1#! /usr/bin/env perl 2# Copyright 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# June 2015 18# 19# ChaCha20 for ARMv8. 20# 21# Performance in cycles per byte out of large buffer. 22# 23# IALU/gcc-4.9 3xNEON+1xIALU 6xNEON+2xIALU 24# 25# Apple A7 5.50/+49% 3.33 1.70 26# Cortex-A53 8.40/+80% 4.72 4.72(*) 27# Cortex-A57 8.06/+43% 4.90 4.43(**) 28# Denver 4.50/+82% 2.63 2.67(*) 29# X-Gene 9.50/+46% 8.82 8.89(*) 30# Mongoose 8.00/+44% 3.64 3.25 31# Kryo 8.17/+50% 4.83 4.65 32# 33# (*) it's expected that doubling interleave factor doesn't help 34# all processors, only those with higher NEON latency and 35# higher instruction issue rate; 36# (**) expected improvement was actually higher; 37 38$flavour=shift; 39$output=shift; 40 41$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 42( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or 43( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or 44die "can't locate arm-xlate.pl"; 45 46open OUT,"| \"$^X\" $xlate $flavour $output"; 47*STDOUT=*OUT; 48 49sub AUTOLOAD() # thunk [simplified] x86-style perlasm 50{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./; 51 my $arg = pop; 52 $arg = "#$arg" if ($arg*1 eq $arg); 53 $code .= "\t$opcode\t".join(',',@_,$arg)."\n"; 54} 55 56my ($out,$inp,$len,$key,$ctr) = map("x$_",(0..4)); 57 58my @x=map("x$_",(5..17,19..21)); 59my @d=map("x$_",(22..28,30)); 60 61sub ROUND { 62my ($a0,$b0,$c0,$d0)=@_; 63my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0)); 64my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1)); 65my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2)); 66 67 ( 68 "&add_32 (@x[$a0],@x[$a0],@x[$b0])", 69 "&add_32 (@x[$a1],@x[$a1],@x[$b1])", 70 "&add_32 (@x[$a2],@x[$a2],@x[$b2])", 71 "&add_32 (@x[$a3],@x[$a3],@x[$b3])", 72 "&eor_32 (@x[$d0],@x[$d0],@x[$a0])", 73 "&eor_32 (@x[$d1],@x[$d1],@x[$a1])", 74 "&eor_32 (@x[$d2],@x[$d2],@x[$a2])", 75 "&eor_32 (@x[$d3],@x[$d3],@x[$a3])", 76 "&ror_32 (@x[$d0],@x[$d0],16)", 77 "&ror_32 (@x[$d1],@x[$d1],16)", 78 "&ror_32 (@x[$d2],@x[$d2],16)", 79 "&ror_32 (@x[$d3],@x[$d3],16)", 80 81 "&add_32 (@x[$c0],@x[$c0],@x[$d0])", 82 "&add_32 (@x[$c1],@x[$c1],@x[$d1])", 83 "&add_32 (@x[$c2],@x[$c2],@x[$d2])", 84 "&add_32 (@x[$c3],@x[$c3],@x[$d3])", 85 "&eor_32 (@x[$b0],@x[$b0],@x[$c0])", 86 "&eor_32 (@x[$b1],@x[$b1],@x[$c1])", 87 "&eor_32 (@x[$b2],@x[$b2],@x[$c2])", 88 "&eor_32 (@x[$b3],@x[$b3],@x[$c3])", 89 "&ror_32 (@x[$b0],@x[$b0],20)", 90 "&ror_32 (@x[$b1],@x[$b1],20)", 91 "&ror_32 (@x[$b2],@x[$b2],20)", 92 "&ror_32 (@x[$b3],@x[$b3],20)", 93 94 "&add_32 (@x[$a0],@x[$a0],@x[$b0])", 95 "&add_32 (@x[$a1],@x[$a1],@x[$b1])", 96 "&add_32 (@x[$a2],@x[$a2],@x[$b2])", 97 "&add_32 (@x[$a3],@x[$a3],@x[$b3])", 98 "&eor_32 (@x[$d0],@x[$d0],@x[$a0])", 99 "&eor_32 (@x[$d1],@x[$d1],@x[$a1])", 100 "&eor_32 (@x[$d2],@x[$d2],@x[$a2])", 101 "&eor_32 (@x[$d3],@x[$d3],@x[$a3])", 102 "&ror_32 (@x[$d0],@x[$d0],24)", 103 "&ror_32 (@x[$d1],@x[$d1],24)", 104 "&ror_32 (@x[$d2],@x[$d2],24)", 105 "&ror_32 (@x[$d3],@x[$d3],24)", 106 107 "&add_32 (@x[$c0],@x[$c0],@x[$d0])", 108 "&add_32 (@x[$c1],@x[$c1],@x[$d1])", 109 "&add_32 (@x[$c2],@x[$c2],@x[$d2])", 110 "&add_32 (@x[$c3],@x[$c3],@x[$d3])", 111 "&eor_32 (@x[$b0],@x[$b0],@x[$c0])", 112 "&eor_32 (@x[$b1],@x[$b1],@x[$c1])", 113 "&eor_32 (@x[$b2],@x[$b2],@x[$c2])", 114 "&eor_32 (@x[$b3],@x[$b3],@x[$c3])", 115 "&ror_32 (@x[$b0],@x[$b0],25)", 116 "&ror_32 (@x[$b1],@x[$b1],25)", 117 "&ror_32 (@x[$b2],@x[$b2],25)", 118 "&ror_32 (@x[$b3],@x[$b3],25)" 119 ); 120} 121 122$code.