about summary refs log tree commit diff
path: root/sysdeps/ieee754/ldbl-128ibm
diff options
context:
space:
mode:
authorJakub Jelinek <jakub@redhat.com>2007-07-31 13:33:18 +0000
committerJakub Jelinek <jakub@redhat.com>2007-07-31 13:33:18 +0000
commit32c075e1f01849e161724bbd400ba77244e482cc (patch)
tree5f083a3f352104f32bb6c902d57fa3f294bd8d4d /sysdeps/ieee754/ldbl-128ibm
parentd6220e9ee38c1c9285221b023346201ec5f511b3 (diff)
downloadglibc-32c075e1f01849e161724bbd400ba77244e482cc.tar.gz
glibc-32c075e1f01849e161724bbd400ba77244e482cc.tar.xz
glibc-32c075e1f01849e161724bbd400ba77244e482cc.zip
.
Diffstat (limited to 'sysdeps/ieee754/ldbl-128ibm')
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c75
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/printf_fphex.c35
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c37
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c107
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c23
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c26
6 files changed, 192 insertions, 111 deletions
diff --git a/sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c b/sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c
index 8a2d45e2d8..21d1e62dab 100644
--- a/sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 1995, 1996, 1997, 1998, 1999, 2002, 2003, 2004, 2006
+/* Copyright (C) 1995, 1996, 1997, 1998, 1999, 2002, 2003, 2004, 2006, 2007
 	Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
@@ -31,19 +31,20 @@ long double
 __mpn_construct_long_double (mp_srcptr frac_ptr, int expt, int sign)
 {
   union ibm_extended_long_double u;
-  unsigned long hidden2, lzcount;
+  unsigned long lzcount;
   unsigned long long hi, lo;
+  int exponent2;
 
   u.ieee.negative = sign;
   u.ieee.negative2 = sign;
   u.ieee.exponent = expt + IBM_EXTENDED_LONG_DOUBLE_BIAS;
-  u.ieee.exponent2 = expt - 53 + IBM_EXTENDED_LONG_DOUBLE_BIAS;
+  u.ieee.exponent2 = 0;
+  exponent2 = expt - 53 + IBM_EXTENDED_LONG_DOUBLE_BIAS;
 
 #if BITS_PER_MP_LIMB == 32
   /* The low order 53 bits (52 + hidden) go into the lower double */
   lo = frac_ptr[0];
   lo |= (frac_ptr[1] & ((1LL << (53 - 32)) - 1)) << 32;
-  hidden2 = (frac_ptr[1] >> (52 - 32)) & ((mp_limb_t) 1);
   /* The high order 53 bits (52 + hidden) go into the upper double */
   hi = (frac_ptr[1] >> (53 - 32)) & ((1 << 11) - 1);
   hi |= ((unsigned long long) frac_ptr[2]) << 11;
@@ -51,7 +52,6 @@ __mpn_construct_long_double (mp_srcptr frac_ptr, int expt, int sign)
 #elif BITS_PER_MP_LIMB == 64
   /* The low order 53 bits (52 + hidden) go into the lower double */
   lo = frac_ptr[0] & (((mp_limb_t) 1 << 53) - 1);
-  hidden2 = (frac_ptr[0] >> 52) & ((mp_limb_t) 1);
   /* The high order 53 bits (52 + hidden) go into the upper double */
   hi = (frac_ptr[0] >> 53) & (((mp_limb_t) 1 << 11) - 1);
   hi |= (frac_ptr[1] << 11);
@@ -59,14 +59,62 @@ __mpn_construct_long_double (mp_srcptr frac_ptr, int expt, int sign)
   #error "mp_limb size " BITS_PER_MP_LIMB "not accounted for"
 #endif
 
