| 1 |
dnl X86-64 mpn_redc_1 optimised for Intel Nehalem and Westmere. |
| 2 |
|
| 3 |
dnl Contributed to the GNU project by Torbjörn Granlund. |
| 4 |
|
| 5 |
dnl Copyright 2003-2005, 2007, 2008, 2011-2013 Free Software Foundation, Inc. |
| 6 |
|
| 7 |
dnl This file is part of the GNU MP Library. |
| 8 |
dnl |
| 9 |
dnl The GNU MP Library is free software; you can redistribute it and/or modify |
| 10 |
dnl it under the terms of either: |
| 11 |
dnl |
| 12 |
dnl * the GNU Lesser General Public License as published by the Free |
| 13 |
dnl Software Foundation; either version 3 of the License, or (at your |
| 14 |
dnl option) any later version. |
| 15 |
dnl |
| 16 |
dnl or |
| 17 |
dnl |
| 18 |
dnl * the GNU General Public License as published by the Free Software |
| 19 |
dnl Foundation; either version 2 of the License, or (at your option) any |
| 20 |
dnl later version. |
| 21 |
dnl |
| 22 |
dnl or both in parallel, as here. |
| 23 |
dnl |
| 24 |
dnl The GNU MP Library is distributed in the hope that it will be useful, but |
| 25 |
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 26 |
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 27 |
dnl for more details. |
| 28 |
dnl |
| 29 |
dnl You should have received copies of the GNU General Public License and the |
| 30 |
dnl GNU Lesser General Public License along with the GNU MP Library. If not, |
| 31 |
dnl see https://www.gnu.org/licenses/. |
| 32 |
|
| 33 |
include(`../config.m4') |
| 34 |
|
| 35 |
C cycles/limb |
| 36 |
C AMD K8,K9 ? |
| 37 |
C AMD K10 ? |
| 38 |
C AMD bull ? |
| 39 |
C AMD pile ? |
| 40 |
C AMD steam ? |
| 41 |
C AMD bobcat ? |
| 42 |
C AMD jaguar ? |
| 43 |
C Intel P4 ? |
| 44 |
C Intel core ? |
| 45 |
C Intel NHM ? |
| 46 |
C Intel SBR ? |
| 47 |
C Intel IBR ? |
| 48 |
C Intel HWL ? |
| 49 |
C Intel BWL ? |
| 50 |
C Intel atom ? |
| 51 |
C VIA nano ? |
| 52 |
|
| 53 |
C The inner loops of this code are the result of running a code generation and |
| 54 |
C optimisation tool suite written by David Harvey and Torbjörn Granlund. |
| 55 |
|
| 56 |
C TODO |
| 57 |
C * Micro-optimise, none performed thus far. |
| 58 |
C * Consider inlining mpn_add_n. |
| 59 |
C * Single basecases out before the pushes. |
| 60 |
|
| 61 |
C When playing with pointers, set this to $2 to fall back to conservative |
| 62 |
C indexing in wind-down code. |
| 63 |
define(`I',`$1') |
| 64 |
|
| 65 |
define(`rp', `%rdi') C rcx |
| 66 |
