88ad44ee5e57f77cce1bca383e9f704a9f93e6d0
[openssl.git] / crypto / sha / asm / sha512-armv8.pl
1 #! /usr/bin/env perl
2 # Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved.
3 #
4 # Licensed under the OpenSSL license (the "License").  You may not use
5 # this file except in compliance with the License.  You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
8
9 # ====================================================================
10 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
11 # project. The module is, however, dual licensed under OpenSSL and
12 # CRYPTOGAMS licenses depending on where you obtain it. For further
13 # details see http://www.openssl.org/~appro/cryptogams/.
14 #
15 # Permission to use under GPLv2 terms is granted.
16 # ====================================================================
17 #
18 # SHA256/512 for ARMv8.
19 #
20 # Performance in cycles per processed byte and improvement coefficient
21 # over code generated with "default" compiler:
22 #
23 #               SHA256-hw       SHA256(*)       SHA512
24 # Apple A7      1.97            10.5 (+33%)     6.73 (-1%(**))
25 # Cortex-A53    2.38            15.5 (+115%)    10.0 (+150%(***))
26 # Cortex-A57    2.31            11.6 (+86%)     7.51 (+260%(***))
27 # Denver        2.01            10.5 (+26%)     6.70 (+8%)
28 # X-Gene                        20.0 (+100%)    12.8 (+300%(***))
29 # Mongoose      2.36            13.0 (+50%)     8.36 (+33%)
30 #
31 # (*)   Software SHA256 results are of lesser relevance, presented
32 #       mostly for informational purposes.
33 # (**)  The result is a trade-off: it's possible to improve it by
34 #       10% (or by 1 cycle per round), but at the cost of 20% loss
35 #       on Cortex-A53 (or by 4 cycles per round).
36 # (***) Super-impressive coefficients over gcc-generated code are
37 #       indication of some compiler "pathology", most notably code
38 #       generated with -mgeneral-regs-only is significantly faster
39 #       and the gap is only 40-90%.
40 #
41 # October 2016.
42 #
43 # Originally it was reckoned that it makes no sense to implement NEON
44 # version of SHA256 for 64-bit processors. This is because performance
45 # improvement on most wide-spread Cortex-A5x processors was observed
46 # to be marginal, same on Cortex-A53 and ~10% on A57. But then it was
47 # observed that 32-bit NEON SHA256 performs significantly better than
48 # 64-bit scalar version on *some* of the more recent processors. As
49 # result 64-bit NEON version of SHA256 was added to provide best
50 # all-round performance. For example it executes ~30% faster on X-Gene
51 # and Mongoose. [For reference, NEON version of SHA512 is bound to
52 # deliver much less improvement, likely *negative* on Cortex-A5x.
53 # Which is why NEON support is limited to SHA256.]
54
55 $output=pop;
56 $flavour=pop;
57
58 if ($flavour && $flavour ne "void") {
59     $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
60     ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
61     ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
62     die "can't locate arm-xlate.pl";
63
64     open OUT,"| \"$^X\" $xlate $flavour $output";
65     *STDOUT=*OUT;
66 } else {
67     open STDOUT,">$output";
68 }
69
70 if ($output =~ /512/) {
71         $BITS=512;
72         $SZ=8;
73         @Sigma0=(28,34,39);
74         @Sigma1=(14,18,41);
75         @sigma0=(1,  8, 7);
76         @sigma1=(19,61, 6);
77         $rounds=80;
78         $reg_t="x";
79 } else {
80         $BITS=256;
81         $SZ=4;
82         @Sigma0=( 2,13,22);
83         @Sigma1=( 6,11,25);
84         @sigma0=( 7,18, 3);
85         @sigma1=(17,19,10);
86         $rounds=64;
87         $reg_t="w";
88 }
89
90 $func="sha${BITS}_block_data_order";
91
92 ($ctx,$inp,$num,$Ktbl)=map("x$_",(0..2,30));
93
94 @X=map("$reg_t$_",(3..15,0..2));
95 @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("$reg_t$_",(20..27));
96 ($t0,$t1,$t2,$t3)=map("$reg_t$_",(16,17,19,28));
97
98 sub BODY_00_xx {
99 my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
100 my $j=($i+1)&15;
101 my ($T0,$T1,$T2)=(@X[($i-8)&15],@X[($i-9)&15],@X[($i-10)&15]);
102    $T0=@X[$i+3] if ($i<11);
103
104 $code.=<<___    if ($i<16);
105 #ifndef __AARCH64EB__
106         rev     @X[$i],@X[$i]                   // $i
107 #endif
108 ___
109 $code.=<<___    if ($i<13 && ($i&1));
110         ldp     @X[$i+1],@X[$i+2],[$inp],#2*$SZ
111 ___
112 $code.=<<___    if ($i==13);
113         ldp     @X[14],@X[15],[$inp]
114 ___
115 $code.=<<___    if ($i>=14);
116         ldr     @X[($i-11)&15],[sp,#`$SZ*(($i-11)%4)`]
117 ___
118 $code.=<<___    if ($i>0 && $i<16);
119         add     $a,$a,$t1                       // h+=Sigma0(a)
120 ___
121 $code.=<<___    if ($i>=11);
122         str     @X[($i-8)&15],[sp,#`$SZ*(($i-8)%4)`]
123 ___
124 # While ARMv8 specifies merged rotate-n-logical operation such as
125 # 'eor x,y,z,ror#n', it was found to negatively affect performance
126 # on Apple A7. The reason seems to be that it requires even 'y' to
127 # be available earlier. This means that such merged instruction is
128 # not necessarily best choice on critical path... On the other hand
129 # Cortex-A5x handles merged instructions much better than disjoint
130 # rotate and logical... See (**) footnote above.
