Add an optimised version of the float and double helper functions. These

are only used on armv6 when the vfp unit is detected. They will also be
available for the upcoming armv6hf platform, however while not used by
default there will need to be defined for any software that calls them
directly.

6 files changed, 537 insertions, 124 deletions

diff --git a/lib/libc/arm/aeabi/Makefile.inc b/lib/libc/arm/aeabi/Makefile.inc
index 7d8e242..46ca764 100644
--- a/lib/libc/arm/aeabi/Makefile.inc
+++ b/lib/libc/arm/aeabi/Makefile.inc
@@ -6,6 +6,10 @@ SRCS+=	aeabi_atexit.c		\
 	aeabi_double.c		\
 	aeabi_float.c		\
 	aeabi_unwind_cpp.c
+.if ${MACHINE_ARCH:Marmv6*}
+SRCS+=	aeabi_vfp_double.S	\
+	aeabi_vfp_float.S
+.endif
 
 # Add the aeabi_mem* functions. While they live in compiler-rt they call into
 # libc. This causes issues when other parts of libc call these functions.
diff --git a/lib/libc/arm/aeabi/aeabi_double.c b/lib/libc/arm/aeabi/aeabi_double.c
index 5f9065c..a32f6a1 100644
--- a/lib/libc/arm/aeabi/aeabi_double.c
+++ b/lib/libc/arm/aeabi/aeabi_double.c
@@ -32,70 +32,45 @@ __FBSDID("$FreeBSD$");
 #include "milieu.h"
 #include "softfloat.h"
 
-flag __unorddf2(float64, float64);
-
-int __aeabi_dcmpeq(float64 a, float64 b)
-{
-	return float64_eq(a, b);
-}
-
-int __aeabi_dcmplt(float64 a, float64 b)
-{
-	return float64_lt(a, b);
-}
-
-int __aeabi_dcmple(float64 a, float64 b)
-{
-	return float64_le(a, b);
-}
-
-int __aeabi_dcmpge(float64 a, float64 b)
-{
-	return float64_le(b, a);
-}
-
-int __aeabi_dcmpgt(float64 a, float64 b)
-{
-	return float64_lt(b, a);
-}
-
-int __aeabi_dcmpun(float64 a, float64 b)
-{
-	return __unorddf2(a, b);
-}
+#include "aeabi_vfp.h"
 
-int __aeabi_d2iz(float64 a)
-{
-	return float64_to_int32_round_to_zero(a);
-}
+extern int _libc_arm_fpu_present;
 
-float32 __aeabi_d2f(float64 a)
-{
-	return float64_to_float32(a);
-}
-
-float64 __aeabi_i2d(int a)
-{
-	return int32_to_float64(a);
-}
-
-float64 __aeabi_dadd(float64 a, float64 b)
-{
-	return float64_add(a, b);
-}
-
-float64 __aeabi_ddiv(float64 a, float64 b)
-{
-	return float64_div(a, b);
-}
+flag __unorddf2(float64, float64);
 
-float64 __aeabi_dmul(float64 a, float64 b)
-{
-	return float64_mul(a, b);
-}
+/* These are written in asm and are only called from this file */
+int __aeabi_dcmpeq_vfp(float64, float64);
+int __aeabi_dcmplt_vfp(float64, float64);
+int __aeabi_dcmple_vfp(float64, float64);
+int __aeabi_dcmpgt_vfp(float64, float64);
+int __aeabi_dcmpge_vfp(float64, float64);
+int __aeabi_dcmpun_vfp(float64, float64);
+int __aeabi_d2iz_vfp(float64);
+float32 __aeabi_d2f_vfp(float64);
+float64 __aeabi_i2d_vfp(int);
+float64 __aeabi_dadd_vfp(float64, float64);
+float64 __aeabi_ddiv_vfp(float64, float64);
+float64 __aeabi_dmul_vfp(float64, float64);
+float64 __aeabi_dsub_vfp(float64, float64);
 