=<<___; 123#include <GFp/arm_arch.h> 124 125.extern GFp_armcap_P 126.hidden GFp_armcap_P 127 128.section .rodata 129 130.align 5 131.Lsigma: 132.quad 0x3320646e61707865,0x6b20657479622d32 // endian-neutral 133.Lone: 134.long 1,0,0,0 135.asciz "ChaCha20 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>" 136 137.text 138 139.globl GFp_ChaCha20_ctr32 140.type GFp_ChaCha20_ctr32,%function 141.align 5 142GFp_ChaCha20_ctr32: 143 AARCH64_VALID_CALL_TARGET 144 cbz $len,.Labort 145#if __has_feature(hwaddress_sanitizer) && __clang_major__ >= 10 146 adrp @x[0],:pg_hi21_nc:GFp_armcap_P 147#else 148 adrp @x[0],:pg_hi21:GFp_armcap_P 149#endif 150 cmp $len,#192 151 b.lo .Lshort 152 ldr w17,[@x[0],:lo12:GFp_armcap_P] 153 tst w17,#ARMV7_NEON 154 b.ne ChaCha20_neon 155 156.Lshort: 157 AARCH64_SIGN_LINK_REGISTER 158 stp x29,x30,[sp,#-96]! 159 add x29,sp,#0 160 161 adrp @x[0],:pg_hi21:.Lsigma 162 add @x[0],@x[0],:lo12:.Lsigma 163 stp x19,x20,[sp,#16] 164 stp x21,x22,[sp,#32] 165 stp x23,x24,[sp,#48] 166 stp x25,x26,[sp,#64] 167 stp x27,x28,[sp,#80] 168 sub sp,sp,#64 169 170 ldp @d[0],@d[1],[@x[0]] // load sigma 171 ldp @d[2],@d[3],[$key] // load key 172 ldp @d[4],@d[5],[$key,#16] 173 ldp @d[6],@d[7],[$ctr] // load counter 174#ifdef __ARMEB__ 175 ror @d[2],@d[2],#32 176 ror @d[3],@d[3],#32 177 ror @d[4],@d[4],#32 178 ror @d[5],@d[5],#32 179 ror @d[6],@d[6],#32 180 ror @d[7],@d[7],#32 181#endif 182 183.Loop_outer: 184 mov.32 @x[0],@d[0] // unpack key block 185 lsr @x[1],@d[0],#32 186 mov.32 @x[2],@d[1] 187 lsr @x[3],@d[1],#32 188 mov.32 @x[4],@d[2] 189 lsr @x[5],@d[2],#32 190 mov.32 @x[6],@d[3] 191 lsr @x[7],@d[3],#32 192 mov.32 @x[8],@d[4] 193 lsr @x[9],@d[4],#32 194 mov.32 @x[10],@d[5] 195 lsr @x[11],@d[5],#32 196 mov.32 @x[12],@d[6] 197 lsr @x[13],@d[6],#32 198 mov.32 @x[14],@d[7] 199 lsr @x[15],@d[7],#32 200 201 mov $ctr,#10 202 subs $len,$len,#64 203.Loop: 204 sub $ctr,$ctr,#1 205___ 206 foreach (&ROUND(0, 4, 8,12)) { eval; } 207 foreach (&ROUND(0, 5,10,15)) { eval; } 208$code.=<<___; 209 cbnz $ctr,.Loop 210 211 add.32 @x[0],@x[0],@d[0] // accumulate key block 212 add @x[1],@x[1],@d[0],lsr#32 213 add.32 @x[2],@x[2],@d[1] 214 add @x[3],@x[3],@d[1],lsr#32 215 add.32 @x[4],@x[4],@d[2] 216 add @x[5],@x[5],@d[2],lsr#32 217 add.32 @x[6],@x[6],@d[3] 218 add @x[7],@x[7],@d[3],lsr#32 219 add.32 @x[8],@x[8],@d[4] 220 add @x[9],@x[9],@d[4],lsr#32 221 add.32 @x[10],@x[10],@d[5] 222 add @x[11],@x[11],@d[5],lsr#32 223 add.32 @x[12],@x[12],@d[6] 224 add @x[13],@x[13],@d[6],lsr#32 225 add.32 @x[14],@x[14],@d[7] 226 add @x[15],@x[15],@d[7],lsr#32 227 228 b.lo .Ltail 229 230 add @x[0],@x[0],@x[1],lsl#32 // pack 231 add @x[2],@x[2],@x[3],lsl#32 232 ldp @x[1],@x[3],[$inp,#0] // load input 233 add @x[4],@x[4],@x[5],lsl#32 234 add @x[6],@x[6],@x[7],lsl#32 235 ldp @x[5],@x[7],[$inp,#16] 236 add @x[8],@x[8],@x[9],lsl#32 237 add @x[10],@x[10],@x[11],lsl#32 238 ldp @x[9],@x[11],[$inp,#32] 239 add @x[12],@x[12],@x[13],lsl#32 240 add @x[14],@x[14],@x[15],lsl#32 241 ldp @x[13],@x[15],[$inp,#48] 242 add $inp,$inp,#64 243#ifdef __ARMEB__ 244 rev @x[0],@x[0] 245 rev @x[2],@x[2] 246 rev @x[4],@x[4] 247 rev @x[6],@x[6] 248 rev @x[8],@x[8] 249 rev @x[10],@x[10] 250 rev @x[12],@x[12] 251 rev @x[14],@x[14] 252#endif 253 eor @x[0],@x[0],@x[1] 254 eor @x[2],@x[2],@x[3] 255 eor @x[4],@x[4],@x[5] 256 eor @x[6],@x[6],@x[7] 257 eor @x[8],@x[8],@x[9] 258 eor @x[10],@x[10],@x[11] 259 eor @x[12],@x[12],@x[13] 260 eor @x[14],@x[14],@x[15] 261 262 stp @x[0],@x[2],[$out,#0] // store output 263 add @d[6],@d[6],#1 // increment counter 264 stp @x[4],@x[6],[$out,#16] 265 stp @x[8],@x[10],[$out,#32] 266 stp @x[12],@x[14],[$out,#48] 267 add $out,$out,#64 268 269 b.hi .Loop_outer 270 271 ldp x19,x20,[x29,#16] 272 add sp,sp,#64 273 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],#96 278 AARCH64_VALIDATE_LINK_REGISTER 279.Labort: 280 ret 281 282.align 4 283.Ltail: 284 add $len,$len,#64 285.Less_than_64: 286 sub $out,$out,#1 287 add $inp,$inp,$len 288 add $out,$out,$len 289 add $ctr,sp,$len 290 neg $len,$len 291 