define(`up', `%rsi') C rdx |
| 67 |
define(`mp_param', `%rdx') C r8 |
| 68 |
define(`n', `%rcx') C r9 |
| 69 |
define(`u0inv', `%r8') C stack |
| 70 |
|
| 71 |
define(`i', `%r14') |
| 72 |
define(`j', `%r15') |
| 73 |
define(`mp', `%r12') |
| 74 |
define(`q0', `%r13') |
| 75 |
|
| 76 |
C rax rbx rcx rdx rdi rsi rbp r8 r9 r10 r11 r12 r13 r14 r15 |
| 77 |
|
| 78 |
ABI_SUPPORT(DOS64) |
| 79 |
ABI_SUPPORT(STD64) |
| 80 |
|
| 81 |
define(`ALIGNx', `ALIGN(16)') |
| 82 |
|
| 83 |
ASM_START() |
| 84 |
TEXT |
| 85 |
ALIGN(32) |
| 86 |
PROLOGUE(mpn_redc_1) |
| 87 |
FUNC_ENTRY(4) |
| 88 |
IFDOS(` mov 56(%rsp), %r8 ') |
| 89 |
push %rbx |
| 90 |
push %rbp |
| 91 |
push %r12 |
| 92 |
push %r13 |
| 93 |
push %r14 |
| 94 |
push %r15 |
| 95 |
|
| 96 |
mov (up), q0 |
| 97 |
mov n, j C outer loop induction var |
| 98 |
lea (mp_param,n,8), mp |
| 99 |
lea (up,n,8), up |
| 100 |
neg n |
| 101 |
imul u0inv, q0 C first iteration q0 |
| 102 |
|
| 103 |
test $1, R8(n) |
| 104 |
jz L(bx0) |
| 105 |
|
| 106 |
L(bx1): test $2, R8(n) |
| 107 |
jz L(b3) |
| 108 |
|
| 109 |
L(b1): cmp $-1, R32(n) |
| 110 |
jz L(n1) |
| 111 |
|
| 112 |
L(otp1):lea 3(n), i |
| 113 |
mov (mp,n,8), %rax |
| 114 |
mov (up,n,8), %rbp |
| 115 |
mul q0 |
| 116 |
add %rax, %rbp |
| 117 |
mov $0, R32(%r9) |
| 118 |
mov 8(mp,n,8), %rax |
| 119 |
adc %rdx, %r9 |
| 120 |
mul q0 |
| 121 |
mov $0, R32(%r11) |
| 122 |
mov 8(up,n,8), %rbx |
| 123 |
add %rax, %rbx |
| 124 |
mov 16(mp,n,8), %rax |
| 125 |
adc %rdx, %r11 |
| 126 |
add %r9, %rbx |
| 127 |
adc $0, %r11 |
| 128 |
mov 16(up,n,8), %rbp |
| 129 |
mul q0 |
| 130 |
add %rax, %rbp |
| 131 |
mov $0, R32(%r9) |
| 132 |
mov 24(mp,n,8), %rax |
| 133 |
adc %rdx, %r9 |
| 134 |
mov %rbx, 8(up,n,8) |
| 135 |
imul u0inv, %rbx C next q limb |
| 136 |
jmp L(e1) |
| 137 |
|
| 138 |
ALIGNx |
| 139 |
L(tp1): mul q0 |
| 140 |
add %rax, %rbp |
| 141 |
mov $0, R32(%r9) |
| 142 |
mov -16(mp,i,8), %rax |
| 143 |
adc %rdx, %r9 |
| 144 |
mul q0 |
| 145 |
add %r11, %rbp |
| 146 |
mov $0, R32(%r11) |
| 147 |
mov -16(up,i,8), %r10 |
| 148 |
adc $0, %r9 |
| 149 |
add %rax, %r10 |
| 150 |
mov -8(mp,i,8), %rax |
| 151 |
adc %rdx, %r11 |
| 152 |
mov %rbp, -24(up,i,8) |
| 153 |
add %r9, %r10 |
| 154 |
adc $0, %r11 |
| 155 |
mov -8(up,i,8), %rbp |
| 156 |
mul q0 |
| 157 |
add %rax, %rbp |
| 158 |
mov $0, R32(%r9) |
| 159 |
mov (mp,i,8), %rax |
| 160 |
adc %rdx, %r9 |
| 161 |
mov %r10, -16(up,i,8) |
| 162 |
L(e1): add %r11, %rbp |
