PowerPC floating point little-endian [4 of 15]

http://sourceware.org/ml/libc-alpha/2013-08/msg00084.html

Another batch of ieee854 macros and union replacement.  These four
files also have bugs fixed with this patch.  The fact that the two
doubles in an IBM long double may have different signs means that
negation and absolute value operations can't just twiddle one sign bit
as you can with ieee864 style extended double.  fmodl, remainderl,
erfl and erfcl all had errors of this type.  erfl also returned +1 for
large magnitude negative input where it should return -1.  The hypotl
error is innocuous since the value adjusted twice is only used as a
flag.  The e_hypotl.c tests for large "a" and small "b" are mutually
exclusive because we've already exited when x/y > 2**120.  That allows
some further small simplifications.

	[BZ #15734], [BZ #15735]
	* sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Rewrite
	all uses of ieee875 long double macros and unions.  Simplify test
	for 0.0L.  Correct |x|<|y| and |x|=|y| test.  Use
	ldbl_extract_mantissa value for ix,iy exponents.  Properly
	normalize after ldbl_extract_mantissa, and don't add hidden bit
	already handled.  Don't treat low word of ieee854 mantissa like
	low word of IBM long double and mask off bit when testing for
	zero.
	* sysdeps/ieee754/ldbl-128ibm/e_hypotl.c (__ieee754_hypotl): Rewrite
	all uses of ieee875 long double macros and unions.  Simplify tests
	for 0.0L and inf.  Correct double adjustment of k.  Delete dead code
	adjusting ha,hb.  Simplify code setting kld.  Delete two600 and
	two1022, instead use their values.  Recognise that tests for large
	"a" and small "b" are mutually exclusive.  Rename vars.  Comment.
	* sysdeps/ieee754/ldbl-128ibm/e_remainderl.c (__ieee754_remainderl):
	Rewrite all uses of ieee875 long double macros and unions.  Simplify
	test for 0.0L and nan.  Correct negation.
	* sysdeps/ieee754/ldbl-128ibm/s_erfl.c (__erfl): Rewrite all uses of
	ieee875 long double macros and unions.  Correct output for large
	magnitude x.  Correct absolute value calculation.
	(__erfcl): Likewise.
	* math/libm-test.inc: Add tests for errors discovered in IBM long
	double versions of fmodl, remainderl, erfl and erfcl.

1 files changed, 66 insertions, 59 deletions

diff --git a/sysdeps/ieee754/ldbl-128ibm/e_fmodl.c b/sysdeps/ieee754/ldbl-128ibm/e_fmodl.c
index a60963c..a140fb3 100644
--- a/sysdeps/ieee754/ldbl-128ibm/e_fmodl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/e_fmodl.c
@@ -27,76 +27,83 @@ static const long double one = 1.0, Zero[] = {0.0, -0.0,};
 long double
 __ieee754_fmodl (long double x, long double y)
 {
-	int64_t n,hx,hy,hz,ix,iy,sx, i;
-	u_int64_t lx,ly,lz;
-	int temp;
+	int64_t hx, hy, hz, sx, sy;
+	uint64_t lx, ly, lz;
+	int n, ix, iy;
+	double xhi, xlo, yhi, ylo;
 
-	GET_LDOUBLE_WORDS64(hx,lx,x);
-	GET_LDOUBLE_WORDS64(hy,ly,y);
+	ldbl_unpack (x, &xhi, &xlo);
+	EXTRACT_WORDS64 (hx, xhi);
+	EXTRACT_WORDS64 (lx, xlo);
+	ldbl_unpack (y, &yhi, &ylo);
+	EXTRACT_WORDS64 (hy, yhi);
+	EXTRACT_WORDS64 (ly, ylo);
 	sx = hx&0x8000000000000000ULL;		/* sign of x */
-	hx ^=sx;				/* |x| */
-	hy &= 0x7fffffffffffffffLL;		/* |y| */
+	hx ^= sx;				/* |x| */
+	sy = hy&0x8000000000000000ULL;		/* sign of y */
+	hy ^= sy;				/* |y| */
 
     /* purge off exception values */
-	if(__builtin_expect((hy|(ly&0x7fffffffffffffff))==0 ||
+	if(__builtin_expect(hy==0 ||
 			    (hx>=0x7ff0000000000000LL)|| /* y=0,or x not finite */
 			    (hy>0x7ff0000000000000LL),0))	/* or y is NaN */
 	    return (x*y)/(x*y);
-	if(__builtin_expect(hx<=hy,0)) {
-	    if((hx<hy)||(lx<ly)) return x;	/* |x|<|y| return x */
-	    if(lx==ly)
-		return Zero[(u_int64_t)sx>>63];	/* |x|=|y| return x*0*/
+	if (__builtin_expect (hx <= hy, 0))
+	  {
+	    /* If |x| < |y| return x.  */
+	    if (hx < hy)
+	      return x;
+	    /* At this point the absolute value of the high doubles of
+	       x and y must be equal.  */
+	    /* If the low double of y is the same sign as the high
+	       double of y (ie. the low double increases |y|)...  */
+	    if (((ly ^ sy) & 0x8000000000000000LL) == 0
+		/* ... then a different sign low double to high double
+		   for x or same sign but lower magnitude...  */
+		&& (int64_t) (lx ^ sx) < (int64_t) (ly ^ sy))
+	      /* ... means |x| < |y|.  */
+	      return x;
+	    /* If the low double of x differs in sign to the high
+	       double of x (ie. the low double decreases |x|)...  */
+	    if (((lx ^ sx) & 0x8000000000000000LL) != 0
+		/* ... then a different sign low double to high double
+		   for y with lower magnitude (we've already caught
+		   the same sign for y case above)...  */
+		&& (int64_t) (lx ^ sx) > (int64_t) (ly ^ sy))
+	      /* ... means |x| < |y|.  */
+	      return x;
+	    /* If |x| == |y| return x*0.  */
+	    if ((lx ^ sx) == (ly ^ sy))
+	      return Zero[(uint64_t) sx >> 63];
 	}
 
