| 1 |
/* -*- Mode: Asm -*- */ |
| 2 |
;; Copyright (C) 2012-2025 Free Software Foundation, Inc. |
| 3 |
;; Contributed by Sean D'Epagnier ([email protected]) |
| 4 |
;; Georg-Johann Lay ([email protected]) |
| 5 |
|
| 6 |
;; This file 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 |
;; In addition to the permissions in the GNU General Public License, the |
| 12 |
;; Free Software Foundation gives you unlimited permission to link the |
| 13 |
;; compiled version of this file into combinations with other programs, |
| 14 |
;; and to distribute those combinations without any restriction coming |
| 15 |
;; from the use of this file. (The General Public License restrictions |
| 16 |
;; do apply in other respects; for example, they cover modification of |
| 17 |
;; the file, and distribution when not linked into a combine |
| 18 |
;; executable.) |
| 19 |
|
| 20 |
;; This file is distributed in the hope that it will be useful, but |
| 21 |
;; WITHOUT ANY WARRANTY; without even the implied warranty of |
| 22 |
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 23 |
;; General Public License for more details. |
| 24 |
|
| 25 |
;; You should have received a copy of the GNU General Public License |
| 26 |
;; along with this program; see the file COPYING. If not, write to |
| 27 |
;; the Free Software Foundation, 51 Franklin Street, Fifth Floor, |
| 28 |
;; Boston, MA 02110-1301, USA. |
| 29 |
|
| 30 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 31 |
;; Fixed point library routines for AVR |
| 32 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 33 |
|
| 34 |
.section .text.libgcc.fixed, "ax", @progbits |
| 35 |
|
| 36 |
#ifndef __AVR_TINY__ |
| 37 |
|
| 38 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 39 |
;; Conversions to float |
| 40 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 41 |
|
| 42 |
#if defined (L_fractqqsf) |
| 43 |
DEFUN __fractqqsf |
| 44 |
;; Move in place for SA -> SF conversion |
| 45 |
clr r22 |
| 46 |
mov r23, r24 |
| 47 |
;; Sign-extend |
| 48 |
lsl r24 |
| 49 |
sbc r24, r24 |
| 50 |
mov r25, r24 |
| 51 |
XJMP __fractsasf |
| 52 |
ENDF __fractqqsf |
| 53 |
#endif /* L_fractqqsf */ |
| 54 |
|
| 55 |
#if defined (L_fractuqqsf) |
| 56 |
DEFUN __fractuqqsf |
| 57 |
;; Move in place for USA -> SF conversion |
| 58 |
clr r22 |
| 59 |
mov r23, r24 |
| 60 |
;; Zero-extend |
| 61 |
clr r24 |
| 62 |
clr r25 |
| 63 |
XJMP __fractusasf |
| 64 |
ENDF __fractuqqsf |
| 65 |
#endif /* L_fractuqqsf */ |
| 66 |
|
| 67 |
#if defined (L_fracthqsf) |
| 68 |
DEFUN __fracthqsf |
| 69 |
;; Move in place for SA -> SF conversion |
| 70 |
wmov 22, 24 |
| 71 |
;; Sign-extend |
| 72 |
lsl r25 |
| 73 |
sbc r24, r24 |
| 74 |
mov r25, r24 |
| 75 |
XJMP __fractsasf |
| 76 |
ENDF __fracthqsf |
| 77 |
#endif /* L_fracthqsf */ |
| 78 |
|
| 79 |
#if defined (L_fractuhqsf) |
| 80 |
DEFUN __fractuhqsf |
| 81 |
;; Move in place for USA -> SF conversion |
| 82 |
wmov 22, 24 |
| 83 |
;; Zero-extend |
| 84 |
clr r24 |
| 85 |
clr r25 |
| 86 |
XJMP __fractusasf |
| 87 |
ENDF __fractuhqsf |
| 88 |
#endif /* L_fractuhqsf */ |
| 89 |
|
| 90 |
#if defined (L_fracthasf) |
| 91 |
DEFUN __fracthasf |
| 92 |
;; Move in place for SA -> SF conversion |
| 93 |
clr r22 |
| 94 |
mov r23, r24 |
| 95 |
mov r24, r25 |
| 96 |
;; Sign-extend |
| 97 |
lsl r25 |
| 98 |
sbc r25, r25 |
| 99 |
XJMP __fractsasf |
| 100 |
ENDF __fracthasf |
| 101 |
#endif /* L_fracthasf */ |
| 102 |
|
| 103 |
#if defined (L_fractuhasf) |
| 104 |
DEFUN __fractuhasf |
| 105 |
;; Move in place for USA -> SF conversion |
| 106 |
clr r22 |
| 107 |
mov r23, r24 |
| 108 |
mov r24, r25 |
| 109 |
;; Zero-extend |
| 110 |
clr r25 |
| 111 |
XJMP __fractusasf |
| 112 |
ENDF __fractuhasf |
| 113 |
#endif /* L_fractuhasf */ |
| 114 |
|
| 115 |
|
| 116 |
#if defined (L_fractsqsf) |
| 117 |
DEFUN __fractsqsf |
| 118 |
XCALL __floatsisf |
| 119 |
;; Divide non-zero results by 2^31 to move the |
| 120 |
;; decimal point into place |
| 121 |
tst r25 |
| 122 |
breq 0f |
| 123 |
subi r24, exp_lo (31) |
| 124 |
sbci r25, exp_hi (31) |
| 125 |
0: ret |
| 126 |
ENDF __fractsqsf |
| 127 |
#endif /* L_fractsqsf */ |
| 128 |
|
| 129 |
#if defined (L_fractusqsf) |
| 130 |
DEFUN __fractusqsf |
| 131 |
XCALL __floatunsisf |
| 132 |
;; Divide non-zero results by 2^32 to move the |
| 133 |
;; decimal point into place |
| 134 |
cpse r25, __zero_reg__ |
| 135 |
subi r25, exp_hi (32) |
| 136 |
ret |
| 137 |
ENDF __fractusqsf |
| 138 |
#endif /* L_fractusqsf */ |
| 139 |
|
| 140 |
#if defined (L_fractsasf) |
| 141 |
DEFUN __fractsasf |
| 142 |
XCALL __floatsisf |
| 143 |
;; Divide non-zero results by 2^15 to move the |
| 144 |
;; decimal point into place |
| 145 |
tst r25 |
| 146 |
breq 0f |
| 147 |
subi r24, exp_lo (15) |
| 148 |
sbci r25, exp_hi (15) |
| 149 |
0: ret |
| 150 |
ENDF __fractsasf |
| 151 |
#endif /* L_fractsasf */ |
| 152 |
|
| 153 |
#if defined (L_fractusasf) |
| 154 |
DEFUN __fractusasf |
| 155 |
XCALL __floatunsisf |
| 156 |
;; Divide non-zero results by 2^16 to move the |
| 157 |
;; decimal point into place |
| 158 |
cpse r25, __zero_reg__ |
| 159 |
subi r25, exp_hi (16) |
| 160 |
ret |
| 161 |
ENDF __fractusasf |
| 162 |
#endif /* L_fractusasf */ |
| 163 |
|
| 164 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 165 |
;; Conversions from float |
| 166 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 167 |
|
| 168 |
#if defined (L_fractsfqq) |
| 169 |
DEFUN __fractsfqq |
| 170 |
;; Multiply with 2^{24+7} to get a QQ result in r25 |
| 171 |
subi r24, exp_lo (-31) |
| 172 |
sbci r25, exp_hi (-31) |
| 173 |
XCALL __fixsfsi |
| 174 |
mov r24, r25 |
| 175 |
ret |
| 176 |
ENDF __fractsfqq |
| 177 |
#endif /* L_fractsfqq */ |
| 178 |
|
| 179 |
#if defined (L_fractsfuqq) |
| 180 |
DEFUN __fractsfuqq |
| 181 |
;; Multiply with 2^{24+8} to get a UQQ result in r25 |
| 182 |
subi r25, exp_hi (-32) |
| 183 |
XCALL __fixunssfsi |
| 184 |
mov r24, r25 |
| 185 |
ret |
| 186 |
ENDF __fractsfuqq |
| 187 |
#endif /* L_fractsfuqq */ |
| 188 |
|
| 189 |
#if defined (L_fractsfha) |
| 190 |
DEFUN __fractsfha |
| 191 |
;; Multiply with 2^{16+7} to get a HA result in r25:r24 |
| 192 |
subi r24, exp_lo (-23) |
| 193 |
sbci r25, exp_hi (-23) |
| 194 |
XJMP __fixsfsi |
| 195 |
ENDF __fractsfha |
| 196 |
#endif /* L_fractsfha */ |
| 197 |
|
| 198 |
#if defined (L_fractsfuha) |
| 199 |
DEFUN __fractsfuha |
| 200 |
;; Multiply with 2^24 to get a UHA result in r25:r24 |
| 201 |
subi r25, exp_hi (-24) |
| 202 |
XJMP __fixunssfsi |
| 203 |
ENDF __fractsfuha |
| 204 |
#endif /* L_fractsfuha */ |
| 205 |
|
| 206 |
#if defined (L_fractsfhq) |
| 207 |
FALIAS __fractsfsq |
| 208 |
|
| 209 |
DEFUN __fractsfhq |
| 210 |
;; Multiply with 2^{16+15} to get a HQ result in r25:r24 |
| 211 |
;; resp. with 2^31 to get a SQ result in r25:r22 |
| 212 |
subi r24, exp_lo (-31) |
| 213 |
sbci r25, exp_hi (-31) |
| 214 |
XJMP __fixsfsi |
| 215 |
ENDF __fractsfhq |
| 216 |
#endif /* L_fractsfhq */ |
| 217 |
|
| 218 |
#if defined (L_fractsfuhq) |
| 219 |
FALIAS __fractsfusq |
