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# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL 11# project. The module is, however, dual licensed under OpenSSL and 12# CRYPTOGAMS licenses depending on where you obtain it. For further 13# details see http://www.openssl.org/~appro/cryptogams/. 14# 15# Permission to use under GPLv2 terms is granted. 16# ==================================================================== 17# 18# SHA256/512 for ARMv8. 19# 20# Performance in cycles per processed byte and improvement coefficient 21# over code generated with "default" compiler: 22# 23# SHA256-hw SHA256(*) SHA512 24# Apple A7 1.97 10.5 (+33%) 6.73 (-1%(**)) 25# Cortex-A53 2.38 15.5 (+115%) 10.0 (+150%(***)) 26# Cortex-A57 2.31 11.6 (+86%) 7.51 (+260%(***)) 27# Denver 2.01 10.5 (+26%) 6.70 (+8%) 28# X-Gene 20.0 (+100%) 12.8 (+300%(***)) 29# Mongoose 2.36 13.0 (+50%) 8.36 (+33%) 30# 31# (*) Software SHA256 results are of lesser relevance, presented 32# mostly for informational purposes. 33# (**) The result is a trade-off: it's possible to improve it by 34# 10% (or by 1 cycle per round), but at the cost of 20% loss 35# on Cortex-A53 (or by 4 cycles per round). 36# (***) Super-impressive coefficients over gcc-generated code are 37# indication of some compiler "pathology", most notably code 38# generated with -mgeneral-regs-only is significanty faster 39# and the gap is only 40-90%. 40# 41# October 2016. 42# 43# Originally it was reckoned that it makes no sense to implement NEON 44# version of SHA256 for 64-bit processors. This is because performance 45# improvement on most wide-spread Cortex-A5x processors was observed 46# to be marginal, same on Cortex-A53 and ~10% on A57. But then it was 47# observed that 32-bit NEON SHA256 performs significantly better than 48# 64-bit scalar version on *some* of the more recent processors. As 49# result 64-bit NEON version of SHA256 was added to provide best 50# all-round performance. For example it executes ~30% faster on X-Gene 51# and Mongoose. [For reference, NEON version of SHA512 is bound to 52# deliver much less improvement, likely *negative* on Cortex-A5x. 53# Which is why NEON support is limited to SHA256.] 54 55$output=pop; 56$flavour=pop; 57 58if ($flavour && $flavour ne "void") { 59 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 60 ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or 61 ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or 62 die "can't locate arm-xlate.pl"; 63 64 open OUT,"| \"$^X\" $xlate $flavour $output"; 65 *STDOUT=*OUT; 66} else { 67 open STDOUT,">$output"; 68} 69 70if ($output =~ /512/) { 71 $BITS=512; 72 $SZ=8; 73 @Sigma0=(28,34,39); 74 @Sigma1=(14,18,41); 75 @sigma0=(1, 8, 7); 76 @sigma1=(19,61, 6); 77 $rounds=80; 78 $reg_t="x"; 79} else { 80 $BITS=256; 81 $SZ=4; 82 @Sigma0=( 2,13,22); 83 @Sigma1=( 6,11,25); 84 @sigma0=( 7,18, 3); 85 @sigma1=(17,19,10); 