3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> 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 # ====================================================================
10 # This module implements support for Intel AES-NI extension. In
11 # OpenSSL context it's used with Intel engine, but can also be used as
12 # drop-in replacement for crypto/aes/asm/aes-x86_64.pl [see below for
17 # Given aes(enc|dec) instructions' latency asymptotic performance for
18 # non-parallelizable modes such as CBC encrypt is 3.75 cycles per byte
19 # processed with 128-bit key. And given their throughput asymptotic
20 # performance for parallelizable modes is 1.25 cycles per byte. Being
21 # asymptotic limit it's not something you commonly achieve in reality,
22 # but how close does one get? Below are results collected for
23 # different modes and block sized. Pairs of numbers are for en-/
26 # 16-byte 64-byte 256-byte 1-KB 8-KB
27 # ECB 4.25/4.25 1.38/1.38 1.28/1.28 1.26/1.26 1.26/1.26
28 # CTR 5.42/5.42 1.92/1.92 1.44/1.44 1.28/1.28 1.26/1.26
29 # CBC 4.38/4.43 4.15/1.43 4.07/1.32 4.07/1.29 4.06/1.28
30 # CCM 5.66/9.42 4.42/5.41 4.16/4.40 4.09/4.15 4.06/4.07
31 # OFB 5.42/5.42 4.64/4.64 4.44/4.44 4.39/4.39 4.38/4.38
32 # CFB 5.73/5.85 5.56/5.62 5.48/5.56 5.47/5.55 5.47/5.55
34 # ECB, CTR, CBC and CCM results are free from EVP overhead. This means
35 # that otherwise used 'openssl speed -evp aes-128-??? -engine aesni
36 # [-decrypt]' will exhibit 10-15% worse results for smaller blocks.
37 # The results were collected with specially crafted speed.c benchmark
38 # in order to compare them with results reported in "Intel Advanced
39 # Encryption Standard (AES) New Instruction Set" White Paper Revision
40 # 3.0 dated May 2010. All above results are consistently better. This
41 # module also provides better performance for block sizes smaller than
42 # 128 bytes in points *not* represented in the above table.
44 # Looking at the results for 8-KB buffer.
46 # CFB and OFB results are far from the limit, because implementation
47 # uses "generic" CRYPTO_[c|o]fb128_encrypt interfaces relying on
48 # single-block aesni_encrypt, which is not the most optimal way to go.
49 # CBC encrypt result is unexpectedly high and there is no documented
50 # explanation for it. Seemingly there is a small penalty for feeding
51 # the result back to AES unit the way it's done in CBC mode. There is
52 # nothing one can do and the result appears optimal. CCM result is
53 # identical to CBC, because CBC-MAC is essentially CBC encrypt without
54 # saving output. CCM CTR "stays invisible," because it's neatly
55 # interleaved wih CBC-MAC. This provides ~30% improvement over
56 # "straghtforward" CCM implementation with CTR and CBC-MAC performed
57 # disjointly. Parallelizable modes practically achieve the theoretical
60 # Looking at how results vary with buffer size.
62 # Curves are practically saturated at 1-KB buffer size. In most cases
63 # "256-byte" performance is >95%, and "64-byte" is ~90% of "8-KB" one.
64 # CTR curve doesn't follow this pattern and is "slowest" changing one
65 # with "256-byte" result being 87% of "8-KB." This is because overhead
66 # in CTR mode is most computationally intensive. Small-block CCM
67 # decrypt is slower than encrypt, because first CTR and last CBC-MAC
68 # iterations can't be interleaved.
70 # Results for 192- and 256-bit keys.
72 # EVP-free results were observed to scale perfectly with number of
73 # rounds for larger block sizes, i.e. 192-bit result being 10/12 times
74 # lower and 256-bit one - 10/14. Well, in CBC encrypt case differences
75 # are a tad smaller, because the above mentioned penalty biases all
76 # results by same constant value. In similar way function call
77 # overhead affects small-block performance, as well as OFB and CFB
78 # results. Differences are not large, most common coefficients are
79 # 10/11.7 and 10/13.4 (as opposite to 10/12.0 and 10/14.0), but one
80 # observe even 10/11.2 and 10/12.4 (CTR, OFB, CFB)...
84 # While Westmere processor features 6 cycles latency for aes[enc|dec]
85 # instructions, which can be scheduled every second cycle, Sandy
86 # Bridge spends 8 cycles per instruction, but it can schedule them
87 # every cycle. This means that code targeting Westmere would perform
88 # suboptimally on Sandy Bridge. Therefore this update.
90 # In addition, non-parallelizable CBC encrypt (as well as CCM) is
91 # optimized. Relative improvement might appear modest, 8% on Westmere,
92 # but in absolute terms it's 3.77 cycles per byte encrypted with
93 # 128-bit key on Westmere, and 5.07 - on Sandy Bridge. These numbers
94 # should be compared to asymptotic limits of 3.75 for Westmere and
95 # 5.00 for Sandy Bridge. Actually, the fact that they get this close
96 # to asymptotic limits is quite amazing. Indeed, the limit is
97 # calculated as latency times number of rounds, 10 for 128-bit key,
98 # and divided by 16, the number of bytes in block, or in other words
99 # it accounts *solely* for aesenc instructions. But there are extra
100 # instructions, and numbers so close to the asymptotic limits mean
101 # that it's as if it takes as little as *one* additional cycle to
102 # execute all of them. How is it possible? It is possible thanks to
103 # out-of-order execution logic, which manages to overlap post-
104 # processing of previous block, things like saving the output, with
105 # actual encryption of current block, as well as pre-processing of
106 # current block, things like fetching input and xor-ing it with
107 # 0-round element of the key schedule, with actual encryption of
108 # previous block. Keep this in mind...
110 # For parallelizable modes, such as ECB, CBC decrypt, CTR, higher
111 # performance is achieved by interleaving instructions working on
112 # independent blocks. In which case asymptotic limit for such modes
113 # can be obtained by dividing above mentioned numbers by AES
114 # instructions' interleave factor. Westmere can execute at most 3
115 # instructions at a time, meaning that optimal interleave factor is 3,
116 # and that's where the "magic" number of 1.25 come from. "Optimal
117 # interleave factor" means that increase of interleave factor does
118 # not improve performance. The formula has proven to reflect reality
119 # pretty well on Westmere... Sandy Bridge on the other hand can
120 # execute up to 8 AES instructions at a time, so how does varying
121 # interleave factor affect the performance? Here is table for ECB
122 # (numbers are cycles per byte processed with 128-bit key):
124 # instruction interleave factor 3x 6x 8x
125 # theoretical asymptotic limit 1.67 0.83 0.625
126 # measured performance for 8KB block 1.05 0.86 0.84
128 # "as if" interleave factor 4.7x 5.8x 6.0x
130 # Further data for other parallelizable modes:
132 # CBC decrypt 1.16 0.93 0.93
135 # Well, given 3x column it's probably inappropriate to call the limit
136 # asymptotic, if it can be surpassed, isn't it? What happens there?
137 # Rewind to CBC paragraph for the answer. Yes, out-of-order execution
138 # magic is responsible for this. Processor overlaps not only the
139 # additional instructions with AES ones, but even AES instuctions
140 # processing adjacent triplets of independent blocks. In the 6x case
141 # additional instructions still claim disproportionally small amount
142 # of additional cycles, but in 8x case number of instructions must be
143 # a tad too high for out-of-order logic to cope with, and AES unit
144 # remains underutilized... As you can see 8x interleave is hardly
145 # justifiable, so there no need to feel bad that 32-bit aesni-x86.pl
146 # utilizies 6x interleave because of limited register bank capacity.
148 # Higher interleave factors do have negative impact on Westmere
149 # performance. While for ECB mode it's negligible ~1.5%, other
150 # parallelizables perform ~5% worse, which is outweighed by ~25%
151 # improvement on Sandy Bridge. To balance regression on Westmere
152 # CTR mode was implemented with 6x aesenc interleave factor.
156 # Add aesni_xts_[en|de]crypt. Westmere spends 1.33 cycles processing
157 # one byte out of 8KB with 128-bit key, Sandy Bridge - 0.97. Just like
158 # in CTR mode AES instruction interleave factor was chosen to be 6x.
160 ######################################################################
161 # For reference, AMD Bulldozer spends 5.77 cycles per byte processed
162 # with 128-bit key in CBC encrypt and 0.76 cycles in CBC decrypt, 0.70
163 # in ECB, 0.71 in CTR, 0.95 in XTS... This means that aes[enc|dec]
164 # instruction latency is 9 cycles and that they can be issued every
167 $PREFIX="aesni"; # if $PREFIX is set to "AES", the script
168 # generates drop-in replacement for
169 # crypto/aes/asm/aes-x86_64.pl:-)
173 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
175 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
177 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
178 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
179 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
180 die "can't locate x86_64-xlate.pl";
182 open OUT,"| \"$^X\" $xlate $flavour $output";
185 $movkey = $PREFIX eq "aesni" ? "movups" : "movups";
186 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") : # Win64 order
187 ("%rdi","%rsi","%rdx","%rcx"); # Unix order
191 $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!!
192 # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ...
196 $key="%rcx"; # input to and changed by aesni_[en|de]cryptN !!!
197 $ivp="%r8"; # cbc, ctr, ...
199 $rnds_="%r10d"; # backup copy for $rounds
200 $key_="%r11"; # backup copy for $key
202 # %xmm register layout
203 $rndkey0="%xmm0"; $rndkey1="%xmm1";
204 $inout0="%xmm2"; $inout1="%xmm3";
205 $inout2="%xmm4"; $inout3="%xmm5";
206 $inout4="%xmm6"; $inout5="%xmm7";
207 $inout6="%xmm8"; $inout7="%xmm9";
209 $in2="%xmm6"; $in1="%xmm7"; # used in CBC decrypt, CTR, ...
210 $in0="%xmm8"; $iv="%xmm9";
212 # Inline version of internal aesni_[en|de]crypt1.
214 # Why folded loop? Because aes[enc|dec] is slow enough to accommodate
215 # cycles which take care of loop variables...
217 sub aesni_generate1 {
218 my ($p,$key,$rounds,$inout,$ivec)=@_; $inout=$inout0 if (!defined($inout));
221 $movkey ($key),$rndkey0
222 $movkey 16($key),$rndkey1
224 $code.=<<___ if (defined($ivec));
229 $code.=<<___ if (!defined($ivec));
231 xorps $rndkey0,$inout
235 aes${p} $rndkey1,$inout
237 $movkey ($key),$rndkey1
239 jnz .Loop_${p}1_$sn # loop body is 16 bytes
240 aes${p}last $rndkey1,$inout
243 # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key);
245 { my ($inp,$out,$key) = @_4args;
248 .globl ${PREFIX}_encrypt
249 .type ${PREFIX}_encrypt,\@abi-omnipotent
252 movups ($inp),$inout0 # load input
253 mov 240($key),$rounds # key->rounds
255 &aesni_generate1("enc",$key,$rounds);
257 movups $inout0,($out) # output
259 .size ${PREFIX}_encrypt,.-${PREFIX}_encrypt
261 .globl ${PREFIX}_decrypt
262 .type ${PREFIX}_decrypt,\@abi-omnipotent
265 movups ($inp),$inout0 # load input
266 mov 240($key),$rounds # key->rounds
268 &aesni_generate1("dec",$key,$rounds);
270 movups $inout0,($out) # output
272 .size ${PREFIX}_decrypt, .-${PREFIX}_decrypt
276 # _aesni_[en|de]cryptN are private interfaces, N denotes interleave
277 # factor. Why 3x subroutine were originally used in loops? Even though
278 # aes[enc|dec] latency was originally 6, it could be scheduled only
279 # every *2nd* cycle. Thus 3x interleave was the one providing optimal
280 # utilization, i.e. when subroutine's throughput is virtually same as
281 # of non-interleaved subroutine [for number of input blocks up to 3].
