Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * fp_util.S |
| 3 | * |
| 4 | * Copyright Roman Zippel, 1997. All rights reserved. |
| 5 | * |
| 6 | * Redistribution and use in source and binary forms, with or without |
| 7 | * modification, are permitted provided that the following conditions |
| 8 | * are met: |
| 9 | * 1. Redistributions of source code must retain the above copyright |
| 10 | * notice, and the entire permission notice in its entirety, |
| 11 | * including the disclaimer of warranties. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer in the |
| 14 | * documentation and/or other materials provided with the distribution. |
| 15 | * 3. The name of the author may not be used to endorse or promote |
| 16 | * products derived from this software without specific prior |
| 17 | * written permission. |
| 18 | * |
| 19 | * ALTERNATIVELY, this product may be distributed under the terms of |
| 20 | * the GNU General Public License, in which case the provisions of the GPL are |
| 21 | * required INSTEAD OF the above restrictions. (This clause is |
| 22 | * necessary due to a potential bad interaction between the GPL and |
| 23 | * the restrictions contained in a BSD-style copyright.) |
| 24 | * |
| 25 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 26 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
| 27 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 28 | * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, |
| 29 | * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
| 30 | * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 31 | * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 32 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, |
| 33 | * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 34 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
| 35 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
| 36 | */ |
| 37 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include "fp_emu.h" |
| 39 | |
| 40 | /* |
| 41 | * Here are lots of conversion and normalization functions mainly |
| 42 | * used by fp_scan.S |
| 43 | * Note that these functions are optimized for "normal" numbers, |
| 44 | * these are handled first and exit as fast as possible, this is |
| 45 | * especially important for fp_normalize_ext/fp_conv_ext2ext, as |
| 46 | * it's called very often. |
| 47 | * The register usage is optimized for fp_scan.S and which register |
| 48 | * is currently at that time unused, be careful if you want change |
| 49 | * something here. %d0 and %d1 is always usable, sometimes %d2 (or |
| 50 | * only the lower half) most function have to return the %a0 |
| 51 | * unmodified, so that the caller can immediately reuse it. |
| 52 | */ |
| 53 | |
| 54 | .globl fp_ill, fp_end |
| 55 | |
| 56 | | exits from fp_scan: |
| 57 | | illegal instruction |
| 58 | fp_ill: |
| 59 | printf ,"fp_illegal\n" |
| 60 | rts |
| 61 | | completed instruction |
| 62 | fp_end: |
| 63 | tst.l (TASK_MM-8,%a2) |
| 64 | jmi 1f |
| 65 | tst.l (TASK_MM-4,%a2) |
| 66 | jmi 1f |
| 67 | tst.l (TASK_MM,%a2) |
| 68 | jpl 2f |
| 69 | 1: printf ,"oops:%p,%p,%p\n",3,%a2@(TASK_MM-8),%a2@(TASK_MM-4),%a2@(TASK_MM) |
| 70 | 2: clr.l %d0 |
| 71 | rts |
| 72 | |
| 73 | .globl fp_conv_long2ext, fp_conv_single2ext |
| 74 | .globl fp_conv_double2ext, fp_conv_ext2ext |
| 75 | .globl fp_normalize_ext, fp_normalize_double |
| 76 | .globl fp_normalize_single, fp_normalize_single_fast |
| 77 | .globl fp_conv_ext2double, fp_conv_ext2single |
| 78 | .globl fp_conv_ext2long, fp_conv_ext2short |
| 79 | .globl fp_conv_ext2byte |
| 80 | .globl fp_finalrounding_single, fp_finalrounding_single_fast |
| 81 | .globl fp_finalrounding_double |
| 82 | .globl fp_finalrounding, fp_finaltest, fp_final |
| 83 | |
| 84 | /* |
| 85 | * First several conversion functions from a source operand |
| 86 | * into the extended format. Note, that only fp_conv_ext2ext |
| 87 | * normalizes the number and is always called after the other |
| 88 | * conversion functions, which only move the information into |
| 89 | * fp_ext structure. |
| 90 | */ |
| 91 | |
| 92 | | fp_conv_long2ext: |
| 93 | | |
| 94 | | args: %d0 = source (32-bit long) |
| 95 | | %a0 = destination (ptr to struct fp_ext) |
| 96 | |
| 97 | fp_conv_long2ext: |
| 98 | printf PCONV,"l2e: %p -> %p(",2,%d0,%a0 |
| 99 | clr.l %d1 | sign defaults to zero |
| 100 | tst.l %d0 |
| 101 | jeq fp_l2e_zero | is source zero? |
| 102 | jpl 1f | positive? |
| 103 | moveq #1,%d1 |
| 104 | neg.l %d0 |
| 105 | 1: swap %d1 |
| 106 | move.w #0x3fff+31,%d1 |
| 107 | move.l %d1,(%a0)+ | set sign / exp |
| 108 | move.l %d0,(%a0)+ | set mantissa |
| 109 | clr.l (%a0) |
| 110 | subq.l #8,%a0 | restore %a0 |
| 111 | printx PCONV,%a0@ |
| 112 | printf PCONV,")\n" |
| 113 | rts |
| 114 | | source is zero |
| 115 | fp_l2e_zero: |
| 116 | clr.l (%a0)+ |
| 117 | clr.l (%a0)+ |
| 118 | clr.l (%a0) |
| 119 | subq.l #8,%a0 |
| 120 | printx PCONV,%a0@ |
| 121 | printf PCONV,")\n" |
| 122 | rts |
| 123 | |
| 124 | | fp_conv_single2ext |
| 125 | | args: %d0 = source (single-precision fp value) |
| 126 | | %a0 = dest (struct fp_ext *) |
| 127 | |
| 128 | fp_conv_single2ext: |
| 129 | printf PCONV,"s2e: %p -> %p(",2,%d0,%a0 |
| 130 | move.l %d0,%d1 |
| 131 | lsl.l #8,%d0 | shift mantissa |
| 132 | lsr.l #8,%d1 | exponent / sign |
| 133 | lsr.l #7,%d1 |
| 134 | lsr.w #8,%d1 |
| 135 | jeq fp_s2e_small | zero / denormal? |
| 136 | cmp.w #0xff,%d1 | NaN / Inf? |
| 137 | jeq fp_s2e_large |
| 138 | bset #31,%d0 | set explizit bit |
| 139 | add.w #0x3fff-0x7f,%d1 | re-bias the exponent. |
| 140 | 9: move.l %d1,(%a0)+ | fp_ext.sign, fp_ext.exp |
| 141 | move.l %d0,(%a0)+ | high lword of fp_ext.mant |
| 142 | clr.l (%a0) | low lword = 0 |
| 143 | subq.l #8,%a0 |
| 144 | printx PCONV,%a0@ |
| 145 | printf PCONV,")\n" |
| 146 | rts |
| 147 | | zeros and denormalized |
| 148 | fp_s2e_small: |
| 149 | | exponent is zero, so explizit bit is already zero too |
| 150 | tst.l %d0 |
| 151 | jeq 9b |
| 152 | move.w #0x4000-0x7f,%d1 |
| 153 | jra 9b |
| 154 | | infinities and NAN |
| 155 | fp_s2e_large: |
| 156 | bclr #31,%d0 | clear explizit bit |
| 157 | move.w #0x7fff,%d1 |
| 158 | jra 9b |
| 159 | |
