| 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) 2006-2010, 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 |
#ifndef headerfilettmathuint_x86_64 |
| 40 |
#define headerfilettmathuint_x86_64 |
| 41 |
|
| 42 |
|
| 43 |
#ifndef TTMATH_NOASM |
| 44 |
#ifdef TTMATH_PLATFORM64 |
| 45 |
|
| 46 |
|
| 47 |
/*! |
| 48 |
\file ttmathuint_x86_64.h |
| 49 |
\brief template class UInt<uint> with assembler code for 64bit x86_64 processors |
| 50 |
|
| 51 |
this file is included at the end of ttmathuint.h |
| 52 |
*/ |
| 53 |
|
| 54 |
#ifndef __GNUC__ |
| 55 |
#include <intrin.h> |
| 56 |
#endif |
| 57 |
|
| 58 |
|
| 59 |
namespace ttmath |
| 60 |
{ |
| 61 |
|
| 62 |
#ifndef __GNUC__ |
| 63 |
|
| 64 |
extern "C" |
| 65 |
{ |
| 66 |
uint __fastcall ttmath_adc_x64(uint* p1, const uint* p2, uint nSize, uint c); |
| 67 |
uint __fastcall ttmath_addindexed_x64(uint* p1, uint nSize, uint nPos, uint nValue); |
| 68 |
uint __fastcall ttmath_addindexed2_x64(uint* p1, uint nSize, uint nPos, uint nValue1, uint nValue2); |
| 69 |
uint __fastcall ttmath_addvector_x64(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result); |
| 70 |
uint __fastcall ttmath_sbb_x64(uint* p1, const uint* p2, uint nSize, uint c); |
| 71 |
uint __fastcall ttmath_subindexed_x64(uint* p1, uint nSize, uint nPos, uint nValue); |
| 72 |
uint __fastcall ttmath_subvector_x64(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result); |
| 73 |
uint __fastcall ttmath_rcl_x64(uint* p1, uint nSize, uint nLowestBit); |
| 74 |
uint __fastcall ttmath_rcr_x64(uint* p1, uint nSize, uint nLowestBit); |
| 75 |
uint __fastcall ttmath_div_x64(uint* pnValHi, uint* pnValLo, uint nDiv); |
| 76 |
uint __fastcall ttmath_rcl2_x64(uint* p1, uint nSize, uint nBits, uint c); |
| 77 |
uint __fastcall ttmath_rcr2_x64(uint* p1, uint nSize, uint nBits, uint c); |
| 78 |
}; |
| 79 |
#endif |
| 80 |
|
| 81 |
|
| 82 |
/*! |
| 83 |
returning the string represents the currect type of the library |
| 84 |
we have following types: |
| 85 |
asm_vc_32 - with asm code designed for Microsoft Visual C++ (32 bits) |
| 86 |
asm_gcc_32 - with asm code designed for GCC (32 bits) |
| 87 |
asm_vc_64 - with asm for VC (64 bit) |
| 88 |
asm_gcc_64 - with asm for GCC (64 bit) |
| 89 |
no_asm_32 - pure C++ version (32 bit) - without any asm code |
| 90 |
no_asm_64 - pure C++ version (64 bit) - without any asm code |
| 91 |
*/ |
| 92 |
template<uint value_size> |
| 93 |
const char * UInt<value_size>::LibTypeStr() |
| 94 |
{ |
| 95 |
#ifndef __GNUC__ |
| 96 |
static const char info[] = "asm_vc_64"; |
| 97 |
#endif |
| 98 |
|
| 99 |
#ifdef __GNUC__ |
| 100 |
static const char info[] = "asm_gcc_64"; |
| 101 |
#endif |
| 102 |
|
| 103 |
return info; |
| 104 |
} |
| 105 |
|
| 106 |
|
| 107 |
/*! |
| 108 |
returning the currect type of the library |
| 109 |
*/ |
| 110 |
template<uint value_size> |
| 111 |
LibTypeCode UInt<value_size>::LibType() |
| 112 |
{ |
| 113 |
#ifndef __GNUC__ |
| 114 |
LibTypeCode info = asm_vc_64; |
| 115 |
#endif |
| 116 |
|
| 117 |
#ifdef __GNUC__ |
| 118 |
LibTypeCode info = asm_gcc_64; |
| 119 |
#endif |
| 120 |
|
| 121 |
return info; |
| 122 |
} |
| 123 |
|
| 124 |
|
| 125 |
/*! |
| 126 |
* |
| 127 |
* basic mathematic functions |
| 128 |
* |
| 129 |
*/ |
| 130 |
|
| 131 |
|
| 132 |
|
| 133 |
/*! |
| 134 |
this method adding ss2 to the this and adding carry if it's defined |
| 135 |
(this = this + ss2 + c) |
| 136 |
|
| 137 |
***this method is created only on a 64bit platform*** |
| 138 |
|
| 139 |
c must be zero or one (might be a bigger value than 1) |
| 140 |
function returns carry (1) (if it was) |
| 141 |
*/ |
| 142 |
template<uint value_size> |
| 143 |
uint UInt<value_size>::Add(const UInt<value_size> & ss2, uint c) |
| 144 |
{ |
| 145 |
uint b = value_size; |
| 146 |
uint * p1 = table; |
| 147 |
const uint * p2 = ss2.table; |
