?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
if_iwm - Recognize IWM_FW_PAGING_BLOCK_CMD wide cmd response correctly.
[dragonfly.git] / crypto / openssl / crypto / bn / asm / x86_64-mont.pl
blob8fb6c994e1efb5322bdc59cd2c1d665546fe4212
1 #!/usr/bin/env perl
2
3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
9
10 # October 2005.
11 #
12 # Montgomery multiplication routine for x86_64. While it gives modest
13 # 9% improvement of rsa4096 sign on Opteron, rsa512 sign runs more
14 # than twice, >2x, as fast. Most common rsa1024 sign is improved by
15 # respectful 50%. It remains to be seen if loop unrolling and
16 # dedicated squaring routine can provide further improvement...
17
18 # July 2011.
19 #
20 # Add dedicated squaring procedure. Performance improvement varies
21 # from platform to platform, but in average it's ~5%/15%/25%/33%
22 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
23
24 # August 2011.
25 #
26 # Unroll and modulo-schedule inner loops in such manner that they
27 # are "fallen through" for input lengths of 8, which is critical for
28 # 1024-bit RSA *sign*. Average performance improvement in comparison
29 # to *initial* version of this module from 2005 is ~0%/30%/40%/45%
30 # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
31
32 # June 2013.
33 #
34 # Optimize reduction in squaring procedure and improve 1024+-bit RSA
35 # sign performance by 10-16% on Intel Sandy Bridge and later
36 # (virtually same on non-Intel processors).
37
38 # August 2013.
39 #
40 # Add MULX/ADOX/ADCX code path.
41
42 $flavour = shift;
43 $output = shift;
44 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
45
46 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
47
48 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
49 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
50 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
51 die "can't locate x86_64-xlate.pl";
52
53 open OUT,"| \"$^X\" $xlate $flavour $output";
54 *STDOUT=*OUT;
55
56 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
57 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
58 $addx = ($1>=2.23);
59 }
60
61 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
62 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
63 $addx = ($1>=2.10);
64 }
65
66 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
67 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
68 $addx = ($1>=12);
69 }
70
71 if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9])\.([0-9]+)/) {
72 my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
73 $addx = ($ver>=3.03);
74 }
75
76 # int bn_mul_mont(
77 $rp="%rdi"; # BN_ULONG *rp,
78 $ap="%rsi"; # const BN_ULONG *ap,
79 $bp="%rdx"; # const BN_ULONG *bp,
80 $np="%rcx"; # const BN_ULONG *np,
81 $n0="%r8"; # const BN_ULONG *n0,
82 $num="%r9"; # int num);
83 $lo0="%r10";
84 $hi0="%r11";
85 $hi1="%r13";
86 $i="%r14";
87 $j="%r15";
88 $m0="%rbx";
89 $m1="%rbp";
90
91 $code=<<___;
92 .text
93
94 .extern OPENSSL_ia32cap_P
95
96 .globl bn_mul_mont
97 .type bn_mul_mont,\@function,6
98 .align 16
99 bn_mul_mont:
100 test \$3,${num}d
101 jnz .Lmul_enter
102 cmp \$8,${num}d
103 jb .Lmul_enter
104 ___
105 $code.=<<___ if ($addx);
106 mov OPENSSL_ia32cap_P+8(%rip),%r11d
107 ___
108 $code.=<<___;
109 cmp $ap,$bp
110 jne .Lmul4x_enter
111 test \$7,${num}d
112 jz .Lsqr8x_enter
113 jmp .Lmul4x_enter
114
115 .align 16
116 .Lmul_enter:
117 push %rbx
118 push %rbp
119 push %r12
120 push %r13
121 push %r14
122 push %r15
123
124 mov ${num}d,${num}d
125 lea 2($num),%r10
126 mov %rsp,%r11
127 neg %r10
128 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+2))
129 and \$-1024,%rsp # minimize TLB usage
130
131 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
132 .Lmul_body:
133 # Some OSes, *cough*-dows, insist on stack being "wired" to
134 # physical memory in strictly sequential manner, i.e. if stack
135 # allocation spans two pages, then reference to farmost one can
136 # be punishable by SEGV. But page walking can do good even on
137 # other OSes, because it guarantees that villain thread hits
138 # the guard page before it can make damage to innocent one...
139 sub %rsp,%r11
140 and \$-4096,%r11
141 .Lmul_page_walk:
142 mov (%rsp,%r11),%r10
143 sub \$4096,%r11
144 .byte 0x66,0x2e # predict non-taken
145 jnc .Lmul_page_walk
146
147 mov $bp,%r12 # reassign $bp
148 ___
149 $bp="%r12";
150 $code.=<<___;
151 mov ($n0),$n0 # pull n0[0] value
152 mov ($bp),$m0 # m0=bp[0]
153 mov ($ap),%rax
154
155 xor $i,$i # i=0
156 xor $j,$j # j=0
157
158 mov $n0,$m1
159 mulq $m0 # ap[0]*bp[0]
160 mov %rax,$lo0
161 mov ($np),%rax
162
163 imulq $lo0,$m1 # "tp[0]"*n0
164 mov %rdx,$hi0
165
166 mulq $m1 # np[0]*m1
167 add %rax,$lo0 # discarded
168 mov 8($ap),%rax
169 adc \$0,%rdx
170 mov %rdx,$hi1
171
172 lea 1($j),$j # j++
