/* ** libgcc support for software floating point. ** Copyright (C) 1991 by Pipeline Associates, Inc. All rights reserved. ** Permission is granted to do *anything* you want with this file, ** commercial or otherwise, provided this message remains intact. So there! ** I would appreciate receiving any updates/patches/changes that anyone ** makes, and am willing to be the repository for said changes (am I ** making a big mistake?). Warning! Only single-precision is actually implemented. This file won't really be much use until double-precision is supported. However, once that is done, this file might eventually become a replacement for libgcc1.c. It might also make possible cross-compilation for an IEEE target machine from a non-IEEE host such as a VAX. If you'd like to work on completing this, please talk to rms@gnu.ai.mit.edu. ** ** Pat Wood ** Pipeline Associates, Inc. ** pipeline!phw@motown.com or ** sun!pipeline!phw or ** uunet!motown!pipeline!phw ** ** 05/01/91 -- V1.0 -- first release to gcc mailing lists ** 05/04/91 -- V1.1 -- added float and double prototypes and return values ** -- fixed problems with adding and subtracting zero ** -- fixed rounding in truncdfsf2 ** -- fixed SWAP define and tested on 386 */ /* ** The following are routines that replace the libgcc soft floating point ** routines that are called automatically when -msoft-float is selected. ** The support single and double precision IEEE format, with provisions ** for byte-swapped machines (tested on 386). Some of the double-precision ** routines work at full precision, but most of the hard ones simply punt ** and call the single precision routines, producing a loss of accuracy. ** long long support is not assumed or included. ** Overall accuracy is close to IEEE (actually 68882) for single-precision ** arithmetic. I think there may still be a 1 in 1000 chance of a bit ** being rounded the wrong way during a multiply. I'm not fussy enough to ** bother with it, but if anyone is, knock yourself out. ** ** Efficiency has only been addressed where it was obvious that something ** would make a big difference. Anyone who wants to do this right for ** best speed should go in and rewrite in assembler. ** ** I have tested this only on a 68030 workstation and 386/ix integrated ** in with -msoft-float. */ /* the following deal with IEEE single-precision numbers */ #define EXCESS 126 #define SIGNBIT 0x80000000 #define HIDDEN (1 << 23) #define SIGN(fp) ((fp) & SIGNBIT) #define EXP(fp) (((fp) >> 23) & 0xFF) #define MANT(fp) (((fp) & 0x7FFFFF) | HIDDEN) #define PACK(s,e,m) ((s) | ((e) << 23) | (m)) /* the following deal with IEEE double-precision numbers */ #define EXCESSD 1022 #define HIDDEND (1 << 20) #define EXPD(fp) (((fp.l.upper) >> 20) & 0x7FF) #define SIGND(fp) ((fp.l.upper) & SIGNBIT) #define MANTD(fp) (((((fp.l.upper) & 0xFFFFF) | HIDDEND) << 10) | \ (fp.l.lower >> 22)) /* define SWAP for 386/960 reverse-byte-order brain-damaged CPUs */ union double_long { double d; #ifdef SWAP struct { unsigned long lower; long upper; } l; #else struct { long upper; unsigned long lower; } l; #endif }; union float_long { float f; long l; }; /* add two floats */ float __addsf3 (float a1, float a2) { register long mant1, mant2; register union float_long fl1, fl2; register int exp1, exp2; int sign = 0; fl1.f = a1; fl2.f = a2; /* check for zero args */ if (!fl1.l) return (fl2.f); if (!fl2.l) return (fl1.f); exp1 = EXP (fl1.l); exp2 = EXP (fl2.l); if (exp1 > exp2 + 25) return (fl1.l); if (exp2 > exp1 + 25) return (fl2.l); /* do everything in excess precision so's we can round later */ mant1 = MANT (fl1.l) << 6; mant2 = MANT (fl2.l) << 6; if (SIGN (fl1.l)) mant1 = -mant1; if (SIGN (fl2.l)) mant2 = -mant2; if (exp1 > exp2) { mant2 >>= exp1 - exp2; } else { mant1 >>= exp2 - exp1; exp1 = exp2; } mant1 += mant2; if (mant1 < 0) { mant1 = -mant1; sign = SIGNBIT; } else if (!mant1) return (0); /* normalize up */ while (!