| 148 |
|
| 149 |
// we don't have to use TTMATH_REFERENCE_ASSERT here |
| 150 |
// this algorithm doesn't require it |
| 151 |
|
| 152 |
#ifndef __GNUC__ |
| 153 |
c = ttmath_adc_x64(p1,p2,b,c); |
| 154 |
#endif |
| 155 |
|
| 156 |
#ifdef __GNUC__ |
| 157 |
uint dummy, dummy2; |
| 158 |
|
| 159 |
/* |
| 160 |
this part should be compiled with gcc |
| 161 |
*/ |
| 162 |
__asm__ __volatile__( |
| 163 |
|
| 164 |
"xorq %%rdx, %%rdx \n" |
| 165 |
"negq %%rax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 166 |
|
| 167 |
"1: \n" |
| 168 |
"movq (%%rsi,%%rdx,8), %%rax \n" |
| 169 |
"adcq %%rax, (%%rbx,%%rdx,8) \n" |
| 170 |
|
| 171 |
"incq %%rdx \n" |
| 172 |
"decq %%rcx \n" |
| 173 |
"jnz 1b \n" |
| 174 |
|
| 175 |
"adcq %%rcx, %%rcx \n" |
| 176 |
|
| 177 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 178 |
: "0" (b), "1" (c), "b" (p1), "S" (p2) |
| 179 |
: "cc", "memory" ); |
| 180 |
|
| 181 |
#endif |
| 182 |
|
| 183 |
TTMATH_LOGC("UInt::Add", c) |
| 184 |
|
| 185 |
return c; |
| 186 |
} |
| 187 |
|
| 188 |
|
| 189 |
|
| 190 |
/*! |
| 191 |
this method adds one word (at a specific position) |
| 192 |
and returns a carry (if it was) |
| 193 |
|
| 194 |
***this method is created only on a 64bit platform*** |
| 195 |
|
| 196 |
|
| 197 |
if we've got (value_size=3): |
| 198 |
table[0] = 10; |
| 199 |
table[1] = 30; |
| 200 |
table[2] = 5; |
| 201 |
and we call: |
| 202 |
AddInt(2,1) |
| 203 |
then it'll be: |
| 204 |
table[0] = 10; |
| 205 |
table[1] = 30 + 2; |
| 206 |
table[2] = 5; |
| 207 |
|
| 208 |
of course if there was a carry from table[2] it would be returned |
| 209 |
*/ |
| 210 |
template<uint value_size> |
| 211 |
uint UInt<value_size>::AddInt(uint value, uint index) |
| 212 |
{ |
| 213 |
uint b = value_size; |
| 214 |
uint * p1 = table; |
| 215 |
uint c; |
| 216 |
|
| 217 |
TTMATH_ASSERT( index < value_size ) |
| 218 |
|
| 219 |
#ifndef __GNUC__ |
| 220 |
c = ttmath_addindexed_x64(p1,b,index,value); |
| 221 |
#endif |
| 222 |
|
| 223 |
|
| 224 |
#ifdef __GNUC__ |
| 225 |
uint dummy, dummy2; |
| 226 |
|
| 227 |
__asm__ __volatile__( |
| 228 |
|
| 229 |
"subq %%rdx, %%rcx \n" |
| 230 |
|
| 231 |
"1: \n" |
| 232 |
"addq %%rax, (%%rbx,%%rdx,8) \n" |
| 233 |
"jnc 2f \n" |
| 234 |
|
| 235 |
"movq $1, %%rax \n" |
| 236 |
"incq %%rdx \n" |
| 237 |
"decq %%rcx \n" |
| 238 |
"jnz 1b \n" |
| 239 |
|
| 240 |
"2: \n" |
| 241 |
"setc %%al \n" |
| 242 |
"movzx %%al, %%rdx \n" |
| 243 |
|
| 244 |
: "=d" (c), "=a" (dummy), "=c" (dummy2) |
| 245 |
: "0" (index), "1" (value), "2" (b), "b" (p1) |
| 246 |
: "cc", "memory" ); |
| 247 |
|
| 248 |
#endif |
| 249 |
|
| 250 |
TTMATH_LOGC("UInt::AddInt", c) |
| 251 |
|
| 252 |
return c; |
| 253 |
} |
| 254 |
|
| 255 |
|
| 256 |
|
| 257 |
/*! |
| 258 |
this method adds only two unsigned words to the existing value |
| 259 |
and these words begin on the 'index' position |
| 260 |
(it's used in the multiplication algorithm 2) |
| 261 |
|
| 262 |
***this method is created only on a 64bit platform*** |
| 263 |
|
| 264 |
index should be equal or smaller than value_size-2 (index <= value_size-2) |
| 265 |
x1 - lower word, x2 - higher word |
| 266 |
|
| 267 |
for example if we've got value_size equal 4 and: |
| 268 |
table[0] = 3 |
| 269 |
table[1] = 4 |
| 270 |
table[2] = 5 |
| 271 |
table[3] = 6 |
| 272 |
then let |
| 273 |
x1 = 10 |
| 274 |
x2 = 20 |
| 275 |
and |
| 276 |
index = 1 |
| 277 |
|
| 278 |
the result of this method will be: |
| 279 |
table[0] = 3 |
| 280 |
table[1] = 4 + x1 = 14 |
| 281 |
table[2] = 5 + x2 = 25 |
| 282 |
table[3] = 6 |
| 283 |
|
| 284 |
and no carry at the end of table[3] |
| 285 |
|
| 286 |
(of course if there was a carry in table[2](5+20) then |
| 287 |
this carry would be passed to the table[3] etc.) |
| 288 |
*/ |
| 289 |
template<uint value_size> |
| 290 |
uint UInt<value_size>::AddTwoInts(uint x2, uint x1, uint index) |