173 jmp .L1st_enter
174
175 .align 16
176 .L1st:
177 add %rax,$hi1
178 mov ($ap,$j,8),%rax
179 adc \$0,%rdx
180 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
181 mov $lo0,$hi0
182 adc \$0,%rdx
183 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
184 mov %rdx,$hi1
185
186 .L1st_enter:
187 mulq $m0 # ap[j]*bp[0]
188 add %rax,$hi0
189 mov ($np,$j,8),%rax
190 adc \$0,%rdx
191 lea 1($j),$j # j++
192 mov %rdx,$lo0
193
194 mulq $m1 # np[j]*m1
195 cmp $num,$j
196 jne .L1st
197
198 add %rax,$hi1
199 mov ($ap),%rax # ap[0]
200 adc \$0,%rdx
201 add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
202 adc \$0,%rdx
203 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
204 mov %rdx,$hi1
205 mov $lo0,$hi0
206
207 xor %rdx,%rdx
208 add $hi0,$hi1
209 adc \$0,%rdx
210 mov $hi1,-8(%rsp,$num,8)
211 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
212
213 lea 1($i),$i # i++
214 jmp .Louter
215 .align 16
216 .Louter:
217 mov ($bp,$i,8),$m0 # m0=bp[i]
218 xor $j,$j # j=0
219 mov $n0,$m1
220 mov (%rsp),$lo0
221 mulq $m0 # ap[0]*bp[i]
222 add %rax,$lo0 # ap[0]*bp[i]+tp[0]
223 mov ($np),%rax
224 adc \$0,%rdx
225
226 imulq $lo0,$m1 # tp[0]*n0
227 mov %rdx,$hi0
228
229 mulq $m1 # np[0]*m1
230 add %rax,$lo0 # discarded
231 mov 8($ap),%rax
232 adc \$0,%rdx
233 mov 8(%rsp),$lo0 # tp[1]
234 mov %rdx,$hi1
235
236 lea 1($j),$j # j++
237 jmp .Linner_enter
238
239 .align 16
240 .Linner:
241 add %rax,$hi1
242 mov ($ap,$j,8),%rax
243 adc \$0,%rdx
244 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
245 mov (%rsp,$j,8),$lo0
246 adc \$0,%rdx
247 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
248 mov %rdx,$hi1
249
250 .Linner_enter:
251 mulq $m0 # ap[j]*bp[i]
252 add %rax,$hi0
253 mov ($np,$j,8),%rax
254 adc \$0,%rdx
255 add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
256 mov %rdx,$hi0
257 adc \$0,$hi0
258 lea 1($j),$j # j++
259
260 mulq $m1 # np[j]*m1
261 cmp $num,$j
262 jne .Linner
263
264 add %rax,$hi1
265 mov ($ap),%rax # ap[0]
266 adc \$0,%rdx
267 add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
268 mov (%rsp,$j,8),$lo0
269 adc \$0,%rdx
270 mov $hi1,-16(%rsp,$j,8) # tp[j-1]
271 mov %rdx,$hi1
272
273 xor %rdx,%rdx
274 add $hi0,$hi1
275 adc \$0,%rdx
276 add $lo0,$hi1 # pull upmost overflow bit
277 adc \$0,%rdx
278 mov $hi1,-8(%rsp,$num,8)
279 mov %rdx,(%rsp,$num,8) # store upmost overflow bit
280
281 lea 1($i),$i # i++
282 cmp $num,$i
283 jb .Louter
284
285 xor $i,$i # i=0 and clear CF!
286 mov (%rsp),%rax # tp[0]
287 lea (%rsp),$ap # borrow ap for tp
288 mov $num,$j # j=num
289 jmp .Lsub
290 .align 16
291 .Lsub: sbb ($np,$i,8),%rax
292 mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
293 mov 8($ap,$i,8),%rax # tp[i+1]
294 lea 1($i),$i # i++
295 dec $j # doesnn't affect CF!
296 jnz .Lsub
297
298 sbb \$0,%rax # handle upmost overflow bit
299 xor $i,$i
300 and %rax,$ap
301 not %rax
302 mov $rp,$np
303 and %rax,$np
304 mov $num,$j # j=num
305 or $np,$ap # ap=borrow?tp:rp
306 .align 16
307 .Lcopy: # copy or in-place refresh
308 mov ($ap,$i,8),%rax
309 mov $i,(%rsp,$i,8) # zap temporary vector
310 mov %rax,($rp,$i,8) # rp[i]=tp[i]
311 lea 1($i),$i
312 sub \$1,$j
313 jnz .Lcopy
314
315 mov 8(%rsp,$num,8),%rsi # restore %rsp
316 mov \$1,%rax
317 mov (%rsi),%r15
318 mov 8(%rsi),%r14
319 mov 16(%rsi),%r13
320 mov 24(%rsi),%r12
321 mov 32(%rsi),%rbp
322 mov 40(%rsi),%rbx
323 lea 48(%rsi),%rsp
324 .Lmul_epilogue:
325 ret
326 .size bn_mul_mont,.-bn_mul_mont
327 ___
328 {{{
329 my @A=("%r10","%r11");
330 my @N=("%r13","%rdi");
331 $code.=<<___;
332 .type bn_mul4x_mont,\@function,6
333 .align 16
334 bn_mul4x_mont:
335 .Lmul4x_enter:
336 ___
337 $code.=<<___ if ($addx);
338 and \$0x80100,%r11d
339 cmp \$0x80100,%r11d
340 je .Lmulx4x_enter
341 ___
342 $code.=<<___;
343 push %rbx
344 push %rbp
345 push %r12
346 push %r13
347 push %r14
348 push %r15
349
350 mov ${num}d,${num}d
351 lea 4($num),%r10
352 mov %rsp,%r11
353 neg %r10
354 lea (%rsp,%r10,8),%rsp # tp=alloca(8*(num+4))
355 and \$-1024,%rsp # minimize TLB usage
356
357 mov %r11,8(%rsp,$num,8) # tp[num+1]=%rsp
358 .Lmul4x_body:
359 sub %rsp,%r11
360 and \$-4096,%r11
361 .Lmul4x_page_walk:
362 mov (%rsp,%r11),%r10
363 sub \$4096,%r11
364 .byte 0x2e # predict non-taken
365 jnc .Lmul4x_page_walk
366
367 mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
368 mov %rdx,%r12 # reassign $bp
369 ___
370 $bp="%r12";
371 $code.=<<___;
372 mov ($n0),$n0 # pull n0[0] value
373 mov ($bp),$m0 # m0=bp[0]
374 mov ($ap),%rax
375
376 xor $i,$i # i=0
377 xor $j,$j # j=0
378
379 mov $n0,$m1
380 mulq $m0 # ap[0]*bp[0]
381 mov %rax,$A[0]
382 mov ($np),%rax
383
384 imulq $A[0],$m1 # "tp[0]"*n0
385 mov %rdx,$A[1]
386
387 mulq $m1 # np[0]*m1
388 add %rax,$A[0] # discarded
389 mov 8($ap),%rax
390 adc \$0,%rdx
391 mov %rdx,$N[1]
392
393 mulq $m0
394 add %rax,$A[1]
