Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Linux/PA-RISC Project (http://www.parisc-linux.org/) |
| 3 | * |
| 4 | * Floating-point emulation code |
| 5 | * Copyright (C) 2001 Hewlett-Packard (Paul Bame) <bame@debian.org> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License as published by |
| 9 | * the Free Software Foundation; either version 2, or (at your option) |
| 10 | * any later version. |
| 11 | * |
| 12 | * This program is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 15 | * GNU General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU General Public License |
| 18 | * along with this program; if not, write to the Free Software |
| 19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 20 | */ |
| 21 | #ifdef __NO_PA_HDRS |
| 22 | PA header file -- do not include this header file for non-PA builds. |
| 23 | #endif |
| 24 | |
Simon Arlott | 7022672 | 2007-05-11 20:42:34 +0100 | [diff] [blame] | 25 | /* 32-bit word grabbing functions */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #define Dbl_firstword(value) Dallp1(value) |
| 27 | #define Dbl_secondword(value) Dallp2(value) |
| 28 | #define Dbl_thirdword(value) dummy_location |
| 29 | #define Dbl_fourthword(value) dummy_location |
| 30 | |
| 31 | #define Dbl_sign(object) Dsign(object) |
| 32 | #define Dbl_exponent(object) Dexponent(object) |
| 33 | #define Dbl_signexponent(object) Dsignexponent(object) |
| 34 | #define Dbl_mantissap1(object) Dmantissap1(object) |
| 35 | #define Dbl_mantissap2(object) Dmantissap2(object) |
| 36 | #define Dbl_exponentmantissap1(object) Dexponentmantissap1(object) |
| 37 | #define Dbl_allp1(object) Dallp1(object) |
| 38 | #define Dbl_allp2(object) Dallp2(object) |
| 39 | |
Simon Arlott | 7022672 | 2007-05-11 20:42:34 +0100 | [diff] [blame] | 40 | /* dbl_and_signs ANDs the sign bits of each argument and puts the result |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | * into the first argument. dbl_or_signs ors those same sign bits */ |
| 42 | #define Dbl_and_signs( src1dst, src2) \ |
| 43 | Dallp1(src1dst) = (Dallp1(src2)|~((unsigned int)1<<31)) & Dallp1(src1dst) |
| 44 | #define Dbl_or_signs( src1dst, src2) \ |
| 45 | Dallp1(src1dst) = (Dallp1(src2)&((unsigned int)1<<31)) | Dallp1(src1dst) |
| 46 | |
| 47 | /* The hidden bit is always the low bit of the exponent */ |
| 48 | #define Dbl_clear_exponent_set_hidden(srcdst) Deposit_dexponent(srcdst,1) |
| 49 | #define Dbl_clear_signexponent_set_hidden(srcdst) \ |
| 50 | Deposit_dsignexponent(srcdst,1) |
| 51 | #define Dbl_clear_sign(srcdst) Dallp1(srcdst) &= ~((unsigned int)1<<31) |
| 52 | #define Dbl_clear_signexponent(srcdst) \ |
| 53 | Dallp1(srcdst) &= Dmantissap1((unsigned int)-1) |
| 54 | |
| 55 | /* Exponent field for doubles has already been cleared and may be |
| 56 | * included in the shift. Here we need to generate two double width |
| 57 | * variable shifts. The insignificant bits can be ignored. |
| 58 | * MTSAR f(varamount) |
| 59 | * VSHD srcdst.high,srcdst.low => srcdst.low |
| 60 | * VSHD 0,srcdst.high => srcdst.high |
| 61 | * This is very difficult to model with C expressions since the shift amount |
| 62 | * could exceed 32. */ |
| 63 | /* varamount must be less than 64 */ |
| 64 | #define Dbl_rightshift(srcdstA, srcdstB, varamount) \ |
| 65 | {if((varamount) >= 32) { \ |
| 66 | Dallp2(srcdstB) = Dallp1(srcdstA) >> (varamount-32); \ |
| 67 | Dallp1(srcdstA)=0; \ |
| 68 | } \ |
| 69 | else if(varamount > 0) { \ |
| 70 | Variable_shift_double(Dallp1(srcdstA), Dallp2(srcdstB), \ |
| 71 | (varamount), Dallp2(srcdstB)); \ |
| 72 | Dallp1(srcdstA) >>= varamount; \ |
| 73 | } } |
| 74 | /* varamount must be less than 64 */ |
| 75 | #define Dbl_rightshift_exponentmantissa(srcdstA, srcdstB, varamount) \ |
| 76 | {if((varamount) >= 32) { \ |
| 77 | Dallp2(srcdstB) = Dexponentmantissap1(srcdstA) >> (varamount-32); \ |
| 78 | Dallp1(srcdstA) &= ((unsigned int)1<<31); /* clear expmant field */ \ |
| 79 | } \ |
| 80 | else if(varamount > 0) { \ |
| 81 | Variable_shift_double(Dexponentmantissap1(srcdstA), Dallp2(srcdstB), \ |
| 82 | (varamount), Dallp2(srcdstB)); \ |
| 83 | Deposit_dexponentmantissap1(srcdstA, \ |
| 84 | (Dexponentmantissap1(srcdstA)>>varamount)); \ |
| 85 | } } |
| 86 | /* varamount must be less than 64 */ |
| 87 | #define Dbl_leftshift(srcdstA, srcdstB, varamount) \ |
| 88 | {if((varamount) >= 32) { \ |
| 89 | Dallp1(srcdstA) = Dallp2(srcdstB) << (varamount-32); \ |
| 90 | Dallp2(srcdstB)=0; \ |
| 91 | } \ |
| 92 | else { \ |
| 93 | if ((varamount) > 0) { \ |
| 94 | Dallp1(srcdstA) = (Dallp1(srcdstA) << (varamount)) | \ |
| 95 | (Dallp2(srcdstB) >> (32-(varamount))); \ |
| 96 | Dallp2(srcdstB) <<= varamount; \ |
| 97 | } \ |
| 98 | } } |
| 99 | #define Dbl_leftshiftby1_withextent(lefta,leftb,right,resulta,resultb) \ |
| 100 | Shiftdouble(Dallp1(lefta), Dallp2(leftb), 31, Dallp1(resulta)); \ |
| 101 | Shiftdouble(Dallp2(leftb), Extall(right), 31, Dallp2(resultb)) |
| 102 | |
| 103 | #define Dbl_rightshiftby1_withextent(leftb,right,dst) \ |
| 104 | Extall(dst) = (Dallp2(leftb) << 31) | ((unsigned int)Extall(right) >> 1) | \ |
| 105 | Extlow(right) |
| 106 | |