| 160 | fp_conv_double2ext: |
| 161 | #ifdef FPU_EMU_DEBUG |
| 162 | getuser.l %a1@(0),%d0,fp_err_ua2,%a1 |
| 163 | getuser.l %a1@(4),%d1,fp_err_ua2,%a1 |
| 164 | printf PCONV,"d2e: %p%p -> %p(",3,%d0,%d1,%a0 |
| 165 | #endif |
| 166 | getuser.l (%a1)+,%d0,fp_err_ua2,%a1 |
| 167 | move.l %d0,%d1 |
| 168 | lsl.l #8,%d0 | shift high mantissa |
| 169 | lsl.l #3,%d0 |
| 170 | lsr.l #8,%d1 | exponent / sign |
| 171 | lsr.l #7,%d1 |
| 172 | lsr.w #5,%d1 |
| 173 | jeq fp_d2e_small | zero / denormal? |
| 174 | cmp.w #0x7ff,%d1 | NaN / Inf? |
| 175 | jeq fp_d2e_large |
| 176 | bset #31,%d0 | set explizit bit |
| 177 | add.w #0x3fff-0x3ff,%d1 | re-bias the exponent. |
| 178 | 9: move.l %d1,(%a0)+ | fp_ext.sign, fp_ext.exp |
| 179 | move.l %d0,(%a0)+ |
| 180 | getuser.l (%a1)+,%d0,fp_err_ua2,%a1 |
| 181 | move.l %d0,%d1 |
| 182 | lsl.l #8,%d0 |
| 183 | lsl.l #3,%d0 |
| 184 | move.l %d0,(%a0) |
| 185 | moveq #21,%d0 |
| 186 | lsr.l %d0,%d1 |
| 187 | or.l %d1,-(%a0) |
| 188 | subq.l #4,%a0 |
| 189 | printx PCONV,%a0@ |
| 190 | printf PCONV,")\n" |
| 191 | rts |
| 192 | | zeros and denormalized |
| 193 | fp_d2e_small: |
| 194 | | exponent is zero, so explizit bit is already zero too |
| 195 | tst.l %d0 |
| 196 | jeq 9b |
| 197 | move.w #0x4000-0x3ff,%d1 |
| 198 | jra 9b |
| 199 | | infinities and NAN |
| 200 | fp_d2e_large: |
| 201 | bclr #31,%d0 | clear explizit bit |
| 202 | move.w #0x7fff,%d1 |
| 203 | jra 9b |
| 204 | |
| 205 | | fp_conv_ext2ext: |
| 206 | | originally used to get longdouble from userspace, now it's |
| 207 | | called before arithmetic operations to make sure the number |
| 208 | | is normalized [maybe rename it?]. |
| 209 | | args: %a0 = dest (struct fp_ext *) |
| 210 | | returns 0 in %d0 for a NaN, otherwise 1 |
| 211 | |
| 212 | fp_conv_ext2ext: |
| 213 | printf PCONV,"e2e: %p(",1,%a0 |
| 214 | printx PCONV,%a0@ |
| 215 | printf PCONV,"), " |
| 216 | move.l (%a0)+,%d0 |
| 217 | cmp.w #0x7fff,%d0 | Inf / NaN? |
| 218 | jeq fp_e2e_large |
| 219 | move.l (%a0),%d0 |
| 220 | jpl fp_e2e_small | zero / denorm? |
| 221 | | The high bit is set, so normalization is irrelevant. |
| 222 | fp_e2e_checkround: |
| 223 | subq.l #4,%a0 |
| 224 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 225 | move.b (%a0),%d0 |
| 226 | jne fp_e2e_round |
| 227 | #endif |
| 228 | printf PCONV,"%p(",1,%a0 |
| 229 | printx PCONV,%a0@ |
| 230 | printf PCONV,")\n" |
| 231 | moveq #1,%d0 |
| 232 | rts |
| 233 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 234 | fp_e2e_round: |
| 235 | fp_set_sr FPSR_EXC_INEX2 |
| 236 | clr.b (%a0) |
| 237 | move.w (FPD_RND,FPDATA),%d2 |
| 238 | jne fp_e2e_roundother | %d2 == 0, round to nearest |
| 239 | tst.b %d0 | test guard bit |
| 240 | jpl 9f | zero is closer |
| 241 | btst #0,(11,%a0) | test lsb bit |
| 242 | jne fp_e2e_doroundup | round to infinity |
| 243 | lsl.b #1,%d0 | check low bits |
| 244 | jeq 9f | round to zero |
| 245 | fp_e2e_doroundup: |
| 246 | addq.l #1,(8,%a0) |
| 247 | jcc 9f |
| 248 | addq.l #1,(4,%a0) |
| 249 | jcc 9f |
| 250 | move.w #0x8000,(4,%a0) |
| 251 | addq.w #1,(2,%a0) |
| 252 | 9: printf PNORM,"%p(",1,%a0 |
| 253 | printx PNORM,%a0@ |
| 254 | printf PNORM,")\n" |
| 255 | rts |
| 256 | fp_e2e_roundother: |
| 257 | subq.w #2,%d2 |
| 258 | jcs 9b | %d2 < 2, round to zero |
| 259 | jhi 1f | %d2 > 2, round to +infinity |
| 260 | tst.b (1,%a0) | to -inf |
| 261 | jne fp_e2e_doroundup | negative, round to infinity |
| 262 | jra 9b | positive, round to zero |
| 263 | 1: tst.b (1,%a0) | to +inf |
| 264 | jeq fp_e2e_doroundup | positive, round to infinity |
| 265 | jra 9b | negative, round to zero |
| 266 | #endif |
| 267 | | zeros and subnormals: |
| 268 | | try to normalize these anyway. |
| 269 | fp_e2e_small: |
| 270 | jne fp_e2e_small1 | high lword zero? |
| 271 | move.l (4,%a0),%d0 |
| 272 | jne fp_e2e_small2 |
| 273 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 274 | clr.l %d0 |
| 275 | move.b (-4,%a0),%d0 |
| 276 | jne fp_e2e_small3 |
| 277 | #endif |
| 278 | | Genuine zero. |
| 279 | clr.w -(%a0) |
| 280 | subq.l #2,%a0 |
| 281 | printf PNORM,"%p(",1,%a0 |
| 282 | printx PNORM,%a0@ |
| 283 | printf PNORM,")\n" |
| 284 | moveq #1,%d0 |
| 285 | rts |
| 286 | | definitely subnormal, need to shift all 64 bits |
| 287 | fp_e2e_small1: |
| 288 | bfffo %d0{#0,#32},%d1 |
| 289 | move.w -(%a0),%d2 |
| 290 | sub.w %d1,%d2 |
| 291 | jcc 1f |
| 292 | | Pathologically small, denormalize. |
| 293 | add.w %d2,%d1 |
| 294 | clr.w %d2 |
| 295 | 1: move.w %d2,(%a0)+ |
| 296 | move.w %d1,%d2 |
| 297 | jeq fp_e2e_checkround |
| 298 | | fancy 64-bit double-shift begins here |
| 299 | lsl.l %d2,%d0 |
| 300 | move.l %d0,(%a0)+ |
| 301 | move.l (%a0),%d0 |
| 302 | move.l %d0,%d1 |
| 303 | lsl.l %d2,%d0 |
| 304 | move.l %d0,(%a0) |
| 305 | neg.w %d2 |
| 306 | and.w #0x1f,%d2 |
| 307 | lsr.l %d2,%d1 |
| 308 | or.l %d1,-(%a0) |
| 309 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 310 | fp_e2e_extra1: |
| 311 | clr.l %d0 |
| 312 | move.b (-4,%a0),%d0 |
| 313 | neg.w %d2 |
| 314 | add.w #24,%d2 |
| 315 | jcc 1f |
| 316 | clr.b (-4,%a0) |
| 317 | lsl.l %d2,%d0 |
| 318 | or.l %d0,(4,%a0) |
| 319 | jra fp_e2e_checkround |
| 320 | 1: addq.w #8,%d2 |
| 321 | lsl.l %d2,%d0 |
| 322 | move.b %d0,(-4,%a0) |
| 323 | lsr.l #8,%d0 |
| 324 | or.l %d0,(4,%a0) |
| 325 | #endif |
| 326 | jra fp_e2e_checkround |
| 327 | | pathologically small subnormal |
| 328 | fp_e2e_small2: |
| 329 | bfffo %d0{#0,#32},%d1 |
| 330 | add.w #32,%d1 |
| 331 | move.w -(%a0),%d2 |
| 332 | sub.w %d1,%d2 |
| 333 | jcc 1f |
| 334 | | Beyond pathologically small, denormalize. |
| 335 | add.w %d2,%d1 |
| 336 | clr.w %d2 |
| 337 | 1: move.w %d2,(%a0)+ |
| 338 | ext.l %d1 |
| 339 | jeq fp_e2e_checkround |
| 340 | clr.l (4,%a0) |
| 341 | sub.w #32,%d2 |
| 342 | jcs 1f |
| 343 | lsl.l %d1,%d0 | lower lword needs only to be shifted |
| 344 | move.l %d0,(%a0) | into the higher lword |
| 345 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 346 | clr.l %d0 |
| 347 | move.b (-4,%a0),%d0 |
| 348 | clr.b (-4,%a0) |
| 349 | neg.w %d1 |
| 350 | add.w #32,%d1 |
| 351 | bfins %d0,(%a0){%d1,#8} |
| 352 | #endif |
| 353 | jra fp_e2e_checkround |
| 354 | 1: neg.w %d1 | lower lword is splitted between |
