arch/arm64/crypto/poly1305-armv8.pl

   1 #!/usr/bin/env perl
   2 # SPDX-License-Identifier: GPL-1.0+ OR BSD-3-Clause
   3 #
   4 # ====================================================================
   5 # Written by Andy Polyakov, @dot-asm, initially for the OpenSSL
   6 # project.
   7 # ====================================================================
   8 #
   9 # This module implements Poly1305 hash for ARMv8.
  10 #
  11 # June 2015
  12 #
  13 # Numbers are cycles per processed byte with poly1305_blocks alone.
  14 #
  15 #               IALU/gcc-4.9    NEON
  16 #
  17 # Apple A7      1.86/+5%        0.72
  18 # Cortex-A53    2.69/+58%       1.47
  19 # Cortex-A57    2.70/+7%        1.14
  20 # Denver        1.64/+50%       1.18(*)
  21 # X-Gene        2.13/+68%       2.27
  22 # Mongoose      1.77/+75%       1.12
  23 # Kryo          2.70/+55%       1.13
  24 # ThunderX2     1.17/+95%       1.36
  25 #
  26 # (*)   estimate based on resources availability is less than 1.0,
  27 #       i.e. measured result is worse than expected, presumably binary
  28 #       translator is not almighty;
  29
  30 $flavour=shift;
  31 $output=shift;
  32
  33 if ($flavour && $flavour ne "void") {
  34     $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  35     ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
  36     ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
  37     die "can't locate arm-xlate.pl";
  38
  39     open STDOUT,"| \"$^X\" $xlate $flavour $output";
  40 } else {
  41     open STDOUT,">$output";
  42 }
  43
  44 my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
  45 my ($mac,$nonce)=($inp,$len);
  46
  47 my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
  48
  49 $code.=<<___;
  50 #ifndef __KERNEL__
  51 # include "arm_arch.h"
  52 .extern OPENSSL_armcap_P
  53 #endif
  54
  55 .text
  56
  57 // forward "declarations" are required for Apple
  58 .globl  poly1305_blocks
  59 .globl  poly1305_emit
  60
  61 .globl  poly1305_init
  62 .type   poly1305_init,%function
  63 .align  5
  64 poly1305_init:
  65         cmp     $inp,xzr
  66         stp     xzr,xzr,[$ctx]          // zero hash value
  67         stp     xzr,xzr,[$ctx,#16]      // [along with is_base2_26]
  68
  69         csel    x0,xzr,x0,eq
  70         b.eq    .Lno_key
  71
  72 #ifndef __KERNEL__
  73         adrp    x17,OPENSSL_armcap_P
  74         ldr     w17,[x17,#:lo12:OPENSSL_armcap_P]
  75 #endif
  76
  77         ldp     $r0,$r1,[$inp]          // load key
  78         mov     $s1,#0xfffffffc0fffffff
  79         movk    $s1,#0x0fff,lsl#48
  80 #ifdef  __AARCH64EB__
  81         rev     $r0,$r0                 // flip bytes
  82         rev     $r1,$r1
  83 #endif
  84         and     $r0,$r0,$s1             // &=0ffffffc0fffffff
  85         and     $s1,$s1,#-4
  86         and     $r1,$r1,$s1             // &=0ffffffc0ffffffc
  87         mov     w#$s1,#-1
  88         stp     $r0,$r1,[$ctx,#32]      // save key value
  89         str     w#$s1,[$ctx,#48]        // impossible key power value
  90
  91 #ifndef __KERNEL__
  92         tst     w17,#ARMV7_NEON
  93
  94         adr     $d0,.Lpoly1305_blocks
  95         adr     $r0,.Lpoly1305_blocks_neon
  96         adr     $d1,.Lpoly1305_emit
  97
  98         csel    $d0,$d0,$r0,eq
  99
 100 # ifdef __ILP32__
 101         stp     w#$d0,w#$d1,[$len]
 102 # else
 103         stp     $d0,$d1,[$len]
 104 # endif
 105 #endif
 106         mov     x0,#1
 107 .Lno_key:
 108         ret
 109 .size   poly1305_init,.-poly1305_init
 110
 111 .type   poly1305_blocks,%function
 112 .align  5
 113 poly1305_blocks:
 114 .Lpoly1305_blocks:
 115         ands    $len,$len,#-16
 116         b.eq    .Lno_data
 117
 118         ldp     $h0,$h1,[$ctx]          // load hash value
 119         ldp     $h2,x17,[$ctx,#16]      // [along with is_base2_26]
 120         ldp     $r0,$r1,[$ctx,#32]      // load key value
 121
 122 #ifdef  __AARCH64EB__
 123         lsr     $d0,$h0,#32
 124         mov     w#$d1,w#$h0
 125         lsr     $d2,$h1,#32
 126         mov     w15,w#$h1
 127         lsr     x16,$h2,#32
 128 #else
 129         mov     w#$d0,w#$h0
 130         lsr     $d1,$h0,#32
 131         mov     w#$d2,w#$h1
 132         lsr     x15,$h1,#32
 133         mov     w16,w#$h2
 134 #endif
 135
 136         add     $d0,$d0,$d1,lsl#26      // base 2^26 -> base 2^64
 137         lsr     $d1,$d2,#12
 138         adds    $d0,$d0,$d2,lsl#52
 139         add     $d1,$d1,x15,lsl#14
 140         adc     $d1,$d1,xzr
 141         lsr     $d2,x16,#24
 142         adds    $d1,$d1,x16,lsl#40
 143         adc     $d2,$d2,xzr
 144
 145         cmp     x17,#0                  // is_base2_26?
