about summary refs log tree commit diff
path: root/sysdeps/sparc/sparc64
diff options
context:
space:
mode:
authorDavid S. Miller <davem@davemloft.net>2013-01-14 21:47:29 -0800
committerDavid S. Miller <davem@davemloft.net>2013-01-14 21:47:29 -0800
commit8b954ab9b808a51d8cf6a90c7d1d46a3366e3274 (patch)
treecbf6dba7b97106b9ce476078cda8e8f0a1e7bf7c /sysdeps/sparc/sparc64
parent65a82e3dd538aabcdbeacf87fcee32f4e355e5b6 (diff)
downloadglibc-8b954ab9b808a51d8cf6a90c7d1d46a3366e3274.tar.gz
glibc-8b954ab9b808a51d8cf6a90c7d1d46a3366e3274.tar.xz
glibc-8b954ab9b808a51d8cf6a90c7d1d46a3366e3274.zip
Optimize sparc {ceil,floor}{,f} using vis2 'siam' instruction.
	* sysdeps/sparc/sparc-ifunc.h (SPARC_ASM_IFUNC2): New macro.
	(SPARC_ASM_VIS2_IFUNC): Likewise.
	(SPARC_ASM_VIS3_VIS2_IFUNC): Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis3.S: Make
	use of 'siam' instruction.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis3.S:
	Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis3.S:
	Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis3.S:
	Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S: Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil-vis2.S: New
	file.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf-vis2.S: New
	file.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor-vis2.S: New
	file.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf-vis2.S: New
	file.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S: New file.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S: New file.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S: New file.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S: New file.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceil.S: Hook in
	new VIS2 routines.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_ceilf.S: Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floor.S: Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_floorf.S:
	Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S: Likewise.
	* sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S: Likewise.
	* sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile: Add new VIS2
	routines to libm-sysdep_routines.
	* sysdeps/sparc/sparc64/fpu/multiarch/Makefile: Likewise.
Diffstat (limited to 'sysdeps/sparc/sparc64')
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/Makefile2
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S57
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S38
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S2
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S56
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S37
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S2
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S57
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S38
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S2
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S56
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S39
-rw-r--r--sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S2
13 files changed, 277 insertions, 111 deletions
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/Makefile b/sysdeps/sparc/sparc64/fpu/multiarch/Makefile
index 83458484a7..7a5a9dd9cc 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/Makefile
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/Makefile
@@ -1,4 +1,6 @@
 ifeq ($(subdir),math)
+libm-sysdep_routines += s_ceil-vis2 s_ceilf-vis2 \
+			s_floor-vis2 s_floorf-vis2
 ifeq ($(have-as-vis3),yes)
 libm-sysdep_routines += m_signbitf-vis3 m_signbit-vis3 s_ceilf-vis3 \
 			s_ceil-vis3 m_finitef-vis3 m_finite-vis3 \
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S
new file mode 100644
index 0000000000..50d96a6efb
--- /dev/null
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis2.S
@@ -0,0 +1,57 @@
+/* ceil function, sparc64 vis2 version.
+   Copyright (C) 2012-2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by David S. Miller <davem@davemloft.net>, 2012.
+
+   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>
+
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
+
+	   We add then subtract (or subtract than add if the initial
+	   value was negative) 2**23 to the value, then subtract it
+	   back out.
+
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.  */
+
+#define TWO_FIFTYTWO	0x43300000		/* 2**52 */
+
+#define ZERO		%f10			/* 0.0 */
+#define SIGN_BIT	%f12			/* -0.0 */
+
+ENTRY (__ceil_vis2)
+	sethi	%hi(TWO_FIFTYTWO), %o2
+	fzero	ZERO
+	sllx	%o2, 32, %o2
+	fnegd	ZERO, SIGN_BIT
+	stx	%o2, [%sp + STACK_BIAS + 128]
+	fabsd	%f0, %f14
+	ldd	[%sp + STACK_BIAS + 128], %f16
+	fcmpd	%fcc3, %f14, %f16
+	fmovduge %fcc3, ZERO, %f16
+	fand	%f0, SIGN_BIT, SIGN_BIT
+	for	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 2
+	faddd	%f0, %f16, %f18
+	siam	(1 << 2) | 0
+	fsubd	%f18, %f16, %f18
+	siam	(0 << 2)
+	retl
+	 for	%f18, SIGN_BIT, %f0
+END (__ceil_vis2)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S
index 9a598ea6fb..6acff09998 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil-vis3.S
@@ -19,57 +19,41 @@
 