| 355 | bfins %d0,(%a0){%d1,#32} | higher and lower lword |
| 356 | #ifndef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 357 | jra fp_e2e_checkround |
| 358 | #else |
| 359 | move.w %d1,%d2 |
| 360 | jra fp_e2e_extra1 |
| 361 | | These are extremely small numbers, that will mostly end up as zero |
| 362 | | anyway, so this is only important for correct rounding. |
| 363 | fp_e2e_small3: |
| 364 | bfffo %d0{#24,#8},%d1 |
| 365 | add.w #40,%d1 |
| 366 | move.w -(%a0),%d2 |
| 367 | sub.w %d1,%d2 |
| 368 | jcc 1f |
| 369 | | Pathologically small, denormalize. |
| 370 | add.w %d2,%d1 |
| 371 | clr.w %d2 |
| 372 | 1: move.w %d2,(%a0)+ |
| 373 | ext.l %d1 |
| 374 | jeq fp_e2e_checkround |
| 375 | cmp.w #8,%d1 |
| 376 | jcs 2f |
| 377 | 1: clr.b (-4,%a0) |
| 378 | sub.w #64,%d1 |
| 379 | jcs 1f |
| 380 | add.w #24,%d1 |
| 381 | lsl.l %d1,%d0 |
| 382 | move.l %d0,(%a0) |
| 383 | jra fp_e2e_checkround |
| 384 | 1: neg.w %d1 |
| 385 | bfins %d0,(%a0){%d1,#8} |
| 386 | jra fp_e2e_checkround |
| 387 | 2: lsl.l %d1,%d0 |
| 388 | move.b %d0,(-4,%a0) |
| 389 | lsr.l #8,%d0 |
| 390 | move.b %d0,(7,%a0) |
| 391 | jra fp_e2e_checkround |
| 392 | #endif |
| 393 | 1: move.l %d0,%d1 | lower lword is splitted between |
| 394 | lsl.l %d2,%d0 | higher and lower lword |
| 395 | move.l %d0,(%a0) |
| 396 | move.l %d1,%d0 |
| 397 | neg.w %d2 |
| 398 | add.w #32,%d2 |
| 399 | lsr.l %d2,%d0 |
| 400 | move.l %d0,-(%a0) |
| 401 | jra fp_e2e_checkround |
| 402 | | Infinities and NaNs |
| 403 | fp_e2e_large: |
| 404 | move.l (%a0)+,%d0 |
| 405 | jne 3f |
| 406 | 1: tst.l (%a0) |
| 407 | jne 4f |
| 408 | moveq #1,%d0 |
| 409 | 2: subq.l #8,%a0 |
| 410 | printf PCONV,"%p(",1,%a0 |
| 411 | printx PCONV,%a0@ |
| 412 | printf PCONV,")\n" |
| 413 | rts |
| 414 | | we have maybe a NaN, shift off the highest bit |
| 415 | 3: lsl.l #1,%d0 |
| 416 | jeq 1b |
| 417 | | we have a NaN, clear the return value |
| 418 | 4: clrl %d0 |
| 419 | jra 2b |
| 420 | |
| 421 | |
| 422 | /* |
| 423 | * Normalization functions. Call these on the output of general |
| 424 | * FP operators, and before any conversion into the destination |
| 425 | * formats. fp_normalize_ext has always to be called first, the |
| 426 | * following conversion functions expect an already normalized |
| 427 | * number. |
| 428 | */ |
| 429 | |
| 430 | | fp_normalize_ext: |
| 431 | | normalize an extended in extended (unpacked) format, basically |
| 432 | | it does the same as fp_conv_ext2ext, additionally it also does |
| 433 | | the necessary postprocessing checks. |
| 434 | | args: %a0 (struct fp_ext *) |
| 435 | | NOTE: it does _not_ modify %a0/%a1 and the upper word of %d2 |
| 436 | |
| 437 | fp_normalize_ext: |
| 438 | printf PNORM,"ne: %p(",1,%a0 |
| 439 | printx PNORM,%a0@ |
| 440 | printf PNORM,"), " |
| 441 | move.l (%a0)+,%d0 |
| 442 | cmp.w #0x7fff,%d0 | Inf / NaN? |
| 443 | jeq fp_ne_large |
| 444 | move.l (%a0),%d0 |
| 445 | jpl fp_ne_small | zero / denorm? |
| 446 | | The high bit is set, so normalization is irrelevant. |
| 447 | fp_ne_checkround: |
| 448 | subq.l #4,%a0 |
| 449 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 450 | move.b (%a0),%d0 |
| 451 | jne fp_ne_round |
| 452 | #endif |
| 453 | printf PNORM,"%p(",1,%a0 |
| 454 | printx PNORM,%a0@ |
| 455 | printf PNORM,")\n" |
| 456 | rts |
| 457 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 458 | fp_ne_round: |
| 459 | fp_set_sr FPSR_EXC_INEX2 |
| 460 | clr.b (%a0) |
| 461 | move.w (FPD_RND,FPDATA),%d2 |
| 462 | jne fp_ne_roundother | %d2 == 0, round to nearest |
| 463 | tst.b %d0 | test guard bit |
| 464 | jpl 9f | zero is closer |
| 465 | btst #0,(11,%a0) | test lsb bit |
| 466 | jne fp_ne_doroundup | round to infinity |
| 467 | lsl.b #1,%d0 | check low bits |
| 468 | jeq 9f | round to zero |
| 469 | fp_ne_doroundup: |
| 470 | addq.l #1,(8,%a0) |
| 471 | jcc 9f |
| 472 | addq.l #1,(4,%a0) |
| 473 | jcc 9f |
| 474 | addq.w #1,(2,%a0) |
| 475 | move.w #0x8000,(4,%a0) |
| 476 | 9: printf PNORM,"%p(",1,%a0 |
| 477 | printx PNORM,%a0@ |
| 478 | printf PNORM,")\n" |
| 479 | rts |
| 480 | fp_ne_roundother: |
| 481 | subq.w #2,%d2 |
| 482 | jcs 9b | %d2 < 2, round to zero |
| 483 | jhi 1f | %d2 > 2, round to +infinity |
| 484 | tst.b (1,%a0) | to -inf |
| 485 | jne fp_ne_doroundup | negative, round to infinity |
| 486 | jra 9b | positive, round to zero |
| 487 | 1: tst.b (1,%a0) | to +inf |
| 488 | jeq fp_ne_doroundup | positive, round to infinity |
| 489 | jra 9b | negative, round to zero |
| 490 | #endif |
| 491 | | Zeros and subnormal numbers |
| 492 | | These are probably merely subnormal, rather than "denormalized" |
| 493 | | numbers, so we will try to make them normal again. |
| 494 | fp_ne_small: |
| 495 | jne fp_ne_small1 | high lword zero? |
| 496 | move.l (4,%a0),%d0 |
| 497 | jne fp_ne_small2 |
| 498 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 499 | clr.l %d0 |
| 500 | move.b (-4,%a0),%d0 |
| 501 | jne fp_ne_small3 |
| 502 | #endif |
| 503 | | Genuine zero. |
| 504 | clr.w -(%a0) |
| 505 | subq.l #2,%a0 |
| 506 | printf PNORM,"%p(",1,%a0 |
| 507 | printx PNORM,%a0@ |
| 508 | printf PNORM,")\n" |
| 509 | rts |
| 510 | | Subnormal. |
| 511 | fp_ne_small1: |
| 512 | bfffo %d0{#0,#32},%d1 |
| 513 | move.w -(%a0),%d2 |
| 514 | sub.w %d1,%d2 |
| 515 | jcc 1f |
| 516 | | Pathologically small, denormalize. |
| 517 | add.w %d2,%d1 |
| 518 | clr.w %d2 |
| 519 | fp_set_sr FPSR_EXC_UNFL |
| 520 | 1: move.w %d2,(%a0)+ |
| 521 | move.w %d1,%d2 |
| 522 | jeq fp_ne_checkround |
| 523 | | This is exactly the same 64-bit double shift as seen above. |
| 524 | lsl.l %d2,%d0 |
| 525 | move.l %d0,(%a0)+ |
| 526 | move.l (%a0),%d0 |
| 527 | move.l %d0,%d1 |
| 528 | lsl.l %d2,%d0 |
| 529 | move.l %d0,(%a0) |
| 530 | neg.w %d2 |
| 531 | and.w #0x1f,%d2 |
| 532 | lsr.l %d2,%d1 |
| 533 | or.l %d1,-(%a0) |
| 534 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 535 | fp_ne_extra1: |
| 536 | clr.l %d0 |
| 537 | move.b (-4,%a0),%d0 |
| 538 | neg.w %d2 |
| 539 | add.w #24,%d2 |
| 540 | jcc 1f |
| 541 | clr.b (-4,%a0) |
| 542 | lsl.l %d2,%d0 |
| 543 | or.l %d0,(4,%a0) |
| 544 | jra fp_ne_checkround |
| 545 | 1: addq.w #8,%d2 |
| 546 | lsl.l %d2,%d0 |
| 547 | move.b %d0,(-4,%a0) |
| 548 | lsr.l #8,%d0 |