395 mov 8($np),%rax
396 adc \$0,%rdx
397 mov %rdx,$A[0]
398
399 mulq $m1
400 add %rax,$N[1]
401 mov 16($ap),%rax
402 adc \$0,%rdx
403 add $A[1],$N[1]
404 lea 4($j),$j # j++
405 adc \$0,%rdx
406 mov $N[1],(%rsp)
407 mov %rdx,$N[0]
408 jmp .L1st4x
409 .align 16
410 .L1st4x:
411 mulq $m0 # ap[j]*bp[0]
412 add %rax,$A[0]
413 mov -16($np,$j,8),%rax
414 adc \$0,%rdx
415 mov %rdx,$A[1]
416
417 mulq $m1 # np[j]*m1
418 add %rax,$N[0]
419 mov -8($ap,$j,8),%rax
420 adc \$0,%rdx
421 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
422 adc \$0,%rdx
423 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
424 mov %rdx,$N[1]
425
426 mulq $m0 # ap[j]*bp[0]
427 add %rax,$A[1]
428 mov -8($np,$j,8),%rax
429 adc \$0,%rdx
430 mov %rdx,$A[0]
431
432 mulq $m1 # np[j]*m1
433 add %rax,$N[1]
434 mov ($ap,$j,8),%rax
435 adc \$0,%rdx
436 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
437 adc \$0,%rdx
438 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
439 mov %rdx,$N[0]
440
441 mulq $m0 # ap[j]*bp[0]
442 add %rax,$A[0]
443 mov ($np,$j,8),%rax
444 adc \$0,%rdx
445 mov %rdx,$A[1]
446
447 mulq $m1 # np[j]*m1
448 add %rax,$N[0]
449 mov 8($ap,$j,8),%rax
450 adc \$0,%rdx
451 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
452 adc \$0,%rdx
453 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
454 mov %rdx,$N[1]
455
456 mulq $m0 # ap[j]*bp[0]
457 add %rax,$A[1]
458 mov 8($np,$j,8),%rax
459 adc \$0,%rdx
460 lea 4($j),$j # j++
461 mov %rdx,$A[0]
462
463 mulq $m1 # np[j]*m1
464 add %rax,$N[1]
465 mov -16($ap,$j,8),%rax
466 adc \$0,%rdx
467 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
468 adc \$0,%rdx
469 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
470 mov %rdx,$N[0]
471 cmp $num,$j
472 jb .L1st4x
473
474 mulq $m0 # ap[j]*bp[0]
475 add %rax,$A[0]
476 mov -16($np,$j,8),%rax
477 adc \$0,%rdx
478 mov %rdx,$A[1]
479
480 mulq $m1 # np[j]*m1
481 add %rax,$N[0]
482 mov -8($ap,$j,8),%rax
483 adc \$0,%rdx
484 add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
485 adc \$0,%rdx
486 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
487 mov %rdx,$N[1]
488
489 mulq $m0 # ap[j]*bp[0]
490 add %rax,$A[1]
491 mov -8($np,$j,8),%rax
492 adc \$0,%rdx
493 mov %rdx,$A[0]
494
495 mulq $m1 # np[j]*m1
496 add %rax,$N[1]
497 mov ($ap),%rax # ap[0]
498 adc \$0,%rdx
499 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
500 adc \$0,%rdx
501 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
502 mov %rdx,$N[0]
503
504 xor $N[1],$N[1]
505 add $A[0],$N[0]
506 adc \$0,$N[1]
507 mov $N[0],-8(%rsp,$j,8)
508 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
509
510 lea 1($i),$i # i++
511 .align 4
512 .Louter4x:
513 mov ($bp,$i,8),$m0 # m0=bp[i]
514 xor $j,$j # j=0
515 mov (%rsp),$A[0]
516 mov $n0,$m1
517 mulq $m0 # ap[0]*bp[i]
518 add %rax,$A[0] # ap[0]*bp[i]+tp[0]
519 mov ($np),%rax
520 adc \$0,%rdx
521
522 imulq $A[0],$m1 # tp[0]*n0
523 mov %rdx,$A[1]
524
525 mulq $m1 # np[0]*m1
526 add %rax,$A[0] # "$N[0]", discarded
527 mov 8($ap),%rax
528 adc \$0,%rdx
529 mov %rdx,$N[1]
530
531 mulq $m0 # ap[j]*bp[i]
532 add %rax,$A[1]
533 mov 8($np),%rax
534 adc \$0,%rdx
535 add 8(%rsp),$A[1] # +tp[1]
536 adc \$0,%rdx
537 mov %rdx,$A[0]
538
539 mulq $m1 # np[j]*m1
540 add %rax,$N[1]
541 mov 16($ap),%rax
542 adc \$0,%rdx
543 add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
544 lea 4($j),$j # j+=2
545 adc \$0,%rdx
546 mov $N[1],(%rsp) # tp[j-1]
547 mov %rdx,$N[0]
548 jmp .Linner4x
549 .align 16
550 .Linner4x:
551 mulq $m0 # ap[j]*bp[i]
552 add %rax,$A[0]
553 mov -16($np,$j,8),%rax
554 adc \$0,%rdx
555 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
556 adc \$0,%rdx
557 mov %rdx,$A[1]
558
559 mulq $m1 # np[j]*m1
560 add %rax,$N[0]
561 mov -8($ap,$j,8),%rax
562 adc \$0,%rdx
563 add $A[0],$N[0]
564 adc \$0,%rdx
565 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
566 mov %rdx,$N[1]
567
568 mulq $m0 # ap[j]*bp[i]
569 add %rax,$A[1]
570 mov -8($np,$j,8),%rax
571 adc \$0,%rdx
572 add -8(%rsp,$j,8),$A[1]
573 adc \$0,%rdx
574 mov %rdx,$A[0]
575
576 mulq $m1 # np[j]*m1
577 add %rax,$N[1]
578 mov ($ap,$j,8),%rax
579 adc \$0,%rdx
580 add $A[1],$N[1]
581 adc \$0,%rdx
582 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
583 mov %rdx,$N[0]
584
585 mulq $m0 # ap[j]*bp[i]
586 add %rax,$A[0]
587 mov ($np,$j,8),%rax
588 adc \$0,%rdx
589 add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
590 adc \$0,%rdx
591 mov %rdx,$A[1]
592
593 mulq $m1 # np[j]*m1
594 add %rax,$N[0]
595 mov 8($ap,$j,8),%rax
596 adc \$0,%rdx
597 add $A[0],$N[0]
598 adc \$0,%rdx
599 mov $N[0],-8(%rsp,$j,8) # tp[j-1]
600 mov %rdx,$N[1]
601
602 mulq $m0 # ap[j]*bp[i]
603 add %rax,$A[1]
604 mov 8($np,$j,8),%rax
605 adc \$0,%rdx
606 add 8(%rsp,$j,8),$A[1]
607 adc \$0,%rdx
608 lea 4($j),$j # j++
609 mov %rdx,$A[0]
610
611 mulq $m1 # np[j]*m1
