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$flavour = shift; 27open STDOUT,">".shift; 28 29($ctx,$inp,$num)=("x0","x1","x2"); 30@Xw=map("w$_",(3..17,19)); 31@Xx=map("x$_",(3..17,19)); 32@V=($A,$B,$C,$D,$E)=map("w$_",(20..24)); 33($t0,$t1,$t2,$K)=map("w$_",(25..28)); 34 35 36sub BODY_00_19 { 37my ($i,$a,$b,$c,$d,$e)=@_; 38my $j=($i+2)&15; 39 40$code.=<<___ if ($i<15 && !($i&1)); 41 lsr @Xx[$i+1],@Xx[$i],#32 42___ 43$code.=<<___ if ($i<14 && !($i&1)); 44 ldr @Xx[$i+2],[$inp,#`($i+2)*4-64`] 45___ 46$code.=<<___ if ($i<14 && ($i&1)); 47#ifdef __ARMEB__ 48 ror @Xx[$i+1],@Xx[$i+1],#32 49#else 50 rev32 @Xx[$i+1],@Xx[$i+1] 51#endif 52___ 53$code.=<<___ if ($i<14); 54 bic $t0,$d,$b 55 and $t1,$c,$b 56 ror $t2,$a,#27 57 add $d,$d,$K // future e+=K 58 orr $t0,$t0,$t1 59 add $e,$e,$t2 // e+=rot(a,5) 60 ror $b,$b,#2 61 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 62 add $e,$e,$t0 // e+=F(b,c,d) 63___ 64$code.=<<___ if ($i==19); 65 movz $K,#0xeba1 66 movk $K,#0x6ed9,lsl#16 67___ 68$code.=<<___ if ($i>=14); 69 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 70 bic $t0,$d,$b 71 and $t1,$c,$b 72 ror $t2,$a,#27 73 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 74 add $d,$d,$K // future e+=K 75 orr $t0,$t0,$t1 76 add $e,$e,$t2 // e+=rot(a,5) 77 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 78 ror $b,$b,#2 79 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 80 add $e,$e,$t0 // e+=F(b,c,d) 81 ror @Xw[$j],@Xw[$j],#31 82___ 83} 84 85sub BODY_40_59 { 86my ($i,$a,$b,$c,$d,$e)=@_; 87my $j=($i+2)&15; 88 89$code.=<<___ if ($i==59); 90 movz $K,#0xc1d6 91 movk $K,#0xca62,lsl#16 92___ 93$code.=<<___; 94 orr $t0,$b,$c 95 and $t1,$b,$c 96 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 97 ror $t2,$a,#27 98 and $t0,$t0,$d 99 add $d,$d,$K // future e+=K 100 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 101 add $e,$e,$t2 // e+=rot(a,5) 102 orr $t0,$t0,$t1 103 ror $b,$b,#2 104 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 105 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 106 add $e,$e,$t0 // e+=F(b,c,d) 107 ror @Xw[$j],@Xw[$j],#31 108___ 109} 110 111sub BODY_20_39 { 112my ($i,$a,$b,$c,$d,$e)=@_; 113my $j=($i+2)&15; 114 115$code.=<<___ if ($i==39); 116 movz $K,#0xbcdc 117 movk $K,#0x8f1b,lsl#16 118___ 119$code.=<<___ if ($i<78); 120 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 121 eor $t0,$d,$b 122 ror $t2,$a,#27 123 add $d,$d,$K // future e+=K 124 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 125 eor $t0,$t0,$c 126 add $e,$e,$t2 // e+=rot(a,5) 127 ror $b,$b,#2 128 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 129 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 130 add $e,$e,$t0 // e+=F(b,c,d) 131 ror @Xw[$j],@Xw[$j],#31 132___ 133$code.=<<___ if ($i==78); 134 ldp @Xw[1],@Xw[2],[$ctx] 135 eor $t0,$d,$b 136 ror $t2,$a,#27 137 add $d,$d,$K // future e+=K 138 eor $t0,$t0,$c 139 add $e,$e,$t2 // e+=rot(a,5) 140 ror $b,$b,#2 141 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 142 add $e,$e,$t0 // e+=F(b,c,d) 143___ 144$code.=<<___ if ($i==79); 145 ldp @Xw[3],@Xw[4],[$ctx,#8] 146 eor $t0,$d,$b 147 ror $t2,$a,#27 148 eor $t0,$t0,$c 149 add $e,$e,$t2 // e+=rot(a,5) 150 ror $b,$b,#2 151 ldr @Xw[5],[$ctx,#16] 152 add $e,$e,$t0 // e+=F(b,c,d) 153___ 154} 155 156$code.=<<___; 157#include "arm_arch.h" 158 159.text 160 161.globl sha1_block_data_order 162.type sha1_block_data_order,%function 163.align 6 164sha1_block_data_order: 165 ldr x16,.LOPENSSL_armcap_P 166 adr x17,.LOPENSSL_armcap_P 167 add x16,x16,x17 168 ldr w16,[x16] 169 tst w16,#ARMV8_SHA1 170 b.ne .Lv8_entry 171 172 stp x29,x30,[sp,#-96]! 