1#! /usr/bin/env perl 2# Copyright 2014-2016 The OpenSSL Project Authors. All Rights Reserved. 3# 4# Licensed under the OpenSSL license (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# 31# (*) Software results are presented mostly for reference purposes. 32# (**) Keep in mind that Denver relies on binary translation, which 33# optimizes compiler output at run-time. 34 35$flavour = shift; 36$output = shift; 37 38$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 39( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or 40( $xlate="${dir}../../../perlasm/arm-xlate.pl" and -f $xlate) or 41die "can't locate arm-xlate.pl"; 42 43open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; 44*STDOUT=*OUT; 45 46($ctx,$inp,$num)=("x0","x1","x2"); 47@Xw=map("w$_",(3..17,19)); 48@Xx=map("x$_",(3..17,19)); 49@V=($A,$B,$C,$D,$E)=map("w$_",(20..24)); 50($t0,$t1,$t2,$K)=map("w$_",(25..28)); 51 52 53sub BODY_00_19 { 54my ($i,$a,$b,$c,$d,$e)=@_; 55my $j=($i+2)&15; 56 57$code.=<<___ if ($i<15 && !($i&1)); 58 lsr @Xx[$i+1],@Xx[$i],#32 59___ 60$code.=<<___ if ($i<14 && !($i&1)); 61 ldr @Xx[$i+2],[$inp,#`($i+2)*4-64`] 62___ 63$code.=<<___ if ($i<14 && ($i&1)); 64#ifdef __AARCH64EB__ 65 ror @Xx[$i+1],@Xx[$i+1],#32 66#else 67 rev32 @Xx[$i+1],@Xx[$i+1] 68#endif 69___ 70$code.=<<___ if ($i<14); 71 bic $t0,$d,$b 72 and $t1,$c,$b 73 ror $t2,$a,#27 74 add $d,$d,$K // future e+=K 75 orr $t0,$t0,$t1 76 add $e,$e,$t2 // e+=rot(a,5) 77 ror $b,$b,#2 78 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 79 add $e,$e,$t0 // e+=F(b,c,d) 80___ 81$code.=<<___ if ($i==19); 82 movz $K,#0xeba1 83 movk $K,#0x6ed9,lsl#16 84___ 85$code.=<<___ if ($i>=14); 86 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 87 bic $t0,$d,$b 88 and $t1,$c,$b 89 ror $t2,$a,#27 90 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 91 add $d,$d,$K // future e+=K 92 orr $t0,$t0,$t1 93 add $e,$e,$t2 // e+=rot(a,5) 94 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 95 ror $b,$b,#2 96 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 97 add $e,$e,$t0 // e+=F(b,c,d) 98 ror @Xw[$j],@Xw[$j],#31 99___ 100} 101 102sub BODY_40_59 { 103my ($i,$a,$b,$c,$d,$e)=@_; 104my $j=($i+2)&15; 105 106$code.=<<___ if ($i==59); 107 movz $K,#0xc1d6 108 movk $K,#0xca62,lsl#16 109___ 110$code.=<<___; 111 orr $t0,$b,$c 112 and $t1,$b,$c 113 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 114 ror $t2,$a,#27 115 and $t0,$t0,$d 116 add $d,$d,$K // future e+=K 117 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 118 add $e,$e,$t2 // e+=rot(a,5) 119 orr $t0,$t0,$t1 120 ror $b,$b,#2 121 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 122 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 123 add $e,$e,$t0 // e+=F(b,c,d) 124 ror @Xw[$j],@Xw[$j],#31 125___ 126} 127 128sub BODY_20_39 { 129my ($i,$a,$b,$c,$d,$e)=@_; 130my $j=($i+2)&15; 131 132$code.