From ee2ee92d62c43f6658d37ddea4c316d2089d0fe9 Mon Sep 17 00:00:00 2001 From: Szabolcs Nagy Date: Tue, 3 Sep 2013 04:09:12 +0000 Subject: math: rewrite remainder functions (remainder, remquo, fmod, modf) * results are exact * modfl follows truncl (raises inexact flag spuriously now) * modf and modff only had cosmetic cleanup * remainder is just a wrapper around remquo now * using iterative shift+subtract for remquo and fmod * ld80 and ld128 are supported as well --- src/math/remquol.c | 270 ++++++++++++++++++++--------------------------------- 1 file changed, 100 insertions(+), 170 deletions(-) (limited to 'src/math/remquol.c') diff --git a/src/math/remquol.c b/src/math/remquol.c index a2e11728..9b065c00 100644 --- a/src/math/remquol.c +++ b/src/math/remquol.c @@ -1,15 +1,3 @@ -/* origin: FreeBSD /usr/src/lib/msun/src/s_remquol.c */ -/*- - * ==================================================== - * 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 "libm.h" #if LDBL_MANT_DIG == 53 && LDBL_MAX_EXP == 1024 @@ -18,177 +6,119 @@ long double remquol(long double x, long double y, int *quo) return remquo(x, y, quo); } #elif (LDBL_MANT_DIG == 64 || LDBL_MANT_DIG == 113) && LDBL_MAX_EXP == 16384 - -#define BIAS (LDBL_MAX_EXP - 1) - -#if LDBL_MANL_SIZE > 32 -typedef uint64_t manl_t; -#else -typedef uint32_t manl_t; -#endif - -#if LDBL_MANH_SIZE > 32 -typedef uint64_t manh_t; -#else -typedef uint32_t manh_t; -#endif - -/* - * These macros add and remove an explicit integer bit in front of the - * fractional mantissa, if the architecture doesn't have such a bit by - * default already. - */ -#ifdef LDBL_IMPLICIT_NBIT -#define SET_NBIT(hx) ((hx) | (1ULL << LDBL_MANH_SIZE)) -#define HFRAC_BITS LDBL_MANH_SIZE -#else -#define SET_NBIT(hx) (hx) -#define HFRAC_BITS (LDBL_MANH_SIZE - 1) -#endif - -#define MANL_SHIFT (LDBL_MANL_SIZE - 1) - -static const long 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. - * - * Assumptions: - * - The low part of the mantissa fits in a manl_t exactly. - * - The high part of the mantissa fits in an int64_t with enough room - * for an explicit integer bit in front of the fractional bits. - */ long double remquol(long double x, long double y, int *quo) { - union IEEEl2bits ux, uy; - int64_t hx,hz; /* We need a carry bit even if LDBL_MANH_SIZE is 32. */ - manh_t hy; - manl_t lx,ly,lz; - int ix,iy,n,q,sx,sxy; - - ux.e = x; - uy.e = y; - sx = ux.bits.sign; - sxy = sx ^ uy.bits.sign; - ux.bits.sign = 0; /* |x| */ - uy.bits.sign = 0; /* |y| */ - x = ux.e; - - /* purge off exception values */ - if ((uy.bits.exp|uy.bits.manh|uy.bits.manl)==0 || /* y=0 */ - (ux.bits.exp == BIAS + LDBL_MAX_EXP) || /* or x not finite */ - (uy.bits.exp == BIAS + LDBL_MAX_EXP && - ((uy.bits.manh&~LDBL_NBIT)|uy.bits.manl)!