 146         add     $s1,$r1,$r1,lsr#2       // s1 = r1 + (r1 >> 2)
 147         csel    $h0,$h0,$d0,eq          // choose between radixes
 148         csel    $h1,$h1,$d1,eq
 149         csel    $h2,$h2,$d2,eq
 150
 151 .Loop:
 152         ldp     $t0,$t1,[$inp],#16      // load input
 153         sub     $len,$len,#16
 154 #ifdef  __AARCH64EB__
 155         rev     $t0,$t0
 156         rev     $t1,$t1
 157 #endif
 158         adds    $h0,$h0,$t0             // accumulate input
 159         adcs    $h1,$h1,$t1
 160
 161         mul     $d0,$h0,$r0             // h0*r0
 162         adc     $h2,$h2,$padbit
 163         umulh   $d1,$h0,$r0
 164
 165         mul     $t0,$h1,$s1             // h1*5*r1
 166         umulh   $t1,$h1,$s1
 167
 168         adds    $d0,$d0,$t0
 169         mul     $t0,$h0,$r1             // h0*r1
 170         adc     $d1,$d1,$t1
 171         umulh   $d2,$h0,$r1
 172
 173         adds    $d1,$d1,$t0
 174         mul     $t0,$h1,$r0             // h1*r0
 175         adc     $d2,$d2,xzr
 176         umulh   $t1,$h1,$r0
 177
 178         adds    $d1,$d1,$t0
 179         mul     $t0,$h2,$s1             // h2*5*r1
 180         adc     $d2,$d2,$t1
 181         mul     $t1,$h2,$r0             // h2*r0
 182
 183         adds    $d1,$d1,$t0
 184         adc     $d2,$d2,$t1
 185
 186         and     $t0,$d2,#-4             // final reduction
 187         and     $h2,$d2,#3
 188         add     $t0,$t0,$d2,lsr#2
 189         adds    $h0,$d0,$t0
 190         adcs    $h1,$d1,xzr
 191         adc     $h2,$h2,xzr
 192
 193         cbnz    $len,.Loop
 194
 195         stp     $h0,$h1,[$ctx]          // store hash value
 196         stp     $h2,xzr,[$ctx,#16]      // [and clear is_base2_26]
 197
 198 .Lno_data:
 199         ret
 200 .size   poly1305_blocks,.-poly1305_blocks
 201
 202 .type   poly1305_emit,%function
 203 .align  5
 204 poly1305_emit:
 205 .Lpoly1305_emit:
 206         ldp     $h0,$h1,[$ctx]          // load hash base 2^64
 207         ldp     $h2,$r0,[$ctx,#16]      // [along with is_base2_26]
 208         ldp     $t0,$t1,[$nonce]        // load nonce
 209
 210 #ifdef  __AARCH64EB__
 211         lsr     $d0,$h0,#32
 212         mov     w#$d1,w#$h0
 213         lsr     $d2,$h1,#32
 214         mov     w15,w#$h1
 215         lsr     x16,$h2,#32
 216 #else
 217         mov     w#$d0,w#$h0
 218         lsr     $d1,$h0,#32
 219         mov     w#$d2,w#$h1
 220         lsr     x15,$h1,#32
 221         mov     w16,w#$h2
 222 #endif
 223
 224         add     $d0,$d0,$d1,lsl#26      // base 2^26 -> base 2^64
 225         lsr     $d1,$d2,#12
 226         adds    $d0,$d0,$d2,lsl#52
 227         add     $d1,$d1,x15,lsl#14
 228         adc     $d1,$d1,xzr
 229         lsr     $d2,x16,#24
 230         adds    $d1,$d1,x16,lsl#40
 231         adc     $d2,$d2,xzr
 232
 233         cmp     $r0,#0                  // is_base2_26?
