| 1 |
/* libgcc routines for 68000 w/o floating-point hardware. |
| 2 |
Copyright (C) 1994-2025 Free Software Foundation, Inc. |
| 3 |
|
| 4 |
This file is part of GCC. |
| 5 |
|
| 6 |
GCC is free software; you can redistribute it and/or modify it |
| 7 |
under the terms of the GNU General Public License as published by the |
| 8 |
Free Software Foundation; either version 3, or (at your option) any |
| 9 |
later version. |
| 10 |
|
| 11 |
This file is distributed in the hope that it will be useful, but |
| 12 |
WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 |
General Public License for more details. |
| 15 |
|
| 16 |
Under Section 7 of GPL version 3, you are granted additional |
| 17 |
permissions described in the GCC Runtime Library Exception, version |
| 18 |
3.1, as published by the Free Software Foundation. |
| 19 |
|
| 20 |
You should have received a copy of the GNU General Public License and |
| 21 |
a copy of the GCC Runtime Library Exception along with this program; |
| 22 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 23 |
<http://www.gnu.org/licenses/>. */ |
| 24 |
|
| 25 |
/* Use this one for any 680x0; assumes no floating point hardware. |
| 26 |
The trailing " '" appearing on some lines is for ANSI preprocessors. Yuk. |
| 27 |
Some of this code comes from MINIX, via the folks at ericsson. |
| 28 |
D. V. Henkel-Wallace ([email protected]) Fete Bastille, 1992 |
| 29 |
*/ |
| 30 |
|
| 31 |
/* These are predefined by new versions of GNU cpp. */ |
| 32 |
|
| 33 |
#ifndef __USER_LABEL_PREFIX__ |
| 34 |
#define __USER_LABEL_PREFIX__ _ |
| 35 |
#endif |
| 36 |
|
| 37 |
#ifndef __REGISTER_PREFIX__ |
| 38 |
#define __REGISTER_PREFIX__ |
| 39 |
#endif |
| 40 |
|
| 41 |
#ifndef __IMMEDIATE_PREFIX__ |
| 42 |
#define __IMMEDIATE_PREFIX__ # |
| 43 |
#endif |
| 44 |
|
| 45 |
/* ANSI concatenation macros. */ |
| 46 |
|
| 47 |
#define CONCAT1(a, b) CONCAT2(a, b) |
| 48 |
#define CONCAT2(a, b) a ## b |
| 49 |
|
| 50 |
/* Use the right prefix for global labels. */ |
| 51 |
|
| 52 |
#define SYM(x) CONCAT1 (__USER_LABEL_PREFIX__, x) |
| 53 |
|
| 54 |
/* Note that X is a function. */ |
| 55 |
|
| 56 |
#ifdef __ELF__ |
| 57 |
#define FUNC(x) .type SYM(x),function |
| 58 |
#else |
| 59 |
/* The .proc pseudo-op is accepted, but ignored, by GAS. We could just |
| 60 |
define this to the empty string for non-ELF systems, but defining it |
| 61 |
to .proc means that the information is available to the assembler if |
| 62 |
the need arises. */ |
| 63 |
#define FUNC(x) .proc |
| 64 |
#endif |
| 65 |
|
| 66 |
/* Use the right prefix for registers. */ |
| 67 |
|
| 68 |
#define REG(x) CONCAT1 (__REGISTER_PREFIX__, x) |
| 69 |
|
| 70 |
/* Use the right prefix for immediate values. */ |
| 71 |
|
| 72 |
#define IMM(x) CONCAT1 (__IMMEDIATE_PREFIX__, x) |
| 73 |
|
| 74 |
#define d0 REG (d0) |
| 75 |
#define d1 REG (d1) |
| 76 |
#define d2 REG (d2) |
| 77 |
#define d3 REG (d3) |
| 78 |
#define d4 REG (d4) |
| 79 |
#define d5 REG (d5) |
| 80 |
#define d6 REG (d6) |
| 81 |
#define d7 REG (d7) |
| 82 |
#define a0 REG (a0) |
| 83 |
#define a1 REG (a1) |
| 84 |
#define a2 REG (a2) |
| 85 |
#define a3 REG (a3) |
| 86 |
#define a4 REG (a4) |
| 87 |
#define a5 REG (a5) |
| 88 |
#define a6 REG (a6) |
| 89 |
#define fp REG (fp) |
| 90 |
#define sp REG (sp) |
| 91 |
#define pc REG (pc) |
| 92 |
|
| 93 |
/* Provide a few macros to allow for PIC code support. |
| 94 |
* With PIC, data is stored A5 relative so we've got to take a bit of special |
| 95 |
* care to ensure that all loads of global data is via A5. PIC also requires |
| 96 |
* jumps and subroutine calls to be PC relative rather than absolute. We cheat |
| 97 |
* a little on this and in the PIC case, we use short offset branches and |
| 98 |
* hope that the final object code is within range (which it should be). |
| 99 |
*/ |
| 100 |
#ifndef __PIC__ |
| 101 |
|
| 102 |
/* Non PIC (absolute/relocatable) versions */ |
| 103 |
|
| 104 |
.macro PICCALL addr |
| 105 |
jbsr \addr |
| 106 |
.endm |
| 107 |
|
| 108 |
.macro PICJUMP addr |
| 109 |
jmp \addr |
| 110 |
.endm |
| 111 |
|
| 112 |
.macro PICLEA sym, reg |
| 113 |
lea \sym, \reg |
| 114 |
.endm |
| 115 |
|
| 116 |
.macro PICPEA sym, areg |
| 117 |
pea \sym |
| 118 |
.endm |
| 119 |
|
| 120 |
#else /* __PIC__ */ |
| 121 |
|
| 122 |
# if defined (__uClinux__) |
| 123 |
|
| 124 |
/* Versions for uClinux */ |
| 125 |
|
| 126 |
# if defined(__ID_SHARED_LIBRARY__) |
| 127 |
|
| 128 |
/* -mid-shared-library versions */ |
| 129 |
|
| 130 |
.macro PICLEA sym, reg |
| 131 |
movel a5@(_current_shared_library_a5_offset_), \reg |
| 132 |
movel \sym@GOT(\reg), \reg |
| 133 |
.endm |
| 134 |
|
| 135 |
.macro PICPEA sym, areg |
| 136 |
movel a5@(_current_shared_library_a5_offset_), \areg |
| 137 |
movel \sym@GOT(\areg), sp@- |
| 138 |
.endm |
| 139 |
|
| 140 |
.macro PICCALL addr |
| 141 |
PICLEA \addr,a0 |
| 142 |
jsr a0@ |
| 143 |
.endm |
| 144 |
|
| 145 |
.macro PICJUMP addr |
| 146 |
PICLEA \addr,a0 |
| 147 |
jmp a0@ |
| 148 |
.endm |
| 149 |
|
| 150 |
# else /* !__ID_SHARED_LIBRARY__ */ |
| 151 |
|
| 152 |
/* Versions for -msep-data */ |
| 153 |
|
| 154 |
.macro PICLEA sym, reg |
| 155 |
movel \sym@GOT(a5), \reg |
| 156 |
.endm |
| 157 |
|
| 158 |
.macro PICPEA sym, areg |
| 159 |
movel \sym@GOT(a5), sp@- |
| 160 |
.endm |
| 161 |
|
| 162 |
.macro PICCALL addr |
| 163 |
#if defined (__mcoldfire__) && !defined (__mcfisab__) && !defined (__mcfisac__) |
| 164 |
lea \addr-.-8,a0 |
| 165 |
jsr pc@(a0) |
| 166 |
#else |
| 167 |
jbsr \addr |
| 168 |
#endif |
| 169 |
.endm |
| 170 |
|
| 171 |
.macro PICJUMP addr |
| 172 |
/* ISA C has no bra.l instruction, and since this assembly file |
| 173 |
gets assembled into multiple object files, we avoid the |
| 174 |
bra instruction entirely. */ |
| 175 |
#if defined (__mcoldfire__) && !defined (__mcfisab__) |
| 176 |
lea \addr-.-8,a0 |
| 177 |
jmp pc@(a0) |
| 178 |
#else |
| 179 |
bra \addr |
| 180 |
#endif |
| 181 |
.endm |
| 182 |
|
| 183 |
# endif |
| 184 |
|
| 185 |
# else /* !__uClinux__ */ |
| 186 |
|
| 187 |
/* Versions for Linux */ |
| 188 |
|
| 189 |
.macro PICLEA sym, reg |
| 190 |
movel #_GLOBAL_OFFSET_TABLE_@GOTPC, \reg |
| 191 |
lea (-6, pc, \reg), \reg |
| 192 |
movel \sym@GOT(\reg), \reg |
| 193 |
.endm |
| 194 |
|
| 195 |
.macro PICPEA sym, areg |
| 196 |
movel #_GLOBAL_OFFSET_TABLE_@GOTPC, \areg |
| 197 |
lea (-6, pc, \areg), \areg |
| 198 |
movel \sym@GOT(\areg), sp@- |
| 199 |
.endm |
| 200 |
|
| 201 |
.macro PICCALL addr |
| 202 |
#if defined (__mcoldfire__) && !defined (__mcfisab__) && !defined (__mcfisac__) |
| 203 |
lea \addr-.-8,a0 |
| 204 |
jsr pc@(a0) |
| 205 |
#else |
| 206 |
jbsr \addr |
| 207 |
#endif |
| 208 |
.endm |
| 209 |
|
| 210 |
.macro PICJUMP addr |
| 211 |
/* ISA C has no bra.l instruction, and since this assembly file |
| 212 |
gets assembled into multiple object files, we avoid the |
| 213 |
bra instruction entirely. */ |
| 214 |
#if defined (__mcoldfire__) && !defined (__mcfisab__) |
| 215 |
lea \addr-.-8,a0 |
| 216 |
jmp pc@(a0) |
| 217 |
#else |
| 218 |
bra \addr |
| 219 |
#endif |
| 220 |
.endm |
| 221 |
|
| 222 |
# endif |
| 223 |
#endif /* __PIC__ */ |
| 224 |
|
| 225 |
|
| 226 |
#ifdef L_floatex |
| 227 |
|
| 228 |
| This is an attempt at a decent floating point (single, double and |
| 229 |
| extended double) code for the GNU C compiler. It should be easy to |
| 230 |
| adapt to other compilers (but beware of the local labels!). |
| 231 |
|
| 232 |
| Starting date: 21 October, 1990 |
| 233 |
|
| 234 |
| It is convenient to introduce the notation (s,e,f) for a floating point |
| 235 |
| number, where s=sign, e=exponent, f=fraction. We will call a floating |
| 236 |
| point number fpn to abbreviate, independently of the precision. |
| 237 |
| Let MAX_EXP be in each case the maximum exponent (255 for floats, 1023 |
| 238 |
| for doubles and 16383 for long doubles). We then have the following |
| 239 |
| different cases: |
| 240 |
| 1. Normalized fpns have 0 < e < MAX_EXP. They correspond to |
| 241 |
| (-1)^s x 1.f x 2^(e-bias-1). |
| 242 |
| 2. Denormalized fpns have e=0. They correspond to numbers of the form |
| 243 |
| (-1)^s x 0.f x 2^(-bias). |
| 244 |
| 3. +/-INFINITY have e=MAX_EXP, f=0. |
| 245 |
| 4. Quiet NaN (Not a Number) have all bits set. |
| 246 |
| 5. Signaling NaN (Not a Number) have s=0, e=MAX_EXP, f=1. |
| 247 |
|
| 248 |
|============================================================================= |
| 249 |
| exceptions |
| 250 |
|============================================================================= |
| 251 |
|
| 252 |
| This is the floating point condition code register (_fpCCR): |
| 253 |
| |
| 254 |
| struct { |
| 255 |
| short _exception_bits; |
| 256 |
| short _trap_enable_bits; |
| 257 |
| short _sticky_bits; |
| 258 |
| short _rounding_mode; |
| 259 |
| short _format; |
| 260 |
| short _last_operation; |
| 261 |
| union { |
| 262 |
| float sf; |
| 263 |
| double df; |
| 264 |
| } _operand1; |
| 265 |
| union { |
| 266 |
| float sf; |
| 267 |
| double df; |
| 268 |
| } _operand2; |
| 269 |
| } _fpCCR; |
| 270 |
|
| 271 |
.data |
| 272 |
.even |
| 273 |
|
| 274 |
.globl SYM (_fpCCR) |
| 275 |
|
| 276 |
SYM (_fpCCR): |
| 277 |
__exception_bits: |
| 278 |
.word 0 |
| 279 |
__trap_enable_bits: |
| 280 |
.word 0 |
| 281 |
__sticky_bits: |
| 282 |
.word 0 |
| 283 |
__rounding_mode: |
| 284 |
.word ROUND_TO_NEAREST |
| 285 |
__format: |
| 286 |
.word NIL |
| 287 |
__last_operation: |
| 288 |
.word NOOP |
| 289 |
__operand1: |
| 290 |
.long 0 |
| 291 |
.long 0 |
| 292 |
__operand2: |
| 293 |
.long 0 |
| 294 |
.long 0 |
| 295 |
|
| 296 |
| Offsets: |
| 297 |
EBITS = __exception_bits - SYM (_fpCCR) |
| 298 |
TRAPE = __trap_enable_bits - SYM (_fpCCR) |
| 299 |
STICK = __sticky_bits - SYM (_fpCCR) |
| 300 |
ROUND = __rounding_mode - SYM (_fpCCR) |
| 301 |
FORMT = __format - SYM (_fpCCR) |
| 302 |
LASTO = __last_operation - SYM (_fpCCR) |
| 303 |
OPER1 = __operand1 - SYM (_fpCCR) |
| 304 |
OPER2 = __operand2 - SYM (_fpCCR) |
| 305 |
|
| 306 |
| The following exception types are supported: |
| 307 |
INEXACT_RESULT = 0x0001 |
| 308 |
UNDERFLOW = 0x0002 |
| 309 |
OVERFLOW = 0x0004 |
| 310 |
DIVIDE_BY_ZERO = 0x0008 |
| 311 |
INVALID_OPERATION = 0x0010 |
| 312 |
|
| 313 |
| The allowed rounding modes are: |
| 314 |
UNKNOWN = -1 |
| 315 |
ROUND_TO_NEAREST = 0 | round result to nearest representable value |
| 316 |
ROUND_TO_ZERO = 1 | round result towards zero |
| 317 |
ROUND_TO_PLUS = 2 | round result towards plus infinity |
| 318 |
ROUND_TO_MINUS = 3 | round result towards minus infinity |
| 319 |
|
| 320 |
| The allowed values of format are: |
| 321 |
NIL = 0 |
| 322 |
SINGLE_FLOAT = 1 |
| 323 |
DOUBLE_FLOAT = 2 |
| 324 |
LONG_FLOAT = 3 |
| 325 |
|
| 326 |
| The allowed values for the last operation are: |
| 327 |
NOOP = 0 |
| 328 |
ADD = 1 |
| 329 |
MULTIPLY = 2 |
| 330 |
DIVIDE = 3 |
| 331 |
NEGATE = 4 |
| 332 |
COMPARE = 5 |
| 333 |
EXTENDSFDF = 6 |
| 334 |
TRUNCDFSF = 7 |
| 335 |
|
| 336 |
|============================================================================= |
| 337 |
| __clear_sticky_bits |
| 338 |
|============================================================================= |
| 339 |
|
| 340 |
| The sticky bits are normally not cleared (thus the name), whereas the |
| 341 |
| exception type and exception value reflect the last computation. |
| 342 |
| This routine is provided to clear them (you can also write to _fpCCR, |
| 343 |
| since it is globally visible). |
| 344 |
|
| 345 |
.globl SYM (__clear_sticky_bit) |
| 346 |
|
| 347 |
.text |
| 348 |
.even |
| 349 |
|
| 350 |
| void __clear_sticky_bits(void); |
| 351 |
SYM (__clear_sticky_bit): |
| 352 |
PICLEA SYM (_fpCCR),a0 |
| 353 |
#ifndef __mcoldfire__ |
| 354 |
movew IMM (0),a0@(STICK) |
| 355 |
#else |
| 356 |
clr.w a0@(STICK) |
| 357 |
#endif |
| 358 |
rts |
| 359 |
|
| 360 |
|============================================================================= |
| 361 |
| $_exception_handler |
| 362 |
|============================================================================= |
| 363 |
|
| 364 |
.globl $_exception_handler |
| 365 |
|
| 366 |
.text |
| 367 |
.even |
| 368 |
|
| 369 |
| This is the common exit point if an exception occurs. |
| 370 |
| NOTE: it is NOT callable from C! |
| 371 |
| It expects the exception type in d7, the format (SINGLE_FLOAT, |
| 372 |
| DOUBLE_FLOAT or LONG_FLOAT) in d6, and the last operation code in d5. |
| 373 |
| It sets the corresponding exception and sticky bits, and the format. |
| 374 |
| Depending on the format if fills the corresponding slots for the |
| 375 |
| operands which produced the exception (all this information is provided |
| 376 |
| so if you write your own exception handlers you have enough information |
| 377 |
| to deal with the problem). |
| 378 |
| Then checks to see if the corresponding exception is trap-enabled, |
| 379 |
| in which case it pushes the address of _fpCCR and traps through |
| 380 |
| trap FPTRAP (15 for the moment). |
| 381 |
|
| 382 |
FPTRAP = 15 |
| 383 |
|
| 384 |
$_exception_handler: |
| 385 |
PICLEA SYM (_fpCCR),a0 |
| 386 |
movew d7,a0@(EBITS) | set __exception_bits |
| 387 |
#ifndef __mcoldfire__ |
| 388 |
orw d7,a0@(STICK) | and __sticky_bits |
| 389 |
#else |
| 390 |
movew a0@(STICK),d4 |
| 391 |
orl d7,d4 |
| 392 |
movew d4,a0@(STICK) |
| 393 |
#endif |
| 394 |
movew d6,a0@(FORMT) | and __format |
| 395 |
movew d5,a0@(LASTO) | and __last_operation |
| 396 |
|
| 397 |
| Now put the operands in place: |
| 398 |
#ifndef __mcoldfire__ |
| 399 |
cmpw IMM (SINGLE_FLOAT),d6 |
| 400 |
#else |
| 401 |
cmpl IMM (SINGLE_FLOAT),d6 |
| 402 |
#endif |
| 403 |
beq 1f |
| 404 |
movel a6@(8),a0@(OPER1) |
| 405 |
movel a6@(12),a0@(OPER1+4) |
| 406 |
movel a6@(16),a0@(OPER2) |
| 407 |
movel a6@(20),a0@(OPER2+4) |
| 408 |
bra 2f |
| 409 |
1: movel a6@(8),a0@(OPER1) |
| 410 |
movel a6@(12),a0@(OPER2) |
| 411 |
2: |
| 412 |
| And check whether the exception is trap-enabled: |
| 413 |
#ifndef __mcoldfire__ |
| 414 |
andw a0@(TRAPE),d7 | is exception trap-enabled? |
| 415 |
#else |
| 416 |
clrl d6 |
| 417 |
movew a0@(TRAPE),d6 |
| 418 |
andl d6,d7 |
| 419 |
#endif |
| 420 |
beq 1f | no, exit |
| 421 |
PICPEA SYM (_fpCCR),a1 | yes, push address of _fpCCR |
| 422 |
trap IMM (FPTRAP) | and trap |
| 423 |
#ifndef __mcoldfire__ |
| 424 |
1: moveml sp@+,d2-d7 | restore data registers |
| 425 |
#else |
| 426 |
1: moveml sp@,d2-d7 |
| 427 |
| XXX if frame pointer is ever removed, stack pointer must |
| 428 |
| be adjusted here. |
| 429 |
#endif |
| 430 |
unlk a6 | and return |
| 431 |
rts |
| 432 |
#endif /* L_floatex */ |
| 433 |
|
| 434 |
#ifdef L_mulsi3 |
| 435 |
.text |
| 436 |
FUNC(__mulsi3) |
| 437 |
.globl SYM (__mulsi3) |
| 438 |
.globl SYM (__mulsi3_internal) |
| 439 |
.hidden SYM (__mulsi3_internal) |
| 440 |
SYM (__mulsi3): |
| 441 |
SYM (__mulsi3_internal): |
| 442 |
movew sp@(4), d0 /* x0 -> d0 */ |
| 443 |
muluw sp@(10), d0 /* x0*y1 */ |
| 444 |
movew sp@(6), d1 /* x1 -> d1 */ |
| 445 |
muluw sp@(8), d1 /* x1*y0 */ |
| 446 |
#ifndef __mcoldfire__ |
| 447 |
addw d1, d0 |
| 448 |
#else |
| 449 |
addl d1, d0 |
| 450 |
#endif |
| 451 |
swap d0 |
| 452 |
clrw d0 |
| 453 |
movew sp@(6), d1 /* x1 -> d1 */ |
| 454 |
muluw sp@(10), d1 /* x1*y1 */ |
| 455 |
addl d1, d0 |
| 456 |
|
| 457 |
rts |
| 458 |
#endif /* L_mulsi3 */ |
| 459 |
|
| 460 |
#ifdef L_udivsi3 |
| 461 |
.text |
| 462 |
FUNC(__udivsi3) |
| 463 |
.globl SYM (__udivsi3) |
| 464 |
.globl SYM (__udivsi3_internal) |
| 465 |
.hidden SYM (__udivsi3_internal) |
| 466 |
SYM (__udivsi3): |
| 467 |
SYM (__udivsi3_internal): |
| 468 |
#ifndef __mcoldfire__ |
| 469 |
movel d2, sp@- |
| 470 |
movel sp@(12), d1 /* d1 = divisor */ |
| 471 |
movel sp@(8), d0 /* d0 = dividend */ |
| 472 |
|
| 473 |
cmpl IMM (0x10000), d1 /* divisor >= 2 ^ 16 ? */ |
| 474 |
jcc L3 /* then try next algorithm */ |
| 475 |
movel d0, d2 |
| 476 |
clrw d2 |
| 477 |
swap d2 |
| 478 |
divu d1, d2 /* high quotient in lower word */ |
| 479 |
movew d2, d0 /* save high quotient */ |
| 480 |
swap d0 |
| 481 |
movew sp@(10), d2 /* get low dividend + high rest */ |
| 482 |
divu d1, d2 /* low quotient */ |
| 483 |
movew d2, d0 |
| 484 |
jra L6 |
| 485 |
|
| 486 |
L3: movel d1, d2 /* use d2 as divisor backup */ |
| 487 |
L4: lsrl IMM (1), d1 /* shift divisor */ |
| 488 |
lsrl IMM (1), d0 /* shift dividend */ |
| 489 |
cmpl IMM (0x10000), d1 /* still divisor >= 2 ^ 16 ? */ |
| 490 |
jcc L4 |
| 491 |
divu d1, d0 /* now we have 16-bit divisor */ |
| 492 |
andl IMM (0xffff), d0 /* mask out divisor, ignore remainder */ |
| 493 |
|
| 494 |
/* Multiply the 16-bit tentative quotient with the 32-bit divisor. Because of |
| 495 |
the operand ranges, this might give a 33-bit product. If this product is |
| 496 |
greater than the dividend, the tentative quotient was too large. */ |
| 497 |
movel d2, d1 |
| 498 |
mulu d0, d1 /* low part, 32 bits */ |
| 499 |
swap d2 |
| 500 |
mulu d0, d2 /* high part, at most 17 bits */ |
| 501 |
swap d2 /* align high part with low part */ |
| 502 |
tstw d2 /* high part 17 bits? */ |
| 503 |
jne L5 /* if 17 bits, quotient was too large */ |
| 504 |
addl d2, d1 /* add parts */ |
| 505 |
jcs L5 /* if sum is 33 bits, quotient was too large */ |
| 506 |
cmpl sp@(8), d1 /* compare the sum with the dividend */ |
| 507 |
jls L6 /* if sum > dividend, quotient was too large */ |
| 508 |
L5: subql IMM (1), d0 /* adjust quotient */ |
| 509 |
|
| 510 |
L6: movel sp@+, d2 |
| 511 |
rts |
| 512 |
|
| 513 |
#else /* __mcoldfire__ */ |
| 514 |
|
| 515 |
/* ColdFire implementation of non-restoring division algorithm from |
| 516 |
Hennessy & Patterson, Appendix A. */ |
| 517 |
link a6,IMM (-12) |
| 518 |
moveml d2-d4,sp@ |
| 519 |
movel a6@(8),d0 |
