crt: arm: Use bit twiddling implementations of remquo*

This avoids issues when the quotient exceeds the range for integer
variables.

Signed-off-by: Martin Storsjö <martin@martin.st>
diff --git a/mingw-w64-crt/Makefile.am b/mingw-w64-crt/Makefile.am
index e07feba..d499e66 100644
--- a/mingw-w64-crt/Makefile.am
+++ b/mingw-w64-crt/Makefile.am
@@ -303,9 +303,9 @@
   math/arm-common/remainder.c \
   math/arm-common/remainderf.c \
   math/arm-common/remainderl.c \
-  math/arm-common/remquo.c \
-  math/arm-common/remquof.c \
   math/arm-common/remquol.c \
+  math/arm-common/s_remquo.c \
+  math/arm-common/s_remquof.c \
   math/arm-common/scalbn.c
 endif
 
@@ -340,9 +340,9 @@
   math/arm-common/remainder.c \
   math/arm-common/remainderf.c \
   math/arm-common/remainderl.c \
-  math/arm-common/remquo.c \
-  math/arm-common/remquof.c \
   math/arm-common/remquol.c \
+  math/arm-common/s_remquo.c \
+  math/arm-common/s_remquof.c \
   math/arm-common/scalbn.c \
   math/arm64/exp2.S \
   math/arm64/exp2f.S \
diff --git a/mingw-w64-crt/math/arm-common/remquo.c b/mingw-w64-crt/math/arm-common/remquo.c
deleted file mode 100644
index d8ec534..0000000
--- a/mingw-w64-crt/math/arm-common/remquo.c
+++ /dev/null
@@ -1,35 +0,0 @@
-/**
- * This file has no copyright assigned and is placed in the Public Domain.
- * This file is part of the mingw-w64 runtime package.
- * No warranty is given; refer to the file DISCLAIMER.PD within this package.
- */
-
-#include <math.h>
-#include <errno.h>
-
-double remquo(double x, double y, int *quo)
-{
-  if (isnan(x))
-    return x;
-  if (isnan(y))
-    return y;
-  if (isinf(x) || y == 0)
-    return NAN;
-  double div = x/y;
-  double integral;
-  double frac = modf(div, &integral);
-  int iintegral = (int)integral;
-  if (frac == 0.5) {
-    if (iintegral & 1)
-      *quo = iintegral + 1;
-    else
-      *quo = iintegral;
-  } else if (frac == -0.5) {
-    if (iintegral & 1)
-      *quo = iintegral - 1;
-    else
-      *quo = iintegral;
-  } else
-    *quo = round(div);
-  return x - *quo * y;
-}
diff --git a/mingw-w64-crt/math/arm-common/remquof.c b/mingw-w64-crt/math/arm-common/remquof.c
deleted file mode 100644
index d928002..0000000
--- a/mingw-w64-crt/math/arm-common/remquof.c
+++ /dev/null
@@ -1,35 +0,0 @@
-/**
- * This file has no copyright assigned and is placed in the Public Domain.
- * This file is part of the mingw-w64 runtime package.
- * No warranty is given; refer to the file DISCLAIMER.PD within this package.
- */
-
-#include <math.h>
-#include <errno.h>
-
-float remquof(float x, float y, int *quo)
-{
-  if (isnan(x))
-    return x;
-  if (isnan(y))
-    return y;
-  if (isinf(x) || y == 0)
-    return NAN;
-  float div = x/y;
-  float integral;
-  float frac = modff(div, &integral);
-  int iintegral = (int)integral;
-  if (frac == 0.5) {
-    if (iintegral & 1)
-      *quo = iintegral + 1;
-    else
-      *quo = iintegral;
-  } else if (frac == -0.5) {
-    if (iintegral & 1)
-      *quo = iintegral - 1;
-    else
-      *quo = iintegral;
-  } else
-    *quo = roundf(div);
-  return x - *quo * y;
-}
diff --git a/mingw-w64-crt/math/arm-common/s_remquo.c b/mingw-w64-crt/math/arm-common/s_remquo.c
new file mode 100644
index 0000000..48b7317
--- /dev/null
+++ b/mingw-w64-crt/math/arm-common/s_remquo.c
@@ -0,0 +1,154 @@
+/* @(#)e_fmod.c 1.3 95/01/18 */
+/*-
+ * ====================================================
+ * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
+ *
+ * Developed at SunSoft, a Sun Microsystems, Inc. business.
