2004-12-22 Steven Munroe <sjmunroe@us.ibm.com>

	* math/libm-test.inc (rint_test_tonearest): New test.
	(rint_test_towardzero): New test.
	(rint_test_downward): New test.
	(rint_test_upward): New test.
	* sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Fix -0.0 case.
	Remove redundant const values.
	* sysdeps/powerpc/powerpc32/fpu/s_ceilf.S: Fix -0.0 case.
	Remove redundant const values.  Use float const.
	* sysdeps/powerpc/powerpc32/fpu/s_floor.S: Fix -0.0 case.
	* sysdeps/powerpc/powerpc32/fpu/s_floorf.S: Fix -0.0 case.
	Use float const.
	* sysdeps/powerpc/powerpc32/fpu/s_rint.S: Fix -0.0 case.
	* sysdeps/powerpc/powerpc32/fpu/s_rintf.S: Fix -0.0 case.
	Use float const.
	* sysdeps/powerpc/powerpc32/fpu/s_round.S: Fix -0.0 case.
	Remove redundant const values.
	* sysdeps/powerpc/powerpc32/fpu/s_roundf.S: Fix -0.0 case.
	Remove redundant const values.  Use float const.
	* sysdeps/powerpc/powerpc32/fpu/s_trunc.S: Fix -0.0 case.
	Remove redundant const values.
	* sysdeps/powerpc/powerpc32/fpu/s_truncf.S: Fix -0.0 case.
	Remove redundant const values.  Use float const.
	* sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Remove redundant const values.
	* sysdeps/powerpc/powerpc64/fpu/s_ceilf.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Remove redundant const values.
	Use float const.
	* sysdeps/powerpc/powerpc64/fpu/s_floor.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.
	* sysdeps/powerpc/powerpc64/fpu/s_floorf.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Use float const.
	* sysdeps/powerpc/powerpc64/fpu/s_rint.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.
	* sysdeps/powerpc/powerpc64/fpu/s_rintf.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Use float const.
	* sysdeps/powerpc/powerpc64/fpu/s_round.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Remove redundant const values.
	* sysdeps/powerpc/powerpc64/fpu/s_roundf.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Remove redundant const values.
	Use float const.
	* sysdeps/powerpc/powerpc64/fpu/s_trunc.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.
	* sysdeps/powerpc/powerpc64/fpu/s_truncf.S: Use EALIGN for Quadword
	alignment.  Fix -0.0 case.  Remove redundant const values.
	Use float const.

22 files changed, 331 insertions, 267 deletions

diff --git a/ChangeLog b/ChangeLog
index 68beedf901..bab27fa5d2 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,51 @@
+2004-12-22  Steven Munroe  <sjmunroe@us.ibm.com>
+
+	* math/libm-test.inc (rint_test_tonearest): New test.
+	(rint_test_towardzero): New test.
+	(rint_test_downward): New test.
+	(rint_test_upward): New test.
+	* sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Fix -0.0 case.
+	Remove redundant const values.
+	* sysdeps/powerpc/powerpc32/fpu/s_ceilf.S: Fix -0.0 case.
+	Remove redundant const values.  Use float const.
+	* sysdeps/powerpc/powerpc32/fpu/s_floor.S: Fix -0.0 case.
+	* sysdeps/powerpc/powerpc32/fpu/s_floorf.S: Fix -0.0 case.
+	Use float const.
+	* sysdeps/powerpc/powerpc32/fpu/s_rint.S: Fix -0.0 case.
+	* sysdeps/powerpc/powerpc32/fpu/s_rintf.S: Fix -0.0 case.
+	Use float const.
+	* sysdeps/powerpc/powerpc32/fpu/s_round.S: Fix -0.0 case.
+	Remove redundant const values.
+	* sysdeps/powerpc/powerpc32/fpu/s_roundf.S: Fix -0.0 case.
+	Remove redundant const values.  Use float const.
+	* sysdeps/powerpc/powerpc32/fpu/s_trunc.S: Fix -0.0 case.
+	Remove redundant const values.
+	* sysdeps/powerpc/powerpc32/fpu/s_truncf.S: Fix -0.0 case.
+	Remove redundant const values.  Use float const.
+	* sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Remove redundant const values.
+	* sysdeps/powerpc/powerpc64/fpu/s_ceilf.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Remove redundant const values.
+	Use float const.
+	* sysdeps/powerpc/powerpc64/fpu/s_floor.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.
+	* sysdeps/powerpc/powerpc64/fpu/s_floorf.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Use float const.
+	* sysdeps/powerpc/powerpc64/fpu/s_rint.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.
+	* sysdeps/powerpc/powerpc64/fpu/s_rintf.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Use float const.
+	* sysdeps/powerpc/powerpc64/fpu/s_round.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Remove redundant const values.
+	* sysdeps/powerpc/powerpc64/fpu/s_roundf.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Remove redundant const values.
+	Use float const.
+	* sysdeps/powerpc/powerpc64/fpu/s_trunc.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.
+	* sysdeps/powerpc/powerpc64/fpu/s_truncf.S: Use EALIGN for Quadword
+	alignment.  Fix -0.0 case.  Remove redundant const values.
+	Use float const.
+
 2004-12-29  Jakub Jelinek  <jakub@redhat.com>
 
