scale.sa revision 1.2
1 * MOTOROLA MICROPROCESSOR & MEMORY TECHNOLOGY GROUP 2 * M68000 Hi-Performance Microprocessor Division 3 * M68040 Software Package 4 * 5 * M68040 Software Package Copyright (c) 1993, 1994 Motorola Inc. 6 * All rights reserved. 7 * 8 * THE SOFTWARE is provided on an "AS IS" basis and without warranty. 9 * To the maximum extent permitted by applicable law, 10 * MOTOROLA DISCLAIMS ALL WARRANTIES WHETHER EXPRESS OR IMPLIED, 11 * INCLUDING IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A 12 * PARTICULAR PURPOSE and any warranty against infringement with 13 * regard to the SOFTWARE (INCLUDING ANY MODIFIED VERSIONS THEREOF) 14 * and any accompanying written materials. 15 * 16 * To the maximum extent permitted by applicable law, 17 * IN NO EVENT SHALL MOTOROLA BE LIABLE FOR ANY DAMAGES WHATSOEVER 18 * (INCLUDING WITHOUT LIMITATION, DAMAGES FOR LOSS OF BUSINESS 19 * PROFITS, BUSINESS INTERRUPTION, LOSS OF BUSINESS INFORMATION, OR 20 * OTHER PECUNIARY LOSS) ARISING OF THE USE OR INABILITY TO USE THE 21 * SOFTWARE. Motorola assumes no responsibility for the maintenance 22 * and support of the SOFTWARE. 23 * 24 * You are hereby granted a copyright license to use, modify, and 25 * distribute the SOFTWARE so long as this entire notice is retained 26 * without alteration in any modified and/or redistributed versions, 27 * and that such modified versions are clearly identified as such. 28 * No licenses are granted by implication, estoppel or otherwise 29 * under any patents or trademarks of Motorola, Inc. 30 31 * 32 * scale.sa 3.3 7/30/91 33 * 34 * The entry point sSCALE computes the destination operand 35 * scaled by the source operand. If the absoulute value of 36 * the source operand is (>= 2^14) an overflow or underflow 37 * is returned. 38 * 39 * The entry point sscale is called from do_func to emulate 40 * the fscale unimplemented instruction. 41 * 42 * Input: Double-extended destination operand in FPTEMP, 43 * double-extended source operand in ETEMP. 44 * 45 * Output: The function returns scale(X,Y) to fp0. 46 * 47 * Modifies: fp0. 48 * 49 * Algorithm: 50 * 51 52 SCALE IDNT 2,1 Motorola 040 Floating Point Software Package 53 54 section 8 55 56 include fpsp.h 57 58 xref t_ovfl2 59 xref t_unfl 60 xref round 61 xref t_resdnrm 62 63 SRC_BNDS dc.w $3fff,$400c 64 65 * 66 * This entry point is used by the unimplemented instruction exception 67 * handler. 