| 107 | #define Dbl_arithrightshiftby1(srcdstA,srcdstB) \ |
| 108 | Shiftdouble(Dallp1(srcdstA),Dallp2(srcdstB),1,Dallp2(srcdstB));\ |
| 109 | Dallp1(srcdstA) = (int)Dallp1(srcdstA) >> 1 |
| 110 | |
| 111 | /* Sign extend the sign bit with an integer destination */ |
| 112 | #define Dbl_signextendedsign(value) Dsignedsign(value) |
| 113 | |
| 114 | #define Dbl_isone_hidden(dbl_value) (Is_dhidden(dbl_value)!=0) |
| 115 | /* Singles and doubles may include the sign and exponent fields. The |
| 116 | * hidden bit and the hidden overflow must be included. */ |
| 117 | #define Dbl_increment(dbl_valueA,dbl_valueB) \ |
| 118 | if( (Dallp2(dbl_valueB) += 1) == 0 ) Dallp1(dbl_valueA) += 1 |
| 119 | #define Dbl_increment_mantissa(dbl_valueA,dbl_valueB) \ |
| 120 | if( (Dmantissap2(dbl_valueB) += 1) == 0 ) \ |
| 121 | Deposit_dmantissap1(dbl_valueA,dbl_valueA+1) |
| 122 | #define Dbl_decrement(dbl_valueA,dbl_valueB) \ |
| 123 | if( Dallp2(dbl_valueB) == 0 ) Dallp1(dbl_valueA) -= 1; \ |
| 124 | Dallp2(dbl_valueB) -= 1 |
| 125 | |
| 126 | #define Dbl_isone_sign(dbl_value) (Is_dsign(dbl_value)!=0) |
| 127 | #define Dbl_isone_hiddenoverflow(dbl_value) (Is_dhiddenoverflow(dbl_value)!=0) |
| 128 | #define Dbl_isone_lowmantissap1(dbl_valueA) (Is_dlowp1(dbl_valueA)!=0) |
| 129 | #define Dbl_isone_lowmantissap2(dbl_valueB) (Is_dlowp2(dbl_valueB)!=0) |
| 130 | #define Dbl_isone_signaling(dbl_value) (Is_dsignaling(dbl_value)!=0) |
| 131 | #define Dbl_is_signalingnan(dbl_value) (Dsignalingnan(dbl_value)==0xfff) |
| 132 | #define Dbl_isnotzero(dbl_valueA,dbl_valueB) \ |
| 133 | (Dallp1(dbl_valueA) || Dallp2(dbl_valueB)) |
| 134 | #define Dbl_isnotzero_hiddenhigh7mantissa(dbl_value) \ |
| 135 | (Dhiddenhigh7mantissa(dbl_value)!=0) |
| 136 | #define Dbl_isnotzero_exponent(dbl_value) (Dexponent(dbl_value)!=0) |
| 137 | #define Dbl_isnotzero_mantissa(dbl_valueA,dbl_valueB) \ |
| 138 | (Dmantissap1(dbl_valueA) || Dmantissap2(dbl_valueB)) |
| 139 | #define Dbl_isnotzero_mantissap1(dbl_valueA) (Dmantissap1(dbl_valueA)!=0) |
| 140 | #define Dbl_isnotzero_mantissap2(dbl_valueB) (Dmantissap2(dbl_valueB)!=0) |
| 141 | #define Dbl_isnotzero_exponentmantissa(dbl_valueA,dbl_valueB) \ |
| 142 | (Dexponentmantissap1(dbl_valueA) || Dmantissap2(dbl_valueB)) |
| 143 | #define Dbl_isnotzero_low4p2(dbl_value) (Dlow4p2(dbl_value)!=0) |
| 144 | #define Dbl_iszero(dbl_valueA,dbl_valueB) (Dallp1(dbl_valueA)==0 && \ |
| 145 | Dallp2(dbl_valueB)==0) |
| 146 | #define Dbl_iszero_allp1(dbl_value) (Dallp1(dbl_value)==0) |
| 147 | #define Dbl_iszero_allp2(dbl_value) (Dallp2(dbl_value)==0) |
| 148 | #define Dbl_iszero_hidden(dbl_value) (Is_dhidden(dbl_value)==0) |
| 149 | #define Dbl_iszero_hiddenoverflow(dbl_value) (Is_dhiddenoverflow(dbl_value)==0) |
| 150 | #define Dbl_iszero_hiddenhigh3mantissa(dbl_value) \ |
| 151 | (Dhiddenhigh3mantissa(dbl_value)==0) |
| 152 | #define Dbl_iszero_hiddenhigh7mantissa(dbl_value) \ |
| 153 | (Dhiddenhigh7mantissa(dbl_value)==0) |
| 154 | #define Dbl_iszero_sign(dbl_value) (Is_dsign(dbl_value)==0) |
| 155 | #define Dbl_iszero_exponent(dbl_value) (Dexponent(dbl_value)==0) |
| 156 | #define Dbl_iszero_mantissa(dbl_valueA,dbl_valueB) \ |
| 157 | (Dmantissap1(dbl_valueA)==0 && Dmantissap2(dbl_valueB)==0) |
| 158 | #define Dbl_iszero_exponentmantissa(dbl_valueA,dbl_valueB) \ |
| 159 | (Dexponentmantissap1(dbl_valueA)==0 && Dmantissap2(dbl_valueB)==0) |
| 160 | #define Dbl_isinfinity_exponent(dbl_value) \ |
| 161 | (Dexponent(dbl_value)==DBL_INFINITY_EXPONENT) |
| 162 | #define Dbl_isnotinfinity_exponent(dbl_value) \ |
| 163 | (Dexponent(dbl_value)!=DBL_INFINITY_EXPONENT) |
| 164 | #define Dbl_isinfinity(dbl_valueA,dbl_valueB) \ |
| 165 | (Dexponent(dbl_valueA)==DBL_INFINITY_EXPONENT && \ |
| 166 | Dmantissap1(dbl_valueA)==0 && Dmantissap2(dbl_valueB)==0) |
| 167 | #define Dbl_isnan(dbl_valueA,dbl_valueB) \ |
| 168 | (Dexponent(dbl_valueA)==DBL_INFINITY_EXPONENT && \ |
| 169 | (Dmantissap1(dbl_valueA)!=0 || Dmantissap2(dbl_valueB)!=0)) |
| 170 | #define Dbl_isnotnan(dbl_valueA,dbl_valueB) \ |
| 171 | (Dexponent(dbl_valueA)!=DBL_INFINITY_EXPONENT || \ |
| 172 | (Dmantissap1(dbl_valueA)==0 && Dmantissap2(dbl_valueB)==0)) |
| 173 | |
| 174 | #define Dbl_islessthan(dbl_op1a,dbl_op1b,dbl_op2a,dbl_op2b) \ |
| 175 | (Dallp1(dbl_op1a) < Dallp1(dbl_op2a) || \ |
| 176 | (Dallp1(dbl_op1a) == Dallp1(dbl_op2a) && \ |
| 177 | Dallp2(dbl_op1b) < Dallp2(dbl_op2b))) |
| 178 | #define Dbl_isgreaterthan(dbl_op1a,dbl_op1b,dbl_op2a,dbl_op2b) \ |
| 179 | (Dallp1(dbl_op1a) > Dallp1(dbl_op2a) || \ |
| 180 | (Dallp1(dbl_op1a) == Dallp1(dbl_op2a) && \ |
| 181 | Dallp2(dbl_op1b) > Dallp2(dbl_op2b))) |
| 182 | #define Dbl_isnotlessthan(dbl_op1a,dbl_op1b,dbl_op2a,dbl_op2b) \ |
| 183 | (Dallp1(dbl_op1a) > Dallp1(dbl_op2a) || \ |
| 184 | (Dallp1(dbl_op1a) == Dallp1(dbl_op2a) && \ |
| 185 | Dallp2(dbl_op1b) >= Dallp2(dbl_op2b))) |
| 186 | #define Dbl_isnotgreaterthan(dbl_op1a,dbl_op1b,dbl_op2a,dbl_op2b) \ |
| 187 | (Dallp1(dbl_op1a) < Dallp1(dbl_op2a) || \ |
| 188 | (Dallp1(dbl_op1a) == Dallp1(dbl_op2a) && \ |
| 189 | Dallp2(dbl_op1b) <= Dallp2(dbl_op2b))) |
| 190 | #define Dbl_isequal(dbl_op1a,dbl_op1b,dbl_op2a,dbl_op2b) \ |
| 191 | ((Dallp1(dbl_op1a) == Dallp1(dbl_op2a)) && \ |