| 520 |
movel a6@(12),d1 |
| 521 |
clrl d2 | clear p |
| 522 |
moveq IMM (31),d4 |
| 523 |
L1: addl d0,d0 | shift reg pair (p,a) one bit left |
| 524 |
addxl d2,d2 |
| 525 |
movl d2,d3 | subtract b from p, store in tmp. |
| 526 |
subl d1,d3 |
| 527 |
jcs L2 | if no carry, |
| 528 |
bset IMM (0),d0 | set the low order bit of a to 1, |
| 529 |
movl d3,d2 | and store tmp in p. |
| 530 |
L2: subql IMM (1),d4 |
| 531 |
jcc L1 |
| 532 |
moveml sp@,d2-d4 | restore data registers |
| 533 |
unlk a6 | and return |
| 534 |
rts |
| 535 |
#endif /* __mcoldfire__ */ |
| 536 |
|
| 537 |
#endif /* L_udivsi3 */ |
| 538 |
|
| 539 |
#ifdef L_divsi3 |
| 540 |
.text |
| 541 |
FUNC(__divsi3) |
| 542 |
.globl SYM (__divsi3) |
| 543 |
.globl SYM (__divsi3_internal) |
| 544 |
.hidden SYM (__divsi3_internal) |
| 545 |
SYM (__divsi3): |
| 546 |
SYM (__divsi3_internal): |
| 547 |
movel d2, sp@- |
| 548 |
|
| 549 |
moveq IMM (1), d2 /* sign of result stored in d2 (=1 or =-1) */ |
| 550 |
movel sp@(12), d1 /* d1 = divisor */ |
| 551 |
jpl L1 |
| 552 |
negl d1 |
| 553 |
#ifndef __mcoldfire__ |
| 554 |
negb d2 /* change sign because divisor <0 */ |
| 555 |
#else |
| 556 |
negl d2 /* change sign because divisor <0 */ |
| 557 |
#endif |
| 558 |
L1: movel sp@(8), d0 /* d0 = dividend */ |
| 559 |
jpl L2 |
| 560 |
negl d0 |
| 561 |
#ifndef __mcoldfire__ |
| 562 |
negb d2 |
| 563 |
#else |
| 564 |
negl d2 |
| 565 |
#endif |
| 566 |
|
| 567 |
L2: movel d1, sp@- |
| 568 |
movel d0, sp@- |
| 569 |
PICCALL SYM (__udivsi3_internal) /* divide abs(dividend) by abs(divisor) */ |
| 570 |
addql IMM (8), sp |
| 571 |
|
| 572 |
tstb d2 |
| 573 |
jpl L3 |
| 574 |
negl d0 |
| 575 |
|
| 576 |
L3: movel sp@+, d2 |
| 577 |
rts |
| 578 |
#endif /* L_divsi3 */ |
| 579 |
|
| 580 |
#ifdef L_umodsi3 |
| 581 |
.text |
| 582 |
FUNC(__umodsi3) |
| 583 |
.globl SYM (__umodsi3) |
| 584 |
SYM (__umodsi3): |
| 585 |
movel sp@(8), d1 /* d1 = divisor */ |
| 586 |
movel sp@(4), d0 /* d0 = dividend */ |
| 587 |
movel d1, sp@- |
| 588 |
movel d0, sp@- |
| 589 |
PICCALL SYM (__udivsi3_internal) |
| 590 |
addql IMM (8), sp |
| 591 |
movel sp@(8), d1 /* d1 = divisor */ |
| 592 |
#ifndef __mcoldfire__ |
| 593 |
movel d1, sp@- |
| 594 |
movel d0, sp@- |
| 595 |
PICCALL SYM (__mulsi3_internal) /* d0 = (a/b)*b */ |
| 596 |
addql IMM (8), sp |
| 597 |
#else |
| 598 |
mulsl d1,d0 |
| 599 |
#endif |
| 600 |
movel sp@(4), d1 /* d1 = dividend */ |
| 601 |
subl d0, d1 /* d1 = a - (a/b)*b */ |
| 602 |
movel d1, d0 |
| 603 |
rts |
| 604 |
#endif /* L_umodsi3 */ |
| 605 |
|
| 606 |
#ifdef L_modsi3 |
| 607 |
.text |
| 608 |
FUNC(__modsi3) |
| 609 |
.globl SYM (__modsi3) |
| 610 |
SYM (__modsi3): |
| 611 |
movel sp@(8), d1 /* d1 = divisor */ |
| 612 |
movel sp@(4), d0 /* d0 = dividend */ |
| 613 |
movel d1, sp@- |
| 614 |
movel d0, sp@- |
| 615 |
PICCALL SYM (__divsi3_internal) |
| 616 |
addql IMM (8), sp |
| 617 |
movel sp@(8), d1 /* d1 = divisor */ |
| 618 |
#ifndef __mcoldfire__ |
| 619 |
movel d1, sp@- |
| 620 |
movel d0, sp@- |
| 621 |
PICCALL SYM (__mulsi3_internal) /* d0 = (a/b)*b */ |
| 622 |
addql IMM (8), sp |
| 623 |
#else |
| 624 |
mulsl d1,d0 |
| 625 |
#endif |
| 626 |
movel sp@(4), d1 /* d1 = dividend */ |
| 627 |
subl d0, d1 /* d1 = a - (a/b)*b */ |
| 628 |
movel d1, d0 |
| 629 |
rts |
| 630 |
#endif /* L_modsi3 */ |
| 631 |
|
| 632 |
|
| 633 |
#ifdef L_double |
| 634 |
|
| 635 |
.globl SYM (_fpCCR) |
| 636 |
.globl $_exception_handler |
| 637 |
|
| 638 |
QUIET_NaN = 0x7fffffff |
| 639 |
|
| 640 |
D_MAX_EXP = 0x07ff |
| 641 |
D_BIAS = 1022 |
| 642 |
DBL_MAX_EXP = D_MAX_EXP - D_BIAS |
| 643 |
DBL_MIN_EXP = 1 - D_BIAS |
| 644 |
DBL_MANT_DIG = 53 |
| 645 |
|
| 646 |
INEXACT_RESULT = 0x0001 |
| 647 |
UNDERFLOW = 0x0002 |
| 648 |
OVERFLOW = 0x0004 |
| 649 |
DIVIDE_BY_ZERO = 0x0008 |
| 650 |
INVALID_OPERATION = 0x0010 |
| 651 |
|
| 652 |
DOUBLE_FLOAT = 2 |
| 653 |
|
| 654 |
NOOP = 0 |
| 655 |
ADD = 1 |
| 656 |
MULTIPLY = 2 |
| 657 |
DIVIDE = 3 |
| 658 |
NEGATE = 4 |
| 659 |
COMPARE = 5 |
| 660 |
EXTENDSFDF = 6 |
| 661 |
TRUNCDFSF = 7 |
| 662 |
|
| 663 |
UNKNOWN = -1 |
| 664 |
ROUND_TO_NEAREST = 0 | round result to nearest representable value |
| 665 |
ROUND_TO_ZERO = 1 | round result towards zero |
| 666 |
ROUND_TO_PLUS = 2 | round result towards plus infinity |
| 667 |
ROUND_TO_MINUS = 3 | round result towards minus infinity |
| 668 |
|
| 669 |
| Entry points: |
| 670 |
|
| 671 |
.globl SYM (__adddf3) |
| 672 |
.globl SYM (__subdf3) |
| 673 |
.globl SYM (__muldf3) |
| 674 |
.globl SYM (__divdf3) |
| 675 |
.globl SYM (__negdf2) |
| 676 |
.globl SYM (__cmpdf2) |
| 677 |
.globl SYM (__cmpdf2_internal) |
| 678 |
.hidden SYM (__cmpdf2_internal) |
| 679 |
|
| 680 |
.text |
| 681 |
.even |
| 682 |
|
| 683 |
| These are common routines to return and signal exceptions. |
| 684 |
|
| 685 |
Ld$den: |
| 686 |
| Return and signal a denormalized number |
| 687 |
orl d7,d0 |
| 688 |
movew IMM (INEXACT_RESULT+UNDERFLOW),d7 |
| 689 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 690 |
PICJUMP $_exception_handler |
| 691 |
|
| 692 |
Ld$infty: |
| 693 |
Ld$overflow: |
| 694 |
| Return a properly signed INFINITY and set the exception flags |
| 695 |
movel IMM (0x7ff00000),d0 |
| 696 |
movel IMM (0),d1 |
| 697 |
orl d7,d0 |
| 698 |
movew IMM (INEXACT_RESULT+OVERFLOW),d7 |
| 699 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 700 |
PICJUMP $_exception_handler |
| 701 |
|
| 702 |
Ld$underflow: |
| 703 |
| Return a properly signed 0 and set the exception flags |
| 704 |
movel IMM (0),d0 |
| 705 |
movel d0,d1 |
| 706 |
orl d7,d0 |
| 707 |
movew IMM (INEXACT_RESULT+UNDERFLOW),d7 |
| 708 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 709 |
PICJUMP $_exception_handler |
| 710 |
|
| 711 |
Ld$inop: |
| 712 |
| Return a quiet NaN and set the exception flags |
| 713 |
movel IMM (QUIET_NaN),d0 |
| 714 |
movel d0,d1 |
| 715 |
bset IMM (31),d1 |
| 716 |
movew IMM (INEXACT_RESULT+INVALID_OPERATION),d7 |
| 717 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 718 |
PICJUMP $_exception_handler |
| 719 |
|
| 720 |
Ld$div$0: |
| 721 |
| Return a properly signed INFINITY and set the exception flags |
| 722 |
movel IMM (0x7ff00000),d0 |
| 723 |
movel IMM (0),d1 |
| 724 |
orl d7,d0 |
| 725 |
movew IMM (INEXACT_RESULT+DIVIDE_BY_ZERO),d7 |
| 726 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 727 |
PICJUMP $_exception_handler |
| 728 |
|
| 729 |
|============================================================================= |
| 730 |
|============================================================================= |
| 731 |
| double precision routines |
| 732 |
|============================================================================= |
| 733 |
|============================================================================= |
| 734 |
|
| 735 |
| A double precision floating point number (double) has the format: |
| 736 |
| |
| 737 |
| struct _double { |
| 738 |
| unsigned int sign : 1; /* sign bit */ |
| 739 |
| unsigned int exponent : 11; /* exponent, shifted by 126 */ |
| 740 |
| unsigned int fraction : 52; /* fraction */ |
| 741 |
| } double; |
| 742 |
| |
| 743 |
| Thus sizeof(double) = 8 (64 bits). |
| 744 |
| |
| 745 |
| All the routines are callable from C programs, and return the result |
| 746 |
| in the register pair d0-d1. They also preserve all registers except |
| 747 |
| d0-d1 and a0-a1. |
| 748 |
|
| 749 |
|============================================================================= |
| 750 |
| __subdf3 |
| 751 |
|============================================================================= |
| 752 |
|
| 753 |
| double __subdf3(double, double); |
| 754 |
FUNC(__subdf3) |
| 755 |
SYM (__subdf3): |
| 756 |
bchg IMM (31),sp@(12) | change sign of second operand |
| 757 |
| and fall through, so we always add |
| 758 |
|============================================================================= |
| 759 |
| __adddf3 |
| 760 |
|============================================================================= |
| 761 |
|
| 762 |
| double __adddf3(double, double); |
| 763 |
FUNC(__adddf3) |
| 764 |
SYM (__adddf3): |
| 765 |
#ifndef __mcoldfire__ |
| 766 |
link a6,IMM (0) | everything will be done in registers |
| 767 |
moveml d2-d7,sp@- | save all data registers and a2 (but d0-d1) |
| 768 |
#else |
| 769 |
link a6,IMM (-24) |
| 770 |
moveml d2-d7,sp@ |
| 771 |
#endif |
| 772 |
movel a6@(8),d0 | get first operand |
| 773 |
movel a6@(12),d1 | |
| 774 |
movel a6@(16),d2 | get second operand |
| 775 |
movel a6@(20),d3 | |
| 776 |
|
| 777 |
movel d0,d7 | get d0's sign bit in d7 ' |
| 778 |
addl d1,d1 | check and clear sign bit of a, and gain one |
| 779 |
addxl d0,d0 | bit of extra precision |
| 780 |
beq Ladddf$b | if zero return second operand |
| 781 |
|
| 782 |
movel d2,d6 | save sign in d6 |
| 783 |
addl d3,d3 | get rid of sign bit and gain one bit of |
| 784 |
addxl d2,d2 | extra precision |
| 785 |
beq Ladddf$a | if zero return first operand |
| 786 |
|
| 787 |
andl IMM (0x80000000),d7 | isolate a's sign bit ' |
| 788 |
swap d6 | and also b's sign bit ' |
| 789 |
#ifndef __mcoldfire__ |
| 790 |
andw IMM (0x8000),d6 | |
| 791 |
orw d6,d7 | and combine them into d7, so that a's sign ' |
| 792 |
| bit is in the high word and b's is in the ' |
| 793 |
| low word, so d6 is free to be used |
| 794 |
#else |
| 795 |
andl IMM (0x8000),d6 |
| 796 |
orl d6,d7 |
| 797 |
#endif |
| 798 |
movel d7,a0 | now save d7 into a0, so d7 is free to |
| 799 |
| be used also |
| 800 |
|
| 801 |
| Get the exponents and check for denormalized and/or infinity. |
| 802 |
|
| 803 |
movel IMM (0x001fffff),d6 | mask for the fraction |
| 804 |
movel IMM (0x00200000),d7 | mask to put hidden bit back |
| 805 |
|
| 806 |
movel d0,d4 | |
| 807 |
andl d6,d0 | get fraction in d0 |
| 808 |
notl d6 | make d6 into mask for the exponent |
| 809 |
andl d6,d4 | get exponent in d4 |
| 810 |
beq Ladddf$a$den | branch if a is denormalized |
| 811 |
cmpl d6,d4 | check for INFINITY or NaN |
| 812 |
beq Ladddf$nf | |
| 813 |
orl d7,d0 | and put hidden bit back |
| 814 |
Ladddf$1: |
| 815 |
swap d4 | shift right exponent so that it starts |
| 816 |
#ifndef __mcoldfire__ |
| 817 |
lsrw IMM (5),d4 | in bit 0 and not bit 20 |
| 818 |
#else |
| 819 |
lsrl IMM (5),d4 | in bit 0 and not bit 20 |
| 820 |
#endif |
| 821 |
| Now we have a's exponent in d4 and fraction in d0-d1 ' |
| 822 |
movel d2,d5 | save b to get exponent |
| 823 |
andl d6,d5 | get exponent in d5 |
| 824 |
beq Ladddf$b$den | branch if b is denormalized |
| 825 |
cmpl d6,d5 | check for INFINITY or NaN |
| 826 |
beq Ladddf$nf |
| 827 |
notl d6 | make d6 into mask for the fraction again |
| 828 |
andl d6,d2 | and get fraction in d2 |
| 829 |
orl d7,d2 | and put hidden bit back |
| 830 |
Ladddf$2: |
| 831 |
swap d5 | shift right exponent so that it starts |
| 832 |
#ifndef __mcoldfire__ |
| 833 |
lsrw IMM (5),d5 | in bit 0 and not bit 20 |
| 834 |
#else |
| 835 |
lsrl IMM (5),d5 | in bit 0 and not bit 20 |
| 836 |
#endif |
| 837 |
|
| 838 |
| Now we have b's exponent in d5 and fraction in d2-d3. ' |
| 839 |
|
| 840 |
| The situation now is as follows: the signs are combined in a0, the |
| 841 |
| numbers are in d0-d1 (a) and d2-d3 (b), and the exponents in d4 (a) |
| 842 |
| and d5 (b). To do the rounding correctly we need to keep all the |
| 843 |
| bits until the end, so we need to use d0-d1-d2-d3 for the first number |
| 844 |
| and d4-d5-d6-d7 for the second. To do this we store (temporarily) the |
| 845 |
| exponents in a2-a3. |
| 846 |
|
| 847 |
#ifndef __mcoldfire__ |
| 848 |
moveml a2-a3,sp@- | save the address registers |
| 849 |
#else |
| 850 |
movel a2,sp@- |
| 851 |
movel a3,sp@- |
| 852 |
movel a4,sp@- |
| 853 |
#endif |
| 854 |
|
| 855 |
movel d4,a2 | save the exponents |
| 856 |
movel d5,a3 | |
| 857 |
|
| 858 |
movel IMM (0),d7 | and move the numbers around |
| 859 |
movel d7,d6 | |
| 860 |
movel d3,d5 | |
| 861 |
movel d2,d4 | |
| 862 |
movel d7,d3 | |
| 863 |
movel d7,d2 | |
| 864 |
|
| 865 |
| Here we shift the numbers until the exponents are the same, and put |
| 866 |
| the largest exponent in a2. |
| 867 |
#ifndef __mcoldfire__ |
| 868 |
exg d4,a2 | get exponents back |
| 869 |
exg d5,a3 | |
| 870 |
cmpw d4,d5 | compare the exponents |
| 871 |
#else |
| 872 |
movel d4,a4 | get exponents back |
| 873 |
movel a2,d4 |
| 874 |
movel a4,a2 |
| 875 |
movel d5,a4 |
| 876 |
movel a3,d5 |
| 877 |
movel a4,a3 |
| 878 |
cmpl d4,d5 | compare the exponents |
| 879 |
#endif |
| 880 |
beq Ladddf$3 | if equal don't shift ' |
| 881 |
bhi 9f | branch if second exponent is higher |
| 882 |
|
| 883 |
| Here we have a's exponent larger than b's, so we have to shift b. We do |
| 884 |
| this by using as counter d2: |
| 885 |
1: movew d4,d2 | move largest exponent to d2 |
| 886 |
#ifndef __mcoldfire__ |
| 887 |
subw d5,d2 | and subtract second exponent |
| 888 |
exg d4,a2 | get back the longs we saved |
| 889 |
exg d5,a3 | |
| 890 |
#else |
| 891 |
subl d5,d2 | and subtract second exponent |
| 892 |
movel d4,a4 | get back the longs we saved |
| 893 |
movel a2,d4 |
| 894 |
movel a4,a2 |
| 895 |
movel d5,a4 |
| 896 |
movel a3,d5 |
| 897 |
movel a4,a3 |
| 898 |
#endif |
| 899 |
| if difference is too large we don't shift (actually, we can just exit) ' |
| 900 |
#ifndef __mcoldfire__ |
| 901 |
cmpw IMM (DBL_MANT_DIG+2),d2 |
| 902 |
#else |
| 903 |
cmpl IMM (DBL_MANT_DIG+2),d2 |
| 904 |
#endif |
| 905 |
bge Ladddf$b$small |
| 906 |
#ifndef __mcoldfire__ |
| 907 |
cmpw IMM (32),d2 | if difference >= 32, shift by longs |
| 908 |
#else |
| 909 |
cmpl IMM (32),d2 | if difference >= 32, shift by longs |
| 910 |
#endif |
| 911 |
bge 5f |
| 912 |
2: |
| 913 |
#ifndef __mcoldfire__ |
| 914 |
cmpw IMM (16),d2 | if difference >= 16, shift by words |
| 915 |
#else |
| 916 |
cmpl IMM (16),d2 | if difference >= 16, shift by words |
| 917 |
#endif |
| 918 |
bge 6f |
| 919 |
bra 3f | enter dbra loop |
| 920 |
|
| 921 |
4: |
| 922 |
#ifndef __mcoldfire__ |
| 923 |
lsrl IMM (1),d4 |
| 924 |
roxrl IMM (1),d5 |
| 925 |
roxrl IMM (1),d6 |
| 926 |
roxrl IMM (1),d7 |
| 927 |
#else |
| 928 |
lsrl IMM (1),d7 |
| 929 |
btst IMM (0),d6 |
| 930 |
beq 10f |
| 931 |
bset IMM (31),d7 |
| 932 |
10: lsrl IMM (1),d6 |
| 933 |
btst IMM (0),d5 |
| 934 |
beq 11f |
| 935 |
bset IMM (31),d6 |
| 936 |
11: lsrl IMM (1),d5 |
| 937 |
btst IMM (0),d4 |
| 938 |
beq 12f |
| 939 |
bset IMM (31),d5 |
| 940 |
12: lsrl IMM (1),d4 |
| 941 |
#endif |
| 942 |
3: |
| 943 |
#ifndef __mcoldfire__ |
| 944 |
dbra d2,4b |
| 945 |
#else |
| 946 |
subql IMM (1),d2 |
| 947 |
bpl 4b |
| 948 |
#endif |
| 949 |
movel IMM (0),d2 |
| 950 |
movel d2,d3 |
| 951 |
bra Ladddf$4 |
| 952 |
5: |
| 953 |
movel d6,d7 |
| 954 |
movel d5,d6 |
| 955 |
movel d4,d5 |
| 956 |
movel IMM (0),d4 |
| 957 |
#ifndef __mcoldfire__ |
| 958 |
subw IMM (32),d2 |
| 959 |
#else |
| 960 |
subl IMM (32),d2 |
| 961 |
#endif |
| 962 |
bra 2b |
| 963 |
6: |
| 964 |
movew d6,d7 |
| 965 |
swap d7 |
| 966 |
movew d5,d6 |
| 967 |
swap d6 |
| 968 |
movew d4,d5 |
| 969 |
swap d5 |
| 970 |
movew IMM (0),d4 |
| 971 |
swap d4 |
| 972 |
#ifndef __mcoldfire__ |
| 973 |
subw IMM (16),d2 |
| 974 |
#else |
| 975 |
subl IMM (16),d2 |
| 976 |
#endif |
| 977 |
bra 3b |
| 978 |
|
| 979 |
9: |
| 980 |
#ifndef __mcoldfire__ |
| 981 |
exg d4,d5 |
| 982 |
movew d4,d6 |
| 983 |
subw d5,d6 | keep d5 (largest exponent) in d4 |
| 984 |
exg d4,a2 |
| 985 |
exg d5,a3 |
| 986 |
#else |
| 987 |
movel d5,d6 |
| 988 |
movel d4,d5 |
| 989 |
movel d6,d4 |
| 990 |
subl d5,d6 |
| 991 |
movel d4,a4 |
| 992 |
movel a2,d4 |
| 993 |
movel a4,a2 |
| 994 |
movel d5,a4 |
| 995 |
movel a3,d5 |
| 996 |
movel a4,a3 |
| 997 |
#endif |
| 998 |
| if difference is too large we don't shift (actually, we can just exit) ' |
| 999 |
#ifndef __mcoldfire__ |
| 1000 |
cmpw IMM (DBL_MANT_DIG+2),d6 |
| 1001 |
#else |
| 1002 |
cmpl IMM (DBL_MANT_DIG+2),d6 |
| 1003 |
#endif |
| 1004 |
bge Ladddf$a$small |
| 1005 |
#ifndef __mcoldfire__ |
| 1006 |
cmpw IMM (32),d6 | if difference >= 32, shift by longs |
| 1007 |
#else |
| 1008 |
cmpl IMM (32),d6 | if difference >= 32, shift by longs |
| 1009 |
#endif |
| 1010 |
bge 5f |
| 1011 |
2: |
| 1012 |
#ifndef __mcoldfire__ |
| 1013 |
cmpw IMM (16),d6 | if difference >= 16, shift by words |
| 1014 |
#else |
| 1015 |
cmpl IMM (16),d6 | if difference >= 16, shift by words |
| 1016 |
#endif |
| 1017 |
bge 6f |
| 1018 |
bra 3f | enter dbra loop |
| 1019 |
|
| 1020 |
4: |
| 1021 |
#ifndef __mcoldfire__ |
| 1022 |
lsrl IMM (1),d0 |
| 1023 |
roxrl IMM (1),d1 |
| 1024 |
roxrl IMM (1),d2 |
| 1025 |
roxrl IMM (1),d3 |
| 1026 |
#else |
| 1027 |
lsrl IMM (1),d3 |
| 1028 |
btst IMM (0),d2 |
| 1029 |
beq 10f |
| 1030 |
bset IMM (31),d3 |
| 1031 |
10: lsrl IMM (1),d2 |
| 1032 |
btst IMM (0),d1 |
| 1033 |
beq 11f |
| 1034 |
bset IMM (31),d2 |
| 1035 |
11: lsrl IMM (1),d1 |
| 1036 |
btst IMM (0),d0 |
| 1037 |
beq 12f |
| 1038 |
bset IMM (31),d1 |
| 1039 |
12: lsrl IMM (1),d0 |
| 1040 |
#endif |
| 1041 |
3: |
| 1042 |
#ifndef __mcoldfire__ |
| 1043 |
dbra d6,4b |
| 1044 |
#else |
| 1045 |
subql IMM (1),d6 |
| 1046 |
bpl 4b |
| 1047 |
#endif |
| 1048 |
movel IMM (0),d7 |
| 1049 |
movel d7,d6 |
| 1050 |
bra Ladddf$4 |
| 1051 |
5: |
| 1052 |
movel d2,d3 |
| 1053 |
movel d1,d2 |
| 1054 |
movel d0,d1 |
| 1055 |
movel IMM (0),d0 |
| 1056 |
#ifndef __mcoldfire__ |
| 1057 |
subw IMM (32),d6 |
| 1058 |
#else |
| 1059 |
subl IMM (32),d6 |
| 1060 |
#endif |
| 1061 |
bra 2b |
| 1062 |
6: |
| 1063 |
movew d2,d3 |
| 1064 |
swap d3 |
| 1065 |
movew d1,d2 |
| 1066 |
swap d2 |
| 1067 |
movew d0,d1 |
| 1068 |
swap d1 |
| 1069 |
movew IMM (0),d0 |
| 1070 |
swap d0 |
| 1071 |
#ifndef __mcoldfire__ |
| 1072 |
subw IMM (16),d6 |
| 1073 |
#else |
| 1074 |
subl IMM (16),d6 |
| 1075 |
#endif |
| 1076 |
bra 3b |
| 1077 |