282 # This is why it makes no sense to implement 2x subroutine.
283 # aes[enc|dec] latency in next processor generation is 8, but the
284 # instructions can be scheduled every cycle. Optimal interleave for
285 # new processor is therefore 8x...
286 sub aesni_generate3 {
288 # As already mentioned it takes in $key and $rounds, which are *not*
289 # preserved. $inout[0-2] is cipher/clear text...
291 .type _aesni_${dir}rypt3,\@abi-omnipotent
294 $movkey ($key),$rndkey0
296 $movkey 16($key),$rndkey1
298 xorps $rndkey0,$inout0
299 xorps $rndkey0,$inout1
300 xorps $rndkey0,$inout2
301 $movkey ($key),$rndkey0
304 aes${dir} $rndkey1,$inout0
305 aes${dir} $rndkey1,$inout1
307 aes${dir} $rndkey1,$inout2
308 $movkey 16($key),$rndkey1
309 aes${dir} $rndkey0,$inout0
310 aes${dir} $rndkey0,$inout1
312 aes${dir} $rndkey0,$inout2
313 $movkey ($key),$rndkey0
316 aes${dir} $rndkey1,$inout0
317 aes${dir} $rndkey1,$inout1
318 aes${dir} $rndkey1,$inout2
319 aes${dir}last $rndkey0,$inout0
320 aes${dir}last $rndkey0,$inout1
321 aes${dir}last $rndkey0,$inout2
323 .size _aesni_${dir}rypt3,.-_aesni_${dir}rypt3
326 # 4x interleave is implemented to improve small block performance,
327 # most notably [and naturally] 4 block by ~30%. One can argue that one
328 # should have implemented 5x as well, but improvement would be <20%,
329 # so it's not worth it...
330 sub aesni_generate4 {
332 # As already mentioned it takes in $key and $rounds, which are *not*
333 # preserved. $inout[0-3] is cipher/clear text...
335 .type _aesni_${dir}rypt4,\@abi-omnipotent
338 $movkey ($key),$rndkey0
340 $movkey 16($key),$rndkey1
342 xorps $rndkey0,$inout0
343 xorps $rndkey0,$inout1
344 xorps $rndkey0,$inout2
345 xorps $rndkey0,$inout3
346 $movkey ($key),$rndkey0
349 aes${dir} $rndkey1,$inout0
350 aes${dir} $rndkey1,$inout1
352 aes${dir} $rndkey1,$inout2
353 aes${dir} $rndkey1,$inout3
354 $movkey 16($key),$rndkey1
355 aes${dir} $rndkey0,$inout0
356 aes${dir} $rndkey0,$inout1
358 aes${dir} $rndkey0,$inout2
359 aes${dir} $rndkey0,$inout3
360 $movkey ($key),$rndkey0
363 aes${dir} $rndkey1,$inout0
364 aes${dir} $rndkey1,$inout1
365 aes${dir} $rndkey1,$inout2
366 aes${dir} $rndkey1,$inout3
367 aes${dir}last $rndkey0,$inout0
368 aes${dir}last $rndkey0,$inout1
369 aes${dir}last $rndkey0,$inout2
370 aes${dir}last $rndkey0,$inout3
372 .size _aesni_${dir}rypt4,.-_aesni_${dir}rypt4
375 sub aesni_generate6 {
377 # As already mentioned it takes in $key and $rounds, which are *not*
378 # preserved. $inout[0-5] is cipher/clear text...
380 .type _aesni_${dir}rypt6,\@abi-omnipotent
383 $movkey ($key),$rndkey0
385 $movkey 16($key),$rndkey1
387 xorps $rndkey0,$inout0
388 pxor $rndkey0,$inout1
389 aes${dir} $rndkey1,$inout0
390 pxor $rndkey0,$inout2
391 aes${dir} $rndkey1,$inout1
392 pxor $rndkey0,$inout3
393 aes${dir} $rndkey1,$inout2
394 pxor $rndkey0,$inout4
395 aes${dir} $rndkey1,$inout3
396 pxor $rndkey0,$inout5
398 aes${dir} $rndkey1,$inout4
399 $movkey ($key),$rndkey0
400 aes${dir} $rndkey1,$inout5
401 jmp .L${dir}_loop6_enter
404 aes${dir} $rndkey1,$inout0
405 aes${dir} $rndkey1,$inout1
407 aes${dir} $rndkey1,$inout2
408 aes${dir} $rndkey1,$inout3
409 aes${dir} $rndkey1,$inout4
410 aes${dir} $rndkey1,$inout5
411 .L${dir}_loop6_enter: # happens to be 16-byte aligned
412 $movkey 16($key),$rndkey1
413 aes${dir} $rndkey0,$inout0
414 aes${dir} $rndkey0,$inout1
416 aes${dir} $rndkey0,$inout2
417 aes${dir} $rndkey0,$inout3
418 aes${dir} $rndkey0,$inout4
419 aes${dir} $rndkey0,$inout5
420 $movkey ($key),$rndkey0
423 aes${dir} $rndkey1,$inout0
424 aes${dir} $rndkey1,$inout1
425 aes${dir} $rndkey1,$inout2
426 aes${dir} $rndkey1,$inout3
427 aes${dir} $rndkey1,$inout4
428 aes${dir} $rndkey1,$inout5
429 aes${dir}last $rndkey0,$inout0
430 aes${dir}last $rndkey0,$inout1
431 aes${dir}last $rndkey0,$inout2
432 aes${dir}last $rndkey0,$inout3
433 aes${dir}last $rndkey0,$inout4
434 aes${dir}last $rndkey0,$inout5
436 .size _aesni_${dir}rypt6,.-_aesni_${dir}rypt6
439 sub aesni_generate8 {
441 # As already mentioned it takes in $key and $rounds, which are *not*
442 # preserved. $inout[0-7] is cipher/clear text...
444 .type _aesni_${dir}rypt8,\@abi-omnipotent
447 $movkey ($key),$rndkey0
449 $movkey 16($key),$rndkey1
451 xorps $rndkey0,$inout0
452 xorps $rndkey0,$inout1
453 aes${dir} $rndkey1,$inout0
454 pxor $rndkey0,$inout2
455 aes${dir} $rndkey1,$inout1
456 pxor $rndkey0,$inout3
457 aes${dir} $rndkey1,$inout2
458 pxor $rndkey0,$inout4
459 aes${dir} $rndkey1,$inout3
460 pxor $rndkey0,$inout5
462 aes${dir} $rndkey1,$inout4
463 pxor $rndkey0,$inout6
464 aes${dir} $rndkey1,$inout5
465 pxor $rndkey0,$inout7
466 $movkey ($key),$rndkey0
467 aes${dir} $rndkey1,$inout6
468 aes${dir} $rndkey1,$inout7
469 $movkey 16($key),$rndkey1
470 jmp .L${dir}_loop8_enter
473 aes${dir} $rndkey1,$inout0
474 aes${dir} $rndkey1,$inout1
476 aes${dir} $rndkey1,$inout2
477 aes${dir} $rndkey1,$inout3
478 aes${dir} $rndkey1,$inout4
479 aes${dir} $rndkey1,$inout5
480 aes${dir} $rndkey1,$inout6
481 aes${dir} $rndkey1,$inout7
482 $movkey 16($key),$rndkey1
483 .L${dir}_loop8_enter: # happens to be 16-byte aligned
484 aes${dir} $rndkey0,$inout0
485 aes${dir} $rndkey0,$inout1
487 aes${dir} $rndkey0,$inout2
488 aes${dir} $rndkey0,$inout3
489 aes${dir} $rndkey0,$inout4
490 aes${dir} $rndkey0,$inout5
491 aes${dir} $rndkey0,$inout6
492 aes${dir} $rndkey0,$inout7
493 $movkey ($key),$rndkey0
496 aes${dir} $rndkey1,$inout0
497 aes${dir} $rndkey1,$inout1
498 aes${dir} $rndkey1,$inout2
499 aes${dir} $rndkey1,$inout3
500 aes${dir} $rndkey1,$inout4
501 aes${dir} $rndkey1,$inout5
502 aes${dir} $rndkey1,$inout6
503 aes${dir} $rndkey1,$inout7
504 aes${dir}last $rndkey0,$inout0
505 aes${dir}last $rndkey0,$inout1
506 aes${dir}last $rndkey0,$inout2
507 aes${dir}last $rndkey0,$inout3
508 aes${dir}last $rndkey0,$inout4
509 aes${dir}last $rndkey0,$inout5
510 aes${dir}last $rndkey0,$inout6
511 aes${dir}last $rndkey0,$inout7
513 .size _aesni_${dir}rypt8,.-_aesni_${dir}rypt8
516 &aesni_generate3("enc") if ($PREFIX eq "aesni");
517 &aesni_generate3("dec");
518 &aesni_generate4("enc") if ($PREFIX eq "aesni");
519 &aesni_generate4("dec");
520 &aesni_generate6("enc") if ($PREFIX eq "aesni");
521 &aesni_generate6("dec");
522 &aesni_generate8("enc") if ($PREFIX eq "aesni");
523 &aesni_generate8("dec");
525 if ($PREFIX eq "aesni") {
526 ########################################################################
527 # void aesni_ecb_encrypt (const void *in, void *out,
528 # size_t length, const AES_KEY *key,
531 .globl aesni_ecb_encrypt
532 .type aesni_ecb_encrypt,\@function,5
538 mov 240($key),$rounds # key->rounds
539 $movkey ($key),$rndkey0
540 mov $key,$key_ # backup $key
541 mov $rounds,$rnds_ # backup $rounds
542 test %r8d,%r8d # 5th argument
544 #--------------------------- ECB ENCRYPT ------------------------------#
548 movdqu ($inp),$inout0
549 movdqu 0x10($inp),$inout1
550 movdqu 0x20($inp),$inout2
551 movdqu 0x30($inp),$inout3
552 movdqu 0x40($inp),$inout4
553 movdqu 0x50($inp),$inout5
554 movdqu 0x60($inp),$inout6
555 movdqu 0x70($inp),$inout7
558 jmp .Lecb_enc_loop8_enter
561 movups $inout0,($out)
562 mov $key_,$key # restore $key
563 movdqu ($inp),$inout0
564 mov $rnds_,$rounds # restore $rounds
565 movups $inout1,0x10($out)
566 movdqu 0x10($inp),$inout1
567 movups $inout2,0x20($out)
568 movdqu 0x20($inp),$inout2
569 movups $inout3,0x30($out)
570 movdqu 0x30($inp),$inout3
571 movups $inout4,0x40($out)
572 movdqu 0x40($inp),$inout4
573 movups $inout5,0x50($out)
574 movdqu 0x50($inp),$inout5
575 movups $inout6,0x60($out)
576 movdqu 0x60($inp),$inout6
577 movups $inout7,0x70($out)
579 movdqu 0x70($inp),$inout7
581 .Lecb_enc_loop8_enter:
588 movups $inout0,($out)
589 mov $key_,$key # restore $key
