| 1 |
#!/usr/bin/env perl |
| 2 |
# |
| 3 |
# ==================================================================== |
| 4 |
# Written by Andy Polyakov <[email protected]> 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 <openssl/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 |
ldr x16,.LOPENSSL_armcap_P |
| 175 |
adr x17,.LOPENSSL_armcap_P |
| 176 |
add x16,x16,x17 |
| 177 |
ldr w16,[x16] |
| 178 |
tst w16,#ARMV8_SHA1 |
| 179 |
b.ne .Lv8_entry |
| 180 |
|
| 181 |
stp x29,x30,[sp,#-96]! |
| 182 |
add x29,sp,#0 |
| 183 |
stp x19,x20,[sp,#16] |
| 184 |
stp x21,x22,[sp,#32] |
| 185 |
stp x23,x24,[sp,#48] |
| 186 |
stp x25,x26,[sp,#64] |
| 187 |
stp x27,x28,[sp,#80] |
| 188 |
|
| 189 |
ldp $A,$B,[$ctx] |
| 190 |
ldp $C,$D,[$ctx,#8] |
| 191 |
ldr $E,[$ctx,#16] |
| 192 |
|
| 193 |
.Loop: |
| 194 |
ldr @Xx[0],[$inp],#64 |
| 195 |
movz $K,#0x7999 |
| 196 |
sub $num,$num,#1 |
| 197 |
movk $K,#0x5a82,lsl#16 |
| 198 |
#ifdef __ARMEB__ |
| 199 |
ror $Xx[0],@Xx[0],#32 |
| 200 |
#else |
| 201 |
rev32 @Xx[0],@Xx[0] |
| 202 |
#endif |
| 203 |
add $E,$E,$K // warm it up |
| 204 |
add $E,$E,@Xw[0] |
| 205 |
___ |
| 206 |
for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); } |
| 207 |
for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } |
| 208 |
for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); } |
| 209 |
for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } |
| 210 |
$code.=<<___; |
| 211 |
add $B,$B,@Xw[2] |
| 212 |
add $C,$C,@Xw[3] |
| 213 |
add $A,$A,@Xw[1] |
| 214 |
add $D,$D,@Xw[4] |
| 215 |
add $E,$E,@Xw[5] |
| 216 |
stp $A,$B,[$ctx] |
| 217 |
stp $C,$D,[$ctx,#8] |
| 218 |
str $E,[$ctx,#16] |
| 219 |
cbnz $num,.Loop |
| 220 |
|
| 221 |
ldp x19,x20,[sp,#16] |
| 222 |
ldp x21,x22,[sp,#32] |
| 223 |
ldp x23,x24,[sp,#48] |
| 224 |
ldp x25,x26,[sp,#64] |
| 225 |
ldp x27,x28,[sp,#80] |
| 226 |
ldr x29,[sp],#96 |
| 227 |
ret |
| 228 |
.size sha1_block_data_order,.-sha1_block_data_order |
| 229 |
___ |
| 230 |
{{{ |
| 231 |
my ($ABCD,$E,$E0,$E1)=map("v$_.16b",(0..3)); |
| 232 |
my @MSG=map("v$_.16b",(4..7)); |
| 233 |
my @Kxx=map("v$_.4s",(16..19)); |
| 234 |
my ($W0,$W1)=("v20.4s","v21.4s"); |
| 235 |
my $ABCD_SAVE="v22.16b"; |
| 236 |
|
| 237 |
$code.=<<___; |
| 238 |
.type sha1_block_armv8,%function |
| 239 |
.align 6 |
| 240 |
sha1_block_armv8: |
| 241 |
.Lv8_entry: |
| 242 |
stp x29,x30,[sp,#-16]! |
| 243 |
add x29,sp,#0 |
| 244 |
|
| 245 |
adr x4,.Lconst |
| 246 |
eor $E,$E,$E |
| 247 |
ld1.32 {$ABCD},[$ctx],#16 |
| 248 |
ld1.32 {$E}[0],[$ctx] |
| 249 |
sub $ctx,$ctx,#16 |
| 250 |
ld1.32 {@Kxx[0]-@Kxx[3]},[x4] |
| 251 |
|
| 252 |
.Loop_hw: |
| 253 |
ld1 {@MSG[0]-@MSG[3]},[$inp],#64 |
| 254 |
sub $num,$num,#1 |
| 255 |
rev32 @MSG[0],@MSG[0] |
| 256 |
rev32 @MSG[1],@MSG[1] |
| 257 |
|
| 258 |
add.i32 $W0,@Kxx[0],@MSG[0] |
| 259 |
