ARMv8 assembly pack: add Qualcomm Kryo results.
[openssl.git] / crypto / sha / asm / sha1-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 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
16 #
17 # SHA1 for ARMv8.
18 #
19 # Performance in cycles per processed byte and improvement coefficient
20 # over code generated with "default" compiler:
21 #
22 #               hardware-assisted       software(*)
23 # Apple A7      2.31                    4.13 (+14%)
24 # Cortex-A53    2.24                    8.03 (+97%)
25 # Cortex-A57    2.35                    7.88 (+74%)
26 # Denver        2.13                    3.97 (+0%)(**)
27 # X-Gene                                8.80 (+200%)
28 # Mongoose      2.05                    6.50 (+160%)
29 # Kryo          1.88                    8.00 (+90%)
30 #
31 # (*)   Software results are presented mostly for reference purposes.
32 # (**)  Keep in mind that Denver relies on binary translation, which
33 #       optimizes compiler output at run-time.
34
35 $flavour = shift;
36 $output  = shift;
37
38 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
39 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
40 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
41 die "can't locate arm-xlate.pl";
42
43 open OUT,"| \"$^X\" $xlate $flavour $output";
44 *STDOUT=*OUT;
45
46 ($ctx,$inp,$num)=("x0","x1","x2");
47 @Xw=map("w$_",(3..17,19));
48 @Xx=map("x$_",(3..17,19));
49 @V=($A,$B,$C,$D,$E)=map("w$_",(20..24));
50 ($t0,$t1,$t2,$K)=map("w$_",(25..28));
51
52
53 sub BODY_00_19 {
54 my ($i,$a,$b,$c,$d,$e)=@_;
55 my $j=($i+2)&15;
56
57 $code.=<<___ if ($i<15 && !($i&1));
58         lsr     @Xx[$i+1],@Xx[$i],#32
59 ___
60 $code.=<<___ if ($i<14 && !($i&1));
61         ldr     @Xx[$i+2],[$inp,#`($i+2)*4-64`]
62 ___
63 $code.=<<___ if ($i<14 && ($i&1));
64 #ifdef  __ARMEB__
65         ror     @Xx[$i+1],@Xx[$i+1],#32
66 #else
67         rev32   @Xx[$i+1],@Xx[$i+1]
68 #endif
69 ___
70 $code.=<<___ if ($i<14);
71         bic     $t0,$d,$b
72         and     $t1,$c,$b
73         ror     $t2,$a,#27
74         add     $d,$d,$K                // future e+=K
75         orr     $t0,$t0,$t1
76         add     $e,$e,$t2               // e+=rot(a,5)
77         ror     $b,$b,#2
78         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
79         add     $e,$e,$t0               // e+=F(b,c,d)
80 ___
81 $code.=<<___ if ($i==19);
82         movz    $K,#0xeba1
83         movk    $K,#0x6ed9,lsl#16
84 ___
85 $code.=<<___ if ($i>=14);
86          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
87         bic     $t0,$d,$b
88         and     $t1,$c,$b
89         ror     $t2,$a,#27
90          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
91         add     $d,$d,$K                // future e+=K
92         orr     $t0,$t0,$t1
93         add     $e,$e,$t2               // e+=rot(a,5)
94          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
95         ror     $b,$b,#2
96         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
97         add     $e,$e,$t0               // e+=F(b,c,d)
98          ror    @Xw[$j],@Xw[$j],#31
99 ___
100 }
101
102 sub BODY_40_59 {
103 my ($i,$a,$b,$c,$d,$e)=@_;
104 my $j=($i+2)&15;
105
106 $code.=<<___ if ($i==59);
107         movz    $K,#0xc1d6
108         movk    $K,#0xca62,lsl#16
109 ___
110 $code.=<<___;
111         orr     $t0,$b,$c
112         and     $t1,$b,$c
113          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
114         ror     $t2,$a,#27
115         and     $t0,$t0,$d
116         add     $d,$d,$K                // future e+=K
117          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
118         add     $e,$e,$t2               // e+=rot(a,5)
119         orr     $t0,$t0,$t1
120         ror     $b,$b,#2
121          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
122         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
123         add     $e,$e,$t0               // e+=F(b,c,d)
124          ror    @Xw[$j],@Xw[$j],#31
125 ___
126 }
127
128 sub BODY_20_39 {
129 my ($i,$a,$b,$c,$d,$e)=@_;
130 my $j=($i+2)&15;