 #include <sysdep.h>
 
-	/* Since changing the rounding mode is extremely expensive, we
-	   try to round up using a method that is rounding mode
-	   agnostic.
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
 
 	   We add then subtract (or subtract than add if the initial
 	   value was negative) 2**23 to the value, then subtract it
 	   back out.
 
-	   This will clear out the fractional portion of the value.
-	   One of two things will happen for non-whole initial values.
-	   Either the rounding mode will round it up, or it will be
-	   rounded down.  If the value started out whole, it will be
-	   equal after the addition and subtraction.  This means we
-	   can accurately detect with one test whether we need to add
-	   another 1.0 to round it up properly.
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.
 
-	   VIS instructions are used to facilitate the formation of
-	   easier constants, and the propagation of the sign bit.  */
+	   We also use VIS3 moves to avoid using the stack to transfer
+	   values between float and integer registers.  */
 
 #define TWO_FIFTYTWO	0x43300000		/* 2**52 */
-#define ONE_DOT_ZERO	0x3ff00000		/* 1.0 */
 
 #define ZERO		%f10			/* 0.0 */
 #define SIGN_BIT	%f12			/* -0.0 */
 
 ENTRY (__ceil_vis3)
 	sethi	%hi(TWO_FIFTYTWO), %o2
-	sethi	%hi(ONE_DOT_ZERO), %o3
 	fzero	ZERO
-
 	sllx	%o2, 32, %o2
 	fnegd	ZERO, SIGN_BIT
-
-	sllx	%o3, 32, %o3
 	movxtod	%o2, %f16
 	fabsd	%f0, %f14
-
 	fcmpd	%fcc3, %f14, %f16
-
 	fmovduge %fcc3, ZERO, %f16
 	fand	%f0, SIGN_BIT, SIGN_BIT
-
 	for	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 2
 	faddd	%f0, %f16, %f18
+	siam	(1 << 2) | 0
 	fsubd	%f18, %f16, %f18
-	fcmpd	%fcc2, %f18, %f0
-	movxtod	%o3, %f20
-
-	fmovduge %fcc2, ZERO, %f20
-	faddd	%f18, %f20, %f0
-	fabsd	%f0, %f0
+	siam	(0 << 2)
 	retl
-	 for	%f0, SIGN_BIT, %f0
+	 for	%f18, SIGN_BIT, %f0
 END (__ceil_vis3)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S
index fa4c7c353f..e7822bc307 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceil.S
@@ -1,6 +1,6 @@
 #include <sparc-ifunc.h>
 
-SPARC_ASM_VIS3_IFUNC(ceil)
+SPARC_ASM_VIS3_VIS2_IFUNC(ceil)
 
 weak_alias (__ceil, ceil)
 
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S
new file mode 100644
index 0000000000..cd5937b38d
--- /dev/null
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis2.S
@@ -0,0 +1,56 @@
+/* Float ceil function, sparc64 vis2 version.
+   Copyright (C) 2012-2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by David S. Miller <davem@davemloft.net>, 2012.
+
+   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>
+
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
+
+	   We add then subtract (or subtract than add if the initial
+	   value was negative) 2**23 to the value, then subtract it
+	   back out.
+
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.  */
+
+#define TWO_TWENTYTHREE	0x4b000000		/* 2**23 */
+
+#define ZERO		%f10			/* 0.0 */
+#define SIGN_BIT	%f12			/* -0.0 */
+
+ENTRY (__ceilf_vis2)
+	sethi	%hi(TWO_TWENTYTHREE), %o2
+	fzeros	ZERO
+	fnegs	ZERO, SIGN_BIT
+	st	%o2, [%sp + STACK_BIAS + 128]
+	fabss	%f1, %f14
+	ld	[%sp + STACK_BIAS + 128], %f16
+	fcmps	%fcc3, %f14, %f16
+	fmovsuge %fcc3, ZERO, %f16
+	fands	%f1, SIGN_BIT, SIGN_BIT
+	fors	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 2
+	fadds	%f1, %f16, %f5
+	siam	(1 << 2) | 0
+	fsubs	%f5, %f16, %f5
+	siam	(0 << 2)
+	retl
+	 fors	%f5, SIGN_BIT, %f0
+END (__ceilf_vis2)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S
index 8c635907d3..b3ec3484db 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf-vis3.S
@@ -19,55 +19,40 @@
 
