN2_1 */
+	cvtsd2ss %xmm3, %xmm0		/* SP result */
+	ret
+
+	.p2align	4
+L(arg_less_2pn27):
+	cmpl	$0, %eax		/* x=0?  */
+	je	L(arg_zero)		/* in case x=0 return sin(+-0)==+-0 */
+	/* Here if |x|<2^-27 */
+	/*
+	 * Special cases here:
+	 *  sin(subnormal) raises inexact/underflow
+	 *  sin(min_normalized) raises inexact/underflow
+	 *  sin(normalized) raises inexact
+	 */
+	movaps	%xmm0, %xmm3		/* Copy of DP x */
+	mulsd	L(DP_SMALL)(%rip), %xmm0 /* x*DP_SMALL */
+	subsd	%xmm0, %xmm3		/* Result is x-x*DP_SMALL */
+	cvtsd2ss %xmm3, %xmm0		/* Result converted to SP */
+	ret
+
+	.p2align	4
+L(arg_zero):
+	movaps	%xmm7, %xmm0		/* SP x */
+	ret
+
+	.p2align	4
+L(arg_inf_or_nan):
+	/* Here if |x| is Inf or NAN */
+	jne	L(skip_errno_setting)	/* in case of x is NaN */
+
+	/* Here if x is Inf. Set errno to EDOM.  */
+	call	JUMPTARGET(__errno_location)
+	movl	$EDOM, (%rax)
+
+	.p2align	4
+L(skip_errno_setting):
+	/* Here if |x| is Inf or NAN. Continued.  */
+	movaps	%xmm7, %xmm0		/* load x */
+	subss	%xmm0, %xmm0		/* Result is NaN */
+	ret
+END(__sinf)
+
+	.section .rodata, "a"
+	.p2align 3
+L(PIO4J): /* Table of j*Pi/4, for j=0,1,..,10 */
+	.long	0x00000000,0x00000000
+	.long	0x54442d18,0x3fe921fb
+	.long	0x54442d18,0x3ff921fb
+	.long	0x7f3321d2,0x4002d97c
+	.long	0x54442d18,0x400921fb
+	.long	0x2955385e,0x400f6a7a
+	.long	0x7f3321d2,0x4012d97c
+	.long	0xe9bba775,0x4015fdbb
+	.long	0x54442d18,0x401921fb
+	.long	0xbeccb2bb,0x401c463a
+	.long	0x2955385e,0x401f6a7a
+	.type L(PIO4J), @object
+	ASM_SIZE_DIRECTIVE(L(PIO4J))
+
+	.p2align 3
+L(_FPI): /* 4/Pi broken into sum of positive DP values */
+	.long	0x00000000,0x00000000
+	.long	0x6c000000,0x3ff45f30
+	.long	0x2a000000,0x3e3c9c88
+	.long	0xa8000000,0x3c54fe13
+	.long	0xd0000000,0x3aaf47d4
+	.long	0x6c000000,0x38fbb81b
+	.long	0xe0000000,0x3714acc9
+	.long	0x7c000000,0x3560e410
+	.long	0x56000000,0x33bca2c7
+	.long	0xac000000,0x31fbd778
+	.long	0xe0000000,0x300b7246
+	.long	0xe8000000,0x2e5d2126
+	.long	0x48000000,0x2c970032
+	.long	0xe8000000,0x2ad77504
+	.long	0xe0000000,0x290921cf
+	.long	0xb0000000,0x274deb1c
+	.long	0xe0000000,0x25829a73
+	.long	0xbe000000,0x23fd1046
+	.long	0x10000000,0x2224baed
+	.long	0x8e000000,0x20709d33
+	.long	0x80000000,0x1e535a2f
+	.long	0x64000000,0x1cef904e
+	.long	0x30000000,0x1b0d6398
+	.long	0x24000000,0x1964ce7d
+	.long	0x16000000,0x17b908bf
+	.type L(_FPI), @object
+	ASM_SIZE_DIRECTIVE(L(_FPI))
+
+/* Coefficients of polynomial
+   for sin(x)~=x+x^3*DP_SIN2_0+x^5*DP_SIN2_1, |x|<2^-5.  */
+	.p2align 3
+L(DP_SIN2_0):
+	.long	0x5543d49d,0xbfc55555
+	.type L(DP_SIN2_0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SIN2_0))
+
+	.p2align 3
+L(DP_SIN2_1):
+	.long	0x75cec8c5,0x3f8110f4
+	.type L(DP_SIN2_1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SIN2_1))
+
+	.p2align 3
+L(DP_ZERONE):
+	.long	0x00000000,0x00000000	/* 0.0 */
+	.long	0x00000000,0xbff00000	/* 1.0 */
+	.type L(DP_ZERONE), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ZERONE))
+
+	.p2align 3
+L(DP_ONES):
+	.long	0x00000000,0x3ff00000	/* +1.0 */
+	.long	0x00000000,0xbff00000	/* -1.0 */
+	.type L(DP_ONES), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ONES))
+
+/* Coefficients of polynomial
+   for sin(t)~=t+t^3*(S0+t^2*(S1+t^2*(S2+t^2*(S3+t^2*S4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_S3):
+	.long	0x64e6b5b4,0x3ec71d72
+	.type L(DP_S3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S3))