| 220 |
|
| 221 |
DEFUN __fractsfuhq |
| 222 |
;; Multiply with 2^{16+16} to get a UHQ result in r25:r24 |
| 223 |
;; resp. with 2^32 to get a USQ result in r25:r22 |
| 224 |
subi r25, exp_hi (-32) |
| 225 |
XJMP __fixunssfsi |
| 226 |
ENDF __fractsfuhq |
| 227 |
#endif /* L_fractsfuhq */ |
| 228 |
|
| 229 |
#if defined (L_fractsfsa) |
| 230 |
DEFUN __fractsfsa |
| 231 |
;; Multiply with 2^15 to get a SA result in r25:r22 |
| 232 |
subi r24, exp_lo (-15) |
| 233 |
sbci r25, exp_hi (-15) |
| 234 |
XJMP __fixsfsi |
| 235 |
ENDF __fractsfsa |
| 236 |
#endif /* L_fractsfsa */ |
| 237 |
|
| 238 |
#if defined (L_fractsfusa) |
| 239 |
DEFUN __fractsfusa |
| 240 |
;; Multiply with 2^16 to get a USA result in r25:r22 |
| 241 |
subi r25, exp_hi (-16) |
| 242 |
XJMP __fixunssfsi |
| 243 |
ENDF __fractsfusa |
| 244 |
#endif /* L_fractsfusa */ |
| 245 |
|
| 246 |
#endif /* ! AVR_TINY */ |
| 247 |
|
| 248 |
;; For multiplication the functions here are called directly from |
| 249 |
;; avr-fixed.md instead of using the standard libcall mechanisms. |
| 250 |
;; This can make better code because GCC knows exactly which |
| 251 |
;; of the call-used registers (not all of them) are clobbered. */ |
| 252 |
|
| 253 |
/******************************************************* |
| 254 |
Fractional Multiplication 8 x 8 without MUL |
| 255 |
*******************************************************/ |
| 256 |
|
| 257 |
#if defined (L_mulqq3) && !defined (__AVR_HAVE_MUL__) |
| 258 |
;;; R23 = R24 * R25 |
| 259 |
;;; Clobbers: __tmp_reg__, R22, R24, R25 |
| 260 |
;;; Rounding: ??? |
| 261 |
DEFUN __mulqq3 |
| 262 |
XCALL __fmuls |
| 263 |
;; TR 18037 requires that (-1) * (-1) does not overflow |
| 264 |
;; The only input that can produce -1 is (-1)^2. |
| 265 |
dec r23 |
| 266 |
brvs 0f |
| 267 |
inc r23 |
| 268 |
0: ret |
| 269 |
ENDF __mulqq3 |
| 270 |
#endif /* L_mulqq3 && ! HAVE_MUL */ |
| 271 |
|
| 272 |
/******************************************************* |
| 273 |
Fractional Multiply .16 x .16 with and without MUL |
| 274 |
*******************************************************/ |
| 275 |
|
| 276 |
#if defined (L_mulhq3) |
| 277 |
;;; Same code with and without MUL, but the interfaces differ: |
| 278 |
;;; no MUL: (R25:R24) = (R22:R23) * (R24:R25) |
| 279 |
;;; Clobbers: ABI, called by optabs |
| 280 |
;;; MUL: (R25:R24) = (R19:R18) * (R27:R26) |
| 281 |
;;; Clobbers: __tmp_reg__, R22, R23 |
| 282 |
;;; Rounding: -0.5 LSB <= error <= 0.5 LSB |
| 283 |
DEFUN __mulhq3 |
| 284 |
XCALL __mulhisi3 |
| 285 |
;; Shift result into place |
| 286 |
lsl r23 |
| 287 |
rol r24 |
| 288 |
rol r25 |
| 289 |
brvs 1f |
| 290 |
;; Round |
| 291 |
#ifndef __AVR_TINY__ |
| 292 |
sbrc r23, 7 |
| 293 |
adiw r24, 1 |
| 294 |
#else |
| 295 |
lsl r23 |
| 296 |
adc r24, __zero_reg__ |
| 297 |
adc r25, __zero_reg__ |
| 298 |
#endif /* AVR_TINY? */ |
| 299 |
ret |
| 300 |
1: ;; Overflow. TR 18037 requires (-1)^2 not to overflow |
| 301 |
ldi r24, lo8 (0x7fff) |
| 302 |
ldi r25, hi8 (0x7fff) |
| 303 |
ret |
| 304 |
ENDF __mulhq3 |
| 305 |
#endif /* defined (L_mulhq3) */ |
| 306 |
|
| 307 |
#if defined (L_muluhq3) |
| 308 |
;;; Same code with and without MUL, but the interfaces differ: |
| 309 |
;;; no MUL: (R25:R24) *= (R23:R22) |
| 310 |
;;; Clobbers: ABI, called by optabs |
| 311 |
;;; MUL: (R25:R24) = (R19:R18) * (R27:R26) |
| 312 |
;;; Clobbers: __tmp_reg__, R22, R23 |
| 313 |
;;; Rounding: -0.5 LSB < error <= 0.5 LSB |
| 314 |
DEFUN __muluhq3 |
| 315 |
XCALL __umulhisi3 |
| 316 |
;; Round |
| 317 |
#ifndef __AVR_TINY__ |
| 318 |
sbrc r23, 7 |
| 319 |
adiw r24, 1 |
| 320 |
#else |
| 321 |
lsl r23 |
| 322 |
adc r24, __zero_reg__ |
| 323 |
adc r25, __zero_reg__ |
| 324 |
#endif /* AVR_TINY? */ |
| 325 |
ret |
| 326 |
ENDF __muluhq3 |
| 327 |
#endif /* L_muluhq3 */ |
| 328 |
|
| 329 |
|
| 330 |
/******************************************************* |
| 331 |
Fixed Multiply 8.8 x 8.8 with and without MUL |
| 332 |
*******************************************************/ |
| 333 |
|
| 334 |
#if defined (L_mulha3) |
| 335 |
;;; Same code with and without MUL, but the interfaces differ: |
| 336 |
;;; no MUL: (R25:R24) = (R22:R23) * (R24:R25) |
| 337 |
;;; Clobbers: ABI, called by optabs |
| 338 |
;;; MUL: (R25:R24) = (R19:R18) * (R27:R26) |
| 339 |
;;; Clobbers: __tmp_reg__, R22, R23 |
| 340 |
;;; Rounding: -0.5 LSB <= error <= 0.5 LSB |
| 341 |
DEFUN __mulha3 |
| 342 |
XCALL __mulhisi3 |
| 343 |
lsl r22 |
| 344 |
rol r23 |
| 345 |
rol r24 |
| 346 |
XJMP __muluha3_round |
| 347 |
ENDF __mulha3 |
| 348 |
#endif /* L_mulha3 */ |
| 349 |
|
| 350 |
#if defined (L_muluha3) |
| 351 |
;;; Same code with and without MUL, but the interfaces differ: |
| 352 |
;;; no MUL: (R25:R24) *= (R23:R22) |
| 353 |
;;; Clobbers: ABI, called by optabs |
| 354 |
;;; MUL: (R25:R24) = (R19:R18) * (R27:R26) |
| 355 |
;;; Clobbers: __tmp_reg__, R22, R23 |
| 356 |
;;; Rounding: -0.5 LSB < error <= 0.5 LSB |
| 357 |
DEFUN __muluha3 |
| 358 |
XCALL __umulhisi3 |
| 359 |
XJMP __muluha3_round |
| 360 |
ENDF __muluha3 |
| 361 |
#endif /* L_muluha3 */ |
| 362 |
|
| 363 |
#if defined (L_muluha3_round) |
| 364 |
DEFUN __muluha3_round |
| 365 |
;; Shift result into place |
| 366 |
mov r25, r24 |
| 367 |
mov r24, r23 |
| 368 |
;; Round |
| 369 |
#ifndef __AVR_TINY__ |
| 370 |
sbrc r22, 7 |
| 371 |
adiw r24, 1 |
| 372 |
#else |
| 373 |
lsl r22 |
| 374 |
adc r24, __zero_reg__ |
| 375 |
adc r25, __zero_reg__ |
| 376 |
#endif /* AVR_TINY? */ |
| 377 |
ret |
| 378 |
ENDF __muluha3_round |
| 379 |
#endif /* L_muluha3_round */ |
| 380 |
|
| 381 |
|
| 382 |
/******************************************************* |
| 383 |
Fixed Multiplication 16.16 x 16.16 |
| 384 |
*******************************************************/ |
| 385 |
|
| 386 |
#ifndef __AVR_TINY__ |
| 387 |
|
| 388 |
;; Bits outside the result (below LSB), used in the signed version |
| 389 |
#define GUARD __tmp_reg__ |
| 390 |
|
| 391 |
#if defined (__AVR_HAVE_MUL__) |
| 392 |
|
| 393 |
;; Multiplier |
| 394 |
#define A0 16 |
| 395 |
#define A1 A0+1 |
| 396 |
#define A2 A1+1 |
| 397 |
#define A3 A2+1 |
| 398 |
|
| 399 |
;; Multiplicand |
| 400 |
#define B0 20 |
| 401 |
#define B1 B0+1 |
| 402 |
#define B2 B1+1 |
| 403 |
#define B3 B2+1 |
| 404 |
|
| 405 |
;; Result |
| 406 |
#define C0 24 |
| 407 |
#define C1 C0+1 |
| 408 |
#define C2 C1+1 |
| 409 |
#define C3 C2+1 |
| 410 |
|
| 411 |
#if defined (L_mulusa3) |
| 412 |
;;; (C3:C0) = (A3:A0) * (B3:B0) |
| 413 |
DEFUN __mulusa3 |
| 414 |
set |
| 415 |
;; Fallthru |
| 416 |
ENDF __mulusa3 |
| 417 |
|
| 418 |
;;; Round for last digit iff T = 1 |
| 419 |
;;; Return guard bits in GUARD (__tmp_reg__). |
| 420 |
;;; Rounding, T = 0: -1.0 LSB < error <= 0 LSB |
| 421 |
;;; Rounding, T = 1: -0.5 LSB < error <= 0.5 LSB |
| 422 |
DEFUN __mulusa3_round |
| 423 |
;; Some of the MUL instructions have LSBs outside the result. |
| 424 |
;; Don't ignore these LSBs in order to tame rounding error. |
| 425 |
;; Use C2/C3 for these LSBs. |
| 426 |
|
| 427 |
clr C0 |