590 movups $inout1,0x10($out)
591 mov $rnds_,$rounds # restore $rounds
592 movups $inout2,0x20($out)
593 movups $inout3,0x30($out)
594 movups $inout4,0x40($out)
595 movups $inout5,0x50($out)
596 movups $inout6,0x60($out)
597 movups $inout7,0x70($out)
603 movups ($inp),$inout0
606 movups 0x10($inp),$inout1
608 movups 0x20($inp),$inout2
611 movups 0x30($inp),$inout3
613 movups 0x40($inp),$inout4
616 movups 0x50($inp),$inout5
618 movdqu 0x60($inp),$inout6
620 movups $inout0,($out)
621 movups $inout1,0x10($out)
622 movups $inout2,0x20($out)
623 movups $inout3,0x30($out)
624 movups $inout4,0x40($out)
625 movups $inout5,0x50($out)
626 movups $inout6,0x60($out)
631 &aesni_generate1("enc",$key,$rounds);
633 movups $inout0,($out)
637 xorps $inout2,$inout2
639 movups $inout0,($out)
640 movups $inout1,0x10($out)
645 movups $inout0,($out)
646 movups $inout1,0x10($out)
647 movups $inout2,0x20($out)
652 movups $inout0,($out)
653 movups $inout1,0x10($out)
654 movups $inout2,0x20($out)
655 movups $inout3,0x30($out)
659 xorps $inout5,$inout5
661 movups $inout0,($out)
662 movups $inout1,0x10($out)
663 movups $inout2,0x20($out)
664 movups $inout3,0x30($out)
665 movups $inout4,0x40($out)
670 movups $inout0,($out)
671 movups $inout1,0x10($out)
672 movups $inout2,0x20($out)
673 movups $inout3,0x30($out)
674 movups $inout4,0x40($out)
675 movups $inout5,0x50($out)
677 \f#--------------------------- ECB DECRYPT ------------------------------#
683 movdqu ($inp),$inout0
684 movdqu 0x10($inp),$inout1
685 movdqu 0x20($inp),$inout2
686 movdqu 0x30($inp),$inout3
687 movdqu 0x40($inp),$inout4
688 movdqu 0x50($inp),$inout5
689 movdqu 0x60($inp),$inout6
690 movdqu 0x70($inp),$inout7
693 jmp .Lecb_dec_loop8_enter
696 movups $inout0,($out)
697 mov $key_,$key # restore $key
698 movdqu ($inp),$inout0
699 mov $rnds_,$rounds # restore $rounds
700 movups $inout1,0x10($out)
701 movdqu 0x10($inp),$inout1
702 movups $inout2,0x20($out)
703 movdqu 0x20($inp),$inout2
704 movups $inout3,0x30($out)
705 movdqu 0x30($inp),$inout3
706 movups $inout4,0x40($out)
707 movdqu 0x40($inp),$inout4
708 movups $inout5,0x50($out)
709 movdqu 0x50($inp),$inout5
710 movups $inout6,0x60($out)
711 movdqu 0x60($inp),$inout6
712 movups $inout7,0x70($out)
714 movdqu 0x70($inp),$inout7
716 .Lecb_dec_loop8_enter:
720 $movkey ($key_),$rndkey0
724 movups $inout0,($out)
725 mov $key_,$key # restore $key
726 movups $inout1,0x10($out)
727 mov $rnds_,$rounds # restore $rounds
728 movups $inout2,0x20($out)
729 movups $inout3,0x30($out)
730 movups $inout4,0x40($out)
731 movups $inout5,0x50($out)
732 movups $inout6,0x60($out)
733 movups $inout7,0x70($out)
739 movups ($inp),$inout0
742 movups 0x10($inp),$inout1
744 movups 0x20($inp),$inout2
747 movups 0x30($inp),$inout3
749 movups 0x40($inp),$inout4
752 movups 0x50($inp),$inout5
754 movups 0x60($inp),$inout6
755 $movkey ($key),$rndkey0
757 movups $inout0,($out)
758 movups $inout1,0x10($out)
759 movups $inout2,0x20($out)
760 movups $inout3,0x30($out)
761 movups $inout4,0x40($out)
762 movups $inout5,0x50($out)
763 movups $inout6,0x60($out)
768 &aesni_generate1("dec",$key,$rounds);
770 movups $inout0,($out)
774 xorps $inout2,$inout2
776 movups $inout0,($out)
777 movups $inout1,0x10($out)
782 movups $inout0,($out)
783 movups $inout1,0x10($out)
784 movups $inout2,0x20($out)
789 movups $inout0,($out)
790 movups $inout1,0x10($out)
791 movups $inout2,0x20($out)
792 movups $inout3,0x30($out)
796 xorps $inout5,$inout5
798 movups $inout0,($out)
799 movups $inout1,0x10($out)
800 movups $inout2,0x20($out)
801 movups $inout3,0x30($out)
802 movups $inout4,0x40($out)
807 movups $inout0,($out)
808 movups $inout1,0x10($out)
809 movups $inout2,0x20($out)
810 movups $inout3,0x30($out)
811 movups $inout4,0x40($out)
812 movups $inout5,0x50($out)
816 .size aesni_ecb_encrypt,.-aesni_ecb_encrypt
820 ######################################################################
821 # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out,
822 # size_t blocks, const AES_KEY *key,
823 # const char *ivec,char *cmac);
825 # Handles only complete blocks, operates on 64-bit counter and
826 # does not update *ivec! Nor does it finalize CMAC value
827 # (see engine/eng_aesni.c for details)
830 my $cmac="%r9"; # 6th argument
832 my $increment="%xmm6";
833 my $bswap_mask="%xmm7";
836 .globl aesni_ccm64_encrypt_blocks
837 .type aesni_ccm64_encrypt_blocks,\@function,6
839 aesni_ccm64_encrypt_blocks:
841 $code.=<<___ if ($win64);
844 movaps %xmm7,0x10(%rsp)
845 movaps %xmm8,0x20(%rsp)
846 movaps %xmm9,0x30(%rsp)
850 mov 240($key),$rounds # key->rounds
852 movdqa .Lincrement64(%rip),$increment
853 movdqa .Lbswap_mask(%rip),$bswap_mask
857 movdqu ($cmac),$inout1
860 pshufb $bswap_mask,$iv
861 jmp .Lccm64_enc_outer
864 $movkey ($key_),$rndkey0
866 movups ($inp),$in0 # load inp
868 xorps $rndkey0,$inout0 # counter
869 $movkey 16($key_),$rndkey1
872 xorps $rndkey0,$inout1 # cmac^=inp
873 $movkey ($key),$rndkey0
876 aesenc $rndkey1,$inout0
878 aesenc $rndkey1,$inout1
879 $movkey 16($key),$rndkey1
880 aesenc $rndkey0,$inout0
882 aesenc $rndkey0,$inout1
883 $movkey 0($key),$rndkey0
884 jnz .Lccm64_enc2_loop
885 aesenc $rndkey1,$inout0
886 aesenc $rndkey1,$inout1
888 aesenclast $rndkey0,$inout0
889 aesenclast $rndkey0,$inout1
893 xorps $inout0,$in0 # inp ^= E(iv)
895 movups $in0,($out) # save output
897 pshufb $bswap_mask,$inout0
898 jnz .Lccm64_enc_outer
900 movups $inout1,($cmac)
902 $code.=<<___ if ($win64);
904 movaps 0x10(%rsp),%xmm7
905 movaps 0x20(%rsp),%xmm8
906 movaps 0x30(%rsp),%xmm9
912 .size aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks
914 ######################################################################
916 .globl aesni_ccm64_decrypt_blocks
917 .type aesni_ccm64_decrypt_blocks,\@function,6
919 aesni_ccm64_decrypt_blocks:
921 $code.=<<___ if ($win64);
924 movaps %xmm7,0x10(%rsp)
925 movaps %xmm8,0x20(%rsp)
926 movaps %xmm9,0x30(%rsp)
930 mov 240($key),$rounds # key->rounds
932 movdqu ($cmac),$inout1
933 movdqa .Lincrement64(%rip),$increment
934 movdqa .Lbswap_mask(%rip),$bswap_mask
939 pshufb $bswap_mask,$iv
941 &aesni_generate1("enc",$key,$rounds);
943 movups ($inp),$in0 # load inp
946 jmp .Lccm64_dec_outer
949 xorps $inout0,$in0 # inp ^= E(iv)
952 movups $in0,($out) # save output
954 pshufb $bswap_mask,$inout0
959 $movkey ($key_),$rndkey0
961 $movkey 16($key_),$rndkey1
964 xorps $rndkey0,$inout0
965 xorps $in0,$inout1 # cmac^=out
966 $movkey ($key),$rndkey0
969 aesenc $rndkey1,$inout0
971 aesenc $rndkey1,$inout1
972 $movkey 16($key),$rndkey1
973 aesenc $rndkey0,$inout0
975 aesenc $rndkey0,$inout1
976 $movkey 0($key),$rndkey0
977 jnz .Lccm64_dec2_loop
978 movups ($inp),$in0 # load inp
980 aesenc $rndkey1,$inout0
981 aesenc $rndkey1,$inout1
983 aesenclast $rndkey0,$inout0
984 aesenclast $rndkey0,$inout1
985 jmp .Lccm64_dec_outer
989 #xorps $in0,$inout1 # cmac^=out
991 &aesni_generate1("enc",$key_,$rounds,$inout1,$in0);
993 movups $inout1,($cmac)
995 $code.=<<___ if ($win64);
997 movaps 0x10(%rsp),%xmm7
998 movaps 0x20(%rsp),%xmm8
999 movaps 0x30(%rsp),%xmm9
1005 .size aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks
1008 ######################################################################
1009 # void aesni_ctr32_encrypt_blocks (const void *in, void *out,
1010 # size_t blocks, const AES_KEY *key,
1011 # const char *ivec);
1013 # Handles only complete blocks, operates on 32-bit counter and
1014 # does not update *ivec! (see crypto/modes/ctr128.c for details)
1016 # Overhaul based on suggestions from Shay Gueron and Vlad Krasnov,
1017 # http://rt.openssl.org/Ticket/Display.html?id=3031&user=guest&pass=guest.
1018 # Keywords are full unroll and modulo-schedule counter calculations
1019 # with zero-round key xor.