Ladddf$3: |
| 1078 |
#ifndef __mcoldfire__ |
| 1079 |
exg d4,a2 |
| 1080 |
exg d5,a3 |
| 1081 |
#else |
| 1082 |
movel d4,a4 |
| 1083 |
movel a2,d4 |
| 1084 |
movel a4,a2 |
| 1085 |
movel d5,a4 |
| 1086 |
movel a3,d5 |
| 1087 |
movel a4,a3 |
| 1088 |
#endif |
| 1089 |
Ladddf$4: |
| 1090 |
| Now we have the numbers in d0--d3 and d4--d7, the exponent in a2, and |
| 1091 |
| the signs in a4. |
| 1092 |
|
| 1093 |
| Here we have to decide whether to add or subtract the numbers: |
| 1094 |
#ifndef __mcoldfire__ |
| 1095 |
exg d7,a0 | get the signs |
| 1096 |
exg d6,a3 | a3 is free to be used |
| 1097 |
#else |
| 1098 |
movel d7,a4 |
| 1099 |
movel a0,d7 |
| 1100 |
movel a4,a0 |
| 1101 |
movel d6,a4 |
| 1102 |
movel a3,d6 |
| 1103 |
movel a4,a3 |
| 1104 |
#endif |
| 1105 |
movel d7,d6 | |
| 1106 |
movew IMM (0),d7 | get a's sign in d7 ' |
| 1107 |
swap d6 | |
| 1108 |
movew IMM (0),d6 | and b's sign in d6 ' |
| 1109 |
eorl d7,d6 | compare the signs |
| 1110 |
bmi Lsubdf$0 | if the signs are different we have |
| 1111 |
| to subtract |
| 1112 |
#ifndef __mcoldfire__ |
| 1113 |
exg d7,a0 | else we add the numbers |
| 1114 |
exg d6,a3 | |
| 1115 |
#else |
| 1116 |
movel d7,a4 |
| 1117 |
movel a0,d7 |
| 1118 |
movel a4,a0 |
| 1119 |
movel d6,a4 |
| 1120 |
movel a3,d6 |
| 1121 |
movel a4,a3 |
| 1122 |
#endif |
| 1123 |
addl d7,d3 | |
| 1124 |
addxl d6,d2 | |
| 1125 |
addxl d5,d1 | |
| 1126 |
addxl d4,d0 | |
| 1127 |
|
| 1128 |
movel a2,d4 | return exponent to d4 |
| 1129 |
movel a0,d7 | |
| 1130 |
andl IMM (0x80000000),d7 | d7 now has the sign |
| 1131 |
|
| 1132 |
#ifndef __mcoldfire__ |
| 1133 |
moveml sp@+,a2-a3 |
| 1134 |
#else |
| 1135 |
movel sp@+,a4 |
| 1136 |
movel sp@+,a3 |
| 1137 |
movel sp@+,a2 |
| 1138 |
#endif |
| 1139 |
|
| 1140 |
| Before rounding normalize so bit #DBL_MANT_DIG is set (we will consider |
| 1141 |
| the case of denormalized numbers in the rounding routine itself). |
| 1142 |
| As in the addition (not in the subtraction!) we could have set |
| 1143 |
| one more bit we check this: |
| 1144 |
btst IMM (DBL_MANT_DIG+1),d0 |
| 1145 |
beq 1f |
| 1146 |
#ifndef __mcoldfire__ |
| 1147 |
lsrl IMM (1),d0 |
| 1148 |
roxrl IMM (1),d1 |
| 1149 |
roxrl IMM (1),d2 |
| 1150 |
roxrl IMM (1),d3 |
| 1151 |
addw IMM (1),d4 |
| 1152 |
#else |
| 1153 |
lsrl IMM (1),d3 |
| 1154 |
btst IMM (0),d2 |
| 1155 |
beq 10f |
| 1156 |
bset IMM (31),d3 |
| 1157 |
10: lsrl IMM (1),d2 |
| 1158 |
btst IMM (0),d1 |
| 1159 |
beq 11f |
| 1160 |
bset IMM (31),d2 |
| 1161 |
11: lsrl IMM (1),d1 |
| 1162 |
btst IMM (0),d0 |
| 1163 |
beq 12f |
| 1164 |
bset IMM (31),d1 |
| 1165 |
12: lsrl IMM (1),d0 |
| 1166 |
addl IMM (1),d4 |
| 1167 |
#endif |
| 1168 |
1: |
| 1169 |
lea pc@(Ladddf$5),a0 | to return from rounding routine |
| 1170 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 1171 |
#ifdef __mcoldfire__ |
| 1172 |
clrl d6 |
| 1173 |
#endif |
| 1174 |
movew a1@(6),d6 | rounding mode in d6 |
| 1175 |
beq Lround$to$nearest |
| 1176 |
#ifndef __mcoldfire__ |
| 1177 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 1178 |
#else |
| 1179 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 1180 |
#endif |
| 1181 |
bhi Lround$to$minus |
| 1182 |
blt Lround$to$zero |
| 1183 |
bra Lround$to$plus |
| 1184 |
Ladddf$5: |
| 1185 |
| Put back the exponent and check for overflow |
| 1186 |
#ifndef __mcoldfire__ |
| 1187 |
cmpw IMM (0x7ff),d4 | is the exponent big? |
| 1188 |
#else |
| 1189 |
cmpl IMM (0x7ff),d4 | is the exponent big? |
| 1190 |
#endif |
| 1191 |
bge 1f |
| 1192 |
bclr IMM (DBL_MANT_DIG-1),d0 |
| 1193 |
#ifndef __mcoldfire__ |
| 1194 |
lslw IMM (4),d4 | put exponent back into position |
| 1195 |
#else |
| 1196 |
lsll IMM (4),d4 | put exponent back into position |
| 1197 |
#endif |
| 1198 |
swap d0 | |
| 1199 |
#ifndef __mcoldfire__ |
| 1200 |
orw d4,d0 | |
| 1201 |
#else |
| 1202 |
orl d4,d0 | |
| 1203 |
#endif |
| 1204 |
swap d0 | |
| 1205 |
bra Ladddf$ret |
| 1206 |
1: |
| 1207 |
moveq IMM (ADD),d5 |
| 1208 |
bra Ld$overflow |
| 1209 |
|
| 1210 |
Lsubdf$0: |
| 1211 |
| Here we do the subtraction. |
| 1212 |
#ifndef __mcoldfire__ |
| 1213 |
exg d7,a0 | put sign back in a0 |
| 1214 |
exg d6,a3 | |
| 1215 |
#else |
| 1216 |
movel d7,a4 |
| 1217 |
movel a0,d7 |
| 1218 |
movel a4,a0 |
| 1219 |
movel d6,a4 |
| 1220 |
movel a3,d6 |
| 1221 |
movel a4,a3 |
| 1222 |
#endif |
| 1223 |
subl d7,d3 | |
| 1224 |
subxl d6,d2 | |
| 1225 |
subxl d5,d1 | |
| 1226 |
subxl d4,d0 | |
| 1227 |
beq Ladddf$ret$1 | if zero just exit |
| 1228 |
bpl 1f | if positive skip the following |
| 1229 |
movel a0,d7 | |
| 1230 |
bchg IMM (31),d7 | change sign bit in d7 |
| 1231 |
movel d7,a0 | |
| 1232 |
negl d3 | |
| 1233 |
negxl d2 | |
| 1234 |
negxl d1 | and negate result |
| 1235 |
negxl d0 | |
| 1236 |
1: |
| 1237 |
movel a2,d4 | return exponent to d4 |
| 1238 |
movel a0,d7 |
| 1239 |
andl IMM (0x80000000),d7 | isolate sign bit |
| 1240 |
#ifndef __mcoldfire__ |
| 1241 |
moveml sp@+,a2-a3 | |
| 1242 |
#else |
| 1243 |
movel sp@+,a4 |
| 1244 |
movel sp@+,a3 |
| 1245 |
movel sp@+,a2 |
| 1246 |
#endif |
| 1247 |
|
| 1248 |
| Before rounding normalize so bit #DBL_MANT_DIG is set (we will consider |
| 1249 |
| the case of denormalized numbers in the rounding routine itself). |
| 1250 |
| As in the addition (not in the subtraction!) we could have set |
| 1251 |
| one more bit we check this: |
| 1252 |
btst IMM (DBL_MANT_DIG+1),d0 |
| 1253 |
beq 1f |
| 1254 |
#ifndef __mcoldfire__ |
| 1255 |
lsrl IMM (1),d0 |
| 1256 |
roxrl IMM (1),d1 |
| 1257 |
roxrl IMM (1),d2 |
| 1258 |
roxrl IMM (1),d3 |
| 1259 |
addw IMM (1),d4 |
| 1260 |
#else |
| 1261 |
lsrl IMM (1),d3 |
| 1262 |
btst IMM (0),d2 |
| 1263 |
beq 10f |
| 1264 |
bset IMM (31),d3 |
| 1265 |
10: lsrl IMM (1),d2 |
| 1266 |
btst IMM (0),d1 |
| 1267 |
beq 11f |
| 1268 |
bset IMM (31),d2 |
| 1269 |
11: lsrl IMM (1),d1 |
| 1270 |
btst IMM (0),d0 |
| 1271 |
beq 12f |
| 1272 |
bset IMM (31),d1 |
| 1273 |
12: lsrl IMM (1),d0 |
| 1274 |
addl IMM (1),d4 |
| 1275 |
#endif |
| 1276 |
1: |
| 1277 |
lea pc@(Lsubdf$1),a0 | to return from rounding routine |
| 1278 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 1279 |
#ifdef __mcoldfire__ |
| 1280 |
clrl d6 |
| 1281 |
#endif |
| 1282 |
movew a1@(6),d6 | rounding mode in d6 |
| 1283 |
beq Lround$to$nearest |
| 1284 |
#ifndef __mcoldfire__ |
| 1285 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 1286 |
#else |
| 1287 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 1288 |
#endif |
| 1289 |
bhi Lround$to$minus |
| 1290 |
blt Lround$to$zero |
| 1291 |
bra Lround$to$plus |
| 1292 |
Lsubdf$1: |
| 1293 |
| Put back the exponent and sign (we don't have overflow). ' |
| 1294 |
bclr IMM (DBL_MANT_DIG-1),d0 |
| 1295 |
#ifndef __mcoldfire__ |
| 1296 |
lslw IMM (4),d4 | put exponent back into position |
| 1297 |
#else |
| 1298 |
lsll IMM (4),d4 | put exponent back into position |
| 1299 |
#endif |
| 1300 |
swap d0 | |
| 1301 |
#ifndef __mcoldfire__ |
| 1302 |
orw d4,d0 | |
| 1303 |
#else |
| 1304 |
orl d4,d0 | |
| 1305 |
#endif |
| 1306 |
swap d0 | |
| 1307 |
bra Ladddf$ret |
| 1308 |
|
| 1309 |
| If one of the numbers was too small (difference of exponents >= |
| 1310 |
| DBL_MANT_DIG+1) we return the other (and now we don't have to ' |
| 1311 |
| check for finiteness or zero). |
| 1312 |
Ladddf$a$small: |
| 1313 |
#ifndef __mcoldfire__ |
| 1314 |
moveml sp@+,a2-a3 |
| 1315 |
#else |
| 1316 |
movel sp@+,a4 |
| 1317 |
movel sp@+,a3 |
| 1318 |
movel sp@+,a2 |
| 1319 |
#endif |
| 1320 |
movel a6@(16),d0 |
| 1321 |
movel a6@(20),d1 |
| 1322 |
PICLEA SYM (_fpCCR),a0 |
| 1323 |
movew IMM (0),a0@ |
| 1324 |
#ifndef __mcoldfire__ |
| 1325 |
moveml sp@+,d2-d7 | restore data registers |
| 1326 |
#else |
| 1327 |
moveml sp@,d2-d7 |
| 1328 |
| XXX if frame pointer is ever removed, stack pointer must |
| 1329 |
| be adjusted here. |
| 1330 |
#endif |
| 1331 |
unlk a6 | and return |
| 1332 |
rts |
| 1333 |
|
| 1334 |
Ladddf$b$small: |
| 1335 |
#ifndef __mcoldfire__ |
| 1336 |
moveml sp@+,a2-a3 |
| 1337 |
#else |
| 1338 |
movel sp@+,a4 |
| 1339 |
movel sp@+,a3 |
| 1340 |
movel sp@+,a2 |
| 1341 |
#endif |
| 1342 |
movel a6@(8),d0 |
| 1343 |
movel a6@(12),d1 |
| 1344 |
PICLEA SYM (_fpCCR),a0 |
| 1345 |
movew IMM (0),a0@ |
| 1346 |
#ifndef __mcoldfire__ |
| 1347 |
moveml sp@+,d2-d7 | restore data registers |
| 1348 |
#else |
| 1349 |
moveml sp@,d2-d7 |
| 1350 |
| XXX if frame pointer is ever removed, stack pointer must |
| 1351 |
| be adjusted here. |
| 1352 |
#endif |
| 1353 |
unlk a6 | and return |
| 1354 |
rts |
| 1355 |
|
| 1356 |
Ladddf$a$den: |
| 1357 |
movel d7,d4 | d7 contains 0x00200000 |
| 1358 |
bra Ladddf$1 |
| 1359 |
|
| 1360 |
Ladddf$b$den: |
| 1361 |
movel d7,d5 | d7 contains 0x00200000 |
| 1362 |
notl d6 |
| 1363 |
bra Ladddf$2 |
| 1364 |
|
| 1365 |
Ladddf$b: |
| 1366 |
| Return b (if a is zero) |
| 1367 |
movel d2,d0 |
| 1368 |
movel d3,d1 |
| 1369 |
bne 1f | Check if b is -0 |
| 1370 |
cmpl IMM (0x80000000),d0 |
| 1371 |
bne 1f |
| 1372 |
andl IMM (0x80000000),d7 | Use the sign of a |
| 1373 |
clrl d0 |
| 1374 |
bra Ladddf$ret |
| 1375 |
Ladddf$a: |
| 1376 |
movel a6@(8),d0 |
| 1377 |
movel a6@(12),d1 |
| 1378 |
1: |
| 1379 |
moveq IMM (ADD),d5 |
| 1380 |
| Check for NaN and +/-INFINITY. |
| 1381 |
movel d0,d7 | |
| 1382 |
andl IMM (0x80000000),d7 | |
| 1383 |
bclr IMM (31),d0 | |
| 1384 |
cmpl IMM (0x7ff00000),d0 | |
| 1385 |
bge 2f | |
| 1386 |
movel d0,d0 | check for zero, since we don't ' |
| 1387 |
bne Ladddf$ret | want to return -0 by mistake |
| 1388 |
movel d1,d1 | |
| 1389 |
bne Ladddf$ret | |
| 1390 |
bclr IMM (31),d7 | |
| 1391 |
bra Ladddf$ret | |
| 1392 |
2: |
| 1393 |
andl IMM (0x000fffff),d0 | check for NaN (nonzero fraction) |
| 1394 |
orl d1,d0 | |
| 1395 |
bne Ld$inop | |
| 1396 |
bra Ld$infty | |
| 1397 |
|
| 1398 |
Ladddf$ret$1: |
| 1399 |
#ifndef __mcoldfire__ |
| 1400 |
moveml sp@+,a2-a3 | restore regs and exit |
| 1401 |
#else |
| 1402 |
movel sp@+,a4 |
| 1403 |
movel sp@+,a3 |
| 1404 |
movel sp@+,a2 |
| 1405 |
#endif |
| 1406 |
|
| 1407 |
Ladddf$ret: |
| 1408 |
| Normal exit. |
| 1409 |
PICLEA SYM (_fpCCR),a0 |
| 1410 |
movew IMM (0),a0@ |
| 1411 |
orl d7,d0 | put sign bit back |
| 1412 |
#ifndef __mcoldfire__ |
| 1413 |
moveml sp@+,d2-d7 |
| 1414 |
#else |
| 1415 |
moveml sp@,d2-d7 |
| 1416 |
| XXX if frame pointer is ever removed, stack pointer must |
| 1417 |
| be adjusted here. |
| 1418 |
#endif |
| 1419 |
unlk a6 |
| 1420 |
rts |
| 1421 |
|
| 1422 |
Ladddf$ret$den: |
| 1423 |
| Return a denormalized number. |
| 1424 |
#ifndef __mcoldfire__ |
| 1425 |
lsrl IMM (1),d0 | shift right once more |
| 1426 |
roxrl IMM (1),d1 | |
| 1427 |
#else |
| 1428 |
lsrl IMM (1),d1 |
| 1429 |
btst IMM (0),d0 |
| 1430 |
beq 10f |
| 1431 |
bset IMM (31),d1 |
| 1432 |
10: lsrl IMM (1),d0 |
| 1433 |
#endif |
| 1434 |
bra Ladddf$ret |
| 1435 |
|
| 1436 |
Ladddf$nf: |
| 1437 |
moveq IMM (ADD),d5 |
| 1438 |
| This could be faster but it is not worth the effort, since it is not |
| 1439 |
| executed very often. We sacrifice speed for clarity here. |
| 1440 |
movel a6@(8),d0 | get the numbers back (remember that we |
| 1441 |
movel a6@(12),d1 | did some processing already) |
| 1442 |
movel a6@(16),d2 | |
| 1443 |
movel a6@(20),d3 | |
| 1444 |
movel IMM (0x7ff00000),d4 | useful constant (INFINITY) |
| 1445 |
movel d0,d7 | save sign bits |
| 1446 |
movel d2,d6 | |
| 1447 |
bclr IMM (31),d0 | clear sign bits |
| 1448 |
bclr IMM (31),d2 | |
| 1449 |
| We know that one of them is either NaN of +/-INFINITY |
| 1450 |
| Check for NaN (if either one is NaN return NaN) |
| 1451 |
cmpl d4,d0 | check first a (d0) |
| 1452 |
bhi Ld$inop | if d0 > 0x7ff00000 or equal and |
| 1453 |
bne 2f |
| 1454 |
tstl d1 | d1 > 0, a is NaN |
| 1455 |
bne Ld$inop | |
| 1456 |
2: cmpl d4,d2 | check now b (d1) |
| 1457 |
bhi Ld$inop | |
| 1458 |
bne 3f |
| 1459 |
tstl d3 | |
| 1460 |
bne Ld$inop | |
| 1461 |
3: |
| 1462 |
| Now comes the check for +/-INFINITY. We know that both are (maybe not |
| 1463 |
| finite) numbers, but we have to check if both are infinite whether we |
| 1464 |
| are adding or subtracting them. |
| 1465 |
eorl d7,d6 | to check sign bits |
| 1466 |
bmi 1f |
| 1467 |
andl IMM (0x80000000),d7 | get (common) sign bit |
| 1468 |
bra Ld$infty |
| 1469 |
1: |
| 1470 |
| We know one (or both) are infinite, so we test for equality between the |
| 1471 |
| two numbers (if they are equal they have to be infinite both, so we |
| 1472 |
| return NaN). |
| 1473 |
cmpl d2,d0 | are both infinite? |
| 1474 |
bne 1f | if d0 <> d2 they are not equal |
| 1475 |
cmpl d3,d1 | if d0 == d2 test d3 and d1 |
| 1476 |
beq Ld$inop | if equal return NaN |
| 1477 |
1: |
| 1478 |
andl IMM (0x80000000),d7 | get a's sign bit ' |
| 1479 |
cmpl d4,d0 | test now for infinity |
| 1480 |
beq Ld$infty | if a is INFINITY return with this sign |
| 1481 |
bchg IMM (31),d7 | else we know b is INFINITY and has |
| 1482 |
bra Ld$infty | the opposite sign |
| 1483 |
|
| 1484 |
|============================================================================= |
| 1485 |
| __muldf3 |
| 1486 |
|============================================================================= |
| 1487 |
|
| 1488 |
| double __muldf3(double, double); |
| 1489 |
FUNC(__muldf3) |
| 1490 |
SYM (__muldf3): |
| 1491 |
#ifndef __mcoldfire__ |
| 1492 |
link a6,IMM (0) |
| 1493 |
moveml d2-d7,sp@- |
| 1494 |
#else |
| 1495 |
link a6,IMM (-24) |
| 1496 |
moveml d2-d7,sp@ |
| 1497 |
#endif |
| 1498 |
movel a6@(8),d0 | get a into d0-d1 |
| 1499 |
movel a6@(12),d1 | |
| 1500 |
movel a6@(16),d2 | and b into d2-d3 |
| 1501 |
movel a6@(20),d3 | |
| 1502 |
movel d0,d7 | d7 will hold the sign of the product |
| 1503 |
eorl d2,d7 | |
| 1504 |
andl IMM (0x80000000),d7 | |
| 1505 |
movel d7,a0 | save sign bit into a0 |
| 1506 |
movel IMM (0x7ff00000),d7 | useful constant (+INFINITY) |
| 1507 |
movel d7,d6 | another (mask for fraction) |
| 1508 |
notl d6 | |
| 1509 |
bclr IMM (31),d0 | get rid of a's sign bit ' |
| 1510 |
movel d0,d4 | |
| 1511 |
orl d1,d4 | |
| 1512 |
beq Lmuldf$a$0 | branch if a is zero |
| 1513 |
movel d0,d4 | |
| 1514 |
bclr IMM (31),d2 | get rid of b's sign bit ' |
| 1515 |
movel d2,d5 | |
| 1516 |
orl d3,d5 | |
| 1517 |
beq Lmuldf$b$0 | branch if b is zero |
| 1518 |
movel d2,d5 | |
| 1519 |
cmpl d7,d0 | is a big? |
| 1520 |
bhi Lmuldf$inop | if a is NaN return NaN |
| 1521 |
beq Lmuldf$a$nf | we still have to check d1 and b ... |
| 1522 |
cmpl d7,d2 | now compare b with INFINITY |
| 1523 |
bhi Lmuldf$inop | is b NaN? |
| 1524 |
beq Lmuldf$b$nf | we still have to check d3 ... |
| 1525 |
| Here we have both numbers finite and nonzero (and with no sign bit). |
| 1526 |
| Now we get the exponents into d4 and d5. |
| 1527 |
andl d7,d4 | isolate exponent in d4 |
| 1528 |
beq Lmuldf$a$den | if exponent zero, have denormalized |
| 1529 |
andl d6,d0 | isolate fraction |
| 1530 |
orl IMM (0x00100000),d0 | and put hidden bit back |
| 1531 |
swap d4 | I like exponents in the first byte |
| 1532 |
#ifndef __mcoldfire__ |
| 1533 |
lsrw IMM (4),d4 | |
| 1534 |
#else |
| 1535 |
lsrl IMM (4),d4 | |
| 1536 |
#endif |
| 1537 |
Lmuldf$1: |
| 1538 |
andl d7,d5 | |
| 1539 |
beq Lmuldf$b$den | |
| 1540 |
andl d6,d2 | |
| 1541 |
orl IMM (0x00100000),d2 | and put hidden bit back |
| 1542 |
swap d5 | |
| 1543 |
#ifndef __mcoldfire__ |
| 1544 |
lsrw IMM (4),d5 | |
| 1545 |
#else |
| 1546 |
lsrl IMM (4),d5 | |
| 1547 |
#endif |
| 1548 |
Lmuldf$2: | |
| 1549 |
#ifndef __mcoldfire__ |
| 1550 |
addw d5,d4 | add exponents |
| 1551 |
subw IMM (D_BIAS+1),d4 | and subtract bias (plus one) |
| 1552 |
#else |
| 1553 |
addl d5,d4 | add exponents |
| 1554 |
subl IMM (D_BIAS+1),d4 | and subtract bias (plus one) |
| 1555 |
#endif |
| 1556 |
|
| 1557 |
| We are now ready to do the multiplication. The situation is as follows: |
| 1558 |
| both a and b have bit 52 ( bit 20 of d0 and d2) set (even if they were |
| 1559 |
| denormalized to start with!), which means that in the product bit 104 |
| 1560 |
| (which will correspond to bit 8 of the fourth long) is set. |
| 1561 |
|
| 1562 |
| Here we have to do the product. |
| 1563 |
| To do it we have to juggle the registers back and forth, as there are not |
| 1564 |
| enough to keep everything in them. So we use the address registers to keep |
| 1565 |
| some intermediate data. |
| 1566 |
|
| 1567 |
#ifndef __mcoldfire__ |
| 1568 |
moveml a2-a3,sp@- | save a2 and a3 for temporary use |
| 1569 |
#else |
| 1570 |
movel a2,sp@- |
| 1571 |
movel a3,sp@- |
| 1572 |
movel a4,sp@- |
| 1573 |
#endif |
| 1574 |
movel IMM (0),a2 | a2 is a null register |
| 1575 |
movel d4,a3 | and a3 will preserve the exponent |
| 1576 |
|
| 1577 |
| First, shift d2-d3 so bit 20 becomes bit 31: |
| 1578 |
#ifndef __mcoldfire__ |
| 1579 |
rorl IMM (5),d2 | rotate d2 5 places right |
| 1580 |
swap d2 | and swap it |
| 1581 |
rorl IMM (5),d3 | do the same thing with d3 |
| 1582 |
swap d3 | |
| 1583 |
movew d3,d6 | get the rightmost 11 bits of d3 |
| 1584 |
andw IMM (0x07ff),d6 | |
| 1585 |
orw d6,d2 | and put them into d2 |
| 1586 |
andw IMM (0xf800),d3 | clear those bits in d3 |
| 1587 |
#else |
| 1588 |
moveq IMM (11),d7 | left shift d2 11 bits |
| 1589 |
lsll d7,d2 |
| 1590 |
movel d3,d6 | get a copy of d3 |
| 1591 |
lsll d7,d3 | left shift d3 11 bits |
| 1592 |
andl IMM (0xffe00000),d6 | get the top 11 bits of d3 |
| 1593 |
moveq IMM (21),d7 | right shift them 21 bits |
| 1594 |
lsrl d7,d6 |
| 1595 |
orl d6,d2 | stick them at the end of d2 |
| 1596 |
#endif |
| 1597 |
|
| 1598 |
movel d2,d6 | move b into d6-d7 |
| 1599 |
movel d3,d7 | move a into d4-d5 |
| 1600 |
movel d0,d4 | and clear d0-d1-d2-d3 (to put result) |
| 1601 |
movel d1,d5 | |
| 1602 |
movel IMM (0),d3 | |
| 1603 |
movel d3,d2 | |
| 1604 |
movel d3,d1 | |
| 1605 |