| 428 |
clr C1 |
| 429 |
mul A0, B0 $ movw C2, r0 |
| 430 |
|
| 431 |
mul A1, B0 $ add C3, r0 $ adc C0, r1 |
| 432 |
mul A0, B1 $ add C3, r0 $ adc C0, r1 $ rol C1 |
| 433 |
|
| 434 |
;; Round if T = 1. Store guarding bits outside the result for rounding |
| 435 |
;; and left-shift by the signed version (function below). |
| 436 |
brtc 0f |
| 437 |
sbrc C3, 7 |
| 438 |
adiw C0, 1 |
| 439 |
0: push C3 |
| 440 |
|
| 441 |
;; The following MULs don't have LSBs outside the result. |
| 442 |
;; C2/C3 is the high part. |
| 443 |
|
| 444 |
mul A0, B2 $ add C0, r0 $ adc C1, r1 $ sbc C2, C2 |
| 445 |
mul A1, B1 $ add C0, r0 $ adc C1, r1 $ sbci C2, 0 |
| 446 |
mul A2, B0 $ add C0, r0 $ adc C1, r1 $ sbci C2, 0 |
| 447 |
neg C2 |
| 448 |
|
| 449 |
mul A0, B3 $ add C1, r0 $ adc C2, r1 $ sbc C3, C3 |
| 450 |
mul A1, B2 $ add C1, r0 $ adc C2, r1 $ sbci C3, 0 |
| 451 |
mul A2, B1 $ add C1, r0 $ adc C2, r1 $ sbci C3, 0 |
| 452 |
mul A3, B0 $ add C1, r0 $ adc C2, r1 $ sbci C3, 0 |
| 453 |
neg C3 |
| 454 |
|
| 455 |
mul A1, B3 $ add C2, r0 $ adc C3, r1 |
| 456 |
mul A2, B2 $ add C2, r0 $ adc C3, r1 |
| 457 |
mul A3, B1 $ add C2, r0 $ adc C3, r1 |
| 458 |
|
| 459 |
mul A2, B3 $ add C3, r0 |
| 460 |
mul A3, B2 $ add C3, r0 |
| 461 |
|
| 462 |
;; Guard bits used in the signed version below. |
| 463 |
pop GUARD |
| 464 |
clr __zero_reg__ |
| 465 |
ret |
| 466 |
ENDF __mulusa3_round |
| 467 |
#endif /* L_mulusa3 */ |
| 468 |
|
| 469 |
#if defined (L_mulsa3) |
| 470 |
;;; (C3:C0) = (A3:A0) * (B3:B0) |
| 471 |
;;; Clobbers: __tmp_reg__, T |
| 472 |
;;; Rounding: -0.5 LSB <= error <= 0.5 LSB |
| 473 |
DEFUN __mulsa3 |
| 474 |
clt |
| 475 |
XCALL __mulusa3_round |
| 476 |
;; A posteriori sign extension of the operands |
| 477 |
tst B3 |
| 478 |
brpl 1f |
| 479 |
sub C2, A0 |
| 480 |
sbc C3, A1 |
| 481 |
1: sbrs A3, 7 |
| 482 |
rjmp 2f |
| 483 |
sub C2, B0 |
| 484 |
sbc C3, B1 |
| 485 |
2: |
| 486 |
;; Shift 1 bit left to adjust for 15 fractional bits |
| 487 |
lsl GUARD |
| 488 |
rol C0 |
| 489 |
rol C1 |
| 490 |
rol C2 |
| 491 |
rol C3 |
| 492 |
;; Round last digit |
| 493 |
lsl GUARD |
| 494 |
adc C0, __zero_reg__ |
| 495 |
adc C1, __zero_reg__ |
| 496 |
adc C2, __zero_reg__ |
| 497 |
adc C3, __zero_reg__ |
| 498 |
ret |
| 499 |
ENDF __mulsa3 |
| 500 |
#endif /* L_mulsa3 */ |
| 501 |
|
| 502 |
#undef A0 |
| 503 |
#undef A1 |
| 504 |
#undef A2 |
| 505 |
#undef A3 |
| 506 |
#undef B0 |
| 507 |
#undef B1 |
| 508 |
#undef B2 |
| 509 |
#undef B3 |
| 510 |
#undef C0 |
| 511 |
#undef C1 |
| 512 |
#undef C2 |
| 513 |
#undef C3 |
| 514 |
|
| 515 |
#else /* __AVR_HAVE_MUL__ */ |
| 516 |
|
| 517 |
#define A0 18 |
| 518 |
#define A1 A0+1 |
| 519 |
#define A2 A0+2 |
| 520 |
#define A3 A0+3 |
| 521 |
|
| 522 |
#define B0 22 |
| 523 |
#define B1 B0+1 |
| 524 |
#define B2 B0+2 |
| 525 |
#define B3 B0+3 |
| 526 |
|
| 527 |
#define C0 22 |
| 528 |
#define C1 C0+1 |
| 529 |
#define C2 C0+2 |
| 530 |
#define C3 C0+3 |
| 531 |
|
| 532 |
;; __tmp_reg__ |
| 533 |
#define CC0 0 |
| 534 |
;; __zero_reg__ |
| 535 |
#define CC1 1 |
| 536 |
#define CC2 16 |
| 537 |
#define CC3 17 |
| 538 |
|
| 539 |
#define AA0 26 |
| 540 |
#define AA1 AA0+1 |
| 541 |
#define AA2 30 |
| 542 |
#define AA3 AA2+1 |
| 543 |
|
| 544 |
#if defined (L_mulsa3) |
| 545 |
;;; (R25:R22) *= (R21:R18) |
| 546 |
;;; Clobbers: ABI, called by optabs |
| 547 |
;;; Rounding: -1 LSB <= error <= 1 LSB |
| 548 |
DEFUN __mulsa3 |
| 549 |
push B0 |
| 550 |
push B1 |
| 551 |
push B3 |
| 552 |
clt |
| 553 |
XCALL __mulusa3_round |
| 554 |
pop r30 |
| 555 |
;; sign-extend B |
| 556 |
bst r30, 7 |
| 557 |
brtc 1f |
| 558 |
;; A1, A0 survived in R27:R26 |
| 559 |
sub C2, AA0 |
| 560 |
sbc C3, AA1 |
| 561 |
1: |
| 562 |
pop AA1 ;; B1 |
| 563 |
pop AA0 ;; B0 |
| 564 |
|
| 565 |
;; sign-extend A. A3 survived in R31 |
| 566 |
bst AA3, 7 |
| 567 |
brtc 2f |
| 568 |
sub C2, AA0 |
| 569 |
sbc C3, AA1 |
| 570 |
2: |
| 571 |
;; Shift 1 bit left to adjust for 15 fractional bits |
| 572 |
lsl GUARD |
| 573 |
rol C0 |
| 574 |
rol C1 |
| 575 |
rol C2 |
| 576 |
rol C3 |
| 577 |
;; Round last digit |
| 578 |
lsl GUARD |
| 579 |
adc C0, __zero_reg__ |
| 580 |
adc C1, __zero_reg__ |
| 581 |
adc C2, __zero_reg__ |
| 582 |
adc C3, __zero_reg__ |
| 583 |
ret |
| 584 |
ENDF __mulsa3 |
| 585 |
#endif /* L_mulsa3 */ |
| 586 |
|
| 587 |
#if defined (L_mulusa3) |
| 588 |
;;; (R25:R22) *= (R21:R18) |
| 589 |
;;; Clobbers: ABI, called by optabs |
| 590 |
;;; Rounding: -1 LSB <= error <= 1 LSB |
| 591 |
DEFUN __mulusa3 |
| 592 |
set |
| 593 |
;; Fallthru |
| 594 |
ENDF __mulusa3 |
| 595 |
|
| 596 |
;;; A[] survives in 26, 27, 30, 31 |
| 597 |
;;; Also used by __mulsa3 with T = 0 |
| 598 |
;;; Round if T = 1 |
| 599 |
;;; Return Guard bits in GUARD (__tmp_reg__), used by signed version. |
| 600 |
DEFUN __mulusa3_round |
| 601 |
push CC2 |
| 602 |
push CC3 |
| 603 |
; clear result |
| 604 |
clr __tmp_reg__ |
| 605 |
wmov CC2, CC0 |
| 606 |
; save multiplicand |
| 607 |
wmov AA0, A0 |
| 608 |
wmov AA2, A2 |
| 609 |
rjmp 3f |
| 610 |
|
| 611 |
;; Loop the integral part |
| 612 |
|
| 613 |
1: ;; CC += A * 2^n; n >= 0 |
| 614 |
add CC0,A0 $ adc CC1,A1 $ adc CC2,A2 $ adc CC3,A3 |
| 615 |
|
| 616 |
2: ;; A <<= 1 |
| 617 |
lsl A0 $ rol A1 $ rol A2 $ rol A3 |
| 618 |
|
| 619 |
3: ;; IBIT(B) >>= 1 |
| 620 |
;; Carry = n-th bit of B; n >= 0 |
| 621 |
lsr B3 |
| 622 |
ror B2 |
| 623 |
brcs 1b |
| 624 |
sbci B3, 0 |
| 625 |
brne 2b |
| 626 |
|
| 627 |
;; Loop the fractional part |
| 628 |
;; B2/B3 is 0 now, use as guard bits for rounding |
| 629 |
;; Restore multiplicand |
| 630 |
wmov A0, AA0 |
| 631 |
wmov A2, AA2 |
| 632 |
rjmp 5f |
| 633 |
|
| 634 |
4: ;; CC += A:Guard * 2^n; n < 0 |
| 635 |
add B3,B2 $ adc CC0,A0 $ adc CC1,A1 $ adc CC2,A2 $ adc CC3,A3 |
| 636 |
5: |
| 637 |
;; A:Guard >>= 1 |
| 638 |
lsr A3 $ ror A2 $ ror A1 $ ror A0 $ ror B2 |
| 639 |
|
| 640 |
;; FBIT(B) <<= 1 |
| 641 |
;; Carry = n-th bit of B; n < 0 |
| 642 |
lsl B0 |
| 643 |
rol B1 |
| 644 |
brcs 4b |
| 645 |
sbci B0, 0 |
| 646 |
brne 5b |
| 647 |
|
| 648 |
;; Save guard bits and set carry for rounding |
| 649 |
push B3 |
| 650 |
lsl B3 |
| 651 |
;; Move result into place |
| 652 |
wmov C2, CC2 |
| 653 |
wmov C0, CC0 |
| 654 |
clr __zero_reg__ |
| 655 |
brtc 6f |
| 656 |
;; Round iff T = 1 |
| 657 |
adc C0, __zero_reg__ |
| 658 |
adc C1, __zero_reg__ |
| 659 |
adc C2, __zero_reg__ |
| 660 |
adc C3, __zero_reg__ |
| 661 |
6: |
| 662 |
pop GUARD |
| 663 |
;; Epilogue |
| 664 |
pop CC3 |
| 665 |
pop CC2 |
| 666 |
ret |
| 667 |
ENDF __mulusa3_round |
| 668 |
#endif /* L_mulusa3 */ |
| 669 |
|
| 670 |
#undef A0 |
| 671 |
#undef A1 |
| 672 |
#undef A2 |
| 673 |
#undef A3 |
| 674 |
#undef B0 |
| 675 |
#undef B1 |
| 676 |
#undef B2 |
| 677 |
#undef B3 |
| 678 |
#undef C0 |
| 679 |
#undef C1 |
| 680 |
#undef C2 |
| 681 |
#undef C3 |
| 682 |
#undef AA0 |
| 683 |
#undef AA1 |
| 684 |
#undef AA2 |
| 685 |
#undef AA3 |
| 686 |
#undef CC0 |
| 687 |
#undef CC1 |
| 688 |
#undef CC2 |
| 689 |
#undef CC3 |
| 690 |
|
| 691 |
#endif /* __AVR_HAVE_MUL__ */ |