292 add @x[0],@x[0],@x[1],lsl#32 // pack 293 add @x[2],@x[2],@x[3],lsl#32 294 add @x[4],@x[4],@x[5],lsl#32 295 add @x[6],@x[6],@x[7],lsl#32 296 add @x[8],@x[8],@x[9],lsl#32 297 add @x[10],@x[10],@x[11],lsl#32 298 add @x[12],@x[12],@x[13],lsl#32 299 add @x[14],@x[14],@x[15],lsl#32 300#ifdef __ARMEB__ 301 rev @x[0],@x[0] 302 rev @x[2],@x[2] 303 rev @x[4],@x[4] 304 rev @x[6],@x[6] 305 rev @x[8],@x[8] 306 rev @x[10],@x[10] 307 rev @x[12],@x[12] 308 rev @x[14],@x[14] 309#endif 310 stp @x[0],@x[2],[sp,#0] 311 stp @x[4],@x[6],[sp,#16] 312 stp @x[8],@x[10],[sp,#32] 313 stp @x[12],@x[14],[sp,#48] 314 315.Loop_tail: 316 ldrb w10,[$inp,$len] 317 ldrb w11,[$ctr,$len] 318 add $len,$len,#1 319 eor w10,w10,w11 320 strb w10,[$out,$len] 321 cbnz $len,.Loop_tail 322 323 stp xzr,xzr,[sp,#0] 324 stp xzr,xzr,[sp,#16] 325 stp xzr,xzr,[sp,#32] 326 stp xzr,xzr,[sp,#48] 327 328 ldp x19,x20,[x29,#16] 329 add sp,sp,#64 330 ldp x21,x22,[x29,#32] 331 ldp x23,x24,[x29,#48] 332 ldp x25,x26,[x29,#64] 333 ldp x27,x28,[x29,#80] 334 ldp x29,x30,[sp],#96 335 AARCH64_VALIDATE_LINK_REGISTER 336 ret 337.size GFp_ChaCha20_ctr32,.-GFp_ChaCha20_ctr32 338___ 339 340{{{ 341my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2,$T0,$T1,$T2,$T3) = 342 map("v$_.4s",(0..7,16..23)); 343my (@K)=map("v$_.4s",(24..30)); 344my $ONE="v31.4s"; 345 346sub NEONROUND { 347my $odd = pop; 348my ($a,$b,$c,$d,$t)=@_; 349 350 ( 351 "&add ('$a','$a','$b')", 352 "&eor ('$d','$d','$a')", 353 "&rev32_16 ('$d','$d')", # vrot ($d,16) 354 355 "&add ('$c','$c','$d')", 356 "&eor ('$t','$b','$c')", 357 "&ushr ('$b','$t',20)", 358 "&sli ('$b','$t',12)", 359 360 "&add ('$a','$a','$b')", 361 "&eor ('$t','$d','$a')", 362 "&ushr ('$d','$t',24)", 363 "&sli ('$d','$t',8)", 364 365 "&add ('$c','$c','$d')", 366 "&eor ('$t','$b','$c')", 367 "&ushr ('$b','$t',25)", 368 "&sli ('$b','$t',7)", 369 370 "&ext ('$c','$c','$c',8)", 371 "&ext ('$d','$d','$d',$odd?4:12)", 372 "&ext ('$b','$b','$b',$odd?12:4)" 373 ); 374} 375 376$code.=<<___; 377 378.type ChaCha20_neon,%function 379.align 5 380ChaCha20_neon: 381 AARCH64_SIGN_LINK_REGISTER 382 stp x29,x30,[sp,#-96]! 383 add x29,sp,#0 384 385 adrp @x[0],:pg_hi21:.Lsigma 386 add @x[0],@x[0],:lo12:.Lsigma 387 stp x19,x20,[sp,#16] 388 stp x21,x22,[sp,#32] 389 stp x23,x24,[sp,#48] 390 stp x25,x26,[sp,#64] 391 stp x27,x28,[sp,#80] 392 cmp $len,#512 393 b.hs .L512_or_more_neon 394 395 sub sp,sp,#64 396 397 ldp @d[0],@d[1],[@x[0]] // load sigma 398 ld1 {@K[0]},[@x[0]],#16 399 ldp @d[2],@d[3],[$key] // load key 400 ldp @d[4],@d[5],[$key,#16] 401 ld1 {@K[1],@K[2]},[$key] 402 ldp @d[6],@d[7],[$ctr] // load counter 403 ld1 {@K[3]},[$ctr] 404 ld1 {$ONE},[@x[0]] 405#ifdef __ARMEB__ 406 rev64 @K[0],@K[0] 407 ror @d[2],@d[2],#32 408 ror @d[3],@d[3],#32 409 ror @d[4],@d[4],#32 410 ror @d[5],@d[5],#32 411 ror @d[6],@d[6],#32 412 ror @d[7],@d[7],#32 413#endif 414 add @K[3],@K[3],$ONE // += 1 415 add @K[4],@K[3],$ONE 416 add @K[5],@K[4],$ONE 417 shl $ONE,$ONE,#2 // 1 -> 4 418 419.Loop_outer_neon: 420 mov.32 @x[0],@d[0] // unpack key block 421 lsr @x[1],@d[0],#32 422 mov $A0,@K[0] 423 mov.32 @x[2],@d[1] 424 lsr @x[3],@d[1],#32 425 mov $A1,@K[0] 426 mov.32 @x[4],@d[2] 427 lsr @x[5],@d[2],#32 428 mov $A2,@K[0] 429 mov.32 @x[6],@d[3] 430 mov $B0,@K[1] 431 lsr @x[7],@d[3],#32 432 mov $B1,@K[1] 433 mov.32 @x[8],@d[4] 434 mov $B2,@K[1] 435 lsr @x[9],@d[4],#32 436 mov $D0,@K[3] 437 mov.32 @x[10],@d[5] 438 mov $D1,@K[4] 439 lsr @x[11],@d[5],#32 440 mov $D2,@K[5] 441 mov.32 @x[12],@d[6] 442 mov $C0,@K[2] 443 lsr @x[13],@d[6],#32 444 mov $C1,@K[2] 445 mov.32 @x[14],@d[7] 446 mov $C2,@K[2] 447 lsr @x[15],@d[7],#32 448 449 mov $ctr,#10 450 subs $len,$len,#256 451.Loop_neon: 452 sub $ctr,$ctr,#1 453___ 454 my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0); 455 