movel d3,d0 | |
| 1606 |
|
| 1607 |
| We use a1 as counter: |
| 1608 |
movel IMM (DBL_MANT_DIG-1),a1 |
| 1609 |
#ifndef __mcoldfire__ |
| 1610 |
exg d7,a1 |
| 1611 |
#else |
| 1612 |
movel d7,a4 |
| 1613 |
movel a1,d7 |
| 1614 |
movel a4,a1 |
| 1615 |
#endif |
| 1616 |
|
| 1617 |
1: |
| 1618 |
#ifndef __mcoldfire__ |
| 1619 |
exg d7,a1 | put counter back in a1 |
| 1620 |
#else |
| 1621 |
movel d7,a4 |
| 1622 |
movel a1,d7 |
| 1623 |
movel a4,a1 |
| 1624 |
#endif |
| 1625 |
addl d3,d3 | shift sum once left |
| 1626 |
addxl d2,d2 | |
| 1627 |
addxl d1,d1 | |
| 1628 |
addxl d0,d0 | |
| 1629 |
addl d7,d7 | |
| 1630 |
addxl d6,d6 | |
| 1631 |
bcc 2f | if bit clear skip the following |
| 1632 |
#ifndef __mcoldfire__ |
| 1633 |
exg d7,a2 | |
| 1634 |
#else |
| 1635 |
movel d7,a4 |
| 1636 |
movel a2,d7 |
| 1637 |
movel a4,a2 |
| 1638 |
#endif |
| 1639 |
addl d5,d3 | else add a to the sum |
| 1640 |
addxl d4,d2 | |
| 1641 |
addxl d7,d1 | |
| 1642 |
addxl d7,d0 | |
| 1643 |
#ifndef __mcoldfire__ |
| 1644 |
exg d7,a2 | |
| 1645 |
#else |
| 1646 |
movel d7,a4 |
| 1647 |
movel a2,d7 |
| 1648 |
movel a4,a2 |
| 1649 |
#endif |
| 1650 |
2: |
| 1651 |
#ifndef __mcoldfire__ |
| 1652 |
exg d7,a1 | put counter in d7 |
| 1653 |
dbf d7,1b | decrement and branch |
| 1654 |
#else |
| 1655 |
movel d7,a4 |
| 1656 |
movel a1,d7 |
| 1657 |
movel a4,a1 |
| 1658 |
subql IMM (1),d7 |
| 1659 |
bpl 1b |
| 1660 |
#endif |
| 1661 |
|
| 1662 |
movel a3,d4 | restore exponent |
| 1663 |
#ifndef __mcoldfire__ |
| 1664 |
moveml sp@+,a2-a3 |
| 1665 |
#else |
| 1666 |
movel sp@+,a4 |
| 1667 |
movel sp@+,a3 |
| 1668 |
movel sp@+,a2 |
| 1669 |
#endif |
| 1670 |
|
| 1671 |
| Now we have the product in d0-d1-d2-d3, with bit 8 of d0 set. The |
| 1672 |
| first thing to do now is to normalize it so bit 8 becomes bit |
| 1673 |
| DBL_MANT_DIG-32 (to do the rounding); later we will shift right. |
| 1674 |
swap d0 |
| 1675 |
swap d1 |
| 1676 |
movew d1,d0 |
| 1677 |
swap d2 |
| 1678 |
movew d2,d1 |
| 1679 |
swap d3 |
| 1680 |
movew d3,d2 |
| 1681 |
movew IMM (0),d3 |
| 1682 |
#ifndef __mcoldfire__ |
| 1683 |
lsrl IMM (1),d0 |
| 1684 |
roxrl IMM (1),d1 |
| 1685 |
roxrl IMM (1),d2 |
| 1686 |
roxrl IMM (1),d3 |
| 1687 |
lsrl IMM (1),d0 |
| 1688 |
roxrl IMM (1),d1 |
| 1689 |
roxrl IMM (1),d2 |
| 1690 |
roxrl IMM (1),d3 |
| 1691 |
lsrl IMM (1),d0 |
| 1692 |
roxrl IMM (1),d1 |
| 1693 |
roxrl IMM (1),d2 |
| 1694 |
roxrl IMM (1),d3 |
| 1695 |
#else |
| 1696 |
moveq IMM (29),d6 |
| 1697 |
lsrl IMM (3),d3 |
| 1698 |
movel d2,d7 |
| 1699 |
lsll d6,d7 |
| 1700 |
orl d7,d3 |
| 1701 |
lsrl IMM (3),d2 |
| 1702 |
movel d1,d7 |
| 1703 |
lsll d6,d7 |
| 1704 |
orl d7,d2 |
| 1705 |
lsrl IMM (3),d1 |
| 1706 |
movel d0,d7 |
| 1707 |
lsll d6,d7 |
| 1708 |
orl d7,d1 |
| 1709 |
lsrl IMM (3),d0 |
| 1710 |
#endif |
| 1711 |
|
| 1712 |
| Now round, check for over- and underflow, and exit. |
| 1713 |
movel a0,d7 | get sign bit back into d7 |
| 1714 |
moveq IMM (MULTIPLY),d5 |
| 1715 |
|
| 1716 |
btst IMM (DBL_MANT_DIG+1-32),d0 |
| 1717 |
beq Lround$exit |
| 1718 |
#ifndef __mcoldfire__ |
| 1719 |
lsrl IMM (1),d0 |
| 1720 |
roxrl IMM (1),d1 |
| 1721 |
addw IMM (1),d4 |
| 1722 |
#else |
| 1723 |
lsrl IMM (1),d1 |
| 1724 |
btst IMM (0),d0 |
| 1725 |
beq 10f |
| 1726 |
bset IMM (31),d1 |
| 1727 |
10: lsrl IMM (1),d0 |
| 1728 |
addl IMM (1),d4 |
| 1729 |
#endif |
| 1730 |
bra Lround$exit |
| 1731 |
|
| 1732 |
Lmuldf$inop: |
| 1733 |
moveq IMM (MULTIPLY),d5 |
| 1734 |
bra Ld$inop |
| 1735 |
|
| 1736 |
Lmuldf$b$nf: |
| 1737 |
moveq IMM (MULTIPLY),d5 |
| 1738 |
movel a0,d7 | get sign bit back into d7 |
| 1739 |
tstl d3 | we know d2 == 0x7ff00000, so check d3 |
| 1740 |
bne Ld$inop | if d3 <> 0 b is NaN |
| 1741 |
bra Ld$overflow | else we have overflow (since a is finite) |
| 1742 |
|
| 1743 |
Lmuldf$a$nf: |
| 1744 |
moveq IMM (MULTIPLY),d5 |
| 1745 |
movel a0,d7 | get sign bit back into d7 |
| 1746 |
tstl d1 | we know d0 == 0x7ff00000, so check d1 |
| 1747 |
bne Ld$inop | if d1 <> 0 a is NaN |
| 1748 |
bra Ld$overflow | else signal overflow |
| 1749 |
|
| 1750 |
| If either number is zero return zero, unless the other is +/-INFINITY or |
| 1751 |
| NaN, in which case we return NaN. |
| 1752 |
Lmuldf$b$0: |
| 1753 |
moveq IMM (MULTIPLY),d5 |
| 1754 |
#ifndef __mcoldfire__ |
| 1755 |
exg d2,d0 | put b (==0) into d0-d1 |
| 1756 |
exg d3,d1 | and a (with sign bit cleared) into d2-d3 |
| 1757 |
movel a0,d0 | set result sign |
| 1758 |
#else |
| 1759 |
movel d0,d2 | put a into d2-d3 |
| 1760 |
movel d1,d3 |
| 1761 |
movel a0,d0 | put result zero into d0-d1 |
| 1762 |
movq IMM(0),d1 |
| 1763 |
#endif |
| 1764 |
bra 1f |
| 1765 |
Lmuldf$a$0: |
| 1766 |
movel a0,d0 | set result sign |
| 1767 |
movel a6@(16),d2 | put b into d2-d3 again |
| 1768 |
movel a6@(20),d3 | |
| 1769 |
bclr IMM (31),d2 | clear sign bit |
| 1770 |
1: cmpl IMM (0x7ff00000),d2 | check for non-finiteness |
| 1771 |
bge Ld$inop | in case NaN or +/-INFINITY return NaN |
| 1772 |
PICLEA SYM (_fpCCR),a0 |
| 1773 |
movew IMM (0),a0@ |
| 1774 |
#ifndef __mcoldfire__ |
| 1775 |
moveml sp@+,d2-d7 |
| 1776 |
#else |
| 1777 |
moveml sp@,d2-d7 |
| 1778 |
| XXX if frame pointer is ever removed, stack pointer must |
| 1779 |
| be adjusted here. |
| 1780 |
#endif |
| 1781 |
unlk a6 |
| 1782 |
rts |
| 1783 |
|
| 1784 |
| If a number is denormalized we put an exponent of 1 but do not put the |
| 1785 |
| hidden bit back into the fraction; instead we shift left until bit 21 |
| 1786 |
| (the hidden bit) is set, adjusting the exponent accordingly. We do this |
| 1787 |
| to ensure that the product of the fractions is close to 1. |
| 1788 |
Lmuldf$a$den: |
| 1789 |
movel IMM (1),d4 |
| 1790 |
andl d6,d0 |
| 1791 |
1: addl d1,d1 | shift a left until bit 20 is set |
| 1792 |
addxl d0,d0 | |
| 1793 |
#ifndef __mcoldfire__ |
| 1794 |
subw IMM (1),d4 | and adjust exponent |
| 1795 |
#else |
| 1796 |
subl IMM (1),d4 | and adjust exponent |
| 1797 |
#endif |
| 1798 |
btst IMM (20),d0 | |
| 1799 |
bne Lmuldf$1 | |
| 1800 |
bra 1b |
| 1801 |
|
| 1802 |
Lmuldf$b$den: |
| 1803 |
movel IMM (1),d5 |
| 1804 |
andl d6,d2 |
| 1805 |
1: addl d3,d3 | shift b left until bit 20 is set |
| 1806 |
addxl d2,d2 | |
| 1807 |
#ifndef __mcoldfire__ |
| 1808 |
subw IMM (1),d5 | and adjust exponent |
| 1809 |
#else |
| 1810 |
subql IMM (1),d5 | and adjust exponent |
| 1811 |
#endif |
| 1812 |
btst IMM (20),d2 | |
| 1813 |
bne Lmuldf$2 | |
| 1814 |
bra 1b |
| 1815 |
|
| 1816 |
|
| 1817 |
|============================================================================= |
| 1818 |
| __divdf3 |
| 1819 |
|============================================================================= |
| 1820 |
|
| 1821 |
| double __divdf3(double, double); |
| 1822 |
FUNC(__divdf3) |
| 1823 |
SYM (__divdf3): |
| 1824 |
#ifndef __mcoldfire__ |
| 1825 |
link a6,IMM (0) |
| 1826 |
moveml d2-d7,sp@- |
| 1827 |
#else |
| 1828 |
link a6,IMM (-24) |
| 1829 |
moveml d2-d7,sp@ |
| 1830 |
#endif |
| 1831 |
movel a6@(8),d0 | get a into d0-d1 |
| 1832 |
movel a6@(12),d1 | |
| 1833 |
movel a6@(16),d2 | and b into d2-d3 |
| 1834 |
movel a6@(20),d3 | |
| 1835 |
movel d0,d7 | d7 will hold the sign of the result |
| 1836 |
eorl d2,d7 | |
| 1837 |
andl IMM (0x80000000),d7 |
| 1838 |
movel d7,a0 | save sign into a0 |
| 1839 |
movel IMM (0x7ff00000),d7 | useful constant (+INFINITY) |
| 1840 |
movel d7,d6 | another (mask for fraction) |
| 1841 |
notl d6 | |
| 1842 |
bclr IMM (31),d0 | get rid of a's sign bit ' |
| 1843 |
movel d0,d4 | |
| 1844 |
orl d1,d4 | |
| 1845 |
beq Ldivdf$a$0 | branch if a is zero |
| 1846 |
movel d0,d4 | |
| 1847 |
bclr IMM (31),d2 | get rid of b's sign bit ' |
| 1848 |
movel d2,d5 | |
| 1849 |
orl d3,d5 | |
| 1850 |
beq Ldivdf$b$0 | branch if b is zero |
| 1851 |
movel d2,d5 |
| 1852 |
cmpl d7,d0 | is a big? |
| 1853 |
bhi Ldivdf$inop | if a is NaN return NaN |
| 1854 |
beq Ldivdf$a$nf | if d0 == 0x7ff00000 we check d1 |
| 1855 |
cmpl d7,d2 | now compare b with INFINITY |
| 1856 |
bhi Ldivdf$inop | if b is NaN return NaN |
| 1857 |
beq Ldivdf$b$nf | if d2 == 0x7ff00000 we check d3 |
| 1858 |
| Here we have both numbers finite and nonzero (and with no sign bit). |
| 1859 |
| Now we get the exponents into d4 and d5 and normalize the numbers to |
| 1860 |
| ensure that the ratio of the fractions is around 1. We do this by |
| 1861 |
| making sure that both numbers have bit #DBL_MANT_DIG-32-1 (hidden bit) |
| 1862 |
| set, even if they were denormalized to start with. |
| 1863 |
| Thus, the result will satisfy: 2 > result > 1/2. |
| 1864 |
andl d7,d4 | and isolate exponent in d4 |
| 1865 |
beq Ldivdf$a$den | if exponent is zero we have a denormalized |
| 1866 |
andl d6,d0 | and isolate fraction |
| 1867 |
orl IMM (0x00100000),d0 | and put hidden bit back |
| 1868 |
swap d4 | I like exponents in the first byte |
| 1869 |
#ifndef __mcoldfire__ |
| 1870 |
lsrw IMM (4),d4 | |
| 1871 |
#else |
| 1872 |
lsrl IMM (4),d4 | |
| 1873 |
#endif |
| 1874 |
Ldivdf$1: | |
| 1875 |
andl d7,d5 | |
| 1876 |
beq Ldivdf$b$den | |
| 1877 |
andl d6,d2 | |
| 1878 |
orl IMM (0x00100000),d2 |
| 1879 |
swap d5 | |
| 1880 |
#ifndef __mcoldfire__ |
| 1881 |
lsrw IMM (4),d5 | |
| 1882 |
#else |
| 1883 |
lsrl IMM (4),d5 | |
| 1884 |
#endif |
| 1885 |
Ldivdf$2: | |
| 1886 |
#ifndef __mcoldfire__ |
| 1887 |
subw d5,d4 | subtract exponents |
| 1888 |
addw IMM (D_BIAS),d4 | and add bias |
| 1889 |
#else |
| 1890 |
subl d5,d4 | subtract exponents |
| 1891 |
addl IMM (D_BIAS),d4 | and add bias |
| 1892 |
#endif |
| 1893 |
|
| 1894 |
| We are now ready to do the division. We have prepared things in such a way |
| 1895 |
| that the ratio of the fractions will be less than 2 but greater than 1/2. |
| 1896 |
| At this point the registers in use are: |
| 1897 |
| d0-d1 hold a (first operand, bit DBL_MANT_DIG-32=0, bit |
| 1898 |
| DBL_MANT_DIG-1-32=1) |
| 1899 |
| d2-d3 hold b (second operand, bit DBL_MANT_DIG-32=1) |
| 1900 |
| d4 holds the difference of the exponents, corrected by the bias |
| 1901 |
| a0 holds the sign of the ratio |
| 1902 |
|
| 1903 |
| To do the rounding correctly we need to keep information about the |
| 1904 |
| nonsignificant bits. One way to do this would be to do the division |
| 1905 |
| using four registers; another is to use two registers (as originally |
| 1906 |
| I did), but use a sticky bit to preserve information about the |
| 1907 |
| fractional part. Note that we can keep that info in a1, which is not |
| 1908 |
| used. |
| 1909 |
movel IMM (0),d6 | d6-d7 will hold the result |
| 1910 |
movel d6,d7 | |
| 1911 |
movel IMM (0),a1 | and a1 will hold the sticky bit |
| 1912 |
|
| 1913 |
movel IMM (DBL_MANT_DIG-32+1),d5 |
| 1914 |
|
| 1915 |
1: cmpl d0,d2 | is a < b? |
| 1916 |
bhi 3f | if b > a skip the following |
| 1917 |
beq 4f | if d0==d2 check d1 and d3 |
| 1918 |
2: subl d3,d1 | |
| 1919 |
subxl d2,d0 | a <-- a - b |
| 1920 |
bset d5,d6 | set the corresponding bit in d6 |
| 1921 |
3: addl d1,d1 | shift a by 1 |
| 1922 |
addxl d0,d0 | |
| 1923 |
#ifndef __mcoldfire__ |
| 1924 |
dbra d5,1b | and branch back |
| 1925 |
#else |
| 1926 |
subql IMM (1), d5 |
| 1927 |
bpl 1b |
| 1928 |
#endif |
| 1929 |
bra 5f |
| 1930 |
4: cmpl d1,d3 | here d0==d2, so check d1 and d3 |
| 1931 |
bhi 3b | if d1 > d2 skip the subtraction |
| 1932 |
bra 2b | else go do it |
| 1933 |
5: |
| 1934 |
| Here we have to start setting the bits in the second long. |
| 1935 |
movel IMM (31),d5 | again d5 is counter |
| 1936 |
|
| 1937 |
1: cmpl d0,d2 | is a < b? |
| 1938 |
bhi 3f | if b > a skip the following |
| 1939 |
beq 4f | if d0==d2 check d1 and d3 |
| 1940 |
2: subl d3,d1 | |
| 1941 |
subxl d2,d0 | a <-- a - b |
| 1942 |
bset d5,d7 | set the corresponding bit in d7 |
| 1943 |
3: addl d1,d1 | shift a by 1 |
| 1944 |
addxl d0,d0 | |
| 1945 |
#ifndef __mcoldfire__ |
| 1946 |
dbra d5,1b | and branch back |
| 1947 |
#else |
| 1948 |
subql IMM (1), d5 |
| 1949 |
bpl 1b |
| 1950 |
#endif |
| 1951 |
bra 5f |
| 1952 |
4: cmpl d1,d3 | here d0==d2, so check d1 and d3 |
| 1953 |
bhi 3b | if d1 > d2 skip the subtraction |
| 1954 |
bra 2b | else go do it |
| 1955 |
5: |
| 1956 |
| Now go ahead checking until we hit a one, which we store in d2. |
| 1957 |
movel IMM (DBL_MANT_DIG),d5 |
| 1958 |
1: cmpl d2,d0 | is a < b? |
| 1959 |
bhi 4f | if b < a, exit |
| 1960 |
beq 3f | if d0==d2 check d1 and d3 |
| 1961 |
2: addl d1,d1 | shift a by 1 |
| 1962 |
addxl d0,d0 | |
| 1963 |
#ifndef __mcoldfire__ |
| 1964 |
dbra d5,1b | and branch back |
| 1965 |
#else |
| 1966 |
subql IMM (1), d5 |
| 1967 |
bpl 1b |
| 1968 |
#endif |
| 1969 |
movel IMM (0),d2 | here no sticky bit was found |
| 1970 |
movel d2,d3 |
| 1971 |
bra 5f |
| 1972 |
3: cmpl d1,d3 | here d0==d2, so check d1 and d3 |
| 1973 |
bhi 2b | if d1 > d2 go back |
| 1974 |
4: |
| 1975 |
| Here put the sticky bit in d2-d3 (in the position which actually corresponds |
| 1976 |
| to it; if you don't do this the algorithm loses in some cases). ' |
| 1977 |
movel IMM (0),d2 |
| 1978 |
movel d2,d3 |
| 1979 |
#ifndef __mcoldfire__ |
| 1980 |
subw IMM (DBL_MANT_DIG),d5 |
| 1981 |
addw IMM (63),d5 |
| 1982 |
cmpw IMM (31),d5 |
| 1983 |
#else |
| 1984 |
subl IMM (DBL_MANT_DIG),d5 |
| 1985 |
addl IMM (63),d5 |
| 1986 |
cmpl IMM (31),d5 |
| 1987 |
#endif |
| 1988 |
bhi 2f |
| 1989 |
1: bset d5,d3 |
| 1990 |
bra 5f |
| 1991 |
#ifndef __mcoldfire__ |
| 1992 |
subw IMM (32),d5 |
| 1993 |
#else |
| 1994 |
subl IMM (32),d5 |
| 1995 |
#endif |
| 1996 |
2: bset d5,d2 |
| 1997 |
5: |
| 1998 |
| Finally we are finished! Move the longs in the address registers to |
| 1999 |
| their final destination: |
| 2000 |
movel d6,d0 |
| 2001 |
movel d7,d1 |
| 2002 |
movel IMM (0),d3 |
| 2003 |
|
| 2004 |
| Here we have finished the division, with the result in d0-d1-d2-d3, with |
| 2005 |
| 2^21 <= d6 < 2^23. Thus bit 23 is not set, but bit 22 could be set. |
| 2006 |
| If it is not, then definitely bit 21 is set. Normalize so bit 22 is |
| 2007 |
| not set: |
| 2008 |
btst IMM (DBL_MANT_DIG-32+1),d0 |
| 2009 |
beq 1f |
| 2010 |
#ifndef __mcoldfire__ |
| 2011 |
lsrl IMM (1),d0 |
| 2012 |
roxrl IMM (1),d1 |
| 2013 |
roxrl IMM (1),d2 |
| 2014 |
roxrl IMM (1),d3 |
| 2015 |
addw IMM (1),d4 |
| 2016 |
#else |
| 2017 |
lsrl IMM (1),d3 |
| 2018 |
btst IMM (0),d2 |
| 2019 |
beq 10f |
| 2020 |
bset IMM (31),d3 |
| 2021 |
10: lsrl IMM (1),d2 |
| 2022 |
btst IMM (0),d1 |
| 2023 |
beq 11f |
| 2024 |
bset IMM (31),d2 |
| 2025 |
11: lsrl IMM (1),d1 |
| 2026 |
btst IMM (0),d0 |
| 2027 |
beq 12f |
| 2028 |
bset IMM (31),d1 |
| 2029 |
12: lsrl IMM (1),d0 |
| 2030 |
addl IMM (1),d4 |
| 2031 |
#endif |
| 2032 |
1: |
| 2033 |
| Now round, check for over- and underflow, and exit. |
| 2034 |
movel a0,d7 | restore sign bit to d7 |
| 2035 |
moveq IMM (DIVIDE),d5 |
| 2036 |
bra Lround$exit |
| 2037 |
|
| 2038 |
Ldivdf$inop: |
| 2039 |
moveq IMM (DIVIDE),d5 |
| 2040 |
bra Ld$inop |
| 2041 |
|
| 2042 |
Ldivdf$a$0: |
| 2043 |
| If a is zero check to see whether b is zero also. In that case return |
| 2044 |
| NaN; then check if b is NaN, and return NaN also in that case. Else |
| 2045 |
| return a properly signed zero. |
| 2046 |
moveq IMM (DIVIDE),d5 |
| 2047 |
bclr IMM (31),d2 | |
| 2048 |
movel d2,d4 | |
| 2049 |
orl d3,d4 | |
| 2050 |
beq Ld$inop | if b is also zero return NaN |
| 2051 |
cmpl IMM (0x7ff00000),d2 | check for NaN |
| 2052 |
bhi Ld$inop | |
| 2053 |
blt 1f | |
| 2054 |
tstl d3 | |
| 2055 |
bne Ld$inop | |
| 2056 |
1: movel a0,d0 | else return signed zero |
| 2057 |
moveq IMM(0),d1 | |
| 2058 |
PICLEA SYM (_fpCCR),a0 | clear exception flags |
| 2059 |
movew IMM (0),a0@ | |
| 2060 |
#ifndef __mcoldfire__ |
| 2061 |
moveml sp@+,d2-d7 | |
| 2062 |
#else |
| 2063 |
moveml sp@,d2-d7 | |
| 2064 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2065 |
| be adjusted here. |
| 2066 |
#endif |
| 2067 |
unlk a6 | |
| 2068 |
rts | |
| 2069 |
|
| 2070 |
Ldivdf$b$0: |
| 2071 |
moveq IMM (DIVIDE),d5 |
| 2072 |
| If we got here a is not zero. Check if a is NaN; in that case return NaN, |
| 2073 |
| else return +/-INFINITY. Remember that a is in d0 with the sign bit |
| 2074 |
| cleared already. |
| 2075 |
movel a0,d7 | put a's sign bit back in d7 ' |
| 2076 |
cmpl IMM (0x7ff00000),d0 | compare d0 with INFINITY |
| 2077 |
bhi Ld$inop | if larger it is NaN |
| 2078 |
tstl d1 | |
| 2079 |
bne Ld$inop | |
| 2080 |
bra Ld$div$0 | else signal DIVIDE_BY_ZERO |
| 2081 |
|
| 2082 |
Ldivdf$b$nf: |
| 2083 |
moveq IMM (DIVIDE),d5 |
| 2084 |
| If d2 == 0x7ff00000 we have to check d3. |
| 2085 |
tstl d3 | |
| 2086 |
bne Ld$inop | if d3 <> 0, b is NaN |
| 2087 |
movel a0,d7 | put a's sign |
| 2088 |
bra Ld$underflow | else b is +/-INFINITY, so signal underflow |
| 2089 |
|
| 2090 |
Ldivdf$a$nf: |
| 2091 |
moveq IMM (DIVIDE),d5 |
| 2092 |
| If d0 == 0x7ff00000 we have to check d1. |
| 2093 |
tstl d1 | |
| 2094 |
bne Ld$inop | if d1 <> 0, a is NaN |
| 2095 |
| If a is INFINITY we have to check b |
| 2096 |
cmpl d7,d2 | compare b with INFINITY |
| 2097 |
bge Ld$inop | if b is NaN or INFINITY return NaN |
| 2098 |
movl a0,d7 | restore sign bit to d7 |
| 2099 |
bra Ld$overflow | else return overflow |
| 2100 |
|
| 2101 |
| If a number is denormalized we put an exponent of 1 but do not put the |
| 2102 |
| bit back into the fraction. |
| 2103 |
Ldivdf$a$den: |
| 2104 |
movel IMM (1),d4 |
| 2105 |
andl d6,d0 |
| 2106 |
1: addl d1,d1 | shift a left until bit 20 is set |
| 2107 |
addxl d0,d0 |
| 2108 |
#ifndef __mcoldfire__ |
| 2109 |
subw IMM (1),d4 | and adjust exponent |
| 2110 |
#else |
| 2111 |
subl IMM (1),d4 | and adjust exponent |
| 2112 |
#endif |
| 2113 |
btst IMM (DBL_MANT_DIG-32-1),d0 |
| 2114 |
bne Ldivdf$1 |
| 2115 |
bra 1b |
| 2116 |
|
| 2117 |
Ldivdf$b$den: |
| 2118 |
movel IMM (1),d5 |
| 2119 |
andl d6,d2 |
| 2120 |
