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, 0.76 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 engine/eng_aesni.c for details)
1017 my ($in0,$in1,$in2,$in3,$one,$ivec)=map("%xmm$_",(10..15));
1021 .globl aesni_ctr32_encrypt_blocks
1022 .type aesni_ctr32_encrypt_blocks,\@function,5
1024 aesni_ctr32_encrypt_blocks:
1026 $code.=<<___ if ($win64);
1027 lea -0xa8(%rsp),%rsp
1028 movaps %xmm6,0x00(%rsp)
1029 movaps %xmm7,0x10(%rsp)
1030 movaps %xmm8,0x20(%rsp)
1031 movaps %xmm9,0x30(%rsp)
1032 movaps %xmm10,0x40(%rsp)
1033 movaps %xmm11,0x50(%rsp)
1034 movaps %xmm12,0x60(%rsp)
1035 movaps %xmm13,0x70(%rsp)
1036 movaps %xmm14,0x80(%rsp)
1037 movaps %xmm15,0x90(%rsp)
1042 je .Lctr32_one_shortcut
1044 movzb 15($ivp),%rax # counter LSB
1045 mov $len,$len_ # backup $len
1046 mov 240($key),$rnds_ # key->rounds
1047 mov $key,$key_ # backup $key
1050 movdqa .Lincrement1(%rip),$one
1051 add \$256,%rax # steps to closest overflow
1056 mov $rnds_,$rounds # restore $rounds
1062 $movkey ($key_),$rndkey0
1070 $movkey 16($key_),$rndkey1
1071 movdqa $rndkey0,$inout0
1072 movdqa $rndkey0,$inout1
1075 movdqa $rndkey0,$inout2
1076 aesenc $rndkey1,$inout0
1080 movdqa $rndkey0,$inout3
1081 aesenc $rndkey1,$inout1
1084 movdqa $rndkey0,$inout4
1085 aesenc $rndkey1,$inout2
1088 movdqa $rndkey0,$inout5
1089 aesenc $rndkey1,$inout3
1092 movdqa $rndkey0,$inout6
1093 aesenc $rndkey1,$inout4
1096 movdqa $rndkey0,$inout7
1097 aesenc $rndkey1,$inout5
1100 $movkey ($key),$rndkey0
1101 aesenc $rndkey1,$inout6
1105 aesenc $rndkey1,$inout7
1106 $movkey 16($key),$rndkey1
1107 movups ($inp),$in0 # load input
1108 movups 0x10($inp),$in1
1109 movups 0x20($inp),$in2
1110 movups 0x30($inp),$in3
1112 call .Lenc_loop8_enter
1114 xorps $in0,$inout0 # xor
1115 movups 0x40($inp),$in0
1117 movups 0x50($inp),$in1
1119 movups 0x60($inp),$in2
1121 movups 0x70($inp),$in3
1124 movups $inout0,($out) # store output
1126 movups $inout1,0x10($out)
1128 movups $inout2,0x20($out)
1130 movups $inout3,0x30($out)
1131 movups $inout4,0x40($out)
1132 movups $inout5,0x50($out)
1133 movups $inout6,0x60($out)
1134 movups $inout7,0x70($out)
1137 $movkey ($key_),$rndkey0
1142 lea 1($rounds,$rounds),$rounds # restore original value
1143 lea 1($rnds_,$rnds_),$rnds_ # restore original value
1148 mov $key_,$key # restore $key
1149 movdqa $ivec,$inout0
1155 movdqa $ivec,$inout1
1157 movups 0x10($inp),$in1
1160 movdqa $ivec,$inout2
1162 movups 0x20($inp),$in2
1166 movdqa $ivec,$inout3
1168 movups 0x30($inp),$in3
1171 movdqa $ivec,$inout4
1176 movdqa $ivec,$inout5
1180 movdqa $ivec,$inout6
1182 xorps $inout7,$inout7
1184 call _aesni_encrypt8
1186 xorps $in0,$inout0 # xor
1187 movups 0x40($inp),$in0
1189 movups 0x50($inp),$in1
1191 movups 0x60($inp),$in2
1194 movups $inout0,($out) # store output
1196 movups $inout1,0x10($out)
1198 movups $inout2,0x20($out)
1200 movups $inout3,0x30($out)
1201 movups $inout4,0x40($out)
1202 movups $inout5,0x50($out)
1203 movups $inout6,0x60($out)
1208 .Lctr32_one_shortcut:
1209 movups ($ivp),$inout0
1212 mov 240($key),$rounds # key->rounds
1215 &aesni_generate1("enc",$key,$rounds);
1219 movups $inout0,($out)
1225 xorps $inout2,$inout2
1226 call _aesni_encrypt3
1227 xorps $in0,$inout0 # xor
1230 movups $inout0,($out) # store output
1231 movups $inout1,0x10($out)
1237 call _aesni_encrypt3
1238 xorps $in0,$inout0 # xor
1241 movups $inout0,($out) # store output
1243 movups $inout1,0x10($out)
1244 movups $inout2,0x20($out)
1250 call _aesni_encrypt4
1251 xorps $in0,$inout0 # xor
1254 movups $inout0,($out) # store output
1256 movups $inout1,0x10($out)
1258 movups $inout2,0x20($out)
1259 movups $inout3,0x30($out)
1265 xorps $inout5,$inout5
1266 call _aesni_encrypt6
1267 xorps $in0,$inout0 # xor
1268 movups 0x40($inp),$in0
1271 movups $inout0,($out) # store output
1273 movups $inout1,0x10($out)
1275 movups $inout2,0x20($out)
1277 movups $inout3,0x30($out)
1278 movups $inout4,0x40($out)
1284 call _aesni_encrypt6
1285 xorps $in0,$inout0 # xor
1286 movups 0x40($inp),$in0
1288 movups 0x50($inp),$in1
1291 movups $inout0,($out) # store output
1293 movups $inout1,0x10($out)
1295 movups $inout2,0x20($out)
1297 movups $inout3,0x30($out)
1298 movups $inout4,0x40($out)
1299 movups $inout5,0x50($out)
1304 jz .Lctr32_really_done
1306 movdqa .Lbswap_mask(%rip),$rndkey1
1307 pshufb $rndkey1,$ivec
1308 psrldq \$14,$one # 256
1311 pshufb $rndkey1,$ivec
1314 jmp .Lctr32_grandloop
1316 .Lctr32_really_done:
1318 $code.=<<___ if ($win64);
1319 movaps 0x00(%rsp),%xmm6
1320 movaps 0x10(%rsp),%xmm7
1321 movaps 0x20(%rsp),%xmm8
1322 movaps 0x30(%rsp),%xmm9
1323 movaps 0x40(%rsp),%xmm10
1324 movaps 0x50(%rsp),%xmm11
1325 movaps 0x60(%rsp),%xmm12
1326 movaps 0x70(%rsp),%xmm13
1327 movaps 0x80(%rsp),%xmm14
1328 movaps 0x90(%rsp),%xmm15
1334 .size aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks
1338 ######################################################################
1339 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len,
1340 # const AES_KEY *key1, const AES_KEY *key2
1341 # const unsigned char iv[16]);
1344 my @tweak=map("%xmm$_",(10..15));
1345 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]);
1346 my ($key2,$ivp,$len_)=("%r8","%r9","%r9");