131
132 $code.=<<___ if ($i==39);
133         movz    $K,#0xbcdc
134         movk    $K,#0x8f1b,lsl#16
135 ___
136 $code.=<<___ if ($i<78);
137          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
138         eor     $t0,$d,$b
139         ror     $t2,$a,#27
140         add     $d,$d,$K                // future e+=K
141          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
142         eor     $t0,$t0,$c
143         add     $e,$e,$t2               // e+=rot(a,5)
144         ror     $b,$b,#2
145          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
146         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
147         add     $e,$e,$t0               // e+=F(b,c,d)
148          ror    @Xw[$j],@Xw[$j],#31
149 ___
150 $code.=<<___ if ($i==78);
151         ldp     @Xw[1],@Xw[2],[$ctx]
152         eor     $t0,$d,$b
153         ror     $t2,$a,#27
154         add     $d,$d,$K                // future e+=K
155         eor     $t0,$t0,$c
156         add     $e,$e,$t2               // e+=rot(a,5)
157         ror     $b,$b,#2
158         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
159         add     $e,$e,$t0               // e+=F(b,c,d)
160 ___
161 $code.=<<___ if ($i==79);
162         ldp     @Xw[3],@Xw[4],[$ctx,#8]
163         eor     $t0,$d,$b
164         ror     $t2,$a,#27
165         eor     $t0,$t0,$c
166         add     $e,$e,$t2               // e+=rot(a,5)
167         ror     $b,$b,#2
168         ldr     @Xw[5],[$ctx,#16]
169         add     $e,$e,$t0               // e+=F(b,c,d)
170 ___
171 }
172
173 $code.=<<___;
174 #include "arm_arch.h"
175
176 .text
177
178 .extern OPENSSL_armcap_P
179 .globl  sha1_block_data_order
180 .type   sha1_block_data_order,%function
181 .align  6
182 sha1_block_data_order:
183 #ifdef  __ILP32__
184         ldrsw   x16,.LOPENSSL_armcap_P
185 #else
186         ldr     x16,.LOPENSSL_armcap_P
187 #endif
188         adr     x17,.LOPENSSL_armcap_P
189         add     x16,x16,x17
190         ldr     w16,[x16]
191         tst     w16,#ARMV8_SHA1
192         b.ne    .Lv8_entry
193
194         stp     x29,x30,[sp,#-96]!
195         add     x29,sp,#0
196         stp     x19,x20,[sp,#16]
197         stp     x21,x22,[sp,#32]
198         stp     x23,x24,[sp,#48]
199         stp     x25,x26,[sp,#64]
200         stp     x27,x28,[sp,#80]
201
202         ldp     $A,$B,[$ctx]
203         ldp     $C,$D,[$ctx,#8]
204         ldr     $E,[$ctx,#16]
205
206 .Loop:
207         ldr     @Xx[0],[$inp],#64
208         movz    $K,#0x7999
209         sub     $num,$num,#1
210         movk    $K,#0x5a82,lsl#16
211 #ifdef  __ARMEB__
212         ror     $Xx[0],@Xx[0],#32
213 #else
214         rev32   @Xx[0],@Xx[0]
215 #endif
216         add     $E,$E,$K                // warm it up
217         add     $E,$E,@Xw[0]
218 ___
219 for($i=0;$i<20;$i++)    { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
220 for(;$i<40;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
221 for(;$i<60;$i++)        { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
222 for(;$i<80;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
223 $code.=<<___;
224         add     $B,$B,@Xw[2]
225         add     $C,$C,@Xw[3]
226         add     $A,$A,@Xw[1]
227         add     $D,$D,@Xw[4]
228         add     $E,$E,@Xw[5]
229         stp     $A,$B,[$ctx]
230         stp     $C,$D,[$ctx,#8]
231         str     $E,[$ctx,#16]
232         cbnz    $num,.Loop
233
234         ldp     x19,x20,[sp,#16]
235         ldp     x21,x22,[sp,#32]
236         ldp     x23,x24,[sp,#48]
237         ldp     x25,x26,[sp,#64]
238         ldp     x27,x28,[sp,#80]
239         ldr     x29,[sp],#96
240         ret
241 .size   sha1_block_data_order,.-sha1_block_data_order
242 ___
243 {{{
244 my ($ABCD,$E,$E0,$E1)=map("v$_.16b",(0..3));
245 my @MSG=map("v$_.16b",(4..7));
246 my @Kxx=map("v$_.4s",(16..19));
247 my ($W0,$W1)=("v20.4s","v21.4s");
248 my $ABCD_SAVE="v22.16b";
249
250 $code.=<<___;
251 .type   sha1_block_armv8,%function
252 .align  6
253 sha1_block_armv8:
254 .Lv8_entry:
255         stp     x29,x30,[sp,#-16]!
