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 Apache License 2.0 (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 # ThunderX2     2.64                    6.36 (+150%)
  31 #
  32 # (*)   Software results are presented mostly for reference purposes.
  33 # (**)  Keep in mind that Denver relies on binary translation, which
  34 #       optimizes compiler output at run-time.
  35
  36 $flavour = shift;
  37 $output  = shift;
  38
  39 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  40 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
  41 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
  42 die "can't locate arm-xlate.pl";
  43
  44 open OUT,"| \"$^X\" $xlate $flavour $output";
  45 *STDOUT=*OUT;
  46
  47 ($ctx,$inp,$num)=("x0","x1","x2");
  48 @Xw=map("w$_",(3..17,19));
  49 @Xx=map("x$_",(3..17,19));
  50 @V=($A,$B,$C,$D,$E)=map("w$_",(20..24));
  51 ($t0,$t1,$t2,$K)=map("w$_",(25..28));
  52
  53
  54 sub BODY_00_19 {
  55 my ($i,$a,$b,$c,$d,$e)=@_;
  56 my $j=($i+2)&15;
  57
  58 $code.=<<___ if ($i<15 && !($i&1));
  59         lsr     @Xx[$i+1],@Xx[$i],#32
  60 ___
  61 $code.=<<___ if ($i<14 && !($i&1));
  62         ldur    @Xx[$i+2],[$inp,#`($i+2)*4-64`]
  63 ___
  64 $code.=<<___ if ($i<14 && ($i&1));
  65 #ifdef  __AARCH64EB__
  66         ror     @Xx[$i+1],@Xx[$i+1],#32
  67 #else
  68         rev32   @Xx[$i+1],@Xx[$i+1]
  69 #endif
  70 ___
  71 $code.=<<___ if ($i<14);
  72         bic     $t0,$d,$b
  73         and     $t1,$c,$b
  74         ror     $t2,$a,#27
  75         add     $d,$d,$K                // future e+=K
  76         orr     $t0,$t0,$t1
  77         add     $e,$e,$t2               // e+=rot(a,5)
  78         ror     $b,$b,#2
  79         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
  80         add     $e,$e,$t0               // e+=F(b,c,d)
  81 ___
  82 $code.=<<___ if ($i==19);
  83         movz    $K,#0xeba1
  84         movk    $K,#0x6ed9,lsl#16
  85 ___
  86 $code.=<<___ if ($i>=14);
  87          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
  88         bic     $t0,$d,$b
  89         and     $t1,$c,$b
  90         ror     $t2,$a,#27
  91          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
  92         add     $d,$d,$K                // future e+=K
  93         orr     $t0,$t0,$t1
  94         add     $e,$e,$t2               // e+=rot(a,5)
  95          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
  96         ror     $b,$b,#2
  97         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
  98         add     $e,$e,$t0               // e+=F(b,c,d)
  99          ror    @Xw[$j],@Xw[$j],#31
 100 ___
 101 }
 102
 103 sub BODY_40_59 {
 104 my ($i,$a,$b,$c,$d,$e)=@_;
 105 my $j=($i+2)&15;
 106
 107 $code.=<<___ if ($i==59);
 108         movz    $K,#0xc1d6
 109         movk    $K,#0xca62,lsl#16
 110 ___
 111 $code.=<<___;
 112         orr     $t0,$b,$c
 113         and     $t1,$b,$c
 114          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
 115         ror     $t2,$a,#27
 116         and     $t0,$t0,$d
 117         add     $d,$d,$K                // future e+=K
 118          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
 119         add     $e,$e,$t2               // e+=rot(a,5)
 120         orr     $t0,$t0,$t1
 121         ror     $b,$b,#2
 122          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
 123         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
 124         add     $e,$e,$t0               // e+=F(b,c,d)
 125          ror    @Xw[$j],@Xw[$j],#31
 126 ___
 127 }
 128
 129 sub BODY_20_39 {
 130 my ($i,$a,$b,$c,$d,$e)=@_;
 131 my $j=($i+2)&15;
 132
 133 $code.=<<___ if ($i==39);
 134         movz    $K,#0xbcdc
 135         movk    $K,#0x8f1b,lsl#16
 136 ___
 137 $code.=<<___ if ($i<78);
 138          eor    @Xw[$j],@Xw[$j],@Xw[($j+2)&15]
