1# Copyright 2010-2016 The OpenSSL Project Authors. All Rights Reserved. 2# 3# Licensed under the Apache License 2.0 (the "License"). You may not use 4# this file except in compliance with the License. You can obtain a copy 5# in the file LICENSE in the source distribution or at 6# https://www.openssl.org/source/license.html 7 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# 16# March, June 2010 17# 18# The module implements "4-bit" GCM GHASH function and underlying 19# single multiplication operation in GF(2^128). "4-bit" means that 20# it uses 256 bytes per-key table [+128 bytes shared table]. GHASH 21# function features so called "528B" variant utilizing additional 22# 256+16 bytes of per-key storage [+512 bytes shared table]. 23# Performance results are for this streamed GHASH subroutine and are 24# expressed in cycles per processed byte, less is better: 25# 26# gcc 3.4.x(*) assembler 27# 28# P4 28.6 14.0 +100% 29# Opteron 19.3 7.7 +150% 30# Core2 17.8 8.1(**) +120% 31# Atom 31.6 16.8 +88% 32# VIA Nano 21.8 10.1 +115% 33# 34# (*) comparison is not completely fair, because C results are 35# for vanilla "256B" implementation, while assembler results 36# are for "528B";-) 37# (**) it's mystery [to me] why Core2 result is not same as for 38# Opteron; 39 40# May 2010 41# 42# Add PCLMULQDQ version performing at 2.02 cycles per processed byte. 43# See ghash-x86.pl for background information and details about coding 44# techniques. 45# 46# Special thanks to David Woodhouse for providing access to a 47# Westmere-based system on behalf of Intel Open Source Technology Centre. 48 49# December 2012 50# 51# Overhaul: aggregate Karatsuba post-processing, improve ILP in 52# reduction_alg9, increase reduction aggregate factor to 4x. As for 53# the latter. ghash-x86.pl discusses that it makes lesser sense to 54# increase aggregate factor. Then why increase here? Critical path 55# consists of 3 independent pclmulqdq instructions, Karatsuba post- 56# processing and reduction. "On top" of this we lay down aggregated 57# multiplication operations, triplets of independent pclmulqdq's. As 58# issue rate for pclmulqdq is limited, it makes lesser sense to 59# aggregate more multiplications than it takes to perform remaining 60# non-multiplication operations. 2x is near-optimal coefficient for 61# contemporary Intel CPUs (therefore modest improvement coefficient), 62# but not for Bulldozer. Latter is because logical SIMD operations 63# are twice as slow in comparison to Intel, so that critical path is 64# longer. A CPU with higher pclmulqdq issue rate would also benefit 65# from higher aggregate factor... 