+  if ((hi & (1LL << 52)) == 0 && (hi | lo) != 0)
+    {
+      /* denormal number  */
+      unsigned long long val = hi ? hi : lo;
+
+      if (sizeof (val) == sizeof (long))
+	lzcount = __builtin_clzl (val);
+      else if ((val >> 32) != 0)
+	lzcount = __builtin_clzl ((long) (val >> 32));
+      else
+	lzcount = __builtin_clzl ((long) val) + 32;
+      if (hi)
+	lzcount = lzcount - 11;
+      else
+	lzcount = lzcount + 42;
+
+      if (lzcount > u.ieee.exponent)
+	{
+	  lzcount = u.ieee.exponent;
+	  u.ieee.exponent = 0;
+	  exponent2 -= lzcount;
+	}
+      else
+	{
+	  u.ieee.exponent -= (lzcount - 1);
+	  exponent2 -= (lzcount - 1);
+	}
+
+      if (lzcount <= 53)
+	{
+	  hi = (hi << lzcount) | (lo >> (53 - lzcount));
+	  lo = (lo << lzcount) & ((1LL << 53) - 1);
+	}
+      else
+	{
+	  hi = lo << (lzcount - 53);
+	  lo = 0;
+	}
+    }
+
   if (lo != 0L)
     {
       /* hidden2 bit of low double controls rounding of the high double.
-	 If hidden2 is '1' then round up hi and adjust lo (2nd mantissa)
+	 If hidden2 is '1' and either the explicit mantissa is non-zero
+	 or hi is odd, then round up hi and adjust lo (2nd mantissa)
 	 plus change the sign of the low double to compensate.  */
-      if (hidden2)
+      if ((lo & (1LL << 52)) != 0
+	  && ((hi & 1) != 0 || (lo & ((1LL << 52) - 1))))
 	{
 	  hi++;
+	  if ((hi & ((1LL << 52) - 1)) == 0)
+	    {
+	      if ((hi & (1LL << 53)) != 0)
+		hi -= 1LL << 52;
+	      u.ieee.exponent++;
+	    }
 	  u.ieee.negative2 = !sign;
 	  lo = (1LL << 53) - lo;
 	}
@@ -85,17 +133,18 @@ __mpn_construct_long_double (mp_srcptr frac_ptr, int expt, int sign)
       if (lzcount > 0)
 	{
 	  lo = lo << lzcount;
-	  u.ieee.exponent2 = u.ieee.exponent2 - lzcount;
+	  exponent2 = exponent2 - lzcount;
 	}
+      if (exponent2 > 0)
+	u.ieee.exponent2 = exponent2;
+      else
+	lo >>= 1 - exponent2;
     }
   else
-    {
-      u.ieee.negative2 = 0;
-      u.ieee.exponent2 = 0;
-    }
+    u.ieee.negative2 = 0;
 
   u.ieee.mantissa3 = lo & 0xffffffffLL;
-  u.ieee.mantissa2 = (lo >> 32) & 0xffffff;
+  u.ieee.mantissa2 = (lo >> 32) & 0xfffff;
   u.ieee.mantissa1 = hi & 0xffffffffLL;
   u.ieee.mantissa0 = (hi >> 32) & ((1LL << (LDBL_MANT_DIG - 86)) - 1);
 