131 $code.=<<___    if ($i<15);
132         ror     $t0,$e,#$Sigma1[0]
133         add     $h,$h,$t2                       // h+=K[i]
134         eor     $T0,$e,$e,ror#`$Sigma1[2]-$Sigma1[1]`
135         and     $t1,$f,$e
136         bic     $t2,$g,$e
137         add     $h,$h,@X[$i&15]                 // h+=X[i]
138         orr     $t1,$t1,$t2                     // Ch(e,f,g)
139         eor     $t2,$a,$b                       // a^b, b^c in next round
140         eor     $t0,$t0,$T0,ror#$Sigma1[1]      // Sigma1(e)
141         ror     $T0,$a,#$Sigma0[0]
142         add     $h,$h,$t1                       // h+=Ch(e,f,g)
143         eor     $t1,$a,$a,ror#`$Sigma0[2]-$Sigma0[1]`
144         add     $h,$h,$t0                       // h+=Sigma1(e)
145         and     $t3,$t3,$t2                     // (b^c)&=(a^b)
146         add     $d,$d,$h                        // d+=h
147         eor     $t3,$t3,$b                      // Maj(a,b,c)
148         eor     $t1,$T0,$t1,ror#$Sigma0[1]      // Sigma0(a)
149         add     $h,$h,$t3                       // h+=Maj(a,b,c)
150         ldr     $t3,[$Ktbl],#$SZ                // *K++, $t2 in next round
151         //add   $h,$h,$t1                       // h+=Sigma0(a)
152 ___
153 $code.=<<___    if ($i>=15);
154         ror     $t0,$e,#$Sigma1[0]
155         add     $h,$h,$t2                       // h+=K[i]
156         ror     $T1,@X[($j+1)&15],#$sigma0[0]
157         and     $t1,$f,$e
158         ror     $T2,@X[($j+14)&15],#$sigma1[0]
159         bic     $t2,$g,$e
160         ror     $T0,$a,#$Sigma0[0]
161         add     $h,$h,@X[$i&15]                 // h+=X[i]
162         eor     $t0,$t0,$e,ror#$Sigma1[1]
163         eor     $T1,$T1,@X[($j+1)&15],ror#$sigma0[1]
164         orr     $t1,$t1,$t2                     // Ch(e,f,g)
165         eor     $t2,$a,$b                       // a^b, b^c in next round
166         eor     $t0,$t0,$e,ror#$Sigma1[2]       // Sigma1(e)
167         eor     $T0,$T0,$a,ror#$Sigma0[1]
168         add     $h,$h,$t1                       // h+=Ch(e,f,g)
169         and     $t3,$t3,$t2                     // (b^c)&=(a^b)
170         eor     $T2,$T2,@X[($j+14)&15],ror#$sigma1[1]
171         eor     $T1,$T1,@X[($j+1)&15],lsr#$sigma0[2]    // sigma0(X[i+1])
172         add     $h,$h,$t0                       // h+=Sigma1(e)
173         eor     $t3,$t3,$b                      // Maj(a,b,c)
174         eor     $t1,$T0,$a,ror#$Sigma0[2]       // Sigma0(a)
175         eor     $T2,$T2,@X[($j+14)&15],lsr#$sigma1[2]   // sigma1(X[i+14])
176         add     @X[$j],@X[$j],@X[($j+9)&15]
177         add     $d,$d,$h                        // d+=h
178         add     $h,$h,$t3                       // h+=Maj(a,b,c)
179         ldr     $t3,[$Ktbl],#$SZ                // *K++, $t2 in next round
180         add     @X[$j],@X[$j],$T1
181         add     $h,$h,$t1                       // h+=Sigma0(a)
182         add     @X[$j],@X[$j],$T2
183 ___
184         ($t2,$t3)=($t3,$t2);
185 }
186
187 $code.=<<___;
188 #ifndef __KERNEL__
189 # include "arm_arch.h"
190 #endif
191
192 .text
193
194 .extern OPENSSL_armcap_P
195 .globl  $func
196 .type   $func,%function
197 .align  6
198 $func:
199 ___
200 $code.=<<___    if ($SZ==4);
201 #ifndef __KERNEL__
202 # ifdef __ILP32__
203         ldrsw   x16,.LOPENSSL_armcap_P
204 # else
205         ldr     x16,.LOPENSSL_armcap_P
206 # endif
207         adr     x17,.LOPENSSL_armcap_P
208         add     x16,x16,x17
209         ldr     w16,[x16]
210         tst     w16,#ARMV8_SHA256