| 549 | or.l %d0,(4,%a0) |
| 550 | #endif |
| 551 | jra fp_ne_checkround |
| 552 | | May or may not be subnormal, if so, only 32 bits to shift. |
| 553 | fp_ne_small2: |
| 554 | bfffo %d0{#0,#32},%d1 |
| 555 | add.w #32,%d1 |
| 556 | move.w -(%a0),%d2 |
| 557 | sub.w %d1,%d2 |
| 558 | jcc 1f |
| 559 | | Beyond pathologically small, denormalize. |
| 560 | add.w %d2,%d1 |
| 561 | clr.w %d2 |
| 562 | fp_set_sr FPSR_EXC_UNFL |
| 563 | 1: move.w %d2,(%a0)+ |
| 564 | ext.l %d1 |
| 565 | jeq fp_ne_checkround |
| 566 | clr.l (4,%a0) |
| 567 | sub.w #32,%d1 |
| 568 | jcs 1f |
| 569 | lsl.l %d1,%d0 | lower lword needs only to be shifted |
| 570 | move.l %d0,(%a0) | into the higher lword |
| 571 | #ifdef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 572 | clr.l %d0 |
| 573 | move.b (-4,%a0),%d0 |
| 574 | clr.b (-4,%a0) |
| 575 | neg.w %d1 |
| 576 | add.w #32,%d1 |
| 577 | bfins %d0,(%a0){%d1,#8} |
| 578 | #endif |
| 579 | jra fp_ne_checkround |
| 580 | 1: neg.w %d1 | lower lword is splitted between |
| 581 | bfins %d0,(%a0){%d1,#32} | higher and lower lword |
| 582 | #ifndef CONFIG_M68KFPU_EMU_EXTRAPREC |
| 583 | jra fp_ne_checkround |
| 584 | #else |
| 585 | move.w %d1,%d2 |
| 586 | jra fp_ne_extra1 |
| 587 | | These are extremely small numbers, that will mostly end up as zero |
| 588 | | anyway, so this is only important for correct rounding. |
| 589 | fp_ne_small3: |
| 590 | bfffo %d0{#24,#8},%d1 |
| 591 | add.w #40,%d1 |
| 592 | move.w -(%a0),%d2 |
| 593 | sub.w %d1,%d2 |
| 594 | jcc 1f |
| 595 | | Pathologically small, denormalize. |
| 596 | add.w %d2,%d1 |
| 597 | clr.w %d2 |
| 598 | 1: move.w %d2,(%a0)+ |
| 599 | ext.l %d1 |
| 600 | jeq fp_ne_checkround |
| 601 | cmp.w #8,%d1 |
| 602 | jcs 2f |
| 603 | 1: clr.b (-4,%a0) |
| 604 | sub.w #64,%d1 |
| 605 | jcs 1f |
| 606 | add.w #24,%d1 |
| 607 | lsl.l %d1,%d0 |
| 608 | move.l %d0,(%a0) |
| 609 | jra fp_ne_checkround |
| 610 | 1: neg.w %d1 |
| 611 | bfins %d0,(%a0){%d1,#8} |
| 612 | jra fp_ne_checkround |
| 613 | 2: lsl.l %d1,%d0 |
| 614 | move.b %d0,(-4,%a0) |
| 615 | lsr.l #8,%d0 |
| 616 | move.b %d0,(7,%a0) |
| 617 | jra fp_ne_checkround |
| 618 | #endif |
| 619 | | Infinities and NaNs, again, same as above. |
| 620 | fp_ne_large: |
| 621 | move.l (%a0)+,%d0 |
| 622 | jne 3f |
| 623 | 1: tst.l (%a0) |
| 624 | jne 4f |
| 625 | 2: subq.l #8,%a0 |
| 626 | printf PNORM,"%p(",1,%a0 |
| 627 | printx PNORM,%a0@ |
| 628 | printf PNORM,")\n" |
| 629 | rts |
| 630 | | we have maybe a NaN, shift off the highest bit |
| 631 | 3: move.l %d0,%d1 |
| 632 | lsl.l #1,%d1 |
| 633 | jne 4f |
| 634 | clr.l (-4,%a0) |
| 635 | jra 1b |
| 636 | | we have a NaN, test if it is signaling |
| 637 | 4: bset #30,%d0 |
| 638 | jne 2b |
| 639 | fp_set_sr FPSR_EXC_SNAN |
| 640 | move.l %d0,(-4,%a0) |
| 641 | jra 2b |
| 642 | |
| 643 | | these next two do rounding as per the IEEE standard. |
| 644 | | values for the rounding modes appear to be: |
| 645 | | 0: Round to nearest |
| 646 | | 1: Round to zero |
| 647 | | 2: Round to -Infinity |
| 648 | | 3: Round to +Infinity |
| 649 | | both functions expect that fp_normalize was already |
| 650 | | called (and extended argument is already normalized |
| 651 | | as far as possible), these are used if there is different |
| 652 | | rounding precision is selected and before converting |
| 653 | | into single/double |
| 654 | |
| 655 | | fp_normalize_double: |
| 656 | | normalize an extended with double (52-bit) precision |
| 657 | | args: %a0 (struct fp_ext *) |
| 658 | |
| 659 | fp_normalize_double: |
| 660 | printf PNORM,"nd: %p(",1,%a0 |
| 661 | printx PNORM,%a0@ |
| 662 | printf PNORM,"), " |
| 663 | move.l (%a0)+,%d2 |
| 664 | tst.w %d2 |
| 665 | jeq fp_nd_zero | zero / denormalized |
| 666 | cmp.w #0x7fff,%d2 |
| 667 | jeq fp_nd_huge | NaN / infinitive. |
| 668 | sub.w #0x4000-0x3ff,%d2 | will the exponent fit? |
| 669 | jcs fp_nd_small | too small. |
| 670 | cmp.w #0x7fe,%d2 |
| 671 | jcc fp_nd_large | too big. |
| 672 | addq.l #4,%a0 |
| 673 | move.l (%a0),%d0 | low lword of mantissa |
| 674 | | now, round off the low 11 bits. |
| 675 | fp_nd_round: |
| 676 | moveq #21,%d1 |
| 677 | lsl.l %d1,%d0 | keep 11 low bits. |
| 678 | jne fp_nd_checkround | Are they non-zero? |
| 679 | | nothing to do here |
| 680 | 9: subq.l #8,%a0 |
| 681 | printf PNORM,"%p(",1,%a0 |
| 682 | printx PNORM,%a0@ |
| 683 | printf PNORM,")\n" |
| 684 | rts |
| 685 | | Be careful with the X bit! It contains the lsb |
| 686 | | from the shift above, it is needed for round to nearest. |
| 687 | fp_nd_checkround: |
| 688 | fp_set_sr FPSR_EXC_INEX2 | INEX2 bit |
| 689 | and.w #0xf800,(2,%a0) | clear bits 0-10 |
| 690 | move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 691 | jne 2f | %d2 == 0, round to nearest |
| 692 | tst.l %d0 | test guard bit |
| 693 | jpl 9b | zero is closer |
| 694 | | here we test the X bit by adding it to %d2 |
| 695 | clr.w %d2 | first set z bit, addx only clears it |
| 696 | addx.w %d2,%d2 | test lsb bit |
| 697 | | IEEE754-specified "round to even" behaviour. If the guard |
| 698 | | bit is set, then the number is odd, so rounding works like |
| 699 | | in grade-school arithmetic (i.e. 1.5 rounds to 2.0) |
| 700 | | Otherwise, an equal distance rounds towards zero, so as not |
| 701 | | to produce an odd number. This is strange, but it is what |
| 702 | | the standard says. |
| 703 | jne fp_nd_doroundup | round to infinity |
| 704 | lsl.l #1,%d0 | check low bits |
| 705 | jeq 9b | round to zero |
| 706 | fp_nd_doroundup: |
| 707 | | round (the mantissa, that is) towards infinity |
| 708 | add.l #0x800,(%a0) |
| 709 | jcc 9b | no overflow, good. |
| 710 | addq.l #1,-(%a0) | extend to high lword |
| 711 | jcc 1f | no overflow, good. |
| 712 | | Yow! we have managed to overflow the mantissa. Since this |
| 713 | | only happens when %d1 was 0xfffff800, it is now zero, so |
| 714 | | reset the high bit, and increment the exponent. |
| 715 | move.w #0x8000,(%a0) |
| 716 | addq.w #1,-(%a0) |
| 717 | cmp.w #0x43ff,(%a0)+ | exponent now overflown? |
| 718 | jeq fp_nd_large | yes, so make it infinity. |
| 719 | 1: subq.l #4,%a0 |
| 720 | printf PNORM,"%p(",1,%a0 |
| 721 | printx PNORM,%a0@ |
| 722 | printf PNORM,")\n" |