612 add %rax,$N[1]
613 mov -16($ap,$j,8),%rax
614 adc \$0,%rdx
615 add $A[1],$N[1]
616 adc \$0,%rdx
617 mov $N[1],-32(%rsp,$j,8) # tp[j-1]
618 mov %rdx,$N[0]
619 cmp $num,$j
620 jb .Linner4x
621
622 mulq $m0 # ap[j]*bp[i]
623 add %rax,$A[0]
624 mov -16($np,$j,8),%rax
625 adc \$0,%rdx
626 add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
627 adc \$0,%rdx
628 mov %rdx,$A[1]
629
630 mulq $m1 # np[j]*m1
631 add %rax,$N[0]
632 mov -8($ap,$j,8),%rax
633 adc \$0,%rdx
634 add $A[0],$N[0]
635 adc \$0,%rdx
636 mov $N[0],-24(%rsp,$j,8) # tp[j-1]
637 mov %rdx,$N[1]
638
639 mulq $m0 # ap[j]*bp[i]
640 add %rax,$A[1]
641 mov -8($np,$j,8),%rax
642 adc \$0,%rdx
643 add -8(%rsp,$j,8),$A[1]
644 adc \$0,%rdx
645 lea 1($i),$i # i++
646 mov %rdx,$A[0]
647
648 mulq $m1 # np[j]*m1
649 add %rax,$N[1]
650 mov ($ap),%rax # ap[0]
651 adc \$0,%rdx
652 add $A[1],$N[1]
653 adc \$0,%rdx
654 mov $N[1],-16(%rsp,$j,8) # tp[j-1]
655 mov %rdx,$N[0]
656
657 xor $N[1],$N[1]
658 add $A[0],$N[0]
659 adc \$0,$N[1]
660 add (%rsp,$num,8),$N[0] # pull upmost overflow bit
661 adc \$0,$N[1]
662 mov $N[0],-8(%rsp,$j,8)
663 mov $N[1],(%rsp,$j,8) # store upmost overflow bit
664
665 cmp $num,$i
666 jb .Louter4x
667 ___
668 {
669 my @ri=("%rax","%rdx",$m0,$m1);
670 $code.=<<___;
671 mov 16(%rsp,$num,8),$rp # restore $rp
672 mov 0(%rsp),@ri[0] # tp[0]
673 pxor %xmm0,%xmm0
674 mov 8(%rsp),@ri[1] # tp[1]
675 shr \$2,$num # num/=4
676 lea (%rsp),$ap # borrow ap for tp
677 xor $i,$i # i=0 and clear CF!
678
679 sub 0($np),@ri[0]
680 mov 16($ap),@ri[2] # tp[2]
681 mov 24($ap),@ri[3] # tp[3]
682 sbb 8($np),@ri[1]
683 lea -1($num),$j # j=num/4-1
684 jmp .Lsub4x
685 .align 16
686 .Lsub4x:
687 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
688 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
689 sbb 16($np,$i,8),@ri[2]
690 mov 32($ap,$i,8),@ri[0] # tp[i+1]
691 mov 40($ap,$i,8),@ri[1]
692 sbb 24($np,$i,8),@ri[3]
693 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
694 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
695 sbb 32($np,$i,8),@ri[0]
696 mov 48($ap,$i,8),@ri[2]
697 mov 56($ap,$i,8),@ri[3]
698 sbb 40($np,$i,8),@ri[1]
699 lea 4($i),$i # i++
700 dec $j # doesnn't affect CF!
701 jnz .Lsub4x
702
703 mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
704 mov 32($ap,$i,8),@ri[0] # load overflow bit
705 sbb 16($np,$i,8),@ri[2]
706 mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
707 sbb 24($np,$i,8),@ri[3]
708 mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
709
710 sbb \$0,@ri[0] # handle upmost overflow bit
711 mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
712 xor $i,$i # i=0
713 and @ri[0],$ap
714 not @ri[0]
715 mov $rp,$np
716 and @ri[0],$np
717 lea -1($num),$j
718 or $np,$ap # ap=borrow?tp:rp
719
720 movdqu ($ap),%xmm1
721 movdqa %xmm0,(%rsp)
722 movdqu %xmm1,($rp)
723 jmp .Lcopy4x
724 .align 16
725 .Lcopy4x: # copy or in-place refresh
726 movdqu 16($ap,$i),%xmm2
727 movdqu 32($ap,$i),%xmm1
728 movdqa %xmm0,16(%rsp,$i)
729 movdqu %xmm2,16($rp,$i)
730 movdqa %xmm0,32(%rsp,$i)
731 movdqu %xmm1,32($rp,$i)
732 lea 32($i),$i
733 dec $j
734 jnz .Lcopy4x
735
736 shl \$2,$num
737 movdqu 16($ap,$i),%xmm2
738 movdqa %xmm0,16(%rsp,$i)
739 movdqu %xmm2,16($rp,$i)
740 ___
741 }
742 $code.=<<___;
743 mov 8(%rsp,$num,8),%rsi # restore %rsp
744 mov \$1,%rax
745 mov (%rsi),%r15
746 mov 8(%rsi),%r14
747 mov 16(%rsi),%r13
748 mov 24(%rsi),%r12
749 mov 32(%rsi),%rbp
750 mov 40(%rsi),%rbx
751 lea 48(%rsi),%rsp
752 .Lmul4x_epilogue:
753 ret
754 .size bn_mul4x_mont,.-bn_mul4x_mont
755 ___
756 }}}
757 \f{{{
758 ######################################################################
759 # void bn_sqr8x_mont(
760 my $rptr="%rdi"; # const BN_ULONG *rptr,
761 my $aptr="%rsi"; # const BN_ULONG *aptr,
762 my $bptr="%rdx"; # not used
763 my $nptr="%rcx"; # const BN_ULONG *nptr,
764 my $n0 ="%r8"; # const BN_ULONG *n0);
765 my $num ="%r9"; # int num, has to be divisible by 8
766
767 my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
768 my @A0=("%r10","%r11");
769 my @A1=("%r12","%r13");
770 my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
771
772 $code.=<<___ if ($addx);
773 .extern bn_sqrx8x_internal # see x86_64-mont5 module
774 ___
775 $code.=<<___;
776 .extern bn_sqr8x_internal # see x86_64-mont5 module
777
778 .type bn_sqr8x_mont,\@function,6
779 .align 32
780 bn_sqr8x_mont:
781 .Lsqr8x_enter:
782 mov %rsp,%rax
783 push %rbx
784 push %rbp
785 push %r12
786 push %r13
787 push %r14
788 push %r15
789
790 mov ${num}d,%r10d
791 shl \$3,${num}d # convert $num to bytes
792 shl \$3+2,%r10 # 4*$num
793 neg $num
794
795 ##############################################################
796 # ensure that stack frame doesn't alias with $aptr modulo
797 # 4096. this is done to allow memory disambiguation logic
798 # do its job.
