| 1 |
/* |
| 2 |
* This file is a part of TTMath Bignum Library |
| 3 |
* and is distributed under the (new) BSD licence. |
| 4 |
* Author: Tomasz Sowa <[email protected]> |
| 5 |
*/ |
| 6 |
|
| 7 |
/* |
| 8 |
* Copyright (c) 2006-2009, Tomasz Sowa |
| 9 |
* All rights reserved. |
| 10 |
* |
| 11 |
* Redistribution and use in source and binary forms, with or without |
| 12 |
* modification, are permitted provided that the following conditions are met: |
| 13 |
* |
| 14 |
* * Redistributions of source code must retain the above copyright notice, |
| 15 |
* this list of conditions and the following disclaimer. |
| 16 |
* |
| 17 |
* * Redistributions in binary form must reproduce the above copyright |
| 18 |
* notice, this list of conditions and the following disclaimer in the |
| 19 |
* documentation and/or other materials provided with the distribution. |
| 20 |
* |
| 21 |
* * Neither the name Tomasz Sowa nor the names of contributors to this |
| 22 |
* project may be used to endorse or promote products derived |
| 23 |
* from this software without specific prior written permission. |
| 24 |
* |
| 25 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 26 |
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 27 |
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 28 |
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
| 29 |
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 30 |
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 31 |
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 32 |
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 33 |
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 34 |
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF |
| 35 |
* THE POSSIBILITY OF SUCH DAMAGE. |
| 36 |
*/ |
| 37 |
|
| 38 |
|
| 39 |
|
| 40 |
#ifndef headerfilettmathuint_x86 |
| 41 |
#define headerfilettmathuint_x86 |
| 42 |
|
| 43 |
|
| 44 |
#ifndef TTMATH_NOASM |
| 45 |
#ifdef TTMATH_PLATFORM32 |
| 46 |
|
| 47 |
|
| 48 |
/*! |
| 49 |
\file ttmathuint_x86.h |
| 50 |
\brief template class UInt<uint> with assembler code for 32bit x86 processors |
| 51 |
|
| 52 |
this file is included at the end of ttmathuint.h |
| 53 |
*/ |
| 54 |
|
| 55 |
|
| 56 |
|
| 57 |
/*! |
| 58 |
\brief a namespace for the TTMath library |
| 59 |
*/ |
| 60 |
namespace ttmath |
| 61 |
{ |
| 62 |
|
| 63 |
/*! |
| 64 |
returning the string represents the currect type of the library |
| 65 |
we have following types: |
| 66 |
asm_vc_32 - with asm code designed for Microsoft Visual C++ (32 bits) |
| 67 |
asm_gcc_32 - with asm code designed for GCC (32 bits) |
| 68 |
asm_vc_64 - with asm for VC (64 bit) |
| 69 |
asm_gcc_64 - with asm for GCC (64 bit) |
| 70 |
no_asm_32 - pure C++ version (32 bit) - without any asm code |
| 71 |
no_asm_64 - pure C++ version (64 bit) - without any asm code |
| 72 |
*/ |
| 73 |
template<uint value_size> |
| 74 |
const char * UInt<value_size>::LibTypeStr() |
| 75 |
{ |
| 76 |
#ifndef __GNUC__ |
| 77 |
static const char info[] = "asm_vc_32"; |
| 78 |
#endif |
| 79 |
|
| 80 |
#ifdef __GNUC__ |
| 81 |
static const char info[] = "asm_gcc_32"; |
| 82 |
#endif |
| 83 |
|
| 84 |
return info; |
| 85 |
} |
| 86 |
|
| 87 |
|
| 88 |
/*! |
| 89 |
returning the currect type of the library |
| 90 |
*/ |
| 91 |
template<uint value_size> |
| 92 |
LibTypeCode UInt<value_size>::LibType() |
| 93 |
{ |
| 94 |
#ifndef __GNUC__ |
| 95 |
LibTypeCode info = asm_vc_32; |
| 96 |
#endif |
| 97 |
|
| 98 |
#ifdef __GNUC__ |
| 99 |
LibTypeCode info = asm_gcc_32; |
| 100 |
#endif |
| 101 |
|
| 102 |
return info; |
| 103 |
} |
| 104 |
|
| 105 |
|
| 106 |
|
| 107 |
/*! |
| 108 |
* |
| 109 |
* basic mathematic functions |
| 110 |
* |
| 111 |
*/ |
| 112 |
|
| 113 |
|
| 114 |
/*! |
| 115 |
adding ss2 to the this and adding carry if it's defined |
| 116 |
(this = this + ss2 + c) |
| 117 |
|
| 118 |
c must be zero or one (might be a bigger value than 1) |
| 119 |
function returns carry (1) (if it has been) |
| 120 |
*/ |
| 121 |
template<uint value_size> |
| 122 |
uint UInt<value_size>::Add(const UInt<value_size> & ss2, uint c) |
| 123 |
{ |
| 124 |
uint b = value_size; |
| 125 |
uint * p1 = table; |
| 126 |
uint * p2 = const_cast<uint*>(ss2.table); |
| 127 |
|
| 128 |
// we don't have to use TTMATH_REFERENCE_ASSERT here |
| 129 |
// this algorithm doesn't require it |
| 130 |
|
| 131 |
#ifndef __GNUC__ |
| 132 |
|
| 133 |
// this part might be compiled with for example visual c |
| 134 |
|
| 135 |
__asm |
| 136 |
{ |
| 137 |
push eax |
| 138 |
push ebx |
| 139 |
push ecx |
| 140 |
push edx |
| 141 |
push esi |
| 142 |
|
| 143 |
mov ecx,[b] |
| 144 |
|
| 145 |
mov ebx,[p1] |
| 146 |
mov esi,[p2] |
| 147 |
|
| 148 |
xor edx,edx // edx=0 |
| 149 |
mov eax,[c] |
| 150 |
neg eax // CF=1 if rax!=0 , CF=0 if rax==0 |
| 151 |
|
| 152 |
ttmath_loop: |
| 153 |
mov eax,[esi+edx*4] |
| 154 |
adc [ebx+edx*4],eax |
| 155 |
|
| 156 |
inc edx |
| 157 |
dec ecx |
| 158 |
jnz ttmath_loop |
| 159 |
|
| 160 |
adc ecx, ecx |
| 161 |
mov [c], ecx |
| 162 |
|
| 163 |
pop esi |
| 164 |
pop edx |
| 165 |
pop ecx |
| 166 |
pop ebx |
