| 1 | /* $Id: extF80_scale_extF80.c 98103 2023-01-17 14:15:46Z vboxsync $ */
|
|---|
| 2 | /** @file
|
|---|
| 3 | * SoftFloat - 387-style fscale.
|
|---|
| 4 | */
|
|---|
| 5 |
|
|---|
| 6 | /*
|
|---|
| 7 | * Copyright (C) 2022-2023 Oracle and/or its affiliates.
|
|---|
| 8 | *
|
|---|
| 9 | * This file is part of VirtualBox base platform packages, as
|
|---|
| 10 | * available from https://www.virtualbox.org.
|
|---|
| 11 | *
|
|---|
| 12 | * This program is free software; you can redistribute it and/or
|
|---|
| 13 | * modify it under the terms of the GNU General Public License
|
|---|
| 14 | * as published by the Free Software Foundation, in version 3 of the
|
|---|
| 15 | * License.
|
|---|
| 16 | *
|
|---|
| 17 | * This program is distributed in the hope that it will be useful, but
|
|---|
| 18 | * WITHOUT ANY WARRANTY; without even the implied warranty of
|
|---|
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|---|
| 20 | * General Public License for more details.
|
|---|
| 21 | *
|
|---|
| 22 | * You should have received a copy of the GNU General Public License
|
|---|
| 23 | * along with this program; if not, see <https://www.gnu.org/licenses>.
|
|---|
| 24 | *
|
|---|
| 25 | * SPDX-License-Identifier: GPL-3.0-only
|
|---|
| 26 | */
|
|---|
| 27 |
|
|---|
| 28 |
|
|---|
| 29 | /*********************************************************************************************************************************
|
|---|
| 30 | * Header Files *
|
|---|
| 31 | *********************************************************************************************************************************/
|
|---|
| 32 | #include "platform.h"
|
|---|
| 33 | #include "internals.h"
|
|---|
| 34 | #include "specialize.h"
|
|---|
| 35 | #include "softfloat.h"
|
|---|
| 36 | #include <iprt/assert.h>
|
|---|
| 37 | #include <iprt/asm.h>
|
|---|
| 38 |
|
|---|
| 39 |
|
|---|
| 40 | /**
|
|---|
| 41 | * Wrapper around softfloat_propagateNaNExtF80UI that returns extFloat80_t
|
|---|
| 42 | * and takes fully unpacked input.
|
|---|
| 43 | */
|
|---|
| 44 | DECLINLINE(extFloat80_t)
|
|---|
| 45 | softfloat_extF80_propagateNaN(bool fSignA, int_fast32_t iExpA, uint_fast64_t uSigA,
|
|---|
| 46 | bool fSignB, int_fast32_t iExpB, uint_fast64_t uSigB SOFTFLOAT_STATE_DECL_COMMA)
|
|---|
| 47 | {
|
|---|
| 48 | struct uint128 uiZ = softfloat_propagateNaNExtF80UI(packToExtF80UI64(fSignA, iExpA), uSigA,
|
|---|
| 49 | packToExtF80UI64(fSignB, iExpB), uSigB SOFTFLOAT_STATE_ARG_COMMA);
|
|---|
| 50 | union extF80M_extF80 Ret;
|
|---|
| 51 | Ret.fM.signExp = uiZ.v64;
|
|---|
| 52 | Ret.fM.signif = uiZ.v0;
|
|---|
| 53 | return Ret.f;
|
|---|
| 54 | }
|
|---|
| 55 |
|
|---|
| 56 |
|
|---|
| 57 | /**
|
|---|
| 58 | * This performs a function similar to extF80_to_i32_r_minMag, but returns
|
|---|
| 59 | * proper MIN/MAX values and no NaNs.
|
|---|
| 60 | *
|
|---|
| 61 | * ASSUMES the input is normalized.
|
|---|
| 62 | *
|
|---|
| 63 | * @returns Values in the range -2^24...+2^24.