1021 my ($in0,$in1,$in2,$in3,$in4,$in5)=map("%xmm$_",(10..15));
1022 my ($key0,$ctr)=("${key_}d","${ivp}d");
1023 my $frame_size = 0x80 + ($win64?160:0);
1026 .globl aesni_ctr32_encrypt_blocks
1027 .type aesni_ctr32_encrypt_blocks,\@function,5
1029 aesni_ctr32_encrypt_blocks:
1032 sub \$$frame_size,%rsp
1033 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
1035 $code.=<<___ if ($win64);
1036 movaps %xmm6,-0xa8(%rax)
1037 movaps %xmm7,-0x98(%rax)
1038 movaps %xmm8,-0x88(%rax)
1039 movaps %xmm9,-0x78(%rax)
1040 movaps %xmm10,-0x68(%rax)
1041 movaps %xmm11,-0x58(%rax)
1042 movaps %xmm12,-0x48(%rax)
1043 movaps %xmm13,-0x38(%rax)
1044 movaps %xmm14,-0x28(%rax)
1045 movaps %xmm15,-0x18(%rax)
1052 je .Lctr32_one_shortcut
1054 movdqu ($ivp),$inout0
1055 movdqu ($key),$rndkey0
1056 mov 12($ivp),$ctr # counter LSB
1057 pxor $rndkey0,$inout0
1058 mov 12($key),$key0 # 0-round key LSB
1059 movdqa $inout0,0x00(%rsp) # populate counter block
1061 movdqa $inout0,0x10(%rsp)
1062 movdqa $inout0,0x20(%rsp)
1063 movdqa $inout0,0x30(%rsp)
1064 movdqa $inout0,0x40(%rsp)
1065 movdqa $inout0,0x50(%rsp)
1066 movdqa $inout0,0x60(%rsp)
1067 movdqa $inout0,0x70(%rsp)
1069 mov 240($key),$rounds # key->rounds
1077 mov %r9d,0x10+12(%rsp)
1079 mov %r10d,0x20+12(%rsp)
1084 mov %r9d,0x30+12(%rsp)
1087 mov %r10d,0x40+12(%rsp)
1092 mov %r9d,0x50+12(%rsp)
1095 mov %r10d,0x60+12(%rsp)
1098 mov %r9d,0x70+12(%rsp)
1100 $movkey 0x10($key),$rndkey1
1102 movdqa 0x10(%rsp),$inout1
1103 movdqa 0x20(%rsp),$inout2
1104 movdqa 0x30(%rsp),$inout3
1105 movdqa 0x40(%rsp),$inout4
1106 movdqa 0x50(%rsp),$inout5
1111 lea 0x80($key),$key # size optimization
1118 movdqa 0x60(%rsp),$inout6
1119 aesenc $rndkey1,$inout0
1121 movdqa 0x70(%rsp),$inout7
1122 aesenc $rndkey1,$inout1
1124 $movkey 0x20-0x80($key),$rndkey0
1125 aesenc $rndkey1,$inout2
1127 aesenc $rndkey1,$inout3
1128 mov %r9d,0x00+12(%rsp)
1130 aesenc $rndkey1,$inout4
1131 aesenc $rndkey1,$inout5
1132 aesenc $rndkey1,$inout6
1133 aesenc $rndkey1,$inout7
1134 $movkey 0x30-0x80($key),$rndkey1
1136 for($i=2;$i<8;$i++) {
1137 my $rndkeyx = ($i&1)?$rndkey1:$rndkey0;
1139 aesenc $rndkeyx,$inout0
1140 aesenc $rndkeyx,$inout1
1142 aesenc $rndkeyx,$inout2
1144 aesenc $rndkeyx,$inout3
1145 mov %r9d,`0x10*($i-1)`+12(%rsp)
1147 aesenc $rndkeyx,$inout4
1148 aesenc $rndkeyx,$inout5
1149 aesenc $rndkeyx,$inout6
1150 aesenc $rndkeyx,$inout7
1151 $movkey `0x20+0x10*$i`-0x80($key),$rndkeyx
1155 aesenc $rndkey0,$inout0
1156 aesenc $rndkey0,$inout1
1158 aesenc $rndkey0,$inout2
1160 aesenc $rndkey0,$inout3
1161 mov %r9d,0x70+12(%rsp)
1162 aesenc $rndkey0,$inout4
1163 aesenc $rndkey0,$inout5
1164 aesenc $rndkey0,$inout6
1165 movdqu 0x00($inp),$in0
1166 aesenc $rndkey0,$inout7
1167 $movkey 0xa0-0x80($key),$rndkey0
1172 aesenc $rndkey1,$inout0
1173 aesenc $rndkey1,$inout1
1174 aesenc $rndkey1,$inout2
1175 aesenc $rndkey1,$inout3
1176 aesenc $rndkey1,$inout4
1177 aesenc $rndkey1,$inout5
1178 aesenc $rndkey1,$inout6
1179 aesenc $rndkey1,$inout7
1180 $movkey 0xb0-0x80($key),$rndkey1
1182 aesenc $rndkey0,$inout0
1183 aesenc $rndkey0,$inout1
1184 aesenc $rndkey0,$inout2
1185 aesenc $rndkey0,$inout3
1186 aesenc $rndkey0,$inout4
1187 aesenc $rndkey0,$inout5
1188 aesenc $rndkey0,$inout6
1189 aesenc $rndkey0,$inout7
1190 $movkey 0xc0-0x80($key),$rndkey0
1193 aesenc $rndkey1,$inout0
1194 aesenc $rndkey1,$inout1
1195 aesenc $rndkey1,$inout2
1196 aesenc $rndkey1,$inout3
1197 aesenc $rndkey1,$inout4
1198 aesenc $rndkey1,$inout5
1199 aesenc $rndkey1,$inout6
1200 aesenc $rndkey1,$inout7
1201 $movkey 0xd0-0x80($key),$rndkey1
1203 aesenc $rndkey0,$inout0
1204 aesenc $rndkey0,$inout1
1205 aesenc $rndkey0,$inout2
1206 aesenc $rndkey0,$inout3
1207 aesenc $rndkey0,$inout4
1208 aesenc $rndkey0,$inout5
1209 aesenc $rndkey0,$inout6
1210 aesenc $rndkey0,$inout7
1211 $movkey 0xe0-0x80($key),$rndkey0
1214 aesenc $rndkey1,$inout0
1215 movdqu 0x10($inp),$in1
1217 aesenc $rndkey1,$inout1
1218 movdqu 0x20($inp),$in2
1220 aesenc $rndkey1,$inout2
1221 movdqu 0x30($inp),$in3
1223 aesenc $rndkey1,$inout3
1224 movdqu 0x40($inp),$in4
1226 aesenc $rndkey1,$inout4
1227 movdqu 0x50($inp),$in5
1229 aesenc $rndkey1,$inout5
1231 aesenc $rndkey1,$inout6
1232 aesenc $rndkey1,$inout7
1233 movdqu 0x60($inp),$rndkey1
1235 aesenclast $in0,$inout0
1236 pxor $rndkey0,$rndkey1
1237 movdqu 0x70($inp),$in0
1239 aesenclast $in1,$inout1
1241 movdqa 0x00(%rsp),$in1 # load next counter block
1242 aesenclast $in2,$inout2
1243 movdqa 0x10(%rsp),$in2
1244 aesenclast $in3,$inout3
1245 movdqa 0x20(%rsp),$in3
1246 aesenclast $in4,$inout4
1247 movdqa 0x30(%rsp),$in4
1248 aesenclast $in5,$inout5
1249 movdqa 0x40(%rsp),$in5
1250 aesenclast $rndkey1,$inout6
1251 movdqa 0x50(%rsp),$rndkey0
1252 aesenclast $in0,$inout7
1253 $movkey 0x10-0x80($key),$rndkey1
1255 movups $inout0,($out) # store output
1257 movups $inout1,0x10($out)
1259 movups $inout2,0x20($out)
1261 movups $inout3,0x30($out)
1263 movups $inout4,0x40($out)
1265 movups $inout5,0x50($out)
1266 movdqa $rndkey0,$inout5
1267 movups $inout6,0x60($out)
1268 movups $inout7,0x70($out)
1276 lea -0x80($key),$key
1283 movdqa 0x60(%rsp),$inout6
1285 $movkey 16($key),$rndkey0
1286 aesenc $rndkey1,$inout0
1288 aesenc $rndkey1,$inout1
1290 aesenc $rndkey1,$inout2
1292 aesenc $rndkey1,$inout3
1293 aesenc $rndkey1,$inout4
1294 aesenc $rndkey1,$inout5
1295 aesenc $rndkey1,$inout6
1296 pxor $inout7,$inout7
1297 $movkey 16($key),$rndkey1
1299 call .Lenc_loop8_enter
1302 movups 0x10($inp),$in1
1303 movups 0x20($inp),$in2
1305 movups 0x30($inp),$in3
1307 movups 0x40($inp),$in0
1309 movups $inout0,($out)
1311 movups $inout1,0x10($out)
1313 movups $inout2,0x20($out)
1314 movups $inout3,0x30($out)
1315 movups $inout4,0x40($out)
1319 movups 0x50($inp),$in1
1321 movups $inout5,0x50($out)
1324 movups 0x60($inp),$in2
1326 movups $inout6,0x60($out)
1331 aesenc $rndkey1,$inout0
1333 aesenc $rndkey1,$inout1
1334 aesenc $rndkey1,$inout2
1335 aesenc $rndkey1,$inout3
1336 $movkey ($key),$rndkey1
1339 aesenclast $rndkey1,$inout0
1340 aesenclast $rndkey1,$inout1
1341 aesenclast $rndkey1,$inout2
1342 aesenclast $rndkey1,$inout3
1346 movups $inout0,($out)
1350 movups 0x10($inp),$in1
1352 movups $inout1,0x10($out)
1355 movups 0x20($inp),$in2
1357 movups $inout2,0x20($out)
1361 movups 0x30($inp),$in3
1363 movups $inout3,0x30($out)
1367 .Lctr32_one_shortcut:
1368 movups ($ivp),$inout0
1370 mov 240($key),$rounds # key->rounds
1372 &aesni_generate1("enc",$key,$rounds);
1375 movups $inout0,($out)
1381 $code.=<<___ if ($win64);
1382 movaps -0xa0(%rbp),%xmm6
1383 movaps -0x90(%rbp),%xmm7
1384 movaps -0x80(%rbp),%xmm8
1385 movaps -0x70(%rbp),%xmm9
1386 movaps -0x60(%rbp),%xmm10
1387 movaps -0x50(%rbp),%xmm11
1388 movaps -0x40(%rbp),%xmm12
1389 movaps -0x30(%rbp),%xmm13
1390 movaps -0x20(%rbp),%xmm14
1391 movaps -0x10(%rbp),%xmm15
1398 .size aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
1402 ######################################################################
1403 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
1404 # const AES_KEY *key1, const AES_KEY *key2
1405 # const unsigned char iv[16]);
1408 my @tweak=map("%xmm$_",(10..15));
1409 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
1410 my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
1411 my $frame_size = 0x60 + ($win64?160:0);
1414 .globl aesni_xts_encrypt
1415 .type aesni_xts_encrypt,\@function,6
1420 sub \$$frame_size,%rsp
1421 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
1423 $code.=<<___ if ($win64);
1424 movaps %xmm6,-0xa8(%rax)
1425 movaps %xmm7,-0x98(%rax)
1426 movaps %xmm8,-0x88(%rax)
1427 movaps %xmm9,-0x78(%rax)
1428 movaps %xmm10,-0x68(%rax)
1429 movaps %xmm11,-0x58(%rax)
1430 movaps %xmm12,-0x48(%rax)
1431 movaps %xmm13,-0x38(%rax)
1432 movaps %xmm14,-0x28(%rax)
1433 movaps %xmm15,-0x18(%rax)
1438 movups ($ivp),@tweak[5] # load clear-text tweak
1439 mov 240(%r8),$rounds # key2->rounds
1440 mov 240($key),$rnds_ # key1->rounds
1442 # generate the tweak
1443 &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1445 mov $key,$key_ # backup $key
1446 mov $rnds_,$rounds # backup $rounds
1447 mov $len,$len_ # backup $len
1450 movdqa .Lxts_magic(%rip),$twmask
1452 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1454 for ($i=0;$i<4;$i++) {
1456 pshufd \$0x13,$twtmp,$twres
1458 movdqa @tweak[5],@tweak[$i]
1459 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1460 pand $twmask,$twres # isolate carry and residue