 139         eor     $t0,$d,$b
 140         ror     $t2,$a,#27
 141         add     $d,$d,$K                // future e+=K
 142          eor    @Xw[$j],@Xw[$j],@Xw[($j+8)&15]
 143         eor     $t0,$t0,$c
 144         add     $e,$e,$t2               // e+=rot(a,5)
 145         ror     $b,$b,#2
 146          eor    @Xw[$j],@Xw[$j],@Xw[($j+13)&15]
 147         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
 148         add     $e,$e,$t0               // e+=F(b,c,d)
 149          ror    @Xw[$j],@Xw[$j],#31
 150 ___
 151 $code.=<<___ if ($i==78);
 152         ldp     @Xw[1],@Xw[2],[$ctx]
 153         eor     $t0,$d,$b
 154         ror     $t2,$a,#27
 155         add     $d,$d,$K                // future e+=K
 156         eor     $t0,$t0,$c
 157         add     $e,$e,$t2               // e+=rot(a,5)
 158         ror     $b,$b,#2
 159         add     $d,$d,@Xw[($i+1)&15]    // future e+=X[i]
 160         add     $e,$e,$t0               // e+=F(b,c,d)
 161 ___
 162 $code.=<<___ if ($i==79);
 163         ldp     @Xw[3],@Xw[4],[$ctx,#8]
 164         eor     $t0,$d,$b
 165         ror     $t2,$a,#27
 166         eor     $t0,$t0,$c
 167         add     $e,$e,$t2               // e+=rot(a,5)
 168         ror     $b,$b,#2
 169         ldr     @Xw[5],[$ctx,#16]
 170         add     $e,$e,$t0               // e+=F(b,c,d)
 171 ___
 172 }
 173
 174 $code.=<<___;
 175 #ifndef __KERNEL__
 176 # include "arm_arch.h"
 177 .extern OPENSSL_armcap_P
 178 #endif
 179
 180 .text
 181
 182 .globl  sha1_block_data_order
 183 .type   sha1_block_data_order,%function
 184 .align  6
 185 sha1_block_data_order:
 186         adrp    x16,OPENSSL_armcap_P
 187         ldr     w16,[x16,#:lo12:OPENSSL_armcap_P]
 188         tst     w16,#ARMV8_SHA1
 189         b.ne    .Lv8_entry
 190
 191         stp     x29,x30,[sp,#-96]!
 192         add     x29,sp,#0
 193         stp     x19,x20,[sp,#16]
 194         stp     x21,x22,[sp,#32]
 195         stp     x23,x24,[sp,#48]
 196         stp     x25,x26,[sp,#64]
 197         stp     x27,x28,[sp,#80]
 198
 199         ldp     $A,$B,[$ctx]
 200         ldp     $C,$D,[$ctx,#8]
 201         ldr     $E,[$ctx,#16]
 202
 203 .Loop:
 204         ldr     @Xx[0],[$inp],#64
 205         movz    $K,#0x7999
 206         sub     $num,$num,#1
 207         movk    $K,#0x5a82,lsl#16
 208 #ifdef  __AARCH64EB__
 209         ror     $Xx[0],@Xx[0],#32
 210 #else
 211         rev32   @Xx[0],@Xx[0]
 212 #endif
 213         add     $E,$E,$K                // warm it up
 214         add     $E,$E,@Xw[0]
 215 ___
 216 for($i=0;$i<20;$i++)    { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
 217 for(;$i<40;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
 218 for(;$i<60;$i++)        { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
 219 for(;$i<80;$i++)        { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
 220 $code.=<<___;
 221         add     $B,$B,@Xw[2]
 222         add     $C,$C,@Xw[3]
 223         add     $A,$A,@Xw[1]
 224         add     $D,$D,@Xw[4]
 225         add     $E,$E,@Xw[5]
 226         stp     $A,$B,[$ctx]
 227         stp     $C,$D,[$ctx,#8]
 228         str     $E,[$ctx,#16]
 229         cbnz    $num,.Loop
 230
 231         ldp     x19,x20,[sp,#16]
 232         ldp     x21,x22,[sp,#32]
 233         ldp     x23,x24,[sp,#48]
 234         ldp     x25,x26,[sp,#64]
 235         ldp     x27,x28,[sp,#80]
 236         ldr     x29,[sp],#96
 237         ret
 238 .size   sha1_block_data_order,.-sha1_block_data_order
 239 ___
 240 {{{
 241 my ($ABCD,$E,$E0,$E1)=map("v$_.16b",(0..3));
 242 my @MSG=map("v$_.16b",(4..7));
 243 my @Kxx=map("v$_.4s",(16..19));
 244 my ($W0,$W1)=("v20.4s","v21.4s");
 245 my $ABCD_SAVE="v22.16b";
 246
 247 $code.=<<___;
 248 .type   sha1_block_armv8,%function
 249 .align  6
 250 sha1_block_armv8:
 251 .Lv8_entry:
 252         stp     x29,x30,[sp,#-16]!