 234         csel    $h0,$h0,$d0,eq          // choose between radixes
 235         csel    $h1,$h1,$d1,eq
 236         csel    $h2,$h2,$d2,eq
 237
 238         adds    $d0,$h0,#5              // compare to modulus
 239         adcs    $d1,$h1,xzr
 240         adc     $d2,$h2,xzr
 241
 242         tst     $d2,#-4                 // see if it's carried/borrowed
 243
 244         csel    $h0,$h0,$d0,eq
 245         csel    $h1,$h1,$d1,eq
 246
 247 #ifdef  __AARCH64EB__
 248         ror     $t0,$t0,#32             // flip nonce words
 249         ror     $t1,$t1,#32
 250 #endif
 251         adds    $h0,$h0,$t0             // accumulate nonce
 252         adc     $h1,$h1,$t1
 253 #ifdef  __AARCH64EB__
 254         rev     $h0,$h0                 // flip output bytes
 255         rev     $h1,$h1
 256 #endif
 257         stp     $h0,$h1,[$mac]          // write result
 258
 259         ret
 260 .size   poly1305_emit,.-poly1305_emit
 261 ___
 262 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
 263 my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
 264 my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
 265 my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
 266 my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
 267 my ($T0,$T1,$MASK) = map("v$_",(29..31));
 268
 269 my ($in2,$zeros)=("x16","x17");
 270 my $is_base2_26 = $zeros;               # borrow
 271
 272 $code.=<<___;
 273 .type   poly1305_mult,%function
 274 .align  5
 275 poly1305_mult:
 276         mul     $d0,$h0,$r0             // h0*r0
 277         umulh   $d1,$h0,$r0
 278
 279         mul     $t0,$h1,$s1             // h1*5*r1
 280         umulh   $t1,$h1,$s1
 281
 282         adds    $d0,$d0,$t0
 283         mul     $t0,$h0,$r1             // h0*r1
 284         adc     $d1,$d1,$t1
 285         umulh   $d2,$h0,$r1
 286
 287         adds    $d1,$d1,$t0
 288         mul     $t0,$h1,$r0             // h1*r0
 289         adc     $d2,$d2,xzr
 290         umulh   $t1,$h1,$r0
 291
 292         adds    $d1,$d1,$t0
 293         mul     $t0,$h2,$s1             // h2*5*r1
 294         adc     $d2,$d2,$t1
 295         mul     $t1,$h2,$r0             // h2*r0
 296
 297         adds    $d1,$d1,$t0
 298         adc     $d2,$d2,$t1
 299
 300         and     $t0,$d2,#-4             // final reduction
 301         and     $h2,$d2,#3
 302         add     $t0,$t0,$d2,lsr#2
 303         adds    $h0,$d0,$t0
 304         adcs    $h1,$d1,xzr
 305         adc     $h2,$h2,xzr
 306
 307         ret
 308 .size   poly1305_mult,.-poly1305_mult
 309
 310 .type   poly1305_splat,%function
 311 .align  4
 312 poly1305_splat:
 313         and     x12,$h0,#0x03ffffff     // base 2^64 -> base 2^26
 314         ubfx    x13,$h0,#26,#26
 315         extr    x14,$h1,$h0,#52
 316         and     x14,x14,#0x03ffffff
 317         ubfx    x15,$h1,#14,#26
 318         extr    x16,$h2,$h1,#40
 319
 320         str     w12,[$ctx,#16*0]        // r0
 321         add     w12,w13,w13,lsl#2       // r1*5
 322         str     w13,[$ctx,#16*1]        // r1
 323         add     w13,w14,w14,lsl#2       // r2*5
 324         str     w12,[$ctx,#16*2]        // s1
 325         str     w14,[$ctx,#16*3]        // r2
 326         add     w14,w15,w15,lsl#2       // r3*5
 327         str     w13,[$ctx,#16*4]        // s2
 328         str     w15,[$ctx,#16*5]        // r3
 329         add     w15,w16,w16,lsl#2       // r4*5
 330         str     w14,[$ctx,#16*6]        // s3
 331         str     w16,[$ctx,#16*7]        // r4
 332         str     w15,[$ctx,#16*8]        // s4
 333
 334         ret
 335 .size   poly1305_splat,.-poly1305_splat
 336
 337 #ifdef  __KERNEL__
 338 .globl  poly1305_blocks_neon
 339 #endif
 340 .type   poly1305_blocks_neon,%function
 341 .align  5
 342 poly1305_blocks_neon:
 343 .Lpoly1305_blocks_neon:
 344         ldr     $is_base2_26,[$ctx,#24]
 345         cmp     $len,#128
 346         b.lo    .Lpoly1305_blocks
 347
 348         .inst   0xd503233f              // paciasp
 349         stp     x29,x30,[sp,#-80]!
 350         add     x29,sp,#0
 351
 352         stp     d8,d9,[sp,#16]          // meet ABI requirements
 353         stp     d10,d11,[sp,#32]
 354         stp     d12,d13,[sp,#48]
 355         stp     d14,d15,[sp,#64]
 356
 357         cbz     $is_base2_26,.Lbase2_64_neon
 358
 359         ldp     w10,w11,[$ctx]          // load hash value base 2^26
 360         ldp     w12,w13,[$ctx,#8]
 361         ldr     w14,[$ctx,#16]
 362
 363         tst     $len,#31
 364         b.eq    .Leven_neon
 365
 366         ldp     $r0,$r1,[$ctx,#32]      // load key value
 367
 368         add     $h0,x10,x11,lsl#26      // base 2^26 -> base 2^64
 369         lsr     $h1,x12,#12
 370         adds    $h0,$h0,x12,lsl#52
 371         add     $h1,$h1,x13,lsl#14
 372         adc     $h1,$h1,xzr
 373         lsr     $h2,x14,#24
 374         adds    $h1,$h1,x14,lsl#40
 375         adc     $d2,$h2,xzr             // can be partially reduced...