| 723 | rts |
| 724 | 2: subq.w #2,%d2 |
| 725 | jcs 9b | %d2 < 2, round to zero |
| 726 | jhi 3f | %d2 > 2, round to +infinity |
| 727 | | Round to +Inf or -Inf. High word of %d2 contains the |
| 728 | | sign of the number, by the way. |
| 729 | swap %d2 | to -inf |
| 730 | tst.b %d2 |
| 731 | jne fp_nd_doroundup | negative, round to infinity |
| 732 | jra 9b | positive, round to zero |
| 733 | 3: swap %d2 | to +inf |
| 734 | tst.b %d2 |
| 735 | jeq fp_nd_doroundup | positive, round to infinity |
| 736 | jra 9b | negative, round to zero |
| 737 | | Exponent underflow. Try to make a denormal, and set it to |
| 738 | | the smallest possible fraction if this fails. |
| 739 | fp_nd_small: |
| 740 | fp_set_sr FPSR_EXC_UNFL | set UNFL bit |
| 741 | move.w #0x3c01,(-2,%a0) | 2**-1022 |
| 742 | neg.w %d2 | degree of underflow |
| 743 | cmp.w #32,%d2 | single or double shift? |
| 744 | jcc 1f |
| 745 | | Again, another 64-bit double shift. |
| 746 | move.l (%a0),%d0 |
| 747 | move.l %d0,%d1 |
| 748 | lsr.l %d2,%d0 |
| 749 | move.l %d0,(%a0)+ |
| 750 | move.l (%a0),%d0 |
| 751 | lsr.l %d2,%d0 |
| 752 | neg.w %d2 |
| 753 | add.w #32,%d2 |
| 754 | lsl.l %d2,%d1 |
| 755 | or.l %d1,%d0 |
| 756 | move.l (%a0),%d1 |
| 757 | move.l %d0,(%a0) |
| 758 | | Check to see if we shifted off any significant bits |
| 759 | lsl.l %d2,%d1 |
| 760 | jeq fp_nd_round | Nope, round. |
| 761 | bset #0,%d0 | Yes, so set the "sticky bit". |
| 762 | jra fp_nd_round | Now, round. |
| 763 | | Another 64-bit single shift and store |
| 764 | 1: sub.w #32,%d2 |
| 765 | cmp.w #32,%d2 | Do we really need to shift? |
| 766 | jcc 2f | No, the number is too small. |
| 767 | move.l (%a0),%d0 |
| 768 | clr.l (%a0)+ |
| 769 | move.l %d0,%d1 |
| 770 | lsr.l %d2,%d0 |
| 771 | neg.w %d2 |
| 772 | add.w #32,%d2 |
| 773 | | Again, check to see if we shifted off any significant bits. |
| 774 | tst.l (%a0) |
| 775 | jeq 1f |
| 776 | bset #0,%d0 | Sticky bit. |
| 777 | 1: move.l %d0,(%a0) |
| 778 | lsl.l %d2,%d1 |
| 779 | jeq fp_nd_round |
| 780 | bset #0,%d0 |
| 781 | jra fp_nd_round |
| 782 | | Sorry, the number is just too small. |
| 783 | 2: clr.l (%a0)+ |
| 784 | clr.l (%a0) |
| 785 | moveq #1,%d0 | Smallest possible fraction, |
| 786 | jra fp_nd_round | round as desired. |
| 787 | | zero and denormalized |
| 788 | fp_nd_zero: |
| 789 | tst.l (%a0)+ |
| 790 | jne 1f |
| 791 | tst.l (%a0) |
| 792 | jne 1f |
| 793 | subq.l #8,%a0 |
| 794 | printf PNORM,"%p(",1,%a0 |
| 795 | printx PNORM,%a0@ |
| 796 | printf PNORM,")\n" |
| 797 | rts | zero. nothing to do. |
| 798 | | These are not merely subnormal numbers, but true denormals, |
| 799 | | i.e. pathologically small (exponent is 2**-16383) numbers. |
| 800 | | It is clearly impossible for even a normal extended number |
| 801 | | with that exponent to fit into double precision, so just |
| 802 | | write these ones off as "too darn small". |
| 803 | 1: fp_set_sr FPSR_EXC_UNFL | Set UNFL bit |
| 804 | clr.l (%a0) |
| 805 | clr.l -(%a0) |
| 806 | move.w #0x3c01,-(%a0) | i.e. 2**-1022 |
| 807 | addq.l #6,%a0 |
| 808 | moveq #1,%d0 |
| 809 | jra fp_nd_round | round. |
| 810 | | Exponent overflow. Just call it infinity. |
| 811 | fp_nd_large: |
| 812 | move.w #0x7ff,%d0 |
| 813 | and.w (6,%a0),%d0 |
| 814 | jeq 1f |
| 815 | fp_set_sr FPSR_EXC_INEX2 |
| 816 | 1: fp_set_sr FPSR_EXC_OVFL |
| 817 | move.w (FPD_RND,FPDATA),%d2 |
| 818 | jne 3f | %d2 = 0 round to nearest |
| 819 | 1: move.w #0x7fff,(-2,%a0) |
| 820 | clr.l (%a0)+ |
| 821 | clr.l (%a0) |
| 822 | 2: subq.l #8,%a0 |
| 823 | printf PNORM,"%p(",1,%a0 |
| 824 | printx PNORM,%a0@ |
| 825 | printf PNORM,")\n" |
| 826 | rts |
| 827 | 3: subq.w #2,%d2 |
| 828 | jcs 5f | %d2 < 2, round to zero |
| 829 | jhi 4f | %d2 > 2, round to +infinity |
| 830 | tst.b (-3,%a0) | to -inf |
| 831 | jne 1b |
| 832 | jra 5f |
| 833 | 4: tst.b (-3,%a0) | to +inf |
| 834 | jeq 1b |
| 835 | 5: move.w #0x43fe,(-2,%a0) |
| 836 | moveq #-1,%d0 |
| 837 | move.l %d0,(%a0)+ |
| 838 | move.w #0xf800,%d0 |
| 839 | move.l %d0,(%a0) |
| 840 | jra 2b |
| 841 | | Infinities or NaNs |
| 842 | fp_nd_huge: |
| 843 | subq.l #4,%a0 |
| 844 | printf PNORM,"%p(",1,%a0 |
| 845 | printx PNORM,%a0@ |
| 846 | printf PNORM,")\n" |
| 847 | rts |
| 848 | |
| 849 | | fp_normalize_single: |
| 850 | | normalize an extended with single (23-bit) precision |
| 851 | | args: %a0 (struct fp_ext *) |
| 852 | |
| 853 | fp_normalize_single: |
| 854 | printf PNORM,"ns: %p(",1,%a0 |
| 855 | printx PNORM,%a0@ |
| 856 | printf PNORM,") " |
| 857 | addq.l #2,%a0 |
| 858 | move.w (%a0)+,%d2 |
| 859 | jeq fp_ns_zero | zero / denormalized |
| 860 | cmp.w #0x7fff,%d2 |
| 861 | jeq fp_ns_huge | NaN / infinitive. |
| 862 | sub.w #0x4000-0x7f,%d2 | will the exponent fit? |
| 863 | jcs fp_ns_small | too small. |
| 864 | cmp.w #0xfe,%d2 |
| 865 | jcc fp_ns_large | too big. |
| 866 | move.l (%a0)+,%d0 | get high lword of mantissa |
| 867 | fp_ns_round: |
| 868 | tst.l (%a0) | check the low lword |
| 869 | jeq 1f |
| 870 | | Set a sticky bit if it is non-zero. This should only |
| 871 | | affect the rounding in what would otherwise be equal- |
| 872 | | distance situations, which is what we want it to do. |
| 873 | bset #0,%d0 |
| 874 | 1: clr.l (%a0) | zap it from memory. |
| 875 | | now, round off the low 8 bits of the hi lword. |
| 876 | tst.b %d0 | 8 low bits. |
| 877 | jne fp_ns_checkround | Are they non-zero? |
| 878 | | nothing to do here |
| 879 | subq.l #8,%a0 |
| 880 | printf PNORM,"%p(",1,%a0 |
| 881 | printx PNORM,%a0@ |
| 882 | printf PNORM,")\n" |
| 883 | rts |
| 884 | fp_ns_checkround: |
| 885 | fp_set_sr FPSR_EXC_INEX2 | INEX2 bit |
| 886 | clr.b -(%a0) | clear low byte of high lword |
| 887 | subq.l #3,%a0 |
| 888 | move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 889 | jne 2f | %d2 == 0, round to nearest |
| 890 | tst.b %d0 | test guard bit |
| 891 | jpl 9f | zero is closer |
| 892 | btst #8,%d0 | test lsb bit |
| 893 | | round to even behaviour, see above. |
| 894 | jne fp_ns_doroundup | round to infinity |
| 895 | lsl.b #1,%d0 | check low bits |
| 896 | jeq 9f | round to zero |
| 897 | fp_ns_doroundup: |
| 898 | | round (the mantissa, that is) towards infinity |
| 899 | add.l #0x100,(%a0) |