86 $rounds=64; 87 $reg_t="w"; 88} 89 90$func="sha${BITS}_block_data_order"; 91 92($ctx,$inp,$num,$Ktbl)=map("x$_",(0..2,30)); 93 94@X=map("$reg_t$_",(3..15,0..2)); 95@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("$reg_t$_",(20..27)); 96($t0,$t1,$t2,$t3)=map("$reg_t$_",(16,17,19,28)); 97 98sub BODY_00_xx { 99my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_; 100my $j=($i+1)&15; 101my ($T0,$T1,$T2)=(@X[($i-8)&15],@X[($i-9)&15],@X[($i-10)&15]); 102 $T0=@X[$i+3] if ($i<11); 103 104$code.=<<___ if ($i<16); 105#ifndef __AARCH64EB__ 106 rev @X[$i],@X[$i] // $i 107#endif 108___ 109$code.=<<___ if ($i<13 && ($i&1)); 110 ldp @X[$i+1],@X[$i+2],[$inp],#2*$SZ 111___ 112$code.=<<___ if ($i==13); 113 ldp @X[14],@X[15],[$inp] 114___ 115$code.=<<___ if ($i>=14); 116 ldr @X[($i-11)&15],[sp,#`$SZ*(($i-11)%4)`] 117___ 118$code.=<<___ if ($i>0 && $i<16); 119 add $a,$a,$t1 // h+=Sigma0(a) 120___ 121$code.=<<___ if ($i>=11); 122 str @X[($i-8)&15],[sp,#`$SZ*(($i-8)%4)`] 123___ 124# While ARMv8 specifies merged rotate-n-logical operation such as 125# 'eor x,y,z,ror#n', it was found to negatively affect performance 126# on Apple A7. The reason seems to be that it requires even 'y' to 127# be available earlier. This means that such merged instruction is 128# not necessarily best choice on critical path... On the other hand 129# Cortex-A5x handles merged instructions much better than disjoint 130# rotate and logical... See (**) footnote above. 131$code.=<<___ if ($i<15); 132 ror $t0,$e,#$Sigma1[0] 133 add $h,$h,$t2 // h+=K[i] 134 eor $T0,$e,$e,ror#`$Sigma1[2]-$Sigma1[1]` 135 and $t1,$f,$e 136 bic $t2,$g,$e 137 add $h,$h,@X[$i&15] // h+=X[i] 138 orr $t1,$t1,$t2 // Ch(e,f,g) 139 eor $t2,$a,$b // a^b, b^c in next round 140 eor $t0,$t0,$T0,ror#$Sigma1[1] // Sigma1(e) 141 ror $T0,$a,#$Sigma0[0] 142 add $h,$h,$t1 // h+=Ch(e,f,g) 143 eor $t1,$a,$a,ror#`$Sigma0[2]-$Sigma0[1]` 144 add $h,$h,$t0 // h+=Sigma1(e) 145 and $t3,$t3,$t2 // (b^c)&=(a^b) 146 add $d,$d,$h // d+=h 147 eor $t3,$t3,$b // Maj(a,b,c) 148 eor $t1,$T0,$t1,ror#$Sigma0[1] // Sigma0(a) 149 add $h,$h,$t3 // h+=Maj(a,b,c) 150 ldr $t3,[$Ktbl],#$SZ // *K++, $t2 in next round 151 //add $h,$h,$t1 // h+=Sigma0(a) 152___ 153$code.=<<___ if ($i>=15); 154 ror $t0,$e,#$Sigma1[0] 155 add $h,$h,$t2 // h+=K[i] 156 ror $T1,@X[($j+1)&15],#$sigma0[0] 157 and $t1,$f,$e 158 ror $T2,@X[($j+14)&15],#$sigma1[0] 159 bic $t2,$g,$e 160 ror $T0,$a,#$Sigma0[0] 161 add $h,$h,@X[$i&15] // h+=X[i] 162 eor $t0,$t0,$e,ror#$Sigma1[1] 163 eor $T1,$T1,@X[($j+1)&15],ror#$sigma0[1] 164 orr $t1,$t1,$t2 // Ch(e,f,g) 165 eor $t2,$a,$b // a^b, b^c in next round 166 eor $t0,$t0,$e,ror#$Sigma1[2] // Sigma1(e) 167 eor $T0,$T0,$a,ror#$Sigma0[1] 168 add $h,$h,$t1 // h+=Ch(e,f,g) 169 and $t3,$t3,$t2 // (b^c)&=(a^b) 170 eor $T2,$T2,@X[($j+14)&15],ror#$sigma1[1] 171 eor $T1,$T1,@X[($j+1)&15],lsr#$sigma0[2] // sigma0(X[i+1]) 172 add $h,$h,$t0 // h+=Sigma1(e) 173 eor $t3,$t3,$b // Maj(a,b,c) 174 eor $t1,$T0,$a,ror#$Sigma0[2] // Sigma0(a) 175 eor $T2,$T2,@X[($j+14)&15],lsr#$sigma1[2] // sigma1(X[i+14]) 176 add @X[$j],@X[$j],@X[($j+9)&15] 177 add $d,$d,$h // d+=h 178 add $h,$h,$t3 // h+=Maj(a,b,c) 179 ldr $t3,[$Ktbl],#$SZ // *K++, $t2 in next round 180 add @X[$j],@X[$j],$T1 181 add $h,$h,$t1 // h+=Sigma0(a) 182 add @X[$j],@X[$j],$T2 183___ 184 ($t2,$t3)=($t3,$t2); 185} 186 187$code.=<<___; 188#ifndef __KERNEL__ 189# include "arm_arch.h" 190#endif 191 192.text 193 194.extern OPENSSL_armcap_P 195.globl $func 196.type $func,%function 197.align 6 198$func: 199___ 200$code.=<<___ if ($SZ==4); 201#ifndef __KERNEL__ 202# ifdef __ILP32__ 203 ldrsw x16,.LOPENSSL_armcap_P 204# else 205 ldr x16,.LOPENSSL_armcap_P 206# endif 207 adr x17,.LOPENSSL_armcap_P 208 add x16,x16,x17 209 ldr w16,[x16] 210 tst w16,#ARMV8_SHA256 211 b.ne .Lv8_entry 212 tst w16,#ARMV7_NEON 213 b.ne .Lneon_entry 214#endif 215___ 216$code.=<<___; 217 stp x29,x30,[sp,#-128]! 218 add x29,sp,#0 219 220 stp x19,x20,[sp,#16] 221 stp x21,x22,[sp,#32] 222 stp x23,x24,[sp,#48] 223 stp x25,x26,[sp,#64] 224 stp x27,x28,[sp,#80] 225 sub sp,sp,#4*$SZ 226 227 ldp $A,$B,[$ctx] // load context 228 ldp $C,$D,[$ctx,#2*$SZ] 229 ldp $E,$F,[$ctx,#4*$SZ] 230 add $num,$inp,$num,lsl#`log(16*$SZ)/log(2)` // end of input 231 ldp $G,$H,[$ctx,#6*$SZ] 232 adr $Ktbl,.LK$BITS 233 stp $ctx,$num,[x29,#96] 234 235.Loop: 236 ldp @X[0],@X[1],[$inp],#2*$SZ 237 ldr $t2,[$Ktbl],#$SZ // *K++ 238 eor $t3,$B,$C // magic seed 239 str $inp,[x29,#112] 240___ 241for ($i=0;$i<16;$i++) { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); } 242$code.=".Loop_16_xx:\n"; 243for (;$i<32;$i++) { &BODY_00_xx($i,@V); unshift(@V,pop(@V)); } 244$code.=<<___; 245 cbnz $t2,.Loop_16_xx 246 247 ldp $ctx,$num,[x29,#96] 248 ldr $inp,[x29,#112] 249 sub $Ktbl,$Ktbl,#`$SZ*($rounds+1)` // rewind 250 251 ldp @X[0],@X[1],[$ctx] 252 ldp @X[2],@X[3],[$ctx,#2*$SZ] 253 add $inp,$inp,#14*$SZ // advance input pointer 254 ldp @X[4],@X[5],[$ctx,#4*$SZ] 255 add $A,$A,@X[0] 256 ldp @X[6],@X[7],[$ctx,#6*$SZ] 257 add $B,$B,@X[1] 258 add $C,$C,@X[2] 259 add $D,$D,@X[3] 260 stp $A,$B,[$ctx] 261 add $E,$E,@X[4] 262 add $F,$F,@X[5] 263 stp $C,$D,[$ctx,#2*$SZ] 264 add $G,$G,@X[6] 265 add $H,$H,@X[7] 266 cmp $inp,$num 267 stp $E,$F,[$ctx,#4*$SZ] 268 stp $G,$H,[$ctx,#6*$SZ] 269 b.ne .Loop 270 271 ldp x19,x20,[x29,#16] 272 add sp,sp,#4*$SZ 273 ldp x21,x22,[x29,#32] 274 ldp x23,x24,[x29,#48] 275 ldp x25,x26,[x29,#64] 276 ldp x27,x28,[x29,#80] 277 ldp x29,x30,[sp],#128 278 ret 279.size $func,.-$func 280 281.align 6 282.type .LK$BITS,%object 283.LK$BITS: 284___ 285$code.