1347 my $frame_size = 0x60 + ($win64?160:0);
1350 .globl aesni_xts_encrypt
1351 .type aesni_xts_encrypt,\@function,6
1356 sub \$$frame_size,%rsp
1357 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
1359 $code.=<<___ if ($win64);
1360 movaps %xmm6,0x60(%rsp)
1361 movaps %xmm7,0x70(%rsp)
1362 movaps %xmm8,0x80(%rsp)
1363 movaps %xmm9,0x90(%rsp)
1364 movaps %xmm10,0xa0(%rsp)
1365 movaps %xmm11,0xb0(%rsp)
1366 movaps %xmm12,0xc0(%rsp)
1367 movaps %xmm13,0xd0(%rsp)
1368 movaps %xmm14,0xe0(%rsp)
1369 movaps %xmm15,0xf0(%rsp)
1374 movups ($ivp),@tweak[5] # load clear-text tweak
1375 mov 240(%r8),$rounds # key2->rounds
1376 mov 240($key),$rnds_ # key1->rounds
1378 # generate the tweak
1379 &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1381 mov $key,$key_ # backup $key
1382 mov $rnds_,$rounds # backup $rounds
1383 mov $len,$len_ # backup $len
1386 movdqa .Lxts_magic(%rip),$twmask
1388 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1390 for ($i=0;$i<4;$i++) {
1392 pshufd \$0x13,$twtmp,$twres
1394 movdqa @tweak[5],@tweak[$i]
1395 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1396 pand $twmask,$twres # isolate carry and residue
1397 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1398 pxor $twres,@tweak[5]
1408 jmp .Lxts_enc_grandloop
1411 .Lxts_enc_grandloop:
1412 pshufd \$0x13,$twtmp,$twres
1413 movdqa @tweak[5],@tweak[4]
1414 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1415 movdqu `16*0`($inp),$inout0 # load input
1416 pand $twmask,$twres # isolate carry and residue
1417 movdqu `16*1`($inp),$inout1
1418 pxor $twres,@tweak[5]
1420 movdqu `16*2`($inp),$inout2
1421 pxor @tweak[0],$inout0 # input^=tweak
1422 movdqu `16*3`($inp),$inout3
1423 pxor @tweak[1],$inout1
1424 movdqu `16*4`($inp),$inout4
1425 pxor @tweak[2],$inout2
1426 movdqu `16*5`($inp),$inout5
1427 lea `16*6`($inp),$inp
1428 pxor @tweak[3],$inout3
1429 $movkey ($key_),$rndkey0
1430 pxor @tweak[4],$inout4
1431 pxor @tweak[5],$inout5
1433 # inline _aesni_encrypt6 and interleave first and last rounds
1435 $movkey 16($key_),$rndkey1
1436 pxor $rndkey0,$inout0
1437 pxor $rndkey0,$inout1
1438 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks
1439 aesenc $rndkey1,$inout0
1441 pxor $rndkey0,$inout2
1442 movdqa @tweak[1],`16*1`(%rsp)
1443 aesenc $rndkey1,$inout1
1444 pxor $rndkey0,$inout3
1445 movdqa @tweak[2],`16*2`(%rsp)
1446 aesenc $rndkey1,$inout2
1447 pxor $rndkey0,$inout4
1448 movdqa @tweak[3],`16*3`(%rsp)
1449 aesenc $rndkey1,$inout3
1450 pxor $rndkey0,$inout5
1451 $movkey ($key),$rndkey0
1453 movdqa @tweak[4],`16*4`(%rsp)
1454 aesenc $rndkey1,$inout4
1455 movdqa @tweak[5],`16*5`(%rsp)
1456 aesenc $rndkey1,$inout5
1458 pcmpgtd @tweak[5],$twtmp
1459 jmp .Lxts_enc_loop6_enter
1463 aesenc $rndkey1,$inout0
1464 aesenc $rndkey1,$inout1
1466 aesenc $rndkey1,$inout2
1467 aesenc $rndkey1,$inout3
1468 aesenc $rndkey1,$inout4
1469 aesenc $rndkey1,$inout5
1470 .Lxts_enc_loop6_enter:
1471 $movkey 16($key),$rndkey1
1472 aesenc $rndkey0,$inout0
1473 aesenc $rndkey0,$inout1
1475 aesenc $rndkey0,$inout2
1476 aesenc $rndkey0,$inout3
1477 aesenc $rndkey0,$inout4
1478 aesenc $rndkey0,$inout5
1479 $movkey ($key),$rndkey0
1482 pshufd \$0x13,$twtmp,$twres
1484 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1485 aesenc $rndkey1,$inout0
1486 pand $twmask,$twres # isolate carry and residue
1487 aesenc $rndkey1,$inout1
1488 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1489 aesenc $rndkey1,$inout2
1490 pxor $twres,@tweak[5]
1491 aesenc $rndkey1,$inout3
1492 aesenc $rndkey1,$inout4
1493 aesenc $rndkey1,$inout5
1494 $movkey 16($key),$rndkey1
1496 pshufd \$0x13,$twtmp,$twres
1498 movdqa @tweak[5],@tweak[0]
1499 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1500 aesenc $rndkey0,$inout0
1501 pand $twmask,$twres # isolate carry and residue
1502 aesenc $rndkey0,$inout1
1503 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1504 aesenc $rndkey0,$inout2
1505 pxor $twres,@tweak[5]
1506 aesenc $rndkey0,$inout3
1507 aesenc $rndkey0,$inout4
1508 aesenc $rndkey0,$inout5
1509 $movkey 32($key),$rndkey0
1511 pshufd \$0x13,$twtmp,$twres
1513 movdqa @tweak[5],@tweak[1]
1514 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1515 aesenc $rndkey1,$inout0
1516 pand $twmask,$twres # isolate carry and residue
1517 aesenc $rndkey1,$inout1
1518 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1519 aesenc $rndkey1,$inout2
1520 pxor $twres,@tweak[5]
1521 aesenc $rndkey1,$inout3
1522 aesenc $rndkey1,$inout4
1523 aesenc $rndkey1,$inout5
1525 pshufd \$0x13,$twtmp,$twres
1527 movdqa @tweak[5],@tweak[2]
1528 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1529 aesenclast $rndkey0,$inout0
1530 pand $twmask,$twres # isolate carry and residue
1531 aesenclast $rndkey0,$inout1
1532 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1533 aesenclast $rndkey0,$inout2
1534 pxor $twres,@tweak[5]
1535 aesenclast $rndkey0,$inout3
1536 aesenclast $rndkey0,$inout4
1537 aesenclast $rndkey0,$inout5
1539 pshufd \$0x13,$twtmp,$twres
1541 movdqa @tweak[5],@tweak[3]
1542 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1543 xorps `16*0`(%rsp),$inout0 # output^=tweak
1544 pand $twmask,$twres # isolate carry and residue
1545 xorps `16*1`(%rsp),$inout1