| 900 | jcc 9f | no overflow, good. |
| 901 | | Overflow. This means that the %d1 was 0xffffff00, so it |
| 902 | | is now zero. We will set the mantissa to reflect this, and |
| 903 | | increment the exponent (checking for overflow there too) |
| 904 | move.w #0x8000,(%a0) |
| 905 | addq.w #1,-(%a0) |
| 906 | cmp.w #0x407f,(%a0)+ | exponent now overflown? |
| 907 | jeq fp_ns_large | yes, so make it infinity. |
| 908 | 9: subq.l #4,%a0 |
| 909 | printf PNORM,"%p(",1,%a0 |
| 910 | printx PNORM,%a0@ |
| 911 | printf PNORM,")\n" |
| 912 | rts |
| 913 | | check nondefault rounding modes |
| 914 | 2: subq.w #2,%d2 |
| 915 | jcs 9b | %d2 < 2, round to zero |
| 916 | jhi 3f | %d2 > 2, round to +infinity |
| 917 | tst.b (-3,%a0) | to -inf |
| 918 | jne fp_ns_doroundup | negative, round to infinity |
| 919 | jra 9b | positive, round to zero |
| 920 | 3: tst.b (-3,%a0) | to +inf |
| 921 | jeq fp_ns_doroundup | positive, round to infinity |
| 922 | jra 9b | negative, round to zero |
| 923 | | Exponent underflow. Try to make a denormal, and set it to |
| 924 | | the smallest possible fraction if this fails. |
| 925 | fp_ns_small: |
| 926 | fp_set_sr FPSR_EXC_UNFL | set UNFL bit |
| 927 | move.w #0x3f81,(-2,%a0) | 2**-126 |
| 928 | neg.w %d2 | degree of underflow |
| 929 | cmp.w #32,%d2 | single or double shift? |
| 930 | jcc 2f |
| 931 | | a 32-bit shift. |
| 932 | move.l (%a0),%d0 |
| 933 | move.l %d0,%d1 |
| 934 | lsr.l %d2,%d0 |
| 935 | move.l %d0,(%a0)+ |
| 936 | | Check to see if we shifted off any significant bits. |
| 937 | neg.w %d2 |
| 938 | add.w #32,%d2 |
| 939 | lsl.l %d2,%d1 |
| 940 | jeq 1f |
| 941 | bset #0,%d0 | Sticky bit. |
| 942 | | Check the lower lword |
| 943 | 1: tst.l (%a0) |
| 944 | jeq fp_ns_round |
| 945 | clr (%a0) |
| 946 | bset #0,%d0 | Sticky bit. |
| 947 | jra fp_ns_round |
| 948 | | Sorry, the number is just too small. |
| 949 | 2: clr.l (%a0)+ |
| 950 | clr.l (%a0) |
| 951 | moveq #1,%d0 | Smallest possible fraction, |
| 952 | jra fp_ns_round | round as desired. |
| 953 | | Exponent overflow. Just call it infinity. |
| 954 | fp_ns_large: |
| 955 | tst.b (3,%a0) |
| 956 | jeq 1f |
| 957 | fp_set_sr FPSR_EXC_INEX2 |
| 958 | 1: fp_set_sr FPSR_EXC_OVFL |
| 959 | move.w (FPD_RND,FPDATA),%d2 |
| 960 | jne 3f | %d2 = 0 round to nearest |
| 961 | 1: move.w #0x7fff,(-2,%a0) |
| 962 | clr.l (%a0)+ |
| 963 | clr.l (%a0) |
| 964 | 2: subq.l #8,%a0 |
| 965 | printf PNORM,"%p(",1,%a0 |
| 966 | printx PNORM,%a0@ |
| 967 | printf PNORM,")\n" |
| 968 | rts |
| 969 | 3: subq.w #2,%d2 |
| 970 | jcs 5f | %d2 < 2, round to zero |
| 971 | jhi 4f | %d2 > 2, round to +infinity |
| 972 | tst.b (-3,%a0) | to -inf |
| 973 | jne 1b |
| 974 | jra 5f |
| 975 | 4: tst.b (-3,%a0) | to +inf |
| 976 | jeq 1b |
| 977 | 5: move.w #0x407e,(-2,%a0) |
| 978 | move.l #0xffffff00,(%a0)+ |
| 979 | clr.l (%a0) |
| 980 | jra 2b |
| 981 | | zero and denormalized |
| 982 | fp_ns_zero: |
| 983 | tst.l (%a0)+ |
| 984 | jne 1f |
| 985 | tst.l (%a0) |
| 986 | jne 1f |
| 987 | subq.l #8,%a0 |
| 988 | printf PNORM,"%p(",1,%a0 |
| 989 | printx PNORM,%a0@ |
| 990 | printf PNORM,")\n" |
| 991 | rts | zero. nothing to do. |
| 992 | | These are not merely subnormal numbers, but true denormals, |
| 993 | | i.e. pathologically small (exponent is 2**-16383) numbers. |
| 994 | | It is clearly impossible for even a normal extended number |
| 995 | | with that exponent to fit into single precision, so just |
| 996 | | write these ones off as "too darn small". |
| 997 | 1: fp_set_sr FPSR_EXC_UNFL | Set UNFL bit |
| 998 | clr.l (%a0) |
| 999 | clr.l -(%a0) |
| 1000 | move.w #0x3f81,-(%a0) | i.e. 2**-126 |
| 1001 | addq.l #6,%a0 |
| 1002 | moveq #1,%d0 |
| 1003 | jra fp_ns_round | round. |
| 1004 | | Infinities or NaNs |
| 1005 | fp_ns_huge: |
| 1006 | subq.l #4,%a0 |
| 1007 | printf PNORM,"%p(",1,%a0 |
| 1008 | printx PNORM,%a0@ |
| 1009 | printf PNORM,")\n" |
| 1010 | rts |
| 1011 | |
| 1012 | | fp_normalize_single_fast: |
| 1013 | | normalize an extended with single (23-bit) precision |
| 1014 | | this is only used by fsgldiv/fsgdlmul, where the |
| 1015 | | operand is not completly normalized. |
| 1016 | | args: %a0 (struct fp_ext *) |
| 1017 | |
| 1018 | fp_normalize_single_fast: |
| 1019 | printf PNORM,"nsf: %p(",1,%a0 |
| 1020 | printx PNORM,%a0@ |
| 1021 | printf PNORM,") " |
| 1022 | addq.l #2,%a0 |
| 1023 | move.w (%a0)+,%d2 |
| 1024 | cmp.w #0x7fff,%d2 |
| 1025 | jeq fp_nsf_huge | NaN / infinitive. |
| 1026 | move.l (%a0)+,%d0 | get high lword of mantissa |
| 1027 | fp_nsf_round: |
| 1028 | tst.l (%a0) | check the low lword |
| 1029 | jeq 1f |
| 1030 | | Set a sticky bit if it is non-zero. This should only |
| 1031 | | affect the rounding in what would otherwise be equal- |
| 1032 | | distance situations, which is what we want it to do. |
| 1033 | bset #0,%d0 |
| 1034 | 1: clr.l (%a0) | zap it from memory. |
| 1035 | | now, round off the low 8 bits of the hi lword. |
| 1036 | tst.b %d0 | 8 low bits. |
| 1037 | jne fp_nsf_checkround | Are they non-zero? |
| 1038 | | nothing to do here |
| 1039 | subq.l #8,%a0 |
| 1040 | printf PNORM,"%p(",1,%a0 |
| 1041 | printx PNORM,%a0@ |
| 1042 | printf PNORM,")\n" |
| 1043 | rts |
| 1044 | fp_nsf_checkround: |
| 1045 | fp_set_sr FPSR_EXC_INEX2 | INEX2 bit |
| 1046 | clr.b -(%a0) | clear low byte of high lword |
| 1047 | subq.l #3,%a0 |
| 1048 | move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 1049 | jne 2f | %d2 == 0, round to nearest |
| 1050 | tst.b %d0 | test guard bit |
| 1051 | jpl 9f | zero is closer |
| 1052 | btst #8,%d0 | test lsb bit |
| 1053 | | round to even behaviour, see above. |
| 1054 | jne fp_nsf_doroundup | round to infinity |
| 1055 | lsl.b #1,%d0 | check low bits |
| 1056 | jeq 9f | round to zero |
| 1057 | fp_nsf_doroundup: |
| 1058 | | round (the mantissa, that is) towards infinity |
| 1059 | add.l #0x100,(%a0) |
| 1060 | jcc 9f | no overflow, good. |
| 1061 | | Overflow. This means that the %d1 was 0xffffff00, so it |
| 1062 | | is now zero. We will set the mantissa to reflect this, and |
| 1063 | | increment the exponent (checking for overflow there too) |
| 1064 | move.w #0x8000,(%a0) |
| 1065 | addq.w #1,-(%a0) |
| 1066 | cmp.w #0x407f,(%a0)+ | exponent now overflown? |