| 167 |
pop eax |
| 168 |
} |
| 169 |
|
| 170 |
|
| 171 |
|
| 172 |
#endif |
| 173 |
|
| 174 |
|
| 175 |
#ifdef __GNUC__ |
| 176 |
uint dummy, dummy2; |
| 177 |
// this part should be compiled with gcc |
| 178 |
|
| 179 |
__asm__ __volatile__( |
| 180 |
|
| 181 |
"xorl %%edx, %%edx \n" |
| 182 |
"negl %%eax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 183 |
|
| 184 |
"1: \n" |
| 185 |
"movl (%%esi,%%edx,4), %%eax \n" |
| 186 |
"adcl %%eax, (%%ebx,%%edx,4) \n" |
| 187 |
|
| 188 |
"incl %%edx \n" |
| 189 |
"decl %%ecx \n" |
| 190 |
"jnz 1b \n" |
| 191 |
|
| 192 |
"adc %%ecx, %%ecx \n" |
| 193 |
|
| 194 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 195 |
: "0" (b), "1" (c), "b" (p1), "S" (p2) |
| 196 |
: "cc", "memory" ); |
| 197 |
#endif |
| 198 |
|
| 199 |
TTMATH_LOGC("UInt::Add", c) |
| 200 |
|
| 201 |
return c; |
| 202 |
} |
| 203 |
|
| 204 |
|
| 205 |
|
| 206 |
/*! |
| 207 |
adding one word (at a specific position) |
| 208 |
and returning a carry (if it has been) |
| 209 |
|
| 210 |
e.g. |
| 211 |
|
| 212 |
if we've got (value_size=3): |
| 213 |
table[0] = 10; |
| 214 |
table[1] = 30; |
| 215 |
table[2] = 5; |
| 216 |
and we call: |
| 217 |
AddInt(2,1) |
| 218 |
then it'll be: |
| 219 |
table[0] = 10; |
| 220 |
table[1] = 30 + 2; |
| 221 |
table[2] = 5; |
| 222 |
|
| 223 |
of course if there was a carry from table[2] it would be returned |
| 224 |
*/ |
| 225 |
template<uint value_size> |
| 226 |
uint UInt<value_size>::AddInt(uint value, uint index) |
| 227 |
{ |
| 228 |
uint b = value_size; |
| 229 |
uint * p1 = table; |
| 230 |
uint c; |
| 231 |
|
| 232 |
TTMATH_ASSERT( index < value_size ) |
| 233 |
|
| 234 |
#ifndef __GNUC__ |
| 235 |
|
| 236 |
__asm |
| 237 |
{ |
| 238 |
push eax |
| 239 |
push ebx |
| 240 |
push ecx |
| 241 |
push edx |
| 242 |
|
| 243 |
mov ecx, [b] |
| 244 |
sub ecx, [index] |
| 245 |
|
| 246 |
mov edx, [index] |
| 247 |
mov ebx, [p1] |
| 248 |
|
| 249 |
mov eax, [value] |
| 250 |
|
| 251 |
ttmath_loop: |
| 252 |
add [ebx+edx*4], eax |
| 253 |
jnc ttmath_end |
| 254 |
|
| 255 |
mov eax, 1 |
| 256 |
inc edx |
| 257 |
dec ecx |
| 258 |
jnz ttmath_loop |
| 259 |
|
| 260 |
ttmath_end: |
| 261 |
setc al |
| 262 |
movzx edx, al |
| 263 |
mov [c], edx |
| 264 |
|
| 265 |
pop edx |
| 266 |
pop ecx |
| 267 |
pop ebx |
| 268 |
pop eax |
| 269 |
} |
| 270 |
|
| 271 |
#endif |
| 272 |
|
| 273 |
|
| 274 |
#ifdef __GNUC__ |
| 275 |
uint dummy, dummy2; |
| 276 |
|
| 277 |
__asm__ __volatile__( |
| 278 |
|
| 279 |
"subl %%edx, %%ecx \n" |
| 280 |
|
| 281 |
"1: \n" |
| 282 |
"addl %%eax, (%%ebx,%%edx,4) \n" |
| 283 |
"jnc 2f \n" |
| 284 |
|
| 285 |
"movl $1, %%eax \n" |
| 286 |
"incl %%edx \n" |
| 287 |
"decl %%ecx \n" |
| 288 |
"jnz 1b \n" |
| 289 |
|
| 290 |
"2: \n" |
| 291 |
"setc %%al \n" |
| 292 |
"movzx %%al, %%edx \n" |
| 293 |
|
| 294 |
: "=d" (c), "=a" (dummy), "=c" (dummy2) |
| 295 |
: "0" (index), "1" (value), "2" (b), "b" (p1) |
| 296 |
: "cc", "memory" ); |
| 297 |
|
| 298 |
#endif |
| 299 |
|
| 300 |
TTMATH_LOGC("UInt::AddInt", c) |
| 301 |
|
| 302 |
return c; |
| 303 |
} |
| 304 |
|
| 305 |
|
| 306 |
|
| 307 |
|
| 308 |
/*! |
| 309 |
adding only two unsigned words to the existing value |
| 310 |
and these words begin on the 'index' position |
| 311 |
(it's used in the multiplication algorithm 2) |
| 312 |
|
| 313 |
index should be equal or smaller than value_size-2 (index <= value_size-2) |
| 314 |
x1 - lower word, x2 - higher word |
| 315 |
|
| 316 |
for example if we've got value_size equal 4 and: |
| 317 |
table[0] = 3 |
| 318 |
table[1] = 4 |
| 319 |
table[2] = 5 |
| 320 |
table[3] = 6 |
| 321 |
then let |
| 322 |
x1 = 10 |
| 323 |
x2 = 20 |
| 324 |
and |
| 325 |
index = 1 |
| 326 |
|
| 327 |
the result of this method will be: |
| 328 |
table[0] = 3 |
| 329 |
table[1] = 4 + x1 = 14 |
| 330 |
table[2] = 5 + x2 = 25 |
| 331 |
table[3] = 6 |
| 332 |
|
| 333 |
and no carry at the end of table[3] |
| 334 |
|
| 335 |
(of course if there was a carry in table[2](5+20) then |
| 336 |
this carry would be passed to the table[3] etc.) |
| 337 |
*/ |
| 338 |
template<uint value_size> |
| 339 |
uint UInt<value_size>::AddTwoInts(uint x2, uint x1, uint index) |
| 340 |
{ |
| 341 |
uint b = value_size; |
| 342 |
uint * p1 = table; |
| 343 |
uint c; |
| 344 |
|
| 345 |
TTMATH_ASSERT( index < value_size - 1 ) |
| 346 |
|
| 347 |
#ifndef __GNUC__ |
| 348 |
__asm |
| 349 |
{ |
| 350 |
push eax |
| 351 |
push ebx |
| 352 |
push ecx |
| 353 |
push edx |
| 354 |
|
| 355 |
mov ecx, [b] |
| 356 |
sub ecx, [index] |
| 357 |
|
| 358 |
mov ebx, [p1] |
| 359 |
mov edx, [index] |
| 360 |
|
| 361 |
mov eax, [x1] |
| 362 |
add [ebx+edx*4], eax |
| 363 |
inc edx |
| 364 |
dec ecx |
| 365 |
|
| 366 |
mov eax, [x2] |
| 367 |
|
| 368 |
ttmath_loop: |
| 369 |
adc [ebx+edx*4], eax |
| 370 |
jnc ttmath_end |
| 371 |
|
| 372 |
mov eax, 0 |
| 373 |
inc edx |
| 374 |
dec ecx |
| 375 |
jnz ttmath_loop |
| 376 |
|
| 377 |
ttmath_end: |
| 378 |
setc al |
| 379 |
movzx edx, al |
| 380 |
mov [c], edx |
| 381 |
|
| 382 |
pop edx |
| 383 |
pop ecx |
| 384 |
pop ebx |
| 385 |
pop eax |
| 386 |