| 692 |
|
| 693 |
#undef GUARD |
| 694 |
|
| 695 |
#endif /* ! AVR_TINY */ |
| 696 |
|
| 697 |
/*********************************************************** |
| 698 |
Fixed unsigned saturated Multiplication 8.8 x 8.8 |
| 699 |
***********************************************************/ |
| 700 |
|
| 701 |
#define C0 22 |
| 702 |
#define C1 C0+1 |
| 703 |
#define C2 C0+2 |
| 704 |
#define C3 C0+3 |
| 705 |
#define SS __tmp_reg__ |
| 706 |
|
| 707 |
#if defined (L_usmuluha3) |
| 708 |
DEFUN __usmuluha3 |
| 709 |
;; Widening multiply |
| 710 |
#ifdef __AVR_HAVE_MUL__ |
| 711 |
;; Adjust interface |
| 712 |
movw R26, R22 |
| 713 |
movw R18, R24 |
| 714 |
#endif /* HAVE MUL */ |
| 715 |
XCALL __umulhisi3 |
| 716 |
tst C3 |
| 717 |
brne .Lmax |
| 718 |
;; Round, target is in C1..C2 |
| 719 |
lsl C0 |
| 720 |
adc C1, __zero_reg__ |
| 721 |
adc C2, __zero_reg__ |
| 722 |
brcs .Lmax |
| 723 |
;; Move result into place |
| 724 |
mov C3, C2 |
| 725 |
mov C2, C1 |
| 726 |
ret |
| 727 |
.Lmax: |
| 728 |
;; Saturate |
| 729 |
ldi C2, 0xff |
| 730 |
ldi C3, 0xff |
| 731 |
ret |
| 732 |
ENDF __usmuluha3 |
| 733 |
#endif /* L_usmuluha3 */ |
| 734 |
|
| 735 |
/*********************************************************** |
| 736 |
Fixed signed saturated Multiplication s8.7 x s8.7 |
| 737 |
***********************************************************/ |
| 738 |
|
| 739 |
#if defined (L_ssmulha3) |
| 740 |
DEFUN __ssmulha3 |
| 741 |
;; Widening multiply |
| 742 |
#ifdef __AVR_HAVE_MUL__ |
| 743 |
;; Adjust interface |
| 744 |
movw R26, R22 |
| 745 |
movw R18, R24 |
| 746 |
#endif /* HAVE MUL */ |
| 747 |
XCALL __mulhisi3 |
| 748 |
;; Adjust decimal point |
| 749 |
lsl C0 |
| 750 |
rol C1 |
| 751 |
rol C2 |
| 752 |
brvs .LsatC3.3 |
| 753 |
;; The 9 MSBs must be the same |
| 754 |
rol C3 |
| 755 |
sbc SS, SS |
| 756 |
cp C3, SS |
| 757 |
brne .LsatSS |
| 758 |
;; Round |
| 759 |
lsl C0 |
| 760 |
adc C1, __zero_reg__ |
| 761 |
adc C2, __zero_reg__ |
| 762 |
brvs .Lmax |
| 763 |
;; Move result into place |
| 764 |
mov C3, C2 |
| 765 |
mov C2, C1 |
| 766 |
ret |
| 767 |
.Lmax: |
| 768 |
;; Load 0x7fff |
| 769 |
clr C3 |
| 770 |
.LsatC3.3: |
| 771 |
;; C3 < 0 --> 0x8000 |
| 772 |
;; C3 >= 0 --> 0x7fff |
| 773 |
mov SS, C3 |
| 774 |
.LsatSS: |
| 775 |
;; Load min / max value: |
| 776 |
;; SS = -1 --> 0x8000 |
| 777 |
;; SS = 0 --> 0x7fff |
| 778 |
ldi C3, 0x7f |
| 779 |
ldi C2, 0xff |
| 780 |
#ifndef __AVR_TINY__ |
| 781 |
sbrc SS, 7 |
| 782 |
adiw C2, 1 |
| 783 |
#else |
| 784 |
lsl SS |
| 785 |
adc C2, __zero_reg__ |
| 786 |
adc C3, __zero_reg__ |
| 787 |
#endif /* AVR_TINY? */ |
| 788 |
ret |
| 789 |
ENDF __ssmulha3 |
| 790 |
#endif /* L_ssmulha3 */ |
| 791 |
|
| 792 |
#undef C0 |
| 793 |
#undef C1 |
| 794 |
#undef C2 |
| 795 |
#undef C3 |
| 796 |
#undef SS |
| 797 |
|
| 798 |
/*********************************************************** |
| 799 |
Fixed unsigned saturated Multiplication 16.16 x 16.16 |
| 800 |
***********************************************************/ |
| 801 |
|
| 802 |
#ifndef __AVR_TINY__ |
| 803 |
|
| 804 |
#define C0 18 |
| 805 |
#define C1 C0+1 |
| 806 |
#define C2 C0+2 |
| 807 |
#define C3 C0+3 |
| 808 |
#define C4 C0+4 |
| 809 |
#define C5 C0+5 |
| 810 |
#define C6 C0+6 |
| 811 |
#define C7 C0+7 |
| 812 |
#define SS __tmp_reg__ |
| 813 |
|
| 814 |
#if defined (L_usmulusa3) |
| 815 |
;; R22[4] = R22[4] *{ssat} R18[4] |
| 816 |
;; Ordinary ABI function |
| 817 |
DEFUN __usmulusa3 |
| 818 |
;; Widening multiply |
| 819 |
XCALL __umulsidi3 |
| 820 |
or C7, C6 |
| 821 |
brne .Lmax |
| 822 |
;; Round, target is in C2..C5 |
| 823 |
lsl C1 |
| 824 |
adc C2, __zero_reg__ |
| 825 |
adc C3, __zero_reg__ |
| 826 |
adc C4, __zero_reg__ |
| 827 |
adc C5, __zero_reg__ |
| 828 |
brcs .Lmax |
| 829 |
;; Move result into place |
| 830 |
wmov C6, C4 |
| 831 |
wmov C4, C2 |
| 832 |
ret |
| 833 |
.Lmax: |
| 834 |
;; Saturate |
| 835 |
ldi C7, 0xff |
| 836 |
ldi C6, 0xff |
| 837 |
wmov C4, C6 |
| 838 |
ret |
| 839 |
ENDF __usmulusa3 |
| 840 |
#endif /* L_usmulusa3 */ |
| 841 |
|
| 842 |
/*********************************************************** |
| 843 |
Fixed signed saturated Multiplication s16.15 x s16.15 |
| 844 |
***********************************************************/ |
| 845 |
|
| 846 |
#if defined (L_ssmulsa3) |
| 847 |
;; R22[4] = R22[4] *{ssat} R18[4] |
| 848 |
;; Ordinary ABI function |
| 849 |
DEFUN __ssmulsa3 |
| 850 |
;; Widening multiply |
| 851 |
XCALL __mulsidi3 |
| 852 |
;; Adjust decimal point |
| 853 |
lsl C1 |
| 854 |
rol C2 |
| 855 |
rol C3 |
| 856 |
rol C4 |
| 857 |
rol C5 |
| 858 |
brvs .LsatC7.7 |
| 859 |
;; The 17 MSBs must be the same |
| 860 |
rol C6 |
| 861 |
rol C7 |
| 862 |
sbc SS, SS |
| 863 |
cp C6, SS |
| 864 |
cpc C7, SS |
| 865 |
brne .LsatSS |
| 866 |
;; Round |
| 867 |
lsl C1 |
| 868 |
adc C2, __zero_reg__ |
| 869 |
adc C3, __zero_reg__ |
| 870 |
adc C4, __zero_reg__ |
| 871 |
adc C5, __zero_reg__ |
| 872 |
brvs .Lmax |
| 873 |
;; Move result into place |
| 874 |
wmov C6, C4 |
| 875 |
wmov C4, C2 |
| 876 |
ret |
| 877 |
|
| 878 |
.Lmax: |
| 879 |
;; Load 0x7fffffff |
| 880 |
clr C7 |
| 881 |
.LsatC7.7: |
| 882 |
;; C7 < 0 --> 0x80000000 |
| 883 |
;; C7 >= 0 --> 0x7fffffff |
| 884 |
lsl C7 |
| 885 |
sbc SS, SS |
| 886 |
.LsatSS: |
| 887 |
;; Load min / max value: |
| 888 |
;; SS = -1 --> 0x80000000 |
| 889 |
;; SS = 0 --> 0x7fffffff |
| 890 |
com SS |
| 891 |
mov C4, SS |
| 892 |
mov C5, C4 |
| 893 |
wmov C6, C4 |
| 894 |
subi C7, 0x80 |
| 895 |
ret |
| 896 |
ENDF __ssmulsa3 |
| 897 |
#endif /* L_ssmulsa3 */ |
| 898 |
|
| 899 |
#undef C0 |
| 900 |
#undef C1 |
| 901 |
#undef C2 |
| 902 |
#undef C3 |
| 903 |
#undef C4 |
| 904 |
#undef C5 |
| 905 |
#undef C6 |
| 906 |
#undef C7 |
| 907 |
#undef SS |
| 908 |
|
| 909 |
#endif /* ! AVR_TINY */ |
| 910 |
|
| 911 |
/******************************************************* |
| 912 |
Fractional Division 8 / 8 |
| 913 |
*******************************************************/ |
| 914 |
|
| 915 |
#define r_divd r25 /* dividend */ |
| 916 |
#define r_quo r24 /* quotient */ |
| 917 |
#define r_div r22 /* divisor */ |
| 918 |
#define r_sign __tmp_reg__ |
| 919 |
|
| 920 |
#if defined (L_divqq3) |
| 921 |
DEFUN __divqq3 |
| 922 |
mov r_sign, r_divd |
| 923 |
eor r_sign, r_div |
| 924 |
sbrc r_div, 7 |
| 925 |
neg r_div |
| 926 |
sbrc r_divd, 7 |
| 927 |
neg r_divd |
| 928 |
XCALL __divqq_helper |
| 929 |
lsr r_quo |
| 930 |
sbrc r_sign, 7 ; negate result if needed |
| 931 |
neg r_quo |
| 932 |
ret |
| 933 |
ENDF __divqq3 |
| 934 |
#endif /* L_divqq3 */ |
| 935 |
|
| 936 |
#if defined (L_udivuqq3) |
| 937 |
DEFUN __udivuqq3 |
| 938 |
cp r_divd, r_div |
| 939 |
brsh 0f |
| 940 |
XJMP __divqq_helper |
| 941 |
;; Result is out of [0, 1) ==> Return 1 - eps. |
| 942 |
0: ldi r_quo, 0xff |
| 943 |
ret |
| 944 |
ENDF __udivuqq3 |
| 945 |
#endif /* L_udivuqq3 */ |
| 946 |
|
| 947 |
|
| 948 |
#if defined (L_divqq_helper) |
| 949 |
DEFUN __divqq_helper |
| 950 |
clr r_quo ; clear quotient |
| 951 |