| 1067 | jeq fp_nsf_large | yes, so make it infinity. |
| 1068 | 9: subq.l #4,%a0 |
| 1069 | printf PNORM,"%p(",1,%a0 |
| 1070 | printx PNORM,%a0@ |
| 1071 | printf PNORM,")\n" |
| 1072 | rts |
| 1073 | | check nondefault rounding modes |
| 1074 | 2: subq.w #2,%d2 |
| 1075 | jcs 9b | %d2 < 2, round to zero |
| 1076 | jhi 3f | %d2 > 2, round to +infinity |
| 1077 | tst.b (-3,%a0) | to -inf |
| 1078 | jne fp_nsf_doroundup | negative, round to infinity |
| 1079 | jra 9b | positive, round to zero |
| 1080 | 3: tst.b (-3,%a0) | to +inf |
| 1081 | jeq fp_nsf_doroundup | positive, round to infinity |
| 1082 | jra 9b | negative, round to zero |
| 1083 | | Exponent overflow. Just call it infinity. |
| 1084 | fp_nsf_large: |
| 1085 | tst.b (3,%a0) |
| 1086 | jeq 1f |
| 1087 | fp_set_sr FPSR_EXC_INEX2 |
| 1088 | 1: fp_set_sr FPSR_EXC_OVFL |
| 1089 | move.w (FPD_RND,FPDATA),%d2 |
| 1090 | jne 3f | %d2 = 0 round to nearest |
| 1091 | 1: move.w #0x7fff,(-2,%a0) |
| 1092 | clr.l (%a0)+ |
| 1093 | clr.l (%a0) |
| 1094 | 2: subq.l #8,%a0 |
| 1095 | printf PNORM,"%p(",1,%a0 |
| 1096 | printx PNORM,%a0@ |
| 1097 | printf PNORM,")\n" |
| 1098 | rts |
| 1099 | 3: subq.w #2,%d2 |
| 1100 | jcs 5f | %d2 < 2, round to zero |
| 1101 | jhi 4f | %d2 > 2, round to +infinity |
| 1102 | tst.b (-3,%a0) | to -inf |
| 1103 | jne 1b |
| 1104 | jra 5f |
| 1105 | 4: tst.b (-3,%a0) | to +inf |
| 1106 | jeq 1b |
| 1107 | 5: move.w #0x407e,(-2,%a0) |
| 1108 | move.l #0xffffff00,(%a0)+ |
| 1109 | clr.l (%a0) |
| 1110 | jra 2b |
| 1111 | | Infinities or NaNs |
| 1112 | fp_nsf_huge: |
| 1113 | subq.l #4,%a0 |
| 1114 | printf PNORM,"%p(",1,%a0 |
| 1115 | printx PNORM,%a0@ |
| 1116 | printf PNORM,")\n" |
| 1117 | rts |
| 1118 | |
| 1119 | | conv_ext2int (macro): |
| 1120 | | Generates a subroutine that converts an extended value to an |
| 1121 | | integer of a given size, again, with the appropriate type of |
| 1122 | | rounding. |
| 1123 | |
| 1124 | | Macro arguments: |
| 1125 | | s: size, as given in an assembly instruction. |
| 1126 | | b: number of bits in that size. |
| 1127 | |
| 1128 | | Subroutine arguments: |
| 1129 | | %a0: source (struct fp_ext *) |
| 1130 | |
| 1131 | | Returns the integer in %d0 (like it should) |
| 1132 | |
| 1133 | .macro conv_ext2int s,b |
| 1134 | .set inf,(1<<(\b-1))-1 | i.e. MAXINT |
| 1135 | printf PCONV,"e2i%d: %p(",2,#\b,%a0 |
| 1136 | printx PCONV,%a0@ |
| 1137 | printf PCONV,") " |
| 1138 | addq.l #2,%a0 |
| 1139 | move.w (%a0)+,%d2 | exponent |
| 1140 | jeq fp_e2i_zero\b | zero / denorm (== 0, here) |
| 1141 | cmp.w #0x7fff,%d2 |
| 1142 | jeq fp_e2i_huge\b | Inf / NaN |
| 1143 | sub.w #0x3ffe,%d2 |
| 1144 | jcs fp_e2i_small\b |
| 1145 | cmp.w #\b,%d2 |
| 1146 | jhi fp_e2i_large\b |
| 1147 | move.l (%a0),%d0 |
| 1148 | move.l %d0,%d1 |
| 1149 | lsl.l %d2,%d1 |
| 1150 | jne fp_e2i_round\b |
| 1151 | tst.l (4,%a0) |
| 1152 | jne fp_e2i_round\b |
| 1153 | neg.w %d2 |
| 1154 | add.w #32,%d2 |
| 1155 | lsr.l %d2,%d0 |
| 1156 | 9: tst.w (-4,%a0) |
| 1157 | jne 1f |
| 1158 | tst.\s %d0 |
| 1159 | jmi fp_e2i_large\b |
| 1160 | printf PCONV,"-> %p\n",1,%d0 |
| 1161 | rts |
| 1162 | 1: neg.\s %d0 |
| 1163 | jeq 1f |
| 1164 | jpl fp_e2i_large\b |
| 1165 | 1: printf PCONV,"-> %p\n",1,%d0 |
| 1166 | rts |
| 1167 | fp_e2i_round\b: |
| 1168 | fp_set_sr FPSR_EXC_INEX2 | INEX2 bit |
| 1169 | neg.w %d2 |
| 1170 | add.w #32,%d2 |
| 1171 | .if \b>16 |
| 1172 | jeq 5f |
| 1173 | .endif |
| 1174 | lsr.l %d2,%d0 |
| 1175 | move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 1176 | jne 2f | %d2 == 0, round to nearest |
| 1177 | tst.l %d1 | test guard bit |
| 1178 | jpl 9b | zero is closer |
| 1179 | btst %d2,%d0 | test lsb bit (%d2 still 0) |
| 1180 | jne fp_e2i_doroundup\b |
| 1181 | lsl.l #1,%d1 | check low bits |
| 1182 | jne fp_e2i_doroundup\b |
| 1183 | tst.l (4,%a0) |
| 1184 | jeq 9b |
| 1185 | fp_e2i_doroundup\b: |
| 1186 | addq.l #1,%d0 |
| 1187 | jra 9b |
| 1188 | | check nondefault rounding modes |
| 1189 | 2: subq.w #2,%d2 |
| 1190 | jcs 9b | %d2 < 2, round to zero |
| 1191 | jhi 3f | %d2 > 2, round to +infinity |
| 1192 | tst.w (-4,%a0) | to -inf |
| 1193 | jne fp_e2i_doroundup\b | negative, round to infinity |
| 1194 | jra 9b | positive, round to zero |
| 1195 | 3: tst.w (-4,%a0) | to +inf |
| 1196 | jeq fp_e2i_doroundup\b | positive, round to infinity |
| 1197 | jra 9b | negative, round to zero |
| 1198 | | we are only want -2**127 get correctly rounded here, |
| 1199 | | since the guard bit is in the lower lword. |
| 1200 | | everything else ends up anyway as overflow. |
| 1201 | .if \b>16 |
| 1202 | 5: move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 1203 | jne 2b | %d2 == 0, round to nearest |
| 1204 | move.l (4,%a0),%d1 | test guard bit |
| 1205 | jpl 9b | zero is closer |
| 1206 | lsl.l #1,%d1 | check low bits |
| 1207 | jne fp_e2i_doroundup\b |
| 1208 | jra 9b |
| 1209 | .endif |
| 1210 | fp_e2i_zero\b: |
| 1211 | clr.l %d0 |
| 1212 | tst.l (%a0)+ |
| 1213 | jne 1f |
| 1214 | tst.l (%a0) |
| 1215 | jeq 3f |
| 1216 | 1: subq.l #4,%a0 |
| 1217 | fp_clr_sr FPSR_EXC_UNFL | fp_normalize_ext has set this bit |
| 1218 | fp_e2i_small\b: |
| 1219 | fp_set_sr FPSR_EXC_INEX2 |
| 1220 | clr.l %d0 |
| 1221 | move.w (FPD_RND,FPDATA),%d2 | rounding mode |
| 1222 | subq.w #2,%d2 |
| 1223 | jcs 3f | %d2 < 2, round to nearest/zero |
| 1224 | jhi 2f | %d2 > 2, round to +infinity |
| 1225 | tst.w (-4,%a0) | to -inf |
| 1226 | jeq 3f |
| 1227 | subq.\s #1,%d0 |
| 1228 | jra 3f |
| 1229 | 2: tst.w (-4,%a0) | to +inf |
| 1230 | jne 3f |
| 1231 | addq.\s #1,%d0 |
| 1232 | 3: printf PCONV,"-> %p\n",1,%d0 |
| 1233 | rts |
| 1234 | fp_e2i_large\b: |
| 1235 | fp_set_sr FPSR_EXC_OPERR |
| 1236 | move.\s #inf,%d0 |
| 1237 | tst.w (-4,%a0) |
| 1238 | jeq 1f |
| 1239 | addq.\s #1,%d0 |
| 1240 | 1: printf PCONV,"-> %p\n",1,%d0 |
| 1241 | rts |
| 1242 | fp_e2i_huge\b: |
| 1243 | move.\s (%a0),%d0 |
| 1244 | tst.l (%a0) |
| 1245 | jne 1f |
| 1246 | tst.l (%a0) |
| 1247 | jeq fp_e2i_large\b |
| 1248 | | fp_normalize_ext has set this bit already |
| 1249 | | and made the number nonsignaling |
| 1250 | 1: fp_tst_sr FPSR_EXC_SNAN |
| 1251 | jne 1f |
| 1252 | fp_set_sr FPSR_EXC_OPERR |