 376
 377         ldp     $d0,$d1,[$inp],#16      // load input
 378         sub     $len,$len,#16
 379         add     $s1,$r1,$r1,lsr#2       // s1 = r1 + (r1 >> 2)
 380
 381 #ifdef  __AARCH64EB__
 382         rev     $d0,$d0
 383         rev     $d1,$d1
 384 #endif
 385         adds    $h0,$h0,$d0             // accumulate input
 386         adcs    $h1,$h1,$d1
 387         adc     $h2,$h2,$padbit
 388
 389         bl      poly1305_mult
 390
 391         and     x10,$h0,#0x03ffffff     // base 2^64 -> base 2^26
 392         ubfx    x11,$h0,#26,#26
 393         extr    x12,$h1,$h0,#52
 394         and     x12,x12,#0x03ffffff
 395         ubfx    x13,$h1,#14,#26
 396         extr    x14,$h2,$h1,#40
 397
 398         b       .Leven_neon
 399
 400 .align  4
 401 .Lbase2_64_neon:
 402         ldp     $r0,$r1,[$ctx,#32]      // load key value
 403
 404         ldp     $h0,$h1,[$ctx]          // load hash value base 2^64
 405         ldr     $h2,[$ctx,#16]
 406
 407         tst     $len,#31
 408         b.eq    .Linit_neon
 409
 410         ldp     $d0,$d1,[$inp],#16      // load input
 411         sub     $len,$len,#16
 412         add     $s1,$r1,$r1,lsr#2       // s1 = r1 + (r1 >> 2)
 413 #ifdef  __AARCH64EB__
 414         rev     $d0,$d0
 415         rev     $d1,$d1
 416 #endif
 417         adds    $h0,$h0,$d0             // accumulate input
 418         adcs    $h1,$h1,$d1
 419         adc     $h2,$h2,$padbit
 420
 421         bl      poly1305_mult
 422
 423 .Linit_neon:
 424         ldr     w17,[$ctx,#48]          // first table element
 425         and     x10,$h0,#0x03ffffff     // base 2^64 -> base 2^26
 426         ubfx    x11,$h0,#26,#26
 427         extr    x12,$h1,$h0,#52
 428         and     x12,x12,#0x03ffffff
 429         ubfx    x13,$h1,#14,#26
 430         extr    x14,$h2,$h1,#40
 431
 432         cmp     w17,#-1                 // is value impossible?
 433         b.ne    .Leven_neon
 434
 435         fmov    ${H0},x10
 436         fmov    ${H1},x11
 437         fmov    ${H2},x12
 438         fmov    ${H3},x13
 439         fmov    ${H4},x14
 440
 441         ////////////////////////////////// initialize r^n table
 442         mov     $h0,$r0                 // r^1
 443         add     $s1,$r1,$r1,lsr#2       // s1 = r1 + (r1 >> 2)
 444         mov     $h1,$r1
 445         mov     $h2,xzr
 446         add     $ctx,$ctx,#48+12
 447         bl      poly1305_splat
 448
 449         bl      poly1305_mult           // r^2
 450         sub     $ctx,$ctx,#4
 451         bl      poly1305_splat
 452
 453         bl      poly1305_mult           // r^3
 454         sub     $ctx,$ctx,#4
 455         bl      poly1305_splat
 456
 457         bl      poly1305_mult           // r^4
 458         sub     $ctx,$ctx,#4
 459         bl      poly1305_splat
 460         sub     $ctx,$ctx,#48           // restore original $ctx
 461         b       .Ldo_neon
 462
 463 .align  4
 464 .Leven_neon:
 465         fmov    ${H0},x10
 466         fmov    ${H1},x11
 467         fmov    ${H2},x12
 468         fmov    ${H3},x13
 469         fmov    ${H4},x14
 470
 471 .Ldo_neon:
 472         ldp     x8,x12,[$inp,#32]       // inp[2:3]
 473         subs    $len,$len,#64
 474         ldp     x9,x13,[$inp,#48]
 475         add     $in2,$inp,#96
 476         adr     $zeros,.Lzeros
 477
 478         lsl     $padbit,$padbit,#24
 479         add     x15,$ctx,#48
 480
 481 #ifdef  __AARCH64EB__
 482         rev     x8,x8
 483         rev     x12,x12
 484         rev     x9,x9
 485         rev     x13,x13
 486 #endif
 487         and     x4,x8,#0x03ffffff       // base 2^64 -> base 2^26
 488         and     x5,x9,#0x03ffffff
 489         ubfx    x6,x8,#26,#26
 490         ubfx    x7,x9,#26,#26
 491         add     x4,x4,x5,lsl#32         // bfi  x4,x5,#32,#32
 492         extr    x8,x12,x8,#52
 493         extr    x9,x13,x9,#52
 494         add     x6,x6,x7,lsl#32         // bfi  x6,x7,#32,#32
 495         fmov    $IN23_0,x4
 496         and     x8,x8,#0x03ffffff
 497         and     x9,x9,#0x03ffffff