=<<___ if ($SZ==8); 286 .quad 0x428a2f98d728ae22,0x7137449123ef65cd 287 .quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc 288 .quad 0x3956c25bf348b538,0x59f111f1b605d019 289 .quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118 290 .quad 0xd807aa98a3030242,0x12835b0145706fbe 291 .quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2 292 .quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1 293 .quad 0x9bdc06a725c71235,0xc19bf174cf692694 294 .quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3 295 .quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65 296 .quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483 297 .quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5 298 .quad 0x983e5152ee66dfab,0xa831c66d2db43210 299 .quad 0xb00327c898fb213f,0xbf597fc7beef0ee4 300 .quad 0xc6e00bf33da88fc2,0xd5a79147930aa725 301 .quad 0x06ca6351e003826f,0x142929670a0e6e70 302 .quad 0x27b70a8546d22ffc,0x2e1b21385c26c926 303 .quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df 304 .quad 0x650a73548baf63de,0x766a0abb3c77b2a8 305 .quad 0x81c2c92e47edaee6,0x92722c851482353b 306 .quad 0xa2bfe8a14cf10364,0xa81a664bbc423001 307 .quad 0xc24b8b70d0f89791,0xc76c51a30654be30 308 .quad 0xd192e819d6ef5218,0xd69906245565a910 309 .quad 0xf40e35855771202a,0x106aa07032bbd1b8 310 .quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53 311 .quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8 312 .quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb 313 .quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3 314 .quad 0x748f82ee5defb2fc,0x78a5636f43172f60 315 .quad 0x84c87814a1f0ab72,0x8cc702081a6439ec 316 .quad 0x90befffa23631e28,0xa4506cebde82bde9 317 .quad 0xbef9a3f7b2c67915,0xc67178f2e372532b 318 .quad 0xca273eceea26619c,0xd186b8c721c0c207 319 .quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178 320 .quad 0x06f067aa72176fba,0x0a637dc5a2c898a6 321 .quad 0x113f9804bef90dae,0x1b710b35131c471b 322 .quad 0x28db77f523047d84,0x32caab7b40c72493 323 .quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c 324 .quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a 325 .quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817 326 .quad 0 // terminator 327___ 328$code.=<<___ if ($SZ==4); 329 .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5 330 .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5 331 .long 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3 332 .long 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174 333 .long 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc 334 .long 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da 335 .long 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7 336 .long 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967 337 .long 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13 338 .long 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85 339 .long 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3 340 .long 0xd192e819,0xd6990624,0xf40e3585,0x106aa070 341 .long 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5 342 .long 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3 343 .long 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208 344 .long 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2 345 .long 0 //terminator 346___ 347$code.