| 192 | (Dallp2(dbl_op1b) == Dallp2(dbl_op2b))) |
| 193 | |
| 194 | #define Dbl_leftshiftby8(dbl_valueA,dbl_valueB) \ |
| 195 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),24,Dallp1(dbl_valueA)); \ |
| 196 | Dallp2(dbl_valueB) <<= 8 |
| 197 | #define Dbl_leftshiftby7(dbl_valueA,dbl_valueB) \ |
| 198 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),25,Dallp1(dbl_valueA)); \ |
| 199 | Dallp2(dbl_valueB) <<= 7 |
| 200 | #define Dbl_leftshiftby4(dbl_valueA,dbl_valueB) \ |
| 201 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),28,Dallp1(dbl_valueA)); \ |
| 202 | Dallp2(dbl_valueB) <<= 4 |
| 203 | #define Dbl_leftshiftby3(dbl_valueA,dbl_valueB) \ |
| 204 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),29,Dallp1(dbl_valueA)); \ |
| 205 | Dallp2(dbl_valueB) <<= 3 |
| 206 | #define Dbl_leftshiftby2(dbl_valueA,dbl_valueB) \ |
| 207 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),30,Dallp1(dbl_valueA)); \ |
| 208 | Dallp2(dbl_valueB) <<= 2 |
| 209 | #define Dbl_leftshiftby1(dbl_valueA,dbl_valueB) \ |
| 210 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),31,Dallp1(dbl_valueA)); \ |
| 211 | Dallp2(dbl_valueB) <<= 1 |
| 212 | |
| 213 | #define Dbl_rightshiftby8(dbl_valueA,dbl_valueB) \ |
| 214 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),8,Dallp2(dbl_valueB)); \ |
| 215 | Dallp1(dbl_valueA) >>= 8 |
| 216 | #define Dbl_rightshiftby4(dbl_valueA,dbl_valueB) \ |
| 217 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),4,Dallp2(dbl_valueB)); \ |
| 218 | Dallp1(dbl_valueA) >>= 4 |
| 219 | #define Dbl_rightshiftby2(dbl_valueA,dbl_valueB) \ |
| 220 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),2,Dallp2(dbl_valueB)); \ |
| 221 | Dallp1(dbl_valueA) >>= 2 |
| 222 | #define Dbl_rightshiftby1(dbl_valueA,dbl_valueB) \ |
| 223 | Shiftdouble(Dallp1(dbl_valueA),Dallp2(dbl_valueB),1,Dallp2(dbl_valueB)); \ |
| 224 | Dallp1(dbl_valueA) >>= 1 |
| 225 | |
| 226 | /* This magnitude comparison uses the signless first words and |
| 227 | * the regular part2 words. The comparison is graphically: |
| 228 | * |
| 229 | * 1st greater? ------------- |
| 230 | * | |
| 231 | * 1st less?-----------------+--------- |
| 232 | * | | |
| 233 | * 2nd greater or equal----->| | |
| 234 | * False True |
| 235 | */ |
| 236 | #define Dbl_ismagnitudeless(leftB,rightB,signlessleft,signlessright) \ |
| 237 | ((signlessleft <= signlessright) && \ |
| 238 | ( (signlessleft < signlessright) || (Dallp2(leftB)<Dallp2(rightB)) )) |
| 239 | |
| 240 | #define Dbl_copytoint_exponentmantissap1(src,dest) \ |
| 241 | dest = Dexponentmantissap1(src) |
| 242 | |
| 243 | /* A quiet NaN has the high mantissa bit clear and at least on other (in this |
| 244 | * case the adjacent bit) bit set. */ |
| 245 | #define Dbl_set_quiet(dbl_value) Deposit_dhigh2mantissa(dbl_value,1) |
| 246 | #define Dbl_set_exponent(dbl_value, exp) Deposit_dexponent(dbl_value,exp) |
| 247 | |
| 248 | #define Dbl_set_mantissa(desta,destb,valuea,valueb) \ |
| 249 | Deposit_dmantissap1(desta,valuea); \ |
| 250 | Dmantissap2(destb) = Dmantissap2(valueb) |
| 251 | #define Dbl_set_mantissap1(desta,valuea) \ |
| 252 | Deposit_dmantissap1(desta,valuea) |
| 253 | #define Dbl_set_mantissap2(destb,valueb) \ |
| 254 | Dmantissap2(destb) = Dmantissap2(valueb) |
| 255 | |
| 256 | #define Dbl_set_exponentmantissa(desta,destb,valuea,valueb) \ |
| 257 | Deposit_dexponentmantissap1(desta,valuea); \ |
| 258 | Dmantissap2(destb) = Dmantissap2(valueb) |
| 259 | #define Dbl_set_exponentmantissap1(dest,value) \ |
| 260 | Deposit_dexponentmantissap1(dest,value) |
| 261 | |
| 262 | #define Dbl_copyfromptr(src,desta,destb) \ |
| 263 | Dallp1(desta) = src->wd0; \ |
| 264 | Dallp2(destb) = src->wd1 |
| 265 | #define Dbl_copytoptr(srca,srcb,dest) \ |
| 266 | dest->wd0 = Dallp1(srca); \ |
| 267 | dest->wd1 = Dallp2(srcb) |
| 268 | |
| 269 | /* An infinity is represented with the max exponent and a zero mantissa */ |
| 270 | #define Dbl_setinfinity_exponent(dbl_value) \ |
| 271 | Deposit_dexponent(dbl_value,DBL_INFINITY_EXPONENT) |
| 272 | #define Dbl_setinfinity_exponentmantissa(dbl_valueA,dbl_valueB) \ |
| 273 | Deposit_dexponentmantissap1(dbl_valueA, \ |
| 274 | (DBL_INFINITY_EXPONENT << (32-(1+DBL_EXP_LENGTH)))); \ |
| 275 | Dmantissap2(dbl_valueB) = 0 |
| 276 | #define Dbl_setinfinitypositive(dbl_valueA,dbl_valueB) \ |
| 277 | Dallp1(dbl_valueA) \ |
| 278 | = (DBL_INFINITY_EXPONENT << (32-(1+DBL_EXP_LENGTH))); \ |
| 279 | Dmantissap2(dbl_valueB) = 0 |
| 280 | #define Dbl_setinfinitynegative(dbl_valueA,dbl_valueB) \ |
| 281 | Dallp1(dbl_valueA) = ((unsigned int)1<<31) | \ |
| 282 | (DBL_INFINITY_EXPONENT << (32-(1+DBL_EXP_LENGTH))); \ |
| 283 | Dmantissap2(dbl_valueB) = 0 |
| 284 | #define Dbl_setinfinity(dbl_valueA,dbl_valueB,sign) \ |
| 285 | Dallp1(dbl_valueA) = ((unsigned int)sign << 31) | \ |
| 286 | (DBL_INFINITY_EXPONENT << (32-(1+DBL_EXP_LENGTH))); \ |
| 287 | Dmantissap2(dbl_valueB) = 0 |
| 288 | |
| 289 | #define Dbl_sethigh4bits(dbl_value, extsign) Deposit_dhigh4p1(dbl_value,extsign) |
| 290 | #define Dbl_set_sign(dbl_value,sign) Deposit_dsign(dbl_value,sign) |
| 291 | #define Dbl_invert_sign(dbl_value) Deposit_dsign(dbl_value,~Dsign(dbl_value)) |