1546 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1547 pxor $twres,@tweak[5]
1549 xorps `16*2`(%rsp),$inout2
1550 movups $inout0,`16*0`($out) # write output
1551 xorps `16*3`(%rsp),$inout3
1552 movups $inout1,`16*1`($out)
1553 xorps `16*4`(%rsp),$inout4
1554 movups $inout2,`16*2`($out)
1555 xorps `16*5`(%rsp),$inout5
1556 movups $inout3,`16*3`($out)
1557 mov $rnds_,$rounds # restore $rounds
1558 movups $inout4,`16*4`($out)
1559 movups $inout5,`16*5`($out)
1560 lea `16*6`($out),$out
1562 jnc .Lxts_enc_grandloop
1564 lea 3($rounds,$rounds),$rounds # restore original value
1565 mov $key_,$key # restore $key
1566 mov $rounds,$rnds_ # backup $rounds
1580 pshufd \$0x13,$twtmp,$twres
1581 movdqa @tweak[5],@tweak[4]
1582 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1583 movdqu ($inp),$inout0
1584 pand $twmask,$twres # isolate carry and residue
1585 movdqu 16*1($inp),$inout1
1586 pxor $twres,@tweak[5]
1588 movdqu 16*2($inp),$inout2
1589 pxor @tweak[0],$inout0
1590 movdqu 16*3($inp),$inout3
1591 pxor @tweak[1],$inout1
1592 movdqu 16*4($inp),$inout4
1594 pxor @tweak[2],$inout2
1595 pxor @tweak[3],$inout3
1596 pxor @tweak[4],$inout4
1598 call _aesni_encrypt6
1600 xorps @tweak[0],$inout0
1601 movdqa @tweak[5],@tweak[0]
1602 xorps @tweak[1],$inout1
1603 xorps @tweak[2],$inout2
1604 movdqu $inout0,($out)
1605 xorps @tweak[3],$inout3
1606 movdqu $inout1,16*1($out)
1607 xorps @tweak[4],$inout4
1608 movdqu $inout2,16*2($out)
1609 movdqu $inout3,16*3($out)
1610 movdqu $inout4,16*4($out)
1616 movups ($inp),$inout0
1618 xorps @tweak[0],$inout0
1620 &aesni_generate1("enc",$key,$rounds);
1622 xorps @tweak[0],$inout0
1623 movdqa @tweak[1],@tweak[0]
1624 movups $inout0,($out)
1630 movups ($inp),$inout0
1631 movups 16($inp),$inout1
1633 xorps @tweak[0],$inout0
1634 xorps @tweak[1],$inout1
1636 call _aesni_encrypt3
1638 xorps @tweak[0],$inout0
1639 movdqa @tweak[2],@tweak[0]
1640 xorps @tweak[1],$inout1
1641 movups $inout0,($out)
1642 movups $inout1,16*1($out)
1648 movups ($inp),$inout0
1649 movups 16*1($inp),$inout1
1650 movups 16*2($inp),$inout2
1652 xorps @tweak[0],$inout0
1653 xorps @tweak[1],$inout1
1654 xorps @tweak[2],$inout2
1656 call _aesni_encrypt3
1658 xorps @tweak[0],$inout0
1659 movdqa @tweak[3],@tweak[0]
1660 xorps @tweak[1],$inout1
1661 xorps @tweak[2],$inout2
1662 movups $inout0,($out)
1663 movups $inout1,16*1($out)
1664 movups $inout2,16*2($out)
1670 movups ($inp),$inout0
1671 movups 16*1($inp),$inout1
1672 movups 16*2($inp),$inout2
1673 xorps @tweak[0],$inout0
1674 movups 16*3($inp),$inout3
1676 xorps @tweak[1],$inout1
1677 xorps @tweak[2],$inout2
1678 xorps @tweak[3],$inout3
1680 call _aesni_encrypt4
1682 xorps @tweak[0],$inout0
1683 movdqa @tweak[5],@tweak[0]
1684 xorps @tweak[1],$inout1
1685 xorps @tweak[2],$inout2
1686 movups $inout0,($out)
1687 xorps @tweak[3],$inout3
1688 movups $inout1,16*1($out)
1689 movups $inout2,16*2($out)
1690 movups $inout3,16*3($out)
1701 movzb ($inp),%eax # borrow $rounds ...
1702 movzb -16($out),%ecx # ... and $key
1710 sub $len_,$out # rewind $out
1711 mov $key_,$key # restore $key
1712 mov $rnds_,$rounds # restore $rounds
1714 movups -16($out),$inout0
1715 xorps @tweak[0],$inout0
1717 &aesni_generate1("enc",$key,$rounds);
1719 xorps @tweak[0],$inout0
1720 movups $inout0,-16($out)
1724 $code.=<<___ if ($win64);
1725 movaps 0x60(%rsp),%xmm6
1726 movaps 0x70(%rsp),%xmm7
1727 movaps 0x80(%rsp),%xmm8
1728 movaps 0x90(%rsp),%xmm9
1729 movaps 0xa0(%rsp),%xmm10
1730 movaps 0xb0(%rsp),%xmm11
1731 movaps 0xc0(%rsp),%xmm12
1732 movaps 0xd0(%rsp),%xmm13
1733 movaps 0xe0(%rsp),%xmm14
1734 movaps 0xf0(%rsp),%xmm15
1741 .size aesni_xts_encrypt,.-aesni_xts_encrypt
1745 .globl aesni_xts_decrypt
1746 .type aesni_xts_decrypt,\@function,6
1751 sub \$$frame_size,%rsp
1752 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
1754 $code.=<<___ if ($win64);
1755 movaps %xmm6,0x60(%rsp)
1756 movaps %xmm7,0x70(%rsp)
1757 movaps %xmm8,0x80(%rsp)
1758 movaps %xmm9,0x90(%rsp)
1759 movaps %xmm10,0xa0(%rsp)
1760 movaps %xmm11,0xb0(%rsp)
1761 movaps %xmm12,0xc0(%rsp)
1762 movaps %xmm13,0xd0(%rsp)
1763 movaps %xmm14,0xe0(%rsp)
1764 movaps %xmm15,0xf0(%rsp)
1769 movups ($ivp),@tweak[5] # load clear-text tweak
1770 mov 240($key2),$rounds # key2->rounds
1771 mov 240($key),$rnds_ # key1->rounds
1773 # generate the tweak
1774 &aesni_generate1("enc",$key2,$rounds,@tweak[5]);
1776 xor %eax,%eax # if ($len%16) len-=16;
1782 mov $key,$key_ # backup $key
1783 mov $rnds_,$rounds # backup $rounds
1784 mov $len,$len_ # backup $len
1787 movdqa .Lxts_magic(%rip),$twmask
1789 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1791 for ($i=0;$i<4;$i++) {
1793 pshufd \$0x13,$twtmp,$twres
1795 movdqa @tweak[5],@tweak[$i]
1796 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1797 pand $twmask,$twres # isolate carry and residue
1798 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1799 pxor $twres,@tweak[5]
1809 jmp .Lxts_dec_grandloop
1812 .Lxts_dec_grandloop:
1813 pshufd \$0x13,$twtmp,$twres
1814 movdqa @tweak[5],@tweak[4]
1815 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1816 movdqu `16*0`($inp),$inout0 # load input
1817 pand $twmask,$twres # isolate carry and residue
1818 movdqu `16*1`($inp),$inout1
1819 pxor $twres,@tweak[5]