799 #
800 lea -64(%rsp,$num,2),%r11
801 mov ($n0),$n0 # *n0
802 sub $aptr,%r11
803 and \$4095,%r11
804 cmp %r11,%r10
805 jb .Lsqr8x_sp_alt
806 sub %r11,%rsp # align with $aptr
807 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
808 jmp .Lsqr8x_sp_done
809
810 .align 32
811 .Lsqr8x_sp_alt:
812 lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
813 lea -64(%rsp,$num,2),%rsp # alloca(frame+2*$num)
814 sub %r10,%r11
815 mov \$0,%r10
816 cmovc %r10,%r11
817 sub %r11,%rsp
818 .Lsqr8x_sp_done:
819 and \$-64,%rsp
820 mov %rax,%r11
821 sub %rsp,%r11
822 and \$-4096,%r11
823 .Lsqr8x_page_walk:
824 mov (%rsp,%r11),%r10
825 sub \$4096,%r11
826 .byte 0x2e # predict non-taken
827 jnc .Lsqr8x_page_walk
828
829 mov $num,%r10
830 neg $num
831
832 mov $n0, 32(%rsp)
833 mov %rax, 40(%rsp) # save original %rsp
834 .Lsqr8x_body:
835
836 movq $nptr, %xmm2 # save pointer to modulus
837 pxor %xmm0,%xmm0
838 movq $rptr,%xmm1 # save $rptr
839 movq %r10, %xmm3 # -$num
840 ___
841 $code.=<<___ if ($addx);
842 mov OPENSSL_ia32cap_P+8(%rip),%eax
843 and \$0x80100,%eax
844 cmp \$0x80100,%eax
845 jne .Lsqr8x_nox
846
847 call bn_sqrx8x_internal # see x86_64-mont5 module
848 # %rax top-most carry
849 # %rbp nptr
850 # %rcx -8*num
851 # %r8 end of tp[2*num]
852 lea (%r8,%rcx),%rbx
853 mov %rcx,$num
854 mov %rcx,%rdx
855 movq %xmm1,$rptr
856 sar \$3+2,%rcx # %cf=0
857 jmp .Lsqr8x_sub
858
859 .align 32
860 .Lsqr8x_nox:
861 ___
862 $code.=<<___;
863 call bn_sqr8x_internal # see x86_64-mont5 module
864 # %rax top-most carry
865 # %rbp nptr
866 # %r8 -8*num
867 # %rdi end of tp[2*num]
868 lea (%rdi,$num),%rbx
869 mov $num,%rcx
870 mov $num,%rdx
871 movq %xmm1,$rptr
872 sar \$3+2,%rcx # %cf=0
873 jmp .Lsqr8x_sub
874
875 .align 32
876 .Lsqr8x_sub:
877 mov 8*0(%rbx),%r12
878 mov 8*1(%rbx),%r13
879 mov 8*2(%rbx),%r14
880 mov 8*3(%rbx),%r15
881 lea 8*4(%rbx),%rbx
882 sbb 8*0(%rbp),%r12
883 sbb 8*1(%rbp),%r13
884 sbb 8*2(%rbp),%r14
885 sbb 8*3(%rbp),%r15
886 lea 8*4(%rbp),%rbp
887 mov %r12,8*0($rptr)
888 mov %r13,8*1($rptr)
889 mov %r14,8*2($rptr)
890 mov %r15,8*3($rptr)
891 lea 8*4($rptr),$rptr
892 inc %rcx # preserves %cf
893 jnz .Lsqr8x_sub
894
895 sbb \$0,%rax # top-most carry
896 lea (%rbx,$num),%rbx # rewind
897 lea ($rptr,$num),$rptr # rewind
898
899 movq %rax,%xmm1
900 pxor %xmm0,%xmm0
901 pshufd \$0,%xmm1,%xmm1
902 mov 40(%rsp),%rsi # restore %rsp
903 jmp .Lsqr8x_cond_copy
904
905 .align 32
906 .Lsqr8x_cond_copy:
907 movdqa 16*0(%rbx),%xmm2
908 movdqa 16*1(%rbx),%xmm3
909 lea 16*2(%rbx),%rbx
910 movdqu 16*0($rptr),%xmm4
911 movdqu 16*1($rptr),%xmm5
912 lea 16*2($rptr),$rptr
913 movdqa %xmm0,-16*2(%rbx) # zero tp
914 movdqa %xmm0,-16*1(%rbx)
915 movdqa %xmm0,-16*2(%rbx,%rdx)
916 movdqa %xmm0,-16*1(%rbx,%rdx)
917 pcmpeqd %xmm1,%xmm0
918 pand %xmm1,%xmm2
919 pand %xmm1,%xmm3
920 pand %xmm0,%xmm4
921 pand %xmm0,%xmm5
922 pxor %xmm0,%xmm0
923 por %xmm2,%xmm4
924 por %xmm3,%xmm5
925 movdqu %xmm4,-16*2($rptr)
926 movdqu %xmm5,-16*1($rptr)
927 add \$32,$num
928 jnz .Lsqr8x_cond_copy
929
930 mov \$1,%rax
931 mov -48(%rsi),%r15
932 mov -40(%rsi),%r14
933 mov -32(%rsi),%r13
934 mov -24(%rsi),%r12
935 mov -16(%rsi),%rbp
936 mov -8(%rsi),%rbx
937 lea (%rsi),%rsp
938 .Lsqr8x_epilogue:
939 ret
940 .size bn_sqr8x_mont,.-bn_sqr8x_mont
941 ___
942 }}}
943 \f
944 if ($addx) {{{
945 my $bp="%rdx"; # original value
946
947 $code.=<<___;
948 .type bn_mulx4x_mont,\@function,6
949 .align 32
950 bn_mulx4x_mont:
951 .Lmulx4x_enter:
952 mov %rsp,%rax
953 push %rbx
954 push %rbp
955 push %r12
956 push %r13
957 push %r14
958 push %r15
959
960 shl \$3,${num}d # convert $num to bytes
961 .byte 0x67
962 xor %r10,%r10
963 sub $num,%r10 # -$num
964 mov ($n0),$n0 # *n0
965 lea -72(%rsp,%r10),%rsp # alloca(frame+$num+8)
966 and \$-128,%rsp
967 mov %rax,%r11
968 sub %rsp,%r11
969 and \$-4096,%r11
970 .Lmulx4x_page_walk:
971 mov (%rsp,%r11),%r10
972 sub \$4096,%r11
973 .byte 0x66,0x2e # predict non-taken
974 jnc .Lmulx4x_page_walk
975
976 lea ($bp,$num),%r10
977 ##############################################################
978 # Stack layout
979 # +0 num
980 # +8 off-loaded &b[i]
981 # +16 end of b[num]
982 # +24 saved n0
983 # +32 saved rp
984 # +40 saved %rsp
985 # +48 inner counter
986 # +56
987 # +64 tmp[num+1]
988 #
989 mov $num,0(%rsp) # save $num
990 shr \$5,$num
991 mov %r10,16(%rsp) # end of b[num]
992 sub \$1,$num
993 mov $n0, 24(%rsp) # save *n0
994 mov $rp, 32(%rsp) # save $rp
995 mov %rax,40(%rsp) # save original %rsp
996 mov $num,48(%rsp) # inner counter
997 jmp .Lmulx4x_body
998
999 .align 32
1000 .Lmulx4x_body:
1001 ___
1002 my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
1003 ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
1004 my $rptr=$bptr;
1005 $code.=<<___;
1006 lea 8($bp),$bptr
1007 mov ($bp),%rdx # b[0], $bp==%rdx actually
1008 lea 64+32(%rsp),$tptr
1009 mov %rdx,$bi
1010
1011 mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
1012 mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
1013 add %rax,%r11
1014 mov $bptr,8(%rsp) # off-load &b[i]
1015 mulx 2*8($aptr),%r12,%r13 # ...