| 292 | #define Dbl_setone_sign(dbl_value) Deposit_dsign(dbl_value,1) |
| 293 | #define Dbl_setone_lowmantissap2(dbl_value) Deposit_dlowp2(dbl_value,1) |
| 294 | #define Dbl_setzero_sign(dbl_value) Dallp1(dbl_value) &= 0x7fffffff |
| 295 | #define Dbl_setzero_exponent(dbl_value) \ |
| 296 | Dallp1(dbl_value) &= 0x800fffff |
| 297 | #define Dbl_setzero_mantissa(dbl_valueA,dbl_valueB) \ |
| 298 | Dallp1(dbl_valueA) &= 0xfff00000; \ |
| 299 | Dallp2(dbl_valueB) = 0 |
| 300 | #define Dbl_setzero_mantissap1(dbl_value) Dallp1(dbl_value) &= 0xfff00000 |
| 301 | #define Dbl_setzero_mantissap2(dbl_value) Dallp2(dbl_value) = 0 |
| 302 | #define Dbl_setzero_exponentmantissa(dbl_valueA,dbl_valueB) \ |
| 303 | Dallp1(dbl_valueA) &= 0x80000000; \ |
| 304 | Dallp2(dbl_valueB) = 0 |
| 305 | #define Dbl_setzero_exponentmantissap1(dbl_valueA) \ |
| 306 | Dallp1(dbl_valueA) &= 0x80000000 |
| 307 | #define Dbl_setzero(dbl_valueA,dbl_valueB) \ |
| 308 | Dallp1(dbl_valueA) = 0; Dallp2(dbl_valueB) = 0 |
| 309 | #define Dbl_setzerop1(dbl_value) Dallp1(dbl_value) = 0 |
| 310 | #define Dbl_setzerop2(dbl_value) Dallp2(dbl_value) = 0 |
| 311 | #define Dbl_setnegativezero(dbl_value) \ |
| 312 | Dallp1(dbl_value) = (unsigned int)1 << 31; Dallp2(dbl_value) = 0 |
| 313 | #define Dbl_setnegativezerop1(dbl_value) Dallp1(dbl_value) = (unsigned int)1<<31 |
| 314 | |
| 315 | /* Use the following macro for both overflow & underflow conditions */ |
| 316 | #define ovfl - |
| 317 | #define unfl + |
| 318 | #define Dbl_setwrapped_exponent(dbl_value,exponent,op) \ |
| 319 | Deposit_dexponent(dbl_value,(exponent op DBL_WRAP)) |
| 320 | |
| 321 | #define Dbl_setlargestpositive(dbl_valueA,dbl_valueB) \ |
| 322 | Dallp1(dbl_valueA) = ((DBL_EMAX+DBL_BIAS) << (32-(1+DBL_EXP_LENGTH))) \ |
| 323 | | ((1<<(32-(1+DBL_EXP_LENGTH))) - 1 ); \ |
| 324 | Dallp2(dbl_valueB) = 0xFFFFFFFF |
| 325 | #define Dbl_setlargestnegative(dbl_valueA,dbl_valueB) \ |
| 326 | Dallp1(dbl_valueA) = ((DBL_EMAX+DBL_BIAS) << (32-(1+DBL_EXP_LENGTH))) \ |
| 327 | | ((1<<(32-(1+DBL_EXP_LENGTH))) - 1 ) \ |
| 328 | | ((unsigned int)1<<31); \ |
| 329 | Dallp2(dbl_valueB) = 0xFFFFFFFF |
| 330 | #define Dbl_setlargest_exponentmantissa(dbl_valueA,dbl_valueB) \ |
| 331 | Deposit_dexponentmantissap1(dbl_valueA, \ |
| 332 | (((DBL_EMAX+DBL_BIAS) << (32-(1+DBL_EXP_LENGTH))) \ |
| 333 | | ((1<<(32-(1+DBL_EXP_LENGTH))) - 1 ))); \ |
| 334 | Dallp2(dbl_valueB) = 0xFFFFFFFF |
| 335 | |
| 336 | #define Dbl_setnegativeinfinity(dbl_valueA,dbl_valueB) \ |
| 337 | Dallp1(dbl_valueA) = ((1<<DBL_EXP_LENGTH) | DBL_INFINITY_EXPONENT) \ |
| 338 | << (32-(1+DBL_EXP_LENGTH)) ; \ |
| 339 | Dallp2(dbl_valueB) = 0 |
| 340 | #define Dbl_setlargest(dbl_valueA,dbl_valueB,sign) \ |
| 341 | Dallp1(dbl_valueA) = ((unsigned int)sign << 31) | \ |
| 342 | ((DBL_EMAX+DBL_BIAS) << (32-(1+DBL_EXP_LENGTH))) | \ |
| 343 | ((1 << (32-(1+DBL_EXP_LENGTH))) - 1 ); \ |
| 344 | Dallp2(dbl_valueB) = 0xFFFFFFFF |
| 345 | |
| 346 | |
| 347 | /* The high bit is always zero so arithmetic or logical shifts will work. */ |
| 348 | #define Dbl_right_align(srcdstA,srcdstB,shift,extent) \ |
| 349 | if( shift >= 32 ) \ |
| 350 | { \ |
| 351 | /* Big shift requires examining the portion shift off \ |
| 352 | the end to properly set inexact. */ \ |
| 353 | if(shift < 64) \ |
| 354 | { \ |
| 355 | if(shift > 32) \ |
| 356 | { \ |
| 357 | Variable_shift_double(Dallp1(srcdstA),Dallp2(srcdstB), \ |
| 358 | shift-32, Extall(extent)); \ |
| 359 | if(Dallp2(srcdstB) << 64 - (shift)) Ext_setone_low(extent); \ |
| 360 | } \ |
| 361 | else Extall(extent) = Dallp2(srcdstB); \ |
| 362 | Dallp2(srcdstB) = Dallp1(srcdstA) >> (shift - 32); \ |
| 363 | } \ |
| 364 | else \ |
| 365 | { \ |
| 366 | Extall(extent) = Dallp1(srcdstA); \ |
| 367 | if(Dallp2(srcdstB)) Ext_setone_low(extent); \ |
| 368 | Dallp2(srcdstB) = 0; \ |
| 369 | } \ |
| 370 | Dallp1(srcdstA) = 0; \ |
| 371 | } \ |
| 372 | else \ |
| 373 | { \ |
| 374 | /* Small alignment is simpler. Extension is easily set. */ \ |
| 375 | if (shift > 0) \ |
| 376 | { \ |
| 377 | Extall(extent) = Dallp2(srcdstB) << 32 - (shift); \ |
| 378 | Variable_shift_double(Dallp1(srcdstA),Dallp2(srcdstB),shift, \ |
| 379 | Dallp2(srcdstB)); \ |
| 380 | Dallp1(srcdstA) >>= shift; \ |
| 381 | } \ |
| 382 | else Extall(extent) = 0; \ |
| 383 | } |
| 384 | |
| 385 | /* |
| 386 | * Here we need to shift the result right to correct for an overshift |
| 387 | * (due to the exponent becoming negative) during normalization. |
| 388 | */ |
| 389 | #define Dbl_fix_overshift(srcdstA,srcdstB,shift,extent) \ |
| 390 | Extall(extent) = Dallp2(srcdstB) << 32 - (shift); \ |
| 391 | Dallp2(srcdstB) = (Dallp1(srcdstA) << 32 - (shift)) | \ |
| 392 | (Dallp2(srcdstB) >> (shift)); \ |
| 393 | Dallp1(srcdstA) = Dallp1(srcdstA) >> shift |
| 394 | |
| 395 | #define Dbl_hiddenhigh3mantissa(dbl_value) Dhiddenhigh3mantissa(dbl_value) |
| 396 | #define Dbl_hidden(dbl_value) Dhidden(dbl_value) |
| 397 | #define Dbl_lowmantissap2(dbl_value) Dlowp2(dbl_value) |
| 398 | |
| 399 | /* The left argument is never smaller than the right argument */ |
| 400 | #define Dbl_subtract(lefta,leftb,righta,rightb,resulta,resultb) \ |