-float64 __aeabi_dsub(float64 a, float64 b)
-{
-	return float64_sub(a, b);
-}
+/*
+ * Depending on the target these will:
+ *  On armv6 with a vfp call the above function, or
+ *  Call the softfloat cunstion in the 3rd argument.
+ */
+int AEABI_FUNC2(dcmpeq, float64, float64_eq)
+int AEABI_FUNC2(dcmplt, float64, float64_lt)
+int AEABI_FUNC2(dcmple, float64, float64_le)
+int AEABI_FUNC2_REV(dcmpge, float64, float64_le)
+int AEABI_FUNC2_REV(dcmpgt, float64, float64_lt)
+int AEABI_FUNC2(dcmpun, float64, __unorddf2)
+
+int AEABI_FUNC(d2iz, float64, float64_to_int32_round_to_zero)
+float32 AEABI_FUNC(d2f, float64, float64_to_float32)
+float64 AEABI_FUNC(i2d, int, int32_to_float64)
+
+float64 AEABI_FUNC2(dadd, float64, float64_add)
+float64 AEABI_FUNC2(ddiv, float64, float64_div)
+float64 AEABI_FUNC2(dmul, float64, float64_mul)
+float64 AEABI_FUNC2(dsub, float64, float64_sub)
 
diff --git a/lib/libc/arm/aeabi/aeabi_float.c b/lib/libc/arm/aeabi/aeabi_float.c
index 97751ad..f8d8cbc 100644
--- a/lib/libc/arm/aeabi/aeabi_float.c
+++ b/lib/libc/arm/aeabi/aeabi_float.c
@@ -32,70 +32,45 @@ __FBSDID("$FreeBSD$");
 #include "milieu.h"
 #include "softfloat.h"
 
-flag __unordsf2(float32, float32);
-
-int __aeabi_fcmpeq(float32 a, float32 b)
-{
-	return float32_eq(a, b);
-}
-
-int __aeabi_fcmplt(float32 a, float32 b)
-{
-	return float32_lt(a, b);
-}
-
-int __aeabi_fcmple(float32 a, float32 b)
-{
-	return float32_le(a, b);
-}
-
-int __aeabi_fcmpge(float32 a, float32 b)
-{
-	return float32_le(b, a);
-}
-
-int __aeabi_fcmpgt(float32 a, float32 b)
-{
-	return float32_lt(b, a);
-}
-
-int __aeabi_fcmpun(float32 a, float32 b)
-{
-	return __unordsf2(a, b);
-}
+#include "aeabi_vfp.h"
 
-int __aeabi_f2iz(float32 a)
-{
-	return float32_to_int32_round_to_zero(a);
-}
+extern int _libc_arm_fpu_present;
 
-float32 __aeabi_f2d(float32 a)
-{
-	return float32_to_float64(a);
-}
-
-float32 __aeabi_i2f(int a)
-{
-	return int32_to_float32(a);
-}
-
-float32 __aeabi_fadd(float32 a, float32 b)
-{
-	return float32_add(a, b);
-}
-
-float32 __aeabi_fdiv(float32 a, float32 b)
-{
-	return float32_div(a, b);
-}
+flag __unordsf2(float32, float32);
 
-float32 __aeabi_fmul(float32 a, float32 b)
-{
-	return float32_mul(a, b);
-}
+/* These are written in asm and are only called from this file */
+int __aeabi_fcmpeq_vfp(float32, float32);
+int __aeabi_fcmplt_vfp(float32, float32);
+int __aeabi_fcmple_vfp(float32, float32);
+int __aeabi_fcmpgt_vfp(float32, float32);
+int __aeabi_fcmpge_vfp(float32, float32);
+int __aeabi_fcmpun_vfp(float32, float32);
+int __aeabi_f2iz_vfp(float32);
+float64 __aeabi_f2d_vfp(float32);
+float32 __aeabi_i2f_vfp(int);
+float32 __aeabi_fadd_vfp(float32, float32);
+float32 __aeabi_fdiv_vfp(float32, float32);
+float32 __aeabi_fmul_vfp(float32, float32);
+float32 __aeabi_fsub_vfp(float32, float32);
 