 #include <sysdep.h>
 
-	/* Since changing the rounding mode is extremely expensive, we
-	   try to round up using a method that is rounding mode
-	   agnostic.
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
 
 	   We add then subtract (or subtract than add if the initial
 	   value was negative) 2**23 to the value, then subtract it
 	   back out.
 
-	   This will clear out the fractional portion of the value.
-	   One of two things will happen for non-whole initial values.
-	   Either the rounding mode will round it up, or it will be
-	   rounded down.  If the value started out whole, it will be
-	   equal after the addition and subtraction.  This means we
-	   can accurately detect with one test whether we need to add
-	   another 1.0 to round it up properly.
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.
 
-	   VIS instructions are used to facilitate the formation of
-	   easier constants, and the propagation of the sign bit.  */
+	   We also use VIS3 moves to avoid using the stack to transfer
+	   values between float and integer registers.  */
 
 #define TWO_TWENTYTHREE	0x4b000000		/* 2**23 */
-#define ONE_DOT_ZERO	0x3f800000		/* 1.0 */
 
 #define ZERO		%f10			/* 0.0 */
 #define SIGN_BIT	%f12			/* -0.0 */
 
 ENTRY (__ceilf_vis3)
 	sethi	%hi(TWO_TWENTYTHREE), %o2
-	sethi	%hi(ONE_DOT_ZERO), %o3
 	fzeros	ZERO
-
 	fnegs	ZERO, SIGN_BIT
-
 	movwtos	%o2, %f16
 	fabss	%f1, %f14
-
 	fcmps	%fcc3, %f14, %f16
-
 	fmovsuge %fcc3, ZERO, %f16
 	fands	%f1, SIGN_BIT, SIGN_BIT
-
 	fors	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 2
 	fadds	%f1, %f16, %f5
+	siam	(1 << 2) | 0
 	fsubs	%f5, %f16, %f5
-	fcmps	%fcc2, %f5, %f1
-	movwtos	%o3, %f9
-
-	fmovsuge %fcc2, ZERO, %f9
-	fadds	%f5, %f9, %f0
-	fabss	%f0, %f0
+	siam	(0 << 2)
 	retl
-	 fors	%f0, SIGN_BIT, %f0
+	 fors	%f5, SIGN_BIT, %f0
 END (__ceilf_vis3)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S
index 3047dd8faa..1c72a5728e 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_ceilf.S
@@ -1,6 +1,6 @@
 #include <sparc-ifunc.h>
 
-SPARC_ASM_VIS3_IFUNC(ceilf)
+SPARC_ASM_VIS3_VIS2_IFUNC(ceilf)
 
 weak_alias (__ceilf, ceilf)
 
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S
new file mode 100644
index 0000000000..5479ceddfe
--- /dev/null
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis2.S
@@ -0,0 +1,57 @@
+/* floor function, sparc64 vis2 version.
+   Copyright (C) 2012-2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by David S. Miller <davem@davemloft.net>, 2012.
+
+   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>
+
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
+
+	   We add then subtract (or subtract than add if the initial
+	   value was negative) 2**23 to the value, then subtract it
+	   back out.
+
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.  */
+
+#define TWO_FIFTYTWO	0x43300000		/* 2**52 */
+
+#define ZERO		%f10			/* 0.0 */
+#define SIGN_BIT	%f12			/* -0.0 */
+
+ENTRY (__floor_vis2)
+	sethi	%hi(TWO_FIFTYTWO), %o2
+	fzero	ZERO
+	sllx	%o2, 32, %o2
+	fnegd	ZERO, SIGN_BIT
+	stx	%o2, [%sp + STACK_BIAS + 128]
+	fabsd	%f0, %f14
+	ldd	[%sp + STACK_BIAS + 128], %f16
+	fcmpd	%fcc3, %f14, %f16
+	fmovduge %fcc3, ZERO, %f16
+	fand	%f0, SIGN_BIT, SIGN_BIT
+	for	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 3
+	faddd	%f0, %f16, %f18
+	siam	(1 << 2) | 0
+	fsubd	%f18, %f16, %f18
+	siam	(0 << 2)
+	retl
+	 for	%f18, SIGN_BIT, %f0
+END (__floor_vis2)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S
index 169d206549..c2ffe9f411 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor-vis3.S
@@ -19,57 +19,41 @@
 