| 401 | if( Dallp2(rightb) > Dallp2(leftb) ) Dallp1(lefta)--; \ |
| 402 | Dallp2(resultb) = Dallp2(leftb) - Dallp2(rightb); \ |
| 403 | Dallp1(resulta) = Dallp1(lefta) - Dallp1(righta) |
| 404 | |
| 405 | /* Subtract right augmented with extension from left augmented with zeros and |
| 406 | * store into result and extension. */ |
| 407 | #define Dbl_subtract_withextension(lefta,leftb,righta,rightb,extent,resulta,resultb) \ |
| 408 | Dbl_subtract(lefta,leftb,righta,rightb,resulta,resultb); \ |
| 409 | if( (Extall(extent) = 0-Extall(extent)) ) \ |
| 410 | { \ |
| 411 | if((Dallp2(resultb)--) == 0) Dallp1(resulta)--; \ |
| 412 | } |
| 413 | |
| 414 | #define Dbl_addition(lefta,leftb,righta,rightb,resulta,resultb) \ |
| 415 | /* If the sum of the low words is less than either source, then \ |
| 416 | * an overflow into the next word occurred. */ \ |
| 417 | Dallp1(resulta) = Dallp1(lefta) + Dallp1(righta); \ |
| 418 | if((Dallp2(resultb) = Dallp2(leftb) + Dallp2(rightb)) < Dallp2(rightb)) \ |
| 419 | Dallp1(resulta)++ |
| 420 | |
| 421 | #define Dbl_xortointp1(left,right,result) \ |
| 422 | result = Dallp1(left) XOR Dallp1(right) |
| 423 | |
| 424 | #define Dbl_xorfromintp1(left,right,result) \ |
| 425 | Dallp1(result) = left XOR Dallp1(right) |
| 426 | |
| 427 | #define Dbl_swap_lower(left,right) \ |
| 428 | Dallp2(left) = Dallp2(left) XOR Dallp2(right); \ |
| 429 | Dallp2(right) = Dallp2(left) XOR Dallp2(right); \ |
| 430 | Dallp2(left) = Dallp2(left) XOR Dallp2(right) |
| 431 | |
| 432 | /* Need to Initialize */ |
| 433 | #define Dbl_makequietnan(desta,destb) \ |
| 434 | Dallp1(desta) = ((DBL_EMAX+DBL_BIAS)+1)<< (32-(1+DBL_EXP_LENGTH)) \ |
| 435 | | (1<<(32-(1+DBL_EXP_LENGTH+2))); \ |
| 436 | Dallp2(destb) = 0 |
| 437 | #define Dbl_makesignalingnan(desta,destb) \ |
| 438 | Dallp1(desta) = ((DBL_EMAX+DBL_BIAS)+1)<< (32-(1+DBL_EXP_LENGTH)) \ |
| 439 | | (1<<(32-(1+DBL_EXP_LENGTH+1))); \ |
| 440 | Dallp2(destb) = 0 |
| 441 | |
| 442 | #define Dbl_normalize(dbl_opndA,dbl_opndB,exponent) \ |
| 443 | while(Dbl_iszero_hiddenhigh7mantissa(dbl_opndA)) { \ |
| 444 | Dbl_leftshiftby8(dbl_opndA,dbl_opndB); \ |
| 445 | exponent -= 8; \ |
| 446 | } \ |
| 447 | if(Dbl_iszero_hiddenhigh3mantissa(dbl_opndA)) { \ |
| 448 | Dbl_leftshiftby4(dbl_opndA,dbl_opndB); \ |
| 449 | exponent -= 4; \ |
| 450 | } \ |
| 451 | while(Dbl_iszero_hidden(dbl_opndA)) { \ |
| 452 | Dbl_leftshiftby1(dbl_opndA,dbl_opndB); \ |
| 453 | exponent -= 1; \ |
| 454 | } |
| 455 | |
| 456 | #define Twoword_add(src1dstA,src1dstB,src2A,src2B) \ |
| 457 | /* \ |
| 458 | * want this macro to generate: \ |
| 459 | * ADD src1dstB,src2B,src1dstB; \ |
| 460 | * ADDC src1dstA,src2A,src1dstA; \ |
| 461 | */ \ |
| 462 | if ((src1dstB) + (src2B) < (src1dstB)) Dallp1(src1dstA)++; \ |
| 463 | Dallp1(src1dstA) += (src2A); \ |
| 464 | Dallp2(src1dstB) += (src2B) |
| 465 | |
| 466 | #define Twoword_subtract(src1dstA,src1dstB,src2A,src2B) \ |
| 467 | /* \ |
| 468 | * want this macro to generate: \ |
| 469 | * SUB src1dstB,src2B,src1dstB; \ |
| 470 | * SUBB src1dstA,src2A,src1dstA; \ |
| 471 | */ \ |
| 472 | if ((src1dstB) < (src2B)) Dallp1(src1dstA)--; \ |
| 473 | Dallp1(src1dstA) -= (src2A); \ |
| 474 | Dallp2(src1dstB) -= (src2B) |
| 475 | |
| 476 | #define Dbl_setoverflow(resultA,resultB) \ |
| 477 | /* set result to infinity or largest number */ \ |
| 478 | switch (Rounding_mode()) { \ |
| 479 | case ROUNDPLUS: \ |
| 480 | if (Dbl_isone_sign(resultA)) { \ |
| 481 | Dbl_setlargestnegative(resultA,resultB); \ |
| 482 | } \ |
| 483 | else { \ |
| 484 | Dbl_setinfinitypositive(resultA,resultB); \ |
| 485 | } \ |
| 486 | break; \ |
| 487 | case ROUNDMINUS: \ |
| 488 | if (Dbl_iszero_sign(resultA)) { \ |
| 489 | Dbl_setlargestpositive(resultA,resultB); \ |
| 490 | } \ |
| 491 | else { \ |
| 492 | Dbl_setinfinitynegative(resultA,resultB); \ |
| 493 | } \ |
| 494 | break; \ |
| 495 | case ROUNDNEAREST: \ |
| 496 | Dbl_setinfinity_exponentmantissa(resultA,resultB); \ |
| 497 | break; \ |
| 498 | case ROUNDZERO: \ |
| 499 | Dbl_setlargest_exponentmantissa(resultA,resultB); \ |
| 500 | } |
| 501 | |
| 502 | #define Dbl_denormalize(opndp1,opndp2,exponent,guard,sticky,inexact) \ |
| 503 | Dbl_clear_signexponent_set_hidden(opndp1); \ |
| 504 | if (exponent >= (1-DBL_P)) { \ |
| 505 | if (exponent >= -31) { \ |
| 506 | guard = (Dallp2(opndp2) >> -exponent) & 1; \ |
| 507 | if (exponent < 0) sticky |= Dallp2(opndp2) << (32+exponent); \ |
| 508 | if (exponent > -31) { \ |
| 509 | Variable_shift_double(opndp1,opndp2,1-exponent,opndp2); \ |
| 510 | Dallp1(opndp1) >>= 1-exponent; \ |
| 511 | } \ |
| 512 | else { \ |
| 513 | Dallp2(opndp2) = Dallp1(opndp1); \ |
| 514 | Dbl_setzerop1(opndp1); \ |
| 515 | } \ |
| 516 | } \ |
| 517 | else { \ |
| 518 | guard = (Dallp1(opndp1) >> -32-exponent) & 1; \ |
| 519 | if (exponent == -32) sticky |= Dallp2(opndp2); \ |
| 520 | else sticky |= (Dallp2(opndp2) | Dallp1(opndp1) << 64+exponent); \ |
| 521 | Dallp2(opndp2) = Dallp1(opndp1) >> -31-exponent; \ |
| 522 | Dbl_setzerop1(opndp1); \ |
| 523 | } \ |
| 524 | inexact = guard | sticky; \ |
| 525 | } \ |
| 526 | else { \ |
| 527 | guard = 0; \ |