 498         ubfx    x10,x12,#14,#26
 499         ubfx    x11,x13,#14,#26
 500         add     x12,$padbit,x12,lsr#40
 501         add     x13,$padbit,x13,lsr#40
 502         add     x8,x8,x9,lsl#32         // bfi  x8,x9,#32,#32
 503         fmov    $IN23_1,x6
 504         add     x10,x10,x11,lsl#32      // bfi  x10,x11,#32,#32
 505         add     x12,x12,x13,lsl#32      // bfi  x12,x13,#32,#32
 506         fmov    $IN23_2,x8
 507         fmov    $IN23_3,x10
 508         fmov    $IN23_4,x12
 509
 510         ldp     x8,x12,[$inp],#16       // inp[0:1]
 511         ldp     x9,x13,[$inp],#48
 512
 513         ld1     {$R0,$R1,$S1,$R2},[x15],#64
 514         ld1     {$S2,$R3,$S3,$R4},[x15],#64
 515         ld1     {$S4},[x15]
 516
 517 #ifdef  __AARCH64EB__
 518         rev     x8,x8
 519         rev     x12,x12
 520         rev     x9,x9
 521         rev     x13,x13
 522 #endif
 523         and     x4,x8,#0x03ffffff       // base 2^64 -> base 2^26
 524         and     x5,x9,#0x03ffffff
 525         ubfx    x6,x8,#26,#26
 526         ubfx    x7,x9,#26,#26
 527         add     x4,x4,x5,lsl#32         // bfi  x4,x5,#32,#32
 528         extr    x8,x12,x8,#52
 529         extr    x9,x13,x9,#52
 530         add     x6,x6,x7,lsl#32         // bfi  x6,x7,#32,#32
 531         fmov    $IN01_0,x4
 532         and     x8,x8,#0x03ffffff
 533         and     x9,x9,#0x03ffffff
 534         ubfx    x10,x12,#14,#26
 535         ubfx    x11,x13,#14,#26
 536         add     x12,$padbit,x12,lsr#40
 537         add     x13,$padbit,x13,lsr#40
 538         add     x8,x8,x9,lsl#32         // bfi  x8,x9,#32,#32
 539         fmov    $IN01_1,x6
 540         add     x10,x10,x11,lsl#32      // bfi  x10,x11,#32,#32
 541         add     x12,x12,x13,lsl#32      // bfi  x12,x13,#32,#32
 542         movi    $MASK.2d,#-1
 543         fmov    $IN01_2,x8
 544         fmov    $IN01_3,x10
 545         fmov    $IN01_4,x12
 546         ushr    $MASK.2d,$MASK.2d,#38
 547
 548         b.ls    .Lskip_loop
 549
 550 .align  4
 551 .Loop_neon:
 552         ////////////////////////////////////////////////////////////////
 553         // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
 554         // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
 555         //   \___________________/
 556         // ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
 557         // ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
 558         //   \___________________/ \____________________/
 559         //
 560         // Note that we start with inp[2:3]*r^2. This is because it
 561         // doesn't depend on reduction in previous iteration.
 562         ////////////////////////////////////////////////////////////////
 563         // d4 = h0*r4 + h1*r3   + h2*r2   + h3*r1   + h4*r0
 564         // d3 = h0*r3 + h1*r2   + h2*r1   + h3*r0   + h4*5*r4
 565         // d2 = h0*r2 + h1*r1   + h2*r0   + h3*5*r4 + h4*5*r3
 566         // d1 = h0*r1 + h1*r0   + h2*5*r4 + h3*5*r3 + h4*5*r2
 567         // d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
 568
 569         subs    $len,$len,#64
 570         umull   $ACC4,$IN23_0,${R4}[2]
 571         csel    $in2,$zeros,$in2,lo
 572         umull   $ACC3,$IN23_0,${R3}[2]
 573         umull   $ACC2,$IN23_0,${R2}[2]
 574          ldp    x8,x12,[$in2],#16       // inp[2:3] (or zero)
 575         umull   $ACC1,$IN23_0,${R1}[2]
 576          ldp    x9,x13,[$in2],#48
 577         umull   $ACC0,$IN23_0,${R0}[2]
 578 #ifdef  __AARCH64EB__
 579          rev    x8,x8
 580          rev    x12,x12
 581          rev    x9,x9
 582          rev    x13,x13
 583 #endif
 584
 585         umlal   $ACC4,$IN23_1,${R3}[2]
 586          and    x4,x8,#0x03ffffff       // base 2^64 -> base 2^26
 587         umlal   $ACC3,$IN23_1,${R2}[2]
 588          and    x5,x9,#0x03ffffff
 589         umlal   $ACC2,$IN23_1,${R1}[2]
 590          ubfx   x6,x8,#26,#26
 591         umlal   $ACC1,$IN23_1,${R0}[2]
 592          ubfx   x7,x9,#26,#26
 593         umlal   $ACC0,$IN23_1,${S4}[2]
