| 1 |
dnl X86-64 mpn_redc_1 optimised for Intel Conroe and Wolfdale. |
| 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 4.5 (fluctuating) |
| 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 |
C * Keep up[i] in registers for basecases (might require pushes). |
| 61 |
|
| 62 |
C When playing with pointers, set this to $2 to fall back to conservative |
| 63 |
C indexing in wind-down code. |
| 64 |
define(`I',`$1') |
| 65 |
|
| 66 |
define(`rp', `%rdi') C rcx |
| 67 |
define(`up', `%rsi') C rdx |
| 68 |
define(`mp_param', `%rdx') C r8 |
| 69 |
define(`n', `%rcx') C r9 |
| 70 |
define(`u0inv', `%r8') C stack |
| 71 |
|
| 72 |
define(`i', `%r14') |
| 73 |
define(`j', `%r15') |
| 74 |
define(`mp', `%r12') |
| 75 |
define(`q0', `%r13') |
| 76 |
|
| 77 |
C rax rbx rcx rdx rdi rsi rbp r8 r9 r10 r11 r12 r13 r14 r15 |
| 78 |
C X q0' n X rp up u0i mp q0 i j |
| 79 |
|
| 80 |
ABI_SUPPORT(DOS64) |
| 81 |
ABI_SUPPORT(STD64) |
| 82 |
|
| 83 |
define(`ALIGNx', `ALIGN(16)') |
| 84 |
|
| 85 |
ASM_START() |
| 86 |
TEXT |
| 87 |
ALIGN(32) |
| 88 |
PROLOGUE(mpn_redc_1) |
| 89 |
FUNC_ENTRY(4) |
| 90 |
IFDOS(` mov 56(%rsp), %r8 ') |
| 91 |
push %rbx |
| 92 |
push %rbp |
| 93 |
push %r12 |
| 94 |
push %r13 |
| 95 |
push %r14 |
| 96 |
push %r15 |
| 97 |
|
| 98 |
mov (up), q0 |
| 99 |
mov n, j C outer loop induction var |
| 100 |
lea (mp_param,n,8), mp |
| 101 |
lea -16(up,n,8), up |
| 102 |
neg n |
| 103 |
imul u0inv, q0 C first iteration q0 |
| 104 |
|
| 105 |
test $1, R8(n) |
| 106 |
jz L(b0) |
| 107 |
|
| 108 |
L(b1): cmp $-1, R32(n) |
| 109 |
jz L(n1) |
| 110 |
cmp $-3, R32(n) |
| 111 |
jz L(n3) |
| 112 |
|
| 113 |
push rp |
| 114 |
|
| 115 |
L(otp1):lea 3(n), i |
| 116 |
mov (mp,n,8), %rax |
| 117 |
mul q0 |
| 118 |
lea (%rax), %rbp |
| 119 |
mov 8(mp,n,8), %rax |
| 120 |
lea (%rdx), %r9 |
| 121 |
mul q0 |
| 122 |
lea (%rax), %r11 |
| 123 |
mov 16(mp,n,8), %rax |
| 124 |
mov 16(up,n,8), %r10 |
| 125 |
lea (%rdx), %rdi |
| 126 |
mul q0 |
| 127 |
add %rbp, %r10 |
| 128 |
lea (%rax), %rbp |
| 129 |
mov 24(mp,n,8), %rax |
| 130 |
adc %r9, %r11 |
| 131 |
mov 24(up,n,8), %rbx |
| 132 |
lea (%rdx), %r9 |
| 133 |
adc $0, %rdi |
| 134 |
mul q0 |
| 135 |
add %r11, %rbx |
| 136 |
lea (%rax), %r11 |
| 137 |
mov 32(mp,n,8), %rax |
| 138 |
adc %rdi, %rbp |
| 139 |
mov %rbx, 24(up,n,8) |
| 140 |
mov 32(up,n,8), %r10 |
| 141 |
lea (%rdx), %rdi |
| 142 |
adc $0, %r9 |
| 143 |