inc __zero_reg__ ; init loop counter, used per shift |
| 952 |
__udivuqq3_loop: |
| 953 |
lsl r_divd ; shift dividend |
| 954 |
brcs 0f ; dividend overflow |
| 955 |
cp r_divd,r_div ; compare dividend & divisor |
| 956 |
brcc 0f ; dividend >= divisor |
| 957 |
rol r_quo ; shift quotient (with CARRY) |
| 958 |
rjmp __udivuqq3_cont |
| 959 |
0: |
| 960 |
sub r_divd,r_div ; restore dividend |
| 961 |
lsl r_quo ; shift quotient (without CARRY) |
| 962 |
__udivuqq3_cont: |
| 963 |
lsl __zero_reg__ ; shift loop-counter bit |
| 964 |
brne __udivuqq3_loop |
| 965 |
com r_quo ; complement result |
| 966 |
; because C flag was complemented in loop |
| 967 |
ret |
| 968 |
ENDF __divqq_helper |
| 969 |
#endif /* L_divqq_helper */ |
| 970 |
|
| 971 |
#undef r_divd |
| 972 |
#undef r_quo |
| 973 |
#undef r_div |
| 974 |
#undef r_sign |
| 975 |
|
| 976 |
|
| 977 |
/******************************************************* |
| 978 |
Fractional Division 16 / 16 |
| 979 |
*******************************************************/ |
| 980 |
#define r_divdL 26 /* dividend Low */ |
| 981 |
#define r_divdH 27 /* dividend Hig */ |
| 982 |
#define r_quoL 24 /* quotient Low */ |
| 983 |
#define r_quoH 25 /* quotient High */ |
| 984 |
#define r_divL 22 /* divisor */ |
| 985 |
#define r_divH 23 /* divisor */ |
| 986 |
#define r_cnt 21 |
| 987 |
|
| 988 |
#if defined (L_divhq3) |
| 989 |
DEFUN __divhq3 |
| 990 |
mov __tmp_reg__, r_divdH |
| 991 |
eor __tmp_reg__, r_divH |
| 992 |
sbrs r_divH, 7 |
| 993 |
rjmp 1f |
| 994 |
NEG2 r_divL |
| 995 |
1: |
| 996 |
sbrs r_divdH, 7 |
| 997 |
rjmp 2f |
| 998 |
NEG2 r_divdL |
| 999 |
2: |
| 1000 |
cp r_divdL, r_divL |
| 1001 |
cpc r_divdH, r_divH |
| 1002 |
breq __divhq3_minus1 ; if equal return -1 |
| 1003 |
XCALL __udivuhq3 |
| 1004 |
lsr r_quoH |
| 1005 |
ror r_quoL |
| 1006 |
brpl 9f |
| 1007 |
;; negate result if needed |
| 1008 |
NEG2 r_quoL |
| 1009 |
9: |
| 1010 |
ret |
| 1011 |
__divhq3_minus1: |
| 1012 |
ldi r_quoH, 0x80 |
| 1013 |
clr r_quoL |
| 1014 |
ret |
| 1015 |
ENDF __divhq3 |
| 1016 |
#endif /* defined (L_divhq3) */ |
| 1017 |
|
| 1018 |
#if defined (L_udivuhq3) |
| 1019 |
DEFUN __udivuhq3 |
| 1020 |
sub r_quoH,r_quoH ; clear quotient and carry |
| 1021 |
;; FALLTHRU |
| 1022 |
ENDF __udivuhq3 |
| 1023 |
|
| 1024 |
DEFUN __udivuha3_common |
| 1025 |
clr r_quoL ; clear quotient |
| 1026 |
ldi r_cnt,16 ; init loop counter |
| 1027 |
__udivuhq3_loop: |
| 1028 |
rol r_divdL ; shift dividend (with CARRY) |
| 1029 |
rol r_divdH |
| 1030 |
brcs __udivuhq3_ep ; dividend overflow |
| 1031 |
cp r_divdL,r_divL ; compare dividend & divisor |
| 1032 |
cpc r_divdH,r_divH |
| 1033 |
brcc __udivuhq3_ep ; dividend >= divisor |
| 1034 |
rol r_quoL ; shift quotient (with CARRY) |
| 1035 |
rjmp __udivuhq3_cont |
| 1036 |
__udivuhq3_ep: |
| 1037 |
sub r_divdL,r_divL ; restore dividend |
| 1038 |
sbc r_divdH,r_divH |
| 1039 |
lsl r_quoL ; shift quotient (without CARRY) |
| 1040 |
__udivuhq3_cont: |
| 1041 |
rol r_quoH ; shift quotient |
| 1042 |
dec r_cnt ; decrement loop counter |
| 1043 |
brne __udivuhq3_loop |
| 1044 |
com r_quoL ; complement result |
| 1045 |
com r_quoH ; because C flag was complemented in loop |
| 1046 |
ret |
| 1047 |
ENDF __udivuha3_common |
| 1048 |
#endif /* defined (L_udivuhq3) */ |
| 1049 |
|
| 1050 |
/******************************************************* |
| 1051 |
Fixed Division 8.8 / 8.8 |
| 1052 |
*******************************************************/ |
| 1053 |
#if defined (L_divha3) |
| 1054 |
DEFUN __divha3 |
| 1055 |
mov __tmp_reg__, r_divdH |
| 1056 |
eor __tmp_reg__, r_divH |
| 1057 |
sbrs r_divH, 7 |
| 1058 |
rjmp 1f |
| 1059 |
NEG2 r_divL |
| 1060 |
1: |
| 1061 |
sbrs r_divdH, 7 |
| 1062 |
rjmp 2f |
| 1063 |
NEG2 r_divdL |
| 1064 |
2: |
| 1065 |
XCALL __udivuha3 |
| 1066 |
lsr r_quoH ; adjust to 7 fractional bits |
| 1067 |
ror r_quoL |
| 1068 |
sbrs __tmp_reg__, 7 ; negate result if needed |
| 1069 |
ret |
| 1070 |
NEG2 r_quoL |
| 1071 |
ret |
| 1072 |
ENDF __divha3 |
| 1073 |
#endif /* defined (L_divha3) */ |
| 1074 |
|
| 1075 |
#if defined (L_udivuha3) |
| 1076 |
DEFUN __udivuha3 |
| 1077 |
mov r_quoH, r_divdL |
| 1078 |
mov r_divdL, r_divdH |
| 1079 |
clr r_divdH |
| 1080 |
lsl r_quoH ; shift quotient into carry |
| 1081 |
XJMP __udivuha3_common ; same as fractional after rearrange |
| 1082 |
ENDF __udivuha3 |
| 1083 |
#endif /* defined (L_udivuha3) */ |
| 1084 |
|
| 1085 |
#undef r_divdL |
| 1086 |
#undef r_divdH |
| 1087 |
#undef r_quoL |
| 1088 |
#undef r_quoH |
| 1089 |
#undef r_divL |
| 1090 |
#undef r_divH |
| 1091 |
#undef r_cnt |
| 1092 |
|
| 1093 |
/******************************************************* |
| 1094 |
Fixed Division 16.16 / 16.16 |
| 1095 |
*******************************************************/ |
| 1096 |
|
| 1097 |
#ifndef __AVR_TINY__ |
| 1098 |
|
| 1099 |
#define r_arg1L 24 /* arg1 gets passed already in place */ |
| 1100 |
#define r_arg1H 25 |
| 1101 |
#define r_arg1HL 26 |
| 1102 |
#define r_arg1HH 27 |
| 1103 |
#define r_divdL 26 /* dividend Low */ |
| 1104 |
#define r_divdH 27 |
| 1105 |
#define r_divdHL 30 |
| 1106 |
#define r_divdHH 31 /* dividend High */ |
| 1107 |
#define r_quoL 22 /* quotient Low */ |
| 1108 |
#define r_quoH 23 |
| 1109 |
#define r_quoHL 24 |
| 1110 |
#define r_quoHH 25 /* quotient High */ |
| 1111 |
#define r_divL 18 /* divisor Low */ |
| 1112 |
#define r_divH 19 |
| 1113 |
#define r_divHL 20 |
| 1114 |
#define r_divHH 21 /* divisor High */ |
| 1115 |
#define r_cnt __zero_reg__ /* loop count (0 after the loop!) */ |
| 1116 |
|
| 1117 |
#if defined (L_divsa3) |
| 1118 |
DEFUN __divsa3 |
| 1119 |
mov r0, r_arg1HH |
| 1120 |
eor r0, r_divHH |
| 1121 |
sbrs r_divHH, 7 |
| 1122 |
rjmp 1f |
| 1123 |
NEG4 r_divL |
| 1124 |
1: |
| 1125 |
sbrs r_arg1HH, 7 |
| 1126 |
rjmp 2f |
| 1127 |
NEG4 r_arg1L |
| 1128 |
2: |
| 1129 |
XCALL __udivusa3 |
| 1130 |
lsr r_quoHH ; adjust to 15 fractional bits |
| 1131 |
ror r_quoHL |
| 1132 |
ror r_quoH |
| 1133 |
ror r_quoL |
| 1134 |
sbrs r0, 7 ; negate result if needed |
| 1135 |
ret |
| 1136 |
;; negate r_quoL |
| 1137 |
XJMP __negsi2 |
| 1138 |
ENDF __divsa3 |
| 1139 |
#endif /* defined (L_divsa3) */ |
| 1140 |
|
| 1141 |
#if defined (L_udivusa3) |
| 1142 |
DEFUN __udivusa3 |
| 1143 |
ldi r_divdHL, 32 ; init loop counter |
| 1144 |
mov r_cnt, r_divdHL |
| 1145 |
clr r_divdHL |
| 1146 |
clr r_divdHH |
| 1147 |
wmov r_quoL, r_divdHL |
| 1148 |
lsl r_quoHL ; shift quotient into carry |
| 1149 |
rol r_quoHH |
| 1150 |
__udivusa3_loop: |
| 1151 |
rol r_divdL ; shift dividend (with CARRY) |
| 1152 |
rol r_divdH |
| 1153 |
rol r_divdHL |
| 1154 |
rol r_divdHH |
| 1155 |
brcs __udivusa3_ep ; dividend overflow |
| 1156 |
cp r_divdL,r_divL ; compare dividend & divisor |
| 1157 |
cpc r_divdH,r_divH |
| 1158 |
cpc r_divdHL,r_divHL |
| 1159 |
cpc r_divdHH,r_divHH |
| 1160 |
brcc __udivusa3_ep ; dividend >= divisor |
| 1161 |
rol r_quoL ; shift quotient (with CARRY) |
| 1162 |
rjmp __udivusa3_cont |
| 1163 |
__udivusa3_ep: |
| 1164 |
sub r_divdL,r_divL ; restore dividend |