211         b.ne    .Lv8_entry
212         tst     w16,#ARMV7_NEON
213         b.ne    .Lneon_entry
214 #endif
215 ___
216 $code.=<<___;
217         stp     x29,x30,[sp,#-128]!
218         add     x29,sp,#0
219
220         stp     x19,x20,[sp,#16]
221         stp     x21,x22,[sp,#32]
222         stp     x23,x24,[sp,#48]
223         stp     x25,x26,[sp,#64]
224         stp     x27,x28,[sp,#80]
225         sub     sp,sp,#4*$SZ
226
227         ldp     $A,$B,[$ctx]                            // load context
228         ldp     $C,$D,[$ctx,#2*$SZ]
229         ldp     $E,$F,[$ctx,#4*$SZ]
230         add     $num,$inp,$num,lsl#`log(16*$SZ)/log(2)` // end of input
231         ldp     $G,$H,[$ctx,#6*$SZ]
232         adr     $Ktbl,.LK$BITS
233         stp     $ctx,$num,[x29,#96]
234
235 .Loop:
236         ldp     @X[0],@X[1],[$inp],#2*$SZ
237         ldr     $t2,[$Ktbl],#$SZ                        // *K++
238         eor     $t3,$B,$C                               // magic seed
239         str     $inp,[x29,#112]
240 ___
241 for ($i=0;$i<16;$i++)   { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
242 $code.=".Loop_16_xx:\n";
243 for (;$i<32;$i++)       { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); }
244 $code.=<<___;
245         cbnz    $t2,.Loop_16_xx
246
247         ldp     $ctx,$num,[x29,#96]
248         ldr     $inp,[x29,#112]
249         sub     $Ktbl,$Ktbl,#`$SZ*($rounds+1)`          // rewind
250
251         ldp     @X[0],@X[1],[$ctx]
252         ldp     @X[2],@X[3],[$ctx,#2*$SZ]
253         add     $inp,$inp,#14*$SZ                       // advance input pointer
254         ldp     @X[4],@X[5],[$ctx,#4*$SZ]
255         add     $A,$A,@X[0]
256         ldp     @X[6],@X[7],[$ctx,#6*$SZ]
257         add     $B,$B,@X[1]
258         add     $C,$C,@X[2]
259         add     $D,$D,@X[3]
260         stp     $A,$B,[$ctx]
261         add     $E,$E,@X[4]
262         add     $F,$F,@X[5]
263         stp     $C,$D,[$ctx,#2*$SZ]
264         add     $G,$G,@X[6]
265         add     $H,$H,@X[7]
266         cmp     $inp,$num
267         stp     $E,$F,[$ctx,#4*$SZ]
268         stp     $G,$H,[$ctx,#6*$SZ]
269         b.ne    .Loop
270
271         ldp     x19,x20,[x29,#16]
272         add     sp,sp,#4*$SZ
273         ldp     x21,x22,[x29,#32]
274         ldp     x23,x24,[x29,#48]
275         ldp     x25,x26,[x29,#64]
276         ldp     x27,x28,[x29,#80]
277         ldp     x29,x30,[sp],#128
278         ret
279 .size   $func,.-$func
280
281 .align  6
282 .type   .LK$BITS,%object
283 .LK$BITS:
284 ___
285 $code.=<<___ if ($SZ==8);
286         .quad   0x428a2f98d728ae22,0x7137449123ef65cd
287         .quad   0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
288         .quad   0x3956c25bf348b538,0x59f111f1b605d019
289         .quad   0x923f82a4af194f9b,0xab1c5ed5da6d8118
290         .quad   0xd807aa98a3030242,0x12835b0145706fbe
291         .quad   0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
292         .quad   0x72be5d74f27b896f,0x80deb1fe3b1696b1
293         .quad   0x9bdc06a725c71235,0xc19bf174cf692694
294         .quad   0xe49b69c19ef14ad2,0xefbe4786384f25e3
295         .quad   0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
296         .quad   0x2de92c6f592b0275,0x4a7484aa6ea6e483
297         .quad   0x5cb0a9dcbd41fbd4,0x76f988da831153b5
298         .quad   0x983e5152ee66dfab,0xa831c66d2db43210
299         .quad   0xb00327c898fb213f,0xbf597fc7beef0ee4
300         .quad   0xc6e00bf33da88fc2,0xd5a79147930aa725
301         .quad   0x06ca6351e003826f,0x142929670a0e6e70