imul u0inv, %rbx C next q limb |
| 144 |
add $2, i |
| 145 |
jns L(ed1) |
| 146 |
|
| 147 |
ALIGNx |
| 148 |
L(tp1): mul q0 |
| 149 |
add %rbp, %r10 |
| 150 |
lea (%rax), %rbp |
| 151 |
mov (mp,i,8), %rax |
| 152 |
adc %r9, %r11 |
| 153 |
mov %r10, -8(up,i,8) |
| 154 |
mov (up,i,8), %r10 |
| 155 |
lea (%rdx), %r9 |
| 156 |
adc $0, %rdi |
| 157 |
mul q0 |
| 158 |
add %r11, %r10 |
| 159 |
lea (%rax), %r11 |
| 160 |
mov 8(mp,i,8), %rax |
| 161 |
adc %rdi, %rbp |
| 162 |
mov %r10, (up,i,8) |
| 163 |
mov 8(up,i,8), %r10 |
| 164 |
lea (%rdx), %rdi |
| 165 |
adc $0, %r9 |
| 166 |
add $2, i |
| 167 |
js L(tp1) |
| 168 |
|
| 169 |
L(ed1): mul q0 |
| 170 |
add %rbp, %r10 |
| 171 |
adc %r9, %r11 |
| 172 |
mov %r10, I(-8(up),-8(up,i,8)) |
| 173 |
mov I((up),(up,i,8)), %r10 |
| 174 |
adc $0, %rdi |
| 175 |
add %r11, %r10 |
| 176 |
adc %rdi, %rax |
| 177 |
mov %r10, I((up),(up,i,8)) |
| 178 |
mov I(8(up),8(up,i,8)), %r10 |
| 179 |
adc $0, %rdx |
| 180 |
add %rax, %r10 |
| 181 |
mov %r10, I(8(up),8(up,i,8)) |
| 182 |
adc $0, %rdx |
| 183 |
mov %rdx, 16(up,n,8) C up[0] |
| 184 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 185 |
lea 8(up), up C up++ |
| 186 |
dec j |
| 187 |
jnz L(otp1) |
| 188 |
jmp L(cj) |
| 189 |
|
| 190 |
L(b0): cmp $-2, R32(n) |
| 191 |
jz L(n2) |
| 192 |
cmp $-4, R32(n) |
| 193 |
jz L(n4) |
| 194 |
|
| 195 |
push rp |
| 196 |
|
| 197 |
L(otp0):lea 4(n), i |
| 198 |
mov (mp,n,8), %rax |
| 199 |
mul q0 |
| 200 |
lea (%rax), %r11 |
| 201 |
mov 8(mp,n,8), %rax |
| 202 |
lea (%rdx), %rdi |
| 203 |
mul q0 |
| 204 |
lea (%rax), %rbp |
| 205 |
mov 16(mp,n,8), %rax |
| 206 |
mov 16(up,n,8), %r10 |
| 207 |
lea (%rdx), %r9 |
| 208 |
mul q0 |
| 209 |
add %r11, %r10 |
| 210 |
lea (%rax), %r11 |
| 211 |
mov 24(mp,n,8), %rax |
| 212 |
adc %rdi, %rbp |
| 213 |
mov 24(up,n,8), %rbx |
| 214 |
lea (%rdx), %rdi |
| 215 |
adc $0, %r9 |
| 216 |
mul q0 |
| 217 |
add %rbp, %rbx |
| 218 |
lea (%rax), %rbp |
| 219 |
mov 32(mp,n,8), %rax |
| 220 |
adc %r9, %r11 |
| 221 |
mov %rbx, 24(up,n,8) |
| 222 |
mov 32(up,n,8), %r10 |
| 223 |
lea (%rdx), %r9 |
| 224 |
adc $0, %rdi |
| 225 |
imul u0inv, %rbx C next q limb |
| 226 |
jmp L(e0) |
| 227 |
|
| 228 |
ALIGNx |
| 229 |
L(tp0): mul q0 |
| 230 |
add %rbp, %r10 |
| 231 |
lea (%rax), %rbp |
| 232 |
mov (mp,i,8), %rax |
| 233 |
adc %r9, %r11 |
| 234 |
mov %r10, -8(up,i,8) |
| 235 |
mov (up,i,8), %r10 |
| 236 |
lea (%rdx), %r9 |
| 237 |