| 1165 |
sbc r_divdH,r_divH |
| 1166 |
sbc r_divdHL,r_divHL |
| 1167 |
sbc r_divdHH,r_divHH |
| 1168 |
lsl r_quoL ; shift quotient (without CARRY) |
| 1169 |
__udivusa3_cont: |
| 1170 |
rol r_quoH ; shift quotient |
| 1171 |
rol r_quoHL |
| 1172 |
rol r_quoHH |
| 1173 |
dec r_cnt ; decrement loop counter |
| 1174 |
brne __udivusa3_loop |
| 1175 |
com r_quoL ; complement result |
| 1176 |
com r_quoH ; because C flag was complemented in loop |
| 1177 |
com r_quoHL |
| 1178 |
com r_quoHH |
| 1179 |
ret |
| 1180 |
ENDF __udivusa3 |
| 1181 |
#endif /* defined (L_udivusa3) */ |
| 1182 |
|
| 1183 |
#undef r_arg1L |
| 1184 |
#undef r_arg1H |
| 1185 |
#undef r_arg1HL |
| 1186 |
#undef r_arg1HH |
| 1187 |
#undef r_divdL |
| 1188 |
#undef r_divdH |
| 1189 |
#undef r_divdHL |
| 1190 |
#undef r_divdHH |
| 1191 |
#undef r_quoL |
| 1192 |
#undef r_quoH |
| 1193 |
#undef r_quoHL |
| 1194 |
#undef r_quoHH |
| 1195 |
#undef r_divL |
| 1196 |
#undef r_divH |
| 1197 |
#undef r_divHL |
| 1198 |
#undef r_divHH |
| 1199 |
#undef r_cnt |
| 1200 |
|
| 1201 |
#endif /* ! AVR_TINY */ |
| 1202 |
|
| 1203 |
|
| 1204 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1205 |
;; Saturation, 1 Byte |
| 1206 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1207 |
|
| 1208 |
;; First Argument and Return Register |
| 1209 |
#define A0 24 |
| 1210 |
|
| 1211 |
#if defined (L_ssabs_1) |
| 1212 |
DEFUN __ssabs_1 |
| 1213 |
sbrs A0, 7 |
| 1214 |
ret |
| 1215 |
neg A0 |
| 1216 |
sbrc A0,7 |
| 1217 |
dec A0 |
| 1218 |
ret |
| 1219 |
ENDF __ssabs_1 |
| 1220 |
#endif /* L_ssabs_1 */ |
| 1221 |
|
| 1222 |
#undef A0 |
| 1223 |
|
| 1224 |
|
| 1225 |
|
| 1226 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1227 |
;; Saturation, 2 Bytes |
| 1228 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1229 |
|
| 1230 |
;; First Argument and Return Register |
| 1231 |
#define A0 24 |
| 1232 |
#define A1 A0+1 |
| 1233 |
|
| 1234 |
#if defined (L_ssneg_2) |
| 1235 |
DEFUN __ssneg_2 |
| 1236 |
NEG2 A0 |
| 1237 |
brvc 0f |
| 1238 |
wsubi A0, 1 |
| 1239 |
0: ret |
| 1240 |
ENDF __ssneg_2 |
| 1241 |
#endif /* L_ssneg_2 */ |
| 1242 |
|
| 1243 |
#if defined (L_ssabs_2) |
| 1244 |
DEFUN __ssabs_2 |
| 1245 |
sbrs A1, 7 |
| 1246 |
ret |
| 1247 |
XJMP __ssneg_2 |
| 1248 |
ENDF __ssabs_2 |
| 1249 |
#endif /* L_ssabs_2 */ |
| 1250 |
|
| 1251 |
#undef A0 |
| 1252 |
#undef A1 |
| 1253 |
|
| 1254 |
|
| 1255 |
|
| 1256 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1257 |
;; Saturation, 4 Bytes |
| 1258 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1259 |
|
| 1260 |
#ifndef __AVR_TINY__ |
| 1261 |
|
| 1262 |
;; First Argument and Return Register |
| 1263 |
#define A0 22 |
| 1264 |
#define A1 A0+1 |
| 1265 |
#define A2 A0+2 |
| 1266 |
#define A3 A0+3 |
| 1267 |
|
| 1268 |
#if defined (L_ssneg_4) |
| 1269 |
DEFUN __ssneg_4 |
| 1270 |
XCALL __negsi2 |
| 1271 |
brvc 0f |
| 1272 |
ldi A3, 0x7f |
| 1273 |
ldi A2, 0xff |
| 1274 |
ldi A1, 0xff |
| 1275 |
ldi A0, 0xff |
| 1276 |
0: ret |
| 1277 |
ENDF __ssneg_4 |
| 1278 |
#endif /* L_ssneg_4 */ |
| 1279 |
|
| 1280 |
#if defined (L_ssabs_4) |
| 1281 |
DEFUN __ssabs_4 |
| 1282 |
sbrs A3, 7 |
| 1283 |
ret |
| 1284 |
XJMP __ssneg_4 |
| 1285 |
ENDF __ssabs_4 |
| 1286 |
#endif /* L_ssabs_4 */ |
| 1287 |
|
| 1288 |
#undef A0 |
| 1289 |
#undef A1 |
| 1290 |
#undef A2 |
| 1291 |
#undef A3 |
| 1292 |
|
| 1293 |
|
| 1294 |
|
| 1295 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1296 |
;; Saturation, 8 Bytes |
| 1297 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1298 |
|
| 1299 |
;; First Argument and Return Register |
| 1300 |
#define A0 18 |
| 1301 |
#define A1 A0+1 |
| 1302 |
#define A2 A0+2 |
| 1303 |
#define A3 A0+3 |
| 1304 |
#define A4 A0+4 |
| 1305 |
#define A5 A0+5 |
| 1306 |
#define A6 A0+6 |
| 1307 |
#define A7 A0+7 |
| 1308 |
|
| 1309 |
#if defined (L_clr_8) |
| 1310 |
FALIAS __usneguta2 |
| 1311 |
FALIAS __usneguda2 |
| 1312 |
FALIAS __usnegudq2 |
| 1313 |
|
| 1314 |
;; Clear Carry and all Bytes |
| 1315 |
DEFUN __clr_8 |
| 1316 |
;; Clear Carry and set Z |
| 1317 |
sub A7, A7 |
| 1318 |
;; FALLTHRU |
| 1319 |
ENDF __clr_8 |
| 1320 |
;; Propagate Carry to all Bytes, Carry unaltered |
| 1321 |
DEFUN __sbc_8 |
| 1322 |
sbc A7, A7 |
| 1323 |
sbc A6, A6 |
| 1324 |
wmov A4, A6 |
| 1325 |
wmov A2, A6 |
| 1326 |
wmov A0, A6 |
| 1327 |
ret |
| 1328 |
ENDF __sbc_8 |
| 1329 |
#endif /* L_clr_8 */ |
| 1330 |
|
| 1331 |
#if defined (L_ssneg_8) |
| 1332 |
FALIAS __ssnegta2 |
| 1333 |
FALIAS __ssnegda2 |
| 1334 |
FALIAS __ssnegdq2 |
| 1335 |
|
| 1336 |
DEFUN __ssneg_8 |
| 1337 |
XCALL __negdi2 |
| 1338 |
brvc 0f |
| 1339 |
;; A[] = 0x7fffffff |
| 1340 |
sec |
| 1341 |
XCALL __sbc_8 |
| 1342 |
ldi A7, 0x7f |
| 1343 |
0: ret |
| 1344 |
ENDF __ssneg_8 |
| 1345 |
#endif /* L_ssneg_8 */ |
| 1346 |
|
| 1347 |
#if defined (L_ssabs_8) |
| 1348 |
FALIAS __ssabsta2 |
| 1349 |
FALIAS __ssabsda2 |
| 1350 |
FALIAS __ssabsdq2 |
| 1351 |
|
| 1352 |
DEFUN __ssabs_8 |
| 1353 |
sbrs A7, 7 |
| 1354 |
ret |
| 1355 |
XJMP __ssneg_8 |
| 1356 |
ENDF __ssabs_8 |
| 1357 |
#endif /* L_ssabs_8 */ |
| 1358 |
|
| 1359 |
;; Second Argument |
| 1360 |
#define B0 10 |
| 1361 |
#define B1 B0+1 |
| 1362 |
#define B2 B0+2 |
| 1363 |
#define B3 B0+3 |
| 1364 |
#define B4 B0+4 |
| 1365 |
#define B5 B0+5 |
| 1366 |
#define B6 B0+6 |
| 1367 |
#define B7 B0+7 |
| 1368 |
|
| 1369 |
#if defined (L_usadd_8) |
| 1370 |
FALIAS __usadduta3 |
| 1371 |
FALIAS __usadduda3 |
| 1372 |
FALIAS __usaddudq3 |
| 1373 |
|
| 1374 |
DEFUN __usadd_8 |
| 1375 |
XCALL __adddi3 |
| 1376 |
brcs 0f |
| 1377 |
ret |
| 1378 |
0: ;; A[] = 0xffffffff |
| 1379 |
XJMP __sbc_8 |
| 1380 |
ENDF __usadd_8 |
| 1381 |
#endif /* L_usadd_8 */ |
| 1382 |
|
| 1383 |
#if defined (L_ussub_8) |
| 1384 |
FALIAS __ussubuta3 |
| 1385 |
FALIAS __ussubuda3 |
| 1386 |
FALIAS __ussubudq3 |
| 1387 |
|
| 1388 |
DEFUN __ussub_8 |
| 1389 |
XCALL __subdi3 |
| 1390 |
brcs 0f |
| 1391 |
ret |
| 1392 |
0: ;; A[] = 0 |
| 1393 |
XJMP __clr_8 |
| 1394 |
ENDF __ussub_8 |
| 1395 |
#endif /* L_ussub_8 */ |
| 1396 |
|
| 1397 |
#if defined (L_ssadd_8) |
| 1398 |
FALIAS __ssaddta3 |
| 1399 |
FALIAS __ssaddda3 |
| 1400 |
FALIAS __ssadddq3 |
| 1401 |
|
| 1402 |
DEFUN __ssadd_8 |
| 1403 |
XCALL __adddi3 |
| 1404 |
brvc 0f |
| 1405 |
;; A = (B >= 0) ? INT64_MAX : INT64_MIN |
| 1406 |
cpi B7, 0x80 |
| 1407 |
XCALL __sbc_8 |
| 1408 |
subi A7, 0x80 |
| 1409 |
0: ret |
| 1410 |
ENDF __ssadd_8 |
| 1411 |
#endif /* L_ssadd_8 */ |
| 1412 |
|
| 1413 |
#if defined (L_sssub_8) |
| 1414 |
FALIAS __sssubta3 |
| 1415 |
FALIAS __sssubda3 |
| 1416 |
FALIAS __sssubdq3 |
| 1417 |
|
| 1418 |
DEFUN __sssub_8 |
| 1419 |
XCALL __subdi3 |
| 1420 |
brvc 0f |
| 1421 |
;; A = (B < 0) ? INT64_MAX : INT64_MIN |
| 1422 |
ldi A7, 0x7f |
| 1423 |
cp A7, B7 |
| 1424 |
XCALL __sbc_8 |
| 1425 |
subi A7, 0x80 |
| 1426 |
0: ret |
| 1427 |
ENDF __sssub_8 |
| 1428 |
#endif /* L_sssub_8 */ |
| 1429 |
|
| 1430 |
#undef A0 |
| 1431 |
#undef A1 |
| 1432 |
#undef A2 |
| 1433 |
#undef A3 |
| 1434 |
#undef A4 |
| 1435 |
#undef A5 |