1: addl d3,d3 | shift b left until bit 20 is set |
| 2121 |
addxl d2,d2 |
| 2122 |
#ifndef __mcoldfire__ |
| 2123 |
subw IMM (1),d5 | and adjust exponent |
| 2124 |
#else |
| 2125 |
subql IMM (1),d5 | and adjust exponent |
| 2126 |
#endif |
| 2127 |
btst IMM (DBL_MANT_DIG-32-1),d2 |
| 2128 |
bne Ldivdf$2 |
| 2129 |
bra 1b |
| 2130 |
|
| 2131 |
Lround$exit: |
| 2132 |
| This is a common exit point for __muldf3 and __divdf3. When they enter |
| 2133 |
| this point the sign of the result is in d7, the result in d0-d1, normalized |
| 2134 |
| so that 2^21 <= d0 < 2^22, and the exponent is in the lower byte of d4. |
| 2135 |
|
| 2136 |
| First check for underlow in the exponent: |
| 2137 |
#ifndef __mcoldfire__ |
| 2138 |
cmpw IMM (-DBL_MANT_DIG-1),d4 |
| 2139 |
#else |
| 2140 |
cmpl IMM (-DBL_MANT_DIG-1),d4 |
| 2141 |
#endif |
| 2142 |
blt Ld$underflow |
| 2143 |
| It could happen that the exponent is less than 1, in which case the |
| 2144 |
| number is denormalized. In this case we shift right and adjust the |
| 2145 |
| exponent until it becomes 1 or the fraction is zero (in the latter case |
| 2146 |
| we signal underflow and return zero). |
| 2147 |
movel d7,a0 | |
| 2148 |
movel IMM (0),d6 | use d6-d7 to collect bits flushed right |
| 2149 |
movel d6,d7 | use d6-d7 to collect bits flushed right |
| 2150 |
#ifndef __mcoldfire__ |
| 2151 |
cmpw IMM (1),d4 | if the exponent is less than 1 we |
| 2152 |
#else |
| 2153 |
cmpl IMM (1),d4 | if the exponent is less than 1 we |
| 2154 |
#endif |
| 2155 |
bge 2f | have to shift right (denormalize) |
| 2156 |
1: |
| 2157 |
#ifndef __mcoldfire__ |
| 2158 |
addw IMM (1),d4 | adjust the exponent |
| 2159 |
lsrl IMM (1),d0 | shift right once |
| 2160 |
roxrl IMM (1),d1 | |
| 2161 |
roxrl IMM (1),d2 | |
| 2162 |
roxrl IMM (1),d3 | |
| 2163 |
roxrl IMM (1),d6 | |
| 2164 |
roxrl IMM (1),d7 | |
| 2165 |
cmpw IMM (1),d4 | is the exponent 1 already? |
| 2166 |
#else |
| 2167 |
addl IMM (1),d4 | adjust the exponent |
| 2168 |
lsrl IMM (1),d7 |
| 2169 |
btst IMM (0),d6 |
| 2170 |
beq 13f |
| 2171 |
bset IMM (31),d7 |
| 2172 |
13: lsrl IMM (1),d6 |
| 2173 |
btst IMM (0),d3 |
| 2174 |
beq 14f |
| 2175 |
bset IMM (31),d6 |
| 2176 |
14: lsrl IMM (1),d3 |
| 2177 |
btst IMM (0),d2 |
| 2178 |
beq 10f |
| 2179 |
bset IMM (31),d3 |
| 2180 |
10: lsrl IMM (1),d2 |
| 2181 |
btst IMM (0),d1 |
| 2182 |
beq 11f |
| 2183 |
bset IMM (31),d2 |
| 2184 |
11: lsrl IMM (1),d1 |
| 2185 |
btst IMM (0),d0 |
| 2186 |
beq 12f |
| 2187 |
bset IMM (31),d1 |
| 2188 |
12: lsrl IMM (1),d0 |
| 2189 |
cmpl IMM (1),d4 | is the exponent 1 already? |
| 2190 |
#endif |
| 2191 |
beq 2f | if not loop back |
| 2192 |
bra 1b | |
| 2193 |
movel a0,d7 | get back sign bit into d7 |
| 2194 |
bra Ld$underflow | safety check, shouldn't execute ' |
| 2195 |
2: orl d6,d2 | this is a trick so we don't lose ' |
| 2196 |
orl d7,d3 | the bits which were flushed right |
| 2197 |
movel a0,d7 | get back sign bit into d7 |
| 2198 |
| Now call the rounding routine (which takes care of denormalized numbers): |
| 2199 |
lea pc@(Lround$0),a0 | to return from rounding routine |
| 2200 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 2201 |
#ifdef __mcoldfire__ |
| 2202 |
clrl d6 |
| 2203 |
#endif |
| 2204 |
movew a1@(6),d6 | rounding mode in d6 |
| 2205 |
beq Lround$to$nearest |
| 2206 |
#ifndef __mcoldfire__ |
| 2207 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 2208 |
#else |
| 2209 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 2210 |
#endif |
| 2211 |
bhi Lround$to$minus |
| 2212 |
blt Lround$to$zero |
| 2213 |
bra Lround$to$plus |
| 2214 |
Lround$0: |
| 2215 |
| Here we have a correctly rounded result (either normalized or denormalized). |
| 2216 |
|
| 2217 |
| Here we should have either a normalized number or a denormalized one, and |
| 2218 |
| the exponent is necessarily larger or equal to 1 (so we don't have to ' |
| 2219 |
| check again for underflow!). We have to check for overflow or for a |
| 2220 |
| denormalized number (which also signals underflow). |
| 2221 |
| Check for overflow (i.e., exponent >= 0x7ff). |
| 2222 |
#ifndef __mcoldfire__ |
| 2223 |
cmpw IMM (0x07ff),d4 |
| 2224 |
#else |
| 2225 |
cmpl IMM (0x07ff),d4 |
| 2226 |
#endif |
| 2227 |
bge Ld$overflow |
| 2228 |
| Now check for a denormalized number (exponent==0): |
| 2229 |
movew d4,d4 |
| 2230 |
beq Ld$den |
| 2231 |
1: |
| 2232 |
| Put back the exponents and sign and return. |
| 2233 |
#ifndef __mcoldfire__ |
| 2234 |
lslw IMM (4),d4 | exponent back to fourth byte |
| 2235 |
#else |
| 2236 |
lsll IMM (4),d4 | exponent back to fourth byte |
| 2237 |
#endif |
| 2238 |
bclr IMM (DBL_MANT_DIG-32-1),d0 |
| 2239 |
swap d0 | and put back exponent |
| 2240 |
#ifndef __mcoldfire__ |
| 2241 |
orw d4,d0 | |
| 2242 |
#else |
| 2243 |
orl d4,d0 | |
| 2244 |
#endif |
| 2245 |
swap d0 | |
| 2246 |
orl d7,d0 | and sign also |
| 2247 |
|
| 2248 |
PICLEA SYM (_fpCCR),a0 |
| 2249 |
movew IMM (0),a0@ |
| 2250 |
#ifndef __mcoldfire__ |
| 2251 |
moveml sp@+,d2-d7 |
| 2252 |
#else |
| 2253 |
moveml sp@,d2-d7 |
| 2254 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2255 |
| be adjusted here. |
| 2256 |
#endif |
| 2257 |
unlk a6 |
| 2258 |
rts |
| 2259 |
|
| 2260 |
|============================================================================= |
| 2261 |
| __negdf2 |
| 2262 |
|============================================================================= |
| 2263 |
|
| 2264 |
| double __negdf2(double, double); |
| 2265 |
FUNC(__negdf2) |
| 2266 |
SYM (__negdf2): |
| 2267 |
#ifndef __mcoldfire__ |
| 2268 |
link a6,IMM (0) |
| 2269 |
moveml d2-d7,sp@- |
| 2270 |
#else |
| 2271 |
link a6,IMM (-24) |
| 2272 |
moveml d2-d7,sp@ |
| 2273 |
#endif |
| 2274 |
moveq IMM (NEGATE),d5 |
| 2275 |
movel a6@(8),d0 | get number to negate in d0-d1 |
| 2276 |
movel a6@(12),d1 | |
| 2277 |
bchg IMM (31),d0 | negate |
| 2278 |
movel d0,d2 | make a positive copy (for the tests) |
| 2279 |
bclr IMM (31),d2 | |
| 2280 |
movel d2,d4 | check for zero |
| 2281 |
orl d1,d4 | |
| 2282 |
beq 2f | if zero (either sign) return +zero |
| 2283 |
cmpl IMM (0x7ff00000),d2 | compare to +INFINITY |
| 2284 |
blt 1f | if finite, return |
| 2285 |
bhi Ld$inop | if larger (fraction not zero) is NaN |
| 2286 |
tstl d1 | if d2 == 0x7ff00000 check d1 |
| 2287 |
bne Ld$inop | |
| 2288 |
movel d0,d7 | else get sign and return INFINITY |
| 2289 |
andl IMM (0x80000000),d7 |
| 2290 |
bra Ld$infty |
| 2291 |
1: PICLEA SYM (_fpCCR),a0 |
| 2292 |
movew IMM (0),a0@ |
| 2293 |
#ifndef __mcoldfire__ |
| 2294 |
moveml sp@+,d2-d7 |
| 2295 |
#else |
| 2296 |
moveml sp@,d2-d7 |
| 2297 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2298 |
| be adjusted here. |
| 2299 |
#endif |
| 2300 |
unlk a6 |
| 2301 |
rts |
| 2302 |
2: bclr IMM (31),d0 |
| 2303 |
bra 1b |
| 2304 |
|
| 2305 |
|============================================================================= |
| 2306 |
| __cmpdf2 |
| 2307 |
|============================================================================= |
| 2308 |
|
| 2309 |
GREATER = 1 |
| 2310 |
LESS = -1 |
| 2311 |
EQUAL = 0 |
| 2312 |
|
| 2313 |
| int __cmpdf2_internal(double, double, int); |
| 2314 |
SYM (__cmpdf2_internal): |
| 2315 |
#ifndef __mcoldfire__ |
| 2316 |
link a6,IMM (0) |
| 2317 |
moveml d2-d7,sp@- | save registers |
| 2318 |
#else |
| 2319 |
link a6,IMM (-24) |
| 2320 |
moveml d2-d7,sp@ |
| 2321 |
#endif |
| 2322 |
moveq IMM (COMPARE),d5 |
| 2323 |
movel a6@(8),d0 | get first operand |
| 2324 |
movel a6@(12),d1 | |
| 2325 |
movel a6@(16),d2 | get second operand |
| 2326 |
movel a6@(20),d3 | |
| 2327 |
| First check if a and/or b are (+/-) zero and in that case clear |
| 2328 |
| the sign bit. |
| 2329 |
movel d0,d6 | copy signs into d6 (a) and d7(b) |
| 2330 |
bclr IMM (31),d0 | and clear signs in d0 and d2 |
| 2331 |
movel d2,d7 | |
| 2332 |
bclr IMM (31),d2 | |
| 2333 |
cmpl IMM (0x7ff00000),d0 | check for a == NaN |
| 2334 |
bhi Lcmpd$inop | if d0 > 0x7ff00000, a is NaN |
| 2335 |
beq Lcmpdf$a$nf | if equal can be INFINITY, so check d1 |
| 2336 |
movel d0,d4 | copy into d4 to test for zero |
| 2337 |
orl d1,d4 | |
| 2338 |
beq Lcmpdf$a$0 | |
| 2339 |
Lcmpdf$0: |
| 2340 |
cmpl IMM (0x7ff00000),d2 | check for b == NaN |
| 2341 |
bhi Lcmpd$inop | if d2 > 0x7ff00000, b is NaN |
| 2342 |
beq Lcmpdf$b$nf | if equal can be INFINITY, so check d3 |
| 2343 |
movel d2,d4 | |
| 2344 |
orl d3,d4 | |
| 2345 |
beq Lcmpdf$b$0 | |
| 2346 |
Lcmpdf$1: |
| 2347 |
| Check the signs |
| 2348 |
eorl d6,d7 |
| 2349 |
bpl 1f |
| 2350 |
| If the signs are not equal check if a >= 0 |
| 2351 |
tstl d6 |
| 2352 |
bpl Lcmpdf$a$gt$b | if (a >= 0 && b < 0) => a > b |
| 2353 |
bmi Lcmpdf$b$gt$a | if (a < 0 && b >= 0) => a < b |
| 2354 |
1: |
| 2355 |
| If the signs are equal check for < 0 |
| 2356 |
tstl d6 |
| 2357 |
bpl 1f |
| 2358 |
| If both are negative exchange them |
| 2359 |
#ifndef __mcoldfire__ |
| 2360 |
exg d0,d2 |
| 2361 |
exg d1,d3 |
| 2362 |
#else |
| 2363 |
movel d0,d7 |
| 2364 |
movel d2,d0 |
| 2365 |
movel d7,d2 |
| 2366 |
movel d1,d7 |
| 2367 |
movel d3,d1 |
| 2368 |
movel d7,d3 |
| 2369 |
#endif |
| 2370 |
1: |
| 2371 |
| Now that they are positive we just compare them as longs (does this also |
| 2372 |
| work for denormalized numbers?). |
| 2373 |
cmpl d0,d2 |
| 2374 |
bhi Lcmpdf$b$gt$a | |b| > |a| |
| 2375 |
bne Lcmpdf$a$gt$b | |b| < |a| |
| 2376 |
| If we got here d0 == d2, so we compare d1 and d3. |
| 2377 |
cmpl d1,d3 |
| 2378 |
bhi Lcmpdf$b$gt$a | |b| > |a| |
| 2379 |
bne Lcmpdf$a$gt$b | |b| < |a| |
| 2380 |
| If we got here a == b. |
| 2381 |
movel IMM (EQUAL),d0 |
| 2382 |
#ifndef __mcoldfire__ |
| 2383 |
moveml sp@+,d2-d7 | put back the registers |
| 2384 |
#else |
| 2385 |
moveml sp@,d2-d7 |
| 2386 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2387 |
| be adjusted here. |
| 2388 |
#endif |
| 2389 |
unlk a6 |
| 2390 |
rts |
| 2391 |
Lcmpdf$a$gt$b: |
| 2392 |
movel IMM (GREATER),d0 |
| 2393 |
#ifndef __mcoldfire__ |
| 2394 |
moveml sp@+,d2-d7 | put back the registers |
| 2395 |
#else |
| 2396 |
moveml sp@,d2-d7 |
| 2397 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2398 |
| be adjusted here. |
| 2399 |
#endif |
| 2400 |
unlk a6 |
| 2401 |
rts |
| 2402 |
Lcmpdf$b$gt$a: |
| 2403 |
movel IMM (LESS),d0 |
| 2404 |
#ifndef __mcoldfire__ |
| 2405 |
moveml sp@+,d2-d7 | put back the registers |
| 2406 |
#else |
| 2407 |
moveml sp@,d2-d7 |
| 2408 |
| XXX if frame pointer is ever removed, stack pointer must |
| 2409 |
| be adjusted here. |
| 2410 |
#endif |
| 2411 |
unlk a6 |
| 2412 |
rts |
| 2413 |
|
| 2414 |
Lcmpdf$a$0: |
| 2415 |
bclr IMM (31),d6 |
| 2416 |
bra Lcmpdf$0 |
| 2417 |
Lcmpdf$b$0: |
| 2418 |
bclr IMM (31),d7 |
| 2419 |
bra Lcmpdf$1 |
| 2420 |
|
| 2421 |
Lcmpdf$a$nf: |
| 2422 |
tstl d1 |
| 2423 |
bne Ld$inop |
| 2424 |
bra Lcmpdf$0 |
| 2425 |
|
| 2426 |
Lcmpdf$b$nf: |
| 2427 |
tstl d3 |
| 2428 |
bne Ld$inop |
| 2429 |
bra Lcmpdf$1 |
| 2430 |
|
| 2431 |
Lcmpd$inop: |
| 2432 |
movl a6@(24),d0 |
| 2433 |
moveq IMM (INEXACT_RESULT+INVALID_OPERATION),d7 |
| 2434 |
moveq IMM (DOUBLE_FLOAT),d6 |
| 2435 |
PICJUMP $_exception_handler |
| 2436 |
|
| 2437 |
| int __cmpdf2(double, double); |
| 2438 |
FUNC(__cmpdf2) |
| 2439 |
SYM (__cmpdf2): |
| 2440 |
link a6,IMM (0) |
| 2441 |
pea 1 |
| 2442 |
movl a6@(20),sp@- |
| 2443 |
movl a6@(16),sp@- |
| 2444 |
movl a6@(12),sp@- |
| 2445 |
movl a6@(8),sp@- |
| 2446 |
PICCALL SYM (__cmpdf2_internal) |
| 2447 |
unlk a6 |
| 2448 |
rts |
| 2449 |
|
| 2450 |
|============================================================================= |
| 2451 |
| rounding routines |
| 2452 |
|============================================================================= |
| 2453 |
|
| 2454 |
| The rounding routines expect the number to be normalized in registers |
| 2455 |
| d0-d1-d2-d3, with the exponent in register d4. They assume that the |
| 2456 |
| exponent is larger or equal to 1. They return a properly normalized number |
| 2457 |
| if possible, and a denormalized number otherwise. The exponent is returned |
| 2458 |
| in d4. |
| 2459 |
|
| 2460 |
Lround$to$nearest: |
| 2461 |
| We now normalize as suggested by D. Knuth ("Seminumerical Algorithms"): |
| 2462 |
| Here we assume that the exponent is not too small (this should be checked |
| 2463 |
| before entering the rounding routine), but the number could be denormalized. |
| 2464 |
|
| 2465 |
| Check for denormalized numbers: |
| 2466 |
1: btst IMM (DBL_MANT_DIG-32),d0 |
| 2467 |
bne 2f | if set the number is normalized |
| 2468 |
| Normalize shifting left until bit #DBL_MANT_DIG-32 is set or the exponent |
| 2469 |
| is one (remember that a denormalized number corresponds to an |
| 2470 |
| exponent of -D_BIAS+1). |
| 2471 |
#ifndef __mcoldfire__ |
| 2472 |
cmpw IMM (1),d4 | remember that the exponent is at least one |
| 2473 |
#else |
| 2474 |
cmpl IMM (1),d4 | remember that the exponent is at least one |
| 2475 |
#endif |
| 2476 |
beq 2f | an exponent of one means denormalized |
| 2477 |
addl d3,d3 | else shift and adjust the exponent |
| 2478 |
addxl d2,d2 | |
| 2479 |
addxl d1,d1 | |
| 2480 |
addxl d0,d0 | |
| 2481 |
#ifndef __mcoldfire__ |
| 2482 |
dbra d4,1b | |
| 2483 |
#else |
| 2484 |
subql IMM (1), d4 |
| 2485 |
bpl 1b |
| 2486 |
#endif |
| 2487 |
2: |
| 2488 |
| Now round: we do it as follows: after the shifting we can write the |
| 2489 |
| fraction part as f + delta, where 1 < f < 2^25, and 0 <= delta <= 2. |
| 2490 |
| If delta < 1, do nothing. If delta > 1, add 1 to f. |
| 2491 |
| If delta == 1, we make sure the rounded number will be even (odd?) |
| 2492 |
| (after shifting). |
| 2493 |
btst IMM (0),d1 | is delta < 1? |
| 2494 |
beq 2f | if so, do not do anything |
| 2495 |
orl d2,d3 | is delta == 1? |
| 2496 |
bne 1f | if so round to even |
| 2497 |
movel d1,d3 | |
| 2498 |
andl IMM (2),d3 | bit 1 is the last significant bit |
| 2499 |
movel IMM (0),d2 | |
| 2500 |
addl d3,d1 | |
| 2501 |
addxl d2,d0 | |
| 2502 |
bra 2f | |
| 2503 |
1: movel IMM (1),d3 | else add 1 |
| 2504 |
movel IMM (0),d2 | |
| 2505 |
addl d3,d1 | |
| 2506 |
addxl d2,d0 |
| 2507 |
| Shift right once (because we used bit #DBL_MANT_DIG-32!). |
| 2508 |
2: |
| 2509 |
#ifndef __mcoldfire__ |
| 2510 |
lsrl IMM (1),d0 |
| 2511 |
roxrl IMM (1),d1 |
| 2512 |
#else |
| 2513 |
lsrl IMM (1),d1 |
| 2514 |
btst IMM (0),d0 |
| 2515 |
beq 10f |
| 2516 |
bset IMM (31),d1 |
| 2517 |
10: lsrl IMM (1),d0 |
| 2518 |
#endif |
| 2519 |
|
| 2520 |
| Now check again bit #DBL_MANT_DIG-32 (rounding could have produced a |
| 2521 |
| 'fraction overflow' ...). |
| 2522 |
btst IMM (DBL_MANT_DIG-32),d0 |
| 2523 |
beq 1f |
| 2524 |
#ifndef __mcoldfire__ |
| 2525 |
lsrl IMM (1),d0 |
| 2526 |
roxrl IMM (1),d1 |
| 2527 |
addw IMM (1),d4 |
| 2528 |
#else |
| 2529 |
lsrl IMM (1),d1 |
| 2530 |
btst IMM (0),d0 |
| 2531 |
beq 10f |
| 2532 |
bset IMM (31),d1 |
| 2533 |
10: lsrl IMM (1),d0 |
| 2534 |
addl IMM (1),d4 |
| 2535 |
#endif |
| 2536 |
1: |
| 2537 |
| If bit #DBL_MANT_DIG-32-1 is clear we have a denormalized number, so we |
| 2538 |
| have to put the exponent to zero and return a denormalized number. |
| 2539 |
btst IMM (DBL_MANT_DIG-32-1),d0 |
| 2540 |
beq 1f |
| 2541 |
jmp a0@ |
| 2542 |
1: movel IMM (0),d4 |
| 2543 |
jmp a0@ |
| 2544 |
|
| 2545 |
Lround$to$zero: |
| 2546 |
Lround$to$plus: |
| 2547 |
Lround$to$minus: |
| 2548 |
jmp a0@ |
| 2549 |
#endif /* L_double */ |
| 2550 |
|
| 2551 |
#ifdef L_float |
| 2552 |
|
| 2553 |
.globl SYM (_fpCCR) |
| 2554 |
.globl $_exception_handler |
| 2555 |
|
| 2556 |
QUIET_NaN = 0x7fffffff |
| 2557 |
SIGNL_NaN = 0x7f800001 |
| 2558 |
INFINITY = 0x7f800000 |
| 2559 |
|
| 2560 |
F_MAX_EXP = 0xff |
| 2561 |
F_BIAS = 126 |
| 2562 |
FLT_MAX_EXP = F_MAX_EXP - F_BIAS |
| 2563 |
FLT_MIN_EXP = 1 - F_BIAS |
| 2564 |
FLT_MANT_DIG = 24 |
| 2565 |
|
| 2566 |
INEXACT_RESULT = 0x0001 |
| 2567 |
UNDERFLOW = 0x0002 |
| 2568 |
OVERFLOW = 0x0004 |
| 2569 |
DIVIDE_BY_ZERO = 0x0008 |
| 2570 |
INVALID_OPERATION = 0x0010 |
| 2571 |
|
| 2572 |
SINGLE_FLOAT = 1 |
| 2573 |
|
| 2574 |
NOOP = 0 |
| 2575 |
ADD = 1 |
| 2576 |
MULTIPLY = 2 |
| 2577 |
DIVIDE = 3 |
| 2578 |
NEGATE = 4 |
| 2579 |
COMPARE = 5 |
| 2580 |
EXTENDSFDF = 6 |
| 2581 |
TRUNCDFSF = 7 |
| 2582 |
|
| 2583 |
UNKNOWN = -1 |
| 2584 |
ROUND_TO_NEAREST = 0 | round result to nearest representable value |
| 2585 |
ROUND_TO_ZERO = 1 | round result towards zero |
| 2586 |
ROUND_TO_PLUS = 2 | round result towards plus infinity |
| 2587 |
ROUND_TO_MINUS = 3 | round result towards minus infinity |
| 2588 |
|
| 2589 |
| Entry points: |
| 2590 |
|
| 2591 |
.globl SYM (__addsf3) |
| 2592 |
.globl SYM (__subsf3) |
| 2593 |
.globl SYM (__mulsf3) |
| 2594 |
.globl SYM (__divsf3) |
| 2595 |
.globl SYM (__negsf2) |
| 2596 |
.globl SYM (__cmpsf2) |
| 2597 |
.globl SYM (__cmpsf2_internal) |
| 2598 |
.hidden SYM (__cmpsf2_internal) |
| 2599 |
|
| 2600 |
| These are common routines to return and signal exceptions. |
| 2601 |
|
| 2602 |
.text |
| 2603 |
.even |
| 2604 |
|
| 2605 |
Lf$den: |
| 2606 |
| Return and signal a denormalized number |
| 2607 |
orl d7,d0 |
| 2608 |
moveq IMM (INEXACT_RESULT+UNDERFLOW),d7 |
| 2609 |
moveq IMM (SINGLE_FLOAT),d6 |
| 2610 |
PICJUMP $_exception_handler |
| 2611 |
|
| 2612 |
Lf$infty: |
| 2613 |
Lf$overflow: |
| 2614 |
| Return a properly signed INFINITY and set the exception flags |
| 2615 |
movel IMM (INFINITY),d0 |
| 2616 |
orl d7,d0 |
| 2617 |
moveq IMM (INEXACT_RESULT+OVERFLOW),d7 |
| 2618 |
moveq IMM (SINGLE_FLOAT),d6 |
| 2619 |
PICJUMP $_exception_handler |
| 2620 |
|
| 2621 |
Lf$underflow: |
| 2622 |
| Return a properly signed 0 and set the exception flags |
| 2623 |
moveq IMM (0),d0 |
| 2624 |
orl d7,d0 |
| 2625 |
moveq IMM (INEXACT_RESULT+UNDERFLOW),d7 |
| 2626 |
moveq IMM (SINGLE_FLOAT),d6 |
| 2627 |