1821 movdqu `16*2`($inp),$inout2
1822 pxor @tweak[0],$inout0 # input^=tweak
1823 movdqu `16*3`($inp),$inout3
1824 pxor @tweak[1],$inout1
1825 movdqu `16*4`($inp),$inout4
1826 pxor @tweak[2],$inout2
1827 movdqu `16*5`($inp),$inout5
1828 lea `16*6`($inp),$inp
1829 pxor @tweak[3],$inout3
1830 $movkey ($key_),$rndkey0
1831 pxor @tweak[4],$inout4
1832 pxor @tweak[5],$inout5
1834 # inline _aesni_decrypt6 and interleave first and last rounds
1836 $movkey 16($key_),$rndkey1
1837 pxor $rndkey0,$inout0
1838 pxor $rndkey0,$inout1
1839 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks
1840 aesdec $rndkey1,$inout0
1842 pxor $rndkey0,$inout2
1843 movdqa @tweak[1],`16*1`(%rsp)
1844 aesdec $rndkey1,$inout1
1845 pxor $rndkey0,$inout3
1846 movdqa @tweak[2],`16*2`(%rsp)
1847 aesdec $rndkey1,$inout2
1848 pxor $rndkey0,$inout4
1849 movdqa @tweak[3],`16*3`(%rsp)
1850 aesdec $rndkey1,$inout3
1851 pxor $rndkey0,$inout5
1852 $movkey ($key),$rndkey0
1854 movdqa @tweak[4],`16*4`(%rsp)
1855 aesdec $rndkey1,$inout4
1856 movdqa @tweak[5],`16*5`(%rsp)
1857 aesdec $rndkey1,$inout5
1859 pcmpgtd @tweak[5],$twtmp
1860 jmp .Lxts_dec_loop6_enter
1864 aesdec $rndkey1,$inout0
1865 aesdec $rndkey1,$inout1
1867 aesdec $rndkey1,$inout2
1868 aesdec $rndkey1,$inout3
1869 aesdec $rndkey1,$inout4
1870 aesdec $rndkey1,$inout5
1871 .Lxts_dec_loop6_enter:
1872 $movkey 16($key),$rndkey1
1873 aesdec $rndkey0,$inout0
1874 aesdec $rndkey0,$inout1
1876 aesdec $rndkey0,$inout2
1877 aesdec $rndkey0,$inout3
1878 aesdec $rndkey0,$inout4
1879 aesdec $rndkey0,$inout5
1880 $movkey ($key),$rndkey0
1883 pshufd \$0x13,$twtmp,$twres
1885 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1886 aesdec $rndkey1,$inout0
1887 pand $twmask,$twres # isolate carry and residue
1888 aesdec $rndkey1,$inout1
1889 pcmpgtd @tweak[5],$twtmp # broadcast upper bits
1890 aesdec $rndkey1,$inout2
1891 pxor $twres,@tweak[5]
1892 aesdec $rndkey1,$inout3
1893 aesdec $rndkey1,$inout4
1894 aesdec $rndkey1,$inout5
1895 $movkey 16($key),$rndkey1
1897 pshufd \$0x13,$twtmp,$twres
1899 movdqa @tweak[5],@tweak[0]
1900 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1901 aesdec $rndkey0,$inout0
1902 pand $twmask,$twres # isolate carry and residue
1903 aesdec $rndkey0,$inout1
1904 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1905 aesdec $rndkey0,$inout2
1906 pxor $twres,@tweak[5]
1907 aesdec $rndkey0,$inout3
1908 aesdec $rndkey0,$inout4
1909 aesdec $rndkey0,$inout5
1910 $movkey 32($key),$rndkey0
1912 pshufd \$0x13,$twtmp,$twres
1914 movdqa @tweak[5],@tweak[1]
1915 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1916 aesdec $rndkey1,$inout0
1917 pand $twmask,$twres # isolate carry and residue
1918 aesdec $rndkey1,$inout1
1919 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1920 aesdec $rndkey1,$inout2
1921 pxor $twres,@tweak[5]
1922 aesdec $rndkey1,$inout3
1923 aesdec $rndkey1,$inout4
1924 aesdec $rndkey1,$inout5
1926 pshufd \$0x13,$twtmp,$twres
1928 movdqa @tweak[5],@tweak[2]
1929 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1930 aesdeclast $rndkey0,$inout0
1931 pand $twmask,$twres # isolate carry and residue
1932 aesdeclast $rndkey0,$inout1
1933 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1934 aesdeclast $rndkey0,$inout2
1935 pxor $twres,@tweak[5]
1936 aesdeclast $rndkey0,$inout3
1937 aesdeclast $rndkey0,$inout4
1938 aesdeclast $rndkey0,$inout5
1940 pshufd \$0x13,$twtmp,$twres
1942 movdqa @tweak[5],@tweak[3]
1943 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1944 xorps `16*0`(%rsp),$inout0 # output^=tweak
1945 pand $twmask,$twres # isolate carry and residue
1946 xorps `16*1`(%rsp),$inout1
1947 pcmpgtd @tweak[5],$twtmp # broadcat upper bits
1948 pxor $twres,@tweak[5]
1950 xorps `16*2`(%rsp),$inout2
1951 movups $inout0,`16*0`($out) # write output
1952 xorps `16*3`(%rsp),$inout3
1953 movups $inout1,`16*1`($out)
1954 xorps `16*4`(%rsp),$inout4
1955 movups $inout2,`16*2`($out)
1956 xorps `16*5`(%rsp),$inout5
1957 movups $inout3,`16*3`($out)
1958 mov $rnds_,$rounds # restore $rounds
1959 movups $inout4,`16*4`($out)
1960 movups $inout5,`16*5`($out)
1961 lea `16*6`($out),$out
1963 jnc .Lxts_dec_grandloop
1965 lea 3($rounds,$rounds),$rounds # restore original value
1966 mov $key_,$key # restore $key
1967 mov $rounds,$rnds_ # backup $rounds
1981 pshufd \$0x13,$twtmp,$twres
1982 movdqa @tweak[5],@tweak[4]
1983 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
1984 movdqu ($inp),$inout0
1985 pand $twmask,$twres # isolate carry and residue
1986 movdqu 16*1($inp),$inout1
1987 pxor $twres,@tweak[5]
1989 movdqu 16*2($inp),$inout2
1990 pxor @tweak[0],$inout0
1991 movdqu 16*3($inp),$inout3
1992 pxor @tweak[1],$inout1
1993 movdqu 16*4($inp),$inout4
1995 pxor @tweak[2],$inout2
1996 pxor @tweak[3],$inout3
1997 pxor @tweak[4],$inout4
1999 call _aesni_decrypt6
2001 xorps @tweak[0],$inout0
2002 xorps @tweak[1],$inout1
2003 xorps @tweak[2],$inout2
2004 movdqu $inout0,($out)
2005 xorps @tweak[3],$inout3
2006 movdqu $inout1,16*1($out)
2007 xorps @tweak[4],$inout4
2008 movdqu $inout2,16*2($out)
2010 movdqu $inout3,16*3($out)
2011 pcmpgtd @tweak[5],$twtmp
2012 movdqu $inout4,16*4($out)
2014 pshufd \$0x13,$twtmp,@tweak[1] # $twres
2018 movdqa @tweak[5],@tweak[0]