302         .quad   0x27b70a8546d22ffc,0x2e1b21385c26c926
303         .quad   0x4d2c6dfc5ac42aed,0x53380d139d95b3df
304         .quad   0x650a73548baf63de,0x766a0abb3c77b2a8
305         .quad   0x81c2c92e47edaee6,0x92722c851482353b
306         .quad   0xa2bfe8a14cf10364,0xa81a664bbc423001
307         .quad   0xc24b8b70d0f89791,0xc76c51a30654be30
308         .quad   0xd192e819d6ef5218,0xd69906245565a910
309         .quad   0xf40e35855771202a,0x106aa07032bbd1b8
310         .quad   0x19a4c116b8d2d0c8,0x1e376c085141ab53
311         .quad   0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
312         .quad   0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
313         .quad   0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
314         .quad   0x748f82ee5defb2fc,0x78a5636f43172f60
315         .quad   0x84c87814a1f0ab72,0x8cc702081a6439ec
316         .quad   0x90befffa23631e28,0xa4506cebde82bde9
317         .quad   0xbef9a3f7b2c67915,0xc67178f2e372532b
318         .quad   0xca273eceea26619c,0xd186b8c721c0c207
319         .quad   0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
320         .quad   0x06f067aa72176fba,0x0a637dc5a2c898a6
321         .quad   0x113f9804bef90dae,0x1b710b35131c471b
322         .quad   0x28db77f523047d84,0x32caab7b40c72493
323         .quad   0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
324         .quad   0x4cc5d4becb3e42b6,0x597f299cfc657e2a
325         .quad   0x5fcb6fab3ad6faec,0x6c44198c4a475817
326         .quad   0       // terminator
327 ___
328 $code.=<<___ if ($SZ==4);
329         .long   0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
330         .long   0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
331         .long   0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
332         .long   0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
333         .long   0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
334         .long   0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
335         .long   0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
336         .long   0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
337         .long   0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
338         .long   0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
339         .long   0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
340         .long   0xd192e819,0xd6990624,0xf40e3585,0x106aa070
341         .long   0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
342         .long   0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
343         .long   0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
344         .long   0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
345         .long   0       //terminator
346 ___
347 $code.=<<___;
348 .size   .LK$BITS,.-.LK$BITS
349 #ifndef __KERNEL__
350 .align  3
351 .LOPENSSL_armcap_P:
352 # ifdef __ILP32__
353         .long   OPENSSL_armcap_P-.
354 # else
355         .quad   OPENSSL_armcap_P-.
356 # endif
357 #endif
358 .asciz  "SHA$BITS block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
359 .align  2
360 ___
361
362 if ($SZ==4) {
363 my $Ktbl="x3";
364
365 my ($ABCD,$EFGH,$abcd)=map("v$_.16b",(0..2));
366 my @MSG=map("v$_.16b",(4..7));
367 my ($W0,$W1)=("v16.4s","v17.4s");
368 my ($ABCD_SAVE,$EFGH_SAVE)=("v18.16b","v19.16b");
369
370 $code.=<<___;
371 #ifndef __KERNEL__
372 .type   sha256_block_armv8,%function
373 .align  6
374 sha256_block_armv8:
375 .Lv8_entry:
376         stp             x29,x30,[sp,#-16]!