| 528 | sticky |= (Dallp1(opndp1) | Dallp2(opndp2)); \ |
| 529 | Dbl_setzero(opndp1,opndp2); \ |
| 530 | inexact = sticky; \ |
| 531 | } |
| 532 | |
| 533 | /* |
| 534 | * The fused multiply add instructions requires a double extended format, |
| 535 | * with 106 bits of mantissa. |
| 536 | */ |
| 537 | #define DBLEXT_THRESHOLD 106 |
| 538 | |
| 539 | #define Dblext_setzero(valA,valB,valC,valD) \ |
| 540 | Dextallp1(valA) = 0; Dextallp2(valB) = 0; \ |
| 541 | Dextallp3(valC) = 0; Dextallp4(valD) = 0 |
| 542 | |
| 543 | |
| 544 | #define Dblext_isnotzero_mantissap3(valC) (Dextallp3(valC)!=0) |
| 545 | #define Dblext_isnotzero_mantissap4(valD) (Dextallp3(valD)!=0) |
| 546 | #define Dblext_isone_lowp2(val) (Dextlowp2(val)!=0) |
| 547 | #define Dblext_isone_highp3(val) (Dexthighp3(val)!=0) |
| 548 | #define Dblext_isnotzero_low31p3(val) (Dextlow31p3(val)!=0) |
| 549 | #define Dblext_iszero(valA,valB,valC,valD) (Dextallp1(valA)==0 && \ |
| 550 | Dextallp2(valB)==0 && Dextallp3(valC)==0 && Dextallp4(valD)==0) |
| 551 | |
| 552 | #define Dblext_copy(srca,srcb,srcc,srcd,desta,destb,destc,destd) \ |
| 553 | Dextallp1(desta) = Dextallp4(srca); \ |
| 554 | Dextallp2(destb) = Dextallp4(srcb); \ |
| 555 | Dextallp3(destc) = Dextallp4(srcc); \ |
| 556 | Dextallp4(destd) = Dextallp4(srcd) |
| 557 | |
| 558 | #define Dblext_swap_lower(leftp2,leftp3,leftp4,rightp2,rightp3,rightp4) \ |
| 559 | Dextallp2(leftp2) = Dextallp2(leftp2) XOR Dextallp2(rightp2); \ |
| 560 | Dextallp2(rightp2) = Dextallp2(leftp2) XOR Dextallp2(rightp2); \ |
| 561 | Dextallp2(leftp2) = Dextallp2(leftp2) XOR Dextallp2(rightp2); \ |
| 562 | Dextallp3(leftp3) = Dextallp3(leftp3) XOR Dextallp3(rightp3); \ |
| 563 | Dextallp3(rightp3) = Dextallp3(leftp3) XOR Dextallp3(rightp3); \ |
| 564 | Dextallp3(leftp3) = Dextallp3(leftp3) XOR Dextallp3(rightp3); \ |
| 565 | Dextallp4(leftp4) = Dextallp4(leftp4) XOR Dextallp4(rightp4); \ |
| 566 | Dextallp4(rightp4) = Dextallp4(leftp4) XOR Dextallp4(rightp4); \ |
| 567 | Dextallp4(leftp4) = Dextallp4(leftp4) XOR Dextallp4(rightp4) |
| 568 | |
| 569 | #define Dblext_setone_lowmantissap4(dbl_value) Deposit_dextlowp4(dbl_value,1) |
| 570 | |
| 571 | /* The high bit is always zero so arithmetic or logical shifts will work. */ |
| 572 | #define Dblext_right_align(srcdstA,srcdstB,srcdstC,srcdstD,shift) \ |
| 573 | {int shiftamt, sticky; \ |
| 574 | shiftamt = shift % 32; \ |
| 575 | sticky = 0; \ |
| 576 | switch (shift/32) { \ |
| 577 | case 0: if (shiftamt > 0) { \ |
| 578 | sticky = Dextallp4(srcdstD) << 32 - (shiftamt); \ |
| 579 | Variable_shift_double(Dextallp3(srcdstC), \ |
| 580 | Dextallp4(srcdstD),shiftamt,Dextallp4(srcdstD)); \ |
| 581 | Variable_shift_double(Dextallp2(srcdstB), \ |
| 582 | Dextallp3(srcdstC),shiftamt,Dextallp3(srcdstC)); \ |
| 583 | Variable_shift_double(Dextallp1(srcdstA), \ |
| 584 | Dextallp2(srcdstB),shiftamt,Dextallp2(srcdstB)); \ |
| 585 | Dextallp1(srcdstA) >>= shiftamt; \ |
| 586 | } \ |
| 587 | break; \ |
| 588 | case 1: if (shiftamt > 0) { \ |
| 589 | sticky = (Dextallp3(srcdstC) << 31 - shiftamt) | \ |
| 590 | Dextallp4(srcdstD); \ |
| 591 | Variable_shift_double(Dextallp2(srcdstB), \ |
| 592 | Dextallp3(srcdstC),shiftamt,Dextallp4(srcdstD)); \ |
| 593 | Variable_shift_double(Dextallp1(srcdstA), \ |
| 594 | Dextallp2(srcdstB),shiftamt,Dextallp3(srcdstC)); \ |
| 595 | } \ |
| 596 | else { \ |
| 597 | sticky = Dextallp4(srcdstD); \ |
| 598 | Dextallp4(srcdstD) = Dextallp3(srcdstC); \ |
| 599 | Dextallp3(srcdstC) = Dextallp2(srcdstB); \ |
| 600 | } \ |
| 601 | Dextallp2(srcdstB) = Dextallp1(srcdstA) >> shiftamt; \ |
| 602 | Dextallp1(srcdstA) = 0; \ |
| 603 | break; \ |
| 604 | case 2: if (shiftamt > 0) { \ |
| 605 | sticky = (Dextallp2(srcdstB) << 31 - shiftamt) | \ |
| 606 | Dextallp3(srcdstC) | Dextallp4(srcdstD); \ |
| 607 | Variable_shift_double(Dextallp1(srcdstA), \ |
| 608 | Dextallp2(srcdstB),shiftamt,Dextallp4(srcdstD)); \ |
| 609 | } \ |
| 610 | else { \ |
| 611 | sticky = Dextallp3(srcdstC) | Dextallp4(srcdstD); \ |
| 612 | Dextallp4(srcdstD) = Dextallp2(srcdstB); \ |
| 613 | } \ |
| 614 | Dextallp3(srcdstC) = Dextallp1(srcdstA) >> shiftamt; \ |
| 615 | Dextallp1(srcdstA) = Dextallp2(srcdstB) = 0; \ |
| 616 | break; \ |
| 617 | case 3: if (shiftamt > 0) { \ |
| 618 | sticky = (Dextallp1(srcdstA) << 31 - shiftamt) | \ |
| 619 | Dextallp2(srcdstB) | Dextallp3(srcdstC) | \ |
| 620 | Dextallp4(srcdstD); \ |
| 621 | } \ |
| 622 | else { \ |
| 623 | sticky = Dextallp2(srcdstB) | Dextallp3(srcdstC) | \ |
| 624 | Dextallp4(srcdstD); \ |
| 625 | } \ |
| 626 | Dextallp4(srcdstD) = Dextallp1(srcdstA) >> shiftamt; \ |
| 627 | Dextallp1(srcdstA) = Dextallp2(srcdstB) = 0; \ |
| 628 | Dextallp3(srcdstC) = 0; \ |
| 629 | break; \ |
| 630 | } \ |
| 631 | if (sticky) Dblext_setone_lowmantissap4(srcdstD); \ |
| 632 | } |
| 633 | |
| 634 | /* The left argument is never smaller than the right argument */ |
| 635 | #define Dblext_subtract(lefta,leftb,leftc,leftd,righta,rightb,rightc,rightd,resulta,resultb,resultc,resultd) \ |
| 636 | if( Dextallp4(rightd) > Dextallp4(leftd) ) \ |
| 637 | if( (Dextallp3(leftc)--) == 0) \ |
| 638 | if( (Dextallp2(leftb)--) == 0) Dextallp1(lefta)--; \ |