 	* sysdeps/ia64/fpu/libm_support.h (__libm_error_support): Use
diff --git a/math/libm-test.inc b/math/libm-test.inc
index 60711fabdc..be05222629 100644
--- a/math/libm-test.inc
+++ b/math/libm-test.inc
@@ -3795,6 +3795,114 @@ rint_test (void)
 }
 
 static void
+rint_test_tonearest (void)
+{
+  int save_round_mode;
+  START (rint_tonearest);
+
+  save_round_mode = fegetround();
+
+  if (!fesetround (FE_TONEAREST))
+  {
+    TEST_f_f (rint, 2.0, 2.0);
+    TEST_f_f (rint, 1.5, 2.0);
+    TEST_f_f (rint, 1.0, 1.0);
+    TEST_f_f (rint, 0.5, 0.0);
+    TEST_f_f (rint, 0.0, 0.0);
+    TEST_f_f (rint, minus_zero, minus_zero);
+    TEST_f_f (rint, -0.5, -0.0);
+    TEST_f_f (rint, -1.0, -1.0);
+    TEST_f_f (rint, -1.5, -2.0);
+    TEST_f_f (rint, -2.0, -2.0);
+  }
+
+  fesetround(save_round_mode);
+
+  END (rint_tonearest);
+}
+
+static void
+rint_test_towardzero (void)
+{
+  int save_round_mode;
+  START (rint_towardzero);
+
+  save_round_mode = fegetround();
+
+  if (!fesetround (FE_TOWARDZERO))
+  {
+    TEST_f_f (rint, 2.0, 2.0);
+    TEST_f_f (rint, 1.5, 1.0);
+    TEST_f_f (rint, 1.0, 1.0);
+    TEST_f_f (rint, 0.5, 0.0);
+    TEST_f_f (rint, 0.0, 0.0);
+    TEST_f_f (rint, minus_zero, minus_zero);
+    TEST_f_f (rint, -0.5, -0.0);
+    TEST_f_f (rint, -1.0, -1.0);
+    TEST_f_f (rint, -1.5, -1.0);
+    TEST_f_f (rint, -2.0, -2.0);
+  }
+
+  fesetround(save_round_mode);
+
+  END (rint_towardzero);
+}
+
+static void
+rint_test_downward (void)
+{
+  int save_round_mode;
+  START (rint_downward);
+
+  save_round_mode = fegetround();
+
+  if (!fesetround (FE_DOWNWARD))
+  {
+    TEST_f_f (rint, 2.0, 2.0);
+    TEST_f_f (rint, 1.5, 1.0);
+    TEST_f_f (rint, 1.0, 1.0);
+    TEST_f_f (rint, 0.5, 0.0);
+    TEST_f_f (rint, 0.0, 0.0);
+    TEST_f_f (rint, minus_zero, minus_zero);
+    TEST_f_f (rint, -0.5, -1.0);
+    TEST_f_f (rint, -1.0, -1.0);
+    TEST_f_f (rint, -1.5, -2.0);
+    TEST_f_f (rint, -2.0, -2.0);
+  }
+
+  fesetround(save_round_mode);
+
+  END (rint_downward);
+}
+
+static void
+rint_test_upward (void)
+{
+  int save_round_mode;
+  START (rint_upward);
+
+  save_round_mode = fegetround();
+
+  if (!fesetround (FE_UPWARD))
+  {
+    TEST_f_f (rint, 2.0, 2.0);
+    TEST_f_f (rint, 1.5, 2.0);
+    TEST_f_f (rint, 1.0, 1.0);
+    TEST_f_f (rint, 0.5, 1.0);
+    TEST_f_f (rint, 0.0, 0.0);
+    TEST_f_f (rint, minus_zero, minus_zero);
+    TEST_f_f (rint, -0.5, -0.0);
+    TEST_f_f (rint, -1.0, -1.0);
+    TEST_f_f (rint, -1.5, -1.0);
+    TEST_f_f (rint, -2.0, -2.0);
+  }
+
+  fesetround(save_round_mode);
+
+  END (rint_upward);
+}
+
+static void
 round_test (void)
 {
   START (round);
@@ -4557,6 +4665,10 @@ main (int argc, char **argv)
   floor_test ();
   nearbyint_test ();
   rint_test ();
+  rint_test_tonearest ();
+  rint_test_towardzero ();
+  rint_test_downward ();
+  rint_test_upward ();
   lrint_test ();
   llrint_test ();
   round_test ();
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S b/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
index d211314bbf..22cf76e54c 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
@@ -26,20 +26,12 @@
 TWO52.0:
 	.long 0x43300000
 	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
 