 594          add    x4,x4,x5,lsl#32         // bfi  x4,x5,#32,#32
 595
 596         umlal   $ACC4,$IN23_2,${R2}[2]
 597          extr   x8,x12,x8,#52
 598         umlal   $ACC3,$IN23_2,${R1}[2]
 599          extr   x9,x13,x9,#52
 600         umlal   $ACC2,$IN23_2,${R0}[2]
 601          add    x6,x6,x7,lsl#32         // bfi  x6,x7,#32,#32
 602         umlal   $ACC1,$IN23_2,${S4}[2]
 603          fmov   $IN23_0,x4
 604         umlal   $ACC0,$IN23_2,${S3}[2]
 605          and    x8,x8,#0x03ffffff
 606
 607         umlal   $ACC4,$IN23_3,${R1}[2]
 608          and    x9,x9,#0x03ffffff
 609         umlal   $ACC3,$IN23_3,${R0}[2]
 610          ubfx   x10,x12,#14,#26
 611         umlal   $ACC2,$IN23_3,${S4}[2]
 612          ubfx   x11,x13,#14,#26
 613         umlal   $ACC1,$IN23_3,${S3}[2]
 614          add    x8,x8,x9,lsl#32         // bfi  x8,x9,#32,#32
 615         umlal   $ACC0,$IN23_3,${S2}[2]
 616          fmov   $IN23_1,x6
 617
 618         add     $IN01_2,$IN01_2,$H2
 619          add    x12,$padbit,x12,lsr#40
 620         umlal   $ACC4,$IN23_4,${R0}[2]
 621          add    x13,$padbit,x13,lsr#40
 622         umlal   $ACC3,$IN23_4,${S4}[2]
 623          add    x10,x10,x11,lsl#32      // bfi  x10,x11,#32,#32
 624         umlal   $ACC2,$IN23_4,${S3}[2]
 625          add    x12,x12,x13,lsl#32      // bfi  x12,x13,#32,#32
 626         umlal   $ACC1,$IN23_4,${S2}[2]
 627          fmov   $IN23_2,x8
 628         umlal   $ACC0,$IN23_4,${S1}[2]
 629          fmov   $IN23_3,x10
 630
 631         ////////////////////////////////////////////////////////////////
 632         // (hash+inp[0:1])*r^4 and accumulate
 633
 634         add     $IN01_0,$IN01_0,$H0
 635          fmov   $IN23_4,x12
 636         umlal   $ACC3,$IN01_2,${R1}[0]
 637          ldp    x8,x12,[$inp],#16       // inp[0:1]
 638         umlal   $ACC0,$IN01_2,${S3}[0]
 639          ldp    x9,x13,[$inp],#48
 640         umlal   $ACC4,$IN01_2,${R2}[0]
 641         umlal   $ACC1,$IN01_2,${S4}[0]
 642         umlal   $ACC2,$IN01_2,${R0}[0]
 643 #ifdef  __AARCH64EB__
 644          rev    x8,x8
 645          rev    x12,x12
 646          rev    x9,x9
 647          rev    x13,x13
 648 #endif
 649
 650         add     $IN01_1,$IN01_1,$H1
 651         umlal   $ACC3,$IN01_0,${R3}[0]
 652         umlal   $ACC4,$IN01_0,${R4}[0]
 653          and    x4,x8,#0x03ffffff       // base 2^64 -> base 2^26
 654         umlal   $ACC2,$IN01_0,${R2}[0]
 655          and    x5,x9,#0x03ffffff
 656         umlal   $ACC0,$IN01_0,${R0}[0]
 657          ubfx   x6,x8,#26,#26
 658         umlal   $ACC1,$IN01_0,${R1}[0]
 659          ubfx   x7,x9,#26,#26
 660
 661         add     $IN01_3,$IN01_3,$H3
 662          add    x4,x4,x5,lsl#32         // bfi  x4,x5,#32,#32
 663         umlal   $ACC3,$IN01_1,${R2}[0]
 664          extr   x8,x12,x8,#52
 665         umlal   $ACC4,$IN01_1,${R3}[0]
 666          extr   x9,x13,x9,#52
 667         umlal   $ACC0,$IN01_1,${S4}[0]
 668          add    x6,x6,x7,lsl#32         // bfi  x6,x7,#32,#32
 669         umlal   $ACC2,$IN01_1,${R1}[0]
 670          fmov   $IN01_0,x4
 671         umlal   $ACC1,$IN01_1,${R0}[0]
 672          and    x8,x8,#0x03ffffff
 673
 674         add     $IN01_4,$IN01_4,$H4
 675          and    x9,x9,#0x03ffffff
 676         umlal   $ACC3,$IN01_3,${R0}[0]
 677          ubfx   x10,x12,#14,#26
 678         umlal   $ACC0,$IN01_3,${S2}[0]
 679          ubfx   x11,x13,#14,#26
 680         umlal   $ACC4,$IN01_3,${R1}[0]
 681          add    x8,x8,x9,lsl#32         // bfi  x8,x9,#32,#32
 682         umlal   $ACC1,$IN01_3,${S3}[0]
 683          fmov   $IN01_1,x6
 684         umlal   $ACC2,$IN01_3,${S4}[0]
 685          add    x12,$padbit,x12,lsr#40
 686
 687         umlal   $ACC3,$IN01_4,${S4}[0]
 688          add    x13,$padbit,x13,lsr#40
 689         umlal   $ACC0,$IN01_4,${S1}[0]
 690          add    x10,x10,x11,lsl#32      // bfi  x10,x11,#32,#32
 691         umlal   $ACC4,$IN01_4,${R0}[0]
 692          add    x12,x12,x13,lsl#32      // bfi  x12,x13,#32,#32
 693         umlal   $ACC1,$IN01_4,${S2}[0]