68 * 69 * 70 * 71 * FSCALE 72 * 73 xdef sscale 74 sscale: 75 fmove.l #0,fpcr ;clr user enabled exc 76 clr.l d1 77 move.w FPTEMP(a6),d1 ;get dest exponent 78 smi L_SCR1(a6) ;use L_SCR1 to hold sign 79 andi.l #$7fff,d1 ;strip sign 80 move.w ETEMP(a6),d0 ;check src bounds 81 andi.w #$7fff,d0 ;clr sign bit 82 cmp2.w SRC_BNDS,d0 83 bcc.b src_in 84 cmpi.w #$400c,d0 ;test for too large 85 bge.w src_out 86 * 87 * The source input is below 1, so we check for denormalized numbers 88 * and set unfl. 89 * 90 src_small: 91 move.b DTAG(a6),d0 92 andi.b #$e0,d0 93 tst.b d0 94 beq.b no_denorm 95 st STORE_FLG(a6) ;dest already contains result 96 or.l #unfl_mask,USER_FPSR(a6) ;set UNFL 97 den_done: 98 lea.l FPTEMP(a6),a0 99 bra t_resdnrm 100 no_denorm: 101 fmove.l USER_FPCR(a6),FPCR 102 fmove.x FPTEMP(a6),fp0 ;simply return dest 103 rts 104 105 106 * 107 * Source is within 2^14 range. To perform the int operation, 108 * move it to d0. 109 * 110 src_in: 111 fmove.x ETEMP(a6),fp0 ;move in src for int 112 fmove.l #rz_mode,fpcr ;force rz for src conversion 113 fmove.l fp0,d0 ;int src to d0 114 fmove.l #0,FPSR ;clr status from above 115 tst.w ETEMP(a6) ;check src sign 116 blt.w src_neg 117 * 118 * Source is positive. Add the src to the dest exponent. 119 * The result can be denormalized, if src = 0, or overflow, 120 * if the result of the add sets a bit in the upper word. 121 * 122 src_pos: 123 tst.w d1 ;check for denorm 124 beq.w dst_dnrm 125 add.l d0,d1 ;add src to dest exp 126 beq.b denorm ;if zero, result is denorm 127 cmpi.l #$7fff,d1 ;test for overflow 128 bge.b ovfl 129 tst.b L_SCR1(a6) 130 beq.b spos_pos 131 or.w #$8000,d1 132 spos_pos: 133 move.w d1,FPTEMP(a6) ;result in FPTEMP 134 fmove.l USER_FPCR(a6),FPCR 135 fmove.x FPTEMP(a6),fp0 ;write result to fp0 136 rts 137 ovfl: 138 tst.b L_SCR1(a6) 139 beq.b sovl_pos 140 or.w #$8000,d1 141 sovl_pos: 142 move.w FPTEMP(a6),ETEMP(a6) ;result in ETEMP 143 move.l FPTEMP_HI(a6),ETEMP_HI(a6) 144 move.l FPTEMP_LO(a6),ETEMP_LO(a6) 145 bra t_ovfl2 146 147 denorm: 148 tst.b L_SCR1(a6) 149 beq.b den_pos 150 or.w #$8000,d1 151 den_pos: 152 tst.l FPTEMP_HI(a6) ;check j bit 153 blt.b nden_exit ;if set, not denorm 154 move.w d1,ETEMP(a6) ;input expected in ETEMP 155 move.l FPTEMP_HI(a6),ETEMP_HI(a6) 156 move.l FPTEMP_LO(a6),ETEMP_LO(a6) 157 or.l #unfl_bit,USER_FPSR(a6) ;set unfl 158 lea.l ETEMP(a6),a0 159 bra t_resdnrm 160 nden_exit: 161 move.w d1,FPTEMP(a6) ;result in FPTEMP 162 fmove.l USER_FPCR(a6),FPCR 163 fmove.x FPTEMP(a6),fp0 ;write result to fp0 164 rts 165 166 * 167 * Source is negative. Add the src to the dest exponent. 168 * (The result exponent will be reduced). The result can be 169 * denormalized. 170 * 171 src_neg: 172 add.l d0,d1 ;add src to dest 173 beq.b denorm ;if zero, result is denorm 174 blt.b fix_dnrm ;if negative, result is 175 * ;needing denormalization 176 tst.b L_SCR1(a6) 177 beq.b sneg_pos 178 or.w #$8000,d1 179 sneg_pos: 180 move.w d1,FPTEMP(a6) ;result in FPTEMP 181 fmove.l USER_FPCR(a6),FPCR 182 fmove.x FPTEMP(a6),fp0 ;write result to fp0 183 rts 184 185 186 * 187 * The result exponent is below denorm value. Test for catastrophic 188 * underflow and force zero if true. If not, try to shift the 189 * mantissa right until a zero exponent exists. 