 	.section	.rodata.cst8,"aM",@progbits,8
 	.align 3
 .LC0:	/* 2**52 */
 	.long 0x43300000
 	.long 0
-.LC1:	/* -0.0 */
-	.long 0x80000000
-	.long 0
 
 	.section	".text"
 ENTRY (__ceil)
@@ -64,27 +56,18 @@ ENTRY (__ceil)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	mflr	r11
-	bl	_GLOBAL_OFFSET_TABLE_@local-4
-	mflr	r10
-	lwz	r9,.LC1@got(10)
-	mtlr	r11
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC1@ha
-	lfd	fp1,.LC1@l(r9)
-#endif
 	blr
 	END (__ceil)
 
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
index 4439dc2338..e7a72186c9 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
@@ -20,26 +20,16 @@
 #include <sysdep.h>
 
 	.section	.rodata
-	.align 3
-	.type	TWO52.0,@object
-	.size	TWO52.0,8
-TWO52.0:
-	.long 0x43300000
-	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
+	.align 2
+	.type	TWO23.0,@object
+	.size	TWO23.0,4
+TWO23.0:
+	.long 0x4b000000
 
-	.section	.rodata.cst8,"aM",@progbits,8
-	.align 3
+	.section	.rodata.cst4,"aM",@progbits,4
+	.align 2
 .LC0:	/* 2**23 */
-	.long 0x41600000
-	.long 0
-.LC1:	/* -0.0 */
-	.long 0x80000000
-	.long 0
+	.long 0x4b000000
 
 	.section	".text"
 ENTRY (__ceilf)
@@ -50,10 +40,10 @@ ENTRY (__ceilf)
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
-	lfd	fp13,0(r9)
+	lfs	fp13,0(r9)
 #else
 	lis	r9,.LC0@ha
-	lfd	fp13,.LC0@l(r9)
+	lfs	fp13,.LC0@l(r9)
 #endif
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
@@ -64,27 +54,18 @@ ENTRY (__ceilf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	mflr	r11
-	bl	_GLOBAL_OFFSET_TABLE_@local-4
-	mflr	r10
-	lwz	r9,.LC1@got(10)
-	mtlr	r11
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC1@ha
-	lfd	fp1,.LC1@l(r9)
-#endif
 	blr
 	END (__ceilf)
 