377         add             x29,sp,#0
378
379         ld1.32          {$ABCD,$EFGH},[$ctx]
380         adr             $Ktbl,.LK256
381
382 .Loop_hw:
383         ld1             {@MSG[0]-@MSG[3]},[$inp],#64
384         sub             $num,$num,#1
385         ld1.32          {$W0},[$Ktbl],#16
386         rev32           @MSG[0],@MSG[0]
387         rev32           @MSG[1],@MSG[1]
388         rev32           @MSG[2],@MSG[2]
389         rev32           @MSG[3],@MSG[3]
390         orr             $ABCD_SAVE,$ABCD,$ABCD          // offload
391         orr             $EFGH_SAVE,$EFGH,$EFGH
392 ___
393 for($i=0;$i<12;$i++) {
394 $code.=<<___;
395         ld1.32          {$W1},[$Ktbl],#16
396         add.i32         $W0,$W0,@MSG[0]
397         sha256su0       @MSG[0],@MSG[1]
398         orr             $abcd,$ABCD,$ABCD
399         sha256h         $ABCD,$EFGH,$W0
400         sha256h2        $EFGH,$abcd,$W0
401         sha256su1       @MSG[0],@MSG[2],@MSG[3]
402 ___
403         ($W0,$W1)=($W1,$W0);    push(@MSG,shift(@MSG));
404 }
405 $code.=<<___;
406         ld1.32          {$W1},[$Ktbl],#16
407         add.i32         $W0,$W0,@MSG[0]
408         orr             $abcd,$ABCD,$ABCD
409         sha256h         $ABCD,$EFGH,$W0
410         sha256h2        $EFGH,$abcd,$W0
411
412         ld1.32          {$W0},[$Ktbl],#16
413         add.i32         $W1,$W1,@MSG[1]
414         orr             $abcd,$ABCD,$ABCD
415         sha256h         $ABCD,$EFGH,$W1
416         sha256h2        $EFGH,$abcd,$W1
417
418         ld1.32          {$W1},[$Ktbl]
419         add.i32         $W0,$W0,@MSG[2]
420         sub             $Ktbl,$Ktbl,#$rounds*$SZ-16     // rewind
421         orr             $abcd,$ABCD,$ABCD
422         sha256h         $ABCD,$EFGH,$W0
423         sha256h2        $EFGH,$abcd,$W0
424
425         add.i32         $W1,$W1,@MSG[3]
426         orr             $abcd,$ABCD,$ABCD
427         sha256h         $ABCD,$EFGH,$W1
428         sha256h2        $EFGH,$abcd,$W1
429
430         add.i32         $ABCD,$ABCD,$ABCD_SAVE
431         add.i32         $EFGH,$EFGH,$EFGH_SAVE
432
433         cbnz            $num,.Loop_hw
434
435         st1.32          {$ABCD,$EFGH},[$ctx]
436
437         ldr             x29,[sp],#16
438         ret
439 .size   sha256_block_armv8,.-sha256_block_armv8
440 #endif
441 ___
442 }
443
444 if ($SZ==4) {   ######################################### NEON stuff #
445 # You'll surely note a lot of similarities with sha256-armv4 module,
446 # and of course it's not a coincidence. sha256-armv4 was used as
447 # initial template, but was adapted for ARMv8 instruction set and
448 # extensively re-tuned for all-round performance.