190 * 191 fix_dnrm: 192 cmpi.w #$ffc0,d1 ;lower bound for normalization 193 blt.w fix_unfl ;if lower, catastrophic unfl 194 move.w d1,d0 ;use d0 for exp 195 move.l d2,-(a7) ;free d2 for norm 196 move.l FPTEMP_HI(a6),d1 197 move.l FPTEMP_LO(a6),d2 198 clr.l L_SCR2(a6) 199 fix_loop: 200 add.w #1,d0 ;drive d0 to 0 201 lsr.l #1,d1 ;while shifting the 202 roxr.l #1,d2 ;mantissa to the right 203 bcc.b no_carry 204 st L_SCR2(a6) ;use L_SCR2 to capture inex 205 no_carry: 206 tst.w d0 ;it is finished when 207 blt.b fix_loop ;d0 is zero or the mantissa 208 tst.b L_SCR2(a6) 209 beq.b tst_zero 210 or.l #unfl_inx_mask,USER_FPSR(a6) 211 * ;set unfl, aunfl, ainex 212 * 213 * Test for zero. If zero, simply use fmove to return +/- zero 214 * to the fpu. 215 * 216 tst_zero: 217 clr.w FPTEMP_EX(a6) 218 tst.b L_SCR1(a6) ;test for sign 219 beq.b tst_con 220 or.w #$8000,FPTEMP_EX(a6) ;set sign bit 221 tst_con: 222 move.l d1,FPTEMP_HI(a6) 223 move.l d2,FPTEMP_LO(a6) 224 move.l (a7)+,d2 225 tst.l d1 226 bne.b not_zero 227 tst.l FPTEMP_LO(a6) 228 bne.b not_zero 229 * 230 * Result is zero. Check for rounding mode to set lsb. If the 231 * mode is rp, and the zero is positive, return smallest denorm. 232 * If the mode is rm, and the zero is negative, return smallest 233 * negative denorm. 234 * 235 btst.b #5,FPCR_MODE(a6) ;test if rm or rp 236 beq.b no_dir 237 btst.b #4,FPCR_MODE(a6) ;check which one 238 beq.b zer_rm 239 zer_rp: 240 tst.b L_SCR1(a6) ;check sign 241 bne.b no_dir ;if set, neg op, no inc 242 move.l #1,FPTEMP_LO(a6) ;set lsb 243 bra.b sm_dnrm 244 zer_rm: 245 tst.b L_SCR1(a6) ;check sign 246 beq.b no_dir ;if clr, neg op, no inc 247 move.l #1,FPTEMP_LO(a6) ;set lsb 248 or.l #neg_mask,USER_FPSR(a6) ;set N 249 bra.b sm_dnrm 250 no_dir: 251 fmove.l USER_FPCR(a6),FPCR 252 fmove.x FPTEMP(a6),fp0 ;use fmove to set cc's 253 rts 254 255 * 256 * The rounding mode changed the zero to a smallest denorm. Call 257 * t_resdnrm with exceptional operand in ETEMP. 258 * 259 sm_dnrm: 260 move.l FPTEMP_EX(a6),ETEMP_EX(a6) 261 move.l FPTEMP_HI(a6),ETEMP_HI(a6) 262 move.l FPTEMP_LO(a6),ETEMP_LO(a6) 263 lea.l ETEMP(a6),a0 264 bra t_resdnrm 265 266 * 267 * Result is still denormalized. 268 * 269 not_zero: 270 or.l #unfl_mask,USER_FPSR(a6) ;set unfl 271 tst.b L_SCR1(a6) ;check for sign 272 beq.b fix_exit 273 or.l #neg_mask,USER_FPSR(a6) ;set N 274 fix_exit: 275 bra.b sm_dnrm 276 277 278 * 279 * The result has underflowed to zero. Return zero and set 280 * unfl, aunfl, and ainex. 