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_floor.S b/sysdeps/powerpc/powerpc32/fpu/s_floor.S
index 143f907b18..812ea7ced2 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_floor.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_floor.S
@@ -56,15 +56,16 @@ ENTRY (__floor)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	fmr	fp1,fp12	/* x must be +0.0 for the 0.0 case.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
 .L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_floorf.S b/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
index 154bc30ff7..ead41d4657 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
@@ -20,18 +20,16 @@
 #include <sysdep.h>
 
 	.section	.rodata
-	.align 3
+	.align 2
 	.type	TWO23.0,@object
-	.size	TWO23.0,8
+	.size	TWO23.0,4
 TWO23.0:
-	.long 0x41600000
-	.long 0
+	.long 0x4b000000
 
-	.section	.rodata.cst8,"aM",@progbits,8
-	.align 3
+	.section	.rodata.cst4,"aM",@progbits,4
+	.align 2
 .LC0:	/* 2**23 */
-	.long 0x41600000
-	.long 0
+	.long 0x4b000000
 
 	.section	".text"
 ENTRY (__floorf)
@@ -42,10 +40,10 @@ ENTRY (__floorf)
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
-	lfd	fp13,0(r9)
+	lfs	fp13,0(r9)
 #else
 	lis	r9,.LC0@ha
-	lfd	fp13,.LC0@l(r9)
+	lfs	fp13,.LC0@l(r9)
 #endif
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
@@ -56,15 +54,16 @@ ENTRY (__floorf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	fmr	fp1,fp12	/* x must be +0.0 for the 0.0 case.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
 .L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_rint.S b/sysdeps/powerpc/powerpc32/fpu/s_rint.S
index dee25f204f..fa02dbc59c 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_rint.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_rint.S
@@ -57,13 +57,14 @@ ENTRY (__rint)
 	bng-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	blr
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = 0.0; */
 .L4:
 	bnllr-	cr6		/* if (x < 0.0)  */
-	fsub	fp1,fp13,fp1	/* x = TWO52 - x;  */
-	fsub	fp0,fp1,fp13	/* x = - (x - TWO52);  */
-	fneg	fp1,fp0
-	blr
+	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
+	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = -0.0; */
 	END (__rint)
 
 weak_alias (__rint, rint)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_rintf.S b/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
index cebf6423af..7825951268 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
@@ -21,18 +21,16 @@
 
 
 	.section	.rodata
-	.align 3
+	.align 2
 	.type	TWO23.0,@object
-	.size	TWO23.0,8
+	.size	TWO23.0,4
 TWO23.0:
-	.long 0x41600000
-	.long 0
+	.long 0x4b000000
 
-	.section	.rodata.cst8,"aM",@progbits,8
-	.align 3
+	.section	.rodata.cst4,"aM",@progbits,4
+	.align 2
 .LC0:	/* 2**23 */
-	.long 0x41600000
-	.long 0
+	.long 0x4b000000
 
 	.section	".text"
 ENTRY (__rintf)
@@ -42,10 +40,10 @@ ENTRY (__rintf)
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
-	lfd	fp13,0(r9)
+	lfs	fp13,0(r9)
 #else
 	lis	r9,.LC0@ha
-	lfd	fp13,.LC0@l(r9)
+	lfs	fp13,.LC0@l(r9)
 #endif
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
@@ -55,13 +53,14 @@ ENTRY (__rintf)
 	bng-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	blr
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = 0.0; */
 .L4:
 	bnllr-	cr6		/* if (x < 0.0)  */
-	fsubs	fp1,fp13,fp1	/* x = TWO23 - x;  */
-	fsubs	fp0,fp1,fp13	/* x = - (x - TWO23);  */
-	fneg	fp1,fp0
-	blr
+	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
+	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = -0.0; */
 	END (__rintf)
 
 weak_alias (__rintf, rintf)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_round.S b/sysdeps/powerpc/powerpc32/fpu/s_round.S
index 13fc74f001..39eab232f6 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_round.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_round.S
@@ -31,11 +31,6 @@ TWO52.0:
 POINTFIVE.0:
 	.long 0x3fe00000
 	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
 