| 291 |
{ |
| 292 |
uint b = value_size; |
| 293 |
uint * p1 = table; |
| 294 |
uint c; |
| 295 |
|
| 296 |
TTMATH_ASSERT( index < value_size - 1 ) |
| 297 |
|
| 298 |
#ifndef __GNUC__ |
| 299 |
c = ttmath_addindexed2_x64(p1,b,index,x1,x2); |
| 300 |
#endif |
| 301 |
|
| 302 |
|
| 303 |
#ifdef __GNUC__ |
| 304 |
uint dummy, dummy2; |
| 305 |
|
| 306 |
__asm__ __volatile__( |
| 307 |
|
| 308 |
"subq %%rdx, %%rcx \n" |
| 309 |
|
| 310 |
"addq %%rsi, (%%rbx,%%rdx,8) \n" |
| 311 |
"incq %%rdx \n" |
| 312 |
"decq %%rcx \n" |
| 313 |
|
| 314 |
"1: \n" |
| 315 |
"adcq %%rax, (%%rbx,%%rdx,8) \n" |
| 316 |
"jnc 2f \n" |
| 317 |
|
| 318 |
"mov $0, %%rax \n" |
| 319 |
"incq %%rdx \n" |
| 320 |
"decq %%rcx \n" |
| 321 |
"jnz 1b \n" |
| 322 |
|
| 323 |
"2: \n" |
| 324 |
"setc %%al \n" |
| 325 |
"movzx %%al, %%rax \n" |
| 326 |
|
| 327 |
: "=a" (c), "=c" (dummy), "=d" (dummy2) |
| 328 |
: "0" (x2), "1" (b), "2" (index), "b" (p1), "S" (x1) |
| 329 |
: "cc", "memory" ); |
| 330 |
|
| 331 |
#endif |
| 332 |
|
| 333 |
TTMATH_LOGC("UInt::AddTwoInts", c) |
| 334 |
|
| 335 |
return c; |
| 336 |
} |
| 337 |
|
| 338 |
|
| 339 |
|
| 340 |
/*! |
| 341 |
this static method addes one vector to the other |
| 342 |
'ss1' is larger in size or equal to 'ss2' |
| 343 |
|
| 344 |
ss1 points to the first (larger) vector |
| 345 |
ss2 points to the second vector |
| 346 |
ss1_size - size of the ss1 (and size of the result too) |
| 347 |
ss2_size - size of the ss2 |
| 348 |
result - is the result vector (which has size the same as ss1: ss1_size) |
| 349 |
|
| 350 |
Example: ss1_size is 5, ss2_size is 3 |
| 351 |
ss1: ss2: result (output): |
| 352 |
5 1 5+1 |
| 353 |
4 3 4+3 |
| 354 |
2 7 2+7 |
| 355 |
6 6 |
| 356 |
9 9 |
| 357 |
of course the carry is propagated and will be returned from the last item |
| 358 |
(this method is used by the Karatsuba multiplication algorithm) |
| 359 |
*/ |
| 360 |
template<uint value_size> |
| 361 |
uint UInt<value_size>::AddVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result) |
| 362 |
{ |
| 363 |
TTMATH_ASSERT( ss1_size >= ss2_size ) |
| 364 |
|
| 365 |
uint c; |
| 366 |
|
| 367 |
#ifndef __GNUC__ |
| 368 |
c = ttmath_addvector_x64(ss1, ss2, ss1_size, ss2_size, result); |
| 369 |
#endif |
| 370 |
|
| 371 |
|
| 372 |
#ifdef __GNUC__ |
| 373 |
uint dummy1, dummy2, dummy3; |
| 374 |
uint rest = ss1_size - ss2_size; |
| 375 |
|
| 376 |
// this part should be compiled with gcc |
| 377 |
|
| 378 |
__asm__ __volatile__( |
| 379 |
"mov %%rdx, %%r8 \n" |
| 380 |
"xor %%rdx, %%rdx \n" // rdx = 0, cf = 0 |
| 381 |
"1: \n" |
| 382 |
"mov (%%rsi,%%rdx,8), %%rax \n" |
| 383 |
"adc (%%rbx,%%rdx,8), %%rax \n" |
| 384 |
"mov %%rax, (%%rdi,%%rdx,8) \n" |
| 385 |
|
| 386 |
"inc %%rdx \n" |
| 387 |
"dec %%rcx \n" |
| 388 |
"jnz 1b \n" |
| 389 |
|
| 390 |
"adc %%rcx, %%rcx \n" // rcx has the cf state |
| 391 |
|
| 392 |
"or %%r8, %%r8 \n" |
| 393 |
"jz 3f \n" |
| 394 |
|
| 395 |
"xor %%rbx, %%rbx \n" // ebx = 0 |
| 396 |
"neg %%rcx \n" // setting cf from rcx |
| 397 |
"mov %%r8, %%rcx \n" // rcx=rest and is != 0 |
| 398 |
"2: \n" |
| 399 |
"mov (%%rsi, %%rdx, 8), %%rax \n" |
| 400 |
"adc %%rbx, %%rax \n" |
| 401 |
"mov %%rax, (%%rdi, %%rdx, 8) \n" |
| 402 |
|
| 403 |
"inc %%rdx \n" |
| 404 |
"dec %%rcx \n" |
| 405 |
"jnz 2b \n" |
| 406 |
|
| 407 |
"adc %%rcx, %%rcx \n" |
| 408 |
"3: \n" |
| 409 |
|
| 410 |
: "=a" (dummy1), "=b" (dummy2), "=c" (c), "=d" (dummy3) |
| 411 |
: "1" (ss2), "2" (ss2_size), "3" (rest), "S" (ss1), "D" (result) |
| 412 |
: "%r8", "cc", "memory" ); |
| 413 |
|
| 414 |
#endif |
| 415 |
|
| 416 |
TTMATH_VECTOR_LOGC("UInt::AddVector", c, result, ss1_size) |
| 417 |
|
| 418 |
return c; |
| 419 |
} |
| 420 |
|
| 421 |
|
| 422 |
|
| 423 |
/*! |
| 424 |
this method's subtracting ss2 from the 'this' and subtracting |
| 425 |
carry if it has been defined |
