www.gitorious.org Git - gnutls:gnutls.git/blob - lib/accelerated/x86/macosx/padlock-x86-64-macosx.s

3 #

4 # Redistribution and use in source and binary forms, with or without

5 # modification, are permitted provided that the following conditions

6 # are met:

7 #

8 # * Redistributions of source code must retain copyright notices,

9 # this list of conditions and the following disclaimer.

10 #

11 # * Redistributions in binary form must reproduce the above

12 # copyright notice, this list of conditions and the following

13 # disclaimer in the documentation and/or other materials

14 # provided with the distribution.

15 #

16 # * Neither the name of the Andy Polyakov nor the names of its

17 # copyright holder and contributors may be used to endorse or

18 # promote products derived from this software without specific

19 # prior written permission.

20 #

21 # ALTERNATIVELY, provided that this notice is retained in full, this

22 # product may be distributed under the terms of the GNU General Public

23 # License (GPL), in which case the provisions of the GPL apply INSTEAD OF

24 # those given above.

25 #

26 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS

27 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT

28 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR

29 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT

30 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,

31 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT

32 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,

33 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY

34 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT

35 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE

36 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

37 #

38 # *** This file is auto-generated ***

39 #

40 .text

41 .globl _padlock_capability

43 .p2align 4

44 _padlock_capability:

45 movq %rbx,%r8

46 xorl %eax,%eax

47 cpuid

48 xorl %eax,%eax

49 cmpl $1953391939,%ebx

50 jne L$noluck

51 cmpl $1215460705,%edx

52 jne L$noluck

53 cmpl $1936487777,%ecx

54 jne L$noluck

55 movl $3221225472,%eax

56 cpuid

57 movl %eax,%edx

58 xorl %eax,%eax

59 cmpl $3221225473,%edx

60 jb L$noluck

61 movl $3221225473,%eax

62 cpuid

63 movl %edx,%eax

64 andl $4294967279,%eax

65 orl $16,%eax

66 L$noluck:

67 movq %r8,%rbx

68 .byte 0xf3,0xc3

71 .globl _padlock_key_bswap

73 .p2align 4

74 _padlock_key_bswap:

75 movl 240(%rdi),%edx

76 L$bswap_loop:

77 movl (%rdi),%eax

78 bswapl %eax

79 movl %eax,(%rdi)

80 leaq 4(%rdi),%rdi

81 subl $1,%edx

82 jnz L$bswap_loop

83 .byte 0xf3,0xc3

86 .globl _padlock_verify_context

88 .p2align 4

89 _padlock_verify_context:

90 movq %rdi,%rdx

91 pushf

92 leaq L$padlock_saved_context(%rip),%rax

93 call _padlock_verify_ctx

94 leaq 8(%rsp),%rsp

95 .byte 0xf3,0xc3

99 .p2align 4

100 _padlock_verify_ctx:

101 movq 8(%rsp),%r8

102 btq $30,%r8

103 jnc L$verified

104 cmpq (%rax),%rdx

105 je L$verified

106 pushf

107 popf

108 L$verified:

109 movq %rdx,(%rax)

110 .byte 0xf3,0xc3

111

112

113 .globl _padlock_reload_key

114

115 .p2align 4

116 _padlock_reload_key:

117 pushf

118 popf

119 .byte 0xf3,0xc3

120

121

122 .globl _padlock_aes_block

123

124 .p2align 4

125 _padlock_aes_block:

126 movq %rbx,%r8

127 movq $1,%rcx

128 leaq 32(%rdx),%rbx

129 leaq 16(%rdx),%rdx

130 .byte 0xf3,0x0f,0xa7,0xc8

131 movq %r8,%rbx

132 .byte 0xf3,0xc3

133

134

135 .globl _padlock_xstore

136

137 .p2align 4

138 _padlock_xstore:

139 movl %esi,%edx

140 .byte 0x0f,0xa7,0xc0

141 .byte 0xf3,0xc3

142

143

144 .globl _padlock_sha1_oneshot

145

146 .p2align 4

147 _padlock_sha1_oneshot:

