www.gitorious.org Git - gnutls:gnutls.git/blob - lib/accelerated/x86/macosx/padlock-x86-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 .file "devel/perlasm/e_padlock-x86.s"

41 .text

42 .globl _padlock_capability

43 .align 4

44 _padlock_capability:

45 L_padlock_capability_begin:

46 pushl %ebx

47 pushfl

48 popl %eax

49 movl %eax,%ecx

50 xorl $2097152,%eax

51 pushl %eax

52 popfl

53 pushfl

54 popl %eax

55 xorl %eax,%ecx

56 xorl %eax,%eax

57 btl $21,%ecx

58 jnc L000noluck

59 .byte 0x0f,0xa2

60 xorl %eax,%eax

61 cmpl $0x746e6543,%ebx

62 jne L000noluck

63 cmpl $0x48727561,%edx

64 jne L000noluck

65 cmpl $0x736c7561,%ecx

66 jne L000noluck

67 movl $3221225472,%eax

68 .byte 0x0f,0xa2

69 movl %eax,%edx

70 xorl %eax,%eax

71 cmpl $3221225473,%edx

72 jb L000noluck

73 movl $1,%eax

74 .byte 0x0f,0xa2

75 orl $15,%eax

76 xorl %ebx,%ebx

77 andl $4095,%eax

78 cmpl $1791,%eax

79 sete %bl

80 movl $3221225473,%eax

81 pushl %ebx

82 .byte 0x0f,0xa2

83 popl %ebx

84 movl %edx,%eax

85 shll $4,%ebx

86 andl $4294967279,%eax

87 orl %ebx,%eax

88 L000noluck:

89 popl %ebx

90 ret

91 .globl _padlock_key_bswap

92 .align 4

93 _padlock_key_bswap:

94 L_padlock_key_bswap_begin:

95 movl 4(%esp),%edx

96 movl 240(%edx),%ecx

97 L001bswap_loop:

98 movl (%edx),%eax

99 bswap %eax

100 movl %eax,(%edx)

101 leal 4(%edx),%edx

102 subl $1,%ecx

103 jnz L001bswap_loop

104 ret

105 .globl _padlock_verify_context

106 .align 4

107 _padlock_verify_context:

108 L_padlock_verify_context_begin:

109 movl 4(%esp),%edx

110 leal Lpadlock_saved_context-L002verify_pic_point,%eax

111 pushfl

112 call __padlock_verify_ctx

113 L002verify_pic_point:

114 leal 4(%esp),%esp

115 ret

116 .align 4

117 __padlock_verify_ctx:

118 addl (%esp),%eax

119 btl $30,4(%esp)

120 jnc L003verified

121 cmpl (%eax),%edx

122 je L003verified

123 pushfl

124 popfl

125 L003verified:

126 movl %edx,(%eax)

127 ret

128 .globl _padlock_reload_key

129 .align 4

130 _padlock_reload_key:

131 L_padlock_reload_key_begin:

132 pushfl

133 popfl

134 ret

135 .globl _padlock_aes_block

136 .align 4

137 _padlock_aes_block:

138 L_padlock_aes_block_begin:

139 pushl %edi

140 pushl %esi

141 pushl %ebx

142 movl 16(%esp),%edi

143 movl 20(%esp),%esi

144 movl 24(%esp),%edx

145 movl $1,%ecx

146 leal 32(%edx),%ebx

147 leal 16(%edx),%edx

148 .byte 243,15,167,200

149 popl %ebx

150 popl %esi

151 popl %edi

152 ret

153 .globl _padlock_ecb_encrypt

154 .align 4

155 _padlock_ecb_encrypt:

156 L_padlock_ecb_encrypt_begin:

157 pushl %ebp

158 pushl %ebx

159 pushl %esi

160 pushl %edi

161 movl 20(%esp),%edi

162 movl 24(%esp),%esi

163 movl 28(%esp),%edx

164 movl 32(%esp),%ecx

165 testl $15,%edx

166 jnz L004ecb_abort

167 testl $15,%ecx

168 jnz L004ecb_abort

169 leal Lpadlock_saved_context-L005ecb_pic_point,%eax

170 pushfl

171 cld

172 call __padlock_verify_ctx

173 L005ecb_pic_point:

174 leal 16(%edx),%edx

175 xorl %eax,%eax

176 xorl %ebx,%ebx

177 testl $32,(%edx)