| 426 |
(this = this - ss2 - c) |
| 427 |
|
| 428 |
***this method is created only on a 64bit platform*** |
| 429 |
|
| 430 |
c must be zero or one (might be a bigger value than 1) |
| 431 |
function returns carry (1) (if it was) |
| 432 |
*/ |
| 433 |
template<uint value_size> |
| 434 |
uint UInt<value_size>::Sub(const UInt<value_size> & ss2, uint c) |
| 435 |
{ |
| 436 |
uint b = value_size; |
| 437 |
uint * p1 = table; |
| 438 |
const uint * p2 = ss2.table; |
| 439 |
|
| 440 |
// we don't have to use TTMATH_REFERENCE_ASSERT here |
| 441 |
// this algorithm doesn't require it |
| 442 |
|
| 443 |
#ifndef __GNUC__ |
| 444 |
c = ttmath_sbb_x64(p1,p2,b,c); |
| 445 |
#endif |
| 446 |
|
| 447 |
|
| 448 |
#ifdef __GNUC__ |
| 449 |
uint dummy, dummy2; |
| 450 |
|
| 451 |
__asm__ __volatile__( |
| 452 |
|
| 453 |
"xorq %%rdx, %%rdx \n" |
| 454 |
"negq %%rax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 455 |
|
| 456 |
"1: \n" |
| 457 |
"movq (%%rsi,%%rdx,8), %%rax \n" |
| 458 |
"sbbq %%rax, (%%rbx,%%rdx,8) \n" |
| 459 |
|
| 460 |
"incq %%rdx \n" |
| 461 |
"decq %%rcx \n" |
| 462 |
"jnz 1b \n" |
| 463 |
|
| 464 |
"adcq %%rcx, %%rcx \n" |
| 465 |
|
| 466 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 467 |
: "0" (b), "1" (c), "b" (p1), "S" (p2) |
| 468 |
: "cc", "memory" ); |
| 469 |
|
| 470 |
#endif |
| 471 |
|
| 472 |
TTMATH_LOGC("UInt::Sub", c) |
| 473 |
|
| 474 |
return c; |
| 475 |
} |
| 476 |
|
| 477 |
|
| 478 |
|
| 479 |
/*! |
| 480 |
this method subtracts one word (at a specific position) |
| 481 |
and returns a carry (if it was) |
| 482 |
|
| 483 |
***this method is created only on a 64bit platform*** |
| 484 |
|
| 485 |
if we've got (value_size=3): |
| 486 |
table[0] = 10; |
| 487 |
table[1] = 30; |
| 488 |
table[2] = 5; |
| 489 |
and we call: |
| 490 |
SubInt(2,1) |
| 491 |
then it'll be: |
| 492 |
table[0] = 10; |
| 493 |
table[1] = 30 - 2; |
| 494 |
table[2] = 5; |
| 495 |
|
| 496 |
of course if there was a carry from table[2] it would be returned |
| 497 |
*/ |
| 498 |
template<uint value_size> |
| 499 |
uint UInt<value_size>::SubInt(uint value, uint index) |
| 500 |
{ |
| 501 |
uint b = value_size; |
| 502 |
uint * p1 = table; |
| 503 |
uint c; |
| 504 |
|
| 505 |
TTMATH_ASSERT( index < value_size ) |
| 506 |
|
| 507 |
#ifndef __GNUC__ |
| 508 |
c = ttmath_subindexed_x64(p1,b,index,value); |
| 509 |
#endif |
| 510 |
|
| 511 |
|
| 512 |
#ifdef __GNUC__ |
| 513 |
uint dummy, dummy2; |
| 514 |
|
| 515 |
__asm__ __volatile__( |
| 516 |
|
| 517 |
"subq %%rdx, %%rcx \n" |
| 518 |
|
| 519 |
"1: \n" |
| 520 |
"subq %%rax, (%%rbx,%%rdx,8) \n" |
| 521 |
"jnc 2f \n" |
| 522 |
|
| 523 |
"movq $1, %%rax \n" |
| 524 |
"incq %%rdx \n" |
| 525 |
"decq %%rcx \n" |
| 526 |
"jnz 1b \n" |
| 527 |
|
| 528 |
"2: \n" |
| 529 |
"setc %%al \n" |
| 530 |
"movzx %%al, %%rdx \n" |
| 531 |
|
| 532 |
: "=d" (c), "=a" (dummy), "=c" (dummy2) |
| 533 |
: "0" (index), "1" (value), "2" (b), "b" (p1) |
| 534 |
: "cc", "memory" ); |
| 535 |
|
| 536 |
#endif |
| 537 |
|
| 538 |
TTMATH_LOGC("UInt::SubInt", c) |
| 539 |
|
| 540 |
return c; |
| 541 |
} |
| 542 |
|
| 543 |
|
| 544 |
/*! |
| 545 |
this static method subtractes one vector from the other |
| 546 |
'ss1' is larger in size or equal to 'ss2' |
| 547 |
|
| 548 |
ss1 points to the first (larger) vector |
| 549 |
ss2 points to the second vector |
| 550 |
ss1_size - size of the ss1 (and size of the result too) |
| 551 |
ss2_size - size of the ss2 |
| 552 |
result - is the result vector (which has size the same as ss1: ss1_size) |
| 553 |
|
| 554 |
Example: ss1_size is 5, ss2_size is 3 |
| 555 |
ss1: ss2: result (output): |
| 556 |
5 1 5-1 |
| 557 |
4 3 4-3 |
| 558 |
2 7 2-7 |
| 559 |
6 6-1 (the borrow from previous item) |
| 560 |
9 9 |
| 561 |
return (carry): 0 |
| 562 |
of course the carry (borrow) is propagated and will be returned from the last item |