449
450 my @V = ($A,$B,$C,$D,$E,$F,$G,$H) = map("w$_",(3..10));
451 my ($t0,$t1,$t2,$t3,$t4) = map("w$_",(11..15));
452 my $Ktbl="x16";
453 my $Xfer="x17";
454 my @X = map("q$_",(0..3));
455 my ($T0,$T1,$T2,$T3,$T4,$T5,$T6,$T7) = map("q$_",(4..7,16..19));
456 my $j=0;
457
458 sub AUTOLOAD()          # thunk [simplified] x86-style perlasm
459 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
460   my $arg = pop;
461     $arg = "#$arg" if ($arg*1 eq $arg);
462     $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
463 }
464
465 sub Dscalar { shift =~ m|[qv]([0-9]+)|?"d$1":""; }
466 sub Dlo     { shift =~ m|[qv]([0-9]+)|?"v$1.d[0]":""; }
467 sub Dhi     { shift =~ m|[qv]([0-9]+)|?"v$1.d[1]":""; }
468
469 sub Xupdate()
470 { use integer;
471   my $body = shift;
472   my @insns = (&$body,&$body,&$body,&$body);
473   my ($a,$b,$c,$d,$e,$f,$g,$h);
474
475         &ext_8          ($T0,@X[0],@X[1],4);    # X[1..4]
476          eval(shift(@insns));
477          eval(shift(@insns));
478          eval(shift(@insns));
479         &ext_8          ($T3,@X[2],@X[3],4);    # X[9..12]
480          eval(shift(@insns));
481          eval(shift(@insns));
482         &mov            (&Dscalar($T7),&Dhi(@X[3]));    # X[14..15]
483          eval(shift(@insns));
484          eval(shift(@insns));
485         &ushr_32        ($T2,$T0,$sigma0[0]);
486          eval(shift(@insns));
487         &ushr_32        ($T1,$T0,$sigma0[2]);
488          eval(shift(@insns));
489         &add_32         (@X[0],@X[0],$T3);      # X[0..3] += X[9..12]
490          eval(shift(@insns));
491         &sli_32         ($T2,$T0,32-$sigma0[0]);
492          eval(shift(@insns));
493          eval(shift(@insns));
494         &ushr_32        ($T3,$T0,$sigma0[1]);
495          eval(shift(@insns));
496          eval(shift(@insns));
497         &eor_8          ($T1,$T1,$T2);
498          eval(shift(@insns));
499          eval(shift(@insns));
500         &sli_32         ($T3,$T0,32-$sigma0[1]);
501          eval(shift(@insns));
502          eval(shift(@insns));
503           &ushr_32      ($T4,$T7,$sigma1[0]);
504          eval(shift(@insns));
505          eval(shift(@insns));
506         &eor_8          ($T1,$T1,$T3);          # sigma0(X[1..4])
507          eval(shift(@insns));
508          eval(shift(@insns));
509           &sli_32       ($T4,$T7,32-$sigma1[0]);
510          eval(shift(@insns));
511          eval(shift(@insns));
512           &ushr_32      ($T5,$T7,$sigma1[2]);
513          eval(shift(@insns));
514          eval(shift(@insns));
515           &ushr_32      ($T3,$T7,$sigma1[1]);
516          eval(shift(@insns));
517          eval(shift(@insns));
518         &add_32         (@X[0],@X[0],$T1);      # X[0..3] += sigma0(X[1..4])
519          eval(shift(@insns));
520          eval(shift(@insns));
521           &sli_u32      ($T3,$T7,32-$sigma1[1]);
522          eval(shift(@insns));
523          eval(shift(@insns));
524           &eor_8        ($T5,$T5,$T4);
525          eval(shift(@insns));
526          eval(shift(@insns));
527          eval(shift(@insns));
528           &eor_8        ($T5,$T5,$T3);          # sigma1(X[14..15])
529          eval(shift(@insns));
530          eval(shift(@insns));
531          eval(shift(@insns));
532         &add_32         (@X[0],@X[0],$T5);      # X[0..1] += sigma1(X[14..15])
533          eval(shift(@insns));
534          eval(shift(@insns));
535          eval(shift(@insns));
536           &ushr_32      ($T6,@X[0],$sigma1[0]);
537          eval(shift(@insns));
538           &ushr_32      ($T7,@X[0],$sigma1[2]);
539          eval(shift(@insns));
540          eval(shift(@insns));
541           &sli_32       ($T6,@X[0],32-$sigma1[0]);
542          eval(shift(@insns));
543           &ushr_32      ($T5,@X[0],$sigma1[1]);
544          eval(shift(@insns));