diff --git a/sysdeps/ieee754/ldbl-128ibm/printf_fphex.c b/sysdeps/ieee754/ldbl-128ibm/printf_fphex.c
index 2a7b70fabf..b2ad25e31f 100644
--- a/sysdeps/ieee754/ldbl-128ibm/printf_fphex.c
+++ b/sysdeps/ieee754/ldbl-128ibm/printf_fphex.c
@@ -1,5 +1,5 @@
 /* Print floating point number in hexadecimal notation according to ISO C99.
-   Copyright (C) 1997,1998,1999,2000,2001,2002,2004,2006
+   Copyright (C) 1997,1998,1999,2000,2001,2002,2004,2006,2007
 	Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -35,21 +35,24 @@ do {									      \
 									      \
       lo = ((long long)eldbl.ieee.mantissa2 << 32) | eldbl.ieee.mantissa3;    \
       hi = ((long long)eldbl.ieee.mantissa0 << 32) | eldbl.ieee.mantissa1;    \
-   /* If the lower double is not a denomal or zero then set the hidden	      \
-      53rd bit.  */							      \
-      if (eldbl.ieee.exponent2 > 0x001)					      \
-	{								      \
-	  lo |= (1ULL << 52);						      \
-	  lo = lo << 7; /* pre-shift lo to match ieee854.  */		      \
-	  /* The lower double is normalized separately from the upper.  We    \
-	     may need to adjust the lower manitissa to reflect this.  */      \
-	  ediff = eldbl.ieee.exponent - eldbl.ieee.exponent2;		      \
-	  if (ediff > 53)						      \
-	    lo = lo >> (ediff-53);					      \
-	}								      \
-  									      \
-      if ((eldbl.ieee.negative != eldbl.ieee.negative2)			      \
-	  && ((eldbl.ieee.exponent2 != 0) && (lo != 0L)))		      \
+      lo <<= 7; /* pre-shift lo to match ieee854.  */			      \
+      /* If the lower double is not a denomal or zero then set the hidden     \
+	 53rd bit.  */							      \
+      if (eldbl.ieee.exponent2 != 0)					      \
+	lo |= (1ULL << (52 + 7));					      \
+      else								      \
+	lo <<= 1;							      \
+      /* The lower double is normalized separately from the upper.  We	      \
+	 may need to adjust the lower manitissa to reflect this.  */	      \
+      ediff = eldbl.ieee.exponent - eldbl.ieee.exponent2;		      \
+      if (ediff > 53 + 63)						      \
+	lo = 0;								      \
+      else if (ediff > 53)						      \
+	lo = lo >> (ediff - 53);					      \
+      else if (eldbl.ieee.exponent2 == 0 && ediff < 53)			      \
+	lo = lo << (53 - ediff);					      \
+      if (eldbl.ieee.negative != eldbl.ieee.negative2			      \
+	  && (eldbl.ieee.exponent2 != 0 || lo != 0L))			      \
 	{								      \
 	  lo = (1ULL << 60) - lo;					      \
 	  if (hi == 0L)							      \
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c b/sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c
index 3ca178a3c5..6999abc250 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c
@@ -1,5 +1,5 @@
 /* Return classification value corresponding to argument.
-   Copyright (C) 1997,1999,2002,2004,2006 Free Software Foundation, Inc.
+   Copyright (C) 1997,1999,2002,2004,2006,2007 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997 and
    		  Jakub Jelinek <jj@ultra.linux.cz>, 1999.
@@ -30,14 +30,16 @@
    * -NaN	fffn nnnn nnnn nnnn xxxx xxxx xxxx xxxx
    * +Inf	7ff0 0000 0000 0000 xxxx xxxx xxxx xxxx
    * -Inf	fff0 0000 0000 0000 xxxx xxxx xxxx xxxx
-   * +0		0000 0000 0000 0000
-   * -0		8000 0000 0000 0000
-   * +normal	001n nnnn nnnn nnnn (smallest)
-   * -normal	801n nnnn nnnn nnnn (smallest)
-   * +normal	7fen nnnn nnnn nnnn (largest)
-   * -normal	ffen nnnn nnnn nnnn (largest)
-   * +denorm	000n nnnn nnnn nnnn
-   * -denorm	800n nnnn nnnn nnnn
+   * +0		0000 0000 0000 0000 xxxx xxxx xxxx xxxx
+   * -0		8000 0000 0000 0000 xxxx xxxx xxxx xxxx
+   * +normal	0360 0000 0000 0000 0000 0000 0000 0000 (smallest)
+   * -normal	8360 0000 0000 0000 0000 0000 0000 0000 (smallest)
+   * +normal	7fef ffff ffff ffff 7c8f ffff ffff fffe (largest)
+   * +normal	ffef ffff ffff ffff fc8f ffff ffff fffe (largest)
+   * +denorm	0360 0000 0000 0000 8000 0000 0000 0001 (largest)
+   * -denorm	8360 0000 0000 0000 0000 0000 0000 0001 (largest)
+   * +denorm	000n nnnn nnnn nnnn xxxx xxxx xxxx xxxx
+   * -denorm	800n nnnn nnnn nnnn xxxx xxxx xxxx xxxx
    */
 