=<<___; 348.size .LK$BITS,.-.LK$BITS 349#ifndef __KERNEL__ 350.align 3 351.LOPENSSL_armcap_P: 352# ifdef __ILP32__ 353 .long OPENSSL_armcap_P-. 354# else 355 .quad OPENSSL_armcap_P-. 356# endif 357#endif 358.asciz "SHA$BITS block transform for ARMv8, CRYPTOGAMS by <appro\@openssl.org>" 359.align 2 360___ 361 362if ($SZ==4) { 363my $Ktbl="x3"; 364 365my ($ABCD,$EFGH,$abcd)=map("v$_.16b",(0..2)); 366my @MSG=map("v$_.16b",(4..7)); 367my ($W0,$W1)=("v16.4s","v17.4s"); 368my ($ABCD_SAVE,$EFGH_SAVE)=("v18.16b","v19.16b"); 369 370$code.=<<___; 371#ifndef __KERNEL__ 372.type sha256_block_armv8,%function 373.align 6 374sha256_block_armv8: 375.Lv8_entry: 376 stp x29,x30,[sp,#-16]! 377 add x29,sp,#0 378 379 ld1.32 {$ABCD,$EFGH},[$ctx] 380 adr $Ktbl,.LK256 381 382.Loop_hw: 383 ld1 {@MSG[0]-@MSG[3]},[$inp],#64 384 sub $num,$num,#1 385 ld1.32 {$W0},[$Ktbl],#16 386 rev32 @MSG[0],@MSG[0] 387 rev32 @MSG[1],@MSG[1] 388 rev32 @MSG[2],@MSG[2] 389 rev32 @MSG[3],@MSG[3] 390 orr $ABCD_SAVE,$ABCD,$ABCD // offload 391 orr $EFGH_SAVE,$EFGH,$EFGH 392___ 393for($i=0;$i<12;$i++) { 394$code.=<<___; 395 ld1.32 {$W1},[$Ktbl],#16 396 add.i32 $W0,$W0,@MSG[0] 397 sha256su0 @MSG[0],@MSG[1] 398 orr $abcd,$ABCD,$ABCD 399 sha256h $ABCD,$EFGH,$W0 400 sha256h2 $EFGH,$abcd,$W0 401 sha256su1 @MSG[0],@MSG[2],@MSG[3] 402___ 403 ($W0,$W1)=($W1,$W0); push(@MSG,shift(@MSG)); 404} 405$code.=<<___; 406 ld1.32 {$W1},[$Ktbl],#16 407 add.i32 $W0,$W0,@MSG[0] 408 orr $abcd,$ABCD,$ABCD 409 sha256h $ABCD,$EFGH,$W0 410 sha256h2 $EFGH,$abcd,$W0 411 412 ld1.32 {$W0},[$Ktbl],#16 413 add.i32 $W1,$W1,@MSG[1] 414 orr $abcd,$ABCD,$ABCD 415 sha256h $ABCD,$EFGH,$W1 416 sha256h2 $EFGH,$abcd,$W1 417 418 ld1.32 {$W1},[$Ktbl] 419 add.i32 $W0,$W0,@MSG[2] 420 sub $Ktbl,$Ktbl,#$rounds*$SZ-16 // rewind 421 orr $abcd,$ABCD,$ABCD 422 sha256h $ABCD,$EFGH,$W0 423 sha256h2 $EFGH,$abcd,$W0 424 425 add.i32 $W1,$W1,@MSG[3] 426 orr $abcd,$ABCD,$ABCD 427 sha256h $ABCD,$EFGH,$W1 428 sha256h2 $EFGH,$abcd,$W1 429 430 add.i32 $ABCD,$ABCD,$ABCD_SAVE 431 add.i32 $EFGH,$EFGH,$EFGH_SAVE 432 433 cbnz $num,.Loop_hw 434 435 st1.32 {$ABCD,$EFGH},[$ctx] 436 437 ldr x29,[sp],#16 438 ret 439.size sha256_block_armv8,.-sha256_block_armv8 440#endif 441___ 442} 443 444if ($SZ==4) { ######################################### NEON stuff # 445# You'll surely note a lot of similarities with sha256-armv4 module, 446# and of course it's not a coincidence. sha256-armv4 was used as 447# initial template, but was adapted for ARMv8 instruction set and 448# extensively re-tuned for all-round performance. 