 	.section	.rodata.cst8,"aM",@progbits,8
 	.align 3
@@ -45,9 +40,6 @@ NEGZERO.0:
 .LC1:	/* 0.5 */
 	.long 0x3fe00000
 	.long 0
-.LC2:	/* -0.0 */
-	.long 0x80000000
-	.long 0
 
 /* double [fp1] round (double x [fp1])
    IEEE 1003.1 round function.  IEEE specifies "round to the nearest 
@@ -89,7 +81,8 @@ ENTRY (__round)
 	fadd	fp1,fp1,fp10	/* x+= 0.5;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
@@ -97,16 +90,10 @@ ENTRY (__round)
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp9,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	lwz	r9,.LC2@got(10)
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC2@ha
-	lfd	fp1,.LC2@l(r9)
-#endif
 	blr
 	END (__round)
 
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_roundf.S b/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
index ea8aaf3add..a9b42f0170 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
@@ -20,34 +20,22 @@
 #include <sysdep.h>
 
 	.section	.rodata
-	.align 3
+	.align 2
 	.type	TWO23.0,@object
-	.size	TWO23.0,8
+	.size	TWO23.0,4
 TWO23.0:
-	.long 0x43300000
-	.long 0
+	.long 0x4b000000
 	.type	POINTFIVE.0,@object
-	.size	POINTFIVE.0,8
+	.size	POINTFIVE.0,4
 POINTFIVE.0:
-	.long 0x3fe00000
-	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
+	.long 0x3f000000
 
-	.section	.rodata.cst8,"aM",@progbits,8
-	.align 3
+	.section	.rodata.cst4,"aM",@progbits,4
+	.align 2
 .LC0:	/* 2**23 */
-	.long 0x41600000
-	.long 0
+	.long 0x4b000000
 .LC1:	/* 0.5 */
-	.long 0x3fe00000
-	.long 0
-.LC2:	/* -0.0 */
-	.long 0x80000000
-	.long 0
+	.long 0x3f000000
 
 /* float [fp1] roundf  (float x [fp1])
    IEEE 1003.1 round function.  IEEE specifies "round to the nearest 
@@ -67,10 +55,10 @@ ENTRY (__roundf )
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
-	lfd	fp13,0(r9)
+	lfs	fp13,0(r9)
 #else
 	lis	r9,.LC0@ha
-	lfd	fp13,.LC0@l(r9)
+	lfs	fp13,.LC0@l(r9)
 #endif
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
@@ -80,16 +68,17 @@ ENTRY (__roundf )
 	mtfsfi	7,1		/* Set rounding mode toward 0.  */
 #ifdef SHARED
 	lwz	r9,.LC1@got(10)
-	lfd	fp10,0(r9)
+	lfs	fp10,0(r9)
 #else
 	lis	r9,.LC1@ha
-	lfd	fp10,.LC1@l(r9)
+	lfs	fp10,.LC1@l(r9)
 #endif
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp10	/* x+= 0.5;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
@@ -97,16 +86,10 @@ ENTRY (__roundf )
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp9,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	lwz	r9,.LC2@got(10)
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC2@ha
-	lfd	fp1,.LC2@l(r9)
-#endif
 	blr
 	END (__roundf)
 
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_trunc.S b/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
index a4be651f8c..08acc00cb2 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
@@ -26,20 +26,12 @@
 TWO52.0:
 	.long 0x43300000
 	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
 
 	.section	.rodata.cst8,"aM",@progbits,8
 	.align 3
 .LC0:	/* 2**52 */
 	.long 0x43300000
 	.long 0
-.LC1:	/* -0.0 */
-	.long 0x80000000
-	.long 0
 
 /* double [fp1] trunc (double x [fp1])
    IEEE 1003.1 trunc function.  IEEE specifies "trunc to the integer
@@ -70,23 +62,18 @@ ENTRY (__trunc)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous truncing mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	lwz	r9,.LC1@got(10)
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC1@ha
-	lfd	fp1,.LC1@l(r9)
-#endif
 	blr
 	END (__trunc)
 
