1| 2| do_func.sa 3.4 2/18/91 3| 4| Do_func performs the unimplemented operation. The operation 5| to be performed is determined from the lower 7 bits of the 6| extension word (except in the case of fmovecr and fsincos). 7| The opcode and tag bits form an index into a jump table in 8| tbldo.sa. Cases of zero, infinity and NaN are handled in 9| do_func by forcing the default result. Normalized and 10| denormalized (there are no unnormalized numbers at this 11| point) are passed onto the emulation code. 12| 13| CMDREG1B and STAG are extracted from the fsave frame 14| and combined to form the table index. The function called 15| will start with a0 pointing to the ETEMP operand. Dyadic 16| functions can find FPTEMP at -12(a0). 17| 18| Called functions return their result in fp0. Sincos returns 19| sin(x) in fp0 and cos(x) in fp1. 20| 21 22| Copyright (C) Motorola, Inc. 1990 23| All Rights Reserved 24| 25| THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA 26| The copyright notice above does not evidence any 27| actual or intended publication of such source code. 28 29DO_FUNC: |idnt 2,1 | Motorola 040 Floating Point Software Package 30 31 |section 8 32 33 .include "fpsp.h" 34 35 |xref t_dz2 36 |xref t_operr 37 |xref t_inx2 38 |xref t_resdnrm 39 |xref dst_nan 40 |xref src_nan 41 |xref nrm_set 42 |xref sto_cos 43 44 |xref tblpre 45 |xref slognp1,slogn,slog10,slog2 46 |xref slognd,slog10d,slog2d 47 |xref smod,srem 48 |xref sscale 49 |xref smovcr 50 51PONE: .long 0x3fff0000,0x80000000,0x00000000 |+1 52MONE: .long 0xbfff0000,0x80000000,0x00000000 |-1 53PZERO: .long 0x00000000,0x00000000,0x00000000 |+0 54MZERO: .long 0x80000000,0x00000000,0x00000000 |-0 55PINF: .long 0x7fff0000,0x00000000,0x00000000 |+inf 56MINF: .long 0xffff0000,0x00000000,0x00000000 |-inf 57QNAN: .long 0x7fff0000,0xffffffff,0xffffffff |non-signaling nan 58PPIBY2: .long 0x3FFF0000,0xC90FDAA2,0x2168C235 |+PI/2 59MPIBY2: .long 0xbFFF0000,0xC90FDAA2,0x2168C235 |-PI/2 60 61 .global do_func 62do_func: 63 clrb CU_ONLY(%a6) 64| 65| Check for fmovecr. It does not follow the format of fp gen 66| unimplemented instructions. The test is on the upper 6 bits; 67| if they are $17, the inst is fmovecr. Call entry smovcr 68| directly. 69| 70 bfextu CMDREG1B(%a6){#0:#6},%d0 |get opclass and src fields 71 cmpil #0x17,%d0 |if op class and size fields are $17, 72| ;it is FMOVECR; if not, continue 73 bnes not_fmovecr 74 jmp smovcr |fmovecr; jmp directly to emulation 75 76not_fmovecr: 77 movew CMDREG1B(%a6),%d0 78 andl #0x7F,%d0 79 cmpil #0x38,%d0 |if the extension is >= $38, 80 bge serror |it is illegal 81 bfextu STAG(%a6){#0:#3},%d1 82 lsll #3,%d0 |make room for STAG 83 addl %d1,%d0 |combine for final index into table 84 leal tblpre,%a1 |start of monster jump table 85 movel (%a1,%d0.w*4),%a1 |real target address 86 leal ETEMP(%a6),%a0 |a0 is pointer to src op 87 movel USER_FPCR(%a6),%d1 88 andl #0xFF,%d1 | discard all but rounding mode/prec 89 fmovel #0,%fpcr 90 jmp (%a1) 91| 92| ERROR 93| 94 .global serror 95serror: 96 st STORE_FLG(%a6) 97 rts 98| 99| These routines load forced values into fp0. They are called 100| by index into tbldo. 