148 movq %rdx,%rcx

149 movq %rdi,%rdx

150 movups (%rdi),%xmm0

151 subq $128+8,%rsp

152 movl 16(%rdi),%eax

153 movaps %xmm0,(%rsp)

154 movq %rsp,%rdi

155 movl %eax,16(%rsp)

156 xorq %rax,%rax

157 .byte 0xf3,0x0f,0xa6,0xc8

158 movaps (%rsp),%xmm0

159 movl 16(%rsp),%eax

160 addq $128+8,%rsp

161 movups %xmm0,(%rdx)

162 movl %eax,16(%rdx)

163 .byte 0xf3,0xc3

164

165

166 .globl _padlock_sha1_blocks

167

168 .p2align 4

169 _padlock_sha1_blocks:

170 movq %rdx,%rcx

171 movq %rdi,%rdx

172 movups (%rdi),%xmm0

173 subq $128+8,%rsp

174 movl 16(%rdi),%eax

175 movaps %xmm0,(%rsp)

176 movq %rsp,%rdi

177 movl %eax,16(%rsp)

178 movq $-1,%rax

179 .byte 0xf3,0x0f,0xa6,0xc8

180 movaps (%rsp),%xmm0

181 movl 16(%rsp),%eax

182 addq $128+8,%rsp

183 movups %xmm0,(%rdx)

184 movl %eax,16(%rdx)

185 .byte 0xf3,0xc3

186

187

188 .globl _padlock_sha256_oneshot

189

190 .p2align 4

191 _padlock_sha256_oneshot:

192 movq %rdx,%rcx

193 movq %rdi,%rdx

194 movups (%rdi),%xmm0

195 subq $128+8,%rsp

196 movups 16(%rdi),%xmm1

197 movaps %xmm0,(%rsp)

198 movq %rsp,%rdi

199 movaps %xmm1,16(%rsp)

200 xorq %rax,%rax

201 .byte 0xf3,0x0f,0xa6,0xd0

202 movaps (%rsp),%xmm0

203 movaps 16(%rsp),%xmm1

204 addq $128+8,%rsp

205 movups %xmm0,(%rdx)

206 movups %xmm1,16(%rdx)

207 .byte 0xf3,0xc3

208

209

210 .globl _padlock_sha256_blocks

211

212 .p2align 4

213 _padlock_sha256_blocks:

214 movq %rdx,%rcx

215 movq %rdi,%rdx

216 movups (%rdi),%xmm0

217 subq $128+8,%rsp

218 movups 16(%rdi),%xmm1

219 movaps %xmm0,(%rsp)

220 movq %rsp,%rdi

221 movaps %xmm1,16(%rsp)

222 movq $-1,%rax

223 .byte 0xf3,0x0f,0xa6,0xd0

224 movaps (%rsp),%xmm0

225 movaps 16(%rsp),%xmm1

226 addq $128+8,%rsp

227 movups %xmm0,(%rdx)

228 movups %xmm1,16(%rdx)

229 .byte 0xf3,0xc3

230

231

232 .globl _padlock_sha512_blocks

233

234 .p2align 4

235 _padlock_sha512_blocks:

236 movq %rdx,%rcx

237 movq %rdi,%rdx

238 movups (%rdi),%xmm0

239 subq $128+8,%rsp

240 movups 16(%rdi),%xmm1

241 movups 32(%rdi),%xmm2

242 movups 48(%rdi),%xmm3

243 movaps %xmm0,(%rsp)

244 movq %rsp,%rdi

245 movaps %xmm1,16(%rsp)

246 movaps %xmm2,32(%rsp)

247 movaps %xmm3,48(%rsp)

248 .byte 0xf3,0x0f,0xa6,0xe0

249 movaps (%rsp),%xmm0

250 movaps 16(%rsp),%xmm1

251 movaps 32(%rsp),%xmm2

252 movaps 48(%rsp),%xmm3

253 addq $128+8,%rsp

254 movups %xmm0,(%rdx)