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_truncf.S b/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
index 9a8dae931b..3b6fe731b4 100644
--- a/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
+++ b/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
@@ -20,26 +20,16 @@
 #include <sysdep.h>
 
 	.section	.rodata
-	.align 3
+	.align 2
 	.type	TWO23.0,@object
-	.size	TWO23.0,8
+	.size	TWO23.0,2
 TWO23.0:
-	.long 0x41600000
-	.long 0
-	.type	NEGZERO.0,@object
-	.size	NEGZERO.0,8
-NEGZERO.0:
-	.long 0x80000000
-	.long 0
+	.long 0x4b000000
 
-	.section	.rodata.cst8,"aM",@progbits,8
-	.align 3
+	.section	.rodata.cst4,"aM",@progbits,4
+	.align 2
 .LC0:	/* 2**23 */
-	.long 0x41600000
-	.long 0
-.LC1:	/* -0.0 */
-	.long 0x80000000
-	.long 0
+	.long 0x4b000000
 
 /* float [fp1] truncf (float x [fp1])
    IEEE 1003.1 trunc function.  IEEE specifies "trunc to the integer
@@ -56,10 +46,10 @@ ENTRY (__truncf)
 	mflr	r10
 	lwz	r9,.LC0@got(10)
 	mtlr	r11
-	lfd	fp13,0(r9)
+	lfs	fp13,0(r9)
 #else
 	lis	r9,.LC0@ha
-	lfd	fp13,.LC0@l(r9)
+	lfs	fp13,.LC0@l(r9)
 #endif
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
@@ -70,23 +60,18 @@ ENTRY (__truncf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous truncing mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-#ifdef SHARED
-	lwz	r9,.LC1@got(10)
-	lfd	fp1,0(r9)
-#else
-	lis	r9,.LC1@ha
-	lfd	fp1,.LC1@l(r9)
-#endif
 	blr
 	END (__truncf)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_ceil.S b/sysdeps/powerpc/powerpc64/fpu/s_ceil.S
index a1bfaa70c2..9809e24d26 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_ceil.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_ceil.S
@@ -22,11 +22,9 @@
 	.section	".toc","aw"
 .LC0:	/* 2**52 */
 	.tc FD_43300000_0[TC],0x4330000000000000
-.LC1:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
 	.section	".text"
 
-ENTRY (__ceil)
+EALIGN (__ceil, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
 	lfd	fp13,.LC0@toc(2)
@@ -39,17 +37,18 @@ ENTRY (__ceil)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC1@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
 	blr
 	END (__ceil)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
index 42eb274389..1ccd133b66 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
@@ -21,15 +21,13 @@
 
 	.section	".toc","aw"
 .LC0:	/* 2**23 */
-	.tc FD_41600000_0[TC],0x4160000000000000
-.LC1:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
+	.tc FD_4b000000_0[TC],0x4b00000000000000
 	.section	".text"
 
-ENTRY (__ceilf)
+EALIGN (__ceilf, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
-	lfd	fp13,.LC0@toc(2)
+	lfs	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
 	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
@@ -39,17 +37,18 @@ ENTRY (__ceilf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC1@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
 	blr
 	END (__ceilf)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_floor.S b/sysdeps/powerpc/powerpc64/fpu/s_floor.S
index 80cbdc5709..183423c2b3 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_floor.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_floor.S
@@ -24,7 +24,7 @@
 	.tc FD_43300000_0[TC],0x4330000000000000
 	.section	".text"
 
-ENTRY (__floor)
+EALIGN (__floor, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
 	lfd	fp13,.LC0@toc(2)
@@ -37,15 +37,16 @@ ENTRY (__floor)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	fmr	fp1,fp12	/* x must be +0.0 for the 0.0 case.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
 .L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_floorf.S b/sysdeps/powerpc/powerpc64/fpu/s_floorf.S
index 20cbb15ebd..bcdbf7823d 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_floorf.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_floorf.S
@@ -21,13 +21,13 @@
 
 	.section	".toc","aw"
 .LC0:	/* 2**23 */
-	.tc FD_41600000_0[TC],0x4160000000000000
+	.tc FD_4b000000_0[TC],0x4b00000000000000
 	.section	".text"
 