| 163 |
adc $0, %r9 |
| 164 |
mul q0 |
| 165 |
mov (up,i,8), %r10 |
| 166 |
mov $0, R32(%r11) |
| 167 |
add %rax, %r10 |
| 168 |
mov 8(mp,i,8), %rax |
| 169 |
adc %rdx, %r11 |
| 170 |
mov %rbp, -8(up,i,8) |
| 171 |
add %r9, %r10 |
| 172 |
adc $0, %r11 |
| 173 |
mov 8(up,i,8), %rbp |
| 174 |
mov %r10, (up,i,8) |
| 175 |
add $4, i |
| 176 |
jnc L(tp1) |
| 177 |
|
| 178 |
L(ed1): mul q0 |
| 179 |
add %rax, %rbp |
| 180 |
adc $0, %rdx |
| 181 |
add %r11, %rbp |
| 182 |
adc $0, %rdx |
| 183 |
mov %rbp, I(-8(up),-24(up,i,8)) |
| 184 |
mov %rdx, (up,n,8) C up[0] |
| 185 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 186 |
lea 8(up), up C up++ |
| 187 |
dec j |
| 188 |
jnz L(otp1) |
| 189 |
jmp L(cj) |
| 190 |
|
| 191 |
L(b3): cmp $-3, R32(n) |
| 192 |
jz L(n3) |
| 193 |
|
| 194 |
L(otp3):lea 5(n), i |
| 195 |
mov (mp,n,8), %rax |
| 196 |
mov (up,n,8), %rbp |
| 197 |
mul q0 |
| 198 |
add %rax, %rbp |
| 199 |
mov $0, R32(%r9) |
| 200 |
mov 8(mp,n,8), %rax |
| 201 |
adc %rdx, %r9 |
| 202 |
mul q0 |
| 203 |
mov 8(up,n,8), %rbx |
| 204 |
mov $0, R32(%r11) |
| 205 |
add %rax, %rbx |
| 206 |
mov 16(mp,n,8), %rax |
| 207 |
adc %rdx, %r11 |
| 208 |
add %r9, %rbx |
| 209 |
adc $0, %r11 |
| 210 |
mov 16(up,n,8), %rbp |
| 211 |
mov %rbx, 8(up,n,8) |
| 212 |
imul u0inv, %rbx C next q limb |
| 213 |
C jmp L(tp3) |
| 214 |
|
| 215 |
ALIGNx |
| 216 |
L(tp3): mul q0 |
| 217 |
add %rax, %rbp |
| 218 |
mov $0, R32(%r9) |
| 219 |
mov -16(mp,i,8), %rax |
| 220 |
adc %rdx, %r9 |
| 221 |
mul q0 |
| 222 |
add %r11, %rbp |
| 223 |
mov $0, R32(%r11) |
| 224 |
mov -16(up,i,8), %r10 |
| 225 |
adc $0, %r9 |
| 226 |
add %rax, %r10 |
| 227 |
mov -8(mp,i,8), %rax |
| 228 |
adc %rdx, %r11 |
| 229 |
mov %rbp, -24(up,i,8) |
| 230 |
add %r9, %r10 |
| 231 |
adc $0, %r11 |
| 232 |
mov -8(up,i,8), %rbp |
| 233 |
mul q0 |
| 234 |
add %rax, %rbp |
| 235 |
mov $0, R32(%r9) |
| 236 |
mov (mp,i,8), %rax |
| 237 |
adc %rdx, %r9 |
| 238 |
mov %r10, -16(up,i,8) |
| 239 |
add %r11, %rbp |
| 240 |
adc $0, %r9 |
| 241 |
mul q0 |
| 242 |
mov (up,i,8), %r10 |
| 243 |
mov $0, R32(%r11) |
| 244 |
add %rax, %r10 |
| 245 |
mov 8(mp,i,8), %rax |
| 246 |
adc %rdx, %r11 |
| 247 |
mov %rbp, -8(up,i,8) |
| 248 |
add %r9, %r10 |
| 249 |
adc $0, %r11 |
| 250 |
mov 8(up,i,8), %rbp |
| 251 |
mov %r10, (up,i,8) |
| 252 |
add $4, i |
| 253 |
jnc L(tp3) |
| 254 |
|
| 255 |
L(ed3): mul q0 |
| 256 |
add %rax, %rbp |
| 257 |
adc $0, %rdx |
| 258 |
add %r11, %rbp |