281 * 282 fix_unfl: 283 or.l #unfl_inx_mask,USER_FPSR(a6) 284 btst.b #5,FPCR_MODE(a6) ;test if rm or rp 285 beq.b no_dir2 286 btst.b #4,FPCR_MODE(a6) ;check which one 287 beq.b zer_rm2 288 zer_rp2: 289 tst.b L_SCR1(a6) ;check sign 290 bne.b no_dir2 ;if set, neg op, no inc 291 clr.l FPTEMP_EX(a6) 292 clr.l FPTEMP_HI(a6) 293 move.l #1,FPTEMP_LO(a6) ;set lsb 294 bra.b sm_dnrm ;return smallest denorm 295 zer_rm2: 296 tst.b L_SCR1(a6) ;check sign 297 beq.b no_dir2 ;if clr, neg op, no inc 298 move.w #$8000,FPTEMP_EX(a6) 299 clr.l FPTEMP_HI(a6) 300 move.l #1,FPTEMP_LO(a6) ;set lsb 301 or.l #neg_mask,USER_FPSR(a6) ;set N 302 bra.w sm_dnrm ;return smallest denorm 303 304 no_dir2: 305 tst.b L_SCR1(a6) 306 bge.b pos_zero 307 neg_zero: 308 clr.l FP_SCR1(a6) ;clear the exceptional operand 309 clr.l FP_SCR1+4(a6) ;for gen_except. 310 clr.l FP_SCR1+8(a6) 311 fmove.s #:80000000,fp0 312 rts 313 pos_zero: 314 clr.l FP_SCR1(a6) ;clear the exceptional operand 315 clr.l FP_SCR1+4(a6) ;for gen_except. 316 clr.l FP_SCR1+8(a6) 317 fmove.s #:00000000,fp0 318 rts 319 320 * 321 * The destination is a denormalized number. It must be handled 322 * by first shifting the bits in the mantissa until it is normalized, 323 * then adding the remainder of the source to the exponent. 324 * 325 dst_dnrm: 326 movem.l d2/d3,-(a7) 327 move.w FPTEMP_EX(a6),d1 328 move.l FPTEMP_HI(a6),d2 329 move.l FPTEMP_LO(a6),d3 330 dst_loop: 331 tst.l d2 ;test for normalized result 332 blt.b dst_norm ;exit loop if so 333 tst.l d0 ;otherwise, test shift count 334 beq.b dst_fin ;if zero, shifting is done 335 subq.l #1,d0 ;dec src 336 add.l d3,d3 337 addx.l d2,d2 338 bra.b dst_loop 339 * 340 * Destination became normalized. Simply add the remaining 341 * portion of the src to the exponent. 342 * 343 dst_norm: 344 add.w d0,d1 ;dst is normalized; add src 345 tst.b L_SCR1(a6) 346 beq.b dnrm_pos 347 or.w #$8000,d1 348 dnrm_pos: 349 movem.w d1,FPTEMP_EX(a6) 350 movem.l d2,FPTEMP_HI(a6) 351 movem.l d3,FPTEMP_LO(a6) 352 fmove.l USER_FPCR(a6),FPCR 353 fmove.x FPTEMP(a6),fp0 354 movem.l (a7)+,d2/d3 355 rts 356 357 * 358 * Destination remained denormalized. Call t_excdnrm with 359 * exceptional operand in ETEMP. 360 * 361 dst_fin: 362 tst.b L_SCR1(a6) ;check for sign 363 beq.b dst_exit 364 or.l #neg_mask,USER_FPSR(a6) ;set N 365 or.w #$8000,d1 366 dst_exit: 367 movem.w d1,ETEMP_EX(a6) 368 movem.l d2,ETEMP_HI(a6) 369 movem.l d3,ETEMP_LO(a6) 370 or.l #unfl_mask,USER_FPSR(a6) ;set unfl 371 movem.l (a7)+,d2/d3 372 lea.l ETEMP(a6),a0 373 bra t_resdnrm 374 375 * 376 * Source is outside of 2^14 range. Test the sign and branch 377 * to the appropriate exception handler. 378 * 379 src_out: 380 tst.b L_SCR1(a6) 381 beq.b scro_pos 382 or.w #$8000,d1 383 scro_pos: 384 move.l FPTEMP_HI(a6),ETEMP_HI(a6) 385 move.l FPTEMP_LO(a6),ETEMP_LO(a6) 386 tst.w ETEMP(a6) 387 blt.b res_neg 388 res_pos: 389 move.w d1,ETEMP(a6) ;result in ETEMP 390 bra t_ovfl2 391 res_neg: 392 move.w d1,ETEMP(a6) ;result in ETEMP 393 lea.l ETEMP(a6),a0 394 bra t_unfl 395 end 396
Indexes created Tue Sep 30 11:09:46 GMT 2025