1461 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1462 pxor $twres,@tweak[5]
1472 jmp .Lxts_enc_grandloop
1475 .Lxts_enc_grandloop:
1476 pshufd \$0x13,$twtmp,$twres
1477 movdqa @tweak[5],@tweak[4]
1478 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1479 movdqu `16*0`($inp),$inout0 # load input
1480 pand $twmask,$twres # isolate carry and residue
1481 movdqu `16*1`($inp),$inout1
1482 pxor $twres,@tweak[5]
1484 movdqu `16*2`($inp),$inout2
1485 pxor @tweak[0],$inout0 # input^=tweak
1486 movdqu `16*3`($inp),$inout3
1487 pxor @tweak[1],$inout1
1488 movdqu `16*4`($inp),$inout4
1489 pxor @tweak[2],$inout2
1490 movdqu `16*5`($inp),$inout5
1491 lea `16*6`($inp),$inp
1492 pxor @tweak[3],$inout3
1493 $movkey ($key_),$rndkey0
1494 pxor @tweak[4],$inout4
1495 pxor @tweak[5],$inout5
1497 # inline _aesni_encrypt6 and interleave first and last rounds
1499 $movkey 16($key_),$rndkey1
1500 pxor $rndkey0,$inout0
1501 pxor $rndkey0,$inout1
1502 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks
1503 aesenc $rndkey1,$inout0
1505 pxor $rndkey0,$inout2
1506 movdqa @tweak[1],`16*1`(%rsp)
1507 aesenc $rndkey1,$inout1
1508 pxor $rndkey0,$inout3
1509 movdqa @tweak[2],`16*2`(%rsp)
1510 aesenc $rndkey1,$inout2
1511 pxor $rndkey0,$inout4
1512 movdqa @tweak[3],`16*3`(%rsp)
1513 aesenc $rndkey1,$inout3
1514 pxor $rndkey0,$inout5
1515 $movkey ($key),$rndkey0
1517 movdqa @tweak[4],`16*4`(%rsp)
1518 aesenc $rndkey1,$inout4
1519 movdqa @tweak[5],`16*5`(%rsp)
1520 aesenc $rndkey1,$inout5
1522 pcmpgtd @tweak[5],$twtmp
1523 jmp .Lxts_enc_loop6_enter
1527 aesenc $rndkey1,$inout0
1528 aesenc $rndkey1,$inout1
1530 aesenc $rndkey1,$inout2
1531 aesenc $rndkey1,$inout3
1532 aesenc $rndkey1,$inout4
1533 aesenc $rndkey1,$inout5
1534 .Lxts_enc_loop6_enter:
1535 $movkey 16($key),$rndkey1
1536 aesenc $rndkey0,$inout0
1537 aesenc $rndkey0,$inout1
1539 aesenc $rndkey0,$inout2
1540 aesenc $rndkey0,$inout3
1541 aesenc $rndkey0,$inout4
1542 aesenc $rndkey0,$inout5
1543 $movkey ($key),$rndkey0
1546 pshufd \$0x13,$twtmp,$twres
1548 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1549 aesenc $rndkey1,$inout0
1550 pand $twmask,$twres # isolate carry and residue
1551 aesenc $rndkey1,$inout1
1552 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1553 aesenc $rndkey1,$inout2
1554 pxor $twres,@tweak[5]
1555 aesenc $rndkey1,$inout3
1556 aesenc $rndkey1,$inout4
1557 aesenc $rndkey1,$inout5
1558 $movkey 16($key),$rndkey1
1560 pshufd \$0x13,$twtmp,$twres
1562 movdqa @tweak[5],@tweak[0]
1563 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1564 aesenc $rndkey0,$inout0
1565 pand $twmask,$twres # isolate carry and residue
1566 aesenc $rndkey0,$inout1
1567 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1568 aesenc $rndkey0,$inout2
1569 pxor $twres,@tweak[5]
1570 aesenc $rndkey0,$inout3
1571 aesenc $rndkey0,$inout4
1572 aesenc $rndkey0,$inout5
1573 $movkey 32($key),$rndkey0
1575 pshufd \$0x13,$twtmp,$twres
1577 movdqa @tweak[5],@tweak[1]
1578 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1579 aesenc $rndkey1,$inout0
1580 pand $twmask,$twres # isolate carry and residue
1581 aesenc $rndkey1,$inout1
1582 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1583 aesenc $rndkey1,$inout2
1584 pxor $twres,@tweak[5]
1585 aesenc $rndkey1,$inout3
1586 aesenc $rndkey1,$inout4
1587 aesenc $rndkey1,$inout5
1589 pshufd \$0x13,$twtmp,$twres
1591 movdqa @tweak[5],@tweak[2]
1592 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1593 aesenclast $rndkey0,$inout0
1594 pand $twmask,$twres # isolate carry and residue
1595 aesenclast $rndkey0,$inout1
1596 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1597 aesenclast $rndkey0,$inout2
1598 pxor $twres,@tweak[5]
1599 aesenclast $rndkey0,$inout3
1600 aesenclast $rndkey0,$inout4
1601 aesenclast $rndkey0,$inout5
1603 pshufd \$0x13,$twtmp,$twres
1605 movdqa @tweak[5],@tweak[3]
1606 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1607 xorps `16*0`(%rsp),$inout0 # output^=tweak
1608 pand $twmask,$twres # isolate carry and residue
1609 xorps `16*1`(%rsp),$inout1
1610 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1611 pxor $twres,@tweak[5]
1613 xorps `16*2`(%rsp),$inout2
1614 movups $inout0,`16*0`($out) # write output
1615 xorps `16*3`(%rsp),$inout3
1616 movups $inout1,`16*1`($out)
1617 xorps `16*4`(%rsp),$inout4
1618 movups $inout2,`16*2`($out)
1619 xorps `16*5`(%rsp),$inout5
1620 movups $inout3,`16*3`($out)
1621 mov $rnds_,$rounds # restore $rounds
1622 movups $inout4,`16*4`($out)
1623 movups $inout5,`16*5`($out)
1624 lea `16*6`($out),$out
1626 jnc .Lxts_enc_grandloop
1628 lea 3($rounds,$rounds),$rounds # restore original value
1629 mov $key_,$key # restore $key
1630 mov $rounds,$rnds_ # backup $rounds
1644 pshufd \$0x13,$twtmp,$twres
1645 movdqa @tweak[5],@tweak[4]
1646 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1647 movdqu ($inp),$inout0
1648 pand $twmask,$twres # isolate carry and residue
1649 movdqu 16*1($inp),$inout1
1650 pxor $twres,@tweak[5]
1652 movdqu 16*2($inp),$inout2
1653 pxor @tweak[0],$inout0
1654 movdqu 16*3($inp),$inout3
1655 pxor @tweak[1],$inout1
1656 movdqu 16*4($inp),$inout4
1658 pxor @tweak[2],$inout2
1659 pxor @tweak[3],$inout3
1660 pxor @tweak[4],$inout4
1662 call _aesni_encrypt6
1664 xorps @tweak[0],$inout0
1665 movdqa @tweak[5],@tweak[0]
1666 xorps @tweak[1],$inout1
1667 xorps @tweak[2],$inout2
1668 movdqu $inout0,($out)
1669 xorps @tweak[3],$inout3
1670 movdqu $inout1,16*1($out)
1671 xorps @tweak[4],$inout4
1672 movdqu $inout2,16*2($out)
1673 movdqu $inout3,16*3($out)
1674 movdqu $inout4,16*4($out)
1680 movups ($inp),$inout0
1682 xorps @tweak[0],$inout0
1684 &aesni_generate1("enc",$key,$rounds);
1686 xorps @tweak[0],$inout0
1687 movdqa @tweak[1],@tweak[0]
1688 movups $inout0,($out)
1694 movups ($inp),$inout0
1695 movups 16($inp),$inout1
1697 xorps @tweak[0],$inout0
1698 xorps @tweak[1],$inout1
1700 call _aesni_encrypt3
1702 xorps @tweak[0],$inout0
1703 movdqa @tweak[2],@tweak[0]
1704 xorps @tweak[1],$inout1
1705 movups $inout0,($out)
1706 movups $inout1,16*1($out)
1712 movups ($inp),$inout0
1713 movups 16*1($inp),$inout1
1714 movups 16*2($inp),$inout2
1716 xorps @tweak[0],$inout0
1717 xorps @tweak[1],$inout1
1718 xorps @tweak[2],$inout2
1720 call _aesni_encrypt3
1722 xorps @tweak[0],$inout0
1723 movdqa @tweak[3],@tweak[0]
1724 xorps @tweak[1],$inout1
1725 xorps @tweak[2],$inout2
1726 movups $inout0,($out)
1727 movups $inout1,16*1($out)
1728 movups $inout2,16*2($out)
1734 movups ($inp),$inout0
1735 movups 16*1($inp),$inout1
1736 movups 16*2($inp),$inout2
1737 xorps @tweak[0],$inout0
1738 movups 16*3($inp),$inout3
1740 xorps @tweak[1],$inout1
1741 xorps @tweak[2],$inout2
1742 xorps @tweak[3],$inout3
1744 call _aesni_encrypt4
1746 xorps @tweak[0],$inout0
1747 movdqa @tweak[5],@tweak[0]
1748 xorps @tweak[1],$inout1
1749 xorps @tweak[2],$inout2
1750 movups $inout0,($out)
1751 xorps @tweak[3],$inout3
1752 movups $inout1,16*1($out)
1753 movups $inout2,16*2($out)
1754 movups $inout3,16*3($out)
1765 movzb ($inp),%eax # borrow $rounds ...
1766 movzb -16($out),%ecx # ... and $key
1774 sub $len_,$out # rewind $out
1775 mov $key_,$key # restore $key
1776 mov $rnds_,$rounds # restore $rounds
1778 movups -16($out),$inout0
1779 xorps @tweak[0],$inout0
1781 &aesni_generate1("enc",$key,$rounds);
1783 xorps @tweak[0],$inout0
1784 movups $inout0,-16($out)
1788 $code.=<<___ if ($win64);
1789 movaps -0xa0(%rbp),%xmm6
1790 movaps -0x90(%rbp),%xmm7
1791 movaps -0x80(%rbp),%xmm8
1792 movaps -0x70(%rbp),%xmm9
1793 movaps -0x60(%rbp),%xmm10
1794 movaps -0x50(%rbp),%xmm11
1795 movaps -0x40(%rbp),%xmm12
1796 movaps -0x30(%rbp),%xmm13
1797 movaps -0x20(%rbp),%xmm14
1798 movaps -0x10(%rbp),%xmm15
1805 .size aesni_xts_encrypt,.-aesni_xts_encrypt
1809 .globl aesni_xts_decrypt
1810 .type aesni_xts_decrypt,\@function,6
1815 sub \$$frame_size,%rsp
1816 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
1818 $code.=<<___ if ($win64);
1819 movaps %xmm6,-0xa8(%rax)
1820 movaps %xmm7,-0x98(%rax)
1821 movaps %xmm8,-0x88(%rax)
1822 movaps %xmm9,-0x78(%rax)
1823 movaps %xmm10,-0x68(%rax)
1824 movaps %xmm11,-0x58(%rax)
1825 movaps %xmm12,-0x48(%rax)
1826 movaps %xmm13,-0x38(%rax)
1827 movaps %xmm14,-0x28(%rax)
1828 movaps %xmm15,-0x18(%rax)
1833 movups ($ivp),@tweak[5] # load clear-text tweak
1834 mov 240($key2),$rounds # key2->rounds