101| 102| Load a signed zero to fp0 and set inex2/ainex 103| 104 .global snzrinx 105snzrinx: 106 btstb #sign_bit,LOCAL_EX(%a0) |get sign of source operand 107 bnes ld_mzinx |if negative, branch 108 bsr ld_pzero |bsr so we can return and set inx 109 bra t_inx2 |now, set the inx for the next inst 110ld_mzinx: 111 bsr ld_mzero |if neg, load neg zero, return here 112 bra t_inx2 |now, set the inx for the next inst 113| 114| Load a signed zero to fp0; do not set inex2/ainex 115| 116 .global szero 117szero: 118 btstb #sign_bit,LOCAL_EX(%a0) |get sign of source operand 119 bne ld_mzero |if neg, load neg zero 120 bra ld_pzero |load positive zero 121| 122| Load a signed infinity to fp0; do not set inex2/ainex 123| 124 .global sinf 125sinf: 126 btstb #sign_bit,LOCAL_EX(%a0) |get sign of source operand 127 bne ld_minf |if negative branch 128 bra ld_pinf 129| 130| Load a signed one to fp0; do not set inex2/ainex 131| 132 .global sone 133sone: 134 btstb #sign_bit,LOCAL_EX(%a0) |check sign of source 135 bne ld_mone 136 bra ld_pone 137| 138| Load a signed pi/2 to fp0; do not set inex2/ainex 139| 140 .global spi_2 141spi_2: 142 btstb #sign_bit,LOCAL_EX(%a0) |check sign of source 143 bne ld_mpi2 144 bra ld_ppi2 145| 146| Load either a +0 or +inf for plus/minus operand 147| 148 .global szr_inf 149szr_inf: 150 btstb #sign_bit,LOCAL_EX(%a0) |check sign of source 151 bne ld_pzero 152 bra ld_pinf 153| 154| Result is either an operr or +inf for plus/minus operand 155| [Used by slogn, slognp1, slog10, and slog2] 156| 157 .global sopr_inf 158sopr_inf: 159 btstb #sign_bit,LOCAL_EX(%a0) |check sign of source 160 bne t_operr 161 bra ld_pinf 162| 163| FLOGNP1 164| 165 .global sslognp1 166sslognp1: 167 fmovemx (%a0),%fp0-%fp0 168 fcmpb #-1,%fp0 169 fbgt slognp1 170 fbeq t_dz2 |if = -1, divide by zero exception 171 fmovel #0,%FPSR |clr N flag 172 bra t_operr |take care of operands < -1 173| 174| FETOXM1 175| 176 .global setoxm1i 177setoxm1i: 178 btstb #sign_bit,LOCAL_EX(%a0) |check sign of source 179 bne ld_mone 180 bra ld_pinf 181| 182| FLOGN 183| 184| Test for 1.0 as an input argument, returning +zero. Also check 185| the sign and return operr if negative. 186| 187 .global sslogn 188sslogn: 189 btstb #sign_bit,LOCAL_EX(%a0) 190 bne t_operr |take care of operands < 0 191 cmpiw #0x3fff,LOCAL_EX(%a0) |test for 1.0 input 192 bne slogn 193 cmpil #0x80000000,LOCAL_HI(%a0) 194 bne slogn 195 tstl LOCAL_LO(%a0) 196 bne slogn 197 fmovex PZERO,%fp0 198 rts 199 200 .global sslognd 201sslognd: 202 btstb #sign_bit,LOCAL_EX(%a0) 203 beq slognd 204 bra t_operr |take care of operands < 0 205 206| 207| FLOG10 208| 209 .global sslog10 210sslog10: 211 btstb #sign_bit,LOCAL_EX(%a0) 212 bne t_operr |take care of operands < 0 213 cmpiw #0x3fff,LOCAL_EX(%a0) |test for 1.0 input 214 bne slog10 215 cmpil #0x80000000,LOCAL_HI(%a0) 216 bne slog10 217 tstl LOCAL_LO(%a0) 218 bne slog10 219 fmovex PZERO,%fp0 220 rts 221 