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