| 1436 |
#undef A6 |
| 1437 |
#undef A7 |
| 1438 |
#undef B0 |
| 1439 |
#undef B1 |
| 1440 |
#undef B2 |
| 1441 |
#undef B3 |
| 1442 |
#undef B4 |
| 1443 |
#undef B5 |
| 1444 |
#undef B6 |
| 1445 |
#undef B7 |
| 1446 |
|
| 1447 |
#endif /* ! AVR_TINY */ |
| 1448 |
|
| 1449 |
|
| 1450 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1451 |
;; Rounding Helpers |
| 1452 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1453 |
|
| 1454 |
#ifdef L_mask1 |
| 1455 |
|
| 1456 |
#define AA 24 |
| 1457 |
#define CC 25 |
| 1458 |
|
| 1459 |
;; R25 = 1 << (R24 & 7) |
| 1460 |
;; CC = 1 << (AA & 7) |
| 1461 |
;; Clobbers: None |
| 1462 |
DEFUN __mask1 |
| 1463 |
;; CC = 2 ^ AA.1 |
| 1464 |
ldi CC, 1 << 2 |
| 1465 |
sbrs AA, 1 |
| 1466 |
ldi CC, 1 << 0 |
| 1467 |
;; CC *= 2 ^ AA.0 |
| 1468 |
sbrc AA, 0 |
| 1469 |
lsl CC |
| 1470 |
;; CC *= 2 ^ AA.2 |
| 1471 |
sbrc AA, 2 |
| 1472 |
swap CC |
| 1473 |
ret |
| 1474 |
ENDF __mask1 |
| 1475 |
|
| 1476 |
#undef AA |
| 1477 |
#undef CC |
| 1478 |
#endif /* L_mask1 */ |
| 1479 |
|
| 1480 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1481 |
|
| 1482 |
;; The rounding point. Any bits smaller than |
| 1483 |
;; 2^{-RP} will be cleared. |
| 1484 |
#define RP R24 |
| 1485 |
|
| 1486 |
#define A0 22 |
| 1487 |
#define A1 A0 + 1 |
| 1488 |
|
| 1489 |
#define C0 24 |
| 1490 |
#define C1 C0 + 1 |
| 1491 |
|
| 1492 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1493 |
;; Rounding, 1 Byte |
| 1494 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1495 |
|
| 1496 |
#ifdef L_roundqq3 |
| 1497 |
|
| 1498 |
;; R24 = round (R22, R24) |
| 1499 |
;; Clobbers: R22, __tmp_reg__ |
| 1500 |
DEFUN __roundqq3 |
| 1501 |
mov __tmp_reg__, C1 |
| 1502 |
subi RP, __QQ_FBIT__ - 1 |
| 1503 |
neg RP |
| 1504 |
;; R25 = 1 << RP (Total offset is FBIT-1 - RP) |
| 1505 |
XCALL __mask1 |
| 1506 |
mov C0, C1 |
| 1507 |
;; Add-Saturate 2^{-RP-1} |
| 1508 |
add A0, C0 |
| 1509 |
brvc 0f |
| 1510 |
ldi C0, 0x7f |
| 1511 |
rjmp 9f |
| 1512 |
0: ;; Mask out bits beyond RP |
| 1513 |
lsl C0 |
| 1514 |
neg C0 |
| 1515 |
and C0, A0 |
| 1516 |
9: mov C1, __tmp_reg__ |
| 1517 |
ret |
| 1518 |
ENDF __roundqq3 |
| 1519 |
#endif /* L_roundqq3 */ |
| 1520 |
|
| 1521 |
#ifdef L_rounduqq3 |
| 1522 |
|
| 1523 |
;; R24 = round (R22, R24) |
| 1524 |
;; Clobbers: R22, __tmp_reg__ |
| 1525 |
DEFUN __rounduqq3 |
| 1526 |
mov __tmp_reg__, C1 |
| 1527 |
subi RP, __UQQ_FBIT__ - 1 |
| 1528 |
neg RP |
| 1529 |
;; R25 = 1 << RP (Total offset is FBIT-1 - RP) |
| 1530 |
XCALL __mask1 |
| 1531 |
mov C0, C1 |
| 1532 |
;; Add-Saturate 2^{-RP-1} |
| 1533 |
add A0, C0 |
| 1534 |
brcc 0f |
| 1535 |
ldi C0, 0xff |
| 1536 |
rjmp 9f |
| 1537 |
0: ;; Mask out bits beyond RP |
| 1538 |
lsl C0 |
| 1539 |
neg C0 |
| 1540 |
and C0, A0 |
| 1541 |
9: mov C1, __tmp_reg__ |
| 1542 |
ret |
| 1543 |
ENDF __rounduqq3 |
| 1544 |
#endif /* L_rounduqq3 */ |
| 1545 |
|
| 1546 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1547 |
;; Rounding, 2 Bytes |
| 1548 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1549 |
|
| 1550 |
#ifdef L_addmask_2 |
| 1551 |
|
| 1552 |
;; [ R25:R24 = 1 << (R24 & 15) |
| 1553 |
;; R23:R22 += 1 << (R24 & 15) ] |
| 1554 |
;; SREG is set according to the addition |
| 1555 |
DEFUN __addmask_2 |
| 1556 |
;; R25 = 1 << (R24 & 7) |
| 1557 |
XCALL __mask1 |
| 1558 |
cpi RP, 1 << 3 |
| 1559 |
sbc C0, C0 |
| 1560 |
;; Swap C0 and C1 if RP.3 was set |
| 1561 |
and C0, C1 |
| 1562 |
eor C1, C0 |
| 1563 |
;; Finally, add the power-of-two: A[] += C[] |
| 1564 |
add A0, C0 |
| 1565 |
adc A1, C1 |
| 1566 |
ret |
| 1567 |
ENDF __addmask_2 |
| 1568 |
#endif /* L_addmask_2 */ |
| 1569 |
|
| 1570 |
#ifdef L_round_s2 |
| 1571 |
|
| 1572 |
;; R25:R24 = round (R23:R22, R24) |
| 1573 |
;; Clobbers: R23, R22 |
| 1574 |
DEFUN __roundhq3 |
| 1575 |
subi RP, __HQ_FBIT__ - __HA_FBIT__ |
| 1576 |
ENDF __roundhq3 |
| 1577 |
DEFUN __roundha3 |
| 1578 |
subi RP, __HA_FBIT__ - 1 |
| 1579 |
neg RP |
| 1580 |
;; [ R25:R24 = 1 << (FBIT-1 - RP) |
| 1581 |
;; R23:R22 += 1 << (FBIT-1 - RP) ] |
| 1582 |
XCALL __addmask_2 |
| 1583 |
XJMP __round_s2_const |
| 1584 |
ENDF __roundha3 |
| 1585 |
|
| 1586 |
#endif /* L_round_s2 */ |
| 1587 |
|
| 1588 |
#ifdef L_round_u2 |
| 1589 |
|
| 1590 |
;; R25:R24 = round (R23:R22, R24) |
| 1591 |
;; Clobbers: R23, R22 |
| 1592 |
DEFUN __rounduhq3 |
| 1593 |
subi RP, __UHQ_FBIT__ - __UHA_FBIT__ |
| 1594 |
ENDF __rounduhq3 |
| 1595 |
DEFUN __rounduha3 |
| 1596 |
subi RP, __UHA_FBIT__ - 1 |
| 1597 |
neg RP |
| 1598 |
;; [ R25:R24 = 1 << (FBIT-1 - RP) |
| 1599 |
;; R23:R22 += 1 << (FBIT-1 - RP) ] |
| 1600 |
XCALL __addmask_2 |
| 1601 |
XJMP __round_u2_const |
| 1602 |
ENDF __rounduha3 |
| 1603 |
|
| 1604 |
#endif /* L_round_u2 */ |
| 1605 |
|
| 1606 |
|
| 1607 |
#ifdef L_round_2_const |
| 1608 |
|
| 1609 |
;; Helpers for 2 byte wide rounding |
| 1610 |
|
| 1611 |
DEFUN __round_s2_const |
| 1612 |
brvc 2f |
| 1613 |
ldi C1, 0x7f |
| 1614 |
rjmp 1f |
| 1615 |
;; FALLTHRU (Barrier) |
| 1616 |
ENDF __round_s2_const |
| 1617 |
|
| 1618 |
DEFUN __round_u2_const |
| 1619 |
brcc 2f |
| 1620 |
ldi C1, 0xff |
| 1621 |
1: |
| 1622 |
ldi C0, 0xff |
| 1623 |
rjmp 9f |
| 1624 |
2: |
| 1625 |
;; Saturation is performed now. |
| 1626 |
;; Currently, we have C[] = 2^{-RP-1} |
| 1627 |
;; C[] = 2^{-RP} |
| 1628 |
lsl C0 |
| 1629 |
rol C1 |
| 1630 |
;; |
| 1631 |
NEG2 C0 |
| 1632 |
;; Clear the bits beyond the rounding point. |
| 1633 |
and C0, A0 |
| 1634 |
and C1, A1 |
| 1635 |
9: ret |
| 1636 |
ENDF __round_u2_const |
| 1637 |
|
| 1638 |
#endif /* L_round_2_const */ |
| 1639 |
|
| 1640 |
#undef A0 |
| 1641 |
#undef A1 |
| 1642 |
#undef C0 |
| 1643 |
#undef C1 |
| 1644 |
|
| 1645 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1646 |
;; Rounding, 4 Bytes |
| 1647 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1648 |
|
| 1649 |
#ifndef __AVR_TINY__ |
| 1650 |
|
| 1651 |
#define A0 18 |
| 1652 |
#define A1 A0 + 1 |
| 1653 |
#define A2 A0 + 2 |
| 1654 |
#define A3 A0 + 3 |
| 1655 |
|
| 1656 |
#define C0 22 |
| 1657 |
#define C1 C0 + 1 |
| 1658 |
#define C2 C0 + 2 |
| 1659 |
#define C3 C0 + 3 |
| 1660 |
|
| 1661 |
#ifdef L_addmask_4 |
| 1662 |
|
| 1663 |
;; [ R25:R22 = 1 << (R24 & 31) |
| 1664 |
;; R21:R18 += 1 << (R24 & 31) ] |
| 1665 |
;; SREG is set according to the addition |
| 1666 |
DEFUN __addmask_4 |
| 1667 |
;; R25 = 1 << (R24 & 7) |
| 1668 |
XCALL __mask1 |
| 1669 |
cpi RP, 1 << 4 |
| 1670 |
sbc C0, C0 |
| 1671 |
sbc C1, C1 |
| 1672 |
;; Swap C2 with C3 if RP.3 is not set |
| 1673 |
cpi RP, 1 << 3 |
| 1674 |
sbc C2, C2 |
| 1675 |
and C2, C3 |
| 1676 |
eor C3, C2 |
| 1677 |
;; Swap C3:C2 with C1:C0 if RP.4 is not set |
| 1678 |
and C0, C2 $ eor C2, C0 |
| 1679 |
and C1, C3 $ eor C3, C1 |
| 1680 |
;; Finally, add the power-of-two: A[] += C[] |
| 1681 |
add A0, C0 |
| 1682 |
adc A1, C1 |
| 1683 |
adc A2, C2 |
| 1684 |
adc A3, C3 |
| 1685 |
ret |
| 1686 |
ENDF __addmask_4 |
| 1687 |
#endif /* L_addmask_4 */ |
| 1688 |
|
| 1689 |
#ifdef L_round_s4 |
| 1690 |
|
| 1691 |
;; R25:R22 = round (R21:R18, R24) |