 253         add     x29,sp,#0
 254
 255         adr     x4,.Lconst
 256         eor     $E,$E,$E
 257         ld1.32  {$ABCD},[$ctx],#16
 258         ld1.32  {$E}[0],[$ctx]
 259         sub     $ctx,$ctx,#16
 260         ld1.32  {@Kxx[0]-@Kxx[3]},[x4]
 261
 262 .Loop_hw:
 263         ld1     {@MSG[0]-@MSG[3]},[$inp],#64
 264         sub     $num,$num,#1
 265         rev32   @MSG[0],@MSG[0]
 266         rev32   @MSG[1],@MSG[1]
 267
 268         add.i32 $W0,@Kxx[0],@MSG[0]
 269         rev32   @MSG[2],@MSG[2]
 270         orr     $ABCD_SAVE,$ABCD,$ABCD  // offload
 271
 272         add.i32 $W1,@Kxx[0],@MSG[1]
 273         rev32   @MSG[3],@MSG[3]
 274         sha1h   $E1,$ABCD
 275         sha1c   $ABCD,$E,$W0            // 0
 276         add.i32 $W0,@Kxx[$j],@MSG[2]
 277         sha1su0 @MSG[0],@MSG[1],@MSG[2]
 278 ___
 279 for ($j=0,$i=1;$i<20-3;$i++) {
 280 my $f=("c","p","m","p")[$i/5];
 281 $code.=<<___;
 282         sha1h   $E0,$ABCD               // $i
 283         sha1$f  $ABCD,$E1,$W1
 284         add.i32 $W1,@Kxx[$j],@MSG[3]
 285         sha1su1 @MSG[0],@MSG[3]
 286 ___
 287 $code.=<<___ if ($i<20-4);
 288         sha1su0 @MSG[1],@MSG[2],@MSG[3]
 289 ___
 290         ($E0,$E1)=($E1,$E0);            ($W0,$W1)=($W1,$W0);
 291         push(@MSG,shift(@MSG));         $j++ if ((($i+3)%5)==0);
 292 }
 293 $code.=<<___;
 294         sha1h   $E0,$ABCD               // $i
 295         sha1p   $ABCD,$E1,$W1
 296         add.i32 $W1,@Kxx[$j],@MSG[3]
 297
 298         sha1h   $E1,$ABCD               // 18
 299         sha1p   $ABCD,$E0,$W0
 300
 301         sha1h   $E0,$ABCD               // 19
 302         sha1p   $ABCD,$E1,$W1
 303
 304         add.i32 $E,$E,$E0
 305         add.i32 $ABCD,$ABCD,$ABCD_SAVE
 306
 307         cbnz    $num,.Loop_hw
 308
 309         st1.32  {$ABCD},[$ctx],#16
 310         st1.32  {$E}[0],[$ctx]
 311
 312         ldr     x29,[sp],#16
 313         ret
 314 .size   sha1_block_armv8,.-sha1_block_armv8
 315 .align  6
 316 .Lconst:
 317 .long   0x5a827999,0x5a827999,0x5a827999,0x5a827999     //K_00_19
 318 .long   0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1     //K_20_39
 319 .long   0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc     //K_40_59
 320 .long   0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6     //K_60_79
 321 .asciz  "SHA1 block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
 322 .align  2
 323 #if !defined(__KERNELL__) && !defined(_WIN64)
 324 .comm   OPENSSL_armcap_P,4,4
 325 #endif
 326 ___
 327 }}}
 328
 329 {   my  %opcode = (
 330         "sha1c"         => 0x5e000000,  "sha1p"         => 0x5e001000,
 331         "sha1m"         => 0x5e002000,  "sha1su0"       => 0x5e003000,
 332         "sha1h"         => 0x5e280800,  "sha1su1"       => 0x5e281800   );
 333
 334     sub unsha1 {
 335         my ($mnemonic,$arg)=@_;
 336
 337         $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
 338         &&
 339         sprintf ".inst\t0x%08x\t//%s %s",
 340                         $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
 341                         $mnemonic,$arg;
 342     }
 343 }
 344
 345 foreach(split("\n",$code)) {
 346
 347         s/\`([^\`]*)\`/eval($1)/geo;
 348
 349         s/\b(sha1\w+)\s+([qv].*)/unsha1($1,$2)/geo;
 350
 351         s/\.\w?32\b//o          and s/\.16b/\.4s/go;
 352         m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go;
 353
 354         print $_,"\n";
 355 }
 356
 357 close STDOUT;