173 add x29,sp,#0 174 stp x19,x20,[sp,#16] 175 stp x21,x22,[sp,#32] 176 stp x23,x24,[sp,#48] 177 stp x25,x26,[sp,#64] 178 stp x27,x28,[sp,#80] 179 180 ldp $A,$B,[$ctx] 181 ldp $C,$D,[$ctx,#8] 182 ldr $E,[$ctx,#16] 183 184.Loop: 185 ldr @Xx[0],[$inp],#64 186 movz $K,#0x7999 187 sub $num,$num,#1 188 movk $K,#0x5a82,lsl#16 189#ifdef __ARMEB__ 190 ror $Xx[0],@Xx[0],#32 191#else 192 rev32 @Xx[0],@Xx[0] 193#endif 194 add $E,$E,$K // warm it up 195 add $E,$E,@Xw[0] 196___ 197for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); } 198for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } 199for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); } 200for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } 201$code.=<<___; 202 add $B,$B,@Xw[2] 203 add $C,$C,@Xw[3] 204 add $A,$A,@Xw[1] 205 add $D,$D,@Xw[4] 206 add $E,$E,@Xw[5] 207 stp $A,$B,[$ctx] 208 stp $C,$D,[$ctx,#8] 209 str $E,[$ctx,#16] 210 cbnz $num,.Loop 211 212 ldp x19,x20,[sp,#16] 213 ldp x21,x22,[sp,#32] 214 ldp x23,x24,[sp,#48] 215 ldp x25,x26,[sp,#64] 216 ldp x27,x28,[sp,#80] 217 ldr x29,[sp],#96 218 ret 219.size sha1_block_data_order,.-sha1_block_data_order 220___ 221{{{ 222my ($ABCD,$E,$E0,$E1)=map("v$_.16b",(0..3)); 223my @MSG=map("v$_.16b",(4..7)); 224my @Kxx=map("v$_.4s",(16..19)); 225my ($W0,$W1)=("v20.4s","v21.4s"); 226my $ABCD_SAVE="v22.16b"; 227 228$code.=<<___; 229.type sha1_block_armv8,%function 230.align 6 231sha1_block_armv8: 232.Lv8_entry: 233 stp x29,x30,[sp,#-16]! 234 add x29,sp,#0 235 236 adr x4,.Lconst 237 eor $E,$E,$E 238 ld1.32 {$ABCD},[$ctx],#16 239 ld1.32 {$E}[0],[$ctx] 240 sub $ctx,$ctx,#16 241 ld1.32 {@Kxx[0]-@Kxx[3]},[x4] 242 243.Loop_hw: 244 ld1 {@MSG[0]-@MSG[3]},[$inp],#64 245 sub $num,$num,#1 246 rev32 @MSG[0],@MSG[0] 247 rev32 @MSG[1],@MSG[1] 248 249 add.i32 $W0,@Kxx[0],@MSG[0] 250 rev32 @MSG[2],@MSG[2] 251 orr $ABCD_SAVE,$ABCD,$ABCD // offload 252 253 add.i32 $W1,@Kxx[0],@MSG[1] 254 rev32 @MSG[3],@MSG[3] 255 sha1h $E1,$ABCD 256 sha1c $ABCD,$E,$W0 // 0 257 add.i32 $W0,@Kxx[$j],@MSG[2] 258 sha1su0 @MSG[0],@MSG[1],@MSG[2] 259___ 260for ($j=0,$i=1;$i<20-3;$i++) { 261my $f=("c","p","m","p")[$i/5]; 262$code.=<<___; 263 sha1h $E0,$ABCD // $i 264 sha1$f $ABCD,$E1,$W1 265 add.i32 $W1,@Kxx[$j],@MSG[3] 266 sha1su1 @MSG[0],@MSG[3] 267___ 268$code.=<<___ if ($i<20-4); 269 sha1su0 @MSG[1],@MSG[2],@MSG[3] 270___ 271 ($E0,$E1)=($E1,$E0); ($W0,$W1)=($W1,$W0); 272 push(@MSG,shift(@MSG)); $j++ if ((($i+3)%5)==0); 273} 274$code.=<<___; 275 sha1h $E0,$ABCD // $i 276 sha1p $ABCD,$E1,$W1 277 add.i32 $W1,@Kxx[$j],@MSG[3] 278 279 sha1h $E1,$ABCD // 18 280 sha1p $ABCD,$E0,$W0 281 282 sha1h $E0,$ABCD // 19 283 sha1p $ABCD,$E1,$W1 284 285 add.i32 $E,$E,$E0 286 add.i32 $ABCD,$ABCD,$ABCD_SAVE 287 288 cbnz $num,.Loop_hw 289 290 st1.32 {$ABCD},[$ctx],#16 291 st1.32 {$E}[0],[$ctx] 292 293 ldr x29,[sp],#16 294 ret 295.size sha1_block_armv8,.-sha1_block_armv8 296.align 6 297.Lconst: 298.long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 //K_00_19 299.long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 //K_20_39 300.long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc //K_40_59 301.long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 //K_60_79 302.LOPENSSL_armcap_P: 303.quad OPENSSL_armcap_P-. 304.asciz "SHA1 block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>" 305.align 2 306.comm OPENSSL_armcap_P,4,4 307___ 308}}} 309 310{ my %opcode = ( 311 "sha1c" => 0x5e000000, "sha1p" => 0x5e001000, 312 "sha1m" => 0x5e002000, "sha1su0" => 0x5e003000, 313 "sha1h" => 0x5e280800, "sha1su1" => 0x5e281800 ); 314 315 sub unsha1 { 316 my ($mnemonic,$arg)=@_; 317 318 $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o 319 && 320 sprintf ".inst\t0x%08x\t//%s %s", 321 $opcode{$mnemonic}|$1|($2<<5)|($3<<16), 322 $mnemonic,$arg; 323 } 324} 325 326foreach(split("\n",$code)) { 327 328 s/\`([^\`]*)\`/eval($1)/geo; 329 330 s/\b(sha1\w+)\s+([qv].*)/unsha1($1,$2)/geo; 331 332 s/\.\w?32\b//o and s/\.16b/\.4s/go; 333 m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go; 334 335 print $_,"\n"; 336} 337 338close STDOUT; 339