 694          fmov   $IN01_2,x8
 695         umlal   $ACC2,$IN01_4,${S3}[0]
 696          fmov   $IN01_3,x10
 697          fmov   $IN01_4,x12
 698
 699         /////////////////////////////////////////////////////////////////
 700         // lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
 701         // and P. Schwabe
 702         //
 703         // [see discussion in poly1305-armv4 module]
 704
 705         ushr    $T0.2d,$ACC3,#26
 706         xtn     $H3,$ACC3
 707          ushr   $T1.2d,$ACC0,#26
 708          and    $ACC0,$ACC0,$MASK.2d
 709         add     $ACC4,$ACC4,$T0.2d      // h3 -> h4
 710         bic     $H3,#0xfc,lsl#24        // &=0x03ffffff
 711          add    $ACC1,$ACC1,$T1.2d      // h0 -> h1
 712
 713         ushr    $T0.2d,$ACC4,#26
 714         xtn     $H4,$ACC4
 715          ushr   $T1.2d,$ACC1,#26
 716          xtn    $H1,$ACC1
 717         bic     $H4,#0xfc,lsl#24
 718          add    $ACC2,$ACC2,$T1.2d      // h1 -> h2
 719
 720         add     $ACC0,$ACC0,$T0.2d
 721         shl     $T0.2d,$T0.2d,#2
 722          shrn   $T1.2s,$ACC2,#26
 723          xtn    $H2,$ACC2
 724         add     $ACC0,$ACC0,$T0.2d      // h4 -> h0
 725          bic    $H1,#0xfc,lsl#24
 726          add    $H3,$H3,$T1.2s          // h2 -> h3
 727          bic    $H2,#0xfc,lsl#24
 728
 729         shrn    $T0.2s,$ACC0,#26
 730         xtn     $H0,$ACC0
 731          ushr   $T1.2s,$H3,#26
 732          bic    $H3,#0xfc,lsl#24
 733          bic    $H0,#0xfc,lsl#24
 734         add     $H1,$H1,$T0.2s          // h0 -> h1
 735          add    $H4,$H4,$T1.2s          // h3 -> h4
 736
 737         b.hi    .Loop_neon
 738
 739 .Lskip_loop:
 740         dup     $IN23_2,${IN23_2}[0]
 741         add     $IN01_2,$IN01_2,$H2
 742
 743         ////////////////////////////////////////////////////////////////
 744         // multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
 745
 746         adds    $len,$len,#32
 747         b.ne    .Long_tail
 748
 749         dup     $IN23_2,${IN01_2}[0]
 750         add     $IN23_0,$IN01_0,$H0
 751         add     $IN23_3,$IN01_3,$H3
 752         add     $IN23_1,$IN01_1,$H1
 753         add     $IN23_4,$IN01_4,$H4
 754
 755 .Long_tail:
 756         dup     $IN23_0,${IN23_0}[0]
 757         umull2  $ACC0,$IN23_2,${S3}
 758         umull2  $ACC3,$IN23_2,${R1}
 759         umull2  $ACC4,$IN23_2,${R2}
 760         umull2  $ACC2,$IN23_2,${R0}
 761         umull2  $ACC1,$IN23_2,${S4}
 762
 763         dup     $IN23_1,${IN23_1}[0]
 764         umlal2  $ACC0,$IN23_0,${R0}
 765         umlal2  $ACC2,$IN23_0,${R2}
 766         umlal2  $ACC3,$IN23_0,${R3}
 767         umlal2  $ACC4,$IN23_0,${R4}
 768         umlal2  $ACC1,$IN23_0,${R1}
 769
 770         dup     $IN23_3,${IN23_3}[0]
 771         umlal2  $ACC0,$IN23_1,${S4}
 772         umlal2  $ACC3,$IN23_1,${R2}
 773         umlal2  $ACC2,$IN23_1,${R1}
 774         umlal2  $ACC4,$IN23_1,${R3}
 775         umlal2  $ACC1,$IN23_1,${R0}
 776
 777         dup     $IN23_4,${IN23_4}[0]
 778         umlal2  $ACC3,$IN23_3,${R0}
 779         umlal2  $ACC4,$IN23_3,${R1}
 780         umlal2  $ACC0,$IN23_3,${S2}
 781         umlal2  $ACC1,$IN23_3,${S3}
 782         umlal2  $ACC2,$IN23_3,${S4}
 783
 784         umlal2  $ACC3,$IN23_4,${S4}
 785         umlal2  $ACC0,$IN23_4,${S1}
 786         umlal2  $ACC4,$IN23_4,${R0}
 787         umlal2  $ACC1,$IN23_4,${S2}
 788         umlal2  $ACC2,$IN23_4,${S3}
 789
 790         b.eq    .Lshort_tail
 791
 792         ////////////////////////////////////////////////////////////////
 793         // (hash+inp[0:1])*r^4:r^3 and accumulate
 794
 795         add     $IN01_0,$IN01_0,$H0
 796         umlal   $ACC3,$IN01_2,${R1}
 797         umlal   $ACC0,$IN01_2,${S3}
 798         umlal   $ACC4,$IN01_2,${R2}
 799         umlal   $ACC1,$IN01_2,${S4}
 800         umlal   $ACC2,$IN01_2,${R0}
 801
 802         add     $IN01_1,$IN01_1,$H1
 803         umlal   $ACC3,$IN01_0,${R3}
 804         umlal   $ACC0,$IN01_0,${R0}
 805         umlal   $ACC4,$IN01_0,${R4}
 806         umlal   $ACC1,$IN01_0,${R1}