178 jnz L006ecb_aligned

179 testl $15,%edi

180 setz %al

181 testl $15,%esi

182 setz %bl

183 testl %ebx,%eax

184 jnz L006ecb_aligned

185 negl %eax

186 movl $512,%ebx

187 notl %eax

188 leal -24(%esp),%ebp

189 cmpl %ebx,%ecx

190 cmovcl %ecx,%ebx

191 andl %ebx,%eax

192 movl %ecx,%ebx

193 negl %eax

194 andl $511,%ebx

195 leal (%eax,%ebp,1),%esp

196 movl $512,%eax

197 cmovzl %eax,%ebx

198 movl %ebp,%eax

199 andl $-16,%ebp

200 andl $-16,%esp

201 movl %eax,16(%ebp)

202 cmpl %ebx,%ecx

203 ja L007ecb_loop

204 movl %esi,%eax

205 cmpl %esp,%ebp

206 cmovel %edi,%eax

207 addl %ecx,%eax

208 negl %eax

209 andl $4095,%eax

210 cmpl $128,%eax

211 movl $-128,%eax

212 cmovael %ebx,%eax

213 andl %eax,%ebx

214 jz L008ecb_unaligned_tail

215 jmp L007ecb_loop

216 .align 4,0x90

217 L007ecb_loop:

218 movl %edi,(%ebp)

219 movl %esi,4(%ebp)

220 movl %ecx,8(%ebp)

221 movl %ebx,%ecx

222 movl %ebx,12(%ebp)

223 testl $15,%edi

224 cmovnzl %esp,%edi

225 testl $15,%esi

226 jz L009ecb_inp_aligned

227 shrl $2,%ecx

228 .byte 243,165

229 subl %ebx,%edi

230 movl %ebx,%ecx

231 movl %edi,%esi

232 L009ecb_inp_aligned:

233 leal -16(%edx),%eax

234 leal 16(%edx),%ebx

235 shrl $4,%ecx

236 .byte 243,15,167,200

237 movl (%ebp),%edi

238 movl 12(%ebp),%ebx

239 testl $15,%edi

240 jz L010ecb_out_aligned

241 movl %ebx,%ecx

242 leal (%esp),%esi

243 shrl $2,%ecx

244 .byte 243,165

245 subl %ebx,%edi

246 L010ecb_out_aligned:

247 movl 4(%ebp),%esi

248 movl 8(%ebp),%ecx

249 addl %ebx,%edi

250 addl %ebx,%esi

251 subl %ebx,%ecx

252 movl $512,%ebx

253 jz L011ecb_break

254 cmpl %ebx,%ecx

255 jae L007ecb_loop

256 L008ecb_unaligned_tail:

257 xorl %eax,%eax

258 cmpl %ebp,%esp

259 cmovel %ecx,%eax

260 subl %eax,%esp

261 movl %edi,%eax

262 movl %ecx,%ebx

263 shrl $2,%ecx

264 leal (%esp),%edi

265 .byte 243,165

266 movl %esp,%esi

267 movl %eax,%edi

268 movl %ebx,%ecx

269 jmp L007ecb_loop

270 .align 4,0x90

271 L011ecb_break:

272 cmpl %ebp,%esp

273 je L012ecb_done

274 pxor %xmm0,%xmm0

275 leal (%esp),%eax

276 L013ecb_bzero:

277 movaps %xmm0,(%eax)

278 leal 16(%eax),%eax

279 cmpl %eax,%ebp

280 ja L013ecb_bzero

281 L012ecb_done:

282 movl 16(%ebp),%ebp

283 leal 24(%ebp),%esp

284 jmp L014ecb_exit

285 .align 4,0x90

286 L006ecb_aligned:

287 leal (%esi,%ecx,1),%ebp

288 negl %ebp

289 andl $4095,%ebp

290 xorl %eax,%eax

291 cmpl $128,%ebp

292 movl $127,%ebp

293 cmovael %eax,%ebp

294 andl %ecx,%ebp

295 subl %ebp,%ecx

296 jz L015ecb_aligned_tail

297 leal -16(%edx),%eax

298 leal 16(%edx),%ebx

299 shrl $4,%ecx

300 .byte 243,15,167,200

301 testl %ebp,%ebp

302 jz L014ecb_exit

303 L015ecb_aligned_tail:

304 movl %ebp,%ecx

305 leal -24(%esp),%ebp

306 movl %ebp,%esp

307 movl %ebp,%eax

308 subl %ecx,%esp

309 andl $-16,%ebp

310 andl $-16,%esp

311 movl %eax,16(%ebp)