| 563 |
(this method is used by the Karatsuba multiplication algorithm) |
| 564 |
*/ |
| 565 |
template<uint value_size> |
| 566 |
uint UInt<value_size>::SubVector(const uint * ss1, const uint * ss2, uint ss1_size, uint ss2_size, uint * result) |
| 567 |
{ |
| 568 |
TTMATH_ASSERT( ss1_size >= ss2_size ) |
| 569 |
|
| 570 |
uint c; |
| 571 |
|
| 572 |
#ifndef __GNUC__ |
| 573 |
c = ttmath_subvector_x64(ss1, ss2, ss1_size, ss2_size, result); |
| 574 |
#endif |
| 575 |
|
| 576 |
|
| 577 |
#ifdef __GNUC__ |
| 578 |
|
| 579 |
// the asm code is nearly the same as in AddVector |
| 580 |
// only two instructions 'adc' are changed to 'sbb' |
| 581 |
|
| 582 |
uint dummy1, dummy2, dummy3; |
| 583 |
uint rest = ss1_size - ss2_size; |
| 584 |
|
| 585 |
__asm__ __volatile__( |
| 586 |
"mov %%rdx, %%r8 \n" |
| 587 |
"xor %%rdx, %%rdx \n" // rdx = 0, cf = 0 |
| 588 |
"1: \n" |
| 589 |
"mov (%%rsi,%%rdx,8), %%rax \n" |
| 590 |
"sbb (%%rbx,%%rdx,8), %%rax \n" |
| 591 |
"mov %%rax, (%%rdi,%%rdx,8) \n" |
| 592 |
|
| 593 |
"inc %%rdx \n" |
| 594 |
"dec %%rcx \n" |
| 595 |
"jnz 1b \n" |
| 596 |
|
| 597 |
"adc %%rcx, %%rcx \n" // rcx has the cf state |
| 598 |
|
| 599 |
"or %%r8, %%r8 \n" |
| 600 |
"jz 3f \n" |
| 601 |
|
| 602 |
"xor %%rbx, %%rbx \n" // ebx = 0 |
| 603 |
"neg %%rcx \n" // setting cf from rcx |
| 604 |
"mov %%r8, %%rcx \n" // rcx=rest and is != 0 |
| 605 |
"2: \n" |
| 606 |
"mov (%%rsi, %%rdx, 8), %%rax \n" |
| 607 |
"sbb %%rbx, %%rax \n" |
| 608 |
"mov %%rax, (%%rdi, %%rdx, 8) \n" |
| 609 |
|
| 610 |
"inc %%rdx \n" |
| 611 |
"dec %%rcx \n" |
| 612 |
"jnz 2b \n" |
| 613 |
|
| 614 |
"adc %%rcx, %%rcx \n" |
| 615 |
"3: \n" |
| 616 |
|
| 617 |
: "=a" (dummy1), "=b" (dummy2), "=c" (c), "=d" (dummy3) |
| 618 |
: "1" (ss2), "2" (ss2_size), "3" (rest), "S" (ss1), "D" (result) |
| 619 |
: "%r8", "cc", "memory" ); |
| 620 |
|
| 621 |
#endif |
| 622 |
|
| 623 |
TTMATH_VECTOR_LOGC("UInt::SubVector", c, result, ss1_size) |
| 624 |
|
| 625 |
return c; |
| 626 |
} |
| 627 |
|
| 628 |
|
| 629 |
/*! |
| 630 |
this method moves all bits into the left hand side |
| 631 |
return value <- this <- c |
| 632 |
|
| 633 |
the lowest *bit* will be held the 'c' and |
| 634 |
the state of one additional bit (on the left hand side) |
| 635 |
will be returned |
| 636 |
|
| 637 |
for example: |
| 638 |
let this is 001010000 |
| 639 |
after Rcl2_one(1) there'll be 010100001 and Rcl2_one returns 0 |
| 640 |
|
| 641 |
***this method is created only on a 64bit platform*** |
| 642 |
*/ |
| 643 |
template<uint value_size> |
| 644 |
uint UInt<value_size>::Rcl2_one(uint c) |
| 645 |
{ |
| 646 |
sint b = value_size; |
| 647 |
uint * p1 = table; |
| 648 |
|
| 649 |
|
| 650 |
#ifndef __GNUC__ |
| 651 |
c = ttmath_rcl_x64(p1,b,c); |
| 652 |
#endif |
| 653 |
|
| 654 |
|
| 655 |
#ifdef __GNUC__ |
| 656 |
uint dummy, dummy2; |
| 657 |
|
| 658 |
__asm__ __volatile__( |
| 659 |
|
| 660 |
"xorq %%rdx, %%rdx \n" // rdx=0 |
| 661 |
"negq %%rax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 662 |
|
| 663 |
"1: \n" |
| 664 |
"rclq $1, (%%rbx, %%rdx, 8) \n" |
| 665 |
|
| 666 |
"incq %%rdx \n" |
| 667 |
"decq %%rcx \n" |
| 668 |
"jnz 1b \n" |
| 669 |
|
| 670 |
"adcq %%rcx, %%rcx \n" |
| 671 |
|
| 672 |
: "=c" (c), "=a" (dummy), "=d" (dummy2) |
| 673 |
: "0" (b), "1" (c), "b" (p1) |
| 674 |
: "cc", "memory" ); |
| 675 |
|
| 676 |
#endif |
| 677 |
|
| 678 |
TTMATH_LOGC("UInt::Rcl2_one", c) |
| 679 |
|
| 680 |
return c; |
| 681 |
} |
| 682 |
|
| 683 |
|
| 684 |
/*! |
| 685 |
this method moves all bits into the right hand side |
| 686 |
c -> this -> return value |
| 687 |
|
| 688 |
the highest *bit* will be held the 'c' and |
| 689 |
the state of one additional bit (on the right hand side) |
| 690 |
will be returned |
| 691 |
|
| 692 |
for example: |
| 693 |
let this is 000000010 |
| 694 |