| 639 | Dextallp4(resultd) = Dextallp4(leftd) - Dextallp4(rightd); \ |
| 640 | if( Dextallp3(rightc) > Dextallp3(leftc) ) \ |
| 641 | if( (Dextallp2(leftb)--) == 0) Dextallp1(lefta)--; \ |
| 642 | Dextallp3(resultc) = Dextallp3(leftc) - Dextallp3(rightc); \ |
| 643 | if( Dextallp2(rightb) > Dextallp2(leftb) ) Dextallp1(lefta)--; \ |
| 644 | Dextallp2(resultb) = Dextallp2(leftb) - Dextallp2(rightb); \ |
| 645 | Dextallp1(resulta) = Dextallp1(lefta) - Dextallp1(righta) |
| 646 | |
| 647 | #define Dblext_addition(lefta,leftb,leftc,leftd,righta,rightb,rightc,rightd,resulta,resultb,resultc,resultd) \ |
| 648 | /* If the sum of the low words is less than either source, then \ |
| 649 | * an overflow into the next word occurred. */ \ |
| 650 | if ((Dextallp4(resultd) = Dextallp4(leftd)+Dextallp4(rightd)) < \ |
| 651 | Dextallp4(rightd)) \ |
| 652 | if((Dextallp3(resultc) = Dextallp3(leftc)+Dextallp3(rightc)+1) <= \ |
| 653 | Dextallp3(rightc)) \ |
| 654 | if((Dextallp2(resultb) = Dextallp2(leftb)+Dextallp2(rightb)+1) \ |
| 655 | <= Dextallp2(rightb)) \ |
| 656 | Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta)+1; \ |
| 657 | else Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta); \ |
| 658 | else \ |
| 659 | if ((Dextallp2(resultb) = Dextallp2(leftb)+Dextallp2(rightb)) < \ |
| 660 | Dextallp2(rightb)) \ |
| 661 | Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta)+1; \ |
| 662 | else Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta); \ |
| 663 | else \ |
| 664 | if ((Dextallp3(resultc) = Dextallp3(leftc)+Dextallp3(rightc)) < \ |
| 665 | Dextallp3(rightc)) \ |
| 666 | if ((Dextallp2(resultb) = Dextallp2(leftb)+Dextallp2(rightb)+1) \ |
| 667 | <= Dextallp2(rightb)) \ |
| 668 | Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta)+1; \ |
| 669 | else Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta); \ |
| 670 | else \ |
| 671 | if ((Dextallp2(resultb) = Dextallp2(leftb)+Dextallp2(rightb)) < \ |
| 672 | Dextallp2(rightb)) \ |
| 673 | Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta)+1; \ |
| 674 | else Dextallp1(resulta) = Dextallp1(lefta)+Dextallp1(righta) |
| 675 | |
| 676 | |
| 677 | #define Dblext_arithrightshiftby1(srcdstA,srcdstB,srcdstC,srcdstD) \ |
| 678 | Shiftdouble(Dextallp3(srcdstC),Dextallp4(srcdstD),1,Dextallp4(srcdstD)); \ |
| 679 | Shiftdouble(Dextallp2(srcdstB),Dextallp3(srcdstC),1,Dextallp3(srcdstC)); \ |
| 680 | Shiftdouble(Dextallp1(srcdstA),Dextallp2(srcdstB),1,Dextallp2(srcdstB)); \ |
| 681 | Dextallp1(srcdstA) = (int)Dextallp1(srcdstA) >> 1 |
| 682 | |
| 683 | #define Dblext_leftshiftby8(valA,valB,valC,valD) \ |
| 684 | Shiftdouble(Dextallp1(valA),Dextallp2(valB),24,Dextallp1(valA)); \ |
| 685 | Shiftdouble(Dextallp2(valB),Dextallp3(valC),24,Dextallp2(valB)); \ |
| 686 | Shiftdouble(Dextallp3(valC),Dextallp4(valD),24,Dextallp3(valC)); \ |
| 687 | Dextallp4(valD) <<= 8 |
| 688 | #define Dblext_leftshiftby4(valA,valB,valC,valD) \ |
| 689 | Shiftdouble(Dextallp1(valA),Dextallp2(valB),28,Dextallp1(valA)); \ |
| 690 | Shiftdouble(Dextallp2(valB),Dextallp3(valC),28,Dextallp2(valB)); \ |
| 691 | Shiftdouble(Dextallp3(valC),Dextallp4(valD),28,Dextallp3(valC)); \ |
| 692 | Dextallp4(valD) <<= 4 |
| 693 | #define Dblext_leftshiftby3(valA,valB,valC,valD) \ |
| 694 | Shiftdouble(Dextallp1(valA),Dextallp2(valB),29,Dextallp1(valA)); \ |
| 695 | Shiftdouble(Dextallp2(valB),Dextallp3(valC),29,Dextallp2(valB)); \ |
| 696 | Shiftdouble(Dextallp3(valC),Dextallp4(valD),29,Dextallp3(valC)); \ |
| 697 | Dextallp4(valD) <<= 3 |
| 698 | #define Dblext_leftshiftby2(valA,valB,valC,valD) \ |
| 699 | Shiftdouble(Dextallp1(valA),Dextallp2(valB),30,Dextallp1(valA)); \ |
| 700 | Shiftdouble(Dextallp2(valB),Dextallp3(valC),30,Dextallp2(valB)); \ |
| 701 | Shiftdouble(Dextallp3(valC),Dextallp4(valD),30,Dextallp3(valC)); \ |
| 702 | Dextallp4(valD) <<= 2 |
| 703 | #define Dblext_leftshiftby1(valA,valB,valC,valD) \ |
| 704 | Shiftdouble(Dextallp1(valA),Dextallp2(valB),31,Dextallp1(valA)); \ |
| 705 | Shiftdouble(Dextallp2(valB),Dextallp3(valC),31,Dextallp2(valB)); \ |
| 706 | Shiftdouble(Dextallp3(valC),Dextallp4(valD),31,Dextallp3(valC)); \ |
| 707 | Dextallp4(valD) <<= 1 |
| 708 | |
| 709 | #define Dblext_rightshiftby4(valueA,valueB,valueC,valueD) \ |
| 710 | Shiftdouble(Dextallp3(valueC),Dextallp4(valueD),4,Dextallp4(valueD)); \ |
| 711 | Shiftdouble(Dextallp2(valueB),Dextallp3(valueC),4,Dextallp3(valueC)); \ |
| 712 | Shiftdouble(Dextallp1(valueA),Dextallp2(valueB),4,Dextallp2(valueB)); \ |
| 713 | Dextallp1(valueA) >>= 4 |
| 714 | #define Dblext_rightshiftby1(valueA,valueB,valueC,valueD) \ |
| 715 | Shiftdouble(Dextallp3(valueC),Dextallp4(valueD),1,Dextallp4(valueD)); \ |
| 716 | Shiftdouble(Dextallp2(valueB),Dextallp3(valueC),1,Dextallp3(valueC)); \ |
| 717 | Shiftdouble(Dextallp1(valueA),Dextallp2(valueB),1,Dextallp2(valueB)); \ |
| 718 | Dextallp1(valueA) >>= 1 |
| 719 | |
| 720 | #define Dblext_xortointp1(left,right,result) Dbl_xortointp1(left,right,result) |
| 721 | |
| 722 | #define Dblext_xorfromintp1(left,right,result) \ |
| 723 | Dbl_xorfromintp1(left,right,result) |
| 724 | |
| 725 | #define Dblext_copytoint_exponentmantissap1(src,dest) \ |