255 movups %xmm1,16(%rdx)

256 movups %xmm2,32(%rdx)

257 movups %xmm3,48(%rdx)

258 .byte 0xf3,0xc3

259

260 .globl _padlock_ecb_encrypt

261

262 .p2align 4

263 _padlock_ecb_encrypt:

264 pushq %rbp

265 pushq %rbx

266

267 xorl %eax,%eax

268 testq $15,%rdx

269 jnz L$ecb_abort

270 testq $15,%rcx

271 jnz L$ecb_abort

272 leaq L$padlock_saved_context(%rip),%rax

273 pushf

274 cld

275 call _padlock_verify_ctx

276 leaq 16(%rdx),%rdx

277 xorl %eax,%eax

278 xorl %ebx,%ebx

279 testl $32,(%rdx)

280 jnz L$ecb_aligned

281 testq $15,%rdi

282 setz %al

283 testq $15,%rsi

284 setz %bl

285 testl %ebx,%eax

286 jnz L$ecb_aligned

287 negq %rax

288 movq $512,%rbx

289 notq %rax

290 leaq (%rsp),%rbp

291 cmpq %rbx,%rcx

292 cmovcq %rcx,%rbx

293 andq %rbx,%rax

294 movq %rcx,%rbx

295 negq %rax

296 andq $512-1,%rbx

297 leaq (%rax,%rbp,1),%rsp

298 movq $512,%rax

299 cmovzq %rax,%rbx

300 cmpq %rbx,%rcx

301 ja L$ecb_loop

302 movq %rsi,%rax

303 cmpq %rsp,%rbp

304 cmoveq %rdi,%rax

305 addq %rcx,%rax

306 negq %rax

307 andq $4095,%rax

308 cmpq $128,%rax

309 movq $-128,%rax

310 cmovaeq %rbx,%rax

311 andq %rax,%rbx

312 jz L$ecb_unaligned_tail

313 jmp L$ecb_loop

314 .p2align 4

315 L$ecb_loop:

316 cmpq %rcx,%rbx

317 cmovaq %rcx,%rbx

318 movq %rdi,%r8

319 movq %rsi,%r9

320 movq %rcx,%r10

321 movq %rbx,%rcx

322 movq %rbx,%r11

323 testq $15,%rdi

324 cmovnzq %rsp,%rdi

325 testq $15,%rsi

326 jz L$ecb_inp_aligned

327 shrq $3,%rcx

328 .byte 0xf3,0x48,0xa5

329 subq %rbx,%rdi

330 movq %rbx,%rcx

331 movq %rdi,%rsi

332 L$ecb_inp_aligned:

333 leaq -16(%rdx),%rax

334 leaq 16(%rdx),%rbx

335 shrq $4,%rcx

336 .byte 0xf3,0x0f,0xa7,200

337 movq %r8,%rdi

338 movq %r11,%rbx

339 testq $15,%rdi

340 jz L$ecb_out_aligned

341 movq %rbx,%rcx

342 leaq (%rsp),%rsi

343 shrq $3,%rcx

344 .byte 0xf3,0x48,0xa5

345 subq %rbx,%rdi

346 L$ecb_out_aligned:

347 movq %r9,%rsi

348 movq %r10,%rcx

349 addq %rbx,%rdi

350 addq %rbx,%rsi

351 subq %rbx,%rcx

352 movq $512,%rbx

353 jz L$ecb_break

354 cmpq %rbx,%rcx

355 jae L$ecb_loop

356 L$ecb_unaligned_tail:

357 xorl %eax,%eax

358 cmpq %rsp,%rbp

359 cmoveq %rcx,%rax

360 movq %rdi,%r8

361 movq %rcx,%rbx

362 subq %rax,%rsp

363 shrq $3,%rcx

364 leaq (%rsp),%rdi

365 .byte 0xf3,0x48,0xa5

366 movq %rsp,%rsi

367 movq %r8,%rdi

368 movq %rbx,%rcx

369 jmp L$ecb_loop

370 .p2align 4

371 L$ecb_break:

372 cmpq %rbp,%rsp

373 je L$ecb_done

374

375 pxor %xmm0,%xmm0

376 leaq (%rsp),%rax

377 L$ecb_bzero:

378 movaps %xmm0,(%rax)