1835 mov 240($key),$rnds_ # key1->rounds
1837 # generate the tweak
1838 &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1840 xor %eax,%eax # if ($len%16) len-=16;
1846 mov $key,$key_ # backup $key
1847 mov $rnds_,$rounds # backup $rounds
1848 mov $len,$len_ # backup $len
1851 movdqa .Lxts_magic(%rip),$twmask
1853 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1855 for ($i=0;$i<4;$i++) {
1857 pshufd \$0x13,$twtmp,$twres
1859 movdqa @tweak[5],@tweak[$i]
1860 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1861 pand $twmask,$twres # isolate carry and residue
1862 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1863 pxor $twres,@tweak[5]
1873 jmp .Lxts_dec_grandloop
1876 .Lxts_dec_grandloop:
1877 pshufd \$0x13,$twtmp,$twres
1878 movdqa @tweak[5],@tweak[4]
1879 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1880 movdqu `16*0`($inp),$inout0 # load input
1881 pand $twmask,$twres # isolate carry and residue
1882 movdqu `16*1`($inp),$inout1
1883 pxor $twres,@tweak[5]
1885 movdqu `16*2`($inp),$inout2
1886 pxor @tweak[0],$inout0 # input^=tweak
1887 movdqu `16*3`($inp),$inout3
1888 pxor @tweak[1],$inout1
1889 movdqu `16*4`($inp),$inout4
1890 pxor @tweak[2],$inout2
1891 movdqu `16*5`($inp),$inout5
1892 lea `16*6`($inp),$inp
1893 pxor @tweak[3],$inout3
1894 $movkey ($key_),$rndkey0
1895 pxor @tweak[4],$inout4
1896 pxor @tweak[5],$inout5
1898 # inline _aesni_decrypt6 and interleave first and last rounds
1900 $movkey 16($key_),$rndkey1
1901 pxor $rndkey0,$inout0
1902 pxor $rndkey0,$inout1
1903 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks
1904 aesdec $rndkey1,$inout0
1906 pxor $rndkey0,$inout2
1907 movdqa @tweak[1],`16*1`(%rsp)
1908 aesdec $rndkey1,$inout1
1909 pxor $rndkey0,$inout3
1910 movdqa @tweak[2],`16*2`(%rsp)
1911 aesdec $rndkey1,$inout2
1912 pxor $rndkey0,$inout4
1913 movdqa @tweak[3],`16*3`(%rsp)
1914 aesdec $rndkey1,$inout3
1915 pxor $rndkey0,$inout5
1916 $movkey ($key),$rndkey0
1918 movdqa @tweak[4],`16*4`(%rsp)
1919 aesdec $rndkey1,$inout4
1920 movdqa @tweak[5],`16*5`(%rsp)
1921 aesdec $rndkey1,$inout5
1923 pcmpgtd @tweak[5],$twtmp
1924 jmp .Lxts_dec_loop6_enter
1928 aesdec $rndkey1,$inout0
1929 aesdec $rndkey1,$inout1
1931 aesdec $rndkey1,$inout2
1932 aesdec $rndkey1,$inout3
1933 aesdec $rndkey1,$inout4
1934 aesdec $rndkey1,$inout5
1935 .Lxts_dec_loop6_enter:
1936 $movkey 16($key),$rndkey1
1937 aesdec $rndkey0,$inout0
1938 aesdec $rndkey0,$inout1
1940 aesdec $rndkey0,$inout2
1941 aesdec $rndkey0,$inout3
1942 aesdec $rndkey0,$inout4
1943 aesdec $rndkey0,$inout5
1944 $movkey ($key),$rndkey0
1947 pshufd \$0x13,$twtmp,$twres
1949 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1950 aesdec $rndkey1,$inout0
1951 pand $twmask,$twres # isolate carry and residue
1952 aesdec $rndkey1,$inout1
1953 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1954 aesdec $rndkey1,$inout2
1955 pxor $twres,@tweak[5]
1956 aesdec $rndkey1,$inout3
1957 aesdec $rndkey1,$inout4
1958 aesdec $rndkey1,$inout5
1959 $movkey 16($key),$rndkey1
1961 pshufd \$0x13,$twtmp,$twres
1963 movdqa @tweak[5],@tweak[0]
1964 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1965 aesdec $rndkey0,$inout0
1966 pand $twmask,$twres # isolate carry and residue
1967 aesdec $rndkey0,$inout1
1968 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1969 aesdec $rndkey0,$inout2
1970 pxor $twres,@tweak[5]
1971 aesdec $rndkey0,$inout3
1972 aesdec $rndkey0,$inout4
1973 aesdec $rndkey0,$inout5
1974 $movkey 32($key),$rndkey0
1976 pshufd \$0x13,$twtmp,$twres
1978 movdqa @tweak[5],@tweak[1]
1979 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1980 aesdec $rndkey1,$inout0
1981 pand $twmask,$twres # isolate carry and residue
1982 aesdec $rndkey1,$inout1
1983 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1984 aesdec $rndkey1,$inout2
1985 pxor $twres,@tweak[5]
1986 aesdec $rndkey1,$inout3
1987 aesdec $rndkey1,$inout4
1988 aesdec $rndkey1,$inout5
1990 pshufd \$0x13,$twtmp,$twres
1992 movdqa @tweak[5],@tweak[2]
1993 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1994 aesdeclast $rndkey0,$inout0
1995 pand $twmask,$twres # isolate carry and residue
1996 aesdeclast $rndkey0,$inout1
1997 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1998 aesdeclast $rndkey0,$inout2
1999 pxor $twres,@tweak[5]
2000 aesdeclast $rndkey0,$inout3
2001 aesdeclast $rndkey0,$inout4
2002 aesdeclast $rndkey0,$inout5
2004 pshufd \$0x13,$twtmp,$twres
2006 movdqa @tweak[5],@tweak[3]
2007 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2008 xorps `16*0`(%rsp),$inout0 # output^=tweak
2009 pand $twmask,$twres # isolate carry and residue
2010 xorps `16*1`(%rsp),$inout1
2011 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
2012 pxor $twres,@tweak[5]
2014 xorps `16*2`(%rsp),$inout2
2015 movups $inout0,`16*0`($out) # write output
2016 xorps `16*3`(%rsp),$inout3
2017 movups $inout1,`16*1`($out)
2018 xorps `16*4`(%rsp),$inout4
2019 movups $inout2,`16*2`($out)
2020 xorps `16*5`(%rsp),$inout5
2021 movups $inout3,`16*3`($out)
2022 mov $rnds_,$rounds # restore $rounds
2023 movups $inout4,`16*4`($out)
2024 movups $inout5,`16*5`($out)
2025 lea `16*6`($out),$out
2027 jnc .Lxts_dec_grandloop
2029 lea 3($rounds,$rounds),$rounds # restore original value
2030 mov $key_,$key # restore $key
2031 mov $rounds,$rnds_ # backup $rounds
2045 pshufd \$0x13,$twtmp,$twres
2046 movdqa @tweak[5],@tweak[4]
2047 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2048 movdqu ($inp),$inout0
2049 pand $twmask,$twres # isolate carry and residue
2050 movdqu 16*1($inp),$inout1
2051 pxor $twres,@tweak[5]
2053 movdqu 16*2($inp),$inout2
2054 pxor @tweak[0],$inout0
2055 movdqu 16*3($inp),$inout3
2056 pxor @tweak[1],$inout1
2057 movdqu 16*4($inp),$inout4
2059 pxor @tweak[2],$inout2
2060 pxor @tweak[3],$inout3
2061 pxor @tweak[4],$inout4
2063 call _aesni_decrypt6
2065 xorps @tweak[0],$inout0
2066 xorps @tweak[1],$inout1
2067 xorps @tweak[2],$inout2
2068 movdqu $inout0,($out)
2069 xorps @tweak[3],$inout3
2070 movdqu $inout1,16*1($out)
2071 xorps @tweak[4],$inout4
2072 movdqu $inout2,16*2($out)
2074 movdqu $inout3,16*3($out)
2075 pcmpgtd @tweak[5],$twtmp
2076 movdqu $inout4,16*4($out)
2078 pshufd \$0x13,$twtmp,@tweak[1] # $twres
2082 movdqa @tweak[5],@tweak[0]
2083 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2084 pand $twmask,@tweak[1] # isolate carry and residue
2085 pxor @tweak[5],@tweak[1]
2090 movups ($inp),$inout0
2092 xorps @tweak[0],$inout0
2094 &aesni_generate1("dec",$key,$rounds);
2096 xorps @tweak[0],$inout0
2097 movdqa @tweak[1],@tweak[0]
2098 movups $inout0,($out)
2099 movdqa @tweak[2],@tweak[1]
2105 movups ($inp),$inout0
2106 movups 16($inp),$inout1
2108 xorps @tweak[0],$inout0
2109 xorps @tweak[1],$inout1
2111 call _aesni_decrypt3
2113 xorps @tweak[0],$inout0
2114 movdqa @tweak[2],@tweak[0]
2115 xorps @tweak[1],$inout1
2116 movdqa @tweak[3],@tweak[1]
2117 movups $inout0,($out)
2118 movups $inout1,16*1($out)
2124 movups ($inp),$inout0
2125 movups 16*1($inp),$inout1
2126 movups 16*2($inp),$inout2
2128 xorps @tweak[0],$inout0
2129 xorps @tweak[1],$inout1
2130 xorps @tweak[2],$inout2
2132 call _aesni_decrypt3
2134 xorps @tweak[0],$inout0
2135 movdqa @tweak[3],@tweak[0]
2136 xorps @tweak[1],$inout1
2137 movdqa @tweak[5],@tweak[1]
2138 xorps @tweak[2],$inout2
2139 movups $inout0,($out)
2140 movups $inout1,16*1($out)
2141 movups $inout2,16*2($out)
2147 pshufd \$0x13,$twtmp,$twres
2148 movdqa @tweak[5],@tweak[4]
2149 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2150 movups ($inp),$inout0
2151 pand $twmask,$twres # isolate carry and residue
2152 movups 16*1($inp),$inout1
2153 pxor $twres,@tweak[5]
2155 movups 16*2($inp),$inout2
2156 xorps @tweak[0],$inout0
2157 movups 16*3($inp),$inout3
2159 xorps @tweak[1],$inout1
2160 xorps @tweak[2],$inout2
2161 xorps @tweak[3],$inout3
2163 call _aesni_decrypt4
2165 xorps @tweak[0],$inout0
2166 movdqa @tweak[4],@tweak[0]
2167 xorps @tweak[1],$inout1
2168 movdqa @tweak[5],@tweak[1]
2169 xorps @tweak[2],$inout2
2170 movups $inout0,($out)
2171 xorps @tweak[3],$inout3
2172 movups $inout1,16*1($out)
2173 movups $inout2,16*2($out)
2174 movups $inout3,16*3($out)
2184 mov $key_,$key # restore $key
2185 mov $rnds_,$rounds # restore $rounds
2187 movups ($inp),$inout0
2188 xorps @tweak[1],$inout0
2190 &aesni_generate1("dec",$key,$rounds);
2192 xorps @tweak[1],$inout0
2193 movups $inout0,($out)
2196 movzb 16($inp),%eax # borrow $rounds ...