-ENTRY (__floorf)
+EALIGN (__floorf, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
-	lfd	fp13,.LC0@toc(2)
+	lfs	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
 	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
@@ -37,15 +37,16 @@ ENTRY (__floorf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	fmr	fp1,fp12	/* x must be +0.0 for the 0.0 case.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
 .L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_rint.S b/sysdeps/powerpc/powerpc64/fpu/s_rint.S
index 79e807269d..0c0e0ba67b 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_rint.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_rint.S
@@ -27,7 +27,7 @@
 	.tc FD_43300000_0[TC],0x4330000000000000
 	.section	".text"
 
-ENTRY (__rint)
+EALIGN (__rint, 4, 0)
 	CALL_MCOUNT 0
 	lfd	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
@@ -38,13 +38,14 @@ ENTRY (__rint)
 	bng-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	blr
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = 0.0; */
 .L4:
 	bnllr-	cr6		/* if (x < 0.0)  */
-	fsub	fp1,fp13,fp1	/* x = TWO52 - x;  */
-	fsub	fp0,fp1,fp13	/* x = - (x - TWO52);  */
-	fneg	fp1,fp0
-	blr
+	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
+	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = -0.0; */
 	END (__rint)
 
 weak_alias (__rint, rint)
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_rintf.S b/sysdeps/powerpc/powerpc64/fpu/s_rintf.S
index eb34dd5e77..e4fa9ba2e6 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_rintf.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_rintf.S
@@ -21,12 +21,12 @@
 
 	.section	".toc","aw"
 .LC0:	/* 2**23 */
-	.tc FD_41600000_0[TC],0x4160000000000000
+	.tc FD_4b000000_0[TC],0x4b00000000000000
 	.section	".text"
 
-ENTRY (__rintf)
+EALIGN (__rintf, 4, 0)
 	CALL_MCOUNT 0
-	lfd	fp13,.LC0@toc(2)
+	lfs	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
 	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
@@ -35,13 +35,14 @@ ENTRY (__rintf)
 	bng-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	blr
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = 0.0; */
 .L4:
 	bnllr-	cr6		/* if (x < 0.0)  */
-	fsubs	fp1,fp13,fp1	/* x = TWO23 - x;  */
-	fsubs	fp0,fp1,fp13	/* x = - (x - TWO23);  */
-	fneg	fp1,fp0
-	blr
+	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
+	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+	blr			/* x = -0.0; */
 	END (__rintf)
 
 weak_alias (__rintf, rintf)
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_round.S b/sysdeps/powerpc/powerpc64/fpu/s_round.S
index c0b6d46fea..b07a7ea15a 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_round.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_round.S
@@ -24,8 +24,6 @@
 	.tc FD_43300000_0[TC],0x4330000000000000
 .LC1:	/* 0.5 */
 	.tc FD_3fe00000_0[TC],0x3fe0000000000000
-.LC2:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
 	.section	".text"
 	
 /* double [fp1] round (double x [fp1])
@@ -38,7 +36,7 @@
    "Round toward Zero" mode and round by adding +-0.5 before rounding
    to the integer value.  */
 
-ENTRY (__round)
+EALIGN (__round, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
 	lfd	fp13,.LC0@toc(2)
@@ -53,7 +51,8 @@ ENTRY (__round)
 	fadd	fp1,fp1,fp10	/* x+= 0.5;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
@@ -61,10 +60,10 @@ ENTRY (__round)
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp9,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
-	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC2@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
+	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
 	END (__round)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_roundf.S b/sysdeps/powerpc/powerpc64/fpu/s_roundf.S
index 23ee4c052b..d2e29fdb8f 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_roundf.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_roundf.S
@@ -21,11 +21,9 @@
 
 	.section	".toc","aw"
 .LC0:	/* 2**23 */
-	.tc FD_41600000_0[TC],0x4160000000000000
+	.tc FD_4b000000_0[TC],0x4b00000000000000
 .LC1:	/* 0.5 */
-	.tc FD_3fe00000_0[TC],0x3fe0000000000000
-.LC2:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
+	.tc FD_3f000000_0[TC],0x3f00000000000000
 	.section	".text"
 	