after Rcr2_one(1) there'll be 100000001 and Rcr2_one returns 0 |
| 695 |
|
| 696 |
***this method is created only on a 64bit platform*** |
| 697 |
*/ |
| 698 |
template<uint value_size> |
| 699 |
uint UInt<value_size>::Rcr2_one(uint c) |
| 700 |
{ |
| 701 |
sint b = value_size; |
| 702 |
uint * p1 = table; |
| 703 |
|
| 704 |
|
| 705 |
#ifndef __GNUC__ |
| 706 |
c = ttmath_rcr_x64(p1,b,c); |
| 707 |
#endif |
| 708 |
|
| 709 |
|
| 710 |
#ifdef __GNUC__ |
| 711 |
uint dummy; |
| 712 |
|
| 713 |
__asm__ __volatile__( |
| 714 |
|
| 715 |
"negq %%rax \n" // CF=1 if rax!=0 , CF=0 if rax==0 |
| 716 |
|
| 717 |
"1: \n" |
| 718 |
"rcrq $1, -8(%%rbx, %%rcx, 8) \n" |
| 719 |
|
| 720 |
"decq %%rcx \n" |
| 721 |
"jnz 1b \n" |
| 722 |
|
| 723 |
"adcq %%rcx, %%rcx \n" |
| 724 |
|
| 725 |
: "=c" (c), "=a" (dummy) |
| 726 |
: "0" (b), "1" (c), "b" (p1) |
| 727 |
: "cc", "memory" ); |
| 728 |
|
| 729 |
#endif |
| 730 |
|
| 731 |
TTMATH_LOGC("UInt::Rcr2_one", c) |
| 732 |
|
| 733 |
return c; |
| 734 |
} |
| 735 |
|
| 736 |
|
| 737 |
|
| 738 |
/*! |
| 739 |
this method moves all bits into the left hand side |
| 740 |
return value <- this <- c |
| 741 |
|
| 742 |
the lowest *bits* will be held the 'c' and |
| 743 |
the state of one additional bit (on the left hand side) |
| 744 |
will be returned |
| 745 |
|
| 746 |
for example: |
| 747 |
let this is 001010000 |
| 748 |
after Rcl2(3, 1) there'll be 010000111 and Rcl2 returns 1 |
| 749 |
|
| 750 |
***this method is created only on a 64bit platform*** |
| 751 |
*/ |
| 752 |
template<uint value_size> |
| 753 |
uint UInt<value_size>::Rcl2(uint bits, uint c) |
| 754 |
{ |
| 755 |
TTMATH_ASSERT( bits>0 && bits<TTMATH_BITS_PER_UINT ) |
| 756 |
|
| 757 |
uint b = value_size; |
| 758 |
uint * p1 = table; |
| 759 |
|
| 760 |
|
| 761 |
#ifndef __GNUC__ |
| 762 |
c = ttmath_rcl2_x64(p1,b,bits,c); |
| 763 |
#endif |
| 764 |
|
| 765 |
|
| 766 |
#ifdef __GNUC__ |
| 767 |
uint dummy, dummy2, dummy3; |
| 768 |
|
| 769 |
__asm__ __volatile__( |
| 770 |
|
| 771 |
"movq %%rcx, %%rsi \n" |
| 772 |
"movq $64, %%rcx \n" |
| 773 |
"subq %%rsi, %%rcx \n" |
| 774 |
"movq $-1, %%rdx \n" |
| 775 |
"shrq %%cl, %%rdx \n" |
| 776 |
"movq %%rdx, %%r8 \n" |
| 777 |
"movq %%rsi, %%rcx \n" |
| 778 |
|
| 779 |
"xorq %%rdx, %%rdx \n" |
| 780 |
"movq %%rdx, %%rsi \n" |
| 781 |
"orq %%rax, %%rax \n" |
| 782 |
"cmovnz %%r8, %%rsi \n" |
| 783 |
|
| 784 |
"1: \n" |
| 785 |
"rolq %%cl, (%%rbx,%%rdx,8) \n" |
| 786 |
|
| 787 |
"movq (%%rbx,%%rdx,8), %%rax \n" |
| 788 |
"andq %%r8, %%rax \n" |
| 789 |
"xorq %%rax, (%%rbx,%%rdx,8) \n" |
| 790 |
"orq %%rsi, (%%rbx,%%rdx,8) \n" |
| 791 |
"movq %%rax, %%rsi \n" |
| 792 |
|
| 793 |
"incq %%rdx \n" |
| 794 |
"decq %%rdi \n" |
| 795 |
"jnz 1b \n" |
| 796 |
|
| 797 |
"and $1, %%rax \n" |
| 798 |
|
| 799 |
: "=a" (c), "=D" (dummy), "=S" (dummy2), "=d" (dummy3) |
| 800 |
: "0" (c), "1" (b), "b" (p1), "c" (bits) |
| 801 |
: "%r8", "cc", "memory" ); |
| 802 |
|
| 803 |
#endif |
| 804 |
|
| 805 |
TTMATH_LOGC("UInt::Rcl2", c) |
| 806 |
|
| 807 |
return c; |
| 808 |
} |
| 809 |
|
| 810 |
|
| 811 |
/*! |
| 812 |
this method moves all bits into the right hand side |
| 813 |
C -> this -> return value |
| 814 |
|
| 815 |
the highest *bits* will be held the 'c' and |
| 816 |
the state of one additional bit (on the right hand side) |
| 817 |
will be returned |
| 818 |
|
| 819 |
for example: |
| 820 |
let this is 000000010 |
| 821 |
after Rcr2(2, 1) there'll be 110000000 and Rcr2 returns 1 |
| 822 |
|
| 823 |
***this method is created only on a 64bit platform*** |
| 824 |
*/ |
| 825 |
template<uint value_size> |
| 826 |
uint UInt<value_size>::Rcr2(uint bits, uint c) |
| 827 |
{ |
| 828 |
TTMATH_ASSERT( bits>0 && bits<TTMATH_BITS_PER_UINT ) |
| 829 |
|
| 830 |
sint b = value_size; |
| 831 |
uint * p1 = table; |
| 832 |
|
| 833 |
|
| 834 |
#ifndef __GNUC__ |
| 835 |