2019 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2020 pand $twmask,@tweak[1] # isolate carry and residue
2021 pxor @tweak[5],@tweak[1]
2026 movups ($inp),$inout0
2028 xorps @tweak[0],$inout0
2030 &aesni_generate1("dec",$key,$rounds);
2032 xorps @tweak[0],$inout0
2033 movdqa @tweak[1],@tweak[0]
2034 movups $inout0,($out)
2035 movdqa @tweak[2],@tweak[1]
2041 movups ($inp),$inout0
2042 movups 16($inp),$inout1
2044 xorps @tweak[0],$inout0
2045 xorps @tweak[1],$inout1
2047 call _aesni_decrypt3
2049 xorps @tweak[0],$inout0
2050 movdqa @tweak[2],@tweak[0]
2051 xorps @tweak[1],$inout1
2052 movdqa @tweak[3],@tweak[1]
2053 movups $inout0,($out)
2054 movups $inout1,16*1($out)
2060 movups ($inp),$inout0
2061 movups 16*1($inp),$inout1
2062 movups 16*2($inp),$inout2
2064 xorps @tweak[0],$inout0
2065 xorps @tweak[1],$inout1
2066 xorps @tweak[2],$inout2
2068 call _aesni_decrypt3
2070 xorps @tweak[0],$inout0
2071 movdqa @tweak[3],@tweak[0]
2072 xorps @tweak[1],$inout1
2073 movdqa @tweak[5],@tweak[1]
2074 xorps @tweak[2],$inout2
2075 movups $inout0,($out)
2076 movups $inout1,16*1($out)
2077 movups $inout2,16*2($out)
2083 pshufd \$0x13,$twtmp,$twres
2084 movdqa @tweak[5],@tweak[4]
2085 paddq @tweak[5],@tweak[5] # psllq 1,$tweak
2086 movups ($inp),$inout0
2087 pand $twmask,$twres # isolate carry and residue
2088 movups 16*1($inp),$inout1
2089 pxor $twres,@tweak[5]
2091 movups 16*2($inp),$inout2
2092 xorps @tweak[0],$inout0
2093 movups 16*3($inp),$inout3
2095 xorps @tweak[1],$inout1
2096 xorps @tweak[2],$inout2
2097 xorps @tweak[3],$inout3
2099 call _aesni_decrypt4
2101 xorps @tweak[0],$inout0
2102 movdqa @tweak[4],@tweak[0]
2103 xorps @tweak[1],$inout1
2104 movdqa @tweak[5],@tweak[1]
2105 xorps @tweak[2],$inout2
2106 movups $inout0,($out)
2107 xorps @tweak[3],$inout3
2108 movups $inout1,16*1($out)
2109 movups $inout2,16*2($out)
2110 movups $inout3,16*3($out)
2120 mov $key_,$key # restore $key
2121 mov $rnds_,$rounds # restore $rounds
2123 movups ($inp),$inout0
2124 xorps @tweak[1],$inout0
2126 &aesni_generate1("dec",$key,$rounds);
2128 xorps @tweak[1],$inout0
2129 movups $inout0,($out)
2132 movzb 16($inp),%eax # borrow $rounds ...
2133 movzb ($out),%ecx # ... and $key
2141 sub $len_,$out # rewind $out
2142 mov $key_,$key # restore $key
2143 mov $rnds_,$rounds # restore $rounds
2145 movups ($out),$inout0
2146 xorps @tweak[0],$inout0
2148 &aesni_generate1("dec",$key,$rounds);
2150 xorps @tweak[0],$inout0
2151 movups $inout0,($out)
2155 $code.=<<___ if ($win64);
2156 movaps 0x60(%rsp),%xmm6
2157 movaps 0x70(%rsp),%xmm7
2158 movaps 0x80(%rsp),%xmm8
2159 movaps 0x90(%rsp),%xmm9
2160 movaps 0xa0(%rsp),%xmm10
2161 movaps 0xb0(%rsp),%xmm11
2162 movaps 0xc0(%rsp),%xmm12
2163 movaps 0xd0(%rsp),%xmm13
2164 movaps 0xe0(%rsp),%xmm14
2165 movaps 0xf0(%rsp),%xmm15
2172 .size aesni_xts_decrypt,.-aesni_xts_decrypt
2176 ########################################################################
2177 # void $PREFIX_cbc_encrypt (const void *inp, void *out,
2178 # size_t length, const AES_KEY *key,
2179 # unsigned char *ivp,const int enc);
2181 my $frame_size = 0x10 + ($win64?0x40:0); # used in decrypt
2183 .globl ${PREFIX}_cbc_encrypt
2184 .type ${PREFIX}_cbc_encrypt,\@function,6
2186 ${PREFIX}_cbc_encrypt:
2187 test $len,$len # check length
2190 mov 240($key),$rnds_ # key->rounds
2191 mov $key,$key_ # backup $key
2192 test %r9d,%r9d # 6th argument
2194 #--------------------------- CBC ENCRYPT ------------------------------#
2195 movups ($ivp),$inout0 # load iv as initial state
2203 movups ($inp),$inout1 # load input
2205 #xorps $inout1,$inout0
2207 &aesni_generate1("enc",$key,$rounds,$inout0,$inout1);
2209 mov $rnds_,$rounds # restore $rounds
2210 mov $key_,$key # restore $key
2211 movups $inout0,0($out) # store output
2217 movups $inout0,($ivp)
2221 mov $len,%rcx # zaps $key
2222 xchg $inp,$out # $inp is %rsi and $out is %rdi now
2223 .long 0x9066A4F3 # rep movsb
2224 mov \$16,%ecx # zero tail
2227 .long 0x9066AAF3 # rep stosb
2228 lea -16(%rdi),%rdi # rewind $out by 1 block
2229 mov $rnds_,$rounds # restore $rounds
2230 mov %rdi,%rsi # $inp and $out are the same
2231 mov $key_,$key # restore $key
2232 xor $len,$len # len=16
2233 jmp .Lcbc_enc_loop # one more spin
2234 \f#--------------------------- CBC DECRYPT ------------------------------#
2239 sub \$$frame_size,%rsp
2240 and \$-16,%rsp # Linux kernel stack can be incorrectly seeded
2242 $code.=<<___ if ($win64);
2243 movaps %xmm6,0x10(%rsp)
2244 movaps %xmm7,0x20(%rsp)
2245 movaps %xmm8,0x30(%rsp)
2246 movaps %xmm9,0x40(%rsp)
2259 jmp .Lcbc_dec_loop8_enter
2262 movaps $rndkey0,(%rsp) # save IV
2263 movups $inout7,($out)
2265 .Lcbc_dec_loop8_enter:
2266 $movkey ($key),$rndkey0
2267 movups ($inp),$inout0 # load input
2268 movups 0x10($inp),$inout1
2269 $movkey 16($key),$rndkey1
2272 movdqu 0x20($inp),$inout2
2273 xorps $rndkey0,$inout0
2274 movdqu 0x30($inp),$inout3
2275 xorps $rndkey0,$inout1
2276 movdqu 0x40($inp),$inout4
2277 aesdec $rndkey1,$inout0
2278 pxor $rndkey0,$inout2
2279 movdqu 0x50($inp),$inout5
2280 aesdec $rndkey1,$inout1
2281 pxor $rndkey0,$inout3
2282 movdqu 0x60($inp),$inout6
2283 aesdec $rndkey1,$inout2
2284 pxor $rndkey0,$inout4
2285 movdqu 0x70($inp),$inout7
2286 aesdec $rndkey1,$inout3