379 leaq 16(%rax),%rax

380 cmpq %rax,%rbp

381 ja L$ecb_bzero

382

383 L$ecb_done:

384 leaq (%rbp),%rsp

385 jmp L$ecb_exit

386

387 .p2align 4

388 L$ecb_aligned:

389 leaq (%rsi,%rcx,1),%rbp

390 negq %rbp

391 andq $4095,%rbp

392 xorl %eax,%eax

393 cmpq $128,%rbp

394 movq $128-1,%rbp

395 cmovaeq %rax,%rbp

396 andq %rcx,%rbp

397 subq %rbp,%rcx

398 jz L$ecb_aligned_tail

399 leaq -16(%rdx),%rax

400 leaq 16(%rdx),%rbx

401 shrq $4,%rcx

402 .byte 0xf3,0x0f,0xa7,200

403 testq %rbp,%rbp

404 jz L$ecb_exit

405

406 L$ecb_aligned_tail:

407 movq %rdi,%r8

408 movq %rbp,%rbx

409 movq %rbp,%rcx

410 leaq (%rsp),%rbp

411 subq %rcx,%rsp

412 shrq $3,%rcx

413 leaq (%rsp),%rdi

414 .byte 0xf3,0x48,0xa5

415 leaq (%r8),%rdi

416 leaq (%rsp),%rsi

417 movq %rbx,%rcx

418 jmp L$ecb_loop

419 L$ecb_exit:

420 movl $1,%eax

421 leaq 8(%rsp),%rsp

422 L$ecb_abort:

423 popq %rbx

424 popq %rbp

425 .byte 0xf3,0xc3

426

427 .globl _padlock_cbc_encrypt

428

429 .p2align 4

430 _padlock_cbc_encrypt:

431 pushq %rbp

432 pushq %rbx

433

434 xorl %eax,%eax

435 testq $15,%rdx

436 jnz L$cbc_abort

437 testq $15,%rcx

438 jnz L$cbc_abort

439 leaq L$padlock_saved_context(%rip),%rax

440 pushf

441 cld

442 call _padlock_verify_ctx

443 leaq 16(%rdx),%rdx

444 xorl %eax,%eax

445 xorl %ebx,%ebx

446 testl $32,(%rdx)