1016 adc %r14,%r12
1017 adc \$0,%r13
1018
1019 mov $mi,$bptr # borrow $bptr
1020 imulq 24(%rsp),$mi # "t[0]"*n0
1021 xor $zero,$zero # cf=0, of=0
1022
1023 mulx 3*8($aptr),%rax,%r14
1024 mov $mi,%rdx
1025 lea 4*8($aptr),$aptr
1026 adcx %rax,%r13
1027 adcx $zero,%r14 # cf=0
1028
1029 mulx 0*8($nptr),%rax,%r10
1030 adcx %rax,$bptr # discarded
1031 adox %r11,%r10
1032 mulx 1*8($nptr),%rax,%r11
1033 adcx %rax,%r10
1034 adox %r12,%r11
1035 .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
1036 mov 48(%rsp),$bptr # counter value
1037 mov %r10,-4*8($tptr)
1038 adcx %rax,%r11
1039 adox %r13,%r12
1040 mulx 3*8($nptr),%rax,%r15
1041 mov $bi,%rdx
1042 mov %r11,-3*8($tptr)
1043 adcx %rax,%r12
1044 adox $zero,%r15 # of=0
1045 lea 4*8($nptr),$nptr
1046 mov %r12,-2*8($tptr)
1047
1048 jmp .Lmulx4x_1st
1049
1050 .align 32
1051 .Lmulx4x_1st:
1052 adcx $zero,%r15 # cf=0, modulo-scheduled
1053 mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
1054 adcx %r14,%r10
1055 mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
1056 adcx %rax,%r11
1057 mulx 2*8($aptr),%r12,%rax # ...
1058 adcx %r14,%r12
1059 mulx 3*8($aptr),%r13,%r14
1060 .byte 0x67,0x67
1061 mov $mi,%rdx
1062 adcx %rax,%r13
1063 adcx $zero,%r14 # cf=0
1064 lea 4*8($aptr),$aptr
1065 lea 4*8($tptr),$tptr
1066
1067 adox %r15,%r10
1068 mulx 0*8($nptr),%rax,%r15
1069 adcx %rax,%r10
1070 adox %r15,%r11
1071 mulx 1*8($nptr),%rax,%r15
1072 adcx %rax,%r11
1073 adox %r15,%r12
1074 mulx 2*8($nptr),%rax,%r15
1075 mov %r10,-5*8($tptr)
1076 adcx %rax,%r12
1077 mov %r11,-4*8($tptr)
1078 adox %r15,%r13
1079 mulx 3*8($nptr),%rax,%r15
1080 mov $bi,%rdx
1081 mov %r12,-3*8($tptr)
1082 adcx %rax,%r13
1083 adox $zero,%r15
1084 lea 4*8($nptr),$nptr
1085 mov %r13,-2*8($tptr)
1086
1087 dec $bptr # of=0, pass cf
1088 jnz .Lmulx4x_1st
1089
1090 mov 0(%rsp),$num # load num
1091 mov 8(%rsp),$bptr # re-load &b[i]
1092 adc $zero,%r15 # modulo-scheduled
1093 add %r15,%r14
1094 sbb %r15,%r15 # top-most carry
1095 mov %r14,-1*8($tptr)
1096 jmp .Lmulx4x_outer
1097
1098 .align 32
1099 .Lmulx4x_outer:
1100 mov ($bptr),%rdx # b[i]
1101 lea 8($bptr),$bptr # b++
1102 sub $num,$aptr # rewind $aptr
1103 mov %r15,($tptr) # save top-most carry
1104 lea 64+4*8(%rsp),$tptr
1105 sub $num,$nptr # rewind $nptr
1106
1107 mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
1108 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1109 mov %rdx,$bi
1110 mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
1111 adox -4*8($tptr),$mi
1112 adcx %r14,%r11
1113 mulx 2*8($aptr),%r15,%r13 # ...
1114 adox -3*8($tptr),%r11
1115 adcx %r15,%r12
1116 adox $zero,%r12
1117 adcx $zero,%r13
1118
1119 mov $bptr,8(%rsp) # off-load &b[i]
1120 .byte 0x67
1121 mov $mi,%r15
1122 imulq 24(%rsp),$mi # "t[0]"*n0
1123 xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
1124
1125 mulx 3*8($aptr),%rax,%r14
1126 mov $mi,%rdx
1127 adox -2*8($tptr),%r12
1128 adcx %rax,%r13
1129 adox -1*8($tptr),%r13
1130 adcx $zero,%r14
1131 lea 4*8($aptr),$aptr
1132 adox $zero,%r14
1133
1134 mulx 0*8($nptr),%rax,%r10
1135 adcx %rax,%r15 # discarded
1136 adox %r11,%r10
1137 mulx 1*8($nptr),%rax,%r11
1138 adcx %rax,%r10
1139 adox %r12,%r11
1140 mulx 2*8($nptr),%rax,%r12
1141 mov %r10,-4*8($tptr)
1142 adcx %rax,%r11
1143 adox %r13,%r12
1144 mulx 3*8($nptr),%rax,%r15
1145 mov $bi,%rdx
1146 mov %r11,-3*8($tptr)
1147 lea 4*8($nptr),$nptr
1148 adcx %rax,%r12
1149 adox $zero,%r15 # of=0
1150 mov 48(%rsp),$bptr # counter value
1151 mov %r12,-2*8($tptr)
1152
1153 jmp .Lmulx4x_inner
1154
1155 .align 32
1156 .Lmulx4x_inner:
1157 mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
1158 adcx $zero,%r15 # cf=0, modulo-scheduled
1159 adox %r14,%r10
1160 mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
1161 adcx 0*8($tptr),%r10
1162 adox %rax,%r11
1163 mulx 2*8($aptr),%r12,%rax # ...
