arch/mips/math-emu/dp_mul.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* IEEE754 floating point arithmetic
   3  * double precision: common utilities
   4  */
   5 /*
   6  * MIPS floating point support
   7  * Copyright (C) 1994-2000 Algorithmics Ltd.
   8  */
   9
  10 #include "ieee754dp.h"
  11
  12 union ieee754dp ieee754dp_mul(union ieee754dp x, union ieee754dp y)
  13 {
  14         int re;
  15         int rs;
  16         u64 rm;
  17         unsigned int lxm;
  18         unsigned int hxm;
  19         unsigned int lym;
  20         unsigned int hym;
  21         u64 lrm;
  22         u64 hrm;
  23         u64 t;
  24         u64 at;
  25
  26         COMPXDP;
  27         COMPYDP;
  28
  29         EXPLODEXDP;
  30         EXPLODEYDP;
  31
  32         ieee754_clearcx();
  33
  34         FLUSHXDP;
  35         FLUSHYDP;
  36
  37         switch (CLPAIR(xc, yc)) {
  38         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_SNAN):
  39         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_SNAN):
  40         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_SNAN):
  41         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_SNAN):
  42         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_SNAN):
  43                 return ieee754dp_nanxcpt(y);
  44
  45         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_SNAN):
  46         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_QNAN):
  47         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_ZERO):
  48         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_NORM):
  49         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_DNORM):
  50         case CLPAIR(IEEE754_CLASS_SNAN, IEEE754_CLASS_INF):
  51                 return ieee754dp_nanxcpt(x);
  52
  53         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_QNAN):
  54         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_QNAN):
  55         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_QNAN):
  56         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_QNAN):
  57                 return y;
  58
  59         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_QNAN):
  60         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_ZERO):
  61         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_NORM):
  62         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_DNORM):
  63         case CLPAIR(IEEE754_CLASS_QNAN, IEEE754_CLASS_INF):
  64                 return x;
  65
  66
  67         /*
  68          * Infinity handling
  69          */
  70         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_ZERO):
  71         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_INF):
  72                 ieee754_setcx(IEEE754_INVALID_OPERATION);
  73                 return ieee754dp_indef();
  74
  75         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_INF):
  76         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_INF):
  77         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_NORM):
  78         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_DNORM):
  79         case CLPAIR(IEEE754_CLASS_INF, IEEE754_CLASS_INF):
  80                 return ieee754dp_inf(xs ^ ys);
  81
  82         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_ZERO):
  83         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_NORM):
  84         case CLPAIR(IEEE754_CLASS_ZERO, IEEE754_CLASS_DNORM):
  85         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_ZERO):
  86         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_ZERO):
  87                 return ieee754dp_zero(xs ^ ys);
  88
  89
  90         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_DNORM):
  91                 DPDNORMX;
  92                 fallthrough;
  93         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_DNORM):
  94                 DPDNORMY;
  95                 break;
  96
  97         case CLPAIR(IEEE754_CLASS_DNORM, IEEE754_CLASS_NORM):
  98                 DPDNORMX;
  99                 break;
 100
 101         case CLPAIR(IEEE754_CLASS_NORM, IEEE754_CLASS_NORM):
 102                 break;
 103         }
 104         /* rm = xm * ym, re = xe+ye basically */
 105         assert(xm & DP_HIDDEN_BIT);
 106         assert(ym & DP_HIDDEN_BIT);
 107
 108         re = xe + ye;
 109         rs = xs ^ ys;
 110
 111         /* shunt to top of word */
 112         xm <<= 64 - (DP_FBITS + 1);
 113         ym <<= 64 - (DP_FBITS + 1);
 114
 115         /*
 116          * Multiply 64 bits xm, ym to give high 64 bits rm with stickness.
 117          */
 118
 119         lxm = xm;
 120         hxm = xm >> 32;
 121         lym = ym;
 122         hym = ym >> 32;
 123
 124         lrm = DPXMULT(lxm, lym);
 125         hrm = DPXMULT(hxm, hym);
 126
 127         t = DPXMULT(lxm, hym);
 128
 129         at = lrm + (t << 32);
 130         hrm += at < lrm;
 131         lrm = at;
 132
 133         hrm = hrm + (t >> 32);
 134
 135         t = DPXMULT(hxm, lym);
 136
 137         at = lrm + (t << 32);
 138         hrm += at < lrm;
 139         lrm = at;
 140
 141         hrm = hrm + (t >> 32);
 142
 143         rm = hrm | (lrm != 0);
 144
 145         /*
 146          * Sticky shift down to normal rounding precision.
 147          */
 148         if ((s64) rm < 0) {
 149                 rm = (rm >> (64 - (DP_FBITS + 1 + 3))) |
 150                      ((rm << (DP_FBITS + 1 + 3)) != 0);
 151                 re++;
 152         } else {
 153                 rm = (rm >> (64 - (DP_FBITS + 1 + 3 + 1))) |
 154                      ((rm << (DP_FBITS + 1 + 3 + 1)) != 0);
 155         }
 156         assert(rm & (DP_HIDDEN_BIT << 3));
 157
 158         return ieee754dp_format(rs, re, rm);
 159 }