 int
@@ -59,12 +61,23 @@ ___fpclassifyl (long double x)
       /* +/-zero or +/- normal or +/- denormal */
       if (hx & 0x7fffffffffffffffULL) {
 	  /* +/- normal or +/- denormal */
-	  if ((hx & 0x7ff0000000000000ULL) >= 0x0360000000000000ULL) {
+	  if ((hx & 0x7ff0000000000000ULL) > 0x0360000000000000ULL) {
 	      /* +/- normal */
 	      retval = FP_NORMAL;
 	  } else {
-	      /* +/- denormal */
-	      retval = FP_SUBNORMAL;
+	      if ((hx & 0x7ff0000000000000ULL) == 0x0360000000000000ULL) {
+		  if ((lx & 0x7fffffffffffffff)	/* lower is non-zero */
+		  && ((lx^hx) & 0x8000000000000000ULL)) { /* and sign differs */
+		      /* +/- denormal */
+		      retval = FP_SUBNORMAL;
+		  } else {
+		      /* +/- normal */
+		      retval = FP_NORMAL;
+		  }
+	      } else {
+		  /* +/- denormal */
+		  retval = FP_SUBNORMAL;
+	      }
 	  }
       } else {
 	  /* +/- zero */
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c b/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c
index e35ce50829..39d0e6a5e3 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c
@@ -24,8 +24,8 @@ static char rcsid[] = "$NetBSD: $";
  *   Special cases:
  */
 
-#include "math.h"
-#include "math_private.h"
+#include <math.h>
+#include <math_private.h>
 #include <math_ldbl_opt.h>
 
 #ifdef __STDC__
@@ -35,7 +35,7 @@ static char rcsid[] = "$NetBSD: $";
 	long double x,y;
 #endif
 {
-	int64_t hx,hy,ihx,ihy,ilx,ily;
+	int64_t hx,hy,ihx,ihy,ilx;
 	u_int64_t lx,ly;
 
 	GET_LDOUBLE_WORDS64(hx,lx,x);
@@ -43,7 +43,6 @@ static char rcsid[] = "$NetBSD: $";
 	ihx = hx&0x7fffffffffffffffLL;		/* |hx| */
 	ilx = lx&0x7fffffffffffffffLL;		/* |lx| */
 	ihy = hy&0x7fffffffffffffffLL;		/* |hy| */
-	ily = ly&0x7fffffffffffffffLL;		/* |ly| */
 