2287 pxor $rndkey0,$inout5
2289 aesdec $rndkey1,$inout4
2290 pxor $rndkey0,$inout6
2291 aesdec $rndkey1,$inout5
2292 pxor $rndkey0,$inout7
2293 $movkey ($key),$rndkey0
2294 aesdec $rndkey1,$inout6
2295 aesdec $rndkey1,$inout7
2296 $movkey 16($key),$rndkey1
2298 call .Ldec_loop8_enter
2300 movups ($inp),$rndkey1 # re-load input
2301 movups 0x10($inp),$rndkey0
2302 xorps (%rsp),$inout0 # ^= IV
2303 xorps $rndkey1,$inout1
2304 movups 0x20($inp),$rndkey1
2305 xorps $rndkey0,$inout2
2306 movups 0x30($inp),$rndkey0
2307 xorps $rndkey1,$inout3
2308 movups 0x40($inp),$rndkey1
2309 xorps $rndkey0,$inout4
2310 movups 0x50($inp),$rndkey0
2311 xorps $rndkey1,$inout5
2312 movups 0x60($inp),$rndkey1
2313 xorps $rndkey0,$inout6
2314 movups 0x70($inp),$rndkey0 # IV
2315 xorps $rndkey1,$inout7
2316 movups $inout0,($out)
2317 movups $inout1,0x10($out)
2318 movups $inout2,0x20($out)
2319 movups $inout3,0x30($out)
2320 mov $rnds_,$rounds # restore $rounds
2321 movups $inout4,0x40($out)
2322 mov $key_,$key # restore $key
2323 movups $inout5,0x50($out)
2325 movups $inout6,0x60($out)
2330 movaps $inout7,$inout0
2333 jle .Lcbc_dec_tail_collected
2334 movups $inout0,($out)
2335 lea 1($rnds_,$rnds_),$rounds
2338 movups ($inp),$inout0
2343 movups 0x10($inp),$inout1
2348 movups 0x20($inp),$inout2
2353 movups 0x30($inp),$inout3
2357 movups 0x40($inp),$inout4
2361 movups 0x50($inp),$inout5
2365 movups 0x60($inp),$inout6
2366 movaps $iv,(%rsp) # save IV
2367 call _aesni_decrypt8
2368 movups ($inp),$rndkey1
2369 movups 0x10($inp),$rndkey0
2370 xorps (%rsp),$inout0 # ^= IV
2371 xorps $rndkey1,$inout1
2372 movups 0x20($inp),$rndkey1
2373 xorps $rndkey0,$inout2
2374 movups 0x30($inp),$rndkey0
2375 xorps $rndkey1,$inout3
2376 movups 0x40($inp),$rndkey1
2377 xorps $rndkey0,$inout4
2378 movups 0x50($inp),$rndkey0
2379 xorps $rndkey1,$inout5
2380 movups 0x60($inp),$iv # IV
2381 xorps $rndkey0,$inout6
2382 movups $inout0,($out)
2383 movups $inout1,0x10($out)
2384 movups $inout2,0x20($out)
2385 movups $inout3,0x30($out)
2386 movups $inout4,0x40($out)
2387 movups $inout5,0x50($out)
2389 movaps $inout6,$inout0
2391 jmp .Lcbc_dec_tail_collected
2395 &aesni_generate1("dec",$key,$rounds);
2400 jmp .Lcbc_dec_tail_collected
2403 xorps $inout2,$inout2
2404 call _aesni_decrypt3
2407 movups $inout0,($out)
2409 movaps $inout1,$inout0
2412 jmp .Lcbc_dec_tail_collected
2415 call _aesni_decrypt3
2418 movups $inout0,($out)
2420 movups $inout1,0x10($out)
2422 movaps $inout2,$inout0
2425 jmp .Lcbc_dec_tail_collected
2428 call _aesni_decrypt4
2430 movups 0x30($inp),$iv
2432 movups $inout0,($out)
2434 movups $inout1,0x10($out)
2436 movups $inout2,0x20($out)
2437 movaps $inout3,$inout0
2440 jmp .Lcbc_dec_tail_collected
2443 xorps $inout5,$inout5
2444 call _aesni_decrypt6
2445 movups 0x10($inp),$rndkey1
2446 movups 0x20($inp),$rndkey0
2449 xorps $rndkey1,$inout2
2450 movups 0x30($inp),$rndkey1
2451 xorps $rndkey0,$inout3
2452 movups 0x40($inp),$iv
2453 xorps $rndkey1,$inout4
2454 movups $inout0,($out)
2455 movups $inout1,0x10($out)
2456 movups $inout2,0x20($out)
2457 movups $inout3,0x30($out)
2459 movaps $inout4,$inout0
2461 jmp .Lcbc_dec_tail_collected
2464 call _aesni_decrypt6
2465 movups 0x10($inp),$rndkey1
2466 movups 0x20($inp),$rndkey0
2469 xorps $rndkey1,$inout2
2470 movups 0x30($inp),$rndkey1
2471 xorps $rndkey0,$inout3
2472 movups 0x40($inp),$rndkey0
2473 xorps $rndkey1,$inout4
2474 movups 0x50($inp),$iv
2475 xorps $rndkey0,$inout5
2476 movups $inout0,($out)
2477 movups $inout1,0x10($out)
2478 movups $inout2,0x20($out)
2479 movups $inout3,0x30($out)
2480 movups $inout4,0x40($out)
2482 movaps $inout5,$inout0
2484 jmp .Lcbc_dec_tail_collected
2486 .Lcbc_dec_tail_collected:
2489 jnz .Lcbc_dec_tail_partial
2490 movups $inout0,($out)
2493 .Lcbc_dec_tail_partial:
2494 movaps $inout0,(%rsp)
2499 .long 0x9066A4F3 # rep movsb
2503 $code.=<<___ if ($win64);
2504 movaps 0x10(%rsp),%xmm6
2505 movaps 0x20(%rsp),%xmm7
2506 movaps 0x30(%rsp),%xmm8
2507 movaps 0x40(%rsp),%xmm9
2514 .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt
2517 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey,
2518 # int bits, AES_KEY *key)
2519 { my ($inp,$bits,$key) = @_4args;
2523 .globl ${PREFIX}_set_decrypt_key
2524 .type ${PREFIX}_set_decrypt_key,\@abi-omnipotent
2526 ${PREFIX}_set_decrypt_key:
2527 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8
2528 call __aesni_set_encrypt_key
2529 shl \$4,$bits # rounds-1 after _aesni_set_encrypt_key
2532 lea 16($key,$bits),$inp # points at the end of key schedule
2534 $movkey ($key),%xmm0 # just swap
2535 $movkey ($inp),%xmm1
2536 $movkey %xmm0,($inp)
2537 $movkey %xmm1,($key)
2542 $movkey ($key),%xmm0 # swap and inverse
2543 $movkey ($inp),%xmm1
2548 $movkey %xmm0,16($inp)
2549 $movkey %xmm1,-16($key)
2551 ja .Ldec_key_inverse
2553 $movkey ($key),%xmm0 # inverse middle
2555 $movkey %xmm0,($inp)
2559 .LSEH_end_set_decrypt_key:
2560 .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key
2563 # This is based on submission by
2565 # Huang Ying <ying.huang@intel.com>
2566 # Vinodh Gopal <vinodh.gopal@intel.com>
2569 # Agressively optimized in respect to aeskeygenassist's critical path
2570 # and is contained in %xmm0-5 to meet Win64 ABI requirement.