PICJUMP $_exception_handler |
| 2628 |
|
| 2629 |
Lf$inop: |
| 2630 |
| Return a quiet NaN and set the exception flags |
| 2631 |
movel IMM (QUIET_NaN),d0 |
| 2632 |
moveq IMM (INEXACT_RESULT+INVALID_OPERATION),d7 |
| 2633 |
moveq IMM (SINGLE_FLOAT),d6 |
| 2634 |
PICJUMP $_exception_handler |
| 2635 |
|
| 2636 |
Lf$div$0: |
| 2637 |
| Return a properly signed INFINITY and set the exception flags |
| 2638 |
movel IMM (INFINITY),d0 |
| 2639 |
orl d7,d0 |
| 2640 |
moveq IMM (INEXACT_RESULT+DIVIDE_BY_ZERO),d7 |
| 2641 |
moveq IMM (SINGLE_FLOAT),d6 |
| 2642 |
PICJUMP $_exception_handler |
| 2643 |
|
| 2644 |
|============================================================================= |
| 2645 |
|============================================================================= |
| 2646 |
| single precision routines |
| 2647 |
|============================================================================= |
| 2648 |
|============================================================================= |
| 2649 |
|
| 2650 |
| A single precision floating point number (float) has the format: |
| 2651 |
| |
| 2652 |
| struct _float { |
| 2653 |
| unsigned int sign : 1; /* sign bit */ |
| 2654 |
| unsigned int exponent : 8; /* exponent, shifted by 126 */ |
| 2655 |
| unsigned int fraction : 23; /* fraction */ |
| 2656 |
| } float; |
| 2657 |
| |
| 2658 |
| Thus sizeof(float) = 4 (32 bits). |
| 2659 |
| |
| 2660 |
| All the routines are callable from C programs, and return the result |
| 2661 |
| in the single register d0. They also preserve all registers except |
| 2662 |
| d0-d1 and a0-a1. |
| 2663 |
|
| 2664 |
|============================================================================= |
| 2665 |
| __subsf3 |
| 2666 |
|============================================================================= |
| 2667 |
|
| 2668 |
| float __subsf3(float, float); |
| 2669 |
FUNC(__subsf3) |
| 2670 |
SYM (__subsf3): |
| 2671 |
bchg IMM (31),sp@(8) | change sign of second operand |
| 2672 |
| and fall through |
| 2673 |
|============================================================================= |
| 2674 |
| __addsf3 |
| 2675 |
|============================================================================= |
| 2676 |
|
| 2677 |
| float __addsf3(float, float); |
| 2678 |
FUNC(__addsf3) |
| 2679 |
SYM (__addsf3): |
| 2680 |
#ifndef __mcoldfire__ |
| 2681 |
link a6,IMM (0) | everything will be done in registers |
| 2682 |
moveml d2-d7,sp@- | save all data registers but d0-d1 |
| 2683 |
#else |
| 2684 |
link a6,IMM (-24) |
| 2685 |
moveml d2-d7,sp@ |
| 2686 |
#endif |
| 2687 |
movel a6@(8),d0 | get first operand |
| 2688 |
movel a6@(12),d1 | get second operand |
| 2689 |
movel d0,a0 | get d0's sign bit ' |
| 2690 |
addl d0,d0 | check and clear sign bit of a |
| 2691 |
beq Laddsf$b | if zero return second operand |
| 2692 |
movel d1,a1 | save b's sign bit ' |
| 2693 |
addl d1,d1 | get rid of sign bit |
| 2694 |
beq Laddsf$a | if zero return first operand |
| 2695 |
|
| 2696 |
| Get the exponents and check for denormalized and/or infinity. |
| 2697 |
|
| 2698 |
movel IMM (0x00ffffff),d4 | mask to get fraction |
| 2699 |
movel IMM (0x01000000),d5 | mask to put hidden bit back |
| 2700 |
|
| 2701 |
movel d0,d6 | save a to get exponent |
| 2702 |
andl d4,d0 | get fraction in d0 |
| 2703 |
notl d4 | make d4 into a mask for the exponent |
| 2704 |
andl d4,d6 | get exponent in d6 |
| 2705 |
beq Laddsf$a$den | branch if a is denormalized |
| 2706 |
cmpl d4,d6 | check for INFINITY or NaN |
| 2707 |
beq Laddsf$nf |
| 2708 |
swap d6 | put exponent into first word |
| 2709 |
orl d5,d0 | and put hidden bit back |
| 2710 |
Laddsf$1: |
| 2711 |
| Now we have a's exponent in d6 (second byte) and the mantissa in d0. ' |
| 2712 |
movel d1,d7 | get exponent in d7 |
| 2713 |
andl d4,d7 | |
| 2714 |
beq Laddsf$b$den | branch if b is denormalized |
| 2715 |
cmpl d4,d7 | check for INFINITY or NaN |
| 2716 |
beq Laddsf$nf |
| 2717 |
swap d7 | put exponent into first word |
| 2718 |
notl d4 | make d4 into a mask for the fraction |
| 2719 |
andl d4,d1 | get fraction in d1 |
| 2720 |
orl d5,d1 | and put hidden bit back |
| 2721 |
Laddsf$2: |
| 2722 |
| Now we have b's exponent in d7 (second byte) and the mantissa in d1. ' |
| 2723 |
|
| 2724 |
| Note that the hidden bit corresponds to bit #FLT_MANT_DIG-1, and we |
| 2725 |
| shifted right once, so bit #FLT_MANT_DIG is set (so we have one extra |
| 2726 |
| bit). |
| 2727 |
|
| 2728 |
movel d1,d2 | move b to d2, since we want to use |
| 2729 |
| two registers to do the sum |
| 2730 |
movel IMM (0),d1 | and clear the new ones |
| 2731 |
movel d1,d3 | |
| 2732 |
|
| 2733 |
| Here we shift the numbers in registers d0 and d1 so the exponents are the |
| 2734 |
| same, and put the largest exponent in d6. Note that we are using two |
| 2735 |
| registers for each number (see the discussion by D. Knuth in "Seminumerical |
| 2736 |
| Algorithms"). |
| 2737 |
#ifndef __mcoldfire__ |
| 2738 |
cmpw d6,d7 | compare exponents |
| 2739 |
#else |
| 2740 |
cmpl d6,d7 | compare exponents |
| 2741 |
#endif |
| 2742 |
beq Laddsf$3 | if equal don't shift ' |
| 2743 |
bhi 5f | branch if second exponent largest |
| 2744 |
1: |
| 2745 |
subl d6,d7 | keep the largest exponent |
| 2746 |
negl d7 |
| 2747 |
#ifndef __mcoldfire__ |
| 2748 |
lsrw IMM (8),d7 | put difference in lower byte |
| 2749 |
#else |
| 2750 |
lsrl IMM (8),d7 | put difference in lower byte |
| 2751 |
#endif |
| 2752 |
| if difference is too large we don't shift (actually, we can just exit) ' |
| 2753 |
#ifndef __mcoldfire__ |
| 2754 |
cmpw IMM (FLT_MANT_DIG+2),d7 |
| 2755 |
#else |
| 2756 |
cmpl IMM (FLT_MANT_DIG+2),d7 |
| 2757 |
#endif |
| 2758 |
bge Laddsf$b$small |
| 2759 |
#ifndef __mcoldfire__ |
| 2760 |
cmpw IMM (16),d7 | if difference >= 16 swap |
| 2761 |
#else |
| 2762 |
cmpl IMM (16),d7 | if difference >= 16 swap |
| 2763 |
#endif |
| 2764 |
bge 4f |
| 2765 |
2: |
| 2766 |
#ifndef __mcoldfire__ |
| 2767 |
subw IMM (1),d7 |
| 2768 |
#else |
| 2769 |
subql IMM (1), d7 |
| 2770 |
#endif |
| 2771 |
3: |
| 2772 |
#ifndef __mcoldfire__ |
| 2773 |
lsrl IMM (1),d2 | shift right second operand |
| 2774 |
roxrl IMM (1),d3 |
| 2775 |
dbra d7,3b |
| 2776 |
#else |
| 2777 |
lsrl IMM (1),d3 |
| 2778 |
btst IMM (0),d2 |
| 2779 |
beq 10f |
| 2780 |
bset IMM (31),d3 |
| 2781 |
10: lsrl IMM (1),d2 |
| 2782 |
subql IMM (1), d7 |
| 2783 |
bpl 3b |
| 2784 |
#endif |
| 2785 |
bra Laddsf$3 |
| 2786 |
4: |
| 2787 |
movew d2,d3 |
| 2788 |
swap d3 |
| 2789 |
movew d3,d2 |
| 2790 |
swap d2 |
| 2791 |
#ifndef __mcoldfire__ |
| 2792 |
subw IMM (16),d7 |
| 2793 |
#else |
| 2794 |
subl IMM (16),d7 |
| 2795 |
#endif |
| 2796 |
bne 2b | if still more bits, go back to normal case |
| 2797 |
bra Laddsf$3 |
| 2798 |
5: |
| 2799 |
#ifndef __mcoldfire__ |
| 2800 |
exg d6,d7 | exchange the exponents |
| 2801 |
#else |
| 2802 |
eorl d6,d7 |
| 2803 |
eorl d7,d6 |
| 2804 |
eorl d6,d7 |
| 2805 |
#endif |
| 2806 |
subl d6,d7 | keep the largest exponent |
| 2807 |
negl d7 | |
| 2808 |
#ifndef __mcoldfire__ |
| 2809 |
lsrw IMM (8),d7 | put difference in lower byte |
| 2810 |
#else |
| 2811 |
lsrl IMM (8),d7 | put difference in lower byte |
| 2812 |
#endif |
| 2813 |
| if difference is too large we don't shift (and exit!) ' |
| 2814 |
#ifndef __mcoldfire__ |
| 2815 |
cmpw IMM (FLT_MANT_DIG+2),d7 |
| 2816 |
#else |
| 2817 |
cmpl IMM (FLT_MANT_DIG+2),d7 |
| 2818 |
#endif |
| 2819 |
bge Laddsf$a$small |
| 2820 |
#ifndef __mcoldfire__ |
| 2821 |
cmpw IMM (16),d7 | if difference >= 16 swap |
| 2822 |
#else |
| 2823 |
cmpl IMM (16),d7 | if difference >= 16 swap |
| 2824 |
#endif |
| 2825 |
bge 8f |
| 2826 |
6: |
| 2827 |
#ifndef __mcoldfire__ |
| 2828 |
subw IMM (1),d7 |
| 2829 |
#else |
| 2830 |
subl IMM (1),d7 |
| 2831 |
#endif |
| 2832 |
7: |
| 2833 |
#ifndef __mcoldfire__ |
| 2834 |
lsrl IMM (1),d0 | shift right first operand |
| 2835 |
roxrl IMM (1),d1 |
| 2836 |
dbra d7,7b |
| 2837 |
#else |
| 2838 |
lsrl IMM (1),d1 |
| 2839 |
btst IMM (0),d0 |
| 2840 |
beq 10f |
| 2841 |
bset IMM (31),d1 |
| 2842 |
10: lsrl IMM (1),d0 |
| 2843 |
subql IMM (1),d7 |
| 2844 |
bpl 7b |
| 2845 |
#endif |
| 2846 |
bra Laddsf$3 |
| 2847 |
8: |
| 2848 |
movew d0,d1 |
| 2849 |
swap d1 |
| 2850 |
movew d1,d0 |
| 2851 |
swap d0 |
| 2852 |
#ifndef __mcoldfire__ |
| 2853 |
subw IMM (16),d7 |
| 2854 |
#else |
| 2855 |
subl IMM (16),d7 |
| 2856 |
#endif |
| 2857 |
bne 6b | if still more bits, go back to normal case |
| 2858 |
| otherwise we fall through |
| 2859 |
|
| 2860 |
| Now we have a in d0-d1, b in d2-d3, and the largest exponent in d6 (the |
| 2861 |
| signs are stored in a0 and a1). |
| 2862 |
|
| 2863 |
Laddsf$3: |
| 2864 |
| Here we have to decide whether to add or subtract the numbers |
| 2865 |
#ifndef __mcoldfire__ |
| 2866 |
exg d6,a0 | get signs back |
| 2867 |
exg d7,a1 | and save the exponents |
| 2868 |
#else |
| 2869 |
movel d6,d4 |
| 2870 |
movel a0,d6 |
| 2871 |
movel d4,a0 |
| 2872 |
movel d7,d4 |
| 2873 |
movel a1,d7 |
| 2874 |
movel d4,a1 |
| 2875 |
#endif |
| 2876 |
eorl d6,d7 | combine sign bits |
| 2877 |
bmi Lsubsf$0 | if negative a and b have opposite |
| 2878 |
| sign so we actually subtract the |
| 2879 |
| numbers |
| 2880 |
|
| 2881 |
| Here we have both positive or both negative |
| 2882 |
#ifndef __mcoldfire__ |
| 2883 |
exg d6,a0 | now we have the exponent in d6 |
| 2884 |
#else |
| 2885 |
movel d6,d4 |
| 2886 |
movel a0,d6 |
| 2887 |
movel d4,a0 |
| 2888 |
#endif |
| 2889 |
movel a0,d7 | and sign in d7 |
| 2890 |
andl IMM (0x80000000),d7 |
| 2891 |
| Here we do the addition. |
| 2892 |
addl d3,d1 |
| 2893 |
addxl d2,d0 |
| 2894 |
| Note: now we have d2, d3, d4 and d5 to play with! |
| 2895 |
|
| 2896 |
| Put the exponent, in the first byte, in d2, to use the "standard" rounding |
| 2897 |
| routines: |
| 2898 |
movel d6,d2 |
| 2899 |
#ifndef __mcoldfire__ |
| 2900 |
lsrw IMM (8),d2 |
| 2901 |
#else |
| 2902 |
lsrl IMM (8),d2 |
| 2903 |
#endif |
| 2904 |
|
| 2905 |
| Before rounding normalize so bit #FLT_MANT_DIG is set (we will consider |
| 2906 |
| the case of denormalized numbers in the rounding routine itself). |
| 2907 |
| As in the addition (not in the subtraction!) we could have set |
| 2908 |
| one more bit we check this: |
| 2909 |
btst IMM (FLT_MANT_DIG+1),d0 |
| 2910 |
beq 1f |
| 2911 |
#ifndef __mcoldfire__ |
| 2912 |
lsrl IMM (1),d0 |
| 2913 |
roxrl IMM (1),d1 |
| 2914 |
#else |
| 2915 |
lsrl IMM (1),d1 |
| 2916 |
btst IMM (0),d0 |
| 2917 |
beq 10f |
| 2918 |
bset IMM (31),d1 |
| 2919 |
10: lsrl IMM (1),d0 |
| 2920 |
#endif |
| 2921 |
addl IMM (1),d2 |
| 2922 |
1: |
| 2923 |
lea pc@(Laddsf$4),a0 | to return from rounding routine |
| 2924 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 2925 |
#ifdef __mcoldfire__ |
| 2926 |
clrl d6 |
| 2927 |
#endif |
| 2928 |
movew a1@(6),d6 | rounding mode in d6 |
| 2929 |
beq Lround$to$nearest |
| 2930 |
#ifndef __mcoldfire__ |
| 2931 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 2932 |
#else |
| 2933 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 2934 |
#endif |
| 2935 |
bhi Lround$to$minus |
| 2936 |
blt Lround$to$zero |
| 2937 |
bra Lround$to$plus |
| 2938 |
Laddsf$4: |
| 2939 |
| Put back the exponent, but check for overflow. |
| 2940 |
#ifndef __mcoldfire__ |
| 2941 |
cmpw IMM (0xff),d2 |
| 2942 |
#else |
| 2943 |
cmpl IMM (0xff),d2 |
| 2944 |
#endif |
| 2945 |
bge 1f |
| 2946 |
bclr IMM (FLT_MANT_DIG-1),d0 |
| 2947 |
#ifndef __mcoldfire__ |
| 2948 |
lslw IMM (7),d2 |
| 2949 |
#else |
| 2950 |
lsll IMM (7),d2 |
| 2951 |
#endif |
| 2952 |
swap d2 |
| 2953 |
orl d2,d0 |
| 2954 |
bra Laddsf$ret |
| 2955 |
1: |
| 2956 |
moveq IMM (ADD),d5 |
| 2957 |
bra Lf$overflow |
| 2958 |
|
| 2959 |
Lsubsf$0: |
| 2960 |
| We are here if a > 0 and b < 0 (sign bits cleared). |
| 2961 |
| Here we do the subtraction. |
| 2962 |
movel d6,d7 | put sign in d7 |
| 2963 |
andl IMM (0x80000000),d7 |
| 2964 |
|
| 2965 |
subl d3,d1 | result in d0-d1 |
| 2966 |
subxl d2,d0 | |
| 2967 |
beq Laddsf$ret | if zero just exit |
| 2968 |
bpl 1f | if positive skip the following |
| 2969 |
bchg IMM (31),d7 | change sign bit in d7 |
| 2970 |
negl d1 |
| 2971 |
negxl d0 |
| 2972 |
1: |
| 2973 |
#ifndef __mcoldfire__ |
| 2974 |
exg d2,a0 | now we have the exponent in d2 |
| 2975 |
lsrw IMM (8),d2 | put it in the first byte |
| 2976 |
#else |
| 2977 |
movel d2,d4 |
| 2978 |
movel a0,d2 |
| 2979 |
movel d4,a0 |
| 2980 |
lsrl IMM (8),d2 | put it in the first byte |
| 2981 |
#endif |
| 2982 |
|
| 2983 |
| Now d0-d1 is positive and the sign bit is in d7. |
| 2984 |
|
| 2985 |
| Note that we do not have to normalize, since in the subtraction bit |
| 2986 |
| #FLT_MANT_DIG+1 is never set, and denormalized numbers are handled by |
| 2987 |
| the rounding routines themselves. |
| 2988 |
lea pc@(Lsubsf$1),a0 | to return from rounding routine |
| 2989 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 2990 |
#ifdef __mcoldfire__ |
| 2991 |
clrl d6 |
| 2992 |
#endif |
| 2993 |
movew a1@(6),d6 | rounding mode in d6 |
| 2994 |
beq Lround$to$nearest |
| 2995 |
#ifndef __mcoldfire__ |
| 2996 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 2997 |
#else |
| 2998 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 2999 |
#endif |
| 3000 |
bhi Lround$to$minus |
| 3001 |
blt Lround$to$zero |
| 3002 |
bra Lround$to$plus |
| 3003 |
Lsubsf$1: |
| 3004 |
| Put back the exponent (we can't have overflow!). ' |
| 3005 |
bclr IMM (FLT_MANT_DIG-1),d0 |
| 3006 |
#ifndef __mcoldfire__ |
| 3007 |
lslw IMM (7),d2 |
| 3008 |
#else |
| 3009 |
lsll IMM (7),d2 |
| 3010 |
#endif |
| 3011 |
swap d2 |
| 3012 |
orl d2,d0 |
| 3013 |
bra Laddsf$ret |
| 3014 |
|
| 3015 |
| If one of the numbers was too small (difference of exponents >= |
| 3016 |
| FLT_MANT_DIG+2) we return the other (and now we don't have to ' |
| 3017 |
| check for finiteness or zero). |
| 3018 |
Laddsf$a$small: |
| 3019 |
movel a6@(12),d0 |
| 3020 |
PICLEA SYM (_fpCCR),a0 |
| 3021 |
movew IMM (0),a0@ |
| 3022 |
#ifndef __mcoldfire__ |
| 3023 |
moveml sp@+,d2-d7 | restore data registers |
| 3024 |
#else |
| 3025 |
moveml sp@,d2-d7 |
| 3026 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3027 |
| be adjusted here. |
| 3028 |
#endif |
| 3029 |
unlk a6 | and return |
| 3030 |
rts |
| 3031 |
|
| 3032 |
Laddsf$b$small: |
| 3033 |
movel a6@(8),d0 |
| 3034 |
PICLEA SYM (_fpCCR),a0 |
| 3035 |
movew IMM (0),a0@ |
| 3036 |
#ifndef __mcoldfire__ |
| 3037 |
moveml sp@+,d2-d7 | restore data registers |
| 3038 |
#else |
| 3039 |
moveml sp@,d2-d7 |
| 3040 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3041 |
| be adjusted here. |
| 3042 |
#endif |
| 3043 |
unlk a6 | and return |
| 3044 |
rts |
| 3045 |
|
| 3046 |
| If the numbers are denormalized remember to put exponent equal to 1. |
| 3047 |
|
| 3048 |
Laddsf$a$den: |
| 3049 |
movel d5,d6 | d5 contains 0x01000000 |
| 3050 |
swap d6 |
| 3051 |
bra Laddsf$1 |
| 3052 |
|
| 3053 |
Laddsf$b$den: |
| 3054 |
movel d5,d7 |
| 3055 |
swap d7 |
| 3056 |
notl d4 | make d4 into a mask for the fraction |
| 3057 |
| (this was not executed after the jump) |
| 3058 |
bra Laddsf$2 |
| 3059 |
|
| 3060 |
| The rest is mainly code for the different results which can be |
| 3061 |
| returned (checking always for +/-INFINITY and NaN). |
| 3062 |
|
| 3063 |
Laddsf$b: |
| 3064 |
| Return b (if a is zero). |
| 3065 |
movel a6@(12),d0 |
| 3066 |
cmpl IMM (0x80000000),d0 | Check if b is -0 |
| 3067 |
bne 1f |
| 3068 |
movel a0,d7 |
| 3069 |
andl IMM (0x80000000),d7 | Use the sign of a |
| 3070 |
clrl d0 |
| 3071 |
bra Laddsf$ret |
| 3072 |
Laddsf$a: |
| 3073 |
| Return a (if b is zero). |
| 3074 |
movel a6@(8),d0 |
| 3075 |
1: |
| 3076 |
moveq IMM (ADD),d5 |
| 3077 |
| We have to check for NaN and +/-infty. |
| 3078 |
movel d0,d7 |
| 3079 |
andl IMM (0x80000000),d7 | put sign in d7 |
| 3080 |
bclr IMM (31),d0 | clear sign |
| 3081 |
cmpl IMM (INFINITY),d0 | check for infty or NaN |
| 3082 |
bge 2f |
| 3083 |
movel d0,d0 | check for zero (we do this because we don't ' |
| 3084 |
bne Laddsf$ret | want to return -0 by mistake |
| 3085 |
bclr IMM (31),d7 | if zero be sure to clear sign |
| 3086 |
bra Laddsf$ret | if everything OK just return |
| 3087 |
2: |
| 3088 |
| The value to be returned is either +/-infty or NaN |
| 3089 |
andl IMM (0x007fffff),d0 | check for NaN |
| 3090 |
bne Lf$inop | if mantissa not zero is NaN |
| 3091 |
bra Lf$infty |
| 3092 |
|
| 3093 |
Laddsf$ret: |
| 3094 |
| Normal exit (a and b nonzero, result is not NaN nor +/-infty). |
| 3095 |
| We have to clear the exception flags (just the exception type). |
| 3096 |
PICLEA SYM (_fpCCR),a0 |
| 3097 |
movew IMM (0),a0@ |
| 3098 |
orl d7,d0 | put sign bit |
| 3099 |
#ifndef __mcoldfire__ |
| 3100 |
moveml sp@+,d2-d7 | restore data registers |
| 3101 |
#else |
| 3102 |
moveml sp@,d2-d7 |
| 3103 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3104 |
| be adjusted here. |
| 3105 |
#endif |
| 3106 |
unlk a6 | and return |
| 3107 |
rts |
| 3108 |
|
| 3109 |
Laddsf$ret$den: |
| 3110 |
| Return a denormalized number (for addition we don't signal underflow) ' |
| 3111 |
lsrl IMM (1),d0 | remember to shift right back once |
| 3112 |
bra Laddsf$ret | and return |
| 3113 |
|
| 3114 |
| Note: when adding two floats of the same sign if either one is |
| 3115 |
| NaN we return NaN without regard to whether the other is finite or |
| 3116 |
| not. When subtracting them (i.e., when adding two numbers of |
| 3117 |
| opposite signs) things are more complicated: if both are INFINITY |
| 3118 |
| we return NaN, if only one is INFINITY and the other is NaN we return |
| 3119 |
| NaN, but if it is finite we return INFINITY with the corresponding sign. |
| 3120 |
|
| 3121 |
Laddsf$nf: |
| 3122 |
moveq IMM (ADD),d5 |
| 3123 |