256         add     x29,sp,#0
257
258         adr     x4,.Lconst
259         eor     $E,$E,$E
260         ld1.32  {$ABCD},[$ctx],#16
261         ld1.32  {$E}[0],[$ctx]
262         sub     $ctx,$ctx,#16
263         ld1.32  {@Kxx[0]-@Kxx[3]},[x4]
264
265 .Loop_hw:
266         ld1     {@MSG[0]-@MSG[3]},[$inp],#64
267         sub     $num,$num,#1
268         rev32   @MSG[0],@MSG[0]
269         rev32   @MSG[1],@MSG[1]
270
271         add.i32 $W0,@Kxx[0],@MSG[0]
272         rev32   @MSG[2],@MSG[2]
273         orr     $ABCD_SAVE,$ABCD,$ABCD  // offload
274
275         add.i32 $W1,@Kxx[0],@MSG[1]
276         rev32   @MSG[3],@MSG[3]
277         sha1h   $E1,$ABCD
278         sha1c   $ABCD,$E,$W0            // 0
279         add.i32 $W0,@Kxx[$j],@MSG[2]
280         sha1su0 @MSG[0],@MSG[1],@MSG[2]
281 ___
282 for ($j=0,$i=1;$i<20-3;$i++) {
283 my $f=("c","p","m","p")[$i/5];
284 $code.=<<___;
285         sha1h   $E0,$ABCD               // $i
286         sha1$f  $ABCD,$E1,$W1
287         add.i32 $W1,@Kxx[$j],@MSG[3]
288         sha1su1 @MSG[0],@MSG[3]
289 ___
290 $code.=<<___ if ($i<20-4);
291         sha1su0 @MSG[1],@MSG[2],@MSG[3]
292 ___
293         ($E0,$E1)=($E1,$E0);            ($W0,$W1)=($W1,$W0);
294         push(@MSG,shift(@MSG));         $j++ if ((($i+3)%5)==0);
295 }
296 $code.=<<___;
297         sha1h   $E0,$ABCD               // $i
298         sha1p   $ABCD,$E1,$W1
299         add.i32 $W1,@Kxx[$j],@MSG[3]
300
301         sha1h   $E1,$ABCD               // 18
302         sha1p   $ABCD,$E0,$W0
303
304         sha1h   $E0,$ABCD               // 19
305         sha1p   $ABCD,$E1,$W1
306
307         add.i32 $E,$E,$E0
308         add.i32 $ABCD,$ABCD,$ABCD_SAVE
309
310         cbnz    $num,.Loop_hw
311
312         st1.32  {$ABCD},[$ctx],#16
313         st1.32  {$E}[0],[$ctx]
314
315         ldr     x29,[sp],#16
316         ret
317 .size   sha1_block_armv8,.-sha1_block_armv8
318 .align  6
319 .Lconst:
320 .long   0x5a827999,0x5a827999,0x5a827999,0x5a827999     //K_00_19
321 .long   0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1     //K_20_39
322 .long   0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc     //K_40_59
323 .long   0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6     //K_60_79
324 .LOPENSSL_armcap_P:
325 #ifdef  __ILP32__
326 .long   OPENSSL_armcap_P-.
327 #else
328 .quad   OPENSSL_armcap_P-.
329 #endif
330 .asciz  "SHA1 block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
331 .align  2
332 .comm   OPENSSL_armcap_P,4,4
333 ___
334 }}}
335
336 {   my  %opcode = (
337         "sha1c"         => 0x5e000000,  "sha1p"         => 0x5e001000,
338         "sha1m"         => 0x5e002000,  "sha1su0"       => 0x5e003000,
339         "sha1h"         => 0x5e280800,  "sha1su1"       => 0x5e281800   );
340
341     sub unsha1 {
342         my ($mnemonic,$arg)=@_;
343
344         $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
345         &&
346         sprintf ".inst\t0x%08x\t//%s %s",
347                         $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
348                         $mnemonic,$arg;
349     }
350 }
351
352 foreach(split("\n",$code)) {
353
354         s/\`([^\`]*)\`/eval($1)/geo;
355
356         s/\b(sha1\w+)\s+([qv].*)/unsha1($1,$2)/geo;
357
358         s/\.\w?32\b//o          and s/\.16b/\.4s/go;
359         m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go;
360
361         print $_,"\n";
362 }
363
364 close STDOUT;