=<<___ if ($i==39); 133 movz $K,#0xbcdc 134 movk $K,#0x8f1b,lsl#16 135___ 136$code.=<<___ if ($i<78); 137 eor @Xw[$j],@Xw[$j],@Xw[($j+2)&15] 138 eor $t0,$d,$b 139 ror $t2,$a,#27 140 add $d,$d,$K // future e+=K 141 eor @Xw[$j],@Xw[$j],@Xw[($j+8)&15] 142 eor $t0,$t0,$c 143 add $e,$e,$t2 // e+=rot(a,5) 144 ror $b,$b,#2 145 eor @Xw[$j],@Xw[$j],@Xw[($j+13)&15] 146 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 147 add $e,$e,$t0 // e+=F(b,c,d) 148 ror @Xw[$j],@Xw[$j],#31 149___ 150$code.=<<___ if ($i==78); 151 ldp @Xw[1],@Xw[2],[$ctx] 152 eor $t0,$d,$b 153 ror $t2,$a,#27 154 add $d,$d,$K // future e+=K 155 eor $t0,$t0,$c 156 add $e,$e,$t2 // e+=rot(a,5) 157 ror $b,$b,#2 158 add $d,$d,@Xw[($i+1)&15] // future e+=X[i] 159 add $e,$e,$t0 // e+=F(b,c,d) 160___ 161$code.=<<___ if ($i==79); 162 ldp @Xw[3],@Xw[4],[$ctx,#8] 163 eor $t0,$d,$b 164 ror $t2,$a,#27 165 eor $t0,$t0,$c 166 add $e,$e,$t2 // e+=rot(a,5) 167 ror $b,$b,#2 168 ldr @Xw[5],[$ctx,#16] 169 add $e,$e,$t0 // e+=F(b,c,d) 170___ 171} 172 173$code.=<<___; 174#include <openssl/arm_arch.h> 175 176.text 177 178.extern OPENSSL_armcap_P 179.hidden OPENSSL_armcap_P 180.globl sha1_block_data_order 181.type sha1_block_data_order,%function 182.align 6 183sha1_block_data_order: 184 // Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later. 185 AARCH64_VALID_CALL_TARGET 186#if __has_feature(hwaddress_sanitizer) && __clang_major__ >= 10 187 adrp x16,:pg_hi21_nc:OPENSSL_armcap_P 188#else 189 adrp x16,:pg_hi21:OPENSSL_armcap_P 190#endif 191 ldr w16,[x16,:lo12:OPENSSL_armcap_P] 192 tst w16,#ARMV8_SHA1 193 b.ne .Lv8_entry 194 195 stp x29,x30,[sp,#-96]! 196 add x29,sp,#0 197 stp x19,x20,[sp,#16] 198 stp x21,x22,[sp,#32] 199 stp x23,x24,[sp,#48] 200 stp x25,x26,[sp,#64] 201 stp x27,x28,[sp,#80] 202 203 ldp $A,$B,[$ctx] 204 ldp $C,$D,[$ctx,#8] 205 ldr $E,[$ctx,#16] 206 207.Loop: 208 ldr @Xx[0],[$inp],#64 209 movz $K,#0x7999 210 sub $num,$num,#1 211 movk $K,#0x5a82,lsl#16 212#ifdef __AARCH64EB__ 213 ror $Xx[0],@Xx[0],#32 214#else 215 rev32 @Xx[0],@Xx[0] 216#endif 217 add $E,$E,$K // warm it up 218 add $E,$E,@Xw[0] 219___ 220for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); } 221for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } 222for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); } 223for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } 224$code.=<<___; 225 add $B,$B,@Xw[2] 226 add $C,$C,@Xw[3] 227 add $A,$A,@Xw[1] 228 add $D,$D,@Xw[4] 229 add $E,$E,@Xw[5] 230 stp $A,$B,[$ctx] 231 stp $C,$D,[$ctx,#8] 232 str $E,[$ctx,#16] 233 cbnz $num,.Loop 234 235 ldp x19,x20,[sp,#16] 236 ldp x21,x22,[sp,#32] 237 ldp x23,x24,[sp,#48] 238 ldp x25,x26,[sp,#64] 239 ldp x27,x28,[sp,#80] 240 ldr x29,[sp],#96 241 ret 242.size sha1_block_data_order,.-sha1_block_data_order 243___ 244{{{ 245my ($ABCD,$E,$E0,$E1)=map("v$_.16b",(0..3)); 246my @MSG=map("v$_.16b",(4..7)); 247my @Kxx=map("v$_.4s",(16..19)); 248my ($W0,$W1)=("v20.4s","v21.4s"); 249my $ABCD_SAVE="v22.16b"; 250 251$code.