 807         umlal   $ACC2,$IN01_0,${R2}
 808
 809         add     $IN01_3,$IN01_3,$H3
 810         umlal   $ACC3,$IN01_1,${R2}
 811         umlal   $ACC0,$IN01_1,${S4}
 812         umlal   $ACC4,$IN01_1,${R3}
 813         umlal   $ACC1,$IN01_1,${R0}
 814         umlal   $ACC2,$IN01_1,${R1}
 815
 816         add     $IN01_4,$IN01_4,$H4
 817         umlal   $ACC3,$IN01_3,${R0}
 818         umlal   $ACC0,$IN01_3,${S2}
 819         umlal   $ACC4,$IN01_3,${R1}
 820         umlal   $ACC1,$IN01_3,${S3}
 821         umlal   $ACC2,$IN01_3,${S4}
 822
 823         umlal   $ACC3,$IN01_4,${S4}
 824         umlal   $ACC0,$IN01_4,${S1}
 825         umlal   $ACC4,$IN01_4,${R0}
 826         umlal   $ACC1,$IN01_4,${S2}
 827         umlal   $ACC2,$IN01_4,${S3}
 828
 829 .Lshort_tail:
 830         ////////////////////////////////////////////////////////////////
 831         // horizontal add
 832
 833         addp    $ACC3,$ACC3,$ACC3
 834          ldp    d8,d9,[sp,#16]          // meet ABI requirements
 835         addp    $ACC0,$ACC0,$ACC0
 836          ldp    d10,d11,[sp,#32]
 837         addp    $ACC4,$ACC4,$ACC4
 838          ldp    d12,d13,[sp,#48]
 839         addp    $ACC1,$ACC1,$ACC1
 840          ldp    d14,d15,[sp,#64]
 841         addp    $ACC2,$ACC2,$ACC2
 842          ldr    x30,[sp,#8]
 843
 844         ////////////////////////////////////////////////////////////////
 845         // lazy reduction, but without narrowing
 846
 847         ushr    $T0.2d,$ACC3,#26
 848         and     $ACC3,$ACC3,$MASK.2d
 849          ushr   $T1.2d,$ACC0,#26
 850          and    $ACC0,$ACC0,$MASK.2d
 851
 852         add     $ACC4,$ACC4,$T0.2d      // h3 -> h4
 853          add    $ACC1,$ACC1,$T1.2d      // h0 -> h1
 854
 855         ushr    $T0.2d,$ACC4,#26
 856         and     $ACC4,$ACC4,$MASK.2d
 857          ushr   $T1.2d,$ACC1,#26
 858          and    $ACC1,$ACC1,$MASK.2d
 859          add    $ACC2,$ACC2,$T1.2d      // h1 -> h2
 860
 861         add     $ACC0,$ACC0,$T0.2d
 862         shl     $T0.2d,$T0.2d,#2
 863          ushr   $T1.2d,$ACC2,#26
 864          and    $ACC2,$ACC2,$MASK.2d
 865         add     $ACC0,$ACC0,$T0.2d      // h4 -> h0
 866          add    $ACC3,$ACC3,$T1.2d      // h2 -> h3
 867
 868         ushr    $T0.2d,$ACC0,#26
 869         and     $ACC0,$ACC0,$MASK.2d
 870          ushr   $T1.2d,$ACC3,#26
 871          and    $ACC3,$ACC3,$MASK.2d
 872         add     $ACC1,$ACC1,$T0.2d      // h0 -> h1
 873          add    $ACC4,$ACC4,$T1.2d      // h3 -> h4
 874
 875         ////////////////////////////////////////////////////////////////
 876         // write the result, can be partially reduced
 877
 878         st4     {$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
 879         mov     x4,#1
 880         st1     {$ACC4}[0],[$ctx]
 881         str     x4,[$ctx,#8]            // set is_base2_26
 882
 883         ldr     x29,[sp],#80
 884          .inst  0xd50323bf              // autiasp
 885         ret
 886 .size   poly1305_blocks_neon,.-poly1305_blocks_neon
 887
 888 .align  5
 889 .Lzeros:
 890 .long   0,0,0,0,0,0,0,0
 891 .asciz  "Poly1305 for ARMv8, CRYPTOGAMS by \@dot-asm"
 892 .align  2
 893 #if !defined(__KERNEL__) && !defined(_WIN64)
 894 .comm   OPENSSL_armcap_P,4,4
 895 .hidden OPENSSL_armcap_P
 896 #endif
 897 ___
 898
 899 foreach (split("\n",$code)) {
 900         s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/                      or
 901         s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/     or
 902         (m/\bdup\b/ and (s/\.[24]s/.2d/g or 1))                 or
 903         (m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1))       or
 904         (m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1))             or
 905         (m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1))            or
 906         (m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
 907
 908         s/\.[124]([sd])\[/.$1\[/;
 909         s/w#x([0-9]+)/w$1/g;
 910
 911         print $_,"\n";
 912 }
 913 close STDOUT;