linux/arch/x86/math-emu/poly_l2.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*---------------------------------------------------------------------------+
   3 |  poly_l2.c                                                                |
   4 |                                                                           |
   5 | Compute the base 2 log of a FPU_REG, using a polynomial approximation.    |
   6 |                                                                           |
   7 | Copyright (C) 1992,1993,1994,1997                                         |
   8 |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
   9 |                  E-mail   billm@suburbia.net                              |
  10 |                                                                           |
  11 |                                                                           |
  12 +---------------------------------------------------------------------------*/
  13
  14#include "exception.h"
  15#include "reg_constant.h"
  16#include "fpu_emu.h"
  17#include "fpu_system.h"
  18#include "control_w.h"
  19#include "poly.h"
  20
  21static void log2_kernel(FPU_REG const *arg, u_char argsign,
  22                        Xsig * accum_result, long int *expon);
  23
  24/*--- poly_l2() -------------------------------------------------------------+
  25 |   Base 2 logarithm by a polynomial approximation.                         |
  26 +---------------------------------------------------------------------------*/
  27void poly_l2(FPU_REG *st0_ptr, FPU_REG *st1_ptr, u_char st1_sign)
  28{
  29        long int exponent, expon, expon_expon;
  30        Xsig accumulator, expon_accum, yaccum;
  31        u_char sign, argsign;
  32        FPU_REG x;
  33        int tag;
  34
  35        exponent = exponent16(st0_ptr);
  36
  37        /* From st0_ptr, make a number > sqrt(2)/2 and < sqrt(2) */
  38        if (st0_ptr->sigh > (unsigned)0xb504f334) {
  39                /* Treat as  sqrt(2)/2 < st0_ptr < 1 */
  40                significand(&x) = -significand(st0_ptr);
  41                setexponent16(&x, -1);
  42                exponent++;
  43                argsign = SIGN_NEG;
  44        } else {
  45                /* Treat as  1 <= st0_ptr < sqrt(2) */
  46                x.sigh = st0_ptr->sigh - 0x80000000;
  47                x.sigl = st0_ptr->sigl;
  48                setexponent16(&x, 0);
  49                argsign = SIGN_POS;
  50        }
  51        tag = FPU_normalize_nuo(&x);
  52
  53        if (tag == TAG_Zero) {
  54                expon = 0;
  55                accumulator.msw = accumulator.midw = accumulator.lsw = 0;
  56        } else {
  57                log2_kernel(&x, argsign, &accumulator, &expon);
  58        }
  59
  60        if (exponent < 0) {
  61                sign = SIGN_NEG;
  62                exponent = -exponent;
  63        } else
  64                sign = SIGN_POS;
  65        expon_accum.msw = exponent;
  66        expon_accum.midw = expon_accum.lsw = 0;
  67        if (exponent) {
  68                expon_expon = 31 + norm_Xsig(&expon_accum);
  69                shr_Xsig(&accumulator, expon_expon - expon);
  70
  71                if (sign ^ argsign)
  72                        negate_Xsig(&accumulator);
  73                add_Xsig_Xsig(&accumulator, &expon_accum);
  74        } else {
  75                expon_expon = expon;
  76                sign = argsign;
  77        }
  78
  79        yaccum.lsw = 0;
  80        XSIG_LL(yaccum) = significand(st1_ptr);
  81        mul_Xsig_Xsig(&accumulator, &yaccum);
  82
  83        expon_expon += round_Xsig(&accumulator);
  84
  85        if (accumulator.msw == 0) {
  86                FPU_copy_to_reg1(&CONST_Z, TAG_Zero);
  87                return;
  88        }
  89
  90        significand(st1_ptr) = XSIG_LL(accumulator);
  91        setexponent16(st1_ptr, expon_expon + exponent16(st1_ptr) + 1);
  92
  93        tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign ^ st1_sign);
  94        FPU_settagi(1, tag);
  95
  96        set_precision_flag_up();        /* 80486 appears to always do this */
  97
  98        return;
  99
 100}
 101
 102/*--- poly_l2p1() -----------------------------------------------------------+
 103 |   Base 2 logarithm by a polynomial approximation.                         |
 104 |   log2(x+1)                                                               |
 105 +---------------------------------------------------------------------------*/
 106int poly_l2p1(u_char sign0, u_char sign1,
 107              FPU_REG * st0_ptr, FPU_REG * st1_ptr, FPU_REG * dest)
 108{
 109        u_char tag;
 110        long int exponent;
 111        Xsig accumulator, yaccum;
 112
 113        if (exponent16(st0_ptr) < 0) {
 114                log2_kernel(st0_ptr, sign0, &accumulator, &exponent);
 115
 116                yaccum.lsw = 0;
 117                XSIG_LL(yaccum) = significand(st1_ptr);
 118                mul_Xsig_Xsig(&accumulator, &yaccum);