545          eval(shift(@insns));
546           &eor_8        ($T7,$T7,$T6);
547          eval(shift(@insns));
548          eval(shift(@insns));
549           &sli_32       ($T5,@X[0],32-$sigma1[1]);
550          eval(shift(@insns));
551          eval(shift(@insns));
552         &ld1_32         ("{$T0}","[$Ktbl], #16");
553          eval(shift(@insns));
554           &eor_8        ($T7,$T7,$T5);          # sigma1(X[16..17])
555          eval(shift(@insns));
556          eval(shift(@insns));
557         &eor_8          ($T5,$T5,$T5);
558          eval(shift(@insns));
559          eval(shift(@insns));
560         &mov            (&Dhi($T5), &Dlo($T7));
561          eval(shift(@insns));
562          eval(shift(@insns));
563          eval(shift(@insns));
564         &add_32         (@X[0],@X[0],$T5);      # X[2..3] += sigma1(X[16..17])
565          eval(shift(@insns));
566          eval(shift(@insns));
567          eval(shift(@insns));
568         &add_32         ($T0,$T0,@X[0]);
569          while($#insns>=1) { eval(shift(@insns)); }
570         &st1_32         ("{$T0}","[$Xfer], #16");
571          eval(shift(@insns));
572
573         push(@X,shift(@X));             # "rotate" X[]
574 }
575
576 sub Xpreload()
577 { use integer;
578   my $body = shift;
579   my @insns = (&$body,&$body,&$body,&$body);
580   my ($a,$b,$c,$d,$e,$f,$g,$h);
581
582          eval(shift(@insns));
583          eval(shift(@insns));
584         &ld1_8          ("{@X[0]}","[$inp],#16");
585          eval(shift(@insns));
586          eval(shift(@insns));
587         &ld1_32         ("{$T0}","[$Ktbl],#16");
588          eval(shift(@insns));
589          eval(shift(@insns));
590          eval(shift(@insns));
591          eval(shift(@insns));
592         &rev32          (@X[0],@X[0]);
593          eval(shift(@insns));
594          eval(shift(@insns));
595          eval(shift(@insns));
596          eval(shift(@insns));
597         &add_32         ($T0,$T0,@X[0]);
598          foreach (@insns) { eval; }     # remaining instructions
599         &st1_32         ("{$T0}","[$Xfer], #16");
600
601         push(@X,shift(@X));             # "rotate" X[]
602 }
603
604 sub body_00_15 () {
605         (
606         '($a,$b,$c,$d,$e,$f,$g,$h)=@V;'.
607         '&add   ($h,$h,$t1)',                   # h+=X[i]+K[i]
608         '&add   ($a,$a,$t4);'.                  # h+=Sigma0(a) from the past
609         '&and   ($t1,$f,$e)',
610         '&bic   ($t4,$g,$e)',
611         '&eor   ($t0,$e,$e,"ror#".($Sigma1[1]-$Sigma1[0]))',
612         '&add   ($a,$a,$t2)',                   # h+=Maj(a,b,c) from the past
613         '&orr   ($t1,$t1,$t4)',                 # Ch(e,f,g)
614         '&eor   ($t0,$t0,$e,"ror#".($Sigma1[2]-$Sigma1[0]))',   # Sigma1(e)
615         '&eor   ($t4,$a,$a,"ror#".($Sigma0[1]-$Sigma0[0]))',
616         '&add   ($h,$h,$t1)',                   # h+=Ch(e,f,g)
617         '&ror   ($t0,$t0,"#$Sigma1[0]")',
618         '&eor   ($t2,$a,$b)',                   # a^b, b^c in next round
619         '&eor   ($t4,$t4,$a,"ror#".($Sigma0[2]-$Sigma0[0]))',   # Sigma0(a)
620         '&add   ($h,$h,$t0)',                   # h+=Sigma1(e)
621         '&ldr   ($t1,sprintf "[sp,#%d]",4*(($j+1)&15))  if (($j&15)!=15);'.
622         '&ldr   ($t1,"[$Ktbl]")                         if ($j==15);'.
623         '&and   ($t3,$t3,$t2)',                 # (b^c)&=(a^b)
624         '&ror   ($t4,$t4,"#$Sigma0[0]")',
625         '&add   ($d,$d,$h)',                    # d+=h
626         '&eor   ($t3,$t3,$b)',                  # Maj(a,b,c)
627         '$j++;  unshift(@V,pop(@V)); ($t2,$t3)=($t3,$t2);'
628         )
629 }
630
631 $code.=<<___;
632 #ifdef  __KERNEL__
633 .globl  sha256_block_neon
634 #endif
635 .type   sha256_block_neon,%function
636 .align  4
637 sha256_block_neon:
638 .Lneon_entry:
639         stp     x29, x30, [sp, #-16]!