 #include <sysdep.h>
 
-	/* Since changing the rounding mode is extremely expensive, we
-	   try to round up using a method that is rounding mode
-	   agnostic.
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
 
 	   We add then subtract (or subtract than add if the initial
 	   value was negative) 2**23 to the value, then subtract it
 	   back out.
 
-	   This will clear out the fractional portion of the value.
-	   One of two things will happen for non-whole initial values.
-	   Either the rounding mode will round it up, or it will be
-	   rounded down.  If the value started out whole, it will be
-	   equal after the addition and subtraction.  This means we
-	   can accurately detect with one test whether we need to add
-	   another 1.0 to round it up properly.
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.
 
-	   VIS instructions are used to facilitate the formation of
-	   easier constants, and the propagation of the sign bit.  */
+	   We also use VIS3 moves to avoid using the stack to transfer
+	   values between float and integer registers.  */
 
 #define TWO_FIFTYTWO	0x43300000		/* 2**52 */
-#define ONE_DOT_ZERO	0x3ff00000		/* 1.0 */
 
 #define ZERO		%f10			/* 0.0 */
 #define SIGN_BIT	%f12			/* -0.0 */
 
 ENTRY (__floor_vis3)
 	sethi	%hi(TWO_FIFTYTWO), %o2
-	sethi	%hi(ONE_DOT_ZERO), %o3
 	fzero	ZERO
-
 	sllx	%o2, 32, %o2
 	fnegd	ZERO, SIGN_BIT
-
-	sllx	%o3, 32, %o3
 	movxtod	%o2, %f16
 	fabsd	%f0, %f14
-
 	fcmpd	%fcc3, %f14, %f16
-
 	fmovduge %fcc3, ZERO, %f16
 	fand	%f0, SIGN_BIT, SIGN_BIT
-
 	for	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 3
 	faddd	%f0, %f16, %f18
+	siam	(1 << 2) | 0
 	fsubd	%f18, %f16, %f18
-	fcmpd	%fcc2, %f18, %f0
-	movxtod	%o3, %f20
-
-	fmovdule %fcc2, ZERO, %f20
-	fsubd	%f18, %f20, %f0
-	fabsd	%f0, %f0
+	siam	(0 << 2)
 	retl
-	 for	%f0, SIGN_BIT, %f0
+	 for	%f18, SIGN_BIT, %f0
 END (__floor_vis3)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S
index 6ae9947a7a..989ccab56d 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floor.S
@@ -1,6 +1,6 @@
 #include <sparc-ifunc.h>
 
-SPARC_ASM_VIS3_IFUNC(floor)
+SPARC_ASM_VIS3_VIS2_IFUNC(floor)
 
 weak_alias (__floor, floor)
 