|
|---|
| 64 | */
|
|---|
| 65 | static int_fast32_t convertToInt(bool fSign, uint_fast64_t uSig, int_fast32_t iExp)
|
|---|
| 66 | {
|
|---|
| 67 | iExp -= RTFLOAT80U_EXP_BIAS;
|
|---|
| 68 | if (iExp < 0)
|
|---|
| 69 | return 0;
|
|---|
| 70 |
|
|---|
| 71 | /* Restrict the range to -2^24...+2^24 to prevent overflows during scaling. */
|
|---|
| 72 | if (iExp >= 24)
|
|---|
| 73 | return fSign ? INT32_MIN / 128 : INT32_MAX / 128;
|
|---|
| 74 |
|
|---|
| 75 | int_fast32_t iRet = (int_fast32_t)(uSig >> (63 - iExp));
|
|---|
| 76 | if (fSign)
|
|---|
| 77 | return -iRet;
|
|---|
| 78 | return iRet;
|
|---|
| 79 | }
|
|---|
| 80 |
|
|---|
| 81 |
|
|---|
| 82 | /**
|
|---|
| 83 | * VBox: scale @a a by 2^truncateToInt(@a b)
|
|---|
| 84 | *
|
|---|
| 85 | * This function accepts and deals correctly with denormals (pseudo and
|
|---|
| 86 | * otherwise).
|
|---|
| 87 | */
|
|---|
| 88 | extFloat80_t extF80_scale_extF80(extFloat80_t a, extFloat80_t b, softfloat_state_t *pState)
|
|---|
| 89 | {
|
|---|
| 90 | static union extF80M_extF80 const s_extF80Indefinite = EXTF80M_EXTF80_INIT(defaultNaNExtF80UI64, defaultNaNExtF80UI0);
|
|---|
| 91 | static union extF80M_extF80 const s_aExtF80Zero[2] =
|
|---|
| 92 | {
|
|---|
| 93 | EXTF80M_EXTF80_INIT3_C(0, 0, 0), EXTF80M_EXTF80_INIT3_C(1, 0, 0),
|
|---|
| 94 | };
|
|---|
| 95 | static union extF80M_extF80 const s_aExtF80Infinity[2] =
|
|---|
| 96 | {
|
|---|
| 97 | EXTF80M_EXTF80_INIT3(0, RT_BIT_64(63), RTFLOAT80U_EXP_MAX),
|
|---|
| 98 | EXTF80M_EXTF80_INIT3(1, RT_BIT_64(63), RTFLOAT80U_EXP_MAX),
|
|---|
| 99 | };
|
|---|
| 100 |
|
|---|
| 101 | /*
|
|---|
| 102 | * Unpack the input.
|
|---|
| 103 | */
|
|---|
| 104 | bool const fSignA = signExtF80UI64(a.signExp);
|
|---|
| 105 | int_fast32_t iExpA = expExtF80UI64(a.signExp);
|
|---|
| 106 | uint_fast64_t uSigA = a.signif;
|
|---|
| 107 |
|
|---|
| 108 | bool const fSignB = signExtF80UI64(b.signExp);
|
|---|
| 109 | int_fast32_t iExpB = expExtF80UI64(b.signExp);
|
|---|
| 110 | uint_fast64_t uSigB = b.signif;
|
|---|
| 111 |
|
|---|
| 112 | /*
|
|---|
| 113 | * Deal with funny input.
|
|---|
| 114 | */
|
|---|
| 115 | /* Invalid first. We ASSUME subnormals are rejected here. */
|
|---|
| 116 | if ( RTFLOAT80U_IS_387_INVALID_EX(uSigA, iExpA)
|
|---|
| 117 | || RTFLOAT80U_IS_387_INVALID_EX(uSigB, iExpB))
|
|---|
| 118 | {
|
|---|
| 119 | softfloat_raiseFlags(softfloat_flag_invalid, pState);
|
|---|
| 120 | return s_extF80Indefinite.f;
|
|---|
| 121 | }
|
|---|
| 122 |
|
|---|
| 123 | /* Then NaNs and indefinites (special NaNs): */
|
|---|
| 124 | if ( RTFLOAT80U_IS_INDEFINITE_OR_QUIET_OR_SIGNALLING_NAN_EX(uSigA, iExpA)
|
|---|
| 125 | || RTFLOAT80U_IS_INDEFINITE_OR_QUIET_OR_SIGNALLING_NAN_EX(uSigB, iExpB))
|
|---|
| 126 | return softfloat_extF80_propagateNaN(fSignA, iExpA, uSigA, fSignB, iExpB, uSigB, pState);
|
|---|