| 1692 |
;; Clobbers: R18...R21 |
| 1693 |
DEFUN __roundsq3 |
| 1694 |
subi RP, __SQ_FBIT__ - __SA_FBIT__ |
| 1695 |
ENDF __roundsq3 |
| 1696 |
DEFUN __roundsa3 |
| 1697 |
subi RP, __SA_FBIT__ - 1 |
| 1698 |
neg RP |
| 1699 |
;; [ R25:R22 = 1 << (FBIT-1 - RP) |
| 1700 |
;; R21:R18 += 1 << (FBIT-1 - RP) ] |
| 1701 |
XCALL __addmask_4 |
| 1702 |
XJMP __round_s4_const |
| 1703 |
ENDF __roundsa3 |
| 1704 |
|
| 1705 |
#endif /* L_round_s4 */ |
| 1706 |
|
| 1707 |
#ifdef L_round_u4 |
| 1708 |
|
| 1709 |
;; R25:R22 = round (R21:R18, R24) |
| 1710 |
;; Clobbers: R18...R21 |
| 1711 |
DEFUN __roundusq3 |
| 1712 |
subi RP, __USQ_FBIT__ - __USA_FBIT__ |
| 1713 |
ENDF __roundusq3 |
| 1714 |
DEFUN __roundusa3 |
| 1715 |
subi RP, __USA_FBIT__ - 1 |
| 1716 |
neg RP |
| 1717 |
;; [ R25:R22 = 1 << (FBIT-1 - RP) |
| 1718 |
;; R21:R18 += 1 << (FBIT-1 - RP) ] |
| 1719 |
XCALL __addmask_4 |
| 1720 |
XJMP __round_u4_const |
| 1721 |
ENDF __roundusa3 |
| 1722 |
|
| 1723 |
#endif /* L_round_u4 */ |
| 1724 |
|
| 1725 |
|
| 1726 |
#ifdef L_round_4_const |
| 1727 |
|
| 1728 |
;; Helpers for 4 byte wide rounding |
| 1729 |
|
| 1730 |
DEFUN __round_s4_const |
| 1731 |
brvc 2f |
| 1732 |
ldi C3, 0x7f |
| 1733 |
rjmp 1f |
| 1734 |
;; FALLTHRU (Barrier) |
| 1735 |
ENDF __round_s4_const |
| 1736 |
|
| 1737 |
DEFUN __round_u4_const |
| 1738 |
brcc 2f |
| 1739 |
ldi C3, 0xff |
| 1740 |
1: |
| 1741 |
ldi C2, 0xff |
| 1742 |
ldi C1, 0xff |
| 1743 |
ldi C0, 0xff |
| 1744 |
rjmp 9f |
| 1745 |
2: |
| 1746 |
;; Saturation is performed now. |
| 1747 |
;; Currently, we have C[] = 2^{-RP-1} |
| 1748 |
;; C[] = 2^{-RP} |
| 1749 |
lsl C0 |
| 1750 |
rol C1 |
| 1751 |
rol C2 |
| 1752 |
rol C3 |
| 1753 |
XCALL __negsi2 |
| 1754 |
;; Clear the bits beyond the rounding point. |
| 1755 |
and C0, A0 |
| 1756 |
and C1, A1 |
| 1757 |
and C2, A2 |
| 1758 |
and C3, A3 |
| 1759 |
9: ret |
| 1760 |
ENDF __round_u4_const |
| 1761 |
|
| 1762 |
#endif /* L_round_4_const */ |
| 1763 |
|
| 1764 |
#undef A0 |
| 1765 |
#undef A1 |
| 1766 |
#undef A2 |
| 1767 |
#undef A3 |
| 1768 |
#undef C0 |
| 1769 |
#undef C1 |
| 1770 |
#undef C2 |
| 1771 |
#undef C3 |
| 1772 |
|
| 1773 |
#undef RP |
| 1774 |
|
| 1775 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1776 |
;; Rounding, 8 Bytes |
| 1777 |
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; |
| 1778 |
|
| 1779 |
#define RP 16 |
| 1780 |
#define FBITm1 31 |
| 1781 |
|
| 1782 |
#define C0 18 |
| 1783 |
#define C1 C0 + 1 |
| 1784 |
#define C2 C0 + 2 |
| 1785 |
#define C3 C0 + 3 |
| 1786 |
#define C4 C0 + 4 |
| 1787 |
#define C5 C0 + 5 |
| 1788 |
#define C6 C0 + 6 |
| 1789 |
#define C7 C0 + 7 |
| 1790 |
|
| 1791 |
#define A0 16 |
| 1792 |
#define A1 17 |
| 1793 |
#define A2 26 |
| 1794 |
#define A3 27 |
| 1795 |
#define A4 28 |
| 1796 |
#define A5 29 |
| 1797 |
#define A6 30 |
| 1798 |
#define A7 31 |
| 1799 |
|
| 1800 |
|
| 1801 |
#ifdef L_rounddq3 |
| 1802 |
;; R25:R18 = round (R25:R18, R16) |
| 1803 |
;; Clobbers: ABI |
| 1804 |
DEFUN __rounddq3 |
| 1805 |
ldi FBITm1, __DQ_FBIT__ - 1 |
| 1806 |
clt |
| 1807 |
XJMP __round_x8 |
| 1808 |
ENDF __rounddq3 |
| 1809 |
#endif /* L_rounddq3 */ |
| 1810 |
|
| 1811 |
#ifdef L_roundudq3 |
| 1812 |
;; R25:R18 = round (R25:R18, R16) |
| 1813 |
;; Clobbers: ABI |
| 1814 |
DEFUN __roundudq3 |
| 1815 |
ldi FBITm1, __UDQ_FBIT__ - 1 |
| 1816 |
set |
| 1817 |
XJMP __round_x8 |
| 1818 |
ENDF __roundudq3 |
| 1819 |
#endif /* L_roundudq3 */ |
| 1820 |
|
| 1821 |
#ifdef L_roundda3 |
| 1822 |
;; R25:R18 = round (R25:R18, R16) |
| 1823 |
;; Clobbers: ABI |
| 1824 |
DEFUN __roundda3 |
| 1825 |
ldi FBITm1, __DA_FBIT__ - 1 |
| 1826 |
clt |
| 1827 |
XJMP __round_x8 |
| 1828 |
ENDF __roundda3 |
| 1829 |
#endif /* L_roundda3 */ |
| 1830 |
|
| 1831 |
#ifdef L_rounduda3 |
| 1832 |
;; R25:R18 = round (R25:R18, R16) |
| 1833 |
;; Clobbers: ABI |
| 1834 |
DEFUN __rounduda3 |
| 1835 |
ldi FBITm1, __UDA_FBIT__ - 1 |
| 1836 |
set |
| 1837 |
XJMP __round_x8 |
| 1838 |
ENDF __rounduda3 |
| 1839 |
#endif /* L_rounduda3 */ |
| 1840 |
|
| 1841 |
#ifdef L_roundta3 |
| 1842 |
;; R25:R18 = round (R25:R18, R16) |
| 1843 |
;; Clobbers: ABI |
| 1844 |
DEFUN __roundta3 |
| 1845 |
ldi FBITm1, __TA_FBIT__ - 1 |
| 1846 |
clt |
| 1847 |
XJMP __round_x8 |
| 1848 |
ENDF __roundta3 |
| 1849 |
#endif /* L_roundta3 */ |
| 1850 |
|
| 1851 |
#ifdef L_rounduta3 |
| 1852 |
;; R25:R18 = round (R25:R18, R16) |
| 1853 |
;; Clobbers: ABI |
| 1854 |
DEFUN __rounduta3 |
| 1855 |
ldi FBITm1, __UTA_FBIT__ - 1 |
| 1856 |
set |
| 1857 |
XJMP __round_x8 |
| 1858 |
ENDF __rounduta3 |
| 1859 |
#endif /* L_rounduta3 */ |
| 1860 |
|
| 1861 |
|
| 1862 |
#ifdef L_round_x8 |
| 1863 |
DEFUN __round_x8 |
| 1864 |
push r16 |
| 1865 |
push r17 |
| 1866 |
push r28 |
| 1867 |
push r29 |
| 1868 |
;; Compute log2 of addend from rounding point |
| 1869 |
sub RP, FBITm1 |
| 1870 |
neg RP |
| 1871 |
;; Move input to work register A[] |
| 1872 |
push C0 |
| 1873 |
mov A1, C1 |
| 1874 |
wmov A2, C2 |
| 1875 |
wmov A4, C4 |
| 1876 |
wmov A6, C6 |
| 1877 |
;; C[] = 1 << (FBIT-1 - RP) |
| 1878 |
XCALL __clr_8 |
| 1879 |
inc C0 |
| 1880 |
XCALL __ashldi3 |
| 1881 |
pop A0 |
| 1882 |
;; A[] += C[] |
| 1883 |
add A0, C0 |
| 1884 |
adc A1, C1 |
| 1885 |
adc A2, C2 |
| 1886 |
adc A3, C3 |
| 1887 |
adc A4, C4 |
| 1888 |
adc A5, C5 |
| 1889 |
adc A6, C6 |
| 1890 |
adc A7, C7 |
| 1891 |
brts 1f |
| 1892 |
;; Signed |
| 1893 |
brvc 3f |
| 1894 |
;; Signed overflow: A[] = 0x7f... |
| 1895 |
brvs 2f |
| 1896 |
1: ;; Unsigned |
| 1897 |
brcc 3f |
| 1898 |
;; Unsigned overflow: A[] = 0xff... |
| 1899 |
2: ldi C7, 0xff |
| 1900 |
ldi C6, 0xff |
| 1901 |
wmov C0, C6 |
| 1902 |
wmov C2, C6 |
| 1903 |
wmov C4, C6 |
| 1904 |
bld C7, 7 |
| 1905 |
rjmp 9f |
| 1906 |
3: |
| 1907 |
;; C[] = -C[] - C[] |
| 1908 |
push A0 |
| 1909 |
ldi r16, 1 |
| 1910 |
XCALL __ashldi3 |
| 1911 |
pop A0 |
| 1912 |
XCALL __negdi2 |
| 1913 |
;; Clear the bits beyond the rounding point. |
| 1914 |
and C0, A0 |
| 1915 |
and C1, A1 |
| 1916 |
and C2, A2 |
| 1917 |
and C3, A3 |
| 1918 |
and C4, A4 |
| 1919 |
and C5, A5 |
| 1920 |
and C6, A6 |
| 1921 |
and C7, A7 |
| 1922 |
9: ;; Epilogue |
| 1923 |
pop r29 |
| 1924 |
pop r28 |
| 1925 |
pop r17 |
| 1926 |
pop r16 |
| 1927 |
ret |
| 1928 |
ENDF __round_x8 |
| 1929 |
|
| 1930 |
#endif /* L_round_x8 */ |
| 1931 |
|
| 1932 |
#undef A0 |
| 1933 |
#undef A1 |
| 1934 |
#undef A2 |
| 1935 |
#undef A3 |
| 1936 |
#undef A4 |
| 1937 |
#undef A5 |
| 1938 |
#undef A6 |
| 1939 |
#undef A7 |
| 1940 |
|
| 1941 |
#undef C0 |
| 1942 |
#undef C1 |
| 1943 |
#undef C2 |
| 1944 |
#undef C3 |
| 1945 |
#undef C4 |
| 1946 |
#undef C5 |
| 1947 |
#undef C6 |
| 1948 |
#undef C7 |
| 1949 |
|
| 1950 |
#undef RP |
| 1951 |
#undef FBITm1 |
| 1952 |
|
| 1953 |
#endif /* ! AVR_TINY */ |
| 1954 |
|
| 1955 |
;; Supply implementations / symbols for the bit-banging functions |
| 1956 |
;; __builtin_avr_bitsfx and __builtin_avr_fxbits |
| 1957 |
#ifdef L_ret |
| 1958 |
DEFUN __ret |
| 1959 |
ret |
| 1960 |
ENDF __ret |
| 1961 |
#endif /* L_ret */ |