| 726 | Dbl_copytoint_exponentmantissap1(src,dest) |
| 727 | |
| 728 | #define Dblext_ismagnitudeless(leftB,rightB,signlessleft,signlessright) \ |
| 729 | Dbl_ismagnitudeless(leftB,rightB,signlessleft,signlessright) |
| 730 | |
| 731 | #define Dbl_copyto_dblext(src1,src2,dest1,dest2,dest3,dest4) \ |
| 732 | Dextallp1(dest1) = Dallp1(src1); Dextallp2(dest2) = Dallp2(src2); \ |
| 733 | Dextallp3(dest3) = 0; Dextallp4(dest4) = 0 |
| 734 | |
| 735 | #define Dblext_set_sign(dbl_value,sign) Dbl_set_sign(dbl_value,sign) |
| 736 | #define Dblext_clear_signexponent_set_hidden(srcdst) \ |
| 737 | Dbl_clear_signexponent_set_hidden(srcdst) |
| 738 | #define Dblext_clear_signexponent(srcdst) Dbl_clear_signexponent(srcdst) |
| 739 | #define Dblext_clear_sign(srcdst) Dbl_clear_sign(srcdst) |
| 740 | #define Dblext_isone_hidden(dbl_value) Dbl_isone_hidden(dbl_value) |
| 741 | |
| 742 | /* |
| 743 | * The Fourword_add() macro assumes that integers are 4 bytes in size. |
| 744 | * It will break if this is not the case. |
| 745 | */ |
| 746 | |
| 747 | #define Fourword_add(src1dstA,src1dstB,src1dstC,src1dstD,src2A,src2B,src2C,src2D) \ |
| 748 | /* \ |
| 749 | * want this macro to generate: \ |
| 750 | * ADD src1dstD,src2D,src1dstD; \ |
| 751 | * ADDC src1dstC,src2C,src1dstC; \ |
| 752 | * ADDC src1dstB,src2B,src1dstB; \ |
| 753 | * ADDC src1dstA,src2A,src1dstA; \ |
| 754 | */ \ |
| 755 | if ((unsigned int)(src1dstD += (src2D)) < (unsigned int)(src2D)) { \ |
| 756 | if ((unsigned int)(src1dstC += (src2C) + 1) <= \ |
| 757 | (unsigned int)(src2C)) { \ |
| 758 | if ((unsigned int)(src1dstB += (src2B) + 1) <= \ |
| 759 | (unsigned int)(src2B)) src1dstA++; \ |
| 760 | } \ |
| 761 | else if ((unsigned int)(src1dstB += (src2B)) < \ |
| 762 | (unsigned int)(src2B)) src1dstA++; \ |
| 763 | } \ |
| 764 | else { \ |
| 765 | if ((unsigned int)(src1dstC += (src2C)) < \ |
| 766 | (unsigned int)(src2C)) { \ |
| 767 | if ((unsigned int)(src1dstB += (src2B) + 1) <= \ |
| 768 | (unsigned int)(src2B)) src1dstA++; \ |
| 769 | } \ |
| 770 | else if ((unsigned int)(src1dstB += (src2B)) < \ |
| 771 | (unsigned int)(src2B)) src1dstA++; \ |
| 772 | } \ |
| 773 | src1dstA += (src2A) |
| 774 | |
| 775 | #define Dblext_denormalize(opndp1,opndp2,opndp3,opndp4,exponent,is_tiny) \ |
| 776 | {int shiftamt, sticky; \ |
| 777 | is_tiny = TRUE; \ |
| 778 | if (exponent == 0 && (Dextallp3(opndp3) || Dextallp4(opndp4))) { \ |
| 779 | switch (Rounding_mode()) { \ |
| 780 | case ROUNDPLUS: \ |
| 781 | if (Dbl_iszero_sign(opndp1)) { \ |
| 782 | Dbl_increment(opndp1,opndp2); \ |
| 783 | if (Dbl_isone_hiddenoverflow(opndp1)) \ |
| 784 | is_tiny = FALSE; \ |
| 785 | Dbl_decrement(opndp1,opndp2); \ |
| 786 | } \ |
| 787 | break; \ |
| 788 | case ROUNDMINUS: \ |
| 789 | if (Dbl_isone_sign(opndp1)) { \ |
| 790 | Dbl_increment(opndp1,opndp2); \ |
| 791 | if (Dbl_isone_hiddenoverflow(opndp1)) \ |
| 792 | is_tiny = FALSE; \ |
| 793 | Dbl_decrement(opndp1,opndp2); \ |
| 794 | } \ |
| 795 | break; \ |
| 796 | case ROUNDNEAREST: \ |
| 797 | if (Dblext_isone_highp3(opndp3) && \ |
| 798 | (Dblext_isone_lowp2(opndp2) || \ |
| 799 | Dblext_isnotzero_low31p3(opndp3))) { \ |
| 800 | Dbl_increment(opndp1,opndp2); \ |
| 801 | if (Dbl_isone_hiddenoverflow(opndp1)) \ |
| 802 | is_tiny = FALSE; \ |
| 803 | Dbl_decrement(opndp1,opndp2); \ |
| 804 | } \ |
| 805 | break; \ |
| 806 | } \ |
| 807 | } \ |
| 808 | Dblext_clear_signexponent_set_hidden(opndp1); \ |
| 809 | if (exponent >= (1-QUAD_P)) { \ |
| 810 | shiftamt = (1-exponent) % 32; \ |
| 811 | switch((1-exponent)/32) { \ |
| 812 | case 0: sticky = Dextallp4(opndp4) << 32-(shiftamt); \ |
| 813 | Variableshiftdouble(opndp3,opndp4,shiftamt,opndp4); \ |
| 814 | Variableshiftdouble(opndp2,opndp3,shiftamt,opndp3); \ |
| 815 | Variableshiftdouble(opndp1,opndp2,shiftamt,opndp2); \ |
| 816 | Dextallp1(opndp1) >>= shiftamt; \ |
| 817 | break; \ |
| 818 | case 1: sticky = (Dextallp3(opndp3) << 32-(shiftamt)) | \ |
| 819 | Dextallp4(opndp4); \ |
| 820 | Variableshiftdouble(opndp2,opndp3,shiftamt,opndp4); \ |
| 821 | Variableshiftdouble(opndp1,opndp2,shiftamt,opndp3); \ |
| 822 | Dextallp2(opndp2) = Dextallp1(opndp1) >> shiftamt; \ |
| 823 | Dextallp1(opndp1) = 0; \ |
| 824 | break; \ |
| 825 | case 2: sticky = (Dextallp2(opndp2) << 32-(shiftamt)) | \ |
| 826 | Dextallp3(opndp3) | Dextallp4(opndp4); \ |
| 827 | Variableshiftdouble(opndp1,opndp2,shiftamt,opndp4); \ |
| 828 | Dextallp3(opndp3) = Dextallp1(opndp1) >> shiftamt; \ |
| 829 | Dextallp1(opndp1) = Dextallp2(opndp2) = 0; \ |
| 830 | break; \ |
| 831 | case 3: sticky = (Dextallp1(opndp1) << 32-(shiftamt)) | \ |
| 832 | Dextallp2(opndp2) | Dextallp3(opndp3) | \ |
| 833 | Dextallp4(opndp4); \ |
| 834 | Dextallp4(opndp4) = Dextallp1(opndp1) >> shiftamt; \ |
| 835 | Dextallp1(opndp1) = Dextallp2(opndp2) = 0; \ |
| 836 | Dextallp3(opndp3) = 0; \ |
| 837 | break; \ |
| 838 | } \ |
| 839 | } \ |
| 840 | else { \ |
| 841 | sticky = Dextallp1(opndp1) | Dextallp2(opndp2) | \ |
| 842 | Dextallp3(opndp3) | Dextallp4(opndp4); \ |
| 843 | Dblext_setzero(opndp1,opndp2,opndp3,opndp4); \ |
| 844 | } \ |
| 845 | if (sticky) Dblext_setone_lowmantissap4(opndp4); \ |
| 846 | exponent = 0; \ |
| 847 | } |