-    /* determine ix = ilogb(x) */
-	if(__builtin_expect(hx<0x0010000000000000LL,0)) {	/* subnormal x */
-	    if(hx==0) {
-		for (ix = -1043, i=lx; i>0; i<<=1) ix -=1;
-	    } else {
-		for (ix = -1022, i=(hx<<11); i>0; i<<=1) ix -=1;
-	    }
-	} else ix = (hx>>52)-0x3ff;
-
-    /* determine iy = ilogb(y) */
-	if(__builtin_expect(hy<0x0010000000000000LL,0)) {	/* subnormal y */
-	    if(hy==0) {
-		for (iy = -1043, i=ly; i>0; i<<=1) iy -=1;
-	    } else {
-		for (iy = -1022, i=(hy<<11); i>0; i<<=1) iy -=1;
-	    }
-	} else iy = (hy>>52)-0x3ff;
-
     /* Make the IBM extended format 105 bit mantissa look like the ieee854 112
        bit mantissa so the following operations will give the correct
        result.  */
-	ldbl_extract_mantissa(&hx, &lx, &temp, x);
-	ldbl_extract_mantissa(&hy, &ly, &temp, y);
+	ldbl_extract_mantissa(&hx, &lx, &ix, x);
+	ldbl_extract_mantissa(&hy, &ly, &iy, y);
 
-    /* set up {hx,lx}, {hy,ly} and align y to x */
-	if(__builtin_expect(ix >= -1022, 1))
-	    hx = 0x0001000000000000LL|(0x0000ffffffffffffLL&hx);
-	else {		/* subnormal x, shift x to normal */
-	    n = -1022-ix;
-	    if(n<=63) {
-		hx = (hx<<n)|(lx>>(64-n));
-		lx <<= n;
-	    } else {
-		hx = lx<<(n-64);
-		lx = 0;
-	    }
-	}
-	if(__builtin_expect(iy >= -1022, 1))
-	    hy = 0x0001000000000000LL|(0x0000ffffffffffffLL&hy);
-	else {		/* subnormal y, shift y to normal */
-	    n = -1022-iy;
-	    if(n<=63) {
-		hy = (hy<<n)|(ly>>(64-n));
-		ly <<= n;
-	    } else {
-		hy = ly<<(n-64);
-		ly = 0;
-	    }
-	}
+	if (__builtin_expect (ix == -IEEE754_DOUBLE_BIAS, 0))
+	  {
+	    /* subnormal x, shift x to normal.  */
+	    while ((hx & (1LL << 48)) == 0)
+	      {
+		hx = (hx << 1) | (lx >> 63);
+		lx = lx << 1;
+		ix -= 1;
+	      }
+	  }
+
+	if (__builtin_expect (iy == -IEEE754_DOUBLE_BIAS, 0))
+	  {
+	    /* subnormal y, shift y to normal.  */
+	    while ((hy & (1LL << 48)) == 0)
+	      {
+		hy = (hy << 1) | (ly >> 63);
+		ly = ly << 1;
+		iy -= 1;
+	      }
+	  }
 
     /* fix point fmod */
 	n = ix - iy;
@@ -104,7 +111,7 @@ __ieee754_fmodl (long double x, long double y)
 	    hz=hx-hy;lz=lx-ly; if(lx<ly) hz -= 1;
 	    if(hz<0){hx = hx+hx+(lx>>63); lx = lx+lx;}
 	    else {
-		if((hz|(lz&0x7fffffffffffffff))==0)		/* return sign(x)*0 */
+		if((hz|lz)==0)		/* return sign(x)*0 */
 		    return Zero[(u_int64_t)sx>>63];
 		hx = hz+hz+(lz>>63); lx = lz+lz;
 	    }
@@ -113,7 +120,7 @@ __ieee754_fmodl (long double x, long double y)
 	if(hz>=0) {hx=hz;lx=lz;}
 
     /* convert back to floating value and restore the sign */
-	if((hx|(lx&0x7fffffffffffffff))==0)			/* return sign(x)*0 */
+	if((hx|lx)==0)			/* return sign(x)*0 */
 	    return Zero[(u_int64_t)sx>>63];
 	while(hx<0x0001000000000000LL) {	/* normalize x */
 	    hx = hx+hx+(lx>>63); lx = lx+lx;