my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0); 456 my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0); 457 my @thread3=&ROUND(0,4,8,12); 458 459 foreach (@thread0) { 460 eval; eval(shift(@thread3)); 461 eval(shift(@thread1)); eval(shift(@thread3)); 462 eval(shift(@thread2)); eval(shift(@thread3)); 463 } 464 465 @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1); 466 @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1); 467 @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1); 468 @thread3=&ROUND(0,5,10,15); 469 470 foreach (@thread0) { 471 eval; eval(shift(@thread3)); 472 eval(shift(@thread1)); eval(shift(@thread3)); 473 eval(shift(@thread2)); eval(shift(@thread3)); 474 } 475$code.=<<___; 476 cbnz $ctr,.Loop_neon 477 478 add.32 @x[0],@x[0],@d[0] // accumulate key block 479 add $A0,$A0,@K[0] 480 add @x[1],@x[1],@d[0],lsr#32 481 add $A1,$A1,@K[0] 482 add.32 @x[2],@x[2],@d[1] 483 add $A2,$A2,@K[0] 484 add @x[3],@x[3],@d[1],lsr#32 485 add $C0,$C0,@K[2] 486 add.32 @x[4],@x[4],@d[2] 487 add $C1,$C1,@K[2] 488 add @x[5],@x[5],@d[2],lsr#32 489 add $C2,$C2,@K[2] 490 add.32 @x[6],@x[6],@d[3] 491 add $D0,$D0,@K[3] 492 add @x[7],@x[7],@d[3],lsr#32 493 add.32 @x[8],@x[8],@d[4] 494 add $D1,$D1,@K[4] 495 add @x[9],@x[9],@d[4],lsr#32 496 add.32 @x[10],@x[10],@d[5] 497 add $D2,$D2,@K[5] 498 add @x[11],@x[11],@d[5],lsr#32 499 add.32 @x[12],@x[12],@d[6] 500 add $B0,$B0,@K[1] 501 add @x[13],@x[13],@d[6],lsr#32 502 add.32 @x[14],@x[14],@d[7] 503 add $B1,$B1,@K[1] 504 add @x[15],@x[15],@d[7],lsr#32 505 add $B2,$B2,@K[1] 506 507 b.lo .Ltail_neon 508 509 add @x[0],@x[0],@x[1],lsl#32 // pack 510 add @x[2],@x[2],@x[3],lsl#32 511 ldp @x[1],@x[3],[$inp,#0] // load input 512 add @x[4],@x[4],@x[5],lsl#32 513 add @x[6],@x[6],@x[7],lsl#32 514 ldp @x[5],@x[7],[$inp,#16] 515 add @x[8],@x[8],@x[9],lsl#32 516 add @x[10],@x[10],@x[11],lsl#32 517 ldp @x[9],@x[11],[$inp,#32] 518 add @x[12],@x[12],@x[13],lsl#32 519 add @x[14],@x[14],@x[15],lsl#32 520 ldp @x[13],@x[15],[$inp,#48] 521 add $inp,$inp,#64 522#ifdef __ARMEB__ 523 rev @x[0],@x[0] 524 rev @x[2],@x[2] 525 rev @x[4],@x[4] 526 rev @x[6],@x[6] 527 rev @x[8],@x[8] 528 rev @x[10],@x[10] 529 rev @x[12],@x[12] 530 rev @x[14],@x[14] 531#endif 532 ld1.8 {$T0-$T3},[$inp],#64 533 eor @x[0],@x[0],@x[1] 534 eor @x[2],@x[2],@x[3] 535 eor @x[4],@x[4],@x[5] 536 eor @x[6],@x[6],@x[7] 537 eor @x[8],@x[8],@x[9] 538 eor $A0,$A0,$T0 539 eor @x[10],@x[10],@x[11] 540 eor $B0,$B0,$T1 541 eor @x[12],@x[12],@x[13] 542 eor $C0,$C0,$T2 543 eor @x[14],@x[14],@x[15] 544 eor $D0,$D0,$T3 545 ld1.8 {$T0-$T3},[$inp],#64 546 547 stp @x[0],@x[2],[$out,#0] // store output 548 add @d[6],@d[6],#4 // increment counter 549 stp @x[4],@x[6],[$out,#16] 550 add @K[3],@K[3],$ONE // += 4 551 stp @x[8],@x[10],[$out,#32] 552 add @K[4],@K[4],$ONE 553 stp @x[12],@x[14],[$out,#48] 554 add @K[5],@K[5],$ONE 555 add $out,$out,#64 556 557 st1.8 {$A0-$D0},[$out],#64 558 ld1.8 {$A0-$D0},[$inp],#64 559 560 eor $A1,$A1,$T0 561 eor $B1,$B1,$T1 562 eor $C1,$C1,$T2 563 eor $D1,$D1,$T3 564 st1.8 {$A1-$D1},[$out],#64 565 566 eor $A2,$A2,$A0 567 eor $B2,$B2,$B0 568 eor $C2,$C2,$C0 569 eor $D2,$D2,$D0 570 st1.8 {$A2-$D2},[$out],#64 571 572 b.hi .Loop_outer_neon 573 574 ldp x19,x20,[x29,#16] 575 add sp,sp,#64 576 ldp x21,x22,[x29,#32] 577 ldp x23,x24,[x29,#48] 578 ldp x25,x26,[x29,#64] 579 ldp x27,x28,[x29,#80] 580 ldp x29,x30,[sp],#96 581 AARCH64_VALIDATE_LINK_REGISTER 582 ret 583 584.Ltail_neon: 585 add $len,$len,#256 586 cmp $len,#64 587 b.lo .Less_than_64 588 589 add @x[0],@x[0],@x[1],lsl#32 // pack 590 add @x[2],@x[2],@x[3],lsl#32 591 ldp @x[1],@x[3],[$inp,#0] // load input 592 add @x[4],@x[4],@x[5],lsl#32 593 add @x[6],@x[6],@x[7],lsl#32 594 ldp @x[5],@x[7],[$inp,#16] 595 add @x[8],@x[8],@x[9],lsl#32 596 add @x[10],@x[10],@x[11],lsl#32 597 ldp @x[9],@x[11],[$inp,#32] 598 add @x[12],@x[12],@x[13],lsl#32 599 add @x[14],@x[14],@x[15],lsl#32 600 ldp @x[13],@x[15],[$inp,#48] 601 add $inp,$inp,#64 602#ifdef __ARMEB__ 603 rev @x[0],@x[0] 604 rev @x[2],@x[2] 605 rev @x[4],@x[4] 606 rev @x[6],@x[6] 607 rev @x[8],@x[8] 608 rev @x[10],@x[10] 609 rev @x[12],@x[12] 610 rev @x[14],@x[14] 611#endif 612 eor @x[0],@x[0],@x[1] 613 eor @x[2],@x[2],@x[3] 614 eor @x[4],@x[4],@x[5] 615 eor @x[6],@x[6],@x[7] 616 eor @x[8],@x[8],@x[9] 617 eor @x[10],@x[10],@x[11] 618 eor @x[12],@x[12],@x[13] 619 eor @x[14],@x[14],@x[15] 620 621 stp @x[0],@x[2],[$out,#0] // store output 622 add @d[6],@d[6],#4 // increment counter 623 stp @x[4],@x[6],[$out,#16] 624 stp @x[8],@x[10],[$out,#32] 625 stp @x[12],@x[14],[$out,#48] 626 add $out,$out,#64 627 b.eq .Ldone_neon 628 sub $len,$len,#64 629 cmp $len,#64 630 b.lo .Less_than_128 631 632 ld1.8 {$T0-$T3},[$inp],#64 633 eor $A0,$A0,$T0 634 eor $B0,$B0,$T1 635 eor $C0,$C0,$T2 636 eor $D0,$D0,$T3 637 st1.8 {$A0-$D0},[$out],#64 638 b.eq .Ldone_neon 639 sub $len,$len,#64 640 cmp $len,#64 641 b.lo .Less_than_192 642 643 ld1.8 {$T0-$T3},[$inp],#64 644 eor $A1,$A1,$T0 645 eor $B1,$B1,$T1 646 eor $C1,$C1,$T2 647 eor $D1,$D1,$T3 648 st1.8 {$A1-$D1},[$out],#64 649 b.eq .Ldone_neon 650 sub $len,$len,#64 651 652 st1.8 {$A2-$D2},[sp] 653 b .Last_neon 654 655.Less_than_128: 656 st1.8 {$A0-$D0},[sp] 657 b .Last_neon 658.Less_than_192: 659 st1.8 {$A1-$D1},[sp] 660 b .Last_neon 661 662.align 4 663.Last_neon: 664 sub $out,$out,#1 665 add $inp,$inp,$len 666 add $out,$out,$len 667 add $ctr,sp,$len 668 neg $len,$len 669 670.Loop_tail_neon: 671 ldrb w10,[$inp,$len] 672 ldrb w11,[$ctr,$len] 673 add $len,$len,#1 674 eor w10,w10,w11 675 strb w10,[$out,$len] 676 cbnz $len,.Loop_tail_neon 677 678 stp xzr,xzr,[sp,#0] 679 stp xzr,xzr,[sp,#16] 680 stp xzr,xzr,[sp,#32] 681 stp xzr,xzr,[sp,#48] 682 683.Ldone_neon: 684 ldp x19,x20,[x29,#16] 685 add sp,sp,#64 686 ldp x21,x22,[x29,#32] 687 ldp x23,x24,[x29,#48] 688 ldp x25,x26,[x29,#64] 689 ldp x27,x28,[x29,#80] 690 ldp x29,x30,[sp],#96 691 AARCH64_VALIDATE_LINK_REGISTER 692 ret 693.size ChaCha20_neon,.-ChaCha20_neon 694___ 695{ 696my ($T0,$T1,$T2,$T3,$T4,$T5)=@K; 697my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2, 698 $A3,$B3,$C3,$D3,$A4,$B4,$C4,$D4,$A5,$B5,$C5,$D5) = map("v$_.4s",(0..23)); 699 700$code.=<<___; 701.type ChaCha20_512_neon,%function 702.align 5 703ChaCha20_512_neon: 704 AARCH64_SIGN_LINK_REGISTER 705 stp x29,x30,[sp,#-96]! 706 add x29,sp,#0 707 708 adrp @x[0],:pg_hi21:.Lsigma 709 add @x[0],@x[0],:lo12:.Lsigma 710 stp x19,x20,[sp,#16] 711 stp x21,x22,[sp,#32] 712 stp x23,x24,[sp,#48] 713 stp x25,x26,[sp,#64] 714 stp x27,x28,[sp,#80] 715 716.L512_or_more_neon: 717 sub sp,sp,#128+64 718 719 ldp @d[0],@d[1],[@x[0]] // load sigma 720 ld1 {@K[0]},[@x[0]],#16 721 ldp @d[2],@d[3],[$key] // load key 722 ldp @d[4],@d[5],[$key,#16] 723 ld1 {@K[1],@K[2]},[$key] 724 ldp @d[6],@d[7],[$ctr] // load counter 725 ld1 {@K[3]},[$ctr] 726 ld1 {$ONE},[@x[0]] 727#ifdef __ARMEB__ 728 rev64 @K[0],@K[0] 729 ror @d[2],@d[2],#32 730 ror @d[3],@d[3],#32 731 ror @d[4],@d[4],#32 732 ror @d[5],@d[5],#32 733 ror @d[6],@d[6],#32 734 ror @d[7],@d[7],#32 735#endif 736 add @K[3],@K[3],$ONE // += 1 737 stp @K[0],@K[1],[sp,#0] // off-load key block, invariant part 738 add @K[3],@K[3],$ONE // not typo 739 str @K[2],[sp,#32] 740 add @K[4],@K[3],$ONE 741 add @K[5],@K[4],$ONE 742 add @K[6],@K[5],$ONE 743 shl $ONE,$ONE,#2 // 1 -> 4 744 745 stp d8,d9,[sp,#128+0] // meet ABI requirements 746 stp d10,d11,[sp,#128+16] 747 stp d12,d13,[sp,#128+32] 748 stp d14,d15,[sp,#128+48] 749 750 sub $len,$len,#512 // not typo 751 752.Loop_outer_512_neon: 753 mov $A0,@K[0] 754 mov $A1,@K[0] 755 mov $A2,@K[0] 756 