66# 67# Westmere 1.78(+13%) 68# Sandy Bridge 1.80(+8%) 69# Ivy Bridge 1.80(+7%) 70# Haswell 0.55(+93%) (if system doesn't support AVX) 71# Broadwell 0.45(+110%)(if system doesn't support AVX) 72# Skylake 0.44(+110%)(if system doesn't support AVX) 73# Bulldozer 1.49(+27%) 74# Silvermont 2.88(+13%) 75# Knights L 2.12(-) (if system doesn't support AVX) 76# Goldmont 1.08(+24%) 77 78# March 2013 79# 80# ... 8x aggregate factor AVX code path is using reduction algorithm 81# suggested by Shay Gueron[1]. Even though contemporary AVX-capable 82# CPUs such as Sandy and Ivy Bridge can execute it, the code performs 83# sub-optimally in comparison to above mentioned version. But thanks 84# to Ilya Albrekht and Max Locktyukhin of Intel Corp. we knew that 85# it performs in 0.41 cycles per byte on Haswell processor, in 86# 0.29 on Broadwell, and in 0.36 on Skylake. 87# 88# Knights Landing achieves 1.09 cpb. 89# 90# [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest 91 92# Generated once from 93# https://github.com/openssl/openssl/blob/5ffc3324/crypto/modes/asm/ghash-x86_64.pl 94# and modified for ICP. Modification are kept at a bare minimum to ease later 95# upstream merges. 96 97#if defined(__x86_64__) && defined(HAVE_AVX) && \ 98 defined(HAVE_AES) && defined(HAVE_PCLMULQDQ) 99 100#define _ASM 101#include <sys/asm_linkage.h> 102 103.text 104 105/* Windows userland links with OpenSSL */ 106#if !defined (_WIN32) || defined (_KERNEL) 107ENTRY_ALIGN(gcm_gmult_clmul, 16) 108 109.cfi_startproc 110 ENDBR 111 112.L_gmult_clmul: 113 movdqu (%rdi),%xmm0 114 movdqa .Lbswap_mask(%rip),%xmm5 115 movdqu (%rsi),%xmm2 116 movdqu 32(%rsi),%xmm4 117.byte 102,15,56,0,197 118 movdqa %xmm0,%xmm1 119 pshufd $78,%xmm0,%xmm3 120 pxor %xmm0,%xmm3 121.byte 102,15,58,68,194,0 122.byte 102,15,58,68,202,17 123.byte 102,15,58,68,220,0 124 pxor %xmm0,%xmm3 125 pxor %xmm1,%xmm3 126 127 movdqa %xmm3,%xmm4 128 psrldq $8,%xmm3 129 pslldq $8,%xmm4 130 pxor %xmm3,%xmm1 131 pxor %xmm4,%xmm0 132 133 movdqa %xmm0,%xmm4 134 movdqa %xmm0,%xmm3 135 psllq $5,%xmm0 136 pxor %xmm0,%xmm3 137 psllq $1,%xmm0 138 pxor %xmm3,%xmm0 139 psllq $57,%xmm0 140 movdqa %xmm0,%xmm3 141 pslldq $8,%xmm0 142 psrldq $8,%xmm3 143 pxor %xmm4,%xmm0 144 pxor %xmm3,%xmm1 145 146 147 movdqa %xmm0,%xmm4 148 psrlq $1,%xmm0 149 pxor %xmm4,%xmm1 150 pxor %xmm0,%xmm4 151 psrlq $5,%xmm0 152 pxor %xmm4,%xmm0 153 psrlq $1,%xmm0 154 pxor %xmm1,%xmm0 155.byte 102,15,56,0,197 156 movdqu %xmm0,(%rdi) 157 RET 158.cfi_endproc 159SET_SIZE(gcm_gmult_clmul) 160#endif /* !_WIN32 || _KERNEL */ 161 162ENTRY_ALIGN(gcm_init_htab_avx, 32) 163.cfi_startproc 164 ENDBR 165 vzeroupper 166 167 vmovdqu (%rsi),%xmm2 168 // KCF/ICP stores H in network byte order with the hi qword first 169 // so we need to swap all bytes, not the 2 qwords. 