2197 movzb ($out),%ecx # ... and $key
2205 sub $len_,$out # rewind $out
2206 mov $key_,$key # restore $key
2207 mov $rnds_,$rounds # restore $rounds
2209 movups ($out),$inout0
2210 xorps @tweak[0],$inout0
2212 &aesni_generate1("dec",$key,$rounds);
2214 xorps @tweak[0],$inout0
2215 movups $inout0,($out)
2219 $code.=<<___ if ($win64);
2220 movaps -0xa0(%rbp),%xmm6
2221 movaps -0x90(%rbp),%xmm7
2222 movaps -0x80(%rbp),%xmm8
2223 movaps -0x70(%rbp),%xmm9
2224 movaps -0x60(%rbp),%xmm10
2225 movaps -0x50(%rbp),%xmm11
2226 movaps -0x40(%rbp),%xmm12
2227 movaps -0x30(%rbp),%xmm13
2228 movaps -0x20(%rbp),%xmm14
2229 movaps -0x10(%rbp),%xmm15
2236 .size aesni_xts_decrypt,.-aesni_xts_decrypt
2240 ########################################################################
2241 # void $PREFIX_cbc_encrypt (const void *inp, void *out,
2242 # size_t length, const AES_KEY *key,
2243 # unsigned char *ivp,const int enc);
2245 my $frame_size = 0x10 + ($win64?0x40:0); # used in decrypt
2247 .globl ${PREFIX}_cbc_encrypt
2248 .type ${PREFIX}_cbc_encrypt,\@function,6
2250 ${PREFIX}_cbc_encrypt:
2251 test $len,$len # check length
2254 mov 240($key),$rnds_ # key->rounds
2255 mov $key,$key_ # backup $key
2256 test %r9d,%r9d # 6th argument
2258 #--------------------------- CBC ENCRYPT ------------------------------#
2259 movups ($ivp),$inout0 # load iv as initial state
2267 movups ($inp),$inout1 # load input
2269 #xorps $inout1,$inout0
2271 &aesni_generate1("enc",$key,$rounds,$inout0,$inout1);
2273 mov $rnds_,$rounds # restore $rounds
2274 mov $key_,$key # restore $key
2275 movups $inout0,0($out) # store output
2281 movups $inout0,($ivp)
2285 mov $len,%rcx # zaps $key
2286 xchg $inp,$out # $inp is %rsi and $out is %rdi now
2287 .long 0x9066A4F3 # rep movsb
2288 mov \$16,%ecx # zero tail
2291 .long 0x9066AAF3 # rep stosb
2292 lea -16(%rdi),%rdi # rewind $out by 1 block
2293 mov $rnds_,$rounds # restore $rounds
2294 mov %rdi,%rsi # $inp and $out are the same
2295 mov $key_,$key # restore $key
2296 xor $len,$len # len=16
2297 jmp .Lcbc_enc_loop # one more spin
2298 \f#--------------------------- CBC DECRYPT ------------------------------#
2303 sub \$$frame_size,%rsp
2304 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
2306 $code.=<<___ if ($win64);
2307 movaps %xmm6,0x10(%rsp)
2308 movaps %xmm7,0x20(%rsp)
2309 movaps %xmm8,0x30(%rsp)
2310 movaps %xmm9,0x40(%rsp)
2323 jmp .Lcbc_dec_loop8_enter
2326 movaps $rndkey0,(%rsp) # save IV
2327 movups $inout7,($out)
2329 .Lcbc_dec_loop8_enter:
2330 $movkey ($key),$rndkey0
2331 movups ($inp),$inout0 # load input
2332 movups 0x10($inp),$inout1
2333 $movkey 16($key),$rndkey1
2336 movdqu 0x20($inp),$inout2
2337 xorps $rndkey0,$inout0
2338 movdqu 0x30($inp),$inout3
2339 xorps $rndkey0,$inout1
2340 movdqu 0x40($inp),$inout4
2341 aesdec $rndkey1,$inout0
2342 pxor $rndkey0,$inout2
2343 movdqu 0x50($inp),$inout5
2344 aesdec $rndkey1,$inout1
2345 pxor $rndkey0,$inout3
2346 movdqu 0x60($inp),$inout6
2347 aesdec $rndkey1,$inout2
2348 pxor $rndkey0,$inout4
2349 movdqu 0x70($inp),$inout7
2350 aesdec $rndkey1,$inout3
2351 pxor $rndkey0,$inout5
2353 aesdec $rndkey1,$inout4
2354 pxor $rndkey0,$inout6
2355 aesdec $rndkey1,$inout5
2356 pxor $rndkey0,$inout7
2357 $movkey ($key),$rndkey0
2358 aesdec $rndkey1,$inout6
2359 aesdec $rndkey1,$inout7
2360 $movkey 16($key),$rndkey1
2362 call .Ldec_loop8_enter
2364 movups ($inp),$rndkey1 # re-load input
2365 movups 0x10($inp),$rndkey0
2366 xorps (%rsp),$inout0 # ^= IV
2367 xorps $rndkey1,$inout1
2368 movups 0x20($inp),$rndkey1
2369 xorps $rndkey0,$inout2
2370 movups 0x30($inp),$rndkey0
2371 xorps $rndkey1,$inout3
2372 movups 0x40($inp),$rndkey1
2373 xorps $rndkey0,$inout4
2374 movups 0x50($inp),$rndkey0
2375 xorps $rndkey1,$inout5
2376 movups 0x60($inp),$rndkey1
2377 xorps $rndkey0,$inout6
2378 movups 0x70($inp),$rndkey0 # IV
2379 xorps $rndkey1,$inout7
2380 movups $inout0,($out)
2381 movups $inout1,0x10($out)
2382 movups $inout2,0x20($out)
2383 movups $inout3,0x30($out)
2384 mov $rnds_,$rounds # restore $rounds
2385 movups $inout4,0x40($out)
2386 mov $key_,$key # restore $key
2387 movups $inout5,0x50($out)
2389 movups $inout6,0x60($out)
2394 movaps $inout7,$inout0
2397 jle .Lcbc_dec_tail_collected
2398 movups $inout0,($out)
2399 lea 1($rnds_,$rnds_),$rounds
2402 movups ($inp),$inout0
2407 movups 0x10($inp),$inout1
2412 movups 0x20($inp),$inout2
2417 movups 0x30($inp),$inout3
2421 movups 0x40($inp),$inout4
2425 movups 0x50($inp),$inout5
2429 movups 0x60($inp),$inout6
2430 movaps $iv,(%rsp) # save IV
2431 call _aesni_decrypt8
2432 movups ($inp),$rndkey1
2433 movups 0x10($inp),$rndkey0
2434 xorps (%rsp),$inout0 # ^= IV
2435 xorps $rndkey1,$inout1
2436 movups 0x20($inp),$rndkey1
2437 xorps $rndkey0,$inout2
2438 movups 0x30($inp),$rndkey0
2439 xorps $rndkey1,$inout3
2440 movups 0x40($inp),$rndkey1
2441 xorps $rndkey0,$inout4
2442 movups 0x50($inp),$rndkey0
2443 xorps $rndkey1,$inout5
2444 movups 0x60($inp),$iv # IV
2445 xorps $rndkey0,$inout6
2446 movups $inout0,($out)
2447 movups $inout1,0x10($out)
2448 movups $inout2,0x20($out)
2449 movups $inout3,0x30($out)
2450 movups $inout4,0x40($out)
2451 movups $inout5,0x50($out)
2453 movaps $inout6,$inout0
2455 jmp .Lcbc_dec_tail_collected
2459 &aesni_generate1("dec",$key,$rounds);
2464 jmp .Lcbc_dec_tail_collected
2467 xorps $inout2,$inout2
2468 call _aesni_decrypt3
2471 movups $inout0,($out)
2473 movaps $inout1,$inout0
2476 jmp .Lcbc_dec_tail_collected
2479 call _aesni_decrypt3
2482 movups $inout0,($out)
2484 movups $inout1,0x10($out)
2486 movaps $inout2,$inout0
2489 jmp .Lcbc_dec_tail_collected
2492 call _aesni_decrypt4
2494 movups 0x30($inp),$iv
2496 movups $inout0,($out)
2498 movups $inout1,0x10($out)
2500 movups $inout2,0x20($out)
2501 movaps $inout3,$inout0
2504 jmp .Lcbc_dec_tail_collected
2507 xorps $inout5,$inout5
2508 call _aesni_decrypt6
2509 movups 0x10($inp),$rndkey1
2510 movups 0x20($inp),$rndkey0
2513 xorps $rndkey1,$inout2
2514 movups 0x30($inp),$rndkey1
2515 xorps $rndkey0,$inout3
2516 movups 0x40($inp),$iv
2517 xorps $rndkey1,$inout4
2518 movups $inout0,($out)
2519 movups $inout1,0x10($out)
2520 movups $inout2,0x20($out)
2521 movups $inout3,0x30($out)
2523 movaps $inout4,$inout0
2525 jmp .Lcbc_dec_tail_collected
2528 call _aesni_decrypt6
2529 movups 0x10($inp),$rndkey1
2530 movups 0x20($inp),$rndkey0
2533 xorps $rndkey1,$inout2
2534 movups 0x30($inp),$rndkey1
2535 xorps $rndkey0,$inout3
2536 movups 0x40($inp),$rndkey0
2537 xorps $rndkey1,$inout4
2538 movups 0x50($inp),$iv
2539 xorps $rndkey0,$inout5
2540 movups $inout0,($out)
2541 movups $inout1,0x10($out)
2542 movups $inout2,0x20($out)
2543 movups $inout3,0x30($out)
2544 movups $inout4,0x40($out)
2546 movaps $inout5,$inout0
2548 jmp .Lcbc_dec_tail_collected
2550 .Lcbc_dec_tail_collected:
2553 jnz .Lcbc_dec_tail_partial
2554 movups $inout0,($out)
2557 .Lcbc_dec_tail_partial:
2558 movaps $inout0,(%rsp)
2563 .long 0x9066A4F3 # rep movsb
2567 $code.=<<___ if ($win64);
2568 movaps 0x10(%rsp),%xmm6
2569 movaps 0x20(%rsp),%xmm7
2570 movaps 0x30(%rsp),%xmm8
2571 movaps 0x40(%rsp),%xmm9
2578 .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
2581 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
2582 # int bits, AES_KEY *key)
2583 { my ($inp,$bits,$key) = @_4args;
2587 .globl ${PREFIX}_set_decrypt_key
2588 .type ${PREFIX}_set_decrypt_key,\@abi-omnipotent
2590 ${PREFIX}_set_decrypt_key:
2591 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8
2592 call __aesni_set_encrypt_key
2593 shl \$4,$bits # rounds-1 after _aesni_set_encrypt_key
2596 lea 16($key,$bits),$inp # points at the end of key schedule
2598 $movkey ($key),%xmm0 # just swap
2599 $movkey ($inp),%xmm1
2600 $movkey %xmm0,($inp)
2601 $movkey %xmm1,($key)
2606 $movkey ($key),%xmm0 # swap and inverse
2607 $movkey ($inp),%xmm1
2612 $movkey %xmm0,16($inp)
2613 $movkey %xmm1,-16($key)
2615 ja .Ldec_key_inverse
2617 $movkey ($key),%xmm0 # inverse middle
2619 $movkey %xmm0,($inp)
2623 .LSEH_end_set_decrypt_key:
2624 .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
2627 # This is based on submission by
2629 # Huang Ying <ying.huang@intel.com>
2630 # Vinodh Gopal <vinodh.gopal@intel.com>
2633 # Agressively optimized in respect to aeskeygenassist's critical path
2634 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
2637 .globl ${PREFIX}_set_encrypt_key
2638 .type ${PREFIX}_set_encrypt_key,\@abi-omnipotent
2640 ${PREFIX}_set_encrypt_key:
2641 __aesni_set_encrypt_key:
2642 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8
2649 movups ($inp),%xmm0 # pull first 128 bits of *userKey
2650 xorps %xmm4,%xmm4 # low dword of xmm4 is assumed 0
2660 mov \$9,$bits # 10 rounds for 128-bit key
2661 $movkey %xmm0,($key) # round 0
2662 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 1
2663 call .Lkey_expansion_128_cold
2664 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 2
2665 call .Lkey_expansion_128
2666 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 3
2667 call .Lkey_expansion_128
2668 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 4
2669 call .Lkey_expansion_128
2670 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 5
2671 call .Lkey_expansion_128
2672 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 6
2673 call .Lkey_expansion_128
2674 aeskeygenassist \$0x40,%xmm0,%xmm1 # round 7
2675 call .Lkey_expansion_128
2676 aeskeygenassist \$0x80,%xmm0,%xmm1 # round 8
2677 call .Lkey_expansion_128
2678 aeskeygenassist \$0x1b,%xmm0,%xmm1 # round 9
2679 call .Lkey_expansion_128
2680 aeskeygenassist \$0x36,%xmm0,%xmm1 # round 10
2681 call .Lkey_expansion_128
2682 $movkey %xmm0,(%rax)
2683 mov $bits,80(%rax) # 240(%rdx)
2689 movq 16($inp),%xmm2 # remaining 1/3 of *userKey
2690 mov \$11,$bits # 12 rounds for 192
2691 $movkey %xmm0,($key) # round 0
2692 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 1,2
2693 call .Lkey_expansion_192a_cold
2694 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 2,3