 	if((((ihx&0x7ff0000000000000LL)==0x7ff0000000000000LL)&&
 	    ((ihx&0x000fffffffffffffLL)!=0)) ||   /* x is nan */
@@ -53,55 +52,67 @@ static char rcsid[] = "$NetBSD: $";
 	if(x==y)
 	    return y;		/* x=y, return y */
 	if(ihx == 0 && ilx == 0) {			/* x == 0 */
-	    SET_LDOUBLE_WORDS64(x,hy&0x8000000000000000ULL,1);/* return +-minsubnormal */
-	    y = x*x;
-	    if(y==x) return y; else return x;	/* raise underflow flag */
+	    long double u;
+	    hy = (hy & 0x8000000000000000ULL) | 1;
+	    SET_LDOUBLE_WORDS64(x,hy,0ULL);/* return +-minsubnormal */
+	    u = math_opt_barrier (x);
+	    u = u * u;
+	    math_force_eval (u);		/* raise underflow flag */
+	    return x;
 	}
-	if(ihx>=0) {			/* x > 0 */
-	    if(ihx>ihy||((ihx==ihy)&&(ilx>ily))) {	/* x > y, x -= ulp */
-
-	        if(ilx==0)
-		    hx--;
-		else
-		    lx--;
-	    } else {				/* x < y, x += ulp */
-	        if((hx==0x7fefffffffffffffLL)&&(lx==0x7c8ffffffffffffeLL))
-		  {
-		    SET_LDOUBLE_WORDS64(x,0x7ff0000000000000,0x8000000000000000);
-		    return x;
-		  }
-	        else if((hx==0xffefffffffffffffLL)&&(lx==0xfc8ffffffffffffeLL))
-		  {
-		    SET_LDOUBLE_WORDS64(x,0xfff0000000000000,0x8000000000000000);
-		    return x;
-		  }
-		else if((lx&0x7fffffffffffffff)==0) hx++;
-		else
-		  lx++;
+	
+	long double u;
+	if(x > y) {	/* x > y, x -= ulp */
+	    if((hx==0xffefffffffffffffLL)&&(lx==0xfc8ffffffffffffeLL))
+	      return x+x;	/* overflow, return -inf */
+	    if (hx >= 0x7ff0000000000000LL) {
+	      SET_LDOUBLE_WORDS64(u,0x7fefffffffffffffLL,0x7c8ffffffffffffeLL);
+	      return u;
 	    }
-	} else {				/* x < 0 */
-	    if(ihy>=0||ihx>ihy||((ihx==ihy)&&(ilx>ily))){/* x < y, x -= ulp */
-		if((lx&0x7fffffffffffffff)==0)
-		    hx--;
-		else
-		    lx--;
-	    } else {				/* x > y, x += ulp */
-		if((lx&0x7fffffffffffffff)==0) hx++;
-		else
-		  lx++;
+	    if(ihx <= 0x0360000000000000LL) {  /* x <= LDBL_MIN */
+	      u = math_opt_barrier (x);
+	      x -= __LDBL_DENORM_MIN__;
+	      if (ihx < 0x0360000000000000LL
+		  || (hx > 0 && (int64_t) lx <= 0)
+		  || (hx < 0 && (int64_t) lx > 1)) {
+		u = u * u;
+		math_force_eval (u);		/* raise underflow flag */
+	      }
+	      return x;
 	    }
-	}
-	hy = hx&0x7ff0000000000000LL;
-	if(hy==0x7ff0000000000000LL) return x+x;/* overflow  */
-	if(hy==0) {				/* underflow */
-	    y = x*x;
-	    if(y!=x) {		/* raise underflow flag */
-	        SET_LDOUBLE_WORDS64(y,hx,lx);
-		return y;
+	    if (ihx < 0x06a0000000000000LL) { /* ulp will denormal */
+	      SET_LDOUBLE_WORDS64(u,(hx&0x7ff0000000000000LL),0ULL);
+	      u *= 0x1.0000000000000p-105L;
+	    } else
+	      SET_LDOUBLE_WORDS64(u,(hx&0x7ff0000000000000LL)-0x0690000000000000LL,0ULL);
+	    return x - u;
+	} else {				/* x < y, x += ulp */
+	    if((hx==0x7fefffffffffffffLL)&&(lx==0x7c8ffffffffffffeLL))
+	      return x+x;	/* overflow, return +inf */
+	    if ((u_int64_t) hx >= 0xfff0000000000000ULL) {
+	      SET_LDOUBLE_WORDS64(u,0xffefffffffffffffLL,0xfc8ffffffffffffeLL);
+	      return u;
+	    }
+	    if(ihx <= 0x0360000000000000LL) {  /* x <= LDBL_MIN */
+	      u = math_opt_barrier (x);
+	      x += __LDBL_DENORM_MIN__;
+	      if (ihx < 0x0360000000000000LL
+		  || (hx > 0 && (int64_t) lx < 0 && lx != 0x8000000000000001LL)
+		  || (hx < 0 && (int64_t) lx >= 0)) {
+		u = u * u;
+		math_force_eval (u);		/* raise underflow flag */
+	      }
+	      if (x == 0.0L)	/* handle negative __LDBL_DENORM_MIN__ case */
+		x = -0.0L;
+	      return x;
 	    }
+	    if (ihx < 0x06a0000000000000LL) { /* ulp will denormal */
+	      SET_LDOUBLE_WORDS64(u,(hx&0x7ff0000000000000LL),0ULL);
+	      u *= 0x1.0000000000000p-105L;
+	    } else
+	      SET_LDOUBLE_WORDS64(u,(hx&0x7ff0000000000000LL)-0x0690000000000000LL,0ULL);
+	    return x + u;
 	}
-	SET_LDOUBLE_WORDS64(x,hx,lx);
-	return x;
 }
 strong_alias (__nextafterl, __nexttowardl)
 long_double_symbol (libm, __nextafterl, nextafterl);
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c b/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c
index 3335100592..e2f6521f57 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c
@@ -26,7 +26,7 @@ static char rcsid[] = "$NetBSD: $";
  */
 