+
+	.p2align 3
+L(DP_S1):
+	.long	0x10c2688b,0x3f811111
+	.type L(DP_S1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S1))
+
+	.p2align 3
+L(DP_S4):
+	.long	0x1674b58a,0xbe5a947e
+	.type L(DP_S4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S4))
+
+	.p2align 3
+L(DP_S2):
+	.long	0x8b4bd1f9,0xbf2a019f
+	.type L(DP_S2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S2))
+
+	.p2align 3
+L(DP_S0):
+	.long	0x55551cd9,0xbfc55555
+	.type L(DP_S0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_S0))
+
+	.p2align 3
+L(DP_SMALL):
+	.long	0x00000000,0x3cd00000	/* 2^(-50) */
+	.type L(DP_SMALL), @object
+	ASM_SIZE_DIRECTIVE(L(DP_SMALL))
+
+/* Coefficients of polynomial
+   for cos(t)~=1.0+t^2*(C0+t^2*(C1+t^2*(C2+t^2*(C3+t^2*C4)))), |t|<Pi/4.  */
+	.p2align 3
+L(DP_C3):
+	.long	0x9ac43cc0,0x3efa00eb
+	.type L(DP_C3), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C3))
+
+	.p2align 3
+L(DP_C1):
+	.long	0x545c50c7,0x3fa55555
+	.type L(DP_C1), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C1))
+
+	.p2align 3
+L(DP_C4):
+	.long	0xdd8844d7,0xbe923c97
+	.type L(DP_C4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C4))
+
+	.p2align 3
+L(DP_C2):
+	.long	0x348b6874,0xbf56c16b
+	.type L(DP_C2), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C2))
+
+	.p2align 3
+L(DP_C0):
+	.long	0xfffe98ae,0xbfdfffff
+	.type L(DP_C0), @object
+	ASM_SIZE_DIRECTIVE(L(DP_C0))
+
+	.p2align 3
+L(DP_PIO4):
+	.long	0x54442d18,0x3fe921fb	/* Pi/4 */
+	.type L(DP_PIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4))
+
+	.p2align 3
+L(DP_2POW52):
+	.long	0x00000000,0x43300000	/* +2^52 */
+	.long	0x00000000,0xc3300000	/* -2^52 */
+	.type L(DP_2POW52), @object
+	ASM_SIZE_DIRECTIVE(L(DP_2POW52))
+
+	.p2align 3
+L(DP_INVPIO4):
+	.long	0x6dc9c883,0x3ff45f30	/* 4/Pi */
+	.type L(DP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(DP_INVPIO4))
+
+	.p2align 3
+L(DP_PIO4HI):
+	.long	0x54000000,0xbfe921fb	/* High part of Pi/4 */
+	.type L(DP_PIO4HI), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4HI))
+
+	.p2align 3
+L(DP_PIO4LO):
+	.long	0x11A62633,0xbe010b46	/* Low part of Pi/4 */
+	.type L(DP_PIO4LO), @object
+	ASM_SIZE_DIRECTIVE(L(DP_PIO4LO))
+
+	.p2align 2
+L(SP_INVPIO4):
+	.long	0x3fa2f983		/* 4/Pi */
+	.type L(SP_INVPIO4), @object
+	ASM_SIZE_DIRECTIVE(L(SP_INVPIO4))
+
+	.p2align 4
+L(DP_ABS_MASK): /* Mask for getting DP absolute value */
+	.long	0xffffffff,0x7fffffff
+	.long	0xffffffff,0x7fffffff
+	.type L(DP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(DP_ABS_MASK))
+
+	.p2align 3
+L(DP_HI_MASK): /* Mask for getting high 21 bits of DP value */
+	.long	0x00000000,0xffffffff
+	.type L(DP_HI_MASK),@object
+	ASM_SIZE_DIRECTIVE(L(DP_HI_MASK))
+
+	.p2align 4
+L(SP_ABS_MASK): /* Mask for getting SP absolute value */
+	.long	0x7fffffff,0x7fffffff
+	.long	0x7fffffff,0x7fffffff
+	.type L(SP_ABS_MASK), @object
+	ASM_SIZE_DIRECTIVE(L(SP_ABS_MASK))
+
+weak_alias(__sinf, sinf)
diff --git a/libc/sysdeps/x86_64/tst-xmmymm.sh b/libc/sysdeps/x86_64/tst-xmmymm.sh
index da8af7e68..64efc6e02 100755
--- a/libc/sysdeps/x86_64/tst-xmmymm.sh
+++ b/libc/sysdeps/x86_64/tst-xmmymm.sh
@@ -1,4 +1,24 @@
 #! /bin/bash
+# Make sure no code in ld.so uses xmm/ymm registers on x86-64.
+# Copyright (C) 2009-2012 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/>.
+
+set -e
+
 objpfx="$1"
 
 tmp=$(mktemp ${objpfx}tst-xmmymm.XXXXXX)