|
| 387 |
} |
| 388 |
#endif |
| 389 |
|
| 390 |
|
| 391 |
#ifdef __GNUC__ |
| 392 |
uint dummy, dummy2; |
| 393 |
|
| 394 |
__asm__ __volatile__( |
| 395 |
|
| 396 |
"subl %%edx, %%ecx \n" |
| 397 |
|
| 398 |
"addl %%esi, (%%ebx,%%edx,4) \n" |
| 399 |
"incl %%edx \n" |
| 400 |
"decl %%ecx \n" |
| 401 |
|
| 402 |
"1: \n" |
| 403 |
"adcl %%eax, (%%ebx,%%edx,4) \n" |
| 404 |
"jnc 2f \n" |
| 405 |
|
| 406 |
"mov $0, %%eax \n" |
| 407 |
"incl %%edx \n" |
| 408 |
"decl %%ecx \n" |
| 409 |
"jnz 1b \n" |
| 410 |
|
| 411 |
"2: \n" |
| 412 |
"setc %%al \n" |
| 413 |
"movzx %%al, %%eax \n" |
| 414 |
|
| 415 |
: "=a" (c), "=c" (dummy), "=d" (dummy2) |
| 416 |
: "0" (x2), "1" (b), "2" (index), "b" (p1), "S" (x1) |
| 417 |
: "cc", "memory" ); |
| 418 |
|
| 419 |
#endif |
| 420 |
|
| 421 |
TTMATH_LOGC("UInt::AddTwoInts", c) |
| 422 |
|
| 423 |
return c; |
| 424 |
} |
| 425 |
|
| 426 |
|
| 427 |
|
| 428 |
/*! |
| 429 |
this static method addes one vector to the other |
| 430 |
'ss1' is larger in size or equal to 'ss2' |
| 431 |
|
| 432 |
ss1 points to the first (larger) vector |
| 433 |
ss2 points to the second vector |
| 434 |
ss1_size - size of the ss1 (and size of the result too) |
| 435 |
ss2_size - size of the ss2 |
| 436 |
result - is the result vector (which has size the same as ss1: ss1_size) |
| 437 |
|
| 438 |
Example: ss1_size is 5, ss2_size is 3 |
| 439 |
ss1: ss2: result (output): |
| 440 |
5 1 5+1 |
| 441 |
4 3 4+3 |
| 442 |
2 7 2+7 |
| 443 |
6 6 |
| 444 |
9 9 |
| 445 |
of course the carry is propagated and will be returned from the last item |
| 446 |
(this method is used by the Karatsuba multiplication algorithm) |
| 447 |
*/ |
| 448 |
template<uint value_size> |
| 449 |
uint UInt<value_size>::AddVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result) |
| 450 |
{ |
| 451 |
TTMATH_ASSERT( ss1_size >= ss2_size ) |
| 452 |
|
| 453 |
uint rest = ss1_size - ss2_size; |
| 454 |
uint c; |
| 455 |
|
| 456 |
#ifndef __GNUC__ |
| 457 |
|
| 458 |
// this part might be compiled with for example visual c |
| 459 |
__asm |
| 460 |
{ |
| 461 |
pushad |
| 462 |
|
| 463 |
mov ecx, [ss2_size] |
| 464 |
xor edx, edx // edx = 0, cf = 0 |
| 465 |
|
| 466 |
mov esi, [ss1] |
| 467 |
mov ebx, [ss2] |
| 468 |
mov edi, [result] |
| 469 |
|
| 470 |
ttmath_loop: |
| 471 |
mov eax, [esi+edx*4] |
| 472 |
adc eax, [ebx+edx*4] |
| 473 |
mov [edi+edx*4], eax |
| 474 |
|
| 475 |
inc edx |
| 476 |
dec ecx |
| 477 |
jnz ttmath_loop |
| 478 |
|
| 479 |
adc ecx, ecx // ecx has the cf state |
| 480 |
|
| 481 |
mov ebx, [rest] |
| 482 |
or ebx, ebx |
| 483 |
jz ttmath_end |
| 484 |
|
| 485 |
xor ebx, ebx // ebx = 0 |
| 486 |
neg ecx // setting cf from ecx |
| 487 |
mov ecx, [rest] // ecx is != 0 |
| 488 |
|
| 489 |
ttmath_loop2: |
| 490 |
mov eax, [esi+edx*4] |
| 491 |
adc eax, ebx |
| 492 |
mov [edi+edx*4], eax |
| 493 |
|
| 494 |
inc edx |
| 495 |
dec ecx |
| 496 |
jnz ttmath_loop2 |
| 497 |
|
| 498 |
adc ecx, ecx |
| 499 |
|
| 500 |
ttmath_end: |
| 501 |
mov [c], ecx |
| 502 |
|
| 503 |
popad |
| 504 |
} |
| 505 |
|
| 506 |
#endif |
| 507 |
|
| 508 |
|
| 509 |
#ifdef __GNUC__ |
| 510 |
|
| 511 |
// this part should be compiled with gcc |
| 512 |
uint dummy1, dummy2, dummy3; |
| 513 |
|
| 514 |
__asm__ __volatile__( |
| 515 |
"push %%edx \n" |
| 516 |
"xor %%edx, %%edx \n" // edx = 0, cf = 0 |
| 517 |
"1: \n" |
| 518 |
"mov (%%esi,%%edx,4), %%eax \n" |
| 519 |
"adc (%%ebx,%%edx,4), %%eax \n" |
| 520 |
"mov %%eax, (%%edi,%%edx,4) \n" |
| 521 |
|
| 522 |
"inc %%edx \n" |
| 523 |
"dec %%ecx \n" |
| 524 |
"jnz 1b \n" |
| 525 |
|
| 526 |
"adc %%ecx, %%ecx \n" // ecx has the cf state |
| 527 |
"pop %%eax \n" // eax = rest |
| 528 |
|
| 529 |
"or %%eax, %%eax \n" |
| 530 |
"jz 3f \n" |
| 531 |
|
| 532 |
"xor %%ebx, %%ebx \n" // ebx = 0 |
| 533 |
"neg %%ecx \n" // setting cf from ecx |
| 534 |
"mov %%eax, %%ecx \n" // ecx=rest and is != 0 |
| 535 |
"2: \n" |
| 536 |
"mov (%%esi, %%edx, 4), %%eax \n" |
| 537 |
"adc %%ebx, %%eax \n" |
| 538 |
"mov %%eax, (%%edi, %%edx, 4) \n" |
| 539 |
|
| 540 |
"inc %%edx \n" |
| 541 |
"dec %%ecx \n" |
| 542 |
"jnz 2b \n" |
| 543 |
|
| 544 |
"adc %%ecx, %%ecx \n" |
| 545 |
"3: \n" |
| 546 |
|
| 547 |
: "=a" (dummy1), "=b" (dummy2), "=c" (c), "=d" (dummy3) |
| 548 |
: "1" (ss2), "2" (ss2_size), "3" (rest), "S" (ss1), "D" (result) |
| 549 |
: "cc", "memory" ); |
| 550 |
|
| 551 |
#endif |
| 552 |
|
| 553 |
TTMATH_VECTOR_LOGC("UInt::AddVector", c, result, ss1_size) |
| 554 |
|
| 555 |
return c; |
| 556 |
} |
| 557 |
|
| 558 |
|
| 559 |
/*! |
| 560 |
subtracting ss2 from the 'this' and subtracting |
| 561 |
carry if it has been defined |
| 562 |
(this = this - ss2 - c) |
| 563 |
|
| 564 |
c must be zero or one (might be a bigger value than 1) |
| 565 |
function returns carry (1) (if it has been) |
| 566 |
*/ |
| 567 |
template<uint value_size> |
| 568 |
uint UInt<value_size>::Sub(const UInt<value_size> & ss2, uint c) |
| 569 |
{ |
| 570 |
uint b = value_size; |
| 571 |
uint * p1 = table; |
| 572 |
uint * p2 = const_cast<uint*>(ss2.table); |
| 573 |
|
| 574 |
// we don't have to use TTMATH_REFERENCE_ASSERT here |