1164 adcx 1*8($tptr),%r11
1165 adox %r14,%r12
1166 mulx 3*8($aptr),%r13,%r14
1167 mov $mi,%rdx
1168 adcx 2*8($tptr),%r12
1169 adox %rax,%r13
1170 adcx 3*8($tptr),%r13
1171 adox $zero,%r14 # of=0
1172 lea 4*8($aptr),$aptr
1173 lea 4*8($tptr),$tptr
1174 adcx $zero,%r14 # cf=0
1175
1176 adox %r15,%r10
1177 mulx 0*8($nptr),%rax,%r15
1178 adcx %rax,%r10
1179 adox %r15,%r11
1180 mulx 1*8($nptr),%rax,%r15
1181 adcx %rax,%r11
1182 adox %r15,%r12
1183 mulx 2*8($nptr),%rax,%r15
1184 mov %r10,-5*8($tptr)
1185 adcx %rax,%r12
1186 adox %r15,%r13
1187 mulx 3*8($nptr),%rax,%r15
1188 mov $bi,%rdx
1189 mov %r11,-4*8($tptr)
1190 mov %r12,-3*8($tptr)
1191 adcx %rax,%r13
1192 adox $zero,%r15
1193 lea 4*8($nptr),$nptr
1194 mov %r13,-2*8($tptr)
1195
1196 dec $bptr # of=0, pass cf
1197 jnz .Lmulx4x_inner
1198
1199 mov 0(%rsp),$num # load num
1200 mov 8(%rsp),$bptr # re-load &b[i]
1201 adc $zero,%r15 # modulo-scheduled
1202 sub 0*8($tptr),$zero # pull top-most carry
1203 adc %r15,%r14
1204 sbb %r15,%r15 # top-most carry
1205 mov %r14,-1*8($tptr)
1206
1207 cmp 16(%rsp),$bptr
1208 jne .Lmulx4x_outer
1209
1210 lea 64(%rsp),$tptr
1211 sub $num,$nptr # rewind $nptr
1212 neg %r15
1213 mov $num,%rdx
1214 shr \$3+2,$num # %cf=0
1215 mov 32(%rsp),$rptr # restore rp
1216 jmp .Lmulx4x_sub
1217
1218 .align 32
1219 .Lmulx4x_sub:
1220 mov 8*0($tptr),%r11
1221 mov 8*1($tptr),%r12
1222 mov 8*2($tptr),%r13
1223 mov 8*3($tptr),%r14
1224 lea 8*4($tptr),$tptr
1225 sbb 8*0($nptr),%r11
1226 sbb 8*1($nptr),%r12
1227 sbb 8*2($nptr),%r13
1228 sbb 8*3($nptr),%r14
1229 lea 8*4($nptr),$nptr
1230 mov %r11,8*0($rptr)
1231 mov %r12,8*1($rptr)
1232 mov %r13,8*2($rptr)
1233 mov %r14,8*3($rptr)
1234 lea 8*4($rptr),$rptr
1235 dec $num # preserves %cf
1236 jnz .Lmulx4x_sub
1237
1238 sbb \$0,%r15 # top-most carry
1239 lea 64(%rsp),$tptr
1240 sub %rdx,$rptr # rewind
1241
1242 movq %r15,%xmm1
1243 pxor %xmm0,%xmm0
1244 pshufd \$0,%xmm1,%xmm1
1245 mov 40(%rsp),%rsi # restore %rsp
1246 jmp .Lmulx4x_cond_copy
1247
1248 .align 32
1249 .Lmulx4x_cond_copy:
1250 movdqa 16*0($tptr),%xmm2
1251 movdqa 16*1($tptr),%xmm3
1252 lea 16*2($tptr),$tptr
1253 movdqu 16*0($rptr),%xmm4
1254 movdqu 16*1($rptr),%xmm5
1255 lea 16*2($rptr),$rptr
1256 movdqa %xmm0,-16*2($tptr) # zero tp
1257 movdqa %xmm0,-16*1($tptr)
1258 pcmpeqd %xmm1,%xmm0
1259 pand %xmm1,%xmm2
1260 pand %xmm1,%xmm3
1261 pand %xmm0,%xmm4
1262 pand %xmm0,%xmm5
1263 pxor %xmm0,%xmm0
1264 por %xmm2,%xmm4
1265 por %xmm3,%xmm5
1266 movdqu %xmm4,-16*2($rptr)
1267 movdqu %xmm5,-16*1($rptr)
1268 sub \$32,%rdx
1269 jnz .Lmulx4x_cond_copy
1270
1271 mov %rdx,($tptr)
1272
1273 mov \$1,%rax
1274 mov -48(%rsi),%r15
1275 mov -40(%rsi),%r14
1276 mov -32(%rsi),%r13
1277 mov -24(%rsi),%r12
1278 mov -16(%rsi),%rbp
1279 mov -8(%rsi),%rbx
1280 lea (%rsi),%rsp
1281 .Lmulx4x_epilogue:
1282 ret
1283 .size bn_mulx4x_mont,.-bn_mulx4x_mont
1284 ___
1285 }}}
1286 $code.=<<___;
1287 .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1288 .align 16
1289 ___
1290
1291 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1292 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1293 if ($win64) {
1294 $rec="%rcx";
1295 $frame="%rdx";
1296 $context="%r8";
1297 $disp="%r9";
1298
1299 $code.=<<___;
1300 .extern __imp_RtlVirtualUnwind
1301 .type mul_handler,\@abi-omnipotent
1302 .align 16
1303 mul_handler:
1304 push %rsi
1305 push %rdi
1306 push %rbx
1307 push %rbp
1308 push %r12
1309 push %r13
1310 push %r14
1311 push %r15
1312 pushfq
1313 sub \$64,%rsp
1314
1315 mov 120($context),%rax # pull context->Rax
1316 mov 248($context),%rbx # pull context->Rip
1317
1318 mov 8($disp),%rsi # disp->ImageBase
1319 mov 56($disp),%r11 # disp->HandlerData
1320
1321 mov 0(%r11),%r10d # HandlerData[0]
1322 lea (%rsi,%r10),%r10 # end of prologue label
1323 cmp %r10,%rbx # context->Rip<end of prologue label
1324 jb .Lcommon_seh_tail
1325
1326 mov 152($context),%rax # pull context->Rsp
1327
1328 mov 4(%r11),%r10d # HandlerData[1]
1329 lea (%rsi,%r10),%r10 # epilogue label
1330 cmp %r10,%rbx # context->Rip>=epilogue label
1331 jae .Lcommon_seh_tail
1332
1333 mov 192($context),%r10 # pull $num
1334 mov 8(%rax,%r10,8),%rax # pull saved stack pointer
1335 lea 48(%rax),%rax
1336
1337 mov -8(%rax),%rbx
1338 mov -16(%rax),%rbp