222 .global sslog10d 223sslog10d: 224 btstb #sign_bit,LOCAL_EX(%a0) 225 beq slog10d 226 bra t_operr |take care of operands < 0 227 228| 229| FLOG2 230| 231 .global sslog2 232sslog2: 233 btstb #sign_bit,LOCAL_EX(%a0) 234 bne t_operr |take care of operands < 0 235 cmpiw #0x3fff,LOCAL_EX(%a0) |test for 1.0 input 236 bne slog2 237 cmpil #0x80000000,LOCAL_HI(%a0) 238 bne slog2 239 tstl LOCAL_LO(%a0) 240 bne slog2 241 fmovex PZERO,%fp0 242 rts 243 244 .global sslog2d 245sslog2d: 246 btstb #sign_bit,LOCAL_EX(%a0) 247 beq slog2d 248 bra t_operr |take care of operands < 0 249 250| 251| FMOD 252| 253pmodt: 254| ;$21 fmod 255| ;dtag,stag 256 .long smod | 00,00 norm,norm = normal 257 .long smod_oper | 00,01 norm,zero = nan with operr 258 .long smod_fpn | 00,10 norm,inf = fpn 259 .long smod_snan | 00,11 norm,nan = nan 260 .long smod_zro | 01,00 zero,norm = +-zero 261 .long smod_oper | 01,01 zero,zero = nan with operr 262 .long smod_zro | 01,10 zero,inf = +-zero 263 .long smod_snan | 01,11 zero,nan = nan 264 .long smod_oper | 10,00 inf,norm = nan with operr 265 .long smod_oper | 10,01 inf,zero = nan with operr 266 .long smod_oper | 10,10 inf,inf = nan with operr 267 .long smod_snan | 10,11 inf,nan = nan 268 .long smod_dnan | 11,00 nan,norm = nan 269 .long smod_dnan | 11,01 nan,zero = nan 270 .long smod_dnan | 11,10 nan,inf = nan 271 .long smod_dnan | 11,11 nan,nan = nan 272 273 .global pmod 274pmod: 275 clrb FPSR_QBYTE(%a6) | clear quotient field 276 bfextu STAG(%a6){#0:#3},%d0 |stag = d0 277 bfextu DTAG(%a6){#0:#3},%d1 |dtag = d1 278 279| 280| Alias extended denorms to norms for the jump table. 281| 282 bclrl #2,%d0 283 bclrl #2,%d1 284 285 lslb #2,%d1 286 orb %d0,%d1 |d1{3:2} = dtag, d1{1:0} = stag 287| ;Tag values: 288| ;00 = norm or denorm 289| ;01 = zero 290| ;10 = inf 291| ;11 = nan 292 lea pmodt,%a1 293 movel (%a1,%d1.w*4),%a1 294 jmp (%a1) 295 296smod_snan: 297 bra src_nan 298smod_dnan: 299 bra dst_nan 300smod_oper: 301 bra t_operr 302smod_zro: 303 moveb ETEMP(%a6),%d1 |get sign of src op 304 moveb FPTEMP(%a6),%d0 |get sign of dst op 305 eorb %d0,%d1 |get exor of sign bits 306 btstl #7,%d1 |test for sign 307 beqs smod_zsn |if clr, do not set sign big 308 bsetb #q_sn_bit,FPSR_QBYTE(%a6) |set q-byte sign bit 309smod_zsn: 310 btstl #7,%d0 |test if + or - 311 beq ld_pzero |if pos then load +0 312 bra ld_mzero |else neg load -0 313 314smod_fpn: 315 moveb ETEMP(%a6),%d1 |get sign of src op 316 moveb FPTEMP(%a6),%d0 |get sign of dst op 317 eorb %d0,%d1 |get exor of sign bits 318 btstl #7,%d1 |test for sign 319 beqs smod_fsn |if clr, do not set sign big 320 bsetb #q_sn_bit,FPSR_QBYTE(%a6) |set q-byte sign bit 321smod_fsn: 322 tstb DTAG(%a6) |filter out denormal destination case 323 bpls smod_nrm | 324 leal FPTEMP(%a6),%a0 |a0<- addr(FPTEMP) 325 bra t_resdnrm |force UNFL(but exact) result 326smod_nrm: 327 fmovel USER_FPCR(%a6),%fpcr |use user's rmode and precision 328 fmovex FPTEMP(%a6),%fp0 |return dest to fp0 329 rts 330 331| 332| FREM 333| 334premt: 335| ;$25 frem 336| ;dtag,stag 