| This could be faster but it is not worth the effort, since it is not |
| 3124 |
| executed very often. We sacrifice speed for clarity here. |
| 3125 |
movel a6@(8),d0 | get the numbers back (remember that we |
| 3126 |
movel a6@(12),d1 | did some processing already) |
| 3127 |
movel IMM (INFINITY),d4 | useful constant (INFINITY) |
| 3128 |
movel d0,d2 | save sign bits |
| 3129 |
movel d0,d7 | into d7 as well as we may need the sign |
| 3130 |
| bit before jumping to LfSinfty |
| 3131 |
movel d1,d3 |
| 3132 |
bclr IMM (31),d0 | clear sign bits |
| 3133 |
bclr IMM (31),d1 |
| 3134 |
| We know that one of them is either NaN of +/-INFINITY |
| 3135 |
| Check for NaN (if either one is NaN return NaN) |
| 3136 |
cmpl d4,d0 | check first a (d0) |
| 3137 |
bhi Lf$inop |
| 3138 |
cmpl d4,d1 | check now b (d1) |
| 3139 |
bhi Lf$inop |
| 3140 |
| Now comes the check for +/-INFINITY. We know that both are (maybe not |
| 3141 |
| finite) numbers, but we have to check if both are infinite whether we |
| 3142 |
| are adding or subtracting them. |
| 3143 |
eorl d3,d2 | to check sign bits |
| 3144 |
bmi 1f |
| 3145 |
andl IMM (0x80000000),d7 | get (common) sign bit |
| 3146 |
bra Lf$infty |
| 3147 |
1: |
| 3148 |
| We know one (or both) are infinite, so we test for equality between the |
| 3149 |
| two numbers (if they are equal they have to be infinite both, so we |
| 3150 |
| return NaN). |
| 3151 |
cmpl d1,d0 | are both infinite? |
| 3152 |
beq Lf$inop | if so return NaN |
| 3153 |
|
| 3154 |
andl IMM (0x80000000),d7 | get a's sign bit ' |
| 3155 |
cmpl d4,d0 | test now for infinity |
| 3156 |
beq Lf$infty | if a is INFINITY return with this sign |
| 3157 |
bchg IMM (31),d7 | else we know b is INFINITY and has |
| 3158 |
bra Lf$infty | the opposite sign |
| 3159 |
|
| 3160 |
|============================================================================= |
| 3161 |
| __mulsf3 |
| 3162 |
|============================================================================= |
| 3163 |
|
| 3164 |
| float __mulsf3(float, float); |
| 3165 |
FUNC(__mulsf3) |
| 3166 |
SYM (__mulsf3): |
| 3167 |
#ifndef __mcoldfire__ |
| 3168 |
link a6,IMM (0) |
| 3169 |
moveml d2-d7,sp@- |
| 3170 |
#else |
| 3171 |
link a6,IMM (-24) |
| 3172 |
moveml d2-d7,sp@ |
| 3173 |
#endif |
| 3174 |
movel a6@(8),d0 | get a into d0 |
| 3175 |
movel a6@(12),d1 | and b into d1 |
| 3176 |
movel d0,d7 | d7 will hold the sign of the product |
| 3177 |
eorl d1,d7 | |
| 3178 |
andl IMM (0x80000000),d7 |
| 3179 |
movel IMM (INFINITY),d6 | useful constant (+INFINITY) |
| 3180 |
movel d6,d5 | another (mask for fraction) |
| 3181 |
notl d5 | |
| 3182 |
movel IMM (0x00800000),d4 | this is to put hidden bit back |
| 3183 |
bclr IMM (31),d0 | get rid of a's sign bit ' |
| 3184 |
movel d0,d2 | |
| 3185 |
beq Lmulsf$a$0 | branch if a is zero |
| 3186 |
bclr IMM (31),d1 | get rid of b's sign bit ' |
| 3187 |
movel d1,d3 | |
| 3188 |
beq Lmulsf$b$0 | branch if b is zero |
| 3189 |
cmpl d6,d0 | is a big? |
| 3190 |
bhi Lmulsf$inop | if a is NaN return NaN |
| 3191 |
beq Lmulsf$inf | if a is INFINITY we have to check b |
| 3192 |
cmpl d6,d1 | now compare b with INFINITY |
| 3193 |
bhi Lmulsf$inop | is b NaN? |
| 3194 |
beq Lmulsf$overflow | is b INFINITY? |
| 3195 |
| Here we have both numbers finite and nonzero (and with no sign bit). |
| 3196 |
| Now we get the exponents into d2 and d3. |
| 3197 |
andl d6,d2 | and isolate exponent in d2 |
| 3198 |
beq Lmulsf$a$den | if exponent is zero we have a denormalized |
| 3199 |
andl d5,d0 | and isolate fraction |
| 3200 |
orl d4,d0 | and put hidden bit back |
| 3201 |
swap d2 | I like exponents in the first byte |
| 3202 |
#ifndef __mcoldfire__ |
| 3203 |
lsrw IMM (7),d2 | |
| 3204 |
#else |
| 3205 |
lsrl IMM (7),d2 | |
| 3206 |
#endif |
| 3207 |
Lmulsf$1: | number |
| 3208 |
andl d6,d3 | |
| 3209 |
beq Lmulsf$b$den | |
| 3210 |
andl d5,d1 | |
| 3211 |
orl d4,d1 | |
| 3212 |
swap d3 | |
| 3213 |
#ifndef __mcoldfire__ |
| 3214 |
lsrw IMM (7),d3 | |
| 3215 |
#else |
| 3216 |
lsrl IMM (7),d3 | |
| 3217 |
#endif |
| 3218 |
Lmulsf$2: | |
| 3219 |
#ifndef __mcoldfire__ |
| 3220 |
addw d3,d2 | add exponents |
| 3221 |
subw IMM (F_BIAS+1),d2 | and subtract bias (plus one) |
| 3222 |
#else |
| 3223 |
addl d3,d2 | add exponents |
| 3224 |
subl IMM (F_BIAS+1),d2 | and subtract bias (plus one) |
| 3225 |
#endif |
| 3226 |
|
| 3227 |
| We are now ready to do the multiplication. The situation is as follows: |
| 3228 |
| both a and b have bit FLT_MANT_DIG-1 set (even if they were |
| 3229 |
| denormalized to start with!), which means that in the product |
| 3230 |
| bit 2*(FLT_MANT_DIG-1) (that is, bit 2*FLT_MANT_DIG-2-32 of the |
| 3231 |
| high long) is set. |
| 3232 |
|
| 3233 |
| To do the multiplication let us move the number a little bit around ... |
| 3234 |
movel d1,d6 | second operand in d6 |
| 3235 |
movel d0,d5 | first operand in d4-d5 |
| 3236 |
movel IMM (0),d4 |
| 3237 |
movel d4,d1 | the sums will go in d0-d1 |
| 3238 |
movel d4,d0 |
| 3239 |
|
| 3240 |
| now bit FLT_MANT_DIG-1 becomes bit 31: |
| 3241 |
lsll IMM (31-FLT_MANT_DIG+1),d6 |
| 3242 |
|
| 3243 |
| Start the loop (we loop #FLT_MANT_DIG times): |
| 3244 |
moveq IMM (FLT_MANT_DIG-1),d3 |
| 3245 |
1: addl d1,d1 | shift sum |
| 3246 |
addxl d0,d0 |
| 3247 |
lsll IMM (1),d6 | get bit bn |
| 3248 |
bcc 2f | if not set skip sum |
| 3249 |
addl d5,d1 | add a |
| 3250 |
addxl d4,d0 |
| 3251 |
2: |
| 3252 |
#ifndef __mcoldfire__ |
| 3253 |
dbf d3,1b | loop back |
| 3254 |
#else |
| 3255 |
subql IMM (1),d3 |
| 3256 |
bpl 1b |
| 3257 |
#endif |
| 3258 |
|
| 3259 |
| Now we have the product in d0-d1, with bit (FLT_MANT_DIG - 1) + FLT_MANT_DIG |
| 3260 |
| (mod 32) of d0 set. The first thing to do now is to normalize it so bit |
| 3261 |
| FLT_MANT_DIG is set (to do the rounding). |
| 3262 |
#ifndef __mcoldfire__ |
| 3263 |
rorl IMM (6),d1 |
| 3264 |
swap d1 |
| 3265 |
movew d1,d3 |
| 3266 |
andw IMM (0x03ff),d3 |
| 3267 |
andw IMM (0xfd00),d1 |
| 3268 |
#else |
| 3269 |
movel d1,d3 |
| 3270 |
lsll IMM (8),d1 |
| 3271 |
addl d1,d1 |
| 3272 |
addl d1,d1 |
| 3273 |
moveq IMM (22),d5 |
| 3274 |
lsrl d5,d3 |
| 3275 |
orl d3,d1 |
| 3276 |
andl IMM (0xfffffd00),d1 |
| 3277 |
#endif |
| 3278 |
lsll IMM (8),d0 |
| 3279 |
addl d0,d0 |
| 3280 |
addl d0,d0 |
| 3281 |
#ifndef __mcoldfire__ |
| 3282 |
orw d3,d0 |
| 3283 |
#else |
| 3284 |
orl d3,d0 |
| 3285 |
#endif |
| 3286 |
|
| 3287 |
moveq IMM (MULTIPLY),d5 |
| 3288 |
|
| 3289 |
btst IMM (FLT_MANT_DIG+1),d0 |
| 3290 |
beq Lround$exit |
| 3291 |
#ifndef __mcoldfire__ |
| 3292 |
lsrl IMM (1),d0 |
| 3293 |
roxrl IMM (1),d1 |
| 3294 |
addw IMM (1),d2 |
| 3295 |
#else |
| 3296 |
lsrl IMM (1),d1 |
| 3297 |
btst IMM (0),d0 |
| 3298 |
beq 10f |
| 3299 |
bset IMM (31),d1 |
| 3300 |
10: lsrl IMM (1),d0 |
| 3301 |
addql IMM (1),d2 |
| 3302 |
#endif |
| 3303 |
bra Lround$exit |
| 3304 |
|
| 3305 |
Lmulsf$inop: |
| 3306 |
moveq IMM (MULTIPLY),d5 |
| 3307 |
bra Lf$inop |
| 3308 |
|
| 3309 |
Lmulsf$overflow: |
| 3310 |
moveq IMM (MULTIPLY),d5 |
| 3311 |
bra Lf$overflow |
| 3312 |
|
| 3313 |
Lmulsf$inf: |
| 3314 |
moveq IMM (MULTIPLY),d5 |
| 3315 |
| If either is NaN return NaN; else both are (maybe infinite) numbers, so |
| 3316 |
| return INFINITY with the correct sign (which is in d7). |
| 3317 |
cmpl d6,d1 | is b NaN? |
| 3318 |
bhi Lf$inop | if so return NaN |
| 3319 |
bra Lf$overflow | else return +/-INFINITY |
| 3320 |
|
| 3321 |
| If either number is zero return zero, unless the other is +/-INFINITY, |
| 3322 |
| or NaN, in which case we return NaN. |
| 3323 |
Lmulsf$b$0: |
| 3324 |
| Here d1 (==b) is zero. |
| 3325 |
movel a6@(8),d1 | get a again to check for non-finiteness |
| 3326 |
bra 1f |
| 3327 |
Lmulsf$a$0: |
| 3328 |
movel a6@(12),d1 | get b again to check for non-finiteness |
| 3329 |
1: bclr IMM (31),d1 | clear sign bit |
| 3330 |
cmpl IMM (INFINITY),d1 | and check for a large exponent |
| 3331 |
bge Lf$inop | if b is +/-INFINITY or NaN return NaN |
| 3332 |
movel d7,d0 | else return signed zero |
| 3333 |
PICLEA SYM (_fpCCR),a0 | |
| 3334 |
movew IMM (0),a0@ | |
| 3335 |
#ifndef __mcoldfire__ |
| 3336 |
moveml sp@+,d2-d7 | |
| 3337 |
#else |
| 3338 |
moveml sp@,d2-d7 |
| 3339 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3340 |
| be adjusted here. |
| 3341 |
#endif |
| 3342 |
unlk a6 | |
| 3343 |
rts | |
| 3344 |
|
| 3345 |
| If a number is denormalized we put an exponent of 1 but do not put the |
| 3346 |
| hidden bit back into the fraction; instead we shift left until bit 23 |
| 3347 |
| (the hidden bit) is set, adjusting the exponent accordingly. We do this |
| 3348 |
| to ensure that the product of the fractions is close to 1. |
| 3349 |
Lmulsf$a$den: |
| 3350 |
movel IMM (1),d2 |
| 3351 |
andl d5,d0 |
| 3352 |
1: addl d0,d0 | shift a left (until bit 23 is set) |
| 3353 |
#ifndef __mcoldfire__ |
| 3354 |
subw IMM (1),d2 | and adjust exponent |
| 3355 |
#else |
| 3356 |
subql IMM (1),d2 | and adjust exponent |
| 3357 |
#endif |
| 3358 |
btst IMM (FLT_MANT_DIG-1),d0 |
| 3359 |
bne Lmulsf$1 | |
| 3360 |
bra 1b | else loop back |
| 3361 |
|
| 3362 |
Lmulsf$b$den: |
| 3363 |
movel IMM (1),d3 |
| 3364 |
andl d5,d1 |
| 3365 |
1: addl d1,d1 | shift b left until bit 23 is set |
| 3366 |
#ifndef __mcoldfire__ |
| 3367 |
subw IMM (1),d3 | and adjust exponent |
| 3368 |
#else |
| 3369 |
subql IMM (1),d3 | and adjust exponent |
| 3370 |
#endif |
| 3371 |
btst IMM (FLT_MANT_DIG-1),d1 |
| 3372 |
bne Lmulsf$2 | |
| 3373 |
bra 1b | else loop back |
| 3374 |
|
| 3375 |
|============================================================================= |
| 3376 |
| __divsf3 |
| 3377 |
|============================================================================= |
| 3378 |
|
| 3379 |
| float __divsf3(float, float); |
| 3380 |
FUNC(__divsf3) |
| 3381 |
SYM (__divsf3): |
| 3382 |
#ifndef __mcoldfire__ |
| 3383 |
link a6,IMM (0) |
| 3384 |
moveml d2-d7,sp@- |
| 3385 |
#else |
| 3386 |
link a6,IMM (-24) |
| 3387 |
moveml d2-d7,sp@ |
| 3388 |
#endif |
| 3389 |
movel a6@(8),d0 | get a into d0 |
| 3390 |
movel a6@(12),d1 | and b into d1 |
| 3391 |
movel d0,d7 | d7 will hold the sign of the result |
| 3392 |
eorl d1,d7 | |
| 3393 |
andl IMM (0x80000000),d7 | |
| 3394 |
movel IMM (INFINITY),d6 | useful constant (+INFINITY) |
| 3395 |
movel d6,d5 | another (mask for fraction) |
| 3396 |
notl d5 | |
| 3397 |
movel IMM (0x00800000),d4 | this is to put hidden bit back |
| 3398 |
bclr IMM (31),d0 | get rid of a's sign bit ' |
| 3399 |
movel d0,d2 | |
| 3400 |
beq Ldivsf$a$0 | branch if a is zero |
| 3401 |
bclr IMM (31),d1 | get rid of b's sign bit ' |
| 3402 |
movel d1,d3 | |
| 3403 |
beq Ldivsf$b$0 | branch if b is zero |
| 3404 |
cmpl d6,d0 | is a big? |
| 3405 |
bhi Ldivsf$inop | if a is NaN return NaN |
| 3406 |
beq Ldivsf$inf | if a is INFINITY we have to check b |
| 3407 |
cmpl d6,d1 | now compare b with INFINITY |
| 3408 |
bhi Ldivsf$inop | if b is NaN return NaN |
| 3409 |
beq Ldivsf$underflow |
| 3410 |
| Here we have both numbers finite and nonzero (and with no sign bit). |
| 3411 |
| Now we get the exponents into d2 and d3 and normalize the numbers to |
| 3412 |
| ensure that the ratio of the fractions is close to 1. We do this by |
| 3413 |
| making sure that bit #FLT_MANT_DIG-1 (hidden bit) is set. |
| 3414 |
andl d6,d2 | and isolate exponent in d2 |
| 3415 |
beq Ldivsf$a$den | if exponent is zero we have a denormalized |
| 3416 |
andl d5,d0 | and isolate fraction |
| 3417 |
orl d4,d0 | and put hidden bit back |
| 3418 |
swap d2 | I like exponents in the first byte |
| 3419 |
#ifndef __mcoldfire__ |
| 3420 |
lsrw IMM (7),d2 | |
| 3421 |
#else |
| 3422 |
lsrl IMM (7),d2 | |
| 3423 |
#endif |
| 3424 |
Ldivsf$1: | |
| 3425 |
andl d6,d3 | |
| 3426 |
beq Ldivsf$b$den | |
| 3427 |
andl d5,d1 | |
| 3428 |
orl d4,d1 | |
| 3429 |
swap d3 | |
| 3430 |
#ifndef __mcoldfire__ |
| 3431 |
lsrw IMM (7),d3 | |
| 3432 |
#else |
| 3433 |
lsrl IMM (7),d3 | |
| 3434 |
#endif |
| 3435 |
Ldivsf$2: | |
| 3436 |
#ifndef __mcoldfire__ |
| 3437 |
subw d3,d2 | subtract exponents |
| 3438 |
addw IMM (F_BIAS),d2 | and add bias |
| 3439 |
#else |
| 3440 |
subl d3,d2 | subtract exponents |
| 3441 |
addl IMM (F_BIAS),d2 | and add bias |
| 3442 |
#endif |
| 3443 |
|
| 3444 |
| We are now ready to do the division. We have prepared things in such a way |
| 3445 |
| that the ratio of the fractions will be less than 2 but greater than 1/2. |
| 3446 |
| At this point the registers in use are: |
| 3447 |
| d0 holds a (first operand, bit FLT_MANT_DIG=0, bit FLT_MANT_DIG-1=1) |
| 3448 |
| d1 holds b (second operand, bit FLT_MANT_DIG=1) |
| 3449 |
| d2 holds the difference of the exponents, corrected by the bias |
| 3450 |
| d7 holds the sign of the ratio |
| 3451 |
| d4, d5, d6 hold some constants |
| 3452 |
movel d7,a0 | d6-d7 will hold the ratio of the fractions |
| 3453 |
movel IMM (0),d6 | |
| 3454 |
movel d6,d7 |
| 3455 |
|
| 3456 |
moveq IMM (FLT_MANT_DIG+1),d3 |
| 3457 |
1: cmpl d0,d1 | is a < b? |
| 3458 |
bhi 2f | |
| 3459 |
bset d3,d6 | set a bit in d6 |
| 3460 |
subl d1,d0 | if a >= b a <-- a-b |
| 3461 |
beq 3f | if a is zero, exit |
| 3462 |
2: addl d0,d0 | multiply a by 2 |
| 3463 |
#ifndef __mcoldfire__ |
| 3464 |
dbra d3,1b |
| 3465 |
#else |
| 3466 |
subql IMM (1),d3 |
| 3467 |
bpl 1b |
| 3468 |
#endif |
| 3469 |
|
| 3470 |
| Now we keep going to set the sticky bit ... |
| 3471 |
moveq IMM (FLT_MANT_DIG),d3 |
| 3472 |
1: cmpl d0,d1 |
| 3473 |
ble 2f |
| 3474 |
addl d0,d0 |
| 3475 |
#ifndef __mcoldfire__ |
| 3476 |
dbra d3,1b |
| 3477 |
#else |
| 3478 |
subql IMM(1),d3 |
| 3479 |
bpl 1b |
| 3480 |
#endif |
| 3481 |
movel IMM (0),d1 |
| 3482 |
bra 3f |
| 3483 |
2: movel IMM (0),d1 |
| 3484 |
#ifndef __mcoldfire__ |
| 3485 |
subw IMM (FLT_MANT_DIG),d3 |
| 3486 |
addw IMM (31),d3 |
| 3487 |
#else |
| 3488 |
subl IMM (FLT_MANT_DIG),d3 |
| 3489 |
addl IMM (31),d3 |
| 3490 |
#endif |
| 3491 |
bset d3,d1 |
| 3492 |
3: |
| 3493 |
movel d6,d0 | put the ratio in d0-d1 |
| 3494 |
movel a0,d7 | get sign back |
| 3495 |
|
| 3496 |
| Because of the normalization we did before we are guaranteed that |
| 3497 |
| d0 is smaller than 2^26 but larger than 2^24. Thus bit 26 is not set, |
| 3498 |
| bit 25 could be set, and if it is not set then bit 24 is necessarily set. |
| 3499 |
btst IMM (FLT_MANT_DIG+1),d0 |
| 3500 |
beq 1f | if it is not set, then bit 24 is set |
| 3501 |
lsrl IMM (1),d0 | |
| 3502 |
#ifndef __mcoldfire__ |
| 3503 |
addw IMM (1),d2 | |
| 3504 |
#else |
| 3505 |
addl IMM (1),d2 | |
| 3506 |
#endif |
| 3507 |
1: |
| 3508 |
| Now round, check for over- and underflow, and exit. |
| 3509 |
moveq IMM (DIVIDE),d5 |
| 3510 |
bra Lround$exit |
| 3511 |
|
| 3512 |
Ldivsf$inop: |
| 3513 |
moveq IMM (DIVIDE),d5 |
| 3514 |
bra Lf$inop |
| 3515 |
|
| 3516 |
Ldivsf$overflow: |
| 3517 |
moveq IMM (DIVIDE),d5 |
| 3518 |
bra Lf$overflow |
| 3519 |
|
| 3520 |
Ldivsf$underflow: |
| 3521 |
moveq IMM (DIVIDE),d5 |
| 3522 |
bra Lf$underflow |
| 3523 |
|
| 3524 |
Ldivsf$a$0: |
| 3525 |
moveq IMM (DIVIDE),d5 |
| 3526 |
| If a is zero check to see whether b is zero also. In that case return |
| 3527 |
| NaN; then check if b is NaN, and return NaN also in that case. Else |
| 3528 |
| return a properly signed zero. |
| 3529 |
andl IMM (0x7fffffff),d1 | clear sign bit and test b |
| 3530 |
beq Lf$inop | if b is also zero return NaN |
| 3531 |
cmpl IMM (INFINITY),d1 | check for NaN |
| 3532 |
bhi Lf$inop | |
| 3533 |
movel d7,d0 | else return signed zero |
| 3534 |
PICLEA SYM (_fpCCR),a0 | |
| 3535 |
movew IMM (0),a0@ | |
| 3536 |
#ifndef __mcoldfire__ |
| 3537 |
moveml sp@+,d2-d7 | |
| 3538 |
#else |
| 3539 |
moveml sp@,d2-d7 | |
| 3540 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3541 |
| be adjusted here. |
| 3542 |
#endif |
| 3543 |
unlk a6 | |
| 3544 |
rts | |
| 3545 |
|
| 3546 |
Ldivsf$b$0: |
| 3547 |
moveq IMM (DIVIDE),d5 |
| 3548 |
| If we got here a is not zero. Check if a is NaN; in that case return NaN, |
| 3549 |
| else return +/-INFINITY. Remember that a is in d0 with the sign bit |
| 3550 |
| cleared already. |
| 3551 |
cmpl IMM (INFINITY),d0 | compare d0 with INFINITY |
| 3552 |
bhi Lf$inop | if larger it is NaN |
| 3553 |
bra Lf$div$0 | else signal DIVIDE_BY_ZERO |
| 3554 |
|
| 3555 |
Ldivsf$inf: |
| 3556 |
moveq IMM (DIVIDE),d5 |
| 3557 |
| If a is INFINITY we have to check b |
| 3558 |
cmpl IMM (INFINITY),d1 | compare b with INFINITY |
| 3559 |
bge Lf$inop | if b is NaN or INFINITY return NaN |
| 3560 |
bra Lf$overflow | else return overflow |
| 3561 |
|
| 3562 |
| If a number is denormalized we put an exponent of 1 but do not put the |
| 3563 |
| bit back into the fraction. |
| 3564 |
Ldivsf$a$den: |
| 3565 |
movel IMM (1),d2 |
| 3566 |
andl d5,d0 |
| 3567 |
1: addl d0,d0 | shift a left until bit FLT_MANT_DIG-1 is set |
| 3568 |
#ifndef __mcoldfire__ |
| 3569 |
subw IMM (1),d2 | and adjust exponent |
| 3570 |
#else |
| 3571 |
subl IMM (1),d2 | and adjust exponent |
| 3572 |
#endif |
| 3573 |
btst IMM (FLT_MANT_DIG-1),d0 |
| 3574 |
bne Ldivsf$1 |
| 3575 |
bra 1b |
| 3576 |
|
| 3577 |
Ldivsf$b$den: |
| 3578 |
movel IMM (1),d3 |
| 3579 |
andl d5,d1 |
| 3580 |
1: addl d1,d1 | shift b left until bit FLT_MANT_DIG is set |
| 3581 |
#ifndef __mcoldfire__ |
| 3582 |
subw IMM (1),d3 | and adjust exponent |
| 3583 |
#else |
| 3584 |
subl IMM (1),d3 | and adjust exponent |
| 3585 |
#endif |
| 3586 |
btst IMM (FLT_MANT_DIG-1),d1 |
| 3587 |
bne Ldivsf$2 |
| 3588 |
bra 1b |
| 3589 |
|
| 3590 |
Lround$exit: |
| 3591 |
| This is a common exit point for __mulsf3 and __divsf3. |
| 3592 |
|
| 3593 |
| First check for underlow in the exponent: |
| 3594 |
#ifndef __mcoldfire__ |
| 3595 |
cmpw IMM (-FLT_MANT_DIG-1),d2 |
| 3596 |
#else |
| 3597 |
cmpl IMM (-FLT_MANT_DIG-1),d2 |
| 3598 |
#endif |