 /* float [fp1] roundf  (float x [fp1])
@@ -38,22 +36,23 @@
    "Round toward Zero" mode and round by adding +-0.5 before rounding
    to the integer value.  */
 
-ENTRY (__roundf )
+EALIGN (__roundf, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
-	lfd	fp13,.LC0@toc(2)
+	lfs	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
 	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
 	fcmpu	cr6,fp1,fp12	/* if (x > 0.0)  */
 	bnllr-	cr7
 	mtfsfi	7,1		/* Set rounding mode toward 0.  */
-	lfd	fp10,.LC1@toc(2)
+	lfs	fp10,.LC1@toc(2)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp10	/* x+= 0.5;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
 	blr
 .L4:
@@ -61,10 +60,10 @@ ENTRY (__roundf )
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp9,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
-	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC2@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
+	mtfsf	0x01,fp11	/* restore previous rounding mode.  */	
 	blr
 	END (__roundf)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_trunc.S b/sysdeps/powerpc/powerpc64/fpu/s_trunc.S
index 3ddd298525..d69e371b61 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_trunc.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_trunc.S
@@ -22,8 +22,6 @@
 	.section	".toc","aw"
 .LC0:	/* 2**52 */
 	.tc FD_43300000_0[TC],0x4330000000000000
-.LC2:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
 	.section	".text"
 	
 /* double [fp1] trunc (double x [fp1])
@@ -33,7 +31,7 @@
    We set "round toward Zero" mode and trunc by adding +-2**52 then
    subtracting +-2**52.  */
 
-ENTRY (__trunc)
+EALIGN (__trunc, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
 	lfd	fp13,.LC0@toc(2)
@@ -46,17 +44,18 @@ ENTRY (__trunc)
 	ble-	cr6,.L4
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous truncing mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
 	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC2@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
 	blr
 	END (__trunc)
 
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_truncf.S b/sysdeps/powerpc/powerpc64/fpu/s_truncf.S
index b38b722a6f..15f53da8ca 100644
--- a/sysdeps/powerpc/powerpc64/fpu/s_truncf.S
+++ b/sysdeps/powerpc/powerpc64/fpu/s_truncf.S
@@ -21,9 +21,7 @@
 
 	.section	".toc","aw"
 .LC0:	/* 2**23 */
-	.tc FD_41600000_0[TC],0x4160000000000000
-.LC2:	/* -0.0 */
-	.tc FD_80000000_0[TC],0x8000000000000000
+	.tc FD_4b000000_0[TC],0x4b00000000000000
 	.section	".text"
 	
 /* float [fp1] truncf (float x [fp1])
@@ -33,10 +31,10 @@
    We set "round toward Zero" mode and trunc by adding +-2**23 then
    subtracting +-2**23.  */
 
-ENTRY (__truncf)
+EALIGN (__truncf, 4, 0)
 	CALL_MCOUNT 0
 	mffs	fp11		/* Save current FPU rounding mode.  */
-	lfd	fp13,.LC0@toc(2)
+	lfs	fp13,.LC0@toc(2)
 	fabs	fp0,fp1
 	fsubs	fp12,fp13,fp13	/* generate 0.0  */
 	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
@@ -46,17 +44,18 @@ ENTRY (__truncf)
 	ble-	cr6,.L4
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-.L9:	
+	fabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = 0.0; */
 	mtfsf	0x01,fp11	/* restore previous truncing mode.  */
 	blr
 .L4:
 	bge-	cr6,.L9		/* if (x < 0.0)  */
 	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
 	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fcmpu	cr5,fp1,fp12	/* if (x > 0.0)  */
+	fnabs	fp1,fp1		/* if (x == 0.0)  */
+				/* x = -0.0; */
+.L9:
 	mtfsf	0x01,fp11	/* restore previous rounding mode.  */
-	bnelr+	cr5
-	lfd	fp1,.LC2@toc(2)	/* x must be -0.0 for the 0.0 case.  */	
 	blr
 	END (__truncf)