640         mov     x29, sp
641         sub     sp,sp,#16*4
642
643         adr     $Ktbl,.LK256
644         add     $num,$inp,$num,lsl#6    // len to point at the end of inp
645
646         ld1.8   {@X[0]},[$inp], #16
647         ld1.8   {@X[1]},[$inp], #16
648         ld1.8   {@X[2]},[$inp], #16
649         ld1.8   {@X[3]},[$inp], #16
650         ld1.32  {$T0},[$Ktbl], #16
651         ld1.32  {$T1},[$Ktbl], #16
652         ld1.32  {$T2},[$Ktbl], #16
653         ld1.32  {$T3},[$Ktbl], #16
654         rev32   @X[0],@X[0]             // yes, even on
655         rev32   @X[1],@X[1]             // big-endian
656         rev32   @X[2],@X[2]
657         rev32   @X[3],@X[3]
658         mov     $Xfer,sp
659         add.32  $T0,$T0,@X[0]
660         add.32  $T1,$T1,@X[1]
661         add.32  $T2,$T2,@X[2]
662         st1.32  {$T0-$T1},[$Xfer], #32
663         add.32  $T3,$T3,@X[3]
664         st1.32  {$T2-$T3},[$Xfer]
665         sub     $Xfer,$Xfer,#32
666
667         ldp     $A,$B,[$ctx]
668         ldp     $C,$D,[$ctx,#8]
669         ldp     $E,$F,[$ctx,#16]
670         ldp     $G,$H,[$ctx,#24]
671         ldr     $t1,[sp,#0]
672         mov     $t2,wzr
673         eor     $t3,$B,$C
674         mov     $t4,wzr
675         b       .L_00_48
676
677 .align  4
678 .L_00_48:
679 ___
680         &Xupdate(\&body_00_15);
681         &Xupdate(\&body_00_15);
682         &Xupdate(\&body_00_15);
683         &Xupdate(\&body_00_15);
684 $code.=<<___;
685         cmp     $t1,#0                          // check for K256 terminator
686         ldr     $t1,[sp,#0]
687         sub     $Xfer,$Xfer,#64
688         bne     .L_00_48
689
690         sub     $Ktbl,$Ktbl,#256                // rewind $Ktbl
691         cmp     $inp,$num
692         mov     $Xfer, #64
693         csel    $Xfer, $Xfer, xzr, eq
694         sub     $inp,$inp,$Xfer                 // avoid SEGV
695         mov     $Xfer,sp
696 ___
697         &Xpreload(\&body_00_15);
698         &Xpreload(\&body_00_15);
699         &Xpreload(\&body_00_15);
700         &Xpreload(\&body_00_15);
701 $code.=<<___;
702         add     $A,$A,$t4                       // h+=Sigma0(a) from the past
703         ldp     $t0,$t1,[$ctx,#0]
704         add     $A,$A,$t2                       // h+=Maj(a,b,c) from the past
705         ldp     $t2,$t3,[$ctx,#8]
706         add     $A,$A,$t0                       // accumulate
707         add     $B,$B,$t1
708         ldp     $t0,$t1,[$ctx,#16]
709         add     $C,$C,$t2
710         add     $D,$D,$t3
711         ldp     $t2,$t3,[$ctx,#24]
712         add     $E,$E,$t0
713         add     $F,$F,$t1
714          ldr    $t1,[sp,#0]
715         stp     $A,$B,[$ctx,#0]
716         add     $G,$G,$t2
717          mov    $t2,wzr
718         stp     $C,$D,[$ctx,#8]
719         add     $H,$H,$t3
720         stp     $E,$F,[$ctx,#16]
721          eor    $t3,$B,$C
722         stp     $G,$H,[$ctx,#24]
723          mov    $t4,wzr
724          mov    $Xfer,sp
725         b.ne    .L_00_48
726
727         ldr     x29,[x29]
728         add     sp,sp,#16*4+16
729         ret
730 .size   sha256_block_neon,.-sha256_block_neon
731 ___
732 }
733
734 $code.=<<___;
735 #ifndef __KERNEL__
736 .comm   OPENSSL_armcap_P,4,4
737 #endif
738 ___
739
740 {   my  %opcode = (
741         "sha256h"       => 0x5e004000,  "sha256h2"      => 0x5e005000,
742         "sha256su0"     => 0x5e282800,  "sha256su1"     => 0x5e006000   );
743
744     sub unsha256 {
745         my ($mnemonic,$arg)=@_;
746
747         $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
748         &&
749         sprintf ".inst\t0x%08x\t//%s %s",
750                         $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
751                         $mnemonic,$arg;
752     }
753 }
754
755 open SELF,$0;
756 while(<SELF>) {
757         next if (/^#!/);
758         last if (!s/^#/\/\// and !/^$/);
759         print;
760 }
761 close SELF;
762
763 foreach(split("\n",$code)) {
764
765         s/\`([^\`]*)\`/eval($1)/ge;
766
767         s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/ge;
768
769         s/\bq([0-9]+)\b/v$1.16b/g;              # old->new registers
770
771         s/\.[ui]?8(\s)/$1/;
772         s/\.\w?32\b//           and s/\.16b/\.4s/g;
773         m/(ld|st)1[^\[]+\[0\]/  and s/\.4s/\.s/g;
774
775         print $_,"\n";
776 }
777
778 close STDOUT;