c = ttmath_rcr2_x64(p1,b,bits,c); |
| 836 |
#endif |
| 837 |
|
| 838 |
|
| 839 |
#ifdef __GNUC__ |
| 840 |
uint dummy, dummy2, dummy3; |
| 841 |
|
| 842 |
__asm__ __volatile__( |
| 843 |
|
| 844 |
"movq %%rcx, %%rsi \n" |
| 845 |
"movq $64, %%rcx \n" |
| 846 |
"subq %%rsi, %%rcx \n" |
| 847 |
"movq $-1, %%rdx \n" |
| 848 |
"shlq %%cl, %%rdx \n" |
| 849 |
"movq %%rdx, %%R8 \n" |
| 850 |
"movq %%rsi, %%rcx \n" |
| 851 |
|
| 852 |
"xorq %%rdx, %%rdx \n" |
| 853 |
"movq %%rdx, %%rsi \n" |
| 854 |
"addq %%rdi, %%rdx \n" |
| 855 |
"decq %%rdx \n" |
| 856 |
"orq %%rax, %%rax \n" |
| 857 |
"cmovnz %%R8, %%rsi \n" |
| 858 |
|
| 859 |
"1: \n" |
| 860 |
"rorq %%cl, (%%rbx,%%rdx,8) \n" |
| 861 |
|
| 862 |
"movq (%%rbx,%%rdx,8), %%rax \n" |
| 863 |
"andq %%R8, %%rax \n" |
| 864 |
"xorq %%rax, (%%rbx,%%rdx,8) \n" |
| 865 |
"orq %%rsi, (%%rbx,%%rdx,8) \n" |
| 866 |
"movq %%rax, %%rsi \n" |
| 867 |
|
| 868 |
"decq %%rdx \n" |
| 869 |
"decq %%rdi \n" |
| 870 |
"jnz 1b \n" |
| 871 |
|
| 872 |
"rolq $1, %%rax \n" |
| 873 |
"andq $1, %%rax \n" |
| 874 |
|
| 875 |
: "=a" (c), "=D" (dummy), "=S" (dummy2), "=d" (dummy3) |
| 876 |
: "0" (c), "1" (b), "b" (p1), "c" (bits) |
| 877 |
: "%r8", "cc", "memory" ); |
| 878 |
|
| 879 |
#endif |
| 880 |
|
| 881 |
TTMATH_LOGC("UInt::Rcr2", c) |
| 882 |
|
| 883 |
return c; |
| 884 |
} |
| 885 |
|
| 886 |
|
| 887 |
/* |
| 888 |
this method returns the number of the highest set bit in one 64-bit word |
| 889 |
if the 'x' is zero this method returns '-1' |
| 890 |
|
| 891 |
***this method is created only on a 64bit platform*** |
| 892 |
*/ |
| 893 |
template<uint value_size> |
| 894 |
sint UInt<value_size>::FindLeadingBitInWord(uint x) |
| 895 |
{ |
| 896 |
sint result; |
| 897 |
|
| 898 |
|
| 899 |
#ifndef __GNUC__ |
| 900 |
|
| 901 |
unsigned long nIndex = 0; |
| 902 |
|
| 903 |
if( _BitScanReverse64(&nIndex,x) == 0 ) |
| 904 |
result = -1; |
| 905 |
else |
| 906 |
result = nIndex; |
| 907 |
|
| 908 |
#endif |
| 909 |
|
| 910 |
|
| 911 |
#ifdef __GNUC__ |
| 912 |
uint dummy; |
| 913 |
|
| 914 |
__asm__ ( |
| 915 |
|
| 916 |
"movq $-1, %1 \n" |
| 917 |
"bsrq %2, %0 \n" |
| 918 |
"cmovz %1, %0 \n" |
| 919 |
|
| 920 |
: "=r" (result), "=&r" (dummy) |
| 921 |
: "r" (x) |
| 922 |
: "cc" ); |
| 923 |
|
| 924 |
#endif |
| 925 |
|
| 926 |
|
| 927 |
return result; |
| 928 |
} |
| 929 |
|
| 930 |
|
| 931 |
/* |
| 932 |
this method returns the number of the highest set bit in one 64-bit word |
| 933 |
if the 'x' is zero this method returns '-1' |
| 934 |
|
| 935 |
***this method is created only on a 64bit platform*** |
| 936 |
*/ |
| 937 |
template<uint value_size> |
| 938 |
sint UInt<value_size>::FindLowestBitInWord(uint x) |
| 939 |
{ |
| 940 |
sint result; |
| 941 |
|
| 942 |
|
| 943 |
#ifndef __GNUC__ |
| 944 |
|
| 945 |
unsigned long nIndex = 0; |
| 946 |
|
| 947 |
if( _BitScanForward64(&nIndex,x) == 0 ) |
| 948 |
result = -1; |
| 949 |
else |
| 950 |
result = nIndex; |
| 951 |
|
| 952 |
#endif |
| 953 |
|
| 954 |
|
| 955 |
#ifdef __GNUC__ |
| 956 |
uint dummy; |
| 957 |
|
| 958 |
__asm__ ( |
| 959 |
|
| 960 |
"movq $-1, %1 \n" |
| 961 |
"bsfq %2, %0 \n" |
| 962 |
"cmovz %1, %0 \n" |
| 963 |
|
| 964 |
: "=r" (result), "=&r" (dummy) |
| 965 |
: "r" (x) |
| 966 |
: "cc" ); |
| 967 |
|
| 968 |
#endif |
| 969 |
|
| 970 |
|
| 971 |
return result; |
| 972 |
} |
| 973 |
|
| 974 |
|
| 975 |
/*! |
| 976 |
this method sets a special bit in the 'value' |
| 977 |
and returns the last state of the bit (zero or one) |
| 978 |
|
| 979 |
***this method is created only on a 64bit platform*** |
| 980 |
|
| 981 |
bit is from <0,63> |
| 982 |
|
| 983 |
e.g. |
| 984 |
uint x = 100; |
| 985 |
uint bit = SetBitInWord(x, 3); |
| 986 |
now: x = 108 and bit = 0 |
| 987 |
*/ |
| 988 |
template<uint value_size> |
| 989 |
uint UInt<value_size>::SetBitInWord(uint & value, uint bit) |
| 990 |
{ |
| 991 |