| 575 |
// this algorithm doesn't require it |
| 576 |
|
| 577 |
#ifndef __GNUC__ |
| 578 |
|
| 579 |
__asm |
| 580 |
{ |
| 581 |
push eax |
| 582 |
push ebx |
| 583 |
push ecx |
| 584 |
push edx |
| 585 |
push esi |
| 586 |
|
| 587 |
mov ecx,[b] |
| 588 |
|
| 589 |
mov ebx,[p1] |
| 590 |
mov esi,[p2] |
| 591 |
|
| 592 |
xor edx,edx // edx=0 |
| 593 |
mov eax,[c] |
| 594 |
neg eax // CF=1 if rax!=0 , CF=0 if rax==0 |
| 595 |
|
| 596 |
ttmath_loop: |
| 597 |
mov eax,[esi+edx*4] |
| 598 |
sbb [ebx+edx*4],eax |
| 599 |
|
| 600 |
inc edx |
| 601 |
dec ecx |
| 602 |
jnz ttmath_loop |
| 603 |
|
| 604 |
adc ecx, ecx |
| 605 |
mov [c], ecx |
| 606 |
|
| 607 |
pop esi |
| 608 |
pop edx |
| 609 |
pop ecx |
| 610 |
pop ebx |
| 611 |
pop eax |
| 612 |
} |
| 613 |
|
| 614 |
#endif |
| 615 |
|
| 616 |
|
| 617 |
#ifdef __GNUC__ |
| 618 |
uint dummy, dummy2; |
| 619 |
|
| 620 |
__asm__ __volatile__( |
| 621 |
|
| 622 |
"xorl %%edx, %%edx \n" |
| 623 |
"negl %%eax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 624 |
|
| 625 |
"1: \n" |
| 626 |
"movl (%%esi,%%edx,4), %%eax \n" |
| 627 |
"sbbl %%eax, (%%ebx,%%edx,4) \n" |
| 628 |
|
| 629 |
"incl %%edx \n" |
| 630 |
"decl %%ecx \n" |
| 631 |
"jnz 1b \n" |
| 632 |
|
| 633 |
"adc %%ecx, %%ecx \n" |
| 634 |
|
| 635 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 636 |
: "0" (b), "1" (c), "b" (p1), "S" (p2) |
| 637 |
: "cc", "memory" ); |
| 638 |
|
| 639 |
#endif |
| 640 |
|
| 641 |
TTMATH_LOGC("UInt::Sub", c) |
| 642 |
|
| 643 |
return c; |
| 644 |
} |
| 645 |
|
| 646 |
|
| 647 |
|
| 648 |
|
| 649 |
/*! |
| 650 |
this method subtracts one word (at a specific position) |
| 651 |
and returns a carry (if it was) |
| 652 |
|
| 653 |
e.g. |
| 654 |
|
| 655 |
if we've got (value_size=3): |
| 656 |
table[0] = 10; |
| 657 |
table[1] = 30; |
| 658 |
table[2] = 5; |
| 659 |
and we call: |
| 660 |
SubInt(2,1) |
| 661 |
then it'll be: |
| 662 |
table[0] = 10; |
| 663 |
table[1] = 30 - 2; |
| 664 |
table[2] = 5; |
| 665 |
|
| 666 |
of course if there was a carry from table[2] it would be returned |
| 667 |
*/ |
| 668 |
template<uint value_size> |
| 669 |
uint UInt<value_size>::SubInt(uint value, uint index) |
| 670 |
{ |
| 671 |
uint b = value_size; |
| 672 |
uint * p1 = table; |
| 673 |
uint c; |
| 674 |
|
| 675 |
TTMATH_ASSERT( index < value_size ) |
| 676 |
|
| 677 |
#ifndef __GNUC__ |
| 678 |
|
| 679 |
__asm |
| 680 |
{ |
| 681 |
push eax |
| 682 |
push ebx |
| 683 |
push ecx |
| 684 |
push edx |
| 685 |
|
| 686 |
mov ecx, [b] |
| 687 |
sub ecx, [index] |
| 688 |
|
| 689 |
mov edx, [index] |
| 690 |
mov ebx, [p1] |
| 691 |
|
| 692 |
mov eax, [value] |
| 693 |
|
| 694 |
ttmath_loop: |
| 695 |
sub [ebx+edx*4], eax |
| 696 |
jnc ttmath_end |
| 697 |
|
| 698 |
mov eax, 1 |
| 699 |
inc edx |
| 700 |
dec ecx |
| 701 |
jnz ttmath_loop |
| 702 |
|
| 703 |
ttmath_end: |
| 704 |
setc al |
| 705 |
movzx edx, al |
| 706 |
mov [c], edx |
| 707 |
|
| 708 |
pop edx |
| 709 |
pop ecx |
| 710 |
pop ebx |
| 711 |
pop eax |
| 712 |
} |
| 713 |
|
| 714 |
#endif |
| 715 |
|
| 716 |
|
| 717 |
#ifdef __GNUC__ |
| 718 |
uint dummy, dummy2; |
| 719 |
|
| 720 |
__asm__ __volatile__( |
| 721 |
|
| 722 |
"subl %%edx, %%ecx \n" |
| 723 |
|
| 724 |
"1: \n" |
| 725 |
"subl %%eax, (%%ebx,%%edx,4) \n" |
| 726 |
"jnc 2f \n" |
| 727 |
|
| 728 |
"movl $1, %%eax \n" |
| 729 |
"incl %%edx \n" |
| 730 |
"decl %%ecx \n" |
| 731 |
"jnz 1b \n" |
| 732 |
|
| 733 |
"2: \n" |
| 734 |
"setc %%al \n" |
| 735 |
"movzx %%al, %%edx \n" |
| 736 |
|
| 737 |
: "=d" (c), "=a" (dummy), "=c" (dummy2) |
| 738 |
: "0" (index), "1" (value), "2" (b), "b" (p1) |
| 739 |
: "cc", "memory" ); |
| 740 |
|
| 741 |
#endif |
| 742 |
|
| 743 |
TTMATH_LOGC("UInt::SubInt", c) |
| 744 |
|
| 745 |
return c; |
| 746 |
} |
| 747 |
|
| 748 |
|
| 749 |
|
| 750 |
/*! |
| 751 |
this static method subtractes one vector from the other |
| 752 |
'ss1' is larger in size or equal to 'ss2' |
| 753 |
|
| 754 |
ss1 points to the first (larger) vector |
| 755 |
ss2 points to the second vector |
| 756 |
ss1_size - size of the ss1 (and size of the result too) |
| 757 |
ss2_size - size of the ss2 |
| 758 |
result - is the result vector (which has size the same as ss1: ss1_size) |
| 759 |
|
| 760 |
Example: ss1_size is 5, ss2_size is 3 |
| 761 |
ss1: ss2: result (output): |
| 762 |
5 1 5-1 |
| 763 |
4 3 4-3 |
| 764 |
2 7 2-7 |
| 765 |
6 6-1 (the borrow from previous item) |
| 766 |
9 9 |
| 767 |
return (carry): 0 |
| 768 |
of course the carry (borrow) is propagated and will be returned from the last item |
| 769 |
(this method is used by the Karatsuba multiplication algorithm) |
| 770 |
*/ |
| 771 |
template<uint value_size> |
| 772 |
uint UInt<value_size>::SubVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result) |
| 773 |
{ |
| 774 |
TTMATH_ASSERT( ss1_size >= ss2_size ) |
| 775 |
|
| 776 |
uint rest = ss1_size - ss2_size; |
| 777 |
uint c; |
| 778 |
|
| 779 |
#ifndef __GNUC__ |
| 780 |
|
| 781 |
// this part might be compiled with for example visual c |
| 782 |
|
| 783 |
/* |
| 784 |
the asm code is nearly the same as in AddVector |