 #include "math.h"
-#include "math_private.h"
+#include <math_private.h>
 #include <math_ldbl_opt.h>
 #include <float.h>
 
@@ -55,10 +55,12 @@ static char rcsid[] = "$NetBSD: $";
 	   return x+y;
 	if((long double) x==y) return y;	/* x=y, return y */
 	if((ix|lx)==0) {			/* x == 0 */
-	    double x2;
+	    double u;
 	    INSERT_WORDS(x,(u_int32_t)((hy>>32)&0x80000000),1);/* return +-minsub */
-	    x2 = x*x;
-	    if(x2==x) return x2; else return x;	/* raise underflow flag */
+	    u = math_opt_barrier (x);
+	    u = u * u;
+	    math_force_eval (u);		/* raise underflow flag */
+	    return x;
 	}
 	if(hx>=0) {				/* x > 0 */
 	    if (hy<0||(ix>>20)>(iy>>52)
@@ -89,16 +91,13 @@ static char rcsid[] = "$NetBSD: $";
 	if(hy>=0x7ff00000) {
 	  x = x+x;	/* overflow  */
 	  if (FLT_EVAL_METHOD != 0 && FLT_EVAL_METHOD != 1)
-	    /* Force conversion to float.  */
-	    asm ("" : "=m"(x) : "m"(x));
+	    /* Force conversion to double.  */
+	    asm ("" : "+m"(x));
 	  return x;
 	}
-	if(hy<0x00100000) {		/* underflow */
-	    double x2 = x*x;
-	    if(x2!=x) {		/* raise underflow flag */
-	        INSERT_WORDS(x2,hx,lx);
-		return x2;
-	    }
+	if(hy<0x00100000) {
+	    double u = x*x;			/* underflow */
+	    math_force_eval (u);		/* raise underflow flag */
 	}
 	INSERT_WORDS(x,hx,lx);
 	return x;
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c b/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c
index a9373ff822..cf655fad16 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c
@@ -19,8 +19,9 @@ static char rcsid[] = "$NetBSD: $";
 #endif
 
 #include "math.h"
-#include "math_private.h"
+#include <math_private.h>
 #include <math_ldbl_opt.h>
+#include <float.h>
 
 #ifdef __STDC__
 	float __nexttowardf(float x, long double y)
@@ -46,10 +47,12 @@ static char rcsid[] = "$NetBSD: $";
 	   return x+y;
 	if((long double) x==y) return y;	/* x=y, return y */
 	if(ix==0) {				/* x == 0 */
-	    float x2;
+	    float u;
 	    SET_FLOAT_WORD(x,(u_int32_t)((hy>>32)&0x80000000)|1);/* return +-minsub*/
-	    x2 = x*x;
-	    if(x2==x) return x2; else return x;	/* raise underflow flag */
+	    u = math_opt_barrier (x);
+	    u = u * u;
+	    math_force_eval (u);		/* raise underflow flag */
+	    return x;
 	}
 	if(hx>=0) {				/* x > 0 */
 	    if(hy<0||(ix>>23)>(iy>>52)-0x380
@@ -69,13 +72,16 @@ static char rcsid[] = "$NetBSD: $";
 	    }
 	}
 	hy = hx&0x7f800000;
-	if(hy>=0x7f800000) return x+x;	/* overflow  */
+	if(hy>=0x7f800000) {
+	  x = x+x;	/* overflow  */
+	  if (FLT_EVAL_METHOD != 0)
+	    /* Force conversion to float.  */
+	    asm ("" : "+m"(x));
+	  return x;
+	}
 	if(hy<0x00800000) {		/* underflow */
-	    float x2 = x*x;
-	    if(x2!=x) {		/* raise underflow flag */
-	        SET_FLOAT_WORD(x2,hx);
-		return x2;
-	    }
+	    float u = x*x;
+	    math_force_eval (u);	/* raise underflow flag */
 	}
 	SET_FLOAT_WORD(x,hx);
 	return x;