| 259 |
adc $0, %rdx |
| 260 |
mov %rbp, I(-8(up),-24(up,i,8)) |
| 261 |
mov %rdx, (up,n,8) C up[0] |
| 262 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 263 |
lea 8(up), up C up++ |
| 264 |
dec j |
| 265 |
jnz L(otp3) |
| 266 |
C jmp L(cj) |
| 267 |
|
| 268 |
L(cj): |
| 269 |
IFSTD(` lea (up,n,8), up C param 2: up |
| 270 |
lea (up,n,8), %rdx C param 3: up - n |
| 271 |
neg R32(n) ') C param 4: n |
| 272 |
|
| 273 |
IFDOS(` lea (up,n,8), %rdx C param 2: up |
| 274 |
lea (%rdx,n,8), %r8 C param 3: up - n |
| 275 |
neg R32(n) |
| 276 |
mov n, %r9 C param 4: n |
| 277 |
mov rp, %rcx ') C param 1: rp |
| 278 |
|
| 279 |
CALL( mpn_add_n) |
| 280 |
|
| 281 |
L(ret): pop %r15 |
| 282 |
pop %r14 |
| 283 |
pop %r13 |
| 284 |
pop %r12 |
| 285 |
pop %rbp |
| 286 |
pop %rbx |
| 287 |
FUNC_EXIT() |
| 288 |
ret |
| 289 |
|
| 290 |
L(bx0): test $2, R8(n) |
| 291 |
jnz L(b2) |
| 292 |
|
| 293 |
L(b0): |
| 294 |
L(otp0):lea 2(n), i |
| 295 |
mov (mp,n,8), %rax |
| 296 |
mul q0 |
| 297 |
mov $0, R32(%r11) |
| 298 |
mov (up,n,8), %r10 |
| 299 |
add %rax, %r10 |
| 300 |
mov 8(mp,n,8), %rax |
| 301 |
adc %rdx, %r11 |
| 302 |
mov 8(up,n,8), %rbx |
| 303 |
mul q0 |
| 304 |
add %rax, %rbx |
| 305 |
mov $0, R32(%r9) |
| 306 |
mov 16(mp,n,8), %rax |
| 307 |
adc %rdx, %r9 |
| 308 |
add %r11, %rbx |
| 309 |
adc $0, %r9 |
| 310 |
mul q0 |
| 311 |
mov 16(up,n,8), %r10 |
| 312 |
mov $0, R32(%r11) |
| 313 |
add %rax, %r10 |
| 314 |
mov 24(mp,n,8), %rax |
| 315 |
adc %rdx, %r11 |
| 316 |
mov %rbx, 8(up,n,8) |
| 317 |
imul u0inv, %rbx C next q limb |
| 318 |
jmp L(e0) |
| 319 |
|
| 320 |
ALIGNx |
| 321 |
L(tp0): mul q0 |
| 322 |
add %rax, %rbp |
| 323 |
mov $0, R32(%r9) |
| 324 |
mov -16(mp,i,8), %rax |
| 325 |
adc %rdx, %r9 |
| 326 |
mul q0 |
| 327 |
add %r11, %rbp |
| 328 |
mov $0, R32(%r11) |
| 329 |
mov -16(up,i,8), %r10 |
| 330 |
adc $0, %r9 |
| 331 |
add %rax, %r10 |
| 332 |
mov -8(mp,i,8), %rax |
| 333 |
adc %rdx, %r11 |
| 334 |
mov %rbp, -24(up,i,8) |
| 335 |
add %r9, %r10 |
| 336 |
adc $0, %r11 |
| 337 |
mov -8(up,i,8), %rbp |
| 338 |
mul q0 |
| 339 |
add %rax, %rbp |
| 340 |
mov $0, R32(%r9) |
| 341 |
mov (mp,i,8), %rax |
| 342 |
adc %rdx, %r9 |
| 343 |
mov %r10, -16(up,i,8) |
| 344 |
add %r11, %rbp |
| 345 |
adc $0, %r9 |
| 346 |
mul q0 |
| 347 |
mov (up,i,8), %r10 |
| 348 |
mov $0, R32(%r11) |
| 349 |
add %rax, %r10 |
| 350 |
mov 8(mp,i,8), %rax |
| 351 |
adc %rdx, %r11 |
| 352 |
mov %rbp, -8(up,i,8) |