| 127 |
|
|---|
| 128 | /* Normalize denormal inputs: */
|
|---|
| 129 | if (RTFLOAT80U_IS_DENORMAL_OR_PSEUDO_DENORMAL_EX(uSigA, iExpA))
|
|---|
| 130 | {
|
|---|
| 131 | softfloat_raiseFlags(softfloat_flag_denormal, pState);
|
|---|
| 132 | if (uSigA & RT_BIT_64(63))
|
|---|
| 133 | iExpA = 1; /* -16382 */
|
|---|
| 134 | else
|
|---|
| 135 | {
|
|---|
| 136 | /* We must return the denormal a value unchanged when b=zero, intel 10980XE
|
|---|
| 137 | does this at least. Where-as pseudo-denormals are normalized. Go figure. */
|
|---|
| 138 | if (RTFLOAT80U_IS_ZERO_EX(uSigB, iExpB))
|
|---|
| 139 | return a;
|
|---|
| 140 | iExpA = 64 - ASMBitLastSetU64(uSigA);
|
|---|
| 141 | uSigA <<= iExpA;
|
|---|
| 142 | iExpA = 1 - iExpA; /* -16382 - shift */
|
|---|
| 143 | }
|
|---|
| 144 | }
|
|---|
| 145 |
|
|---|
| 146 | if (RTFLOAT80U_IS_DENORMAL_OR_PSEUDO_DENORMAL_EX(uSigB, iExpB))
|
|---|
| 147 | {
|
|---|
| 148 | softfloat_raiseFlags(softfloat_flag_denormal, pState);
|
|---|
| 149 | if (uSigB & RT_BIT_64(63))
|
|---|
| 150 | iExpB = 1; /* -16382 */
|
|---|
| 151 | else
|
|---|
| 152 | {
|
|---|
| 153 | iExpB = 64 - ASMBitLastSetU64(uSigB);
|
|---|
| 154 | uSigB <<= iExpB;
|
|---|
| 155 | iExpB = 1 - iExpB; /* -16382 - shift */
|
|---|
| 156 | }
|
|---|
| 157 | }
|
|---|
| 158 |
|
|---|
| 159 | /* Infinities and zeros: If a is Zero or Infinity, return it as-is unless
|
|---|
| 160 | b=-Infinity & a=+/-Infinity or b=+Infinity & a=+/-zero when we have to
|
|---|
| 161 | raise #I and return indefinite instead.
|
|---|
| 162 |
|
|---|
| 163 | Note! If b is zero, don't, because pseudo-denormals should be returned
|
|---|
| 164 | normalized (intel does that at least). Excpetion is b=zero and
|
|---|
| 165 | a=denormal, which is handled above. */
|
|---|
| 166 | if ( RTFLOAT80U_IS_INF_EX(uSigA, iExpA)
|
|---|
| 167 | || RTFLOAT80U_IS_ZERO_EX(uSigA, iExpA))
|
|---|
| 168 | {
|
|---|
| 169 | if (RTFLOAT80U_IS_INF_EX(uSigB, iExpB) && fSignB == RTFLOAT80U_IS_INF_EX(uSigA, iExpA))
|
|---|
| 170 | {
|
|---|
| 171 | softfloat_raiseFlags(softfloat_flag_invalid, pState);
|
|---|
| 172 | return s_extF80Indefinite.f;
|
|---|
| 173 | }
|
|---|
| 174 | return a;
|
|---|
| 175 | }
|
|---|
| 176 |
|
|---|
| 177 | if (RTFLOAT80U_IS_INF_EX(uSigB, iExpB))
|
|---|
| 178 | {
|
|---|
| 179 | if (fSignB)
|
|---|
| 180 | return s_aExtF80Zero[fSignA].f;
|
|---|
| 181 | return s_aExtF80Infinity[fSignA].f;
|
|---|
| 182 | }
|
|---|
| 183 |
|
|---|
| 184 | /*
|
|---|
| 185 | * Convert b to an integer and do the scaling.
|
|---|
| 186 | */
|
|---|
| 187 | int_fast32_t iScaleFactor = convertToInt(fSignB, uSigB, iExpB);
|
|---|
| 188 | int_fast32_t iScaledExp = iExpA + iScaleFactor;
|
|---|
| 189 | return softfloat_normRoundPackToExtF80(fSignA, iScaledExp, uSigA, 0 /*sigExtra*/, 80 /*precision*/, pState);
|
|---|
| 190 | }
|
|---|
| 191 |
|
|---|