170 vmovdqu .Lbswap_mask(%rip),%xmm4 171 vpshufb %xmm4,%xmm2,%xmm2 172 173 174 vpshufd $255,%xmm2,%xmm4 175 vpsrlq $63,%xmm2,%xmm3 176 vpsllq $1,%xmm2,%xmm2 177 vpxor %xmm5,%xmm5,%xmm5 178 vpcmpgtd %xmm4,%xmm5,%xmm5 179 vpslldq $8,%xmm3,%xmm3 180 vpor %xmm3,%xmm2,%xmm2 181 182 183 vpand .L0x1c2_polynomial(%rip),%xmm5,%xmm5 184 vpxor %xmm5,%xmm2,%xmm2 185 186 vpunpckhqdq %xmm2,%xmm2,%xmm6 187 vmovdqa %xmm2,%xmm0 188 vpxor %xmm2,%xmm6,%xmm6 189 movq $4,%r10 190 jmp .Linit_start_avx 191.balign 32 192.Linit_loop_avx: 193 vpalignr $8,%xmm3,%xmm4,%xmm5 194 vmovdqu %xmm5,-16(%rdi) 195 vpunpckhqdq %xmm0,%xmm0,%xmm3 196 vpxor %xmm0,%xmm3,%xmm3 197 vpclmulqdq $0x11,%xmm2,%xmm0,%xmm1 198 vpclmulqdq $0x00,%xmm2,%xmm0,%xmm0 199 vpclmulqdq $0x00,%xmm6,%xmm3,%xmm3 200 vpxor %xmm0,%xmm1,%xmm4 201 vpxor %xmm4,%xmm3,%xmm3 202 203 vpslldq $8,%xmm3,%xmm4 204 vpsrldq $8,%xmm3,%xmm3 205 vpxor %xmm4,%xmm0,%xmm0 206 vpxor %xmm3,%xmm1,%xmm1 207 vpsllq $57,%xmm0,%xmm3 208 vpsllq $62,%xmm0,%xmm4 209 vpxor %xmm3,%xmm4,%xmm4 210 vpsllq $63,%xmm0,%xmm3 211 vpxor %xmm3,%xmm4,%xmm4 212 vpslldq $8,%xmm4,%xmm3 213 vpsrldq $8,%xmm4,%xmm4 214 vpxor %xmm3,%xmm0,%xmm0 215 vpxor %xmm4,%xmm1,%xmm1 216 217 vpsrlq $1,%xmm0,%xmm4 218 vpxor %xmm0,%xmm1,%xmm1 219 vpxor %xmm4,%xmm0,%xmm0 220 vpsrlq $5,%xmm4,%xmm4 221 vpxor %xmm4,%xmm0,%xmm0 222 vpsrlq $1,%xmm0,%xmm0 223 vpxor %xmm1,%xmm0,%xmm0 224.Linit_start_avx: 225 vmovdqa %xmm0,%xmm5 226 vpunpckhqdq %xmm0,%xmm0,%xmm3 227 vpxor %xmm0,%xmm3,%xmm3 228 vpclmulqdq $0x11,%xmm2,%xmm0,%xmm1 229 vpclmulqdq $0x00,%xmm2,%xmm0,%xmm0 230 vpclmulqdq $0x00,%xmm6,%xmm3,%xmm3 231 vpxor %xmm0,%xmm1,%xmm4 232 vpxor %xmm4,%xmm3,%xmm3 233 234 vpslldq $8,%xmm3,%xmm4 235 vpsrldq $8,%xmm3,%xmm3 236 vpxor %xmm4,%xmm0,%xmm0 237 vpxor %xmm3,%xmm1,%xmm1 238 vpsllq $57,%xmm0,%xmm3 239 vpsllq $62,%xmm0,%xmm4 240 vpxor %xmm3,%xmm4,%xmm4 241 vpsllq $63,%xmm0,%xmm3 242 vpxor %xmm3,%xmm4,%xmm4 243 vpslldq $8,%xmm4,%xmm3 244 vpsrldq $8,%xmm4,%xmm4 245 vpxor %xmm3,%xmm0,%xmm0 246 vpxor %xmm4,%xmm1,%xmm1 247 248 vpsrlq $1,%xmm0,%xmm4 249 vpxor %xmm0,%xmm1,%xmm1 250 vpxor %xmm4,%xmm0,%xmm0 251 vpsrlq $5,%xmm4,%xmm4 252 vpxor %xmm4,%xmm0,%xmm0 253 vpsrlq $1,%xmm0,%xmm0 254 vpxor %xmm1,%xmm0,%xmm0 255 vpshufd $78,%xmm5,%xmm3 256 vpshufd $78,%xmm0,%xmm4 257 vpxor %xmm5,%xmm3,%xmm3 258 vmovdqu %xmm5,0(%rdi) 259 vpxor %xmm0,%xmm4,%xmm4 260 vmovdqu %xmm0,16(%rdi) 261 leaq 48(%rdi),%rdi 262 subq $1,%r10 263 jnz .Linit_loop_avx 264 265 vpalignr $8,%xmm4,%xmm3,%xmm5 266 vmovdqu %xmm5,-16(%rdi) 267 268 vzeroupper 269 RET 270.cfi_endproc 271SET_SIZE(gcm_init_htab_avx) 272 273#if !defined (_WIN32) || defined (_KERNEL) 274ENTRY_ALIGN(gcm_gmult_avx, 32) 275.cfi_startproc 276 ENDBR 277 jmp .L_gmult_clmul 278.cfi_endproc 279SET_SIZE(gcm_gmult_avx) 280 281ENTRY_ALIGN(gcm_ghash_avx, 32) 282.cfi_startproc 283 ENDBR 284 vzeroupper 285 286 vmovdqu (%rdi),%xmm10 287 leaq .L0x1c2_polynomial(%rip),%r10 288 leaq 64(%rsi),%rsi 289 vmovdqu .Lbswap_mask(%rip),%xmm13 290 vpshufb %xmm13,%xmm10,%xmm10 291 cmpq $0x80,%rcx 292 jb .Lshort_avx 293 subq $0x80,%rcx 294 295 vmovdqu 112(%rdx),%xmm14 296 vmovdqu 0-64(%rsi),%xmm6 297 vpshufb %xmm13,%xmm14,%xmm14 298 vmovdqu 32-64(%rsi),%xmm7 299 300 vpunpckhqdq %xmm14,%xmm14,%xmm9 301 vmovdqu 96(%rdx),%xmm15 302 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 303 vpxor %xmm14,%xmm9,%xmm9 304 vpshufb %xmm13,%xmm15,%xmm15 305 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 306 vmovdqu 16-64(%rsi),%xmm6 307 vpunpckhqdq %xmm15,%xmm15,%xmm8 308 vmovdqu 80(%rdx),%xmm14 309 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 310 vpxor %xmm15,%xmm8,%xmm8 311 312 vpshufb %xmm13,%xmm14,%xmm14 313 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 314 vpunpckhqdq %xmm14,%xmm14,%xmm9 315 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 316 vmovdqu 48-64(%rsi),%xmm6 317 vpxor %xmm14,%xmm9,%xmm9 318 vmovdqu 64(%rdx),%xmm15 319 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 320 vmovdqu 80-64(%rsi),%xmm7 321 322 vpshufb %xmm13,%xmm15,%xmm15 323 vpxor %xmm0,%xmm3,%xmm3 324 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 325 vpxor %xmm1,%xmm4,%xmm4 326 vpunpckhqdq %xmm15,%xmm15,%xmm8 327 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 328 vmovdqu 64-64(%rsi),%xmm6 329 vpxor %xmm2,%xmm5,%xmm5 330 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 331 vpxor %xmm15,%xmm8,%xmm8 332 333 vmovdqu 48(%rdx),%xmm14 334 vpxor %xmm3,%xmm0,%xmm0 335 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 336 vpxor %xmm4,%xmm1,%xmm1 337 vpshufb %xmm13,%xmm14,%xmm14 338 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 339 vmovdqu 96-64(%rsi),%xmm6 340 vpxor %xmm5,%xmm2,%xmm2 341 vpunpckhqdq %xmm14,%xmm14,%xmm9 342 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 343 vmovdqu 128-64(%rsi),%xmm7 344 vpxor %xmm14,%xmm9,%xmm9 345 346 vmovdqu 32(%rdx),%xmm15 347 vpxor %xmm0,%xmm3,%xmm3 348 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 349 vpxor %xmm1,%xmm4,%xmm4 350 vpshufb %xmm13,%xmm15,%xmm15 351 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 352 vmovdqu 112-64(%rsi),%xmm6 353 vpxor %xmm2,%xmm5,%xmm5 354 vpunpckhqdq %xmm15,%xmm15,%xmm8 355 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 356 vpxor %xmm15,%xmm8,%xmm8 357 358 vmovdqu 16(%rdx),%xmm14 359 vpxor %xmm3,%xmm0,%xmm0 