2573 .globl ${PREFIX}_set_encrypt_key
2574 .type ${PREFIX}_set_encrypt_key,\@abi-omnipotent
2576 ${PREFIX}_set_encrypt_key:
2577 __aesni_set_encrypt_key:
2578 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8
2585 movups ($inp),%xmm0 # pull first 128 bits of *userKey
2586 xorps %xmm4,%xmm4 # low dword of xmm4 is assumed 0
2596 mov \$9,$bits # 10 rounds for 128-bit key
2597 $movkey %xmm0,($key) # round 0
2598 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 1
2599 call .Lkey_expansion_128_cold
2600 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 2
2601 call .Lkey_expansion_128
2602 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 3
2603 call .Lkey_expansion_128
2604 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 4
2605 call .Lkey_expansion_128
2606 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 5
2607 call .Lkey_expansion_128
2608 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 6
2609 call .Lkey_expansion_128
2610 aeskeygenassist \$0x40,%xmm0,%xmm1 # round 7
2611 call .Lkey_expansion_128
2612 aeskeygenassist \$0x80,%xmm0,%xmm1 # round 8
2613 call .Lkey_expansion_128
2614 aeskeygenassist \$0x1b,%xmm0,%xmm1 # round 9
2615 call .Lkey_expansion_128
2616 aeskeygenassist \$0x36,%xmm0,%xmm1 # round 10
2617 call .Lkey_expansion_128
2618 $movkey %xmm0,(%rax)
2619 mov $bits,80(%rax) # 240(%rdx)
2625 movq 16($inp),%xmm2 # remaining 1/3 of *userKey
2626 mov \$11,$bits # 12 rounds for 192
2627 $movkey %xmm0,($key) # round 0
2628 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 1,2
2629 call .Lkey_expansion_192a_cold
2630 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 2,3
2631 call .Lkey_expansion_192b
2632 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 4,5
2633 call .Lkey_expansion_192a
2634 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 5,6
2635 call .Lkey_expansion_192b
2636 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 7,8
2637 call .Lkey_expansion_192a
2638 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 8,9
2639 call .Lkey_expansion_192b
2640 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 10,11
2641 call .Lkey_expansion_192a
2642 aeskeygenassist \$0x80,%xmm2,%xmm1 # round 11,12
2643 call .Lkey_expansion_192b
2644 $movkey %xmm0,(%rax)
2645 mov $bits,48(%rax) # 240(%rdx)
2651 movups 16($inp),%xmm2 # remaning half of *userKey
2652 mov \$13,$bits # 14 rounds for 256
2654 $movkey %xmm0,($key) # round 0
2655 $movkey %xmm2,16($key) # round 1
2656 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 2
2657 call .Lkey_expansion_256a_cold
2658 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 3
2659 call .Lkey_expansion_256b
2660 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 4
2661 call .Lkey_expansion_256a
2662 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 5
2663 call .Lkey_expansion_256b
2664 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 6
2665 call .Lkey_expansion_256a
2666 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 7
2667 call .Lkey_expansion_256b
2668 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 8
2669 call .Lkey_expansion_256a
2670 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 9
2671 call .Lkey_expansion_256b
2672 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 10
2673 call .Lkey_expansion_256a
2674 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 11
2675 call .Lkey_expansion_256b
2676 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 12
2677 call .Lkey_expansion_256a
2678 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 13
2679 call .Lkey_expansion_256b
2680 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 14
2681 call .Lkey_expansion_256a
2682 $movkey %xmm0,(%rax)
2683 mov $bits,16(%rax) # 240(%rdx)
2693 .LSEH_end_set_encrypt_key:
2696 .Lkey_expansion_128:
2697 $movkey %xmm0,(%rax)
2699 .Lkey_expansion_128_cold:
2700 shufps \$0b00010000,%xmm0,%xmm4
2702 shufps \$0b10001100,%xmm0,%xmm4
2704 shufps \$0b11111111,%xmm1,%xmm1 # critical path
2709 .Lkey_expansion_192a:
2710 $movkey %xmm0,(%rax)
2712 .Lkey_expansion_192a_cold:
2714 .Lkey_expansion_192b_warm:
2715 shufps \$0b00010000,%xmm0,%xmm4
2718 shufps \$0b10001100,%xmm0,%xmm4
2721 pshufd \$0b01010101,%xmm1,%xmm1 # critical path
2724 pshufd \$0b11111111,%xmm0,%xmm3
2729 .Lkey_expansion_192b:
2731 shufps \$0b01000100,%xmm0,%xmm5
2732 $movkey %xmm5,(%rax)
2733 shufps \$0b01001110,%xmm2,%xmm3
2734 $movkey %xmm3,16(%rax)
2736 jmp .Lkey_expansion_192b_warm
2739 .Lkey_expansion_256a:
2740 $movkey %xmm2,(%rax)
2742 .Lkey_expansion_256a_cold:
2743 shufps \$0b00010000,%xmm0,%xmm4
2745 shufps \$0b10001100,%xmm0,%xmm4
2747 shufps \$0b11111111,%xmm1,%xmm1 # critical path
2752 .Lkey_expansion_256b:
2753 $movkey %xmm0,(%rax)
2756 shufps \$0b00010000,%xmm2,%xmm4
2758 shufps \$0b10001100,%xmm2,%xmm4
2760 shufps \$0b10101010,%xmm1,%xmm1 # critical path
2763 .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key
2764 .size __aesni_set_encrypt_key,.-__aesni_set_encrypt_key
2771 .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
2779 .byte 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
2781 .asciz "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>"
2785 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2786 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
2794 .extern __imp_RtlVirtualUnwind
2796 $code.=<<___ if ($PREFIX eq "aesni");
2797 .type ecb_se_handler,\@abi-omnipotent
2811 mov 152($context),%rax # pull context->Rsp
2813 jmp .Lcommon_seh_tail
2814 .size ecb_se_handler,.-ecb_se_handler
2816 .type ccm64_se_handler,\@abi-omnipotent
2830 mov 120($context),%rax # pull context->Rax
2831 mov 248($context),%rbx # pull context->Rip
2833 mov 8($disp),%rsi # disp->ImageBase
2834 mov 56($disp),%r11 # disp->HandlerData
2836 mov 0(%r11),%r10d # HandlerData[0]
2837 lea (%rsi,%r10),%r10 # prologue label
2838 cmp %r10,%rbx # context->Rip<prologue label
2839 jb .Lcommon_seh_tail
2841 mov 152($context),%rax # pull context->Rsp
2843 mov 4(%r11),%r10d # HandlerData[1]
2844 lea (%rsi,%r10),%r10 # epilogue label