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S
new file mode 100644
index 0000000000..935fa853d8
--- /dev/null
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis2.S
@@ -0,0 +1,56 @@
+/* Float floor function, sparc64 vis2 version.
+   Copyright (C) 2012-2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by David S. Miller <davem@davemloft.net>, 2012.
+
+   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>
+
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
+
+	   We add then subtract (or subtract than add if the initial
+	   value was negative) 2**23 to the value, then subtract it
+	   back out.
+
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.  */
+
+#define TWO_TWENTYTHREE	0x4b000000		/* 2**23 */
+
+#define ZERO		%f10			/* 0.0 */
+#define SIGN_BIT	%f12			/* -0.0 */
+
+ENTRY (__floorf_vis2)
+	sethi	%hi(TWO_TWENTYTHREE), %o2
+	fzeros	ZERO
+	fnegs	ZERO, SIGN_BIT
+	st	%o2, [%sp + STACK_BIAS + 128]
+	fabss	%f1, %f14
+	ld	[%sp + STACK_BIAS + 128], %f16
+	fcmps	%fcc3, %f14, %f16
+	fmovsuge %fcc3, ZERO, %f16
+	fands	%f1, SIGN_BIT, SIGN_BIT
+	fors	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 3
+	fadds	%f1, %f16, %f5
+	siam	(1 << 2) | 0
+	fsubs	%f5, %f16, %f5
+	siam	(0 << 2)
+	retl
+	 fors	%f5, SIGN_BIT, %f0
+END (__floorf_vis2)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S
index 65be2977a3..225e17ec05 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf-vis3.S
@@ -19,55 +19,40 @@
 
 #include <sysdep.h>
 
-	/* Since changing the rounding mode is extremely expensive, we
-	   try to round up using a method that is rounding mode
-	   agnostic.
+	/* 'siam' (Set Interval Arithmetic Mode) is used to quickly override
+	   the rounding mode during this routine.
 
 	   We add then subtract (or subtract than add if the initial
 	   value was negative) 2**23 to the value, then subtract it
 	   back out.
 
-	   This will clear out the fractional portion of the value.
-	   One of two things will happen for non-whole initial values.
-	   Either the rounding mode will round it up, or it will be
-	   rounded down.  If the value started out whole, it will be
-	   equal after the addition and subtraction.  This means we
-	   can accurately detect with one test whether we need to add
-	   another 1.0 to round it up properly.
+	   This will clear out the fractional portion of the value and,
+	   with suitable 'siam' initiated rouding mode settings, round
+	   the final result in the proper direction.
 
-	   VIS instructions are used to facilitate the formation of
-	   easier constants, and the propagation of the sign bit.  */
+	   We also use VIS3 moves to avoid using the stack to transfer
+	   values between float and integer registers.  */
 
 #define TWO_TWENTYTHREE	0x4b000000		/* 2**23 */
-#define ONE_DOT_ZERO	0x3f800000		/* 1.0 */
 
 #define ZERO		%f10			/* 0.0 */
 #define SIGN_BIT	%f12			/* -0.0 */
 
 ENTRY (__floorf_vis3)
 	sethi	%hi(TWO_TWENTYTHREE), %o2
-	sethi	%hi(ONE_DOT_ZERO), %o3
 	fzeros	ZERO
-
 	fnegs	ZERO, SIGN_BIT
-
-	movwtos %o2, %f16
+	movwtos	%o2, %f16
 	fabss	%f1, %f14
-
 	fcmps	%fcc3, %f14, %f16
-
 	fmovsuge %fcc3, ZERO, %f16
 	fands	%f1, SIGN_BIT, SIGN_BIT
-
 	fors	%f16, SIGN_BIT, %f16
+	siam	(1 << 2) | 3
 	fadds	%f1, %f16, %f5
+	siam	(1 << 2) | 0
 	fsubs	%f5, %f16, %f5
-	fcmps	%fcc2, %f5, %f1
-	movwtos	%o3, %f9
-
-	fmovsule %fcc2, ZERO, %f9
-	fsubs	%f5, %f9, %f0
-	fabss	%f0, %f0
+	siam	(0 << 2)
 	retl
-	 fors	%f0, SIGN_BIT, %f0
+	 fors	%f5, SIGN_BIT, %f0
 END (__floorf_vis3)
diff --git a/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S
index 31cda385b7..d2a83cb9b8 100644
--- a/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S
+++ b/sysdeps/sparc/sparc64/fpu/multiarch/s_floorf.S
@@ -1,6 +1,6 @@
 #include <sparc-ifunc.h>
 
-SPARC_ASM_VIS3_IFUNC(floorf)
+SPARC_ASM_VIS3_VIS2_IFUNC(floorf)
 
 weak_alias (__floorf, floorf)