360 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 361 vpxor %xmm4,%xmm1,%xmm1 362 vpshufb %xmm13,%xmm14,%xmm14 363 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 364 vmovdqu 144-64(%rsi),%xmm6 365 vpxor %xmm5,%xmm2,%xmm2 366 vpunpckhqdq %xmm14,%xmm14,%xmm9 367 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 368 vmovdqu 176-64(%rsi),%xmm7 369 vpxor %xmm14,%xmm9,%xmm9 370 371 vmovdqu (%rdx),%xmm15 372 vpxor %xmm0,%xmm3,%xmm3 373 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 374 vpxor %xmm1,%xmm4,%xmm4 375 vpshufb %xmm13,%xmm15,%xmm15 376 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 377 vmovdqu 160-64(%rsi),%xmm6 378 vpxor %xmm2,%xmm5,%xmm5 379 vpclmulqdq $0x10,%xmm7,%xmm9,%xmm2 380 381 leaq 128(%rdx),%rdx 382 cmpq $0x80,%rcx 383 jb .Ltail_avx 384 385 vpxor %xmm10,%xmm15,%xmm15 386 subq $0x80,%rcx 387 jmp .Loop8x_avx 388 389.balign 32 390.Loop8x_avx: 391 vpunpckhqdq %xmm15,%xmm15,%xmm8 392 vmovdqu 112(%rdx),%xmm14 393 vpxor %xmm0,%xmm3,%xmm3 394 vpxor %xmm15,%xmm8,%xmm8 395 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm10 396 vpshufb %xmm13,%xmm14,%xmm14 397 vpxor %xmm1,%xmm4,%xmm4 398 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm11 399 vmovdqu 0-64(%rsi),%xmm6 400 vpunpckhqdq %xmm14,%xmm14,%xmm9 401 vpxor %xmm2,%xmm5,%xmm5 402 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm12 403 vmovdqu 32-64(%rsi),%xmm7 404 vpxor %xmm14,%xmm9,%xmm9 405 406 vmovdqu 96(%rdx),%xmm15 407 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 408 vpxor %xmm3,%xmm10,%xmm10 409 vpshufb %xmm13,%xmm15,%xmm15 410 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 411 vxorps %xmm4,%xmm11,%xmm11 412 vmovdqu 16-64(%rsi),%xmm6 413 vpunpckhqdq %xmm15,%xmm15,%xmm8 414 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 415 vpxor %xmm5,%xmm12,%xmm12 416 vxorps %xmm15,%xmm8,%xmm8 417 418 vmovdqu 80(%rdx),%xmm14 419 vpxor %xmm10,%xmm12,%xmm12 420 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 421 vpxor %xmm11,%xmm12,%xmm12 422 vpslldq $8,%xmm12,%xmm9 423 vpxor %xmm0,%xmm3,%xmm3 424 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 425 vpsrldq $8,%xmm12,%xmm12 426 vpxor %xmm9,%xmm10,%xmm10 427 vmovdqu 48-64(%rsi),%xmm6 428 vpshufb %xmm13,%xmm14,%xmm14 429 vxorps %xmm12,%xmm11,%xmm11 430 vpxor %xmm1,%xmm4,%xmm4 431 vpunpckhqdq %xmm14,%xmm14,%xmm9 432 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 433 vmovdqu 80-64(%rsi),%xmm7 434 vpxor %xmm14,%xmm9,%xmm9 435 vpxor %xmm2,%xmm5,%xmm5 436 437 vmovdqu 64(%rdx),%xmm15 438 vpalignr $8,%xmm10,%xmm10,%xmm12 439 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 440 vpshufb %xmm13,%xmm15,%xmm15 441 vpxor %xmm3,%xmm0,%xmm0 442 