mov $A3,@K[0] 757 mov $A4,@K[0] 758 mov $A5,@K[0] 759 mov $B0,@K[1] 760 mov.32 @x[0],@d[0] // unpack key block 761 mov $B1,@K[1] 762 lsr @x[1],@d[0],#32 763 mov $B2,@K[1] 764 mov.32 @x[2],@d[1] 765 mov $B3,@K[1] 766 lsr @x[3],@d[1],#32 767 mov $B4,@K[1] 768 mov.32 @x[4],@d[2] 769 mov $B5,@K[1] 770 lsr @x[5],@d[2],#32 771 mov $D0,@K[3] 772 mov.32 @x[6],@d[3] 773 mov $D1,@K[4] 774 lsr @x[7],@d[3],#32 775 mov $D2,@K[5] 776 mov.32 @x[8],@d[4] 777 mov $D3,@K[6] 778 lsr @x[9],@d[4],#32 779 mov $C0,@K[2] 780 mov.32 @x[10],@d[5] 781 mov $C1,@K[2] 782 lsr @x[11],@d[5],#32 783 add $D4,$D0,$ONE // +4 784 mov.32 @x[12],@d[6] 785 add $D5,$D1,$ONE // +4 786 lsr @x[13],@d[6],#32 787 mov $C2,@K[2] 788 mov.32 @x[14],@d[7] 789 mov $C3,@K[2] 790 lsr @x[15],@d[7],#32 791 mov $C4,@K[2] 792 stp @K[3],@K[4],[sp,#48] // off-load key block, variable part 793 mov $C5,@K[2] 794 str @K[5],[sp,#80] 795 796 mov $ctr,#5 797 subs $len,$len,#512 798.Loop_upper_neon: 799 sub $ctr,$ctr,#1 800___ 801 my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0); 802 my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0); 803 my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0); 804 my @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0); 805 my @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0); 806 my @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0); 807 my @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15)); 808 my $diff = ($#thread0+1)*6 - $#thread67 - 1; 809 my $i = 0; 810 811 foreach (@thread0) { 812 eval; eval(shift(@thread67)); 813 eval(shift(@thread1)); eval(shift(@thread67)); 814 eval(shift(@thread2)); eval(shift(@thread67)); 815 eval(shift(@thread3)); eval(shift(@thread67)); 816 eval(shift(@thread4)); eval(shift(@thread67)); 817 eval(shift(@thread5)); eval(shift(@thread67)); 818 } 819 820 @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1); 821 @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1); 822 @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1); 823 @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1); 824 @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1); 825 @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1); 826 @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15)); 827 828 foreach (@thread0) { 829 eval; eval(shift(@thread67)); 830 eval(shift(@thread1)); eval(shift(@thread67)); 831 eval(shift(@thread2)); eval(shift(@thread67)); 832 eval(shift(@thread3)); eval(shift(@thread67)); 833 eval(shift(@thread4)); eval(shift(@thread67)); 834 eval(shift(@thread5)); eval(shift(@thread67)); 835 } 836$code.=<<___; 837 cbnz $ctr,.Loop_upper_neon 838 839 add.32 @x[0],@x[0],@d[0] // accumulate key block 840 add @x[1],@x[1],@d[0],lsr#32 841 add.32 @x[2],@x[2],@d[1] 842 add @x[3],@x[3],@d[1],lsr#32 843 add.32 @x[4],@x[4],@d[2] 844 add @x[5],@x[5],@d[2],lsr#32 845 add.32 @x[6],@x[6],@d[3] 846 add @x[7],@x[7],@d[3],lsr#32 847 add.32 @x[8],@x[8],@d[4] 848 add @x[9],@x[9],@d[4],lsr#32 849 add.32 @x[10],@x[10],@d[5] 850 add @x[11],@x[11],@d[5],lsr#32 851 add.32 @x[12],@x[12],@d[6] 852 add @x[13],@x[13],@d[6],lsr#32 853 add.32 @x[14],@x[14],@d[7] 854 add @x[15],@x[15],@d[7],lsr#32 855 856 add @x[0],@x[0],@x[1],lsl#32 // pack 857 add @x[2],@x[2],@x[3],lsl#32 858 ldp @x[1],@x[3],[$inp,#0] // load input 859 add @x[4],@x[4],@x[5],lsl#32 860 add @x[6],@x[6],@x[7],lsl#32 861 ldp @x[5],@x[7],[$inp,#16] 862 add @x[8],@x[8],@x[9],lsl#32 863 add @x[10],@x[10],@x[11],lsl#32 864 ldp @x[9],@x[11],[$inp,#32] 865 add @x[12],@x[12],@x[13],lsl#32 866 add @x[14],@x[14],@x[15],lsl#32 867 ldp @x[13],@x[15],[$inp,#48] 868 add $inp,$inp,#64 869#ifdef __ARMEB__ 870 rev @x[0],@x[0] 871 rev @x[2],@x[2] 872 rev @x[4],@x[4] 873 rev @x[6],@x[6] 874 rev @x[8],@x[8] 875 rev @x[10],@x[10] 876 rev @x[12],@x[12] 877 rev @x[14],@x[14] 878#endif 879 eor @x[0],@x[0],@x[1] 880 eor @x[2],@x[2],@x[3] 881 eor @x[4],@x[4],@x[5] 882 eor @x[6],@x[6],@x[7] 883 eor @x[8],@x[8],@x[9] 884 eor @x[10],@x[10],@x[11] 885 eor @x[12],@x[12],@x[13] 886 eor @x[14],@x[14],@x[15] 887 888 stp @x[0],@x[2],[$out,#0] // store output 889 add @d[6],@d[6],#1 // increment counter 890 mov.32 @x[0],@d[0] // unpack key block 891 lsr @x[1],@d[0],#32 892 stp @x[4],@x[6],[$out,#16] 893 mov.32 @x[2],@d[1] 894 lsr @x[3],@d[1],#32 895 stp @x[8],@x[10],[$out,#32] 896 mov.32 @x[4],@d[2] 897 lsr @x[5],@d[2],#32 898 stp @x[12],@x[14],[$out,#48] 899 add $out,$out,#64 900 mov.32 @x[6],@d[3] 901 lsr @x[7],@d[3],#32 902 mov.32 @x[8],@d[4] 903 lsr @x[9],@d[4],#32 904 mov.32 @x[10],@d[5] 905 lsr @x[11],@d[5],#32 906 mov.32 @x[12],@d[6] 907 lsr @x[13],@d[6],#32 908 mov.32 @x[14],@d[7] 909 lsr @x[15],@d[7],#32 910 911 mov $ctr,#5 912.Loop_lower_neon: 913 sub $ctr,$ctr,#1 914___ 915 @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0); 916 @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0); 917 @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0); 918 @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0); 919 @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0); 920 @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0); 921 @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15)); 922 923 foreach (@thread0) { 924 eval; eval(shift(@thread67)); 925 eval(shift(@thread1)); eval(shift(@thread67)); 926 eval(shift(@thread2)); eval(shift(@thread67)); 927 eval(shift(@thread3)); eval(shift(@thread67)); 928 eval(shift(@thread4)); eval(shift(@thread67)); 929 eval(shift(@thread5)); eval(shift(@thread67)); 930 } 931 932 @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1); 933 @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1); 934 @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1); 935 @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1); 936 @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1); 937 @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1); 938 @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15)); 939 940 foreach (@thread0) { 941 eval; eval(shift(@thread67)); 942 eval(shift(@thread1)); eval(shift(@thread67)); 943 eval(shift(@thread2)); eval(shift(@thread67)); 944 eval(shift(@thread3)); eval(shift(@thread67)); 945 eval(shift(@thread4)); eval(shift(@thread67)); 946 eval(shift(@thread5)); eval(shift(@thread67)); 947 } 948$code.=<<___; 949 cbnz $ctr,.Loop_lower_neon 950 951 add.32 @x[0],@x[0],@d[0] // accumulate key block 952 ldp @K[0],@K[1],[sp,#0] 953 add @x[1],@x[1],@d[0],lsr#32 954 ldp @K[2],@K[3],[sp,#32] 955 add.32 @x[2],@x[2],@d[1] 956 ldp @K[4],@K[5],[sp,#64] 957 add @x[3],@x[3],@d[1],lsr#32 958 add $A0,$A0,@K[0] 959 add.32 @x[4],@x[4],@d[2] 960 add $A1,$A1,@K[0] 961 add @x[5],@x[5],@d[2],lsr#32 962 add $A2,$A2,@K[0] 963 add.32 @x[6],@x[6],@d[3] 964 add $A3,$A3,@K[0] 965 add @x[7],@x[7],@d[3],lsr#32 966 add $A4,$A4,@K[0] 967 add.32 @x[8],@x[8],@d[4] 968 add $A5,$A5,@K[0] 969 add @x[9],@x[9],@d[4],lsr#32 970 add $C0,$C0,@K[2] 971 add.32 @x[10],@x[10],@d[5] 972 add $C1,$C1,@K[2] 973 add @x[11],@x[11],@d[5],lsr#32 974 add $C2,$C2,@K[2] 975 add.32 @x[12],@x[12],@d[6] 976 add $C3,$C3,@K[2] 977 add @x[13],@x[13],@d[6],lsr#32 978 add $C4,$C4,@K[2] 979 add.32 @x[14],@x[14],@d[7] 980 add $C5,$C5,@K[2] 981 add @x[15],@x[15],@d[7],lsr#32 982 add $D4,$D4,$ONE // +4 983 add @x[0],@x[0],@x[1],lsl#32 // pack 984 add $D5,$D5,$ONE // +4 985 add @x[2],@x[2],@x[3],lsl#32 986 add $D0,$D0,@K[3] 987 ldp @x[1],@x[3],[$inp,#0] // load input 988 add $D1,$D1,@K[4] 989 add @x[4],@x[4],@x[5],lsl#32 990 add $D2,$D2,@K[5] 991 add @x[6],@x[6],@x[7],lsl#32 992 add $D3,$D3,@K[6] 993 ldp @x[5],@x[7],[$inp,#16] 994 add $D4,$D4,@K[3] 995 add @x[8],@x[8],@x[9],lsl#32 996 add $D5,$D5,@K[4] 997 add @x[10],@x[10],@x[11],lsl#32 998 add $B0,$B0,@K[1] 999 ldp @x[9],@x[11],[$inp,#32] 1000 add $B1,$B1,@K[1] 1001 add @x[12],@x[12],@x[13],lsl#32 1002 add $B2,$B2,@K[1] 1003 add @x[14],@x[14],@x[15],lsl#32 1004 add $B3,$B3,@K[1] 1005 ldp @x[13],@x[15],[$inp,#48] 1006 add $B4,$B4,@K[1] 1007 add $inp,$inp,#64 1008 add $B5,$B5,@K[1] 1009 1010#ifdef __ARMEB__ 1011 rev @x[0],@x[0] 1012 rev @x[2],@x[2] 1013 rev @x[4],@x[4] 1014 rev @x[6],@x[6] 1015 rev @x[8],@x[8] 1016 rev @x[10],@x[10] 1017 rev @x[12],@x[12] 1018 rev @x[14],@x[14] 1019#endif 1020 ld1.8 {$T0-$T3},[$inp],#64 1021 eor @x[0],@x[0],@x[1] 1022 eor @x[2],@x[2],@x[3] 1023 eor @x[4],@x[4],@x[5] 1024 eor @x[6],@x[6],@x[7] 1025 eor @x[8],@x[8],@x[9] 1026 eor $A0,$A0,$T0 1027 eor @x[10],@x[10],@x[11] 1028 eor $B0,$B0,$T1 1029 eor @x[12],@x[12],@x[13] 1030 eor $C0,$C0,$T2 1031 eor @x[14],@x[14],@x[15] 1032 eor $D0,$D0,$T3 1033 ld1.8 {$T0-$T3},[$inp],#64 1034 1035 stp @x[0],@x[2],[$out,#0] // store output 1036 add @d[6],@d[6],#7 // increment counter 1037 stp @x[4],@x[6],[$out,#16] 1038 stp @x[8],@x[10],[$out,#32] 1039 stp @x[12],@x[14],[$out,#48] 1040 add $out,$out,#64 1041 st1.8 {$A0-$D0},[$out],#64 1042 1043 ld1.8 {$A0-$D0},[$inp],#64 1044 eor $A1,$A1,$T0 1045 eor $B1,$B1,$T1 1046 eor $C1,$C1,$T2 1047 eor $D1,$D1,$T3 1048 st1.8 {$A1-$D1},[$out],#64 1049 1050 ld1.8 {$A1-$D1},[$inp],#64 1051 eor $A2,$A2,$A0 1052 ldp @K[0],@K[1],[sp,#0] 1053 eor $B2,$B2,$B0 1054 ldp @K[2],@K[3],[sp,#32] 1055 eor $C2,$C2,$C0 1056 eor $D2,$D2,$D0 1057 st1.8 {$A2-$D2},[$out],#64 1058 1059 ld1.8 {$A2-$D2},[$inp],#64 1060 eor $A3,$A3,$A1 1061 eor $B3,$B3,$B1 1062 eor $C3,$C3,$C1 1063 eor $D3,$D3,$D1 1064 st1.8 {$A3-$D3},[$out],#64 1065 1066 ld1.8 {$A3-$D3},[$inp],#64 1067 eor $A4,$A4,$A2 1068 eor $B4,$B4,$B2 1069 eor $C4,$C4,$C2 1070 eor $D4,$D4,$D2 1071 st1.8 {$A4-$D4},[$out],#64 1072 1073 shl $A0,$ONE,#1 // 4 -> 8 1074 eor $A5,$A5,$A3 1075 eor $B5,$B5,$B3 1076 eor $C5,$C5,$C3 1077 eor $D5,$D5,$D3 1078 st1.8 {$A5-$D5},[$out],#64 1079 1080 add @K[3],@K[3],$A0 // += 8 1081 add @K[4],@K[4],$A0 1082 add @K[5],@K[5],$A0 1083 add @K[6],@K[6],$A0 1084 1085 b.hs .Loop_outer_512_neon 1086 1087 adds $len,$len,#512 1088 ushr $A0,$ONE,#2 // 4 -> 1 1089 1090 ldp d8,d9,[sp,#128+0] // meet ABI requirements 1091 ldp d10,d11,[sp,#128+16] 1092 ldp d12,d13,[sp,#128+32] 1093 ldp d14,d15,[sp,#128+48] 1094 1095 stp @K[0],$ONE,[sp,#0] // wipe off-load area 1096 stp @K[0],$ONE,[sp,#32] 1097 stp @K[0],$ONE,[sp,#64] 1098 1099 b.eq .Ldone_512_neon 1100 1101 cmp $len,#192 1102 sub @K[3],@K[3],$A0 // -= 1 1103 sub @K[4],@K[4],$A0 1104 sub @K[5],@K[5],$A0 1105 add sp,sp,#128 1106 b.hs .Loop_outer_neon 1107 1108 eor @K[1],@K[1],@K[1] 1109 eor @K[2],@K[2],@K[2] 1110 eor @K[3],@K[3],@K[3] 1111 eor @K[4],@K[4],@K[4] 1112 eor @K[5],@K[5],@K[5] 1113 eor @K[6],@K[6],@K[6] 1114 b .Loop_outer 1115 1116.Ldone_512_neon: 1117 ldp x19,x20,[x29,#16] 1118 add sp,sp,#128+64 1119 ldp x21,x22,[x29,#32] 1120 ldp x23,x24,[x29,#48] 1121 ldp x25,x26,[x29,#64] 1122 ldp x27,x28,[x29,#80] 1123 ldp x29,x30,[sp],#96 1124 AARCH64_VALIDATE_LINK_REGISTER 1125 ret 1126.size ChaCha20_512_neon,.-ChaCha20_512_neon 1127___ 1128} 1129}}} 1130 1131foreach (split("\n",$code)) { 1132 s/\`([^\`]*)\`/eval $1/geo; 1133 1134 (s/\b([a-z]+)\.32\b/$1/ and (s/x([0-9]+)/w$1/g or 1)) or 1135 (m/\b(eor|ext|mov)\b/ and (s/\.4s/\.16b/g or 1)) or 1136 (s/\b((?:ld|st)1)\.8\b/$1/ and (s/\.4s/\.16b/g or 1)) or 1137 (m/\b(ld|st)[rp]\b/ and (s/v([0-9]+)\.4s/q$1/g or 1)) or 1138 (s/\brev32\.16\b/rev32/ and (s/\.4s/\.8h/g or 1)); 1139 1140 #s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo; 1141 1142 print $_,"\n"; 1143} 1144close STDOUT or die "error closing STDOUT"; # flush 1145