+ * Permission to use, copy, modify, and distribute this
+ * software is freely granted, provided that this notice
+ * is preserved.
+ * ====================================================
+ */
+
+#include <sys/cdefs.h>
+
+#include <float.h>
+
+#include <math.h>
+#include "../bsd_private_base.h"
+
+static const double Zero[] = {0.0, -0.0,};
+
+/*
+ * Return the IEEE remainder and set *quo to the last n bits of the
+ * quotient, rounded to the nearest integer.  We choose n=31 because
+ * we wind up computing all the integer bits of the quotient anyway as
+ * a side-effect of computing the remainder by the shift and subtract
+ * method.  In practice, this is far more bits than are needed to use
+ * remquo in reduction algorithms.
+ */
+double
+remquo(double x, double y, int *quo)
+{
+	int32_t n,hx,hy,hz,ix,iy,sx,i;
+	u_int32_t lx,ly,lz,q,sxy;
+
+	EXTRACT_WORDS(hx,lx,x);
+	EXTRACT_WORDS(hy,ly,y);
+	sxy = (hx ^ hy) & 0x80000000;
+	sx = hx&0x80000000;		/* sign of x */
+	hx ^=sx;		/* |x| */
+	hy &= 0x7fffffff;	/* |y| */
+
+    /* purge off exception values */
+	if((hy|ly)==0||(hx>=0x7ff00000)||	/* y=0,or x not finite */
+	  ((hy|((ly|-ly)>>31))>0x7ff00000))	/* or y is NaN */
+	    return nan_mix_op(x, y, *)/nan_mix_op(x, y, *);
+	if(hx<=hy) {
+	    if((hx<hy)||(lx<ly)) {
+		q = 0;
+		goto fixup;	/* |x|<|y| return x or x-y */
+	    }
+	    if(lx==ly) {
+		*quo = (sxy ? -1 : 1);
+		return Zero[(u_int32_t)sx>>31];	/* |x|=|y| return x*0*/
+	    }
+	}
+
+    /* determine ix = ilogb(x) */
+	if(hx<0x00100000) {	/* 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>>20)-1023;
+
+    /* determine iy = ilogb(y) */
+	if(hy<0x00100000) {	/* 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>>20)-1023;
+
+    /* set up {hx,lx}, {hy,ly} and align y to x */
+	if(ix >= -1022)
+	    hx = 0x00100000|(0x000fffff&hx);
+	else {		/* subnormal x, shift x to normal */
+	    n = -1022-ix;
+	    if(n<=31) {
+	        hx = (hx<<n)|(lx>>(32-n));
+	        lx <<= n;
+	    } else {
+		hx = lx<<(n-32);
+		lx = 0;
+	    }
+	}
+	if(iy >= -1022)
+	    hy = 0x00100000|(0x000fffff&hy);
+	else {		/* subnormal y, shift y to normal */
+	    n = -1022-iy;
+	    if(n<=31) {
+	        hy = (hy<<n)|(ly>>(32-n));
+	        ly <<= n;
+	    } else {
+		hy = ly<<(n-32);
+		ly = 0;
+	    }
+	}
+
+    /* fix point fmod */
+	n = ix - iy;
+	q = 0;
+	while(n--) {
+	    hz=hx-hy;lz=lx-ly; if(lx<ly) hz -= 1;
+	    if(hz<0){hx = hx+hx+(lx>>31); lx = lx+lx;}
+	    else {hx = hz+hz+(lz>>31); lx = lz+lz; q++;}
+	    q <<= 1;
+	}
+	hz=hx-hy;lz=lx-ly; if(lx<ly) hz -= 1;
+	if(hz>=0) {hx=hz;lx=lz;q++;}
+
+    /* convert back to floating value and restore the sign */
+	if((hx|lx)==0) {			/* return sign(x)*0 */
+	    q &= 0x7fffffff;
+	    *quo = (sxy ? -q : q);
+	    return Zero[(u_int32_t)sx>>31];
+	}
+	while(hx<0x00100000) {		/* normalize x */
+	    hx = hx+hx+(lx>>31); lx = lx+lx;