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 443 vmovdqu 64-64(%rsi),%xmm6 444 vpunpckhqdq %xmm15,%xmm15,%xmm8 445 vpxor %xmm4,%xmm1,%xmm1 446 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 447 vxorps %xmm15,%xmm8,%xmm8 448 vpxor %xmm5,%xmm2,%xmm2 449 450 vmovdqu 48(%rdx),%xmm14 451 vpclmulqdq $0x10,(%r10),%xmm10,%xmm10 452 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 453 vpshufb %xmm13,%xmm14,%xmm14 454 vpxor %xmm0,%xmm3,%xmm3 455 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 456 vmovdqu 96-64(%rsi),%xmm6 457 vpunpckhqdq %xmm14,%xmm14,%xmm9 458 vpxor %xmm1,%xmm4,%xmm4 459 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 460 vmovdqu 128-64(%rsi),%xmm7 461 vpxor %xmm14,%xmm9,%xmm9 462 vpxor %xmm2,%xmm5,%xmm5 463 464 vmovdqu 32(%rdx),%xmm15 465 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 466 vpshufb %xmm13,%xmm15,%xmm15 467 vpxor %xmm3,%xmm0,%xmm0 468 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 469 vmovdqu 112-64(%rsi),%xmm6 470 vpunpckhqdq %xmm15,%xmm15,%xmm8 471 vpxor %xmm4,%xmm1,%xmm1 472 vpclmulqdq $0x00,%xmm7,%xmm9,%xmm2 473 vpxor %xmm15,%xmm8,%xmm8 474 vpxor %xmm5,%xmm2,%xmm2 475 vxorps %xmm12,%xmm10,%xmm10 476 477 vmovdqu 16(%rdx),%xmm14 478 vpalignr $8,%xmm10,%xmm10,%xmm12 479 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm3 480 vpshufb %xmm13,%xmm14,%xmm14 481 vpxor %xmm0,%xmm3,%xmm3 482 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm4 483 vmovdqu 144-64(%rsi),%xmm6 484 vpclmulqdq $0x10,(%r10),%xmm10,%xmm10 485 vxorps %xmm11,%xmm12,%xmm12 486 vpunpckhqdq %xmm14,%xmm14,%xmm9 487 vpxor %xmm1,%xmm4,%xmm4 488 vpclmulqdq $0x10,%xmm7,%xmm8,%xmm5 489 vmovdqu 176-64(%rsi),%xmm7 490 vpxor %xmm14,%xmm9,%xmm9 491 vpxor %xmm2,%xmm5,%xmm5 492 493 vmovdqu (%rdx),%xmm15 494 vpclmulqdq $0x00,%xmm6,%xmm14,%xmm0 495 vpshufb %xmm13,%xmm15,%xmm15 496 vpclmulqdq $0x11,%xmm6,%xmm14,%xmm1 497 vmovdqu 160-64(%rsi),%xmm6 498 vpxor %xmm12,%xmm15,%xmm15 499 vpclmulqdq $0x10,%xmm7,%xmm9,%xmm2 500 vpxor %xmm10,%xmm15,%xmm15 501 502 leaq 128(%rdx),%rdx 503 subq $0x80,%rcx 504 jnc .Loop8x_avx 505 506 addq $0x80,%rcx 507 jmp .Ltail_no_xor_avx 508 509.balign 32 510.Lshort_avx: 511 vmovdqu -16(%rdx,%rcx,1),%xmm14 512 leaq (%rdx,%rcx,1),%rdx 513 vmovdqu 0-64(%rsi),%xmm6 514 vmovdqu 32-64(%rsi),%xmm7 515 vpshufb %xmm13,%xmm14,%xmm15 516 517 vmovdqa %xmm0,%xmm3 518 vmovdqa %xmm1,%xmm4 519 vmovdqa %xmm2,%xmm5 520 subq $0x10,%rcx 521 jz .Ltail_avx 522 523 vpunpckhqdq %xmm15,%xmm15,%xmm8 524 vpxor %xmm0,%xmm3,%xmm3 525 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 526 vpxor %xmm15,%xmm8,%xmm8 527 vmovdqu -32(%rdx),%xmm14 528 vpxor %xmm1,%xmm4,%xmm4 529 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 