449 450my @V = ($A,$B,$C,$D,$E,$F,$G,$H) = map("w$_",(3..10)); 451my ($t0,$t1,$t2,$t3,$t4) = map("w$_",(11..15)); 452my $Ktbl="x16"; 453my $Xfer="x17"; 454my @X = map("q$_",(0..3)); 455my ($T0,$T1,$T2,$T3,$T4,$T5,$T6,$T7) = map("q$_",(4..7,16..19)); 456my $j=0; 457 458sub AUTOLOAD() # thunk [simplified] x86-style perlasm 459{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./; 460 my $arg = pop; 461 $arg = "#$arg" if ($arg*1 eq $arg); 462 $code .= "\t$opcode\t".join(',',@_,$arg)."\n"; 463} 464 465sub Dscalar { shift =~ m|[qv]([0-9]+)|?"d$1":""; } 466sub Dlo { shift =~ m|[qv]([0-9]+)|?"v$1.d[0]":""; } 467sub Dhi { shift =~ m|[qv]([0-9]+)|?"v$1.d[1]":""; } 468 469sub Xupdate() 470{ use integer; 471 my $body = shift; 472 my @insns = (&$body,&$body,&$body,&$body); 473 my ($a,$b,$c,$d,$e,$f,$g,$h); 474 475 &ext_8 ($T0,@X[0],@X[1],4); # X[1..4] 476 eval(shift(@insns)); 477 eval(shift(@insns)); 478 eval(shift(@insns)); 479 &ext_8 ($T3,@X[2],@X[3],4); # X[9..12] 480 eval(shift(@insns)); 481 eval(shift(@insns)); 482 &mov (&Dscalar($T7),&Dhi(@X[3])); # X[14..15] 483 eval(shift(@insns)); 484 eval(shift(@insns)); 485 &ushr_32 ($T2,$T0,$sigma0[0]); 486 eval(shift(@insns)); 487 &ushr_32 ($T1,$T0,$sigma0[2]); 488 eval(shift(@insns)); 489 &add_32 (@X[0],@X[0],$T3); # X[0..3] += X[9..12] 490 eval(shift(@insns)); 491 &sli_32 ($T2,$T0,32-$sigma0[0]); 492 eval(shift(@insns)); 493 eval(shift(@insns)); 494 &ushr_32 ($T3,$T0,$sigma0[1]); 495 eval(shift(@insns)); 496 eval(shift(@insns)); 497 &eor_8 ($T1,$T1,$T2); 498 eval(shift(@insns)); 499 eval(shift(@insns)); 500 &sli_32 ($T3,$T0,32-$sigma0[1]); 501 eval(shift(@insns)); 502 eval(shift(@insns)); 503 &ushr_32 ($T4,$T7,$sigma1[0]); 504 eval(shift(@insns)); 505 eval(shift(@insns)); 506 &eor_8 ($T1,$T1,$T3); # sigma0(X[1..4]) 507 eval(shift(@insns)); 508 eval(shift(@insns)); 509 &sli_32 ($T4,$T7,32-$sigma1[0]); 510 eval(shift(@insns)); 511 eval(shift(@insns)); 512 &ushr_32 ($T5,$T7,$sigma1[2]); 513 eval(shift(@insns)); 514 eval(shift(@insns)); 515 &ushr_32 ($T3,$T7,$sigma1[1]); 516 eval(shift(@insns)); 517 eval(shift(@insns)); 518 &add_32 (@X[0],@X[0],$T1); # X[0..3] += sigma0(X[1..4]) 519 eval(shift(@insns)); 520 eval(shift(@insns)); 521 &sli_u32 ($T3,$T7,32-$sigma1[1]); 522 eval(shift(@insns)); 523 eval(shift(@insns)); 524 &eor_8 ($T5,$T5,$T4); 525 eval(shift(@insns)); 526 eval(shift(@insns)); 527 eval(shift(@insns)); 528 &eor_8 ($T5,$T5,$T3); # sigma1(X[14..15]) 529 eval(shift(@insns)); 530 eval(shift(@insns)); 531 eval(shift(@insns)); 532 &add_32 (@X[0],@X[0],$T5); # X[0..1] += sigma1(X[14..15]) 533 eval(shift(@insns)); 534 eval(shift(@insns)); 535 eval(shift(@insns)); 536 &ushr_32 ($T6,@X[0],$sigma1[0]); 537 eval(shift(@insns)); 538 &ushr_32 ($T7,@X[0],$sigma1[2]); 539 eval(shift(@insns)); 540 eval(shift(@insns)); 541 &sli_32 ($T6,@X[0],32-$sigma1[0]); 542 eval(shift(@insns)); 543 &ushr_32 ($T5,@X[0],$sigma1[1]); 544 eval(shift(@insns)); 545 eval(shift(@insns)); 546 &eor_8 ($T7,$T7,$T6); 547 eval(shift(@insns)); 548 eval(shift(@insns)); 549 &sli_32 ($T5,@X[0],32-$sigma1[1]); 550 eval(shift(@insns)); 551 eval(shift(@insns)); 552 &ld1_32 ("{$T0}","[$Ktbl], #16"); 553 eval(shift(@insns)); 554 &eor_8 ($T7,$T7,$T5); # sigma1(X[16..17]) 555 eval(shift(@insns)); 556 eval(shift(@insns)); 557 &eor_8 ($T5,$T5,$T5); 558 eval(shift(@insns)); 559 eval(shift(@insns)); 560 &mov (&Dhi($T5), &Dlo($T7)); 561 eval(shift(@insns)); 562 eval(shift(@insns)); 563 eval(shift(@insns)); 564 &add_32 (@X[0],@X[0],$T5); # X[2..3] += sigma1(X[16..17]) 565 eval(shift(@insns)); 566 eval(shift(@insns)); 567 eval(shift(@insns)); 568 &add_32 ($T0,$T0,@X[0]); 569 while($#insns>=1) { eval(shift(@insns)); } 570 &st1_32 ("{$T0}","[$Xfer], #16"); 571 eval(shift(@insns)); 572 573 push(@X,shift(@X)); # "rotate" X[] 574} 575 576sub Xpreload() 577{ use integer; 578 my $body = shift; 579 my @insns = (&$body,&$body,&$body,&$body); 580 my ($a,$b,$c,$d,$e,$f,$g,$h); 581 582 eval(shift(@insns)); 583 eval(shift(@insns)); 584 &ld1_8 ("{@X[0]}","[$inp],#16"); 585 eval(shift(@insns)); 586 eval(shift(@insns)); 587 &ld1_32 ("{$T0}","[$Ktbl],#16"); 588 eval(shift(@insns)); 589 eval(shift(@insns)); 590 eval(shift(@insns)); 591 eval(shift(@insns)); 592 &rev32 (@X[0],@X[0]); 593 eval(shift(@insns)); 594 eval(shift(@insns)); 595 eval(shift(@insns)); 596 eval(shift(@insns)); 597 &add_32 ($T0,$T0,@X[0]); 598 foreach (@insns) { eval; } # remaining instructions 599 &st1_32 ("{$T0}","[$Xfer], #16"); 600 601 push(@X,shift(@X)); # "rotate" X[] 602} 603 604sub body_00_15 () { 605 ( 606 '($a,$b,$c,$d,$e,$f,$g,$h)=@V;'. 607 '&add ($h,$h,$t1)', # h+=X[i]+K[i] 608 '&add ($a,$a,$t4);'. # h+=Sigma0(a) from the past 609 '&and ($t1,$f,$e)', 610 '&bic ($t4,$g,$e)', 611 '&eor ($t0,$e,$e,"ror#".($Sigma1[1]-$Sigma1[0]))', 612 '&add ($a,$a,$t2)', # h+=Maj(a,b,c) from the past 613 '&orr ($t1,$t1,$t4)', # Ch(e,f,g) 614 '&eor ($t0,$t0,$e,"ror#".($Sigma1[2]-$Sigma1[0]))', # Sigma1(e) 615 '&eor ($t4,$a,$a,"ror#".($Sigma0[1]-$Sigma0[0]))', 616 '&add ($h,$h,$t1)', # h+=Ch(e,f,g) 617 '&ror ($t0,$t0,"#$Sigma1[0]")', 618 '&eor ($t2,$a,$b)', # a^b, b^c in next round 619 '&eor ($t4,$t4,$a,"ror#".($Sigma0[2]-$Sigma0[0]))', # Sigma0(a) 620 '&add ($h,$h,$t0)', # h+=Sigma1(e) 621 '&ldr ($t1,sprintf "[sp,#%d]",4*(($j+1)&15)) if (($j&15)!=15);'. 622 '&ldr ($t1,"[$Ktbl]") if ($j==15);'. 623 '&and ($t3,$t3,$t2)', # (b^c)&=(a^b) 624 '&ror ($t4,$t4,"#$Sigma0[0]")', 625 '&add ($d,$d,$h)', # d+=h 626 '&eor ($t3,$t3,$b)', # Maj(a,b,c) 627 '$j++; unshift(@V,pop(@V)); ($t2,$t3)=($t3,$t2);' 628 ) 629} 630 631$code.=<<___; 632#ifdef __KERNEL__ 633.globl sha256_block_neon 634#endif 635.type sha256_block_neon,%function 636.align 4 637sha256_block_neon: 638.Lneon_entry: 639 stp x29, x30, [sp, #-16]! 