adc $0, %rdi |
| 238 |
L(e0): mul q0 |
| 239 |
add %r11, %r10 |
| 240 |
lea (%rax), %r11 |
| 241 |
mov 8(mp,i,8), %rax |
| 242 |
adc %rdi, %rbp |
| 243 |
mov %r10, (up,i,8) |
| 244 |
mov 8(up,i,8), %r10 |
| 245 |
lea (%rdx), %rdi |
| 246 |
adc $0, %r9 |
| 247 |
add $2, i |
| 248 |
js L(tp0) |
| 249 |
|
| 250 |
L(ed0): mul q0 |
| 251 |
add %rbp, %r10 |
| 252 |
adc %r9, %r11 |
| 253 |
mov %r10, I(-8(up),-8(up,i,8)) |
| 254 |
mov I((up),(up,i,8)), %r10 |
| 255 |
adc $0, %rdi |
| 256 |
add %r11, %r10 |
| 257 |
adc %rdi, %rax |
| 258 |
mov %r10, I((up),(up,i,8)) |
| 259 |
mov I(8(up),8(up,i,8)), %r10 |
| 260 |
adc $0, %rdx |
| 261 |
add %rax, %r10 |
| 262 |
mov %r10, I(8(up),8(up,i,8)) |
| 263 |
adc $0, %rdx |
| 264 |
mov %rdx, 16(up,n,8) C up[0] |
| 265 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 266 |
lea 8(up), up C up++ |
| 267 |
dec j |
| 268 |
jnz L(otp0) |
| 269 |
|
| 270 |
L(cj): lea 16(up), up C FIXME |
| 271 |
pop rp |
| 272 |
L(add_n): |
| 273 |
IFSTD(` lea (up,n,8), up C param 2: up |
| 274 |
lea (up,n,8), %rdx C param 3: up - n |
| 275 |
neg R32(n) ') C param 4: n |
| 276 |
|
| 277 |
IFDOS(` lea (up,n,8), %rdx C param 2: up |
| 278 |
lea (%rdx,n,8), %r8 C param 3: up - n |
| 279 |
neg R32(n) |
| 280 |
mov n, %r9 C param 4: n |
| 281 |
mov rp, %rcx ') C param 1: rp |
| 282 |
|
| 283 |
CALL( mpn_add_n) |
| 284 |
|
| 285 |
L(ret): pop %r15 |
| 286 |
pop %r14 |
| 287 |
pop %r13 |
| 288 |
pop %r12 |
| 289 |
pop %rbp |
| 290 |
pop %rbx |
| 291 |
FUNC_EXIT() |
| 292 |
ret |
| 293 |
|
| 294 |
L(n1): mov (mp_param), %rax |
| 295 |
mul q0 |
| 296 |
add 8(up), %rax |
| 297 |
adc 16(up), %rdx |
| 298 |
mov %rdx, (rp) |
| 299 |
mov $0, R32(%rax) |
| 300 |
adc R32(%rax), R32(%rax) |
| 301 |
jmp L(ret) |
| 302 |
|
| 303 |
L(n2): mov (mp_param), %rax |
| 304 |
mov (up), %rbp |
| 305 |
mul q0 |
| 306 |
add %rax, %rbp |
| 307 |
mov %rdx, %r9 |
| 308 |
adc $0, %r9 |
| 309 |
mov -8(mp), %rax |
| 310 |
mov 8(up), %r10 |
| 311 |
mul q0 |
| 312 |
add %rax, %r10 |
| 313 |
mov %rdx, %r11 |
| 314 |
adc $0, %r11 |
| 315 |
add %r9, %r10 |
| 316 |
adc $0, %r11 |
| 317 |
mov %r10, q0 |
| 318 |
imul u0inv, q0 C next q0 |
| 319 |
mov -16(mp), %rax |
| 320 |
mul q0 |
| 321 |
add %rax, %r10 |
| 322 |
mov %rdx, %r9 |
| 323 |
adc $0, %r9 |
| 324 |
mov -8(mp), %rax |
| 325 |
mov 16(up), %r14 |
| 326 |
mul q0 |
| 327 |
add %rax, %r14 |
| 328 |
adc $0, %rdx |
| 329 |
add %r9, %r14 |
| 330 |