-float32 __aeabi_fsub(float32 a, float32 b)
-{
-	return float32_sub(a, b);
-}
+/*
+ * Depending on the target these will:
+ *  On armv6 with a vfp call the above function, or
+ *  Call the softfloat cunstion in the 3rd argument.
+ */
+int AEABI_FUNC2(fcmpeq, float32, float32_eq)
+int AEABI_FUNC2(fcmplt, float32, float32_lt)
+int AEABI_FUNC2(fcmple, float32, float32_le)
+int AEABI_FUNC2_REV(fcmpge, float32, float32_le)
+int AEABI_FUNC2_REV(fcmpgt, float32, float32_lt)
+int AEABI_FUNC2(fcmpun, float32, __unordsf2)
+
+int AEABI_FUNC(f2iz, float32, float32_to_int32_round_to_zero)
+float64 AEABI_FUNC(f2d, float32, float32_to_float64)
+float32 AEABI_FUNC(i2f, int, int32_to_float32)
+
+float32 AEABI_FUNC2(fadd, float32, float32_add)
+float32 AEABI_FUNC2(fdiv, float32, float32_div)
+float32 AEABI_FUNC2(fmul, float32, float32_mul)
+float32 AEABI_FUNC2(fsub, float32, float32_sub)
 
diff --git a/lib/libc/arm/aeabi/aeabi_vfp.h b/lib/libc/arm/aeabi/aeabi_vfp.h
new file mode 100644
index 0000000..e3872e5
--- /dev/null
+++ b/lib/libc/arm/aeabi/aeabi_vfp.h
@@ -0,0 +1,129 @@
+/*
+ * Copyright (C) 2013 Andrew Turner
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ *
+ */
+
+#ifndef AEABI_VFP_H
+#define	AEABI_VFP_H
+
+/*
+ * ASM helper macros. These allow the functions to be changed depending on
+ * the endian-ness we are building for.
+ */
+
+/* Allow the name of the function to be changed depending on the ABI */
+#ifndef __ARM_PCS_VFP
+#define	AEABI_ENTRY(x)	ENTRY(__aeabi_ ## x ## _vfp)
+#define	AEABI_END(x)	END(__aeabi_ ## x ## _vfp)
+#else
+#define	AEABI_ENTRY(x)	ENTRY(__aeabi_ ## x)
+#define	AEABI_END(x)	END(__aeabi_ ## x)
+#endif
+
+/*
+ * These should be used when a function either takes, or returns a floating
+ * point falue. They will load the data from an ARM to a VFP register(s),
+ * or from a VFP to an ARM register
+ */
+#ifdef __ARMEB__
+#define	LOAD_DREG(vreg, reg0, reg1)   vmov vreg, reg1, reg0
+#define	UNLOAD_DREG(reg0, reg1, vreg) vmov reg1, reg0, vreg
+#else
+#define	LOAD_DREG(vreg, reg0, reg1)   vmov vreg, reg0, reg1
+#define	UNLOAD_DREG(reg0, reg1, vreg) vmov reg0, reg1, vreg
+#endif
+
+#define	LOAD_SREGS(vreg0, vreg1, reg0, reg1) vmov vreg0, vreg1, reg0, reg1
+#define	LOAD_SREG(vreg, reg)                 vmov vreg, reg
+#define	UNLOAD_SREG(reg, vreg)               vmov reg, vreg
+
+/*
+ * C Helper macros
+ */
+
+#if 1 && defined(__FreeBSD_ARCH_armv6__) || (defined(__ARM_ARCH) && __ARM_ARCH >= 6)
+/*
+ * Generate a function that will either call into the VFP implementation,
+ * or the soft float version for a given __aeabi_* helper. The function
+ * will take a single argument of the type given by in_type.
+ */
+#define	AEABI_FUNC(name, in_type, soft_func)			\
+__aeabi_ ## name(in_type a)					\
+{								\
+	if (_libc_arm_fpu_present)				\
+		return __aeabi_ ## name ## _vfp(a);		\
+	else							\
+		return soft_func (a);				\
+}
+
+/* As above, but takes two arguments of the same type */
+#define	AEABI_FUNC2(name, in_type, soft_func)			\
+__aeabi_ ## name(in_type a, in_type b)				\
+{								\
+	if (_libc_arm_fpu_present)				\
+		return __aeabi_ ## name ## _vfp(a, b);	\
+	else							\
+		return soft_func (a, b);			\
+}
+
+/* As above, but with the soft float arguments reversed */
+#define	AEABI_FUNC2_REV(name, in_type, soft_func)		\
+__aeabi_ ## name(in_type a, in_type b)				\
+{								\