| 1253 | 1: printf PCONV,"-> %p\n",1,%d0 |
| 1254 | rts |
| 1255 | .endm |
| 1256 | |
| 1257 | fp_conv_ext2long: |
| 1258 | conv_ext2int l,32 |
| 1259 | |
| 1260 | fp_conv_ext2short: |
| 1261 | conv_ext2int w,16 |
| 1262 | |
| 1263 | fp_conv_ext2byte: |
| 1264 | conv_ext2int b,8 |
| 1265 | |
| 1266 | fp_conv_ext2double: |
| 1267 | jsr fp_normalize_double |
| 1268 | printf PCONV,"e2d: %p(",1,%a0 |
| 1269 | printx PCONV,%a0@ |
| 1270 | printf PCONV,"), " |
| 1271 | move.l (%a0)+,%d2 |
| 1272 | cmp.w #0x7fff,%d2 |
| 1273 | jne 1f |
| 1274 | move.w #0x7ff,%d2 |
| 1275 | move.l (%a0)+,%d0 |
| 1276 | jra 2f |
| 1277 | 1: sub.w #0x3fff-0x3ff,%d2 |
| 1278 | move.l (%a0)+,%d0 |
| 1279 | jmi 2f |
| 1280 | clr.w %d2 |
| 1281 | 2: lsl.w #5,%d2 |
| 1282 | lsl.l #7,%d2 |
| 1283 | lsl.l #8,%d2 |
| 1284 | move.l %d0,%d1 |
| 1285 | lsl.l #1,%d0 |
| 1286 | lsr.l #4,%d0 |
| 1287 | lsr.l #8,%d0 |
| 1288 | or.l %d2,%d0 |
| 1289 | putuser.l %d0,(%a1)+,fp_err_ua2,%a1 |
| 1290 | moveq #21,%d0 |
| 1291 | lsl.l %d0,%d1 |
| 1292 | move.l (%a0),%d0 |
| 1293 | lsr.l #4,%d0 |
| 1294 | lsr.l #7,%d0 |
| 1295 | or.l %d1,%d0 |
| 1296 | putuser.l %d0,(%a1),fp_err_ua2,%a1 |
| 1297 | #ifdef FPU_EMU_DEBUG |
| 1298 | getuser.l %a1@(-4),%d0,fp_err_ua2,%a1 |
| 1299 | getuser.l %a1@(0),%d1,fp_err_ua2,%a1 |
| 1300 | printf PCONV,"%p(%08x%08x)\n",3,%a1,%d0,%d1 |
| 1301 | #endif |
| 1302 | rts |
| 1303 | |
| 1304 | fp_conv_ext2single: |
| 1305 | jsr fp_normalize_single |
| 1306 | printf PCONV,"e2s: %p(",1,%a0 |
| 1307 | printx PCONV,%a0@ |
| 1308 | printf PCONV,"), " |
| 1309 | move.l (%a0)+,%d1 |
| 1310 | cmp.w #0x7fff,%d1 |
| 1311 | jne 1f |
| 1312 | move.w #0xff,%d1 |
| 1313 | move.l (%a0)+,%d0 |
| 1314 | jra 2f |
| 1315 | 1: sub.w #0x3fff-0x7f,%d1 |
| 1316 | move.l (%a0)+,%d0 |
| 1317 | jmi 2f |
| 1318 | clr.w %d1 |
| 1319 | 2: lsl.w #8,%d1 |
| 1320 | lsl.l #7,%d1 |
| 1321 | lsl.l #8,%d1 |
| 1322 | bclr #31,%d0 |
| 1323 | lsr.l #8,%d0 |
| 1324 | or.l %d1,%d0 |
| 1325 | printf PCONV,"%08x\n",1,%d0 |
| 1326 | rts |
| 1327 | |
| 1328 | | special return addresses for instr that |
| 1329 | | encode the rounding precision in the opcode |
| 1330 | | (e.g. fsmove,fdmove) |
| 1331 | |
| 1332 | fp_finalrounding_single: |
| 1333 | addq.l #8,%sp |
| 1334 | jsr fp_normalize_ext |
| 1335 | jsr fp_normalize_single |
| 1336 | jra fp_finaltest |
| 1337 | |
| 1338 | fp_finalrounding_single_fast: |
| 1339 | addq.l #8,%sp |
| 1340 | jsr fp_normalize_ext |
| 1341 | jsr fp_normalize_single_fast |
| 1342 | jra fp_finaltest |
| 1343 | |
| 1344 | fp_finalrounding_double: |
| 1345 | addq.l #8,%sp |
| 1346 | jsr fp_normalize_ext |
| 1347 | jsr fp_normalize_double |
| 1348 | jra fp_finaltest |
| 1349 | |
| 1350 | | fp_finaltest: |
| 1351 | | set the emulated status register based on the outcome of an |
| 1352 | | emulated instruction. |
| 1353 | |
| 1354 | fp_finalrounding: |
| 1355 | addq.l #8,%sp |
| 1356 | | printf ,"f: %p\n",1,%a0 |
| 1357 | jsr fp_normalize_ext |
| 1358 | move.w (FPD_PREC,FPDATA),%d0 |
| 1359 | subq.w #1,%d0 |
| 1360 | jcs fp_finaltest |
| 1361 | jne 1f |
| 1362 | jsr fp_normalize_single |
| 1363 | jra 2f |
| 1364 | 1: jsr fp_normalize_double |
| 1365 | 2:| printf ,"f: %p\n",1,%a0 |
| 1366 | fp_finaltest: |
| 1367 | | First, we do some of the obvious tests for the exception |
| 1368 | | status byte and condition code bytes of fp_sr here, so that |
| 1369 | | they do not have to be handled individually by every |
| 1370 | | emulated instruction. |
| 1371 | clr.l %d0 |
| 1372 | addq.l #1,%a0 |
| 1373 | tst.b (%a0)+ | sign |
| 1374 | jeq 1f |
| 1375 | bset #FPSR_CC_NEG-24,%d0 | N bit |
| 1376 | 1: cmp.w #0x7fff,(%a0)+ | exponent |
| 1377 | jeq 2f |
| 1378 | | test for zero |
| 1379 | moveq #FPSR_CC_Z-24,%d1 |
| 1380 | tst.l (%a0)+ |
| 1381 | jne 9f |
| 1382 | tst.l (%a0) |
| 1383 | jne 9f |
| 1384 | jra 8f |
| 1385 | | infinitiv and NAN |
| 1386 | 2: moveq #FPSR_CC_NAN-24,%d1 |
| 1387 | move.l (%a0)+,%d2 |
| 1388 | lsl.l #1,%d2 | ignore high bit |
| 1389 | jne 8f |
| 1390 | tst.l (%a0) |
| 1391 | jne 8f |
| 1392 | moveq #FPSR_CC_INF-24,%d1 |
| 1393 | 8: bset %d1,%d0 |
| 1394 | 9: move.b %d0,(FPD_FPSR+0,FPDATA) | set condition test result |
| 1395 | | move instructions enter here |
| 1396 | | Here, we test things in the exception status byte, and set |
| 1397 | | other things in the accrued exception byte accordingly. |
| 1398 | | Emulated instructions can set various things in the former, |
| 1399 | | as defined in fp_emu.h. |
| 1400 | fp_final: |
| 1401 | move.l (FPD_FPSR,FPDATA),%d0 |
| 1402 | #if 0 |
| 1403 | btst #FPSR_EXC_SNAN,%d0 | EXC_SNAN |
| 1404 | jne 1f |
| 1405 | btst #FPSR_EXC_OPERR,%d0 | EXC_OPERR |
| 1406 | jeq 2f |
| 1407 | 1: bset #FPSR_AEXC_IOP,%d0 | set IOP bit |
| 1408 | 2: btst #FPSR_EXC_OVFL,%d0 | EXC_OVFL |
| 1409 | jeq 1f |
| 1410 | bset #FPSR_AEXC_OVFL,%d0 | set OVFL bit |
| 1411 | 1: btst #FPSR_EXC_UNFL,%d0 | EXC_UNFL |
| 1412 | jeq 1f |
| 1413 | btst #FPSR_EXC_INEX2,%d0 | EXC_INEX2 |
| 1414 | jeq 1f |
| 1415 | bset #FPSR_AEXC_UNFL,%d0 | set UNFL bit |
| 1416 | 1: btst #FPSR_EXC_DZ,%d0 | EXC_INEX1 |
| 1417 | jeq 1f |
| 1418 | bset #FPSR_AEXC_DZ,%d0 | set DZ bit |
| 1419 | 1: btst #FPSR_EXC_OVFL,%d0 | EXC_OVFL |
| 1420 | jne 1f |
| 1421 | btst #FPSR_EXC_INEX2,%d0 | EXC_INEX2 |
| 1422 | jne 1f |
| 1423 | btst #FPSR_EXC_INEX1,%d0 | EXC_INEX1 |
| 1424 | jeq 2f |
| 1425 | 1: bset #FPSR_AEXC_INEX,%d0 | set INEX bit |
| 1426 | 2: move.l %d0,(FPD_FPSR,FPDATA) |
| 1427 | #else |
| 1428 | | same as above, greatly optimized, but untested (yet) |
| 1429 | move.l %d0,%d2 |
| 1430 | lsr.l #5,%d0 |
| 1431 | move.l %d0,%d1 |
| 1432 | lsr.l #4,%d1 |
| 1433 | or.l %d0,%d1 |
| 1434 | and.b #0x08,%d1 |
| 1435 | move.l %d2,%d0 |
| 1436 | lsr.l #6,%d0 |
| 1437 | or.l %d1,%d0 |
| 1438 | move.l %d2,%d1 |
| 1439 | lsr.l #4,%d1 |
| 1440 | or.b #0xdf,%d1 |
| 1441 | and.b %d1,%d0 |
| 1442 | move.l %d2,%d1 |
| 1443 | lsr.l #7,%d1 |
| 1444 | and.b #0x80,%d1 |
| 1445 | or.b %d1,%d0 |
| 1446 | and.b #0xf8,%d0 |
| 1447 | or.b %d0,%d2 |
| 1448 | move.l %d2,(FPD_FPSR,FPDATA) |
| 1449 | #endif |
| 1450 | move.b (FPD_FPSR+2,FPDATA),%d0 |
| 1451 | and.b (FPD_FPCR+2,FPDATA),%d0 |
| 1452 | jeq 1f |
| 1453 | printf ,"send signal!!!\n" |
| 1454 | 1: jra fp_end |