adc $0, %rdx |
| 331 |
xor R32(%rax), R32(%rax) |
| 332 |
add %r11, %r14 |
| 333 |
adc 24(up), %rdx |
| 334 |
mov %r14, (rp) |
| 335 |
mov %rdx, 8(rp) |
| 336 |
adc R32(%rax), R32(%rax) |
| 337 |
jmp L(ret) |
| 338 |
|
| 339 |
ALIGNx |
| 340 |
L(n3): mov -24(mp), %rax |
| 341 |
mov -8(up), %r10 |
| 342 |
mul q0 |
| 343 |
add %rax, %r10 |
| 344 |
mov -16(mp), %rax |
| 345 |
mov %rdx, %r11 |
| 346 |
adc $0, %r11 |
| 347 |
mov (up), %rbp |
| 348 |
mul q0 |
| 349 |
add %rax, %rbp |
| 350 |
mov %rdx, %r9 |
| 351 |
adc $0, %r9 |
| 352 |
mov -8(mp), %rax |
| 353 |
add %r11, %rbp |
| 354 |
mov 8(up), %r10 |
| 355 |
adc $0, %r9 |
| 356 |
mul q0 |
| 357 |
mov %rbp, q0 |
| 358 |
imul u0inv, q0 C next q0 |
| 359 |
add %rax, %r10 |
| 360 |
mov %rdx, %r11 |
| 361 |
adc $0, %r11 |
| 362 |
mov %rbp, (up) |
| 363 |
add %r9, %r10 |
| 364 |
adc $0, %r11 |
| 365 |
mov %r10, 8(up) |
| 366 |
mov %r11, -8(up) C up[0] |
| 367 |
lea 8(up), up C up++ |
| 368 |
dec j |
| 369 |
jnz L(n3) |
| 370 |
|
| 371 |
mov -32(up), %rdx |
| 372 |
mov -24(up), %rbx |
| 373 |
xor R32(%rax), R32(%rax) |
| 374 |
add %rbp, %rdx |
| 375 |
adc %r10, %rbx |
| 376 |
adc 8(up), %r11 |
| 377 |
mov %rdx, (rp) |
| 378 |
mov %rbx, 8(rp) |
| 379 |
mov %r11, 16(rp) |
| 380 |
adc R32(%rax), R32(%rax) |
| 381 |
jmp L(ret) |
| 382 |
|
| 383 |
ALIGNx |
| 384 |
L(n4): mov -32(mp), %rax |
| 385 |
mul q0 |
| 386 |
lea (%rax), %r11 |
| 387 |
mov -24(mp), %rax |
| 388 |
lea (%rdx), %r14 |
| 389 |
mul q0 |
| 390 |
lea (%rax), %rbp |
| 391 |
mov -16(mp), %rax |
| 392 |
mov -16(up), %r10 |
| 393 |
lea (%rdx), %r9 |
| 394 |
mul q0 |
| 395 |
add %r11, %r10 |
| 396 |
lea (%rax), %r11 |
| 397 |
mov -8(mp), %rax |
| 398 |
adc %r14, %rbp |
| 399 |
mov -8(up), %rbx |
| 400 |
lea (%rdx), %r14 |
| 401 |
adc $0, %r9 |
| 402 |
mul q0 |
| 403 |
add %rbp, %rbx |
| 404 |
adc %r9, %r11 |
| 405 |
mov %rbx, -8(up) |
| 406 |
mov (up), %r10 |
| 407 |
adc $0, %r14 |
| 408 |
imul u0inv, %rbx C next q limb |
| 409 |
add %r11, %r10 |
| 410 |
adc %r14, %rax |
| 411 |
mov %r10, (up) |
| 412 |
mov 8(up), %r10 |
| 413 |
adc $0, %rdx |
| 414 |
add %rax, %r10 |
| 415 |
mov %r10, 8(up) |
| 416 |
adc $0, %rdx |
| 417 |
mov %rdx, -16(up) C up[0] |
| 418 |
mov %rbx, q0 C previously computed q limb -> q0 |
| 419 |
lea 8(up), up C up++ |
| 420 |
dec j |
| 421 |
jnz L(n4) |
| 422 |
lea 16(up), up |
| 423 |
jmp L(add_n) |
| 424 |
EPILOGUE() |
| 425 |
ASM_END() |