=0)) /* or y is NaN */ + union ldshape ux = {x}, uy = {y}; + int ex = ux.i.se & 0x7fff; + int ey = uy.i.se & 0x7fff; + int sx = ux.i.se >> 15; + int sy = uy.i.se >> 15; + uint32_t q; + + *quo = 0; + if (y == 0 || isnan(y) || ex == 0x7fff) return (x*y)/(x*y); - if (ux.bits.exp <= uy.bits.exp) { - if ((ux.bits.exp < uy.bits.exp) || - (ux.bits.manh <= uy.bits.manh && - (ux.bits.manh < uy.bits.manh || - ux.bits.manl < uy.bits.manl))) { - q = 0; - goto fixup; /* |x|<|y| return x or x-y */ - } - if (ux.bits.manh == uy.bits.manh && ux.bits.manl == uy.bits.manl) { - *quo = sxy ? -1 : 1; - return Zero[sx]; /* |x|=|y| return x*0*/ - } - } - - /* determine ix = ilogb(x) */ - if (ux.bits.exp == 0) { /* subnormal x */ - ux.e *= 0x1.0p512; - ix = ux.bits.exp - (BIAS + 512); - } else { - ix = ux.bits.exp - BIAS; + if (x == 0) + return x; + + /* normalize x and y */ + if (!ex) { + ux.i.se = ex; + ux.f *= 0x1p120f; + ex = ux.i.se - 120; } - - /* determine iy = ilogb(y) */ - if (uy.bits.exp == 0) { /* subnormal y */ - uy.e *= 0x1.0p512; - iy = uy.bits.exp - (BIAS + 512); - } else { - iy = uy.bits.exp - BIAS; + if (!ey) { + uy.i.se = ey; + uy.f *= 0x1p120f; + ey = uy.i.se - 120; } - /* set up {hx,lx}, {hy,ly} and align y to x */ - hx = SET_NBIT(ux.bits.manh); - hy = SET_NBIT(uy.bits.manh); - lx = ux.bits.manl; - ly = uy.bits.manl; - - /* 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>>MANL_SHIFT); - lx = lx + lx; - } else { - hx = hz + hz + (lz>>MANL_SHIFT); - lx = lz + lz; + if (ex >= ey) { + /* x mod y */ +#if LDBL_MANT_DIG == 64 + uint64_t i, mx, my; + mx = ux.i.m; + my = uy.i.m; + for (; ex > ey; ex--) { + i = mx - my; + if (mx >= my) { + mx = 2*i; + q++; + q <<= 1; + } else if (2*mx < mx) { + mx = 2*mx - my; + q <<= 1; + q++; + } else { + mx = 2*mx; + q <<= 1; + } + } + i = mx - my; + if (mx >= my) { + mx = i; 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[sx]; - } - while (hx < (1ULL<>MANL_SHIFT); - lx = lx + lx; - iy -= 1; - } - ux.bits.manh = hx; /* The integer bit is truncated here if needed. */ - ux.bits.manl = lx; - if (iy < LDBL_MIN_EXP) { - ux.bits.exp = iy + (BIAS + 512); - ux.e *= 0x1p-512; - } else { - ux.bits.exp = iy + BIAS; - } - ux.bits.sign = 0; - x = ux.e; -fixup: - y = fabsl(y); - if (y < LDBL_MIN * 2) { - if (x + x > y || (x + x == y && (q & 1))) { + if (mx == 0) + ex = -120; + else + for (; mx >> 63 == 0; mx *= 2, ex--); + ux.i.m = mx; +#elif LDBL_MANT_DIG == 113 + uint64_t hi, lo, xhi, xlo, yhi, ylo; + xhi = (ux.i2.hi & -1ULL>>16) | 1ULL<<48; + yhi = (uy.i2.hi & -1ULL>>16) | 1ULL<<48; + xlo = ux.i2.lo; + ylo = ux.i2.lo; + for (; ex > ey; ex--) { + hi = xhi - yhi; + lo = xlo - ylo; + if (xlo < ylo) + hi -= 1; + if (hi >> 63 == 0) { + xhi = 2*hi + (lo>>63); + xlo = 2*lo; + q++; + } else { + xhi = 2*xhi + (xlo>>63); + xlo = 2*xlo; + } + q <<= 1; + } + hi = xhi - yhi; + lo = xlo - ylo; + if (xlo < ylo) + hi -= 1; + if (hi >> 63 == 0) { + xhi = hi; + xlo = lo; q++; - x-=y; } - } else if (x > 0.5*y || (x == 0.5*y && (q & 1))) { - q++; - x-=y; + if ((xhi|xlo) == 0) + ex = -120; + else + for (; xhi >> 48 == 0; xhi = 2*xhi + (xlo>>63), xlo = 2*xlo, ex--); + ux.i2.hi = xhi; + ux.i2.lo = xlo; +#endif } - ux.e = x; - ux.bits.sign ^= sx; - x = ux.e; - + /* scale result and decide between |x| and |x|-|y| */ + if (ex <= 0) { + ux.i.se = ex + 120; + ux.f *= 0x1p-120f; + } else + ux.i.se = ex; + x = ux.f; + if (sy) + y = -y; + if (ex == ey || (ex+1 == ey && (2*x > y || (2*x == y && q%2)))) { + x -= y; + q++; + } q &= 0x7fffffff; - *quo = sxy ? -q : q; - return x; + *quo = sx^sy ? -(int)q : (int)q; + return sx ? -x : x; } #endif -- cgit v1.2.1