rev32 @MSG[2],@MSG[2] |
| 260 |
orr $ABCD_SAVE,$ABCD,$ABCD // offload |
| 261 |
|
| 262 |
add.i32 $W1,@Kxx[0],@MSG[1] |
| 263 |
rev32 @MSG[3],@MSG[3] |
| 264 |
sha1h $E1,$ABCD |
| 265 |
sha1c $ABCD,$E,$W0 // 0 |
| 266 |
add.i32 $W0,@Kxx[$j],@MSG[2] |
| 267 |
sha1su0 @MSG[0],@MSG[1],@MSG[2] |
| 268 |
___ |
| 269 |
for ($j=0,$i=1;$i<20-3;$i++) { |
| 270 |
my $f=("c","p","m","p")[$i/5]; |
| 271 |
$code.=<<___; |
| 272 |
sha1h $E0,$ABCD // $i |
| 273 |
sha1$f $ABCD,$E1,$W1 |
| 274 |
add.i32 $W1,@Kxx[$j],@MSG[3] |
| 275 |
sha1su1 @MSG[0],@MSG[3] |
| 276 |
___ |
| 277 |
$code.=<<___ if ($i<20-4); |
| 278 |
sha1su0 @MSG[1],@MSG[2],@MSG[3] |
| 279 |
___ |
| 280 |
($E0,$E1)=($E1,$E0); ($W0,$W1)=($W1,$W0); |
| 281 |
push(@MSG,shift(@MSG)); $j++ if ((($i+3)%5)==0); |
| 282 |
} |
| 283 |
$code.=<<___; |
| 284 |
sha1h $E0,$ABCD // $i |
| 285 |
sha1p $ABCD,$E1,$W1 |
| 286 |
add.i32 $W1,@Kxx[$j],@MSG[3] |
| 287 |
|
| 288 |
sha1h $E1,$ABCD // 18 |
| 289 |
sha1p $ABCD,$E0,$W0 |
| 290 |
|
| 291 |
sha1h $E0,$ABCD // 19 |
| 292 |
sha1p $ABCD,$E1,$W1 |
| 293 |
|
| 294 |
add.i32 $E,$E,$E0 |
| 295 |
add.i32 $ABCD,$ABCD,$ABCD_SAVE |
| 296 |
|
| 297 |
cbnz $num,.Loop_hw |
| 298 |
|
| 299 |
st1.32 {$ABCD},[$ctx],#16 |
| 300 |
st1.32 {$E}[0],[$ctx] |
| 301 |
|
| 302 |
ldr x29,[sp],#16 |
| 303 |
ret |
| 304 |
.size sha1_block_armv8,.-sha1_block_armv8 |
| 305 |
.align 6 |
| 306 |
.Lconst: |
| 307 |
.long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 //K_00_19 |
| 308 |
.long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 //K_20_39 |
| 309 |
.long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc //K_40_59 |
| 310 |
.long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 //K_60_79 |
| 311 |
.LOPENSSL_armcap_P: |
| 312 |
.quad OPENSSL_armcap_P-. |
| 313 |
.asciz "SHA1 block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>" |
| 314 |
.align 2 |
| 315 |
.comm OPENSSL_armcap_P,4,4 |
| 316 |
___ |
| 317 |
}}} |
| 318 |
|
| 319 |
{ my %opcode = ( |
| 320 |
"sha1c" => 0x5e000000, "sha1p" => 0x5e001000, |
| 321 |
"sha1m" => 0x5e002000, "sha1su0" => 0x5e003000, |
| 322 |
"sha1h" => 0x5e280800, "sha1su1" => 0x5e281800 ); |
| 323 |
|
| 324 |
sub unsha1 { |
| 325 |
my ($mnemonic,$arg)=@_; |
| 326 |
|
| 327 |
$arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o |
| 328 |
&& |
| 329 |
sprintf ".inst\t0x%08x\t//%s %s", |
| 330 |
$opcode{$mnemonic}|$1|($2<<5)|($3<<16), |
| 331 |
$mnemonic,$arg; |
| 332 |
} |
| 333 |
} |
| 334 |
|
| 335 |
foreach(split("\n",$code)) { |
| 336 |
|
| 337 |
s/\`([^\`]*)\`/eval($1)/geo; |
| 338 |
|
| 339 |
s/\b(sha1\w+)\s+([qv].*)/unsha1($1,$2)/geo; |
| 340 |
|
| 341 |
s/\.\w?32\b//o and s/\.16b/\.4s/go; |
| 342 |
m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go; |
| 343 |
|
| 344 |
print $_,"\n"; |
| 345 |
} |
| 346 |
|
| 347 |
close STDOUT; |