 119
 120                exponent += round_Xsig(&accumulator);
 121
 122                exponent += exponent16(st1_ptr) + 1;
 123                if (exponent < EXP_WAY_UNDER)
 124                        exponent = EXP_WAY_UNDER;
 125
 126                significand(dest) = XSIG_LL(accumulator);
 127                setexponent16(dest, exponent);
 128
 129                tag = FPU_round(dest, 1, 0, FULL_PRECISION, sign0 ^ sign1);
 130                FPU_settagi(1, tag);
 131
 132                if (tag == TAG_Valid)
 133                        set_precision_flag_up();        /* 80486 appears to always do this */
 134        } else {
 135                /* The magnitude of st0_ptr is far too large. */
 136
 137                if (sign0 != SIGN_POS) {
 138                        /* Trying to get the log of a negative number. */
 139#ifdef PECULIAR_486             /* Stupid 80486 doesn't worry about log(negative). */
 140                        changesign(st1_ptr);
 141#else
 142                        if (arith_invalid(1) < 0)
 143                                return 1;
 144#endif /* PECULIAR_486 */
 145                }
 146
 147                /* 80486 appears to do this */
 148                if (sign0 == SIGN_NEG)
 149                        set_precision_flag_down();
 150                else
 151                        set_precision_flag_up();
 152        }
 153
 154        if (exponent(dest) <= EXP_UNDER)
 155                EXCEPTION(EX_Underflow);
 156
 157        return 0;
 158
 159}
 160
 161#undef HIPOWER
 162#define HIPOWER 10
 163static const unsigned long long logterms[HIPOWER] = {
 164        0x2a8eca5705fc2ef0LL,
 165        0xf6384ee1d01febceLL,
 166        0x093bb62877cdf642LL,
 167        0x006985d8a9ec439bLL,
 168        0x0005212c4f55a9c8LL,
 169        0x00004326a16927f0LL,
 170        0x0000038d1d80a0e7LL,
 171        0x0000003141cc80c6LL,
 172        0x00000002b1668c9fLL,
 173        0x000000002c7a46aaLL
 174};
 175
 176static const unsigned long leadterm = 0xb8000000;
 177
 178/*--- log2_kernel() ---------------------------------------------------------+
 179 |   Base 2 logarithm by a polynomial approximation.                         |
 180 |   log2(x+1)                                                               |
 181 +---------------------------------------------------------------------------*/
 182static void log2_kernel(FPU_REG const *arg, u_char argsign, Xsig *accum_result,
 183                        long int *expon)
 184{
 185        long int exponent, adj;
 186        unsigned long long Xsq;
 187        Xsig accumulator, Numer, Denom, argSignif, arg_signif;
 188
 189        exponent = exponent16(arg);
 190        Numer.lsw = Denom.lsw = 0;
 191        XSIG_LL(Numer) = XSIG_LL(Denom) = significand(arg);
 192        if (argsign == SIGN_POS) {
 193                shr_Xsig(&Denom, 2 - (1 + exponent));
 194                Denom.msw |= 0x80000000;
 195                div_Xsig(&Numer, &Denom, &argSignif);
 196        } else {
 197                shr_Xsig(&Denom, 1 - (1 + exponent));
 198                negate_Xsig(&Denom);
 199                if (Denom.msw & 0x80000000) {
 200                        div_Xsig(&Numer, &Denom, &argSignif);
 201                        exponent++;
 202                } else {
 203                        /* Denom must be 1.0 */
 204                        argSignif.lsw = Numer.lsw;
 205                        argSignif.midw = Numer.midw;
 206                        argSignif.msw = Numer.msw;
 207                }
 208        }
 209
 210#ifndef PECULIAR_486
 211        /* Should check here that  |local_arg|  is within the valid range */
 212        if (exponent >= -2) {
 213                if ((exponent > -2) || (argSignif.msw > (unsigned)0xafb0ccc0)) {
 214                        /* The argument is too large */
 215                }
 216        }
 217#endif /* PECULIAR_486 */
 218
 219        arg_signif.lsw = argSignif.lsw;
 220        XSIG_LL(arg_signif) = XSIG_LL(argSignif);
 221        adj = norm_Xsig(&argSignif);
 222        accumulator.lsw = argSignif.lsw;
 223        XSIG_LL(accumulator) = XSIG_LL(argSignif);
 224        mul_Xsig_Xsig(&accumulator, &accumulator);
 225        shr_Xsig(&accumulator, 2 * (-1 - (1 + exponent + adj)));
 226        Xsq = XSIG_LL(accumulator);
 227        if (accumulator.lsw & 0x80000000)
 228                Xsq++;
 229
 230        accumulator.msw = accumulator.midw = accumulator.lsw = 0;
 231        /* Do the basic fixed point polynomial evaluation */
 232        polynomial_Xsig(&accumulator, &Xsq, logterms, HIPOWER - 1);
 233
 234        mul_Xsig_Xsig(&accumulator, &argSignif);
 235        shr_Xsig(&accumulator, 6 - adj);
 236
 237        mul32_Xsig(&arg_signif, leadterm);
 238        add_two_Xsig(&accumulator, &arg_signif, &exponent);
 239
 240        *expon = exponent + 1;
 241        accum_result->lsw = accumulator.lsw;
 242        accum_result->midw = accumulator.midw;
 243        accum_result->msw = accumulator.msw;
 244
 245}
 246