| 785 |
only two instructions 'adc' are changed to 'sbb' |
| 786 |
*/ |
| 787 |
__asm |
| 788 |
{ |
| 789 |
pushad |
| 790 |
|
| 791 |
mov ecx, [ss2_size] |
| 792 |
xor edx, edx // edx = 0, cf = 0 |
| 793 |
|
| 794 |
mov esi, [ss1] |
| 795 |
mov ebx, [ss2] |
| 796 |
mov edi, [result] |
| 797 |
|
| 798 |
ttmath_loop: |
| 799 |
mov eax, [esi+edx*4] |
| 800 |
sbb eax, [ebx+edx*4] |
| 801 |
mov [edi+edx*4], eax |
| 802 |
|
| 803 |
inc edx |
| 804 |
dec ecx |
| 805 |
jnz ttmath_loop |
| 806 |
|
| 807 |
adc ecx, ecx // ecx has the cf state |
| 808 |
|
| 809 |
mov ebx, [rest] |
| 810 |
or ebx, ebx |
| 811 |
jz ttmath_end |
| 812 |
|
| 813 |
xor ebx, ebx // ebx = 0 |
| 814 |
neg ecx // setting cf from ecx |
| 815 |
mov ecx, [rest] // ecx is != 0 |
| 816 |
|
| 817 |
ttmath_loop2: |
| 818 |
mov eax, [esi+edx*4] |
| 819 |
sbb eax, ebx |
| 820 |
mov [edi+edx*4], eax |
| 821 |
|
| 822 |
inc edx |
| 823 |
dec ecx |
| 824 |
jnz ttmath_loop2 |
| 825 |
|
| 826 |
adc ecx, ecx |
| 827 |
|
| 828 |
ttmath_end: |
| 829 |
mov [c], ecx |
| 830 |
|
| 831 |
popad |
| 832 |
} |
| 833 |
|
| 834 |
#endif |
| 835 |
|
| 836 |
|
| 837 |
#ifdef __GNUC__ |
| 838 |
|
| 839 |
// this part should be compiled with gcc |
| 840 |
uint dummy1, dummy2, dummy3; |
| 841 |
|
| 842 |
__asm__ __volatile__( |
| 843 |
"push %%edx \n" |
| 844 |
"xor %%edx, %%edx \n" // edx = 0, cf = 0 |
| 845 |
"1: \n" |
| 846 |
"mov (%%esi,%%edx,4), %%eax \n" |
| 847 |
"sbb (%%ebx,%%edx,4), %%eax \n" |
| 848 |
"mov %%eax, (%%edi,%%edx,4) \n" |
| 849 |
|
| 850 |
"inc %%edx \n" |
| 851 |
"dec %%ecx \n" |
| 852 |
"jnz 1b \n" |
| 853 |
|
| 854 |
"adc %%ecx, %%ecx \n" // ecx has the cf state |
| 855 |
"pop %%eax \n" // eax = rest |
| 856 |
|
| 857 |
"or %%eax, %%eax \n" |
| 858 |
"jz 3f \n" |
| 859 |
|
| 860 |
"xor %%ebx, %%ebx \n" // ebx = 0 |
| 861 |
"neg %%ecx \n" // setting cf from ecx |
| 862 |
"mov %%eax, %%ecx \n" // ecx=rest and is != 0 |
| 863 |
"2: \n" |
| 864 |
"mov (%%esi, %%edx, 4), %%eax \n" |
| 865 |
"sbb %%ebx, %%eax \n" |
| 866 |
"mov %%eax, (%%edi, %%edx, 4) \n" |
| 867 |
|
| 868 |
"inc %%edx \n" |
| 869 |
"dec %%ecx \n" |
| 870 |
"jnz 2b \n" |
| 871 |
|
| 872 |
"adc %%ecx, %%ecx \n" |
| 873 |
"3: \n" |
| 874 |
|
| 875 |
: "=a" (dummy1), "=b" (dummy2), "=c" (c), "=d" (dummy3) |
| 876 |
: "1" (ss2), "2" (ss2_size), "3" (rest), "S" (ss1), "D" (result) |
| 877 |
: "cc", "memory" ); |
| 878 |
|
| 879 |
#endif |
| 880 |
|
| 881 |
TTMATH_VECTOR_LOGC("UInt::SubVector", c, result, ss1_size) |
| 882 |
|
| 883 |
return c; |
| 884 |
} |
| 885 |
|
| 886 |
|
| 887 |
|
| 888 |
/*! |
| 889 |
this method moves all bits into the left hand side |
| 890 |
return value <- this <- c |
| 891 |
|
| 892 |
the lowest *bit* will be held the 'c' and |
| 893 |
the state of one additional bit (on the left hand side) |
| 894 |
will be returned |
| 895 |
|
| 896 |
for example: |
| 897 |
let this is 001010000 |
| 898 |
after Rcl2_one(1) there'll be 010100001 and Rcl2_one returns 0 |
| 899 |
*/ |
| 900 |
template<uint value_size> |
| 901 |
uint UInt<value_size>::Rcl2_one(uint c) |
| 902 |
{ |
| 903 |
uint b = value_size; |
| 904 |
uint * p1 = table; |
| 905 |
|
| 906 |
#ifndef __GNUC__ |
| 907 |
__asm |
| 908 |
{ |
| 909 |
push ebx |
| 910 |
push ecx |
| 911 |
push edx |
| 912 |
|
| 913 |
mov ebx, [p1] |
| 914 |
xor edx, edx |
| 915 |
mov ecx, [c] |
| 916 |
neg ecx |
| 917 |
mov ecx, [b] |
| 918 |
|
| 919 |
ttmath_loop: |
| 920 |
rcl dword ptr [ebx+edx*4], 1 |
| 921 |
|
| 922 |
inc edx |
| 923 |
dec ecx |
| 924 |
jnz ttmath_loop |
| 925 |
|
| 926 |
adc ecx, ecx |
| 927 |
mov [c], ecx |
| 928 |
|
| 929 |
pop edx |
| 930 |
pop ecx |
| 931 |
pop ebx |
| 932 |
} |
| 933 |
#endif |
| 934 |
|
| 935 |
|
| 936 |
#ifdef __GNUC__ |
| 937 |
uint dummy, dummy2; |
| 938 |
|
| 939 |
__asm__ __volatile__( |
| 940 |
|
| 941 |
"xorl %%edx, %%edx \n" // edx=0 |
| 942 |
"negl %%eax \n" // CF=1 if eax!=0 , CF=0 if eax==0 |
| 943 |
|
| 944 |
"1: \n" |
| 945 |
"rcll $1, (%%ebx, %%edx, 4) \n" |
| 946 |
|
| 947 |
"incl %%edx \n" |
| 948 |
"decl %%ecx \n" |
| 949 |
"jnz 1b \n" |
| 950 |
|
| 951 |
"adcl %%ecx, %%ecx \n" |
| 952 |
|
| 953 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 954 |
: "0" (b), "1" (c), "b" (p1) |
| 955 |
: "cc", "memory" ); |
| 956 |
|
| 957 |
#endif |
| 958 |
|
| 959 |
TTMATH_LOGC("UInt::Rcl2_one", c) |
| 960 |
|
| 961 |
return c; |
| 962 |
} |
| 963 |
|
| 964 |
|
| 965 |
|
| 966 |
/*! |
| 967 |
this method moves all bits into the right hand side |
| 968 |
c -> this -> return value |
| 969 |
|
| 970 |
the highest *bit* will be held the 'c' and |
| 971 |
the state of one additional bit (on the right hand side) |
| 972 |
will be returned |
| 973 |
|
| 974 |
for example: |
| 975 |
let this is 000000010 |
| 976 |
after Rcr2_one(1) there'll be 100000001 and Rcr2_one returns 0 |
| 977 |
*/ |
| 978 |
template<uint value_size> |
| 979 |
uint UInt<value_size>::Rcr2_one(uint c) |
| 980 |
{ |
| 981 |
uint b = value_size; |
| 982 |
uint * p1 = table; |
| 983 |
|
| 984 |
#ifndef __GNUC__ |
| 985 |
__asm |
| 986 |
{ |
| 987 |
push ebx |
| 988 |
push ecx |
| 989 |
|