2695 call .Lkey_expansion_192b
2696 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 4,5
2697 call .Lkey_expansion_192a
2698 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 5,6
2699 call .Lkey_expansion_192b
2700 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 7,8
2701 call .Lkey_expansion_192a
2702 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 8,9
2703 call .Lkey_expansion_192b
2704 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 10,11
2705 call .Lkey_expansion_192a
2706 aeskeygenassist \$0x80,%xmm2,%xmm1 # round 11,12
2707 call .Lkey_expansion_192b
2708 $movkey %xmm0,(%rax)
2709 mov $bits,48(%rax) # 240(%rdx)
2715 movups 16($inp),%xmm2 # remaning half of *userKey
2716 mov \$13,$bits # 14 rounds for 256
2718 $movkey %xmm0,($key) # round 0
2719 $movkey %xmm2,16($key) # round 1
2720 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 2
2721 call .Lkey_expansion_256a_cold
2722 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 3
2723 call .Lkey_expansion_256b
2724 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 4
2725 call .Lkey_expansion_256a
2726 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 5
2727 call .Lkey_expansion_256b
2728 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 6
2729 call .Lkey_expansion_256a
2730 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 7
2731 call .Lkey_expansion_256b
2732 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 8
2733 call .Lkey_expansion_256a
2734 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 9
2735 call .Lkey_expansion_256b
2736 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 10
2737 call .Lkey_expansion_256a
2738 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 11
2739 call .Lkey_expansion_256b
2740 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 12
2741 call .Lkey_expansion_256a
2742 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 13
2743 call .Lkey_expansion_256b
2744 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 14
2745 call .Lkey_expansion_256a
2746 $movkey %xmm0,(%rax)
2747 mov $bits,16(%rax) # 240(%rdx)
2757 .LSEH_end_set_encrypt_key:
2760 .Lkey_expansion_128:
2761 $movkey %xmm0,(%rax)
2763 .Lkey_expansion_128_cold:
2764 shufps \$0b00010000,%xmm0,%xmm4
2766 shufps \$0b10001100,%xmm0,%xmm4
2768 shufps \$0b11111111,%xmm1,%xmm1 # critical path
2773 .Lkey_expansion_192a:
2774 $movkey %xmm0,(%rax)
2776 .Lkey_expansion_192a_cold:
2778 .Lkey_expansion_192b_warm:
2779 shufps \$0b00010000,%xmm0,%xmm4
2782 shufps \$0b10001100,%xmm0,%xmm4
2785 pshufd \$0b01010101,%xmm1,%xmm1 # critical path
2788 pshufd \$0b11111111,%xmm0,%xmm3
2793 .Lkey_expansion_192b:
2795 shufps \$0b01000100,%xmm0,%xmm5
2796 $movkey %xmm5,(%rax)
2797 shufps \$0b01001110,%xmm2,%xmm3
2798 $movkey %xmm3,16(%rax)
2800 jmp .Lkey_expansion_192b_warm
2803 .Lkey_expansion_256a:
2804 $movkey %xmm2,(%rax)
2806 .Lkey_expansion_256a_cold:
2807 shufps \$0b00010000,%xmm0,%xmm4
2809 shufps \$0b10001100,%xmm0,%xmm4
2811 shufps \$0b11111111,%xmm1,%xmm1 # critical path
2816 .Lkey_expansion_256b:
2817 $movkey %xmm0,(%rax)
2820 shufps \$0b00010000,%xmm2,%xmm4
2822 shufps \$0b10001100,%xmm2,%xmm4
2824 shufps \$0b10101010,%xmm1,%xmm1 # critical path
2827 .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
2828 .size __aesni_set_encrypt_key,.-__aesni_set_encrypt_key
2835 .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
2843 .byte 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
2845 .asciz "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
2849 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2850 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
2858 .extern __imp_RtlVirtualUnwind
2860 $code.=<<___ if ($PREFIX eq "aesni");
2861 .type ecb_se_handler,\@abi-omnipotent
2875 mov 152($context),%rax # pull context->Rsp
2877 jmp .Lcommon_seh_tail
2878 .size ecb_se_handler,.-ecb_se_handler
2880 .type ccm64_se_handler,\@abi-omnipotent
2894 mov 120($context),%rax # pull context->Rax
2895 mov 248($context),%rbx # pull context->Rip
2897 mov 8($disp),%rsi # disp->ImageBase
2898 mov 56($disp),%r11 # disp->HandlerData
2900 mov 0(%r11),%r10d # HandlerData[0]
2901 lea (%rsi,%r10),%r10 # prologue label
2902 cmp %r10,%rbx # context->Rip<prologue label
2903 jb .Lcommon_seh_tail
2905 mov 152($context),%rax # pull context->Rsp
2907 mov 4(%r11),%r10d # HandlerData[1]
2908 lea (%rsi,%r10),%r10 # epilogue label
2909 cmp %r10,%rbx # context->Rip>=epilogue label
2910 jae .Lcommon_seh_tail
2912 lea 0(%rax),%rsi # %xmm save area
2913 lea 512($context),%rdi # &context.Xmm6
2914 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax)
2915 .long 0xa548f3fc # cld; rep movsq
2916 lea 0x58(%rax),%rax # adjust stack pointer
2918 jmp .Lcommon_seh_tail
2919 .size ccm64_se_handler,.-ccm64_se_handler
2921 .type ctr_xts_se_handler,\@abi-omnipotent
2935 mov 120($context),%rax # pull context->Rax
2936 mov 248($context),%rbx # pull context->Rip
2938 mov 8($disp),%rsi # disp->ImageBase
2939 mov 56($disp),%r11 # disp->HandlerData
2941 mov 0(%r11),%r10d # HandlerData[0]
2942 lea (%rsi,%r10),%r10 # prologue lable
2943 cmp %r10,%rbx # context->Rip<prologue label
2944 jb .Lcommon_seh_tail
2946 mov 152($context),%rax # pull context->Rsp
2948 mov 4(%r11),%r10d # HandlerData[1]
2949 lea (%rsi,%r10),%r10 # epilogue label
2950 cmp %r10,%rbx # context->Rip>=epilogue label
2951 jae .Lcommon_seh_tail
2953 mov 160($context),%rax # pull context->Rbp
2954 lea -0xa0(%rax),%rsi # %xmm save area
2955 lea 512($context),%rdi # & context.Xmm6
2956 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax)
2957 .long 0xa548f3fc # cld; rep movsq
2959 jmp .Lcommon_rbp_tail
2960 .size ctr_xts_se_handler,.-ctr_xts_se_handler
2963 .type cbc_se_handler,\@abi-omnipotent
2977 mov 152($context),%rax # pull context->Rsp
2978 mov 248($context),%rbx # pull context->Rip
2980 lea .Lcbc_decrypt(%rip),%r10
2981 cmp %r10,%rbx # context->Rip<"prologue" label
2982 jb .Lcommon_seh_tail
2984 lea .Lcbc_decrypt_body(%rip),%r10
2985 cmp %r10,%rbx # context->Rip<cbc_decrypt_body
2986 jb .Lrestore_cbc_rax
2988 lea .Lcbc_ret(%rip),%r10
2989 cmp %r10,%rbx # context->Rip>="epilogue" label
2990 jae .Lcommon_seh_tail
2992 lea 16(%rax),%rsi # %xmm save area
2993 lea 512($context),%rdi # &context.Xmm6
2994 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax)
2995 .long 0xa548f3fc # cld; rep movsq
2998 mov 160($context),%rax # pull context->Rbp
2999 mov (%rax),%rbp # restore saved %rbp
3000 lea 8(%rax),%rax # adjust stack pointer
3001 mov %rbp,160($context) # restore context->Rbp
3002 jmp .Lcommon_seh_tail
3005 mov 120($context),%rax
3010 mov %rax,152($context) # restore context->Rsp
3011 mov %rsi,168($context) # restore context->Rsi
3012 mov %rdi,176($context) # restore context->Rdi
3014 mov 40($disp),%rdi # disp->ContextRecord
3015 mov $context,%rsi # context
3016 mov \$154,%ecx # sizeof(CONTEXT)
3017 .long 0xa548f3fc # cld; rep movsq
3020 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
3021 mov 8(%rsi),%rdx # arg2, disp->ImageBase
3022 mov 0(%rsi),%r8 # arg3, disp->ControlPc
3023 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
3024 mov 40(%rsi),%r10 # disp->ContextRecord
3025 lea 56(%rsi),%r11 # &disp->HandlerData
3026 lea 24(%rsi),%r12 # &disp->EstablisherFrame
3027 mov %r10,32(%rsp) # arg5
3028 mov %r11,40(%rsp) # arg6
3029 mov %r12,48(%rsp) # arg7
3030 mov %rcx,56(%rsp) # arg8, (NULL)
3031 call *__imp_RtlVirtualUnwind(%rip)
3033 mov \$1,%eax # ExceptionContinueSearch
3045 .size cbc_se_handler,.-cbc_se_handler
3050 $code.=<<___ if ($PREFIX eq "aesni");
3051 .rva .LSEH_begin_aesni_ecb_encrypt
3052 .rva .LSEH_end_aesni_ecb_encrypt
3055 .rva .LSEH_begin_aesni_ccm64_encrypt_blocks
3056 .rva .LSEH_end_aesni_ccm64_encrypt_blocks
3057 .rva .LSEH_info_ccm64_enc
3059 .rva .LSEH_begin_aesni_ccm64_decrypt_blocks
3060 .rva .LSEH_end_aesni_ccm64_decrypt_blocks
3061 .rva .LSEH_info_ccm64_dec
3063 .rva .LSEH_begin_aesni_ctr32_encrypt_blocks
3064 .rva .LSEH_end_aesni_ctr32_encrypt_blocks
3065 .rva .LSEH_info_ctr32
3067 .rva .LSEH_begin_aesni_xts_encrypt
3068 .rva .LSEH_end_aesni_xts_encrypt
3069 .rva .LSEH_info_xts_enc
3071 .rva .LSEH_begin_aesni_xts_decrypt
3072 .rva .LSEH_end_aesni_xts_decrypt
3073 .rva .LSEH_info_xts_dec
3076 .rva .LSEH_begin_${PREFIX}_cbc_encrypt
3077 .rva .LSEH_end_${PREFIX}_cbc_encrypt
3080 .rva ${PREFIX}_set_decrypt_key
3081 .rva .LSEH_end_set_decrypt_key
3084 .rva ${PREFIX}_set_encrypt_key
3085 .rva .LSEH_end_set_encrypt_key
3090 $code.=<<___ if ($PREFIX eq "aesni");
3094 .LSEH_info_ccm64_enc:
3096 .rva ccm64_se_handler
3097 .rva .Lccm64_enc_body,.Lccm64_enc_ret # HandlerData[]
3098 .LSEH_info_ccm64_dec:
3100 .rva ccm64_se_handler
3101 .rva .Lccm64_dec_body,.Lccm64_dec_ret # HandlerData[]
3104 .rva ctr_xts_se_handler
3105 .rva .Lctr32_body,.Lctr32_epilogue # HandlerData[]
3108 .rva ctr_xts_se_handler
3109 .rva .Lxts_enc_body,.Lxts_enc_epilogue # HandlerData[]
3112 .rva ctr_xts_se_handler
3113 .rva .Lxts_dec_body,.Lxts_dec_epilogue # HandlerData[]
3120 .byte 0x01,0x04,0x01,0x00
3121 .byte 0x04,0x02,0x00,0x00 # sub rsp,8
3126 local *opcode=shift;
3130 $rex|=0x04 if($dst>=8);
3131 $rex|=0x01 if($src>=8);
3132 push @opcode,$rex|0x40 if($rex);
3139 if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3140 rex(\@opcode,$4,$3);
3141 push @opcode,0x0f,0x3a,0xdf;
3142 push @opcode,0xc0|($3&7)|(($4&7)<<3); # ModR/M
3144 push @opcode,$c=~/^0/?oct($c):$c;
3145 return ".byte\t".join(',',@opcode);
3147 elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3150 "aesenc" => 0xdc, "aesenclast" => 0xdd,
3151 "aesdec" => 0xde, "aesdeclast" => 0xdf
3153 return undef if (!defined($opcodelet{$1}));
3154 rex(\@opcode,$3,$2);
3155 push @opcode,0x0f,0x38,$opcodelet{$1};
3156 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
3157 return ".byte\t".join(',',@opcode);
3162 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
3163 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;