337 .long srem | 00,00 norm,norm = normal 338 .long srem_oper | 00,01 norm,zero = nan with operr 339 .long srem_fpn | 00,10 norm,inf = fpn 340 .long srem_snan | 00,11 norm,nan = nan 341 .long srem_zro | 01,00 zero,norm = +-zero 342 .long srem_oper | 01,01 zero,zero = nan with operr 343 .long srem_zro | 01,10 zero,inf = +-zero 344 .long srem_snan | 01,11 zero,nan = nan 345 .long srem_oper | 10,00 inf,norm = nan with operr 346 .long srem_oper | 10,01 inf,zero = nan with operr 347 .long srem_oper | 10,10 inf,inf = nan with operr 348 .long srem_snan | 10,11 inf,nan = nan 349 .long srem_dnan | 11,00 nan,norm = nan 350 .long srem_dnan | 11,01 nan,zero = nan 351 .long srem_dnan | 11,10 nan,inf = nan 352 .long srem_dnan | 11,11 nan,nan = nan 353 354 .global prem 355prem: 356 clrb FPSR_QBYTE(%a6) |clear quotient field 357 bfextu STAG(%a6){#0:#3},%d0 |stag = d0 358 bfextu DTAG(%a6){#0:#3},%d1 |dtag = d1 359| 360| Alias extended denorms to norms for the jump table. 361| 362 bclr #2,%d0 363 bclr #2,%d1 364 365 lslb #2,%d1 366 orb %d0,%d1 |d1{3:2} = dtag, d1{1:0} = stag 367| ;Tag values: 368| ;00 = norm or denorm 369| ;01 = zero 370| ;10 = inf 371| ;11 = nan 372 lea premt,%a1 373 movel (%a1,%d1.w*4),%a1 374 jmp (%a1) 375 376srem_snan: 377 bra src_nan 378srem_dnan: 379 bra dst_nan 380srem_oper: 381 bra t_operr 382srem_zro: 383 moveb ETEMP(%a6),%d1 |get sign of src op 384 moveb FPTEMP(%a6),%d0 |get sign of dst op 385 eorb %d0,%d1 |get exor of sign bits 386 btstl #7,%d1 |test for sign 387 beqs srem_zsn |if clr, do not set sign big 388 bsetb #q_sn_bit,FPSR_QBYTE(%a6) |set q-byte sign bit 389srem_zsn: 390 btstl #7,%d0 |test if + or - 391 beq ld_pzero |if pos then load +0 392 bra ld_mzero |else neg load -0 393 394srem_fpn: 395 moveb ETEMP(%a6),%d1 |get sign of src op 396 moveb FPTEMP(%a6),%d0 |get sign of dst op 397 eorb %d0,%d1 |get exor of sign bits 398 btstl #7,%d1 |test for sign 399 beqs srem_fsn |if clr, do not set sign big 400 bsetb #q_sn_bit,FPSR_QBYTE(%a6) |set q-byte sign bit 401srem_fsn: 402 tstb DTAG(%a6) |filter out denormal destination case 403 bpls srem_nrm | 404 leal FPTEMP(%a6),%a0 |a0<- addr(FPTEMP) 405 bra t_resdnrm |force UNFL(but exact) result 406srem_nrm: 407 fmovel USER_FPCR(%a6),%fpcr |use user's rmode and precision 408 fmovex FPTEMP(%a6),%fp0 |return dest to fp0 409 rts 410| 411| FSCALE 412| 413pscalet: 414| ;$26 fscale 415| ;dtag,stag 416 .long sscale | 00,00 norm,norm = result 417 .long sscale | 00,01 norm,zero = fpn 418 .long scl_opr | 00,10 norm,inf = nan with operr 419 .long scl_snan | 00,11 norm,nan = nan 420 .long scl_zro | 01,00 zero,norm = +-zero 421 .long scl_zro | 01,01 zero,zero = +-zero 422 .long scl_opr | 01,10 zero,inf = nan with operr 423 .long scl_snan | 01,11 zero,nan = nan 424 .long scl_inf | 10,00 inf,norm = +-inf 425 .long scl_inf | 10,01 inf,zero = +-inf 426 .long scl_opr | 10,10 inf,inf = nan with operr 427 .long scl_snan | 10,11 inf,nan = nan 428 .long scl_dnan | 11,00 nan,norm = nan 429 .long scl_dnan | 11,01 nan,zero = nan 430 .long scl_dnan | 11,10 nan,inf = nan 431 .long scl_dnan | 11,11 nan,nan = nan 432 433 .global pscale 434pscale: 435 bfextu STAG(%a6){#0:#3},%d0 |stag in d0 436 bfextu DTAG(%a6){#0:#3},%d1 |dtag in d1 437 bclrl #2,%d0 |alias denorm into norm 438 bclrl #2,%d1 |alias denorm into norm 439 lslb #2,%d1 440 orb %d0,%d1 |d1{4:2} = dtag, d1{1:0} = stag 441| ;dtag values stag values: 442| ;000 = norm 00 = norm 443| ;001 = zero 01 = zero 444| ;010 = inf 10 = inf 445| ;011 = nan 11 = nan 446| ;100 = dnrm 447| 448| 449 leal pscalet,%a1 |load start of jump table 450 movel (%a1,%d1.w*4),%a1 |load a1 with label depending on tag 451 jmp (%a1) |go to the routine 452 453scl_opr: 454 bra t_operr 455 456scl_dnan: 457 bra dst_nan 458 459scl_zro: 460 btstb #sign_bit,FPTEMP_EX(%a6) |test if + or - 461 beq ld_pzero |if pos then load +0 462 bra ld_mzero |if neg then load -0 463scl_inf: 464 btstb #sign_bit,FPTEMP_EX(%a6) |test if + or - 465 beq ld_pinf |if pos then load +inf 466 bra ld_minf |else neg load -inf 467scl_snan: 468 bra src_nan 469| 470| FSINCOS 471| 472 .global ssincosz 473ssincosz: 474 btstb #sign_bit,ETEMP(%a6) |get sign 475 beqs sincosp 476 fmovex MZERO,%fp0 477 bras sincoscom 478sincosp: 479 fmovex PZERO,%fp0 480sincoscom: 481 fmovemx PONE,%fp1-%fp1 |do not allow FPSR to be affected 482 bra sto_cos |store cosine result 483 484 .global ssincosi 485ssincosi: 486 fmovex QNAN,%fp1 |load NAN 487 bsr sto_cos |store cosine result 488 fmovex QNAN,%fp0 |load NAN 489 bra t_operr 490 491 .global ssincosnan 492ssincosnan: 493 movel ETEMP_EX(%a6),FP_SCR1(%a6) 494 movel ETEMP_HI(%a6),FP_SCR1+4(%a6) 495 movel ETEMP_LO(%a6),FP_SCR1+8(%a6) 496 bsetb #signan_bit,FP_SCR1+4(%a6) 497 fmovemx FP_SCR1(%a6),%fp1-%fp1 498 bsr sto_cos 499 bra src_nan 500| 501| This code forces default values for the zero, inf, and nan cases 502| in the transcendentals code. The CC bits must be set in the 503| stacked FPSR to be correctly reported. 504| 505|**Returns +PI/2 506 .global ld_ppi2 507ld_ppi2: 508 fmovex PPIBY2,%fp0 |load +pi/2 509 bra t_inx2 |set inex2 exc 510 511|**Returns -PI/2 512 .global ld_mpi2 513ld_mpi2: 514 fmovex MPIBY2,%fp0 |load -pi/2 515 orl #neg_mask,USER_FPSR(%a6) |set N bit 516 bra t_inx2 |set inex2 exc 517 518|**Returns +inf 519 .global ld_pinf 520ld_pinf: 521 fmovex PINF,%fp0 |load +inf 522 orl #inf_mask,USER_FPSR(%a6) |set I bit 523 rts 524 525|**Returns -inf 526 .global ld_minf 527ld_minf: 528 fmovex MINF,%fp0 |load -inf 529 orl #neg_mask+inf_mask,USER_FPSR(%a6) |set N and I bits 530 rts 531 532|**Returns +1 533 .global ld_pone 534ld_pone: 535 fmovex PONE,%fp0 |load +1 536 rts 537 538|**Returns -1 539 .global ld_mone 540ld_mone: 541 fmovex MONE,%fp0 |load -1 542 orl #neg_mask,USER_FPSR(%a6) |set N bit 543 rts 544 545|**Returns +0 546 .global ld_pzero 547ld_pzero: 548 fmovex PZERO,%fp0 |load +0 549 orl #z_mask,USER_FPSR(%a6) |set Z bit 550 rts 551 552|**Returns -0 553 .global ld_mzero 554ld_mzero: 555 fmovex MZERO,%fp0 |load -0 556 orl #neg_mask+z_mask,USER_FPSR(%a6) |set N and Z bits 557 rts 558 559 |end 560