| 990 |
mov ebx, [p1] |
| 991 |
mov ecx, [c] |
| 992 |
neg ecx |
| 993 |
mov ecx, [b] |
| 994 |
|
| 995 |
ttmath_loop: |
| 996 |
rcr dword ptr [ebx+ecx*4-4], 1 |
| 997 |
|
| 998 |
dec ecx |
| 999 |
jnz ttmath_loop |
| 1000 |
|
| 1001 |
adc ecx, ecx |
| 1002 |
mov [c], ecx |
| 1003 |
|
| 1004 |
pop ecx |
| 1005 |
pop ebx |
| 1006 |
} |
| 1007 |
#endif |
| 1008 |
|
| 1009 |
|
| 1010 |
#ifdef __GNUC__ |
| 1011 |
uint dummy; |
| 1012 |
|
| 1013 |
__asm__ __volatile__( |
| 1014 |
|
| 1015 |
"negl %%eax \n" // CF=1 if eax!=0 , CF=0 if eax==0 |
| 1016 |
|
| 1017 |
"1: \n" |
| 1018 |
"rcrl $1, -4(%%ebx, %%ecx, 4) \n" |
| 1019 |
|
| 1020 |
"decl %%ecx \n" |
| 1021 |
"jnz 1b \n" |
| 1022 |
|
| 1023 |
"adcl %%ecx, %%ecx \n" |
| 1024 |
|
| 1025 |
: "=c" (c), "=a" (dummy) |
| 1026 |
: "0" (b), "1" (c), "b" (p1) |
| 1027 |
: "cc", "memory" ); |
| 1028 |
|
| 1029 |
#endif |
| 1030 |
|
| 1031 |
TTMATH_LOGC("UInt::Rcr2_one", c) |
| 1032 |
|
| 1033 |
return c; |
| 1034 |
} |
| 1035 |
|
| 1036 |
|
| 1037 |
|
| 1038 |
#ifdef _MSC_VER |
| 1039 |
#pragma warning (disable : 4731) |
| 1040 |
//warning C4731: frame pointer register 'ebp' modified by inline assembly code |
| 1041 |
#endif |
| 1042 |
|
| 1043 |
|
| 1044 |
|
| 1045 |
/*! |
| 1046 |
this method moves all bits into the left hand side |
| 1047 |
return value <- this <- c |
| 1048 |
|
| 1049 |
the lowest *bits* will be held the 'c' and |
| 1050 |
the state of one additional bit (on the left hand side) |
| 1051 |
will be returned |
| 1052 |
|
| 1053 |
for example: |
| 1054 |
let this is 001010000 |
| 1055 |
after Rcl2(3, 1) there'll be 010000111 and Rcl2 returns 1 |
| 1056 |
*/ |
| 1057 |
template<uint value_size> |
| 1058 |
uint UInt<value_size>::Rcl2(uint bits, uint c) |
| 1059 |
{ |
| 1060 |
TTMATH_ASSERT( bits>0 && bits<TTMATH_BITS_PER_UINT ) |
| 1061 |
|
| 1062 |
uint b = value_size; |
| 1063 |
uint * p1 = table; |
| 1064 |
|
| 1065 |
#ifndef __GNUC__ |
| 1066 |
__asm |
| 1067 |
{ |
| 1068 |
push eax |
| 1069 |
push ebx |
| 1070 |
push ecx |
| 1071 |
push edx |
| 1072 |
push esi |
| 1073 |
push edi |
| 1074 |
push ebp |
| 1075 |
|
| 1076 |
mov edi, [b] |
| 1077 |
|
| 1078 |
mov ecx, 32 |
| 1079 |
sub ecx, [bits] |
| 1080 |
mov edx, -1 |
| 1081 |
shr edx, cl |
| 1082 |
|
| 1083 |
mov ecx, [bits] |
| 1084 |
mov ebx, [p1] |
| 1085 |
mov eax, [c] |
| 1086 |
|
| 1087 |
mov ebp, edx // ebp = mask (modified ebp - don't read/write to variables) |
| 1088 |
|
| 1089 |
xor edx, edx // edx = 0 |
| 1090 |
mov esi, edx |
| 1091 |
or eax, eax |
| 1092 |
cmovnz esi, ebp // if(c) esi=mask else esi=0 |
| 1093 |
|
| 1094 |
ttmath_loop: |
| 1095 |
rol dword ptr [ebx+edx*4], cl |
| 1096 |
|
| 1097 |
mov eax, [ebx+edx*4] |
| 1098 |
and eax, ebp |
| 1099 |
xor [ebx+edx*4], eax // clearing bits |
| 1100 |
or [ebx+edx*4], esi // saving old value |
| 1101 |
mov esi, eax |
| 1102 |
|
| 1103 |
inc edx |
| 1104 |
dec edi |
| 1105 |
jnz ttmath_loop |
| 1106 |
|
| 1107 |
pop ebp // restoring ebp |
| 1108 |
|
| 1109 |
and eax, 1 |
| 1110 |
mov [c], eax |
| 1111 |
|
| 1112 |
pop edi |
| 1113 |
pop esi |
| 1114 |
pop edx |
| 1115 |
pop ecx |
| 1116 |
pop ebx |
| 1117 |
pop eax |
| 1118 |
} |
| 1119 |
#endif |
| 1120 |
|
| 1121 |
|
| 1122 |
#ifdef __GNUC__ |
| 1123 |
uint dummy, dummy2, dummy3; |
| 1124 |
|
| 1125 |
__asm__ __volatile__( |
| 1126 |
|
| 1127 |
"push %%ebp \n" |
| 1128 |
|
| 1129 |
"movl %%ecx, %%esi \n" |
| 1130 |
"movl $32, %%ecx \n" |
| 1131 |
"subl %%esi, %%ecx \n" // ecx = 32 - bits |
| 1132 |
"movl $-1, %%edx \n" // edx = -1 (all bits set to one) |
| 1133 |
"shrl %%cl, %%edx \n" // shifting (0 -> edx -> cf) (cl times) |
| 1134 |
"movl %%edx, %%ebp \n" // ebp = edx = mask |
| 1135 |
"movl %%esi, %%ecx \n" |
| 1136 |
|
| 1137 |
"xorl %%edx, %%edx \n" |
| 1138 |
"movl %%edx, %%esi \n" |
| 1139 |
"orl %%eax, %%eax \n" |
| 1140 |
"cmovnz %%ebp, %%esi \n" // if(c) esi=mask else esi=0 |
| 1141 |
|
| 1142 |
"1: \n" |
| 1143 |
"roll %%cl, (%%ebx,%%edx,4) \n" |
| 1144 |
|
| 1145 |
"movl (%%ebx,%%edx,4), %%eax \n" |
| 1146 |
"andl %%ebp, %%eax \n" |
| 1147 |
"xorl %%eax, (%%ebx,%%edx,4) \n" |
| 1148 |
"orl %%esi, (%%ebx,%%edx,4) \n" |
| 1149 |
"movl %%eax, %%esi \n" |
| 1150 |
|
| 1151 |
"incl %%edx \n" |
| 1152 |
"decl %%edi \n" |
| 1153 |
"jnz 1b \n" |
| 1154 |
|
| 1155 |
"and $1, %%eax \n" |
| 1156 |
|
| 1157 |
"pop %%ebp \n" |
| 1158 |
|
| 1159 |
: "=a" (c), "=D" (dummy), "=S" (dummy2), "=d" (dummy3) |
| 1160 |
: "0" (c), "1" (b), "b" (p1), "c" (bits) |
| 1161 |
: "cc", "memory" ); |
| 1162 |
|
| 1163 |
#endif |
| 1164 |
|
| 1165 |
TTMATH_LOGC("UInt::Rcl2", c) |
| 1166 |
|
| 1167 |
return c; |
| 1168 |
} |
| 1169 |
|
| 1170 |
|
| 1171 |
|
| 1172 |
|
| 1173 |
/*! |
| 1174 |
this method moves all bits into the right hand side |
| 1175 |
C -> this -> return value |
| 1176 |
|
| 1177 |
the highest *bits* will be held the 'c' and |
| 1178 |
the state of one additional bit (on the right hand side) |
| 1179 |
will be returned |
| 1180 |
|
| 1181 |
for example: |
| 1182 |
let this is 000000010 |
| 1183 |
after Rcr2(2, 1) there'll be 110000000 and Rcr2 returns 1 |
| 1184 |
*/ |