(mant1 & 0xE0000000)) { mant1 <<= 1; exp1--; } /* normalize down? */ if (mant1 & (1 << 30)) { mant1 >>= 1; exp1++; } /* round to even */ mant1 += (mant1 & 0x40) ? 0x20 : 0x1F; /* normalize down? */ if (mant1 & (1 << 30)) { mant1 >>= 1; exp1++; } /* lose extra precision */ mant1 >>= 6; /* turn off hidden bit */ mant1 &= ~HIDDEN; /* pack up and go home */ fl1.l = PACK (sign, exp1, mant1); return (fl1.f); } /* subtract two floats */ float __subsf3 (float a1, float a2) { register union float_long fl1, fl2; fl1.f = a1; fl2.f = a2; /* check for zero args */ if (!fl2.l) return (fl1.f); if (!fl1.l) return (-fl2.f); /* twiddle sign bit and add */ fl2.l ^= SIGNBIT; return __addsf3 (a1, fl2.f); } /* compare two floats */ long __cmpsf2 (float a1, float a2) { register union float_long fl1, fl2; fl1.f = a1; fl2.f = a2; if (SIGN (fl1.l) && SIGN (fl2.l)) { fl1.l ^= SIGNBIT; fl2.l ^= SIGNBIT; } if (fl1.l < fl2.l) return (-1); if (fl1.l > fl2.l) return (1); return (0); } /* multiply two floats */ float __mulsf3 (float a1, float a2) { register union float_long fl1, fl2; register unsigned long result; register int exp; int sign; fl1.f = a1; fl2.f = a2; if (!fl1.l || !fl2.l) return (0); /* compute sign and exponent */ sign = SIGN (fl1.l) ^ SIGN (fl2.l); exp = EXP (fl1.l) - EXCESS; exp += EXP (fl2.l); fl1.l = MANT (fl1.l); fl2.l = MANT (fl2.l); /* the multiply is done as one 16x16 multiply and two 16x8 multiples */ result = (fl1.l >> 8) * (fl2.l >> 8); result += ((fl1.l & 0xFF) * (fl2.l >> 8)) >> 8; result += ((fl2.l & 0xFF) * (fl1.l >> 8)) >> 8; if (result & 0x80000000) { /* round */ result += 0x80; result >>= 8; } else { /* round */ result += 0x40; result >>= 7; exp--; } result &= ~HIDDEN; /* pack up and go home */ fl1.l = PACK (sign, exp, result); return (fl1.f); } /* divide two floats */ float __divsf3 (float a1, float a2) { register union float_long fl1, fl2; register int result; register int mask; register int exp, sign; fl1.f = a1; fl2.f = a2; /* subtract exponents */ exp = EXP (fl1.l) - EXP (fl2.l) + EXCESS; /* compute sign */ sign = SIGN (fl1.l) ^ SIGN (fl2.l); /* divide by zero??? */ if (!fl2.l) /* return NaN or -NaN */ return (sign ? 0xFFFFFFFF : 0x7FFFFFFF); /* numerator zero??? */ if (!fl1.l) return (0); /* now get mantissas */ fl1.l = MANT (fl1.l); fl2.l = MANT (fl2.l); /* this assures we have 25 bits of precision in the end */ if (fl1.l < fl2.l) { fl1.l <<= 1; exp--; } /* now we perform repeated subtraction of fl2.l from fl1.l */ mask = 0x1000000; result = 0; while (mask) { if (fl1.l >= fl2.l) { result |= mask; fl1.l -= fl2.l; } fl1.l <<= 1; mask >>= 1; } /* round */ result += 1; /* normalize down */ exp++; result >>= 1; result &= ~HIDDEN; /* pack up and go home */ fl1.l = PACK (sign, exp, result); return (fl1.f); } /* convert int to double */ double __floatsidf (register long a1) { register int sign = 0, exp = 31 + EXCESSD; union double_long dl; if (!a1) { dl.l.upper = dl.l.lower = 0; return (dl.d); } if (a1 < 0) { sign = SIGNBIT; a1 = -a1; } while (a1 < 0x1000000) { a1 <<= 4; exp -= 4; } while (a1 < 0x40000000) { a1 <<= 1; exp--; } /* pack up and go home */ dl.l.upper = sign; dl.l.upper |= exp << 20; dl.l.upper |= (a1 >> 10) & ~HIDDEND; dl.l.lower = a1 << 22; return (dl.d); } /* negate a float */ float __negsf2 (float a1) { register union float_long fl1; fl1.f = a1; if (!fl1.l) return (0); fl1.l ^= SIGNBIT; return (fl1.f); } /* negate a double */ double __negdf2 (double a1) { register union double_long dl1; dl1.d = a1; if (!dl1.l.upper && !dl1.l.lower) return (dl1.d); dl1.l.upper ^= SIGNBIT; return (dl1.d); } /* convert float to double */ double __extendsfdf2 (float a1) { register union float_long fl1; register union double_long dl; register int exp; fl1.f = a1; if (!fl1.l) { dl.l.upper = dl.l.lower = 0; return (dl.d); } dl.l.upper = SIGN (fl1.l); exp = EXP (fl1.l) - EXCESS + EXCESSD; dl.l.upper |= exp << 20; dl.l.upper |= (MANT (fl1.l) & ~HIDDEN) >> 3; dl.l.lower = MANT (fl1.l) << 29; return (dl.d); } /* convert double to float */ float __truncdfsf2 (double a1) { register int exp; register long mant; register union float_long fl; register union double_long dl1; dl1.d = a1; if (!dl1.l.upper && !dl1.l.lower) return (0); exp = EXPD (dl1) - EXCESSD + EXCESS; /* shift double mantissa 6 bits so we can round */ mant = MANTD (dl1) >> 6; /* now round and shift down */ mant += 1; mant >>= 1; /* did the round overflow? */ if (mant & 0xFF000000) { mant >>= 1; exp++; } mant &= ~HIDDEN; /* pack up and go home */ fl.l = PACK (SIGND (dl1), exp, mant); return (fl.f); } /* compare two doubles */ long __cmpdf2 (double a1, double a2) { register union double_long dl1, dl2; dl1.d = a1; dl2.d = a2; if (SIGND (dl1) && SIGND (dl2)) { dl1.l.upper ^= SIGNBIT; dl2.l.upper ^= SIGNBIT; } if (dl1.l.upper < dl2.l.upper) return (-1); if (dl1.l.upper > dl2.l.upper) return (1); if (dl1.l.lower < dl2.l.lower) return (-1); if (dl1.l.lower > dl2.l.lower) return (1); return (0); } /* convert double to int */ long __fixdfsi (double a1) { register union double_long dl1; register int exp; register long l; dl1.d = a1; if (!dl1.l.upper && !dl1.l.lower) return (0); exp = EXPD (dl1) - EXCESSD - 31; l = MANTD (dl1); if (exp > 0) return (0x7FFFFFFF | SIGND (dl1)); /* largest integer */ /* shift down until exp = 0 or l = 0 */ if (exp < 0 && exp > -32 && l) l >>= -exp; else return (0); return (SIGND (dl1) ? -l : l); } /* convert double to unsigned int */ unsigned long __fixunsdfsi (double a1) { register union double_long dl1; register int exp; register unsigned long l; dl1.d = a1; if (!dl1.l.upper && !dl1.l.lower) return (0); exp = EXPD (dl1) - EXCESSD - 32; l = (((((dl1.l.upper) & 0xFFFFF) | HIDDEND) << 11) | (dl1.l.lower >> 21)); if (exp > 0) return (0xFFFFFFFF); /* largest integer */ /* shift down until exp = 0 or l = 0 */ if (exp < 0 && exp > -32 && l) l >>= -exp; else return (0); return (l); } /* For now, the hard double-precision routines simply punt and do it in single */ /* addtwo doubles */ double __adddf3 (double a1, double a2) { return ((float) a1 + (float) a2); } /* subtract two doubles */ double __subdf3 (double a1, double a2) { return ((float) a1 - (float) a2); } /* multiply two doubles */ double __muldf3 (double a1, double a2) { return ((float) a1 * (float) a2); } /* divide two doubles */ double __divdf3 (double a1, double a2) { return ((float) a1 / (float) a2); }