312 movl %edi,%eax

313 movl %ecx,%ebx

314 shrl $2,%ecx

315 leal (%esp),%edi

316 .byte 243,165

317 movl %esp,%esi

318 movl %eax,%edi

319 movl %ebx,%ecx

320 jmp L007ecb_loop

321 L014ecb_exit:

322 movl $1,%eax

323 leal 4(%esp),%esp

324 L004ecb_abort:

325 popl %edi

326 popl %esi

327 popl %ebx

328 popl %ebp

329 ret

330 .globl _padlock_cbc_encrypt

331 .align 4

332 _padlock_cbc_encrypt:

333 L_padlock_cbc_encrypt_begin:

334 pushl %ebp

335 pushl %ebx

336 pushl %esi

337 pushl %edi

338 movl 20(%esp),%edi

339 movl 24(%esp),%esi

340 movl 28(%esp),%edx

341 movl 32(%esp),%ecx

342 testl $15,%edx

343 jnz L016cbc_abort

344 testl $15,%ecx

345 jnz L016cbc_abort

346 leal Lpadlock_saved_context-L017cbc_pic_point,%eax

347 pushfl

348 cld

349 call __padlock_verify_ctx

350 L017cbc_pic_point:

351 leal 16(%edx),%edx

352 xorl %eax,%eax

353 xorl %ebx,%ebx

354 testl $32,(%edx)

355 jnz L018cbc_aligned

356 testl $15,%edi

357 setz %al

358 testl $15,%esi

359 setz %bl

360 testl %ebx,%eax

361 jnz L018cbc_aligned

362 negl %eax

363 movl $512,%ebx

364 notl %eax

365 leal -24(%esp),%ebp

366 cmpl %ebx,%ecx

367 cmovcl %ecx,%ebx

368 andl %ebx,%eax

369 movl %ecx,%ebx

370 negl %eax

371 andl $511,%ebx

372 leal (%eax,%ebp,1),%esp

373 movl $512,%eax

374 cmovzl %eax,%ebx

375 movl %ebp,%eax

376 andl $-16,%ebp

377 andl $-16,%esp

378 movl %eax,16(%ebp)

379 cmpl %ebx,%ecx

380 ja L019cbc_loop

381 movl %esi,%eax

382 cmpl %esp,%ebp

383 cmovel %edi,%eax

384 addl %ecx,%eax

385 negl %eax

386 andl $4095,%eax

387 cmpl $64,%eax

388 movl $-64,%eax

389 cmovael %ebx,%eax

390 andl %eax,%ebx

391 jz L020cbc_unaligned_tail

392 jmp L019cbc_loop

393 .align 4,0x90

394 L019cbc_loop:

395 movl %edi,(%ebp)

396 movl %esi,4(%ebp)

397 movl %ecx,8(%ebp)

398 movl %ebx,%ecx

399 movl %ebx,12(%ebp)

400 testl $15,%edi

401 cmovnzl %esp,%edi

402 testl $15,%esi

403 jz L021cbc_inp_aligned

404 shrl $2,%ecx

405 .byte 243,165

406 subl %ebx,%edi

407 movl %ebx,%ecx

408 movl %edi,%esi

409 L021cbc_inp_aligned:

410 leal -16(%edx),%eax

411 leal 16(%edx),%ebx

412 shrl $4,%ecx

413 .byte 243,15,167,208

414 movaps (%eax),%xmm0

415 movaps %xmm0,-16(%edx)