| 1185 |
template<uint value_size> |
| 1186 |
uint UInt<value_size>::Rcr2(uint bits, uint c) |
| 1187 |
{ |
| 1188 |
TTMATH_ASSERT( bits>0 && bits<TTMATH_BITS_PER_UINT ) |
| 1189 |
|
| 1190 |
uint b = value_size; |
| 1191 |
uint * p1 = table; |
| 1192 |
|
| 1193 |
#ifndef __GNUC__ |
| 1194 |
__asm |
| 1195 |
{ |
| 1196 |
push eax |
| 1197 |
push ebx |
| 1198 |
push ecx |
| 1199 |
push edx |
| 1200 |
push esi |
| 1201 |
push edi |
| 1202 |
push ebp |
| 1203 |
|
| 1204 |
mov edi, [b] |
| 1205 |
|
| 1206 |
mov ecx, 32 |
| 1207 |
sub ecx, [bits] |
| 1208 |
mov edx, -1 |
| 1209 |
shl edx, cl |
| 1210 |
|
| 1211 |
mov ecx, [bits] |
| 1212 |
mov ebx, [p1] |
| 1213 |
mov eax, [c] |
| 1214 |
|
| 1215 |
mov ebp, edx // ebp = mask (modified ebp - don't read/write to variables) |
| 1216 |
|
| 1217 |
xor edx, edx // edx = 0 |
| 1218 |
mov esi, edx |
| 1219 |
add edx, edi |
| 1220 |
dec edx // edx is pointing at the end of the table (on last word) |
| 1221 |
or eax, eax |
| 1222 |
cmovnz esi, ebp // if(c) esi=mask else esi=0 |
| 1223 |
|
| 1224 |
ttmath_loop: |
| 1225 |
ror dword ptr [ebx+edx*4], cl |
| 1226 |
|
| 1227 |
mov eax, [ebx+edx*4] |
| 1228 |
and eax, ebp |
| 1229 |
xor [ebx+edx*4], eax // clearing bits |
| 1230 |
or [ebx+edx*4], esi // saving old value |
| 1231 |
mov esi, eax |
| 1232 |
|
| 1233 |
dec edx |
| 1234 |
dec edi |
| 1235 |
jnz ttmath_loop |
| 1236 |
|
| 1237 |
pop ebp // restoring ebp |
| 1238 |
|
| 1239 |
rol eax, 1 // 31bit will be first |
| 1240 |
and eax, 1 |
| 1241 |
mov [c], eax |
| 1242 |
|
| 1243 |
pop edi |
| 1244 |
pop esi |
| 1245 |
pop edx |
| 1246 |
pop ecx |
| 1247 |
pop ebx |
| 1248 |
pop eax |
| 1249 |
} |
| 1250 |
#endif |
| 1251 |
|
| 1252 |
|
| 1253 |
#ifdef __GNUC__ |
| 1254 |
uint dummy, dummy2, dummy3; |
| 1255 |
|
| 1256 |
__asm__ __volatile__( |
| 1257 |
|
| 1258 |
"push %%ebp \n" |
| 1259 |
|
| 1260 |
"movl %%ecx, %%esi \n" |
| 1261 |
"movl $32, %%ecx \n" |
| 1262 |
"subl %%esi, %%ecx \n" // ecx = 32 - bits |
| 1263 |
"movl $-1, %%edx \n" // edx = -1 (all bits set to one) |
| 1264 |
"shll %%cl, %%edx \n" // shifting (cf <- edx <- 0) (cl times) |
| 1265 |
"movl %%edx, %%ebp \n" // ebp = edx = mask |
| 1266 |
"movl %%esi, %%ecx \n" |
| 1267 |
|
| 1268 |
"xorl %%edx, %%edx \n" |
| 1269 |
"movl %%edx, %%esi \n" |
| 1270 |
"addl %%edi, %%edx \n" |
| 1271 |
"decl %%edx \n" // edx is pointing at the end of the table (on last word) |
| 1272 |
"orl %%eax, %%eax \n" |
| 1273 |
"cmovnz %%ebp, %%esi \n" // if(c) esi=mask else esi=0 |
| 1274 |
|
| 1275 |
"1: \n" |
| 1276 |
"rorl %%cl, (%%ebx,%%edx,4) \n" |
| 1277 |
|
| 1278 |
"movl (%%ebx,%%edx,4), %%eax \n" |
| 1279 |
"andl %%ebp, %%eax \n" |
| 1280 |
"xorl %%eax, (%%ebx,%%edx,4) \n" |
| 1281 |
"orl %%esi, (%%ebx,%%edx,4) \n" |
| 1282 |
"movl %%eax, %%esi \n" |
| 1283 |
|
| 1284 |
"decl %%edx \n" |
| 1285 |
"decl %%edi \n" |
| 1286 |
"jnz 1b \n" |
| 1287 |
|
| 1288 |
"roll $1, %%eax \n" |
| 1289 |
"andl $1, %%eax \n" |
| 1290 |
|
| 1291 |
"pop %%ebp \n" |
| 1292 |
|
| 1293 |
: "=a" (c), "=D" (dummy), "=S" (dummy2), "=d" (dummy3) |
| 1294 |
: "0" (c), "1" (b), "b" (p1), "c" (bits) |
| 1295 |
: "cc", "memory" ); |
| 1296 |
|
| 1297 |
#endif |
| 1298 |
|
| 1299 |
TTMATH_LOGC("UInt::Rcr2", c) |
| 1300 |
|
| 1301 |
return c; |
| 1302 |
} |
| 1303 |
|
| 1304 |
|
| 1305 |
#ifdef _MSC_VER |
| 1306 |
#pragma warning (default : 4731) |
| 1307 |
#endif |
| 1308 |
|
| 1309 |
|
| 1310 |
/* |
| 1311 |
this method returns the number of the highest set bit in one 32-bit word |
| 1312 |
if the 'x' is zero this method returns '-1' |
| 1313 |
*/ |
| 1314 |
template<uint value_size> |
| 1315 |
sint UInt<value_size>::FindLeadingBitInWord(uint x) |
| 1316 |
{ |
| 1317 |
sint result; |
| 1318 |
|
| 1319 |
#ifndef __GNUC__ |
| 1320 |
__asm |
| 1321 |
{ |
| 1322 |
push eax |
| 1323 |
push edx |
| 1324 |
|
| 1325 |
mov edx,-1 |
| 1326 |
bsr eax,[x] |
| 1327 |
cmovz eax,edx |
| 1328 |
mov [result], eax |
| 1329 |
|
| 1330 |
pop edx |
| 1331 |
pop eax |
| 1332 |
} |
| 1333 |
#endif |
| 1334 |
|
| 1335 |
|
| 1336 |
#ifdef __GNUC__ |
| 1337 |
uint dummy; |
| 1338 |
|
| 1339 |
__asm__ ( |
| 1340 |
|
| 1341 |
"movl $-1, %1 \n" |
| 1342 |
"bsrl %2, %0 \n" |
| 1343 |
"cmovz %1, %0 \n" |
| 1344 |
|
| 1345 |
: "=r" (result), "=&r" (dummy) |
| 1346 |
: "r" (x) |
| 1347 |
: "cc" ); |
| 1348 |
|
| 1349 |
#endif |
| 1350 |
|
| 1351 |
return result; |
| 1352 |
} |
| 1353 |
|
| 1354 |
|
| 1355 |
|
| 1356 |
/* |
| 1357 |
this method returns the number of the smallest set bit in one 32-bit word |
| 1358 |
if the 'x' is zero this method returns '-1' |
| 1359 |
*/ |
| 1360 |
template<uint value_size> |
| 1361 |
sint UInt<value_size>::FindLowestBitInWord(uint x) |
| 1362 |
{ |
| 1363 |
sint result; |
| 1364 |
|
| 1365 |
#ifndef __GNUC__ |
| 1366 |
__asm |
| 1367 |
{ |
| 1368 |
push eax |
| 1369 |
push edx |
| 1370 |
|
| 1371 |
mov edx,-1 |
| 1372 |
bsf eax,[x] |
| 1373 |
cmovz eax,edx |
| 1374 |
mov [result], eax |
| 1375 |
|
| 1376 |
pop edx |
| 1377 |
pop eax |
| 1378 |
} |
| 1379 |
#endif |
| 1380 |
|
| 1381 |
|
| 1382 |
#ifdef __GNUC__ |
| 1383 |
uint dummy; |