+	    iy -= 1;
+	}
+	if(iy>= -1022) {	/* normalize output */
+	    hx = ((hx-0x00100000)|((iy+1023)<<20));
+	} else {		/* subnormal output */
+	    n = -1022 - iy;
+	    if(n<=20) {
+		lx = (lx>>n)|((u_int32_t)hx<<(32-n));
+		hx >>= n;
+	    } else if (n<=31) {
+		lx = (hx<<(32-n))|(lx>>n); hx = 0;
+	    } else {
+		lx = hx>>(n-32); hx = 0;
+	    }
+	}
+fixup:
+	INSERT_WORDS(x,hx,lx);
+	y = fabs(y);
+	if (y < 0x1p-1021) {
+	    if (x+x>y || (x+x==y && (q & 1))) {
+		q++;
+		x-=y;
+	    }
+	} else if (x>0.5*y || (x==0.5*y && (q & 1))) {
+	    q++;
+	    x-=y;
+	}
+	GET_HIGH_WORD(hx,x);
+	SET_HIGH_WORD(x,hx^sx);
+	q &= 0x7fffffff;
+	*quo = (sxy ? -q : q);
+	return x;
+}
diff --git a/mingw-w64-crt/math/arm-common/s_remquof.c b/mingw-w64-crt/math/arm-common/s_remquof.c
new file mode 100644
index 0000000..f2a4e0d
--- /dev/null
+++ b/mingw-w64-crt/math/arm-common/s_remquof.c
@@ -0,0 +1,121 @@
+/* @(#)e_fmod.c 1.3 95/01/18 */
+/*-
+ * ====================================================
+ * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
+ *
+ * Developed at SunSoft, a Sun Microsystems, Inc. business.
+ * Permission to use, copy, modify, and distribute this
+ * software is freely granted, provided that this notice
+ * is preserved.
+ * ====================================================
+ */
+
+#include <sys/cdefs.h>
+
+#include <math.h>
+#include "../bsd_private_base.h"
+
+static const float Zero[] = {0.0, -0.0,};
+
+/*
+ * Return the IEEE remainder and set *quo to the last n bits of the
+ * quotient, rounded to the nearest integer.  We choose n=31 because
+ * we wind up computing all the integer bits of the quotient anyway as
+ * a side-effect of computing the remainder by the shift and subtract
+ * method.  In practice, this is far more bits than are needed to use
+ * remquo in reduction algorithms.
+ */
+float
+remquof(float x, float y, int *quo)
+{
+	int32_t n,hx,hy,hz,ix,iy,sx,i;
+	u_int32_t q,sxy;
+
+	GET_FLOAT_WORD(hx,x);
+	GET_FLOAT_WORD(hy,y);
+	sxy = (hx ^ hy) & 0x80000000;
+	sx = hx&0x80000000;		/* sign of x */
+	hx ^=sx;		/* |x| */
+	hy &= 0x7fffffff;	/* |y| */
+
+    /* purge off exception values */
+	if(hy==0||hx>=0x7f800000||hy>0x7f800000) /* y=0,NaN;or x not finite */
+	    return nan_mix_op(x, y, *)/nan_mix_op(x, y, *);
+	if(hx<hy) {
+	    q = 0;
+	    goto fixup;	/* |x|<|y| return x or x-y */
+	} else if(hx==hy) {
+	    *quo = (sxy ? -1 : 1);
+	    return Zero[(u_int32_t)sx>>31];	/* |x|=|y| return x*0*/
+	}
+
+    /* determine ix = ilogb(x) */
+	if(hx<0x00800000) {	/* subnormal x */
+	    for (ix = -126,i=(hx<<8); i>0; i<<=1) ix -=1;
+	} else ix = (hx>>23)-127;
+
+    /* determine iy = ilogb(y) */
+	if(hy<0x00800000) {	/* subnormal y */
+	    for (iy = -126,i=(hy<<8); i>0; i<<=1) iy -=1;
+	} else iy = (hy>>23)-127;
+
+    /* set up {hx,lx}, {hy,ly} and align y to x */
+	if(ix >= -126)