1339 mov -24(%rax),%r12
1340 mov -32(%rax),%r13
1341 mov -40(%rax),%r14
1342 mov -48(%rax),%r15
1343 mov %rbx,144($context) # restore context->Rbx
1344 mov %rbp,160($context) # restore context->Rbp
1345 mov %r12,216($context) # restore context->R12
1346 mov %r13,224($context) # restore context->R13
1347 mov %r14,232($context) # restore context->R14
1348 mov %r15,240($context) # restore context->R15
1349
1350 jmp .Lcommon_seh_tail
1351 .size mul_handler,.-mul_handler
1352
1353 .type sqr_handler,\@abi-omnipotent
1354 .align 16
1355 sqr_handler:
1356 push %rsi
1357 push %rdi
1358 push %rbx
1359 push %rbp
1360 push %r12
1361 push %r13
1362 push %r14
1363 push %r15
1364 pushfq
1365 sub \$64,%rsp
1366
1367 mov 120($context),%rax # pull context->Rax
1368 mov 248($context),%rbx # pull context->Rip
1369
1370 mov 8($disp),%rsi # disp->ImageBase
1371 mov 56($disp),%r11 # disp->HandlerData
1372
1373 mov 0(%r11),%r10d # HandlerData[0]
1374 lea (%rsi,%r10),%r10 # end of prologue label
1375 cmp %r10,%rbx # context->Rip<.Lsqr_body
1376 jb .Lcommon_seh_tail
1377
1378 mov 152($context),%rax # pull context->Rsp
1379
1380 mov 4(%r11),%r10d # HandlerData[1]
1381 lea (%rsi,%r10),%r10 # epilogue label
1382 cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
1383 jae .Lcommon_seh_tail
1384
1385 mov 40(%rax),%rax # pull saved stack pointer
1386
1387 mov -8(%rax),%rbx
1388 mov -16(%rax),%rbp
1389 mov -24(%rax),%r12
1390 mov -32(%rax),%r13
1391 mov -40(%rax),%r14
1392 mov -48(%rax),%r15
1393 mov %rbx,144($context) # restore context->Rbx
1394 mov %rbp,160($context) # restore context->Rbp
1395 mov %r12,216($context) # restore context->R12
1396 mov %r13,224($context) # restore context->R13
1397 mov %r14,232($context) # restore context->R14
1398 mov %r15,240($context) # restore context->R15
1399
1400 .Lcommon_seh_tail:
1401 mov 8(%rax),%rdi
1402 mov 16(%rax),%rsi
1403 mov %rax,152($context) # restore context->Rsp
1404 mov %rsi,168($context) # restore context->Rsi
1405 mov %rdi,176($context) # restore context->Rdi
1406
1407 mov 40($disp),%rdi # disp->ContextRecord
1408 mov $context,%rsi # context
1409 mov \$154,%ecx # sizeof(CONTEXT)
1410 .long 0xa548f3fc # cld; rep movsq
1411
1412 mov $disp,%rsi
1413 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1414 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1415 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1416 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1417 mov 40(%rsi),%r10 # disp->ContextRecord
1418 lea 56(%rsi),%r11 # &disp->HandlerData
1419 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1420 mov %r10,32(%rsp) # arg5
1421 mov %r11,40(%rsp) # arg6
1422 mov %r12,48(%rsp) # arg7
1423 mov %rcx,56(%rsp) # arg8, (NULL)
1424 call *__imp_RtlVirtualUnwind(%rip)
1425
1426 mov \$1,%eax # ExceptionContinueSearch
1427 add \$64,%rsp
1428 popfq
1429 pop %r15
1430 pop %r14
1431 pop %r13
1432 pop %r12
1433 pop %rbp
1434 pop %rbx
1435 pop %rdi
1436 pop %rsi
1437 ret
1438 .size sqr_handler,.-sqr_handler
1439
1440 .section .pdata
1441 .align 4
1442 .rva .LSEH_begin_bn_mul_mont
1443 .rva .LSEH_end_bn_mul_mont
1444 .rva .LSEH_info_bn_mul_mont
1445
1446 .rva .LSEH_begin_bn_mul4x_mont
1447 .rva .LSEH_end_bn_mul4x_mont
1448 .rva .LSEH_info_bn_mul4x_mont
1449
1450 .rva .LSEH_begin_bn_sqr8x_mont
1451 .rva .LSEH_end_bn_sqr8x_mont
1452 .rva .LSEH_info_bn_sqr8x_mont
1453 ___
1454 $code.=<<___ if ($addx);
1455 .rva .LSEH_begin_bn_mulx4x_mont
1456 .rva .LSEH_end_bn_mulx4x_mont
1457 .rva .LSEH_info_bn_mulx4x_mont
1458 ___
1459 $code.=<<___;
1460 .section .xdata
1461 .align 8
1462 .LSEH_info_bn_mul_mont:
1463 .byte 9,0,0,0
1464 .rva mul_handler
1465 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
1466 .LSEH_info_bn_mul4x_mont:
1467 .byte 9,0,0,0
1468 .rva mul_handler
1469 .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
1470 .LSEH_info_bn_sqr8x_mont:
1471 .byte 9,0,0,0
1472 .rva sqr_handler
1473 .rva .Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
1474 ___
1475 $code.=<<___ if ($addx);
1476 .LSEH_info_bn_mulx4x_mont:
1477 .byte 9,0,0,0
1478 .rva sqr_handler
1479 .rva .Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
1480 ___
1481 }
1482
1483 print $code;
1484 close STDOUT;
DragonFly BSD