2845 cmp %r10,%rbx # context->Rip>=epilogue label
2846 jae .Lcommon_seh_tail
2848 lea 0(%rax),%rsi # %xmm save area
2849 lea 512($context),%rdi # &context.Xmm6
2850 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax)
2851 .long 0xa548f3fc # cld; rep movsq
2852 lea 0x58(%rax),%rax # adjust stack pointer
2854 jmp .Lcommon_seh_tail
2855 .size ccm64_se_handler,.-ccm64_se_handler
2857 .type ctr32_se_handler,\@abi-omnipotent
2871 mov 120($context),%rax # pull context->Rax
2872 mov 248($context),%rbx # pull context->Rip
2874 lea .Lctr32_body(%rip),%r10
2875 cmp %r10,%rbx # context->Rip<"prologue" label
2876 jb .Lcommon_seh_tail
2878 mov 152($context),%rax # pull context->Rsp
2880 lea .Lctr32_ret(%rip),%r10
2882 jae .Lcommon_seh_tail
2884 lea (%rax),%rsi # %xmm save area
2885 lea 512($context),%rdi # &context.Xmm6
2886 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax)
2887 .long 0xa548f3fc # cld; rep movsq
2888 lea 0xa8(%rax),%rax # adjust stack pointer
2890 jmp .Lcommon_seh_tail
2891 .size ctr32_se_handler,.-ctr32_se_handler
2893 .type xts_se_handler,\@abi-omnipotent
2907 mov 120($context),%rax # pull context->Rax
2908 mov 248($context),%rbx # pull context->Rip
2910 mov 8($disp),%rsi # disp->ImageBase
2911 mov 56($disp),%r11 # disp->HandlerData
2913 mov 0(%r11),%r10d # HandlerData[0]
2914 lea (%rsi,%r10),%r10 # prologue lable
2915 cmp %r10,%rbx # context->Rip<prologue label
2916 jb .Lcommon_seh_tail
2918 mov 152($context),%rax # pull context->Rsp
2920 mov 4(%r11),%r10d # HandlerData[1]
2921 lea (%rsi,%r10),%r10 # epilogue label
2922 cmp %r10,%rbx # context->Rip>=epilogue label
2923 jae .Lcommon_seh_tail
2925 lea 0x60(%rax),%rsi # %xmm save area
2926 lea 512($context),%rdi # & context.Xmm6
2927 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax)
2928 .long 0xa548f3fc # cld; rep movsq
2930 jmp .Lcommon_rbp_tail
2931 .size xts_se_handler,.-xts_se_handler
2934 .type cbc_se_handler,\@abi-omnipotent
2948 mov 152($context),%rax # pull context->Rsp
2949 mov 248($context),%rbx # pull context->Rip
2951 lea .Lcbc_decrypt(%rip),%r10
2952 cmp %r10,%rbx # context->Rip<"prologue" label
2953 jb .Lcommon_seh_tail
2955 lea .Lcbc_decrypt_body(%rip),%r10
2956 cmp %r10,%rbx # context->Rip<cbc_decrypt_body
2957 jb .Lrestore_cbc_rax
2959 lea .Lcbc_ret(%rip),%r10
2960 cmp %r10,%rbx # context->Rip>="epilogue" label
2961 jae .Lcommon_seh_tail
2963 lea 16(%rax),%rsi # %xmm save area
2964 lea 512($context),%rdi # &context.Xmm6
2965 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax)
2966 .long 0xa548f3fc # cld; rep movsq
2969 mov 160($context),%rax # pull context->Rbp
2970 mov (%rax),%rbp # restore saved %rbp
2971 lea 8(%rax),%rax # adjust stack pointer
2972 mov %rbp,160($context) # restore context->Rbp
2973 jmp .Lcommon_seh_tail
2976 mov 120($context),%rax
2981 mov %rax,152($context) # restore context->Rsp
2982 mov %rsi,168($context) # restore context->Rsi
2983 mov %rdi,176($context) # restore context->Rdi
2985 mov 40($disp),%rdi # disp->ContextRecord
2986 mov $context,%rsi # context
2987 mov \$154,%ecx # sizeof(CONTEXT)
2988 .long 0xa548f3fc # cld; rep movsq
2991 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
2992 mov 8(%rsi),%rdx # arg2, disp->ImageBase
2993 mov 0(%rsi),%r8 # arg3, disp->ControlPc
2994 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
2995 mov 40(%rsi),%r10 # disp->ContextRecord
2996 lea 56(%rsi),%r11 # &disp->HandlerData
2997 lea 24(%rsi),%r12 # &disp->EstablisherFrame
2998 mov %r10,32(%rsp) # arg5
2999 mov %r11,40(%rsp) # arg6
3000 mov %r12,48(%rsp) # arg7
3001 mov %rcx,56(%rsp) # arg8, (NULL)
3002 call *__imp_RtlVirtualUnwind(%rip)
3004 mov \$1,%eax # ExceptionContinueSearch
3016 .size cbc_se_handler,.-cbc_se_handler
3021 $code.=<<___ if ($PREFIX eq "aesni");
3022 .rva .LSEH_begin_aesni_ecb_encrypt
3023 .rva .LSEH_end_aesni_ecb_encrypt
3026 .rva .LSEH_begin_aesni_ccm64_encrypt_blocks
3027 .rva .LSEH_end_aesni_ccm64_encrypt_blocks
3028 .rva .LSEH_info_ccm64_enc
3030 .rva .LSEH_begin_aesni_ccm64_decrypt_blocks
3031 .rva .LSEH_end_aesni_ccm64_decrypt_blocks
3032 .rva .LSEH_info_ccm64_dec
3034 .rva .LSEH_begin_aesni_ctr32_encrypt_blocks
3035 .rva .LSEH_end_aesni_ctr32_encrypt_blocks
3036 .rva .LSEH_info_ctr32
3038 .rva .LSEH_begin_aesni_xts_encrypt
3039 .rva .LSEH_end_aesni_xts_encrypt
3040 .rva .LSEH_info_xts_enc
3042 .rva .LSEH_begin_aesni_xts_decrypt
3043 .rva .LSEH_end_aesni_xts_decrypt
3044 .rva .LSEH_info_xts_dec
3047 .rva .LSEH_begin_${PREFIX}_cbc_encrypt
3048 .rva .LSEH_end_${PREFIX}_cbc_encrypt
3051 .rva ${PREFIX}_set_decrypt_key
3052 .rva .LSEH_end_set_decrypt_key
3055 .rva ${PREFIX}_set_encrypt_key
3056 .rva .LSEH_end_set_encrypt_key
3061 $code.=<<___ if ($PREFIX eq "aesni");
3065 .LSEH_info_ccm64_enc:
3067 .rva ccm64_se_handler
3068 .rva .Lccm64_enc_body,.Lccm64_enc_ret # HandlerData[]
3069 .LSEH_info_ccm64_dec:
3071 .rva ccm64_se_handler
3072 .rva .Lccm64_dec_body,.Lccm64_dec_ret # HandlerData[]
3075 .rva ctr32_se_handler
3079 .rva .Lxts_enc_body,.Lxts_enc_epilogue # HandlerData[]
3083 .rva .Lxts_dec_body,.Lxts_dec_epilogue # HandlerData[]
3090 .byte 0x01,0x04,0x01,0x00
3091 .byte 0x04,0x02,0x00,0x00 # sub rsp,8
3096 local *opcode=shift;
3100 $rex|=0x04 if($dst>=8);
3101 $rex|=0x01 if($src>=8);
3102 push @opcode,$rex|0x40 if($rex);
3109 if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3110 rex(\@opcode,$4,$3);
3111 push @opcode,0x0f,0x3a,0xdf;
3112 push @opcode,0xc0|($3&7)|(($4&7)<<3); # ModR/M
3114 push @opcode,$c=~/^0/?oct($c):$c;
3115 return ".byte\t".join(',',@opcode);
3117 elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
3120 "aesenc" => 0xdc, "aesenclast" => 0xdd,
3121 "aesdec" => 0xde, "aesdeclast" => 0xdf
3123 return undef if (!defined($opcodelet{$1}));
3124 rex(\@opcode,$3,$2);
3125 push @opcode,0x0f,0x38,$opcodelet{$1};
3126 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
3127 return ".byte\t".join(',',@opcode);
3132 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
3133 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;