TTMATH_ASSERT( bit < TTMATH_BITS_PER_UINT ) |
| 992 |
|
| 993 |
uint old_bit; |
| 994 |
uint v = value; |
| 995 |
|
| 996 |
|
| 997 |
#ifndef __GNUC__ |
| 998 |
old_bit = _bittestandset64((__int64*)&value,bit) != 0; |
| 999 |
#endif |
| 1000 |
|
| 1001 |
|
| 1002 |
#ifdef __GNUC__ |
| 1003 |
|
| 1004 |
__asm__ ( |
| 1005 |
|
| 1006 |
"btsq %%rbx, %%rax \n" |
| 1007 |
"setc %%bl \n" |
| 1008 |
"movzx %%bl, %%rbx \n" |
| 1009 |
|
| 1010 |
: "=a" (v), "=b" (old_bit) |
| 1011 |
: "0" (v), "1" (bit) |
| 1012 |
: "cc" ); |
| 1013 |
|
| 1014 |
#endif |
| 1015 |
|
| 1016 |
value = v; |
| 1017 |
|
| 1018 |
return old_bit; |
| 1019 |
} |
| 1020 |
|
| 1021 |
|
| 1022 |
/*! |
| 1023 |
* |
| 1024 |
* Multiplication |
| 1025 |
* |
| 1026 |
* |
| 1027 |
*/ |
| 1028 |
|
| 1029 |
|
| 1030 |
/*! |
| 1031 |
multiplication: result_high:result_low = a * b |
| 1032 |
result_high - higher word of the result |
| 1033 |
result_low - lower word of the result |
| 1034 |
|
| 1035 |
this methos never returns a carry |
| 1036 |
this method is used in the second version of the multiplication algorithms |
| 1037 |
|
| 1038 |
***this method is created only on a 64bit platform*** |
| 1039 |
*/ |
| 1040 |
template<uint value_size> |
| 1041 |
void UInt<value_size>::MulTwoWords(uint a, uint b, uint * result_high, uint * result_low) |
| 1042 |
{ |
| 1043 |
/* |
| 1044 |
we must use these temporary variables in order to inform the compilator |
| 1045 |
that value pointed with result1 and result2 has changed |
| 1046 |
|
| 1047 |
this has no effect in visual studio but it's usefull when |
| 1048 |
using gcc and options like -O |
| 1049 |
*/ |
| 1050 |
uint result1_; |
| 1051 |
uint result2_; |
| 1052 |
|
| 1053 |
|
| 1054 |
#ifndef __GNUC__ |
| 1055 |
result1_ = _umul128(a,b,&result2_); |
| 1056 |
#endif |
| 1057 |
|
| 1058 |
|
| 1059 |
#ifdef __GNUC__ |
| 1060 |
|
| 1061 |
__asm__ ( |
| 1062 |
|
| 1063 |
"mulq %%rdx \n" |
| 1064 |
|
| 1065 |
: "=a" (result1_), "=d" (result2_) |
| 1066 |
: "0" (a), "1" (b) |
| 1067 |
: "cc" ); |
| 1068 |
|
| 1069 |
#endif |
| 1070 |
|
| 1071 |
|
| 1072 |
*result_low = result1_; |
| 1073 |
*result_high = result2_; |
| 1074 |
} |
| 1075 |
|
| 1076 |
|
| 1077 |
|
| 1078 |
|
| 1079 |
/*! |
| 1080 |
* |
| 1081 |
* Division |
| 1082 |
* |
| 1083 |
* |
| 1084 |
*/ |
| 1085 |
|
| 1086 |
|
| 1087 |
/*! |
| 1088 |
this method calculates 64bits word a:b / 32bits c (a higher, b lower word) |
| 1089 |
r = a:b / c and rest - remainder |
| 1090 |
|
| 1091 |
***this method is created only on a 64bit platform*** |
| 1092 |
|
| 1093 |
* |
| 1094 |
* WARNING: |
| 1095 |
* if r (one word) is too small for the result or c is equal zero |
| 1096 |
* there'll be a hardware interruption (0) |
| 1097 |
* and probably the end of your program |
| 1098 |
* |
| 1099 |
*/ |
| 1100 |
template<uint value_size> |
| 1101 |
void UInt<value_size>::DivTwoWords(uint a,uint b, uint c, uint * r, uint * rest) |
| 1102 |
{ |
| 1103 |
uint r_; |
| 1104 |
uint rest_; |
| 1105 |
/* |
| 1106 |
these variables have similar meaning like those in |
| 1107 |
the multiplication algorithm MulTwoWords |
| 1108 |
*/ |
| 1109 |
|
| 1110 |
TTMATH_ASSERT( c != 0 ) |
| 1111 |
|
| 1112 |
|
| 1113 |
#ifndef __GNUC__ |
| 1114 |
|
| 1115 |
ttmath_div_x64(&a,&b,c); |
| 1116 |
r_ = a; |
| 1117 |
rest_ = b; |
| 1118 |
|
| 1119 |
#endif |
| 1120 |
|
| 1121 |
|
| 1122 |
#ifdef __GNUC__ |
| 1123 |
|
| 1124 |
__asm__ ( |
| 1125 |
|
| 1126 |
"divq %%rcx \n" |
| 1127 |
|
| 1128 |
: "=a" (r_), "=d" (rest_) |
| 1129 |
: "d" (a), "a" (b), "c" (c) |
| 1130 |
: "cc" ); |
| 1131 |
|
| 1132 |
#endif |
| 1133 |
|
| 1134 |
|
| 1135 |
*r = r_; |
| 1136 |
*rest = rest_; |
| 1137 |
} |
| 1138 |
|
| 1139 |
} //namespace |
| 1140 |
|
| 1141 |
|
| 1142 |
#endif //ifdef TTMATH_PLATFORM64 |
| 1143 |
#endif //ifndef TTMATH_NOASM |
| 1144 |
#endif |
| 1145 |
|
| 1146 |
|