=<<___; 252.type sha1_block_armv8,%function 253.align 6 254sha1_block_armv8: 255 // Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later. 256 AARCH64_VALID_CALL_TARGET 257.Lv8_entry: 258 stp x29,x30,[sp,#-16]! 259 add x29,sp,#0 260 261 adrp x4,:pg_hi21:.Lconst 262 add x4,x4,:lo12:.Lconst 263 eor $E,$E,$E 264 ld1.32 {$ABCD},[$ctx],#16 265 ld1.32 {$E}[0],[$ctx] 266 sub $ctx,$ctx,#16 267 ld1.32 {@Kxx[0]-@Kxx[3]},[x4] 268 269.Loop_hw: 270 ld1 {@MSG[0]-@MSG[3]},[$inp],#64 271 sub $num,$num,#1 272 rev32 @MSG[0],@MSG[0] 273 rev32 @MSG[1],@MSG[1] 274 275 add.i32 $W0,@Kxx[0],@MSG[0] 276 rev32 @MSG[2],@MSG[2] 277 orr $ABCD_SAVE,$ABCD,$ABCD // offload 278 279 add.i32 $W1,@Kxx[0],@MSG[1] 280 rev32 @MSG[3],@MSG[3] 281 sha1h $E1,$ABCD 282 sha1c $ABCD,$E,$W0 // 0 283 add.i32 $W0,@Kxx[$j],@MSG[2] 284 sha1su0 @MSG[0],@MSG[1],@MSG[2] 285___ 286for ($j=0,$i=1;$i<20-3;$i++) { 287my $f=("c","p","m","p")[$i/5]; 288$code.=<<___; 289 sha1h $E0,$ABCD // $i 290 sha1$f $ABCD,$E1,$W1 291 add.i32 $W1,@Kxx[$j],@MSG[3] 292 sha1su1 @MSG[0],@MSG[3] 293___ 294$code.=<<___ if ($i<20-4); 295 sha1su0 @MSG[1],@MSG[2],@MSG[3] 296___ 297 ($E0,$E1)=($E1,$E0); ($W0,$W1)=($W1,$W0); 298 push(@MSG,shift(@MSG)); $j++ if ((($i+3)%5)==0); 299} 300$code.=<<___; 301 sha1h $E0,$ABCD // $i 302 sha1p $ABCD,$E1,$W1 303 add.i32 $W1,@Kxx[$j],@MSG[3] 304 305 sha1h $E1,$ABCD // 18 306 sha1p $ABCD,$E0,$W0 307 308 sha1h $E0,$ABCD // 19 309 sha1p $ABCD,$E1,$W1 310 311 add.i32 $E,$E,$E0 312 add.i32 $ABCD,$ABCD,$ABCD_SAVE 313 314 cbnz $num,.Loop_hw 315 316 st1.32 {$ABCD},[$ctx],#16 317 st1.32 {$E}[0],[$ctx] 318 319 ldr x29,[sp],#16 320 ret 321.size sha1_block_armv8,.-sha1_block_armv8 322.section .rodata 323.align 6 324.Lconst: 325.long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 //K_00_19 326.long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 //K_20_39 327.long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc //K_40_59 328.long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 //K_60_79 329.asciz "SHA1 block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>" 330.align 2 331___ 332}}} 333 334{ my %opcode = ( 335 "sha1c" => 0x5e000000, "sha1p" => 0x5e001000, 336 "sha1m" => 0x5e002000, "sha1su0" => 0x5e003000, 337 "sha1h" => 0x5e280800, "sha1su1" => 0x5e281800 ); 338 339 sub unsha1 { 340 my ($mnemonic,$arg)=@_; 341 342 $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o 343 && 344 sprintf ".inst\t0x%08x\t//%s %s", 345 $opcode{$mnemonic}|$1|($2<<5)|($3<<16), 346 $mnemonic,$arg; 347 } 348} 349 350foreach(split("\n",$code)) { 351 352 s/\`([^\`]*)\`/eval($1)/geo; 353 354 s/\b(sha1\w+)\s+([qv].*)/unsha1($1,$2)/geo; 355 356 s/\.\w?32\b//o and s/\.16b/\.4s/go; 357 m/(ld|st)1[^\[]+\[0\]/o and s/\.4s/\.s/go; 358 359 print $_,"\n"; 360} 361 362close STDOUT or die "error closing STDOUT: $!"; 363