+	if (_libc_arm_fpu_present)				\
+		return __aeabi_ ## name ## _vfp(a, b);	\
+	else							\
+		return soft_func (b, a);			\
+}
+#else
+/*
+ * Helper macros for when we are only able to use the softfloat
+ * version of these functions, i.e. on arm before armv6.
+ */
+#define	AEABI_FUNC(name, in_type, soft_func)			\
+__aeabi_ ## name(in_type a)					\
+{								\
+	return soft_func (a);					\
+}
+
+/* As above, but takes two arguments of the same type */
+#define	AEABI_FUNC2(name, in_type, soft_func)			\
+__aeabi_ ## name(in_type a, in_type b)				\
+{								\
+	return soft_func (a, b);				\
+}
+
+/* As above, but with the soft float arguments reversed */
+#define	AEABI_FUNC2_REV(name, in_type, soft_func)		\
+__aeabi_ ## name(in_type a, in_type b)				\
+{								\
+	return soft_func (b, a);				\
+}
+#endif
+
+#endif
+
diff --git a/lib/libc/arm/aeabi/aeabi_vfp_double.S b/lib/libc/arm/aeabi/aeabi_vfp_double.S
new file mode 100644
index 0000000..842412b
--- /dev/null
+++ b/lib/libc/arm/aeabi/aeabi_vfp_double.S
@@ -0,0 +1,170 @@
+/*
+ * Copyright (C) 2013 Andrew Turner
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ */
+
+#include <machine/asm.h>
+__FBSDID("$FreeBSD$");
+
+#include "aeabi_vfp.h"
+
+.fpu	vfp
+.syntax	unified
+
+/* int __aeabi_dcmpeq(double, double) */
+AEABI_ENTRY(dcmpeq)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movne    r0, #0
+	moveq    r0, #1
+	RET
+AEABI_END(dcmpeq)
+
+/* int __aeabi_dcmplt(double, double) */
+AEABI_ENTRY(dcmplt)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movcs    r0, #0
+	movlt    r0, #1
+	RET
+AEABI_END(dcmplt)
+
+/* int __aeabi_dcmple(double, double) */
+AEABI_ENTRY(dcmple)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movhi    r0, #0
+	movls    r0, #1
+	RET
+AEABI_END(dcmple)
+
+/* int __aeabi_dcmpge(double, double) */
+AEABI_ENTRY(dcmpge)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movlt    r0, #0
+	movge    r0, #1
+	RET
+AEABI_END(dcmpge)
+
+/* int __aeabi_dcmpgt(double, double) */
+AEABI_ENTRY(dcmpgt)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movle    r0, #0
+	movgt    r0, #1
+	RET
+AEABI_END(dcmpgt)
+
+/* int __aeabi_dcmpun(double, double) */
+AEABI_ENTRY(dcmpun)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vcmp.f64 d0, d1
+	vmrs     APSR_nzcv, fpscr
+	movvc    r0, #0
+	movvs    r0, #1
+	RET
+AEABI_END(dcmpun)
+
+/* int __aeabi_d2iz(double) */
+AEABI_ENTRY(d2iz)
+	LOAD_DREG(d0, r0, r1)
+#if 0
+	/*
+	 * This should be the correct instruction, but binutils incorrectly
+	 * encodes it as the version that used FPSCR to determine the rounding.
+	 * When binutils is fixed we can use this again.
+	 */
+	vcvt.s32.f64 s0, d0
+#else
+	ftosizd s0, d0
+#endif
+	vmov         r0, s0
+	RET
+AEABI_END(d2iz)
+
+/* float __aeabi_d2f(double) */
+AEABI_ENTRY(d2f)
+	LOAD_DREG(d0, r0, r1)
+	vcvt.f32.f64 s0, d0
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(d2f)
+
+/* double __aeabi_i2d(int) */
+AEABI_ENTRY(i2d)
+	vmov         s0, r0
+	vcvt.f64.s32 d0, s0
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(i2d)
+
+/* double __aeabi_dadd(double, double) */
+AEABI_ENTRY(dadd)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vadd.f64 d0, d0, d1
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(dadd)
+
+/* double __aeabi_ddiv(double, double) */
+AEABI_ENTRY(ddiv)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vdiv.f64 d0, d0, d1
