crypto/sha/asm/sha1-armv8.pl

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