640 mov x29, sp 641 sub sp,sp,#16*4 642 643 adr $Ktbl,.LK256 644 add $num,$inp,$num,lsl#6 // len to point at the end of inp 645 646 ld1.8 {@X[0]},[$inp], #16 647 ld1.8 {@X[1]},[$inp], #16 648 ld1.8 {@X[2]},[$inp], #16 649 ld1.8 {@X[3]},[$inp], #16 650 ld1.32 {$T0},[$Ktbl], #16 651 ld1.32 {$T1},[$Ktbl], #16 652 ld1.32 {$T2},[$Ktbl], #16 653 ld1.32 {$T3},[$Ktbl], #16 654 rev32 @X[0],@X[0] // yes, even on 655 rev32 @X[1],@X[1] // big-endian 656 rev32 @X[2],@X[2] 657 rev32 @X[3],@X[3] 658 mov $Xfer,sp 659 add.32 $T0,$T0,@X[0] 660 add.32 $T1,$T1,@X[1] 661 add.32 $T2,$T2,@X[2] 662 st1.32 {$T0-$T1},[$Xfer], #32 663 add.32 $T3,$T3,@X[3] 664 st1.32 {$T2-$T3},[$Xfer] 665 sub $Xfer,$Xfer,#32 666 667 ldp $A,$B,[$ctx] 668 ldp $C,$D,[$ctx,#8] 669 ldp $E,$F,[$ctx,#16] 670 ldp $G,$H,[$ctx,#24] 671 ldr $t1,[sp,#0] 672 mov $t2,wzr 673 eor $t3,$B,$C 674 mov $t4,wzr 675 b .L_00_48 676 677.align 4 678.L_00_48: 679___ 680 &Xupdate(\&body_00_15); 681 &Xupdate(\&body_00_15); 682 &Xupdate(\&body_00_15); 683 &Xupdate(\&body_00_15); 684$code.=<<___; 685 cmp $t1,#0 // check for K256 terminator 686 ldr $t1,[sp,#0] 687 sub $Xfer,$Xfer,#64 688 bne .L_00_48 689 690 sub $Ktbl,$Ktbl,#256 // rewind $Ktbl 691 cmp $inp,$num 692 mov $Xfer, #64 693 csel $Xfer, $Xfer, xzr, eq 694 sub $inp,$inp,$Xfer // avoid SEGV 695 mov $Xfer,sp 696___ 697 &Xpreload(\&body_00_15); 698 &Xpreload(\&body_00_15); 699 &Xpreload(\&body_00_15); 700 &Xpreload(\&body_00_15); 701$code.=<<___; 702 add $A,$A,$t4 // h+=Sigma0(a) from the past 703 ldp $t0,$t1,[$ctx,#0] 704 add $A,$A,$t2 // h+=Maj(a,b,c) from the past 705 ldp $t2,$t3,[$ctx,#8] 706 add $A,$A,$t0 // accumulate 707 add $B,$B,$t1 708 ldp $t0,$t1,[$ctx,#16] 709 add $C,$C,$t2 710 add $D,$D,$t3 711 ldp $t2,$t3,[$ctx,#24] 712 add $E,$E,$t0 713 add $F,$F,$t1 714 ldr $t1,[sp,#0] 715 stp $A,$B,[$ctx,#0] 716 add $G,$G,$t2 717 mov $t2,wzr 718 stp $C,$D,[$ctx,#8] 719 add $H,$H,$t3 720 stp $E,$F,[$ctx,#16] 721 eor $t3,$B,$C 722 stp $G,$H,[$ctx,#24] 723 mov $t4,wzr 724 mov $Xfer,sp 725 b.ne .L_00_48 726 727 ldr x29,[x29] 728 add sp,sp,#16*4+16 729 ret 730.size sha256_block_neon,.-sha256_block_neon 731___ 732} 733 734$code.=<<___; 735#ifndef __KERNEL__ 736.comm OPENSSL_armcap_P,4,4 737#endif 738___ 739 740{ my %opcode = ( 741 "sha256h" => 0x5e004000, "sha256h2" => 0x5e005000, 742 "sha256su0" => 0x5e282800, "sha256su1" => 0x5e006000 ); 743 744 sub unsha256 { 745 my ($mnemonic,$arg)=@_; 746 747 $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o 748 && 749 sprintf ".inst\t0x%08x\t//%s %s", 750 $opcode{$mnemonic}|$1|($2<<5)|($3<<16), 751 $mnemonic,$arg; 752 } 753} 754 755open SELF,$0; 756while(<SELF>) { 757 next if (/^#!/); 758 last if (!s/^#/\/\// and !/^$/); 759 print; 760} 761close SELF; 762 763foreach(split("\n",$code)) { 764 765 s/\`([^\`]*)\`/eval($1)/ge; 766 767 s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/ge; 768 769 s/\bq([0-9]+)\b/v$1.16b/g; # old->new registers 770 771 s/\.[ui]?8(\s)/$1/; 772 s/\.\w?32\b// and s/\.16b/\.4s/g; 773 m/(ld|st)1[^\[]+\[0\]/ and s/\.4s/\.s/g; 774 775 print $_,"\n"; 776} 777 778close STDOUT; 779