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(ddiv)
+
+/* double __aeabi_dmul(double, double) */
+AEABI_ENTRY(dmul)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vmul.f64 d0, d0, d1
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(dmul)
+
+/* double __aeabi_dsub(double, double) */
+AEABI_ENTRY(dsub)
+	LOAD_DREG(d0, r0, r1)
+	LOAD_DREG(d1, r2, r3)
+	vsub.f64 d0, d0, d1
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(dsub)
+
diff --git a/lib/libc/arm/aeabi/aeabi_vfp_float.S b/lib/libc/arm/aeabi/aeabi_vfp_float.S
new file mode 100644
index 0000000..d81b2b2
--- /dev/null
+++ b/lib/libc/arm/aeabi/aeabi_vfp_float.S
@@ -0,0 +1,160 @@
+/*
+ * Copyright (C) 2013 Andrew Turner
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ */
+
+#include <machine/asm.h>
+__FBSDID("$FreeBSD$");
+
+#include "aeabi_vfp.h"
+
+.fpu	vfp
+.syntax	unified
+
+/* int __aeabi_fcmpeq(float, float) */
+AEABI_ENTRY(fcmpeq)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movne    r0, #0
+	moveq    r0, #1
+	RET
+AEABI_END(fcmpeq)
+
+/* int __aeabi_fcmplt(float, float) */
+AEABI_ENTRY(fcmplt)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movcs    r0, #0
+	movlt    r0, #1
+	RET
+AEABI_END(fcmplt)
+
+/* int __aeabi_fcmple(float, float) */
+AEABI_ENTRY(fcmple)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movhi    r0, #0
+	movls    r0, #1
+	RET
+AEABI_END(fcmple)
+
+/* int __aeabi_fcmpge(float, float) */
+AEABI_ENTRY(fcmpge)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movlt    r0, #0
+	movge    r0, #1
+	RET
+AEABI_END(fcmpge)
+
+/* int __aeabi_fcmpgt(float, float) */
+AEABI_ENTRY(fcmpgt)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movle    r0, #0
+	movgt    r0, #1
+	RET
+AEABI_END(fcmpgt)
+
+/* int __aeabi_fcmpun(float, float) */
+AEABI_ENTRY(fcmpun)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vcmp.f32 s0, s1
+	vmrs     APSR_nzcv, fpscr
+	movvc    r0, #0
+	movvs    r0, #1
+	RET
+AEABI_END(fcmpun)
+
+/* int __aeabi_f2iz(float) */
+AEABI_ENTRY(f2iz)
+	LOAD_SREG(s0, r0)
+#if 0
+	/*
+	 * This should be the correct instruction, but binutils incorrectly
+	 * encodes it as the version that used FPSCR to determine the rounding.
+	 * When binutils is fixed we can use this again.
+	 */
+	vcvt.s32.f32 s0, s0
+#else
+	ftosizs      s0, s0
+#endif
+	vmov         r0, s0
+	RET
+AEABI_END(f2iz)
+
+/* double __aeabi_f2d(float) */
+AEABI_ENTRY(f2d)
+	LOAD_SREG(s0, r0)
+	vcvt.f64.f32 d0, s0
+	UNLOAD_DREG(r0, r1, d0)
+	RET
+AEABI_END(f2d)
+
+/* float __aeabi_i2f(int) */
+AEABI_ENTRY(i2f)
+	vmov         s0, r0
+	vcvt.f32.s32 s0, s0
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(i2f)
+
+/* float __aeabi_fadd(float, float) */
+AEABI_ENTRY(fadd)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vadd.f32 s0, s0, s1
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(fadd)
+
+/* float __aeabi_fmul(float, float) */
+AEABI_ENTRY(fdiv)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vdiv.f32 s0, s0, s1
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(fdiv)
+
+/* float __aeabi_fmul(float, float) */
+AEABI_ENTRY(fmul)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vmul.f32 s0, s0, s1
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(fmul)
+
+/* float __aeabi_fsub(float, float) */
+AEABI_ENTRY(fsub)
+	LOAD_SREGS(s0, s1, r0, r1)
+	vsub.f32 s0, s0, s1
+	UNLOAD_SREG(r0, s0)
+	RET
+AEABI_END(fsub)
+