+	    hx = 0x00800000|(0x007fffff&hx);
+	else {		/* subnormal x, shift x to normal */
+	    n = -126-ix;
+	    hx <<= n;
+	}
+	if(iy >= -126)
+	    hy = 0x00800000|(0x007fffff&hy);
+	else {		/* subnormal y, shift y to normal */
+	    n = -126-iy;
+	    hy <<= n;
+	}
+
+    /* fix point fmod */
+	n = ix - iy;
+	q = 0;
+	while(n--) {
+	    hz=hx-hy;
+	    if(hz<0) hx = hx << 1;
+	    else {hx = hz << 1; q++;}
+	    q <<= 1;
+	}
+	hz=hx-hy;
+	if(hz>=0) {hx=hz;q++;}
+
+    /* convert back to floating value and restore the sign */
+	if(hx==0) {				/* return sign(x)*0 */
+	    q &= 0x7fffffff;
+	    *quo = (sxy ? -q : q);
+	    return Zero[(u_int32_t)sx>>31];
+	}
+	while(hx<0x00800000) {		/* normalize x */
+	    hx <<= 1;
+	    iy -= 1;
+	}
+	if(iy>= -126) {		/* normalize output */
+	    hx = ((hx-0x00800000)|((iy+127)<<23));
+	} else {		/* subnormal output */
+	    n = -126 - iy;
+	    hx >>= n;
+	}
+fixup:
+	SET_FLOAT_WORD(x,hx);
+	y = fabsf(y);
+	if (y < 0x1p-125f) {
+	    if (x+x>y || (x+x==y && (q & 1))) {
+		q++;
+		x-=y;
+	    }
+	} else if (x>0.5f*y || (x==0.5f*y && (q & 1))) {
+	    q++;
+	    x-=y;
+	}
+	GET_FLOAT_WORD(hx,x);
+	SET_FLOAT_WORD(x,hx^sx);
+	q &= 0x7fffffff;
+	*quo = (sxy ? -q : q);
+	return x;
+}
diff --git a/mingw-w64-crt/math/bsd_private_base.h b/mingw-w64-crt/math/bsd_private_base.h
index de5b935..598b815 100644
--- a/mingw-w64-crt/math/bsd_private_base.h
+++ b/mingw-w64-crt/math/bsd_private_base.h
@@ -121,3 +121,28 @@
 } while (0)
 #endif
 #endif /* FLT_EVAL_METHOD */
+
+/*
+ * Mix 0, 1 or 2 NaNs.  First add 0 to each arg.  This normally just turns
+ * signaling NaNs into quiet NaNs by setting a quiet bit.  We do this
+ * because we want to never return a signaling NaN, and also because we
+ * don't want the quiet bit to affect the result.  Then mix the converted
+ * args using the specified operation.
+ *
+ * When one arg is NaN, the result is typically that arg quieted.  When both
+ * args are NaNs, the result is typically the quietening of the arg whose
+ * mantissa is largest after quietening.  When neither arg is NaN, the
+ * result may be NaN because it is indeterminate, or finite for subsequent
+ * construction of a NaN as the indeterminate 0.0L/0.0L.
+ *
+ * Technical complications: the result in bits after rounding to the final
+ * precision might depend on the runtime precision and/or on compiler
+ * optimizations, especially when different register sets are used for
+ * different precisions.  Try to make the result not depend on at least the
+ * runtime precision by always doing the main mixing step in long double
+ * precision.  Try to reduce dependencies on optimizations by adding the
+ * the 0's in different precisions (unless everything is in long double
+ * precision).
+ */
+#define nan_mix(x, y)		(nan_mix_op((x), (y), +))
+#define nan_mix_op(x, y, op)	(((x) + 0.0L) op ((y) + 0))