| 353 |
L(e0): add %r9, %r10 |
| 354 |
adc $0, %r11 |
| 355 |
mov 8(up,i,8), %rbp |
| 356 |
mov %r10, (up,i,8) |
| 357 |
add $4, i |
| 358 |
jnc L(tp0) |
| 359 |
|
| 360 |
L(ed0): mul q0 |
| 361 |
add %rax, %rbp |
| 362 |
adc $0, %rdx |
| 363 |
add %r11, %rbp |
| 364 |
adc $0, %rdx |
| 365 |
mov %rbp, I(-8(up),-24(up,i,8)) |
| 366 |
mov %rdx, (up,n,8) C up[0] |
| 367 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 368 |
lea 8(up), up C up++ |
| 369 |
dec j |
| 370 |
jnz L(otp0) |
| 371 |
jmp L(cj) |
| 372 |
|
| 373 |
L(b2): cmp $-2, R32(n) |
| 374 |
jz L(n2) |
| 375 |
|
| 376 |
L(otp2):lea 4(n), i |
| 377 |
mov (mp,n,8), %rax |
| 378 |
mul q0 |
| 379 |
mov (up,n,8), %r10 |
| 380 |
mov $0, R32(%r11) |
| 381 |
add %rax, %r10 |
| 382 |
mov 8(mp,n,8), %rax |
| 383 |
adc %rdx, %r11 |
| 384 |
mov 8(up,n,8), %rbx |
| 385 |
mul q0 |
| 386 |
add %rax, %rbx |
| 387 |
mov $0, R32(%r9) |
| 388 |
mov 16(mp,n,8), %rax |
| 389 |
adc %rdx, %r9 |
| 390 |
mul q0 |
| 391 |
add %r11, %rbx |
| 392 |
mov $0, R32(%r11) |
| 393 |
mov 16(up,n,8), %r10 |
| 394 |
adc $0, %r9 |
| 395 |
add %rax, %r10 |
| 396 |
mov 24(mp,n,8), %rax |
| 397 |
adc %rdx, %r11 |
| 398 |
mov %rbx, 8(up,n,8) |
| 399 |
imul u0inv, %rbx C next q limb |
| 400 |
jmp L(e2) |
| 401 |
|
| 402 |
ALIGNx |
| 403 |
L(tp2): mul q0 |
| 404 |
add %rax, %rbp |
| 405 |
mov $0, R32(%r9) |
| 406 |
mov -16(mp,i,8), %rax |
| 407 |
adc %rdx, %r9 |
| 408 |
mul q0 |
| 409 |
add %r11, %rbp |
| 410 |
mov $0, R32(%r11) |
| 411 |
mov -16(up,i,8), %r10 |
| 412 |
adc $0, %r9 |
| 413 |
add %rax, %r10 |
| 414 |
mov -8(mp,i,8), %rax |
| 415 |
adc %rdx, %r11 |
| 416 |
mov %rbp, -24(up,i,8) |
| 417 |
L(e2): add %r9, %r10 |
| 418 |
adc $0, %r11 |
| 419 |
mov -8(up,i,8), %rbp |
| 420 |
mul q0 |
| 421 |
add %rax, %rbp |
| 422 |
mov $0, R32(%r9) |
| 423 |
mov (mp,i,8), %rax |
| 424 |
adc %rdx, %r9 |
| 425 |
mov %r10, -16(up,i,8) |
| 426 |
add %r11, %rbp |
| 427 |
adc $0, %r9 |
| 428 |
mul q0 |
| 429 |
mov (up,i,8), %r10 |
| 430 |
mov $0, R32(%r11) |
| 431 |
add %rax, %r10 |
| 432 |
mov 8(mp,i,8), %rax |
| 433 |
adc %rdx, %r11 |
| 434 |
mov %rbp, -8(up,i,8) |
| 435 |
add %r9, %r10 |
| 436 |
adc $0, %r11 |
| 437 |
mov 8(up,i,8), %rbp |
| 438 |
mov %r10, (up,i,8) |
| 439 |
add $4, i |
| 440 |
jnc L(tp2) |
| 441 |
|
| 442 |
L(ed2): mul q0 |
| 443 |
add %rax, %rbp |
| 444 |
adc $0, %rdx |
| 445 |
add %r11, %rbp |
| 446 |
adc $0, %rdx |
| 447 |
mov %rbp, I(-8(up),-24(up,i,8)) |