447 jnz L$cbc_aligned

448 testq $15,%rdi

449 setz %al

450 testq $15,%rsi

451 setz %bl

452 testl %ebx,%eax

453 jnz L$cbc_aligned

454 negq %rax

455 movq $512,%rbx

456 notq %rax

457 leaq (%rsp),%rbp

458 cmpq %rbx,%rcx

459 cmovcq %rcx,%rbx

460 andq %rbx,%rax

461 movq %rcx,%rbx

462 negq %rax

463 andq $512-1,%rbx

464 leaq (%rax,%rbp,1),%rsp

465 movq $512,%rax

466 cmovzq %rax,%rbx

467 cmpq %rbx,%rcx

468 ja L$cbc_loop

469 movq %rsi,%rax

470 cmpq %rsp,%rbp

471 cmoveq %rdi,%rax

472 addq %rcx,%rax

473 negq %rax

474 andq $4095,%rax

475 cmpq $64,%rax

476 movq $-64,%rax

477 cmovaeq %rbx,%rax

478 andq %rax,%rbx

479 jz L$cbc_unaligned_tail

480 jmp L$cbc_loop

481 .p2align 4

482 L$cbc_loop:

483 cmpq %rcx,%rbx

484 cmovaq %rcx,%rbx

485 movq %rdi,%r8

486 movq %rsi,%r9

487 movq %rcx,%r10

488 movq %rbx,%rcx

489 movq %rbx,%r11

490 testq $15,%rdi

491 cmovnzq %rsp,%rdi

492 testq $15,%rsi

493 jz L$cbc_inp_aligned

494 shrq $3,%rcx

495 .byte 0xf3,0x48,0xa5

496 subq %rbx,%rdi

497 movq %rbx,%rcx

498 movq %rdi,%rsi

499 L$cbc_inp_aligned:

500 leaq -16(%rdx),%rax

501 leaq 16(%rdx),%rbx

502 shrq $4,%rcx

503 .byte 0xf3,0x0f,0xa7,208

504 movdqa (%rax),%xmm0

505 movdqa %xmm0,-16(%rdx)

506 movq %r8,%rdi

507 movq %r11,%rbx

508 testq $15,%rdi

509 jz L$cbc_out_aligned

510 movq %rbx,%rcx

511 leaq (%rsp),%rsi

512 shrq $3,%rcx

513 .byte 0xf3,0x48,0xa5

514 subq %rbx,%rdi

515 L$cbc_out_aligned:

516 movq %r9,%rsi

517 movq %r10,%rcx

518 addq %rbx,%rdi

519 addq %rbx,%rsi

520 subq %rbx,%rcx

521 movq $512,%rbx

522 jz L$cbc_break

523 cmpq %rbx,%rcx

524 jae L$cbc_loop

525 L$cbc_unaligned_tail:

526 xorl %eax,%eax

527 cmpq %rsp,%rbp

528 cmoveq %rcx,%rax

529 movq %rdi,%r8

530 movq %rcx,%rbx

531 subq %rax,%rsp

532 shrq $3,%rcx

533 leaq (%rsp),%rdi

534 .byte 0xf3,0x48,0xa5

535 movq %rsp,%rsi

536 movq %r8,%rdi

537 movq %rbx,%rcx

538 jmp L$cbc_loop

539 .p2align 4

540 L$cbc_break:

541 cmpq %rbp,%rsp

542 je L$cbc_done

543

544 pxor %xmm0,%xmm0

545 leaq (%rsp),%rax

546 L$cbc_bzero:

547 movaps %xmm0,(%rax)

548 leaq 16(%rax),%rax

549 cmpq %rax,%rbp

550 ja L$cbc_bzero

551

552 L$cbc_done:

553 leaq (%rbp),%rsp

554 jmp L$cbc_exit

555

556 .p2align 4

557 L$cbc_aligned:

558 leaq (%rsi,%rcx,1),%rbp

559 negq %rbp

560 andq $4095,%rbp

561 xorl %eax,%eax

562 cmpq $64,%rbp

563 movq $64-1,%rbp

564 cmovaeq %rax,%rbp

565 andq %rcx,%rbp

566 subq %rbp,%rcx

567 jz L$cbc_aligned_tail

568 leaq -16(%rdx),%rax

569 leaq 16(%rdx),%rbx

570 shrq $4,%rcx

571 .byte 0xf3,0x0f,0xa7,208

572 movdqa (%rax),%xmm0

573 movdqa %xmm0,-16(%rdx)

574 testq %rbp,%rbp

575 jz L$cbc_exit

576

577 L$cbc_aligned_tail:

578 movq %rdi,%r8

579 movq %rbp,%rbx

580 movq %rbp,%rcx

581 leaq (%rsp),%rbp

582 subq %rcx,%rsp

583 shrq $3,%rcx

584 leaq (%rsp),%rdi

585 .byte 0xf3,0x48,0xa5

586 leaq (%r8),%rdi

587 leaq (%rsp),%rsi

588 movq %rbx,%rcx

589 jmp L$cbc_loop

590 L$cbc_exit:

591 movl $1,%eax

592 leaq 8(%rsp),%rsp

593 L$cbc_abort:

594 popq %rbx

595 popq %rbp

596 .byte 0xf3,0xc3

597

598 .byte 86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,95,54,52,32,109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0

599 .p2align 4

600 .data

601 .p2align 3

602 L$padlock_saved_context:

603 .quad 0