| 1384 |
|
| 1385 |
__asm__ ( |
| 1386 |
|
| 1387 |
"movl $-1, %1 \n" |
| 1388 |
"bsfl %2, %0 \n" |
| 1389 |
"cmovz %1, %0 \n" |
| 1390 |
|
| 1391 |
: "=r" (result), "=&r" (dummy) |
| 1392 |
: "r" (x) |
| 1393 |
: "cc" ); |
| 1394 |
|
| 1395 |
#endif |
| 1396 |
|
| 1397 |
return result; |
| 1398 |
} |
| 1399 |
|
| 1400 |
|
| 1401 |
|
| 1402 |
/*! |
| 1403 |
this method sets a special bit in the 'value' |
| 1404 |
and returns the last state of the bit (zero or one) |
| 1405 |
|
| 1406 |
bit is from <0,31> |
| 1407 |
e.g. |
| 1408 |
uint x = 100; |
| 1409 |
uint bit = SetBitInWord(x, 3); |
| 1410 |
now: x = 108 and bit = 0 |
| 1411 |
*/ |
| 1412 |
template<uint value_size> |
| 1413 |
uint UInt<value_size>::SetBitInWord(uint & value, uint bit) |
| 1414 |
{ |
| 1415 |
TTMATH_ASSERT( bit < TTMATH_BITS_PER_UINT ) |
| 1416 |
|
| 1417 |
uint old_bit; |
| 1418 |
uint v = value; |
| 1419 |
|
| 1420 |
#ifndef __GNUC__ |
| 1421 |
__asm |
| 1422 |
{ |
| 1423 |
push ebx |
| 1424 |
push eax |
| 1425 |
|
| 1426 |
mov eax, [v] |
| 1427 |
mov ebx, [bit] |
| 1428 |
bts eax, ebx |
| 1429 |
mov [v], eax |
| 1430 |
|
| 1431 |
setc bl |
| 1432 |
movzx ebx, bl |
| 1433 |
mov [old_bit], ebx |
| 1434 |
|
| 1435 |
pop eax |
| 1436 |
pop ebx |
| 1437 |
} |
| 1438 |
#endif |
| 1439 |
|
| 1440 |
|
| 1441 |
#ifdef __GNUC__ |
| 1442 |
__asm__ ( |
| 1443 |
|
| 1444 |
"btsl %%ebx, %%eax \n" |
| 1445 |
"setc %%bl \n" |
| 1446 |
"movzx %%bl, %%ebx \n" |
| 1447 |
|
| 1448 |
: "=a" (v), "=b" (old_bit) |
| 1449 |
: "0" (v), "1" (bit) |
| 1450 |
: "cc" ); |
| 1451 |
|
| 1452 |
#endif |
| 1453 |
|
| 1454 |
value = v; |
| 1455 |
|
| 1456 |
return old_bit; |
| 1457 |
} |
| 1458 |
|
| 1459 |
|
| 1460 |
|
| 1461 |
|
| 1462 |
/*! |
| 1463 |
multiplication: result_high:result_low = a * b |
| 1464 |
result_high - higher word of the result |
| 1465 |
result_low - lower word of the result |
| 1466 |
|
| 1467 |
this methos never returns a carry |
| 1468 |
this method is used in the second version of the multiplication algorithms |
| 1469 |
*/ |
| 1470 |
template<uint value_size> |
| 1471 |
void UInt<value_size>::MulTwoWords(uint a, uint b, uint * result_high, uint * result_low) |
| 1472 |
{ |
| 1473 |
/* |
| 1474 |
we must use these temporary variables in order to inform the compilator |
| 1475 |
that value pointed with result1 and result2 has changed |
| 1476 |
|
| 1477 |
this has no effect in visual studio but it's useful when |
| 1478 |
using gcc and options like -Ox |
| 1479 |
*/ |
| 1480 |
uint result1_; |
| 1481 |
uint result2_; |
| 1482 |
|
| 1483 |
#ifndef __GNUC__ |
| 1484 |
|
| 1485 |
__asm |
| 1486 |
{ |
| 1487 |
push eax |
| 1488 |
push edx |
| 1489 |
|
| 1490 |
mov eax, [a] |
| 1491 |
mul dword ptr [b] |
| 1492 |
|
| 1493 |
mov [result2_], edx |
| 1494 |
mov [result1_], eax |
| 1495 |
|
| 1496 |
pop edx |
| 1497 |
pop eax |
| 1498 |
} |
| 1499 |
|
| 1500 |
#endif |
| 1501 |
|
| 1502 |
|
| 1503 |
#ifdef __GNUC__ |
| 1504 |
|
| 1505 |
__asm__ ( |
| 1506 |
|
| 1507 |
"mull %%edx \n" |
| 1508 |
|
| 1509 |
: "=a" (result1_), "=d" (result2_) |
| 1510 |
: "0" (a), "1" (b) |
| 1511 |
: "cc" ); |
| 1512 |
|
| 1513 |
#endif |
| 1514 |
|
| 1515 |
|
| 1516 |
*result_low = result1_; |
| 1517 |
*result_high = result2_; |
| 1518 |
} |
| 1519 |
|
| 1520 |
|
| 1521 |
|
| 1522 |
|
| 1523 |
|
| 1524 |
/*! |
| 1525 |
* |
| 1526 |
* Division |
| 1527 |
* |
| 1528 |
* |
| 1529 |
*/ |
| 1530 |
|
| 1531 |
|
| 1532 |
|
| 1533 |
|
| 1534 |
/*! |
| 1535 |
this method calculates 64bits word a:b / 32bits c (a higher, b lower word) |
| 1536 |
r = a:b / c and rest - remainder |
| 1537 |
|
| 1538 |
* |
| 1539 |
* WARNING: |
| 1540 |
* if r (one word) is too small for the result or c is equal zero |
| 1541 |
* there'll be a hardware interruption (0) |
| 1542 |
* and probably the end of your program |
| 1543 |
* |
| 1544 |
*/ |
| 1545 |
template<uint value_size> |
| 1546 |
void UInt<value_size>::DivTwoWords(uint a, uint b, uint c, uint * r, uint * rest) |
| 1547 |
{ |
| 1548 |
uint r_; |
| 1549 |
uint rest_; |
| 1550 |
/* |
| 1551 |
these variables have similar meaning like those in |
| 1552 |
the multiplication algorithm MulTwoWords |
| 1553 |
*/ |
| 1554 |
|
| 1555 |
TTMATH_ASSERT( c != 0 ) |
| 1556 |
|
| 1557 |
#ifndef __GNUC__ |
| 1558 |
__asm |
| 1559 |
{ |
| 1560 |
push eax |
| 1561 |
push edx |
| 1562 |
|
| 1563 |
mov edx, [a] |
| 1564 |
mov eax, [b] |
| 1565 |
div dword ptr [c] |
| 1566 |
|
| 1567 |
mov [r_], eax |
| 1568 |
mov [rest_], edx |
| 1569 |
|
| 1570 |
pop edx |
| 1571 |
pop eax |
| 1572 |
} |
| 1573 |
#endif |
| 1574 |
|
| 1575 |
|
| 1576 |
#ifdef __GNUC__ |
| 1577 |
|
| 1578 |
__asm__ ( |
| 1579 |
|
| 1580 |
"divl %%ecx \n" |
| 1581 |
|
| 1582 |
: "=a" (r_), "=d" (rest_) |
| 1583 |
: "0" (b), "1" (a), "c" (c) |
| 1584 |
: "cc" ); |
| 1585 |
|
| 1586 |
#endif |
| 1587 |
|
| 1588 |
|
| 1589 |
*r = r_; |
| 1590 |
*rest = rest_; |
| 1591 |
|
| 1592 |
} |
| 1593 |
|
| 1594 |
|
| 1595 |
|
| 1596 |
} //namespace |
| 1597 |
|
| 1598 |
|
| 1599 |
|
| 1600 |
#endif //ifdef TTMATH_PLATFORM32 |
| 1601 |
#endif //ifndef TTMATH_NOASM |
| 1602 |
#endif |