| 448 |
mov %rdx, (up,n,8) C up[0] |
| 449 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 450 |
lea 8(up), up C up++ |
| 451 |
dec j |
| 452 |
jnz L(otp2) |
| 453 |
jmp L(cj) |
| 454 |
|
| 455 |
L(n1): mov (mp_param), %rax |
| 456 |
mul q0 |
| 457 |
add -8(up), %rax |
| 458 |
adc (up), %rdx |
| 459 |
mov %rdx, (rp) |
| 460 |
mov $0, R32(%rax) |
| 461 |
adc R32(%rax), R32(%rax) |
| 462 |
jmp L(ret) |
| 463 |
|
| 464 |
L(n2): mov (mp_param), %rax |
| 465 |
mov -16(up), %rbp |
| 466 |
mul q0 |
| 467 |
add %rax, %rbp |
| 468 |
mov %rdx, %r9 |
| 469 |
adc $0, %r9 |
| 470 |
mov -8(mp), %rax |
| 471 |
mov -8(up), %r10 |
| 472 |
mul q0 |
| 473 |
add %rax, %r10 |
| 474 |
mov %rdx, %r11 |
| 475 |
adc $0, %r11 |
| 476 |
add %r9, %r10 |
| 477 |
adc $0, %r11 |
| 478 |
mov %r10, q0 |
| 479 |
imul u0inv, q0 C next q0 |
| 480 |
mov -16(mp), %rax |
| 481 |
mul q0 |
| 482 |
add %rax, %r10 |
| 483 |
mov %rdx, %r9 |
| 484 |
adc $0, %r9 |
| 485 |
mov -8(mp), %rax |
| 486 |
mov (up), %r14 |
| 487 |
mul q0 |
| 488 |
add %rax, %r14 |
| 489 |
adc $0, %rdx |
| 490 |
add %r9, %r14 |
| 491 |
adc $0, %rdx |
| 492 |
xor R32(%rax), R32(%rax) |
| 493 |
add %r11, %r14 |
| 494 |
adc 8(up), %rdx |
| 495 |
mov %r14, (rp) |
| 496 |
mov %rdx, 8(rp) |
| 497 |
adc R32(%rax), R32(%rax) |
| 498 |
jmp L(ret) |
| 499 |
|
| 500 |
ALIGNx |
| 501 |
L(n3): mov -24(mp), %rax |
| 502 |
mov -24(up), %r10 |
| 503 |
mul q0 |
| 504 |
add %rax, %r10 |
| 505 |
mov -16(mp), %rax |
| 506 |
mov %rdx, %r11 |
| 507 |
adc $0, %r11 |
| 508 |
mov -16(up), %rbp |
| 509 |
mul q0 |
| 510 |
add %rax, %rbp |
| 511 |
mov %rdx, %r9 |
| 512 |
adc $0, %r9 |
| 513 |
mov -8(mp), %rax |
| 514 |
add %r11, %rbp |
| 515 |
mov -8(up), %r10 |
| 516 |
adc $0, %r9 |
| 517 |
mul q0 |
| 518 |
mov %rbp, q0 |
| 519 |
imul u0inv, q0 C next q0 |
| 520 |
add %rax, %r10 |
| 521 |
mov %rdx, %r11 |
| 522 |
adc $0, %r11 |
| 523 |
mov %rbp, -16(up) |
| 524 |
add %r9, %r10 |
| 525 |
adc $0, %r11 |
| 526 |
mov %r10, -8(up) |
| 527 |
mov %r11, -24(up) C up[0] |
| 528 |
lea 8(up), up C up++ |
| 529 |
dec j |
| 530 |
jnz L(n3) |
| 531 |
|
| 532 |
mov -48(up), %rdx |
| 533 |
mov -40(up), %rbx |
| 534 |
xor R32(%rax), R32(%rax) |
| 535 |
add %rbp, %rdx |
| 536 |
adc %r10, %rbx |
| 537 |
adc -8(up), %r11 |
| 538 |
mov %rdx, (rp) |
| 539 |
mov %rbx, 8(rp) |
| 540 |
mov %r11, 16(rp) |
| 541 |
adc R32(%rax), R32(%rax) |
| 542 |
jmp L(ret) |
| 543 |
EPILOGUE() |
| 544 |
ASM_END() |