| 3599 |
blt Lf$underflow |
| 3600 |
| It could happen that the exponent is less than 1, in which case the |
| 3601 |
| number is denormalized. In this case we shift right and adjust the |
| 3602 |
| exponent until it becomes 1 or the fraction is zero (in the latter case |
| 3603 |
| we signal underflow and return zero). |
| 3604 |
movel IMM (0),d6 | d6 is used temporarily |
| 3605 |
#ifndef __mcoldfire__ |
| 3606 |
cmpw IMM (1),d2 | if the exponent is less than 1 we |
| 3607 |
#else |
| 3608 |
cmpl IMM (1),d2 | if the exponent is less than 1 we |
| 3609 |
#endif |
| 3610 |
bge 2f | have to shift right (denormalize) |
| 3611 |
1: |
| 3612 |
#ifndef __mcoldfire__ |
| 3613 |
addw IMM (1),d2 | adjust the exponent |
| 3614 |
lsrl IMM (1),d0 | shift right once |
| 3615 |
roxrl IMM (1),d1 | |
| 3616 |
roxrl IMM (1),d6 | d6 collect bits we would lose otherwise |
| 3617 |
cmpw IMM (1),d2 | is the exponent 1 already? |
| 3618 |
#else |
| 3619 |
addql IMM (1),d2 | adjust the exponent |
| 3620 |
lsrl IMM (1),d6 |
| 3621 |
btst IMM (0),d1 |
| 3622 |
beq 11f |
| 3623 |
bset IMM (31),d6 |
| 3624 |
11: lsrl IMM (1),d1 |
| 3625 |
btst IMM (0),d0 |
| 3626 |
beq 10f |
| 3627 |
bset IMM (31),d1 |
| 3628 |
10: lsrl IMM (1),d0 |
| 3629 |
cmpl IMM (1),d2 | is the exponent 1 already? |
| 3630 |
#endif |
| 3631 |
beq 2f | if not loop back |
| 3632 |
bra 1b | |
| 3633 |
bra Lf$underflow | safety check, shouldn't execute ' |
| 3634 |
2: orl d6,d1 | this is a trick so we don't lose ' |
| 3635 |
| the extra bits which were flushed right |
| 3636 |
| Now call the rounding routine (which takes care of denormalized numbers): |
| 3637 |
lea pc@(Lround$0),a0 | to return from rounding routine |
| 3638 |
PICLEA SYM (_fpCCR),a1 | check the rounding mode |
| 3639 |
#ifdef __mcoldfire__ |
| 3640 |
clrl d6 |
| 3641 |
#endif |
| 3642 |
movew a1@(6),d6 | rounding mode in d6 |
| 3643 |
beq Lround$to$nearest |
| 3644 |
#ifndef __mcoldfire__ |
| 3645 |
cmpw IMM (ROUND_TO_PLUS),d6 |
| 3646 |
#else |
| 3647 |
cmpl IMM (ROUND_TO_PLUS),d6 |
| 3648 |
#endif |
| 3649 |
bhi Lround$to$minus |
| 3650 |
blt Lround$to$zero |
| 3651 |
bra Lround$to$plus |
| 3652 |
Lround$0: |
| 3653 |
| Here we have a correctly rounded result (either normalized or denormalized). |
| 3654 |
|
| 3655 |
| Here we should have either a normalized number or a denormalized one, and |
| 3656 |
| the exponent is necessarily larger or equal to 1 (so we don't have to ' |
| 3657 |
| check again for underflow!). We have to check for overflow or for a |
| 3658 |
| denormalized number (which also signals underflow). |
| 3659 |
| Check for overflow (i.e., exponent >= 255). |
| 3660 |
#ifndef __mcoldfire__ |
| 3661 |
cmpw IMM (0x00ff),d2 |
| 3662 |
#else |
| 3663 |
cmpl IMM (0x00ff),d2 |
| 3664 |
#endif |
| 3665 |
bge Lf$overflow |
| 3666 |
| Now check for a denormalized number (exponent==0). |
| 3667 |
movew d2,d2 |
| 3668 |
beq Lf$den |
| 3669 |
1: |
| 3670 |
| Put back the exponents and sign and return. |
| 3671 |
#ifndef __mcoldfire__ |
| 3672 |
lslw IMM (7),d2 | exponent back to fourth byte |
| 3673 |
#else |
| 3674 |
lsll IMM (7),d2 | exponent back to fourth byte |
| 3675 |
#endif |
| 3676 |
bclr IMM (FLT_MANT_DIG-1),d0 |
| 3677 |
swap d0 | and put back exponent |
| 3678 |
#ifndef __mcoldfire__ |
| 3679 |
orw d2,d0 | |
| 3680 |
#else |
| 3681 |
orl d2,d0 |
| 3682 |
#endif |
| 3683 |
swap d0 | |
| 3684 |
orl d7,d0 | and sign also |
| 3685 |
|
| 3686 |
PICLEA SYM (_fpCCR),a0 |
| 3687 |
movew IMM (0),a0@ |
| 3688 |
#ifndef __mcoldfire__ |
| 3689 |
moveml sp@+,d2-d7 |
| 3690 |
#else |
| 3691 |
moveml sp@,d2-d7 |
| 3692 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3693 |
| be adjusted here. |
| 3694 |
#endif |
| 3695 |
unlk a6 |
| 3696 |
rts |
| 3697 |
|
| 3698 |
|============================================================================= |
| 3699 |
| __negsf2 |
| 3700 |
|============================================================================= |
| 3701 |
|
| 3702 |
| This is trivial and could be shorter if we didn't bother checking for NaN ' |
| 3703 |
| and +/-INFINITY. |
| 3704 |
|
| 3705 |
| float __negsf2(float); |
| 3706 |
FUNC(__negsf2) |
| 3707 |
SYM (__negsf2): |
| 3708 |
#ifndef __mcoldfire__ |
| 3709 |
link a6,IMM (0) |
| 3710 |
moveml d2-d7,sp@- |
| 3711 |
#else |
| 3712 |
link a6,IMM (-24) |
| 3713 |
moveml d2-d7,sp@ |
| 3714 |
#endif |
| 3715 |
moveq IMM (NEGATE),d5 |
| 3716 |
movel a6@(8),d0 | get number to negate in d0 |
| 3717 |
bchg IMM (31),d0 | negate |
| 3718 |
movel d0,d1 | make a positive copy |
| 3719 |
bclr IMM (31),d1 | |
| 3720 |
tstl d1 | check for zero |
| 3721 |
beq 2f | if zero (either sign) return +zero |
| 3722 |
cmpl IMM (INFINITY),d1 | compare to +INFINITY |
| 3723 |
blt 1f | |
| 3724 |
bhi Lf$inop | if larger (fraction not zero) is NaN |
| 3725 |
movel d0,d7 | else get sign and return INFINITY |
| 3726 |
andl IMM (0x80000000),d7 |
| 3727 |
bra Lf$infty |
| 3728 |
1: PICLEA SYM (_fpCCR),a0 |
| 3729 |
movew IMM (0),a0@ |
| 3730 |
#ifndef __mcoldfire__ |
| 3731 |
moveml sp@+,d2-d7 |
| 3732 |
#else |
| 3733 |
moveml sp@,d2-d7 |
| 3734 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3735 |
| be adjusted here. |
| 3736 |
#endif |
| 3737 |
unlk a6 |
| 3738 |
rts |
| 3739 |
2: bclr IMM (31),d0 |
| 3740 |
bra 1b |
| 3741 |
|
| 3742 |
|============================================================================= |
| 3743 |
| __cmpsf2 |
| 3744 |
|============================================================================= |
| 3745 |
|
| 3746 |
GREATER = 1 |
| 3747 |
LESS = -1 |
| 3748 |
EQUAL = 0 |
| 3749 |
|
| 3750 |
| int __cmpsf2_internal(float, float, int); |
| 3751 |
SYM (__cmpsf2_internal): |
| 3752 |
#ifndef __mcoldfire__ |
| 3753 |
link a6,IMM (0) |
| 3754 |
moveml d2-d7,sp@- | save registers |
| 3755 |
#else |
| 3756 |
link a6,IMM (-24) |
| 3757 |
moveml d2-d7,sp@ |
| 3758 |
#endif |
| 3759 |
moveq IMM (COMPARE),d5 |
| 3760 |
movel a6@(8),d0 | get first operand |
| 3761 |
movel a6@(12),d1 | get second operand |
| 3762 |
| Check if either is NaN, and in that case return garbage and signal |
| 3763 |
| INVALID_OPERATION. Check also if either is zero, and clear the signs |
| 3764 |
| if necessary. |
| 3765 |
movel d0,d6 |
| 3766 |
andl IMM (0x7fffffff),d0 |
| 3767 |
beq Lcmpsf$a$0 |
| 3768 |
cmpl IMM (0x7f800000),d0 |
| 3769 |
bhi Lcmpf$inop |
| 3770 |
Lcmpsf$1: |
| 3771 |
movel d1,d7 |
| 3772 |
andl IMM (0x7fffffff),d1 |
| 3773 |
beq Lcmpsf$b$0 |
| 3774 |
cmpl IMM (0x7f800000),d1 |
| 3775 |
bhi Lcmpf$inop |
| 3776 |
Lcmpsf$2: |
| 3777 |
| Check the signs |
| 3778 |
eorl d6,d7 |
| 3779 |
bpl 1f |
| 3780 |
| If the signs are not equal check if a >= 0 |
| 3781 |
tstl d6 |
| 3782 |
bpl Lcmpsf$a$gt$b | if (a >= 0 && b < 0) => a > b |
| 3783 |
bmi Lcmpsf$b$gt$a | if (a < 0 && b >= 0) => a < b |
| 3784 |
1: |
| 3785 |
| If the signs are equal check for < 0 |
| 3786 |
tstl d6 |
| 3787 |
bpl 1f |
| 3788 |
| If both are negative exchange them |
| 3789 |
#ifndef __mcoldfire__ |
| 3790 |
exg d0,d1 |
| 3791 |
#else |
| 3792 |
movel d0,d7 |
| 3793 |
movel d1,d0 |
| 3794 |
movel d7,d1 |
| 3795 |
#endif |
| 3796 |
1: |
| 3797 |
| Now that they are positive we just compare them as longs (does this also |
| 3798 |
| work for denormalized numbers?). |
| 3799 |
cmpl d0,d1 |
| 3800 |
bhi Lcmpsf$b$gt$a | |b| > |a| |
| 3801 |
bne Lcmpsf$a$gt$b | |b| < |a| |
| 3802 |
| If we got here a == b. |
| 3803 |
movel IMM (EQUAL),d0 |
| 3804 |
#ifndef __mcoldfire__ |
| 3805 |
moveml sp@+,d2-d7 | put back the registers |
| 3806 |
#else |
| 3807 |
moveml sp@,d2-d7 |
| 3808 |
#endif |
| 3809 |
unlk a6 |
| 3810 |
rts |
| 3811 |
Lcmpsf$a$gt$b: |
| 3812 |
movel IMM (GREATER),d0 |
| 3813 |
#ifndef __mcoldfire__ |
| 3814 |
moveml sp@+,d2-d7 | put back the registers |
| 3815 |
#else |
| 3816 |
moveml sp@,d2-d7 |
| 3817 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3818 |
| be adjusted here. |
| 3819 |
#endif |
| 3820 |
unlk a6 |
| 3821 |
rts |
| 3822 |
Lcmpsf$b$gt$a: |
| 3823 |
movel IMM (LESS),d0 |
| 3824 |
#ifndef __mcoldfire__ |
| 3825 |
moveml sp@+,d2-d7 | put back the registers |
| 3826 |
#else |
| 3827 |
moveml sp@,d2-d7 |
| 3828 |
| XXX if frame pointer is ever removed, stack pointer must |
| 3829 |
| be adjusted here. |
| 3830 |
#endif |
| 3831 |
unlk a6 |
| 3832 |
rts |
| 3833 |
|
| 3834 |
Lcmpsf$a$0: |
| 3835 |
bclr IMM (31),d6 |
| 3836 |
bra Lcmpsf$1 |
| 3837 |
Lcmpsf$b$0: |
| 3838 |
bclr IMM (31),d7 |
| 3839 |
bra Lcmpsf$2 |
| 3840 |
|
| 3841 |
Lcmpf$inop: |
| 3842 |
movl a6@(16),d0 |
| 3843 |
moveq IMM (INEXACT_RESULT+INVALID_OPERATION),d7 |
| 3844 |
moveq IMM (SINGLE_FLOAT),d6 |
| 3845 |
PICJUMP $_exception_handler |
| 3846 |
|
| 3847 |
| int __cmpsf2(float, float); |
| 3848 |
FUNC(__cmpsf2) |
| 3849 |
SYM (__cmpsf2): |
| 3850 |
link a6,IMM (0) |
| 3851 |
pea 1 |
| 3852 |
movl a6@(12),sp@- |
| 3853 |
movl a6@(8),sp@- |
| 3854 |
PICCALL SYM (__cmpsf2_internal) |
| 3855 |
unlk a6 |
| 3856 |
rts |
| 3857 |
|
| 3858 |
|============================================================================= |
| 3859 |
| rounding routines |
| 3860 |
|============================================================================= |
| 3861 |
|
| 3862 |
| The rounding routines expect the number to be normalized in registers |
| 3863 |
| d0-d1, with the exponent in register d2. They assume that the |
| 3864 |
| exponent is larger or equal to 1. They return a properly normalized number |
| 3865 |
| if possible, and a denormalized number otherwise. The exponent is returned |
| 3866 |
| in d2. |
| 3867 |
|
| 3868 |
Lround$to$nearest: |
| 3869 |
| We now normalize as suggested by D. Knuth ("Seminumerical Algorithms"): |
| 3870 |
| Here we assume that the exponent is not too small (this should be checked |
| 3871 |
| before entering the rounding routine), but the number could be denormalized. |
| 3872 |
|
| 3873 |
| Check for denormalized numbers: |
| 3874 |
1: btst IMM (FLT_MANT_DIG),d0 |
| 3875 |
bne 2f | if set the number is normalized |
| 3876 |
| Normalize shifting left until bit #FLT_MANT_DIG is set or the exponent |
| 3877 |
| is one (remember that a denormalized number corresponds to an |
| 3878 |
| exponent of -F_BIAS+1). |
| 3879 |
#ifndef __mcoldfire__ |
| 3880 |
cmpw IMM (1),d2 | remember that the exponent is at least one |
| 3881 |
#else |
| 3882 |
cmpl IMM (1),d2 | remember that the exponent is at least one |
| 3883 |
#endif |
| 3884 |
beq 2f | an exponent of one means denormalized |
| 3885 |
addl d1,d1 | else shift and adjust the exponent |
| 3886 |
addxl d0,d0 | |
| 3887 |
#ifndef __mcoldfire__ |
| 3888 |
dbra d2,1b | |
| 3889 |
#else |
| 3890 |
subql IMM (1),d2 |
| 3891 |
bpl 1b |
| 3892 |
#endif |
| 3893 |
2: |
| 3894 |
| Now round: we do it as follows: after the shifting we can write the |
| 3895 |
| fraction part as f + delta, where 1 < f < 2^25, and 0 <= delta <= 2. |
| 3896 |
| If delta < 1, do nothing. If delta > 1, add 1 to f. |
| 3897 |
| If delta == 1, we make sure the rounded number will be even (odd?) |
| 3898 |
| (after shifting). |
| 3899 |
btst IMM (0),d0 | is delta < 1? |
| 3900 |
beq 2f | if so, do not do anything |
| 3901 |
tstl d1 | is delta == 1? |
| 3902 |
bne 1f | if so round to even |
| 3903 |
movel d0,d1 | |
| 3904 |
andl IMM (2),d1 | bit 1 is the last significant bit |
| 3905 |
addl d1,d0 | |
| 3906 |
bra 2f | |
| 3907 |
1: movel IMM (1),d1 | else add 1 |
| 3908 |
addl d1,d0 | |
| 3909 |
| Shift right once (because we used bit #FLT_MANT_DIG!). |
| 3910 |
2: lsrl IMM (1),d0 |
| 3911 |
| Now check again bit #FLT_MANT_DIG (rounding could have produced a |
| 3912 |
| 'fraction overflow' ...). |
| 3913 |
btst IMM (FLT_MANT_DIG),d0 |
| 3914 |
beq 1f |
| 3915 |
lsrl IMM (1),d0 |
| 3916 |
#ifndef __mcoldfire__ |
| 3917 |
addw IMM (1),d2 |
| 3918 |
#else |
| 3919 |
addql IMM (1),d2 |
| 3920 |
#endif |
| 3921 |
1: |
| 3922 |
| If bit #FLT_MANT_DIG-1 is clear we have a denormalized number, so we |
| 3923 |
| have to put the exponent to zero and return a denormalized number. |
| 3924 |
btst IMM (FLT_MANT_DIG-1),d0 |
| 3925 |
beq 1f |
| 3926 |
jmp a0@ |
| 3927 |
1: movel IMM (0),d2 |
| 3928 |
jmp a0@ |
| 3929 |
|
| 3930 |
Lround$to$zero: |
| 3931 |
Lround$to$plus: |
| 3932 |
Lround$to$minus: |
| 3933 |
jmp a0@ |
| 3934 |
#endif /* L_float */ |
| 3935 |
|
| 3936 |
| gcc expects the routines __eqdf2, __nedf2, __gtdf2, __gedf2, |
| 3937 |
| __ledf2, __ltdf2 to all return the same value as a direct call to |
| 3938 |
| __cmpdf2 would. In this implementation, each of these routines |
| 3939 |
| simply calls __cmpdf2. It would be more efficient to give the |
| 3940 |
| __cmpdf2 routine several names, but separating them out will make it |
| 3941 |
| easier to write efficient versions of these routines someday. |
| 3942 |
| If the operands recompare unordered unordered __gtdf2 and __gedf2 return -1. |
| 3943 |
| The other routines return 1. |
| 3944 |
|
| 3945 |
#ifdef L_eqdf2 |
| 3946 |
.text |
| 3947 |
FUNC(__eqdf2) |
| 3948 |
.globl SYM (__eqdf2) |
| 3949 |
SYM (__eqdf2): |
| 3950 |
link a6,IMM (0) |
| 3951 |
pea 1 |
| 3952 |
movl a6@(20),sp@- |
| 3953 |
movl a6@(16),sp@- |
| 3954 |
movl a6@(12),sp@- |
| 3955 |
movl a6@(8),sp@- |
| 3956 |
PICCALL SYM (__cmpdf2_internal) |
| 3957 |
unlk a6 |
| 3958 |
rts |
| 3959 |
#endif /* L_eqdf2 */ |
| 3960 |
|
| 3961 |
#ifdef L_nedf2 |
| 3962 |
.text |
| 3963 |
FUNC(__nedf2) |
| 3964 |
.globl SYM (__nedf2) |
| 3965 |
SYM (__nedf2): |
| 3966 |
link a6,IMM (0) |
| 3967 |
pea 1 |
| 3968 |
movl a6@(20),sp@- |
| 3969 |
movl a6@(16),sp@- |
| 3970 |
movl a6@(12),sp@- |
| 3971 |
movl a6@(8),sp@- |
| 3972 |
PICCALL SYM (__cmpdf2_internal) |
| 3973 |
unlk a6 |
| 3974 |
rts |
| 3975 |
#endif /* L_nedf2 */ |
| 3976 |
|
| 3977 |
#ifdef L_gtdf2 |
| 3978 |
.text |
| 3979 |
FUNC(__gtdf2) |
| 3980 |
.globl SYM (__gtdf2) |
| 3981 |
SYM (__gtdf2): |
| 3982 |
link a6,IMM (0) |
| 3983 |
pea -1 |
| 3984 |
movl a6@(20),sp@- |
| 3985 |
movl a6@(16),sp@- |
| 3986 |
movl a6@(12),sp@- |
| 3987 |
movl a6@(8),sp@- |
| 3988 |
PICCALL SYM (__cmpdf2_internal) |
| 3989 |
unlk a6 |
| 3990 |
rts |
| 3991 |
#endif /* L_gtdf2 */ |
| 3992 |
|
| 3993 |
#ifdef L_gedf2 |
| 3994 |
.text |
| 3995 |
FUNC(__gedf2) |
| 3996 |
.globl SYM (__gedf2) |
| 3997 |
SYM (__gedf2): |
| 3998 |
link a6,IMM (0) |
| 3999 |
pea -1 |
| 4000 |
movl a6@(20),sp@- |
| 4001 |
movl a6@(16),sp@- |
| 4002 |
movl a6@(12),sp@- |
| 4003 |
movl a6@(8),sp@- |
| 4004 |
PICCALL SYM (__cmpdf2_internal) |
| 4005 |
unlk a6 |
| 4006 |
rts |
| 4007 |
#endif /* L_gedf2 */ |
| 4008 |
|
| 4009 |
#ifdef L_ltdf2 |
| 4010 |
.text |
| 4011 |
FUNC(__ltdf2) |
| 4012 |
.globl SYM (__ltdf2) |
| 4013 |
SYM (__ltdf2): |
| 4014 |
link a6,IMM (0) |
| 4015 |
pea 1 |
| 4016 |
movl a6@(20),sp@- |
| 4017 |
movl a6@(16),sp@- |
| 4018 |
movl a6@(12),sp@- |
| 4019 |
movl a6@(8),sp@- |
| 4020 |
PICCALL SYM (__cmpdf2_internal) |
| 4021 |
unlk a6 |
| 4022 |
rts |
| 4023 |
#endif /* L_ltdf2 */ |
| 4024 |
|
| 4025 |
#ifdef L_ledf2 |
| 4026 |
.text |
| 4027 |
FUNC(__ledf2) |
| 4028 |
.globl SYM (__ledf2) |
| 4029 |
SYM (__ledf2): |
| 4030 |
link a6,IMM (0) |
| 4031 |
pea 1 |
| 4032 |
movl a6@(20),sp@- |
| 4033 |
movl a6@(16),sp@- |
| 4034 |
movl a6@(12),sp@- |
| 4035 |
movl a6@(8),sp@- |
| 4036 |
PICCALL SYM (__cmpdf2_internal) |
| 4037 |
unlk a6 |
| 4038 |
rts |
| 4039 |
#endif /* L_ledf2 */ |
| 4040 |
|
| 4041 |
| The comments above about __eqdf2, et. al., also apply to __eqsf2, |
| 4042 |
| et. al., except that the latter call __cmpsf2 rather than __cmpdf2. |
| 4043 |
|
| 4044 |
#ifdef L_eqsf2 |
| 4045 |
.text |
| 4046 |
FUNC(__eqsf2) |
| 4047 |
.globl SYM (__eqsf2) |
| 4048 |
SYM (__eqsf2): |
| 4049 |
link a6,IMM (0) |
| 4050 |
pea 1 |
| 4051 |
movl a6@(12),sp@- |
| 4052 |
movl a6@(8),sp@- |
| 4053 |
PICCALL SYM (__cmpsf2_internal) |
| 4054 |
unlk a6 |
| 4055 |
rts |
| 4056 |
#endif /* L_eqsf2 */ |
| 4057 |
|
| 4058 |
#ifdef L_nesf2 |
| 4059 |
.text |
| 4060 |
FUNC(__nesf2) |
| 4061 |
.globl SYM (__nesf2) |
| 4062 |
SYM (__nesf2): |
| 4063 |
link a6,IMM (0) |
| 4064 |
pea 1 |
| 4065 |
movl a6@(12),sp@- |
| 4066 |
movl a6@(8),sp@- |
| 4067 |
PICCALL SYM (__cmpsf2_internal) |
| 4068 |
unlk a6 |
| 4069 |
rts |
| 4070 |
#endif /* L_nesf2 */ |
| 4071 |
|
| 4072 |
#ifdef L_gtsf2 |
| 4073 |
.text |
| 4074 |
FUNC(__gtsf2) |
| 4075 |
.globl SYM (__gtsf2) |
| 4076 |
SYM (__gtsf2): |
| 4077 |
link a6,IMM (0) |
| 4078 |
pea -1 |
| 4079 |
movl a6@(12),sp@- |
| 4080 |
movl a6@(8),sp@- |
| 4081 |
PICCALL SYM (__cmpsf2_internal) |
| 4082 |
unlk a6 |
| 4083 |
rts |
| 4084 |
#endif /* L_gtsf2 */ |
| 4085 |
|
| 4086 |
#ifdef L_gesf2 |
| 4087 |
.text |
| 4088 |
FUNC(__gesf2) |
| 4089 |
.globl SYM (__gesf2) |
| 4090 |
SYM (__gesf2): |
| 4091 |
link a6,IMM (0) |
| 4092 |
pea -1 |
| 4093 |
movl a6@(12),sp@- |
| 4094 |
movl a6@(8),sp@- |
| 4095 |
PICCALL SYM (__cmpsf2_internal) |
| 4096 |
unlk a6 |
| 4097 |
rts |
| 4098 |
#endif /* L_gesf2 */ |
| 4099 |
|
| 4100 |
#ifdef L_ltsf2 |
| 4101 |
.text |
| 4102 |
FUNC(__ltsf2) |
| 4103 |
.globl SYM (__ltsf2) |
| 4104 |
SYM (__ltsf2): |
| 4105 |
link a6,IMM (0) |
| 4106 |
pea 1 |
| 4107 |
movl a6@(12),sp@- |
| 4108 |
movl a6@(8),sp@- |
| 4109 |
PICCALL SYM (__cmpsf2_internal) |
| 4110 |
unlk a6 |
| 4111 |
rts |
| 4112 |
#endif /* L_ltsf2 */ |
| 4113 |
|
| 4114 |
#ifdef L_lesf2 |
| 4115 |
.text |
| 4116 |
FUNC(__lesf2) |
| 4117 |
.globl SYM (__lesf2) |
| 4118 |
SYM (__lesf2): |
| 4119 |
link a6,IMM (0) |
| 4120 |
pea 1 |
| 4121 |
movl a6@(12),sp@- |
| 4122 |
movl a6@(8),sp@- |
| 4123 |
PICCALL SYM (__cmpsf2_internal) |
| 4124 |
unlk a6 |
| 4125 |
rts |
| 4126 |
#endif /* L_lesf2 */ |
| 4127 |
|
| 4128 |
#if defined (__ELF__) && defined (__linux__) |
| 4129 |
/* Make stack non-executable for ELF linux targets. */ |
| 4130 |
.section .note.GNU-stack,"",@progbits |
| 4131 |
#endif |