416 movl (%ebp),%edi

417 movl 12(%ebp),%ebx

418 testl $15,%edi

419 jz L022cbc_out_aligned

420 movl %ebx,%ecx

421 leal (%esp),%esi

422 shrl $2,%ecx

423 .byte 243,165

424 subl %ebx,%edi

425 L022cbc_out_aligned:

426 movl 4(%ebp),%esi

427 movl 8(%ebp),%ecx

428 addl %ebx,%edi

429 addl %ebx,%esi

430 subl %ebx,%ecx

431 movl $512,%ebx

432 jz L023cbc_break

433 cmpl %ebx,%ecx

434 jae L019cbc_loop

435 L020cbc_unaligned_tail:

436 xorl %eax,%eax

437 cmpl %ebp,%esp

438 cmovel %ecx,%eax

439 subl %eax,%esp

440 movl %edi,%eax

441 movl %ecx,%ebx

442 shrl $2,%ecx

443 leal (%esp),%edi

444 .byte 243,165

445 movl %esp,%esi

446 movl %eax,%edi

447 movl %ebx,%ecx

448 jmp L019cbc_loop

449 .align 4,0x90

450 L023cbc_break:

451 cmpl %ebp,%esp

452 je L024cbc_done

453 pxor %xmm0,%xmm0

454 leal (%esp),%eax

455 L025cbc_bzero:

456 movaps %xmm0,(%eax)

457 leal 16(%eax),%eax

458 cmpl %eax,%ebp

459 ja L025cbc_bzero

460 L024cbc_done:

461 movl 16(%ebp),%ebp

462 leal 24(%ebp),%esp

463 jmp L026cbc_exit

464 .align 4,0x90

465 L018cbc_aligned:

466 leal (%esi,%ecx,1),%ebp

467 negl %ebp

468 andl $4095,%ebp

469 xorl %eax,%eax

470 cmpl $64,%ebp

471 movl $63,%ebp

472 cmovael %eax,%ebp

473 andl %ecx,%ebp

474 subl %ebp,%ecx

475 jz L027cbc_aligned_tail

476 leal -16(%edx),%eax

477 leal 16(%edx),%ebx

478 shrl $4,%ecx

479 .byte 243,15,167,208

480 movaps (%eax),%xmm0

481 movaps %xmm0,-16(%edx)

482 testl %ebp,%ebp

483 jz L026cbc_exit

484 L027cbc_aligned_tail:

485 movl %ebp,%ecx

486 leal -24(%esp),%ebp

487 movl %ebp,%esp

488 movl %ebp,%eax

489 subl %ecx,%esp

490 andl $-16,%ebp

491 andl $-16,%esp

492 movl %eax,16(%ebp)

493 movl %edi,%eax

494 movl %ecx,%ebx

495 shrl $2,%ecx

496 leal (%esp),%edi

497 .byte 243,165

498 movl %esp,%esi

499 movl %eax,%edi

500 movl %ebx,%ecx

501 jmp L019cbc_loop

502 L026cbc_exit:

503 movl $1,%eax

504 leal 4(%esp),%esp

505 L016cbc_abort:

506 popl %edi

507 popl %esi

508 popl %ebx

509 popl %ebp

510 ret

511 .globl _padlock_xstore

512 .align 4

513 _padlock_xstore:

514 L_padlock_xstore_begin:

515 pushl %edi

516 movl 8(%esp),%edi

517 movl 12(%esp),%edx

518 .byte 15,167,192

519 popl %edi

520 ret

521 .align 4

522 __win32_segv_handler:

523 movl $1,%eax

524 movl 4(%esp),%edx

525 movl 12(%esp),%ecx

526 cmpl $3221225477,(%edx)

527 jne L028ret

528 addl $4,184(%ecx)

529 movl $0,%eax

530 L028ret:

531 ret

532 .globl _padlock_sha1_oneshot

533 .align 4

534 _padlock_sha1_oneshot:

535 L_padlock_sha1_oneshot_begin:

536 pushl %edi

537 pushl %esi

538 xorl %eax,%eax

539 movl 12(%esp),%edi

540 movl 16(%esp),%esi

541 movl 20(%esp),%ecx

542 movl %esp,%edx

543 addl $-128,%esp

544 movups (%edi),%xmm0

545 andl $-16,%esp

546 movl 16(%edi),%eax

547 movaps %xmm0,(%esp)

548 movl %esp,%edi

549 movl %eax,16(%esp)

550 xorl %eax,%eax

551 .byte 243,15,166,200

552 movaps (%esp),%xmm0

553 movl 16(%esp),%eax

554 movl %edx,%esp

555 movl 12(%esp),%edi

556 movups %xmm0,(%edi)

557 movl %eax,16(%edi)

558 popl %esi

559 popl %edi

560 ret

561 .globl _padlock_sha1_blocks

562 .align 4

563 _padlock_sha1_blocks:

564 L_padlock_sha1_blocks_begin:

565 pushl %edi

566 pushl %esi

567 movl 12(%esp),%edi

568 movl 16(%esp),%esi

569 movl %esp,%edx

570 movl 20(%esp),%ecx

571 addl $-128,%esp

572 movups (%edi),%xmm0

573 andl $-16,%esp

574 movl 16(%edi),%eax

575 movaps %xmm0,(%esp)

576 movl %esp,%edi

577 movl %eax,16(%esp)

578 movl $-1,%eax

579 .byte 243,15,166,200

580 movaps (%esp),%xmm0

581 movl 16(%esp),%eax

582 movl %edx,%esp

583 movl 12(%esp),%edi

584 movups %xmm0,(%edi)

585 movl %eax,16(%edi)

586 popl %esi

587 popl %edi

588 ret

589 .globl _padlock_sha256_oneshot

590 .align 4

591 _padlock_sha256_oneshot:

592 L_padlock_sha256_oneshot_begin:

593 pushl %edi

594 pushl %esi

595 xorl %eax,%eax

596 movl 12(%esp),%edi

597 movl 16(%esp),%esi

598 movl 20(%esp),%ecx

599 movl %esp,%edx

600 addl $-128,%esp

601 movups (%edi),%xmm0

602 andl $-16,%esp

603 movups 16(%edi),%xmm1

604 movaps %xmm0,(%esp)

605 movl %esp,%edi

606 movaps %xmm1,16(%esp)

607 xorl %eax,%eax

608 .byte 243,15,166,208

609 movaps (%esp),%xmm0

610 movaps 16(%esp),%xmm1

611 movl %edx,%esp

612 movl 12(%esp),%edi

613 movups %xmm0,(%edi)

614 movups %xmm1,16(%edi)

615 popl %esi

616 popl %edi

617 ret

618 .globl _padlock_sha256_blocks

619 .align 4

620 _padlock_sha256_blocks:

621 L_padlock_sha256_blocks_begin:

622 pushl %edi

623 pushl %esi

624 movl 12(%esp),%edi

625 movl 16(%esp),%esi

626 movl 20(%esp),%ecx

627 movl %esp,%edx

628 addl $-128,%esp

629 movups (%edi),%xmm0

630 andl $-16,%esp

631 movups 16(%edi),%xmm1

632 movaps %xmm0,(%esp)

633 movl %esp,%edi

634 movaps %xmm1,16(%esp)

635 movl $-1,%eax

636 .byte 243,15,166,208

637 movaps (%esp),%xmm0

638 movaps 16(%esp),%xmm1

639 movl %edx,%esp

640 movl 12(%esp),%edi

641 movups %xmm0,(%edi)

642 movups %xmm1,16(%edi)

643 popl %esi

644 popl %edi

645 ret

646 .globl _padlock_sha512_blocks

647 .align 4

648 _padlock_sha512_blocks:

649 L_padlock_sha512_blocks_begin:

650 pushl %edi

651 pushl %esi

652 movl 12(%esp),%edi

653 movl 16(%esp),%esi

654 movl 20(%esp),%ecx

655 movl %esp,%edx

656 addl $-128,%esp

657 movups (%edi),%xmm0

658 andl $-16,%esp

659 movups 16(%edi),%xmm1

660 movups 32(%edi),%xmm2

661 movups 48(%edi),%xmm3

662 movaps %xmm0,(%esp)

663 movl %esp,%edi

664 movaps %xmm1,16(%esp)

665 movaps %xmm2,32(%esp)

666 movaps %xmm3,48(%esp)

667 .byte 243,15,166,224

668 movaps (%esp),%xmm0

669 movaps 16(%esp),%xmm1

670 movaps 32(%esp),%xmm2

671 movaps 48(%esp),%xmm3

672 movl %edx,%esp

673 movl 12(%esp),%edi

674 movups %xmm0,(%edi)

675 movups %xmm1,16(%edi)

676 movups %xmm2,32(%edi)

677 movups %xmm3,48(%edi)

678 popl %esi

679 popl %edi

680 ret

681 .byte 86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,32

682 .byte 109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65

683 .byte 77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101

684 .byte 110,115,115,108,46,111,114,103,62,0

685 .align 4,0x90

686 .data

687 .align 2,0x90

688 Lpadlock_saved_context:

689 .long 0