530 vmovdqu 16-64(%rsi),%xmm6 531 vpshufb %xmm13,%xmm14,%xmm15 532 vpxor %xmm2,%xmm5,%xmm5 533 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 534 vpsrldq $8,%xmm7,%xmm7 535 subq $0x10,%rcx 536 jz .Ltail_avx 537 538 vpunpckhqdq %xmm15,%xmm15,%xmm8 539 vpxor %xmm0,%xmm3,%xmm3 540 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 541 vpxor %xmm15,%xmm8,%xmm8 542 vmovdqu -48(%rdx),%xmm14 543 vpxor %xmm1,%xmm4,%xmm4 544 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 545 vmovdqu 48-64(%rsi),%xmm6 546 vpshufb %xmm13,%xmm14,%xmm15 547 vpxor %xmm2,%xmm5,%xmm5 548 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 549 vmovdqu 80-64(%rsi),%xmm7 550 subq $0x10,%rcx 551 jz .Ltail_avx 552 553 vpunpckhqdq %xmm15,%xmm15,%xmm8 554 vpxor %xmm0,%xmm3,%xmm3 555 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 556 vpxor %xmm15,%xmm8,%xmm8 557 vmovdqu -64(%rdx),%xmm14 558 vpxor %xmm1,%xmm4,%xmm4 559 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 560 vmovdqu 64-64(%rsi),%xmm6 561 vpshufb %xmm13,%xmm14,%xmm15 562 vpxor %xmm2,%xmm5,%xmm5 563 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 564 vpsrldq $8,%xmm7,%xmm7 565 subq $0x10,%rcx 566 jz .Ltail_avx 567 568 vpunpckhqdq %xmm15,%xmm15,%xmm8 569 vpxor %xmm0,%xmm3,%xmm3 570 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 571 vpxor %xmm15,%xmm8,%xmm8 572 vmovdqu -80(%rdx),%xmm14 573 vpxor %xmm1,%xmm4,%xmm4 574 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 575 vmovdqu 96-64(%rsi),%xmm6 576 vpshufb %xmm13,%xmm14,%xmm15 577 vpxor %xmm2,%xmm5,%xmm5 578 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 579 vmovdqu 128-64(%rsi),%xmm7 580 subq $0x10,%rcx 581 jz .Ltail_avx 582 583 vpunpckhqdq %xmm15,%xmm15,%xmm8 584 vpxor %xmm0,%xmm3,%xmm3 585 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 586 vpxor %xmm15,%xmm8,%xmm8 587 vmovdqu -96(%rdx),%xmm14 588 vpxor %xmm1,%xmm4,%xmm4 589 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 590 vmovdqu 112-64(%rsi),%xmm6 591 vpshufb %xmm13,%xmm14,%xmm15 592 vpxor %xmm2,%xmm5,%xmm5 593 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 594 vpsrldq $8,%xmm7,%xmm7 595 subq $0x10,%rcx 596 jz .Ltail_avx 597 598 vpunpckhqdq %xmm15,%xmm15,%xmm8 599 vpxor %xmm0,%xmm3,%xmm3 600 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 601 vpxor %xmm15,%xmm8,%xmm8 602 vmovdqu -112(%rdx),%xmm14 603 vpxor %xmm1,%xmm4,%xmm4 604 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 605 vmovdqu 144-64(%rsi),%xmm6 606 vpshufb %xmm13,%xmm14,%xmm15 607 vpxor %xmm2,%xmm5,%xmm5 608 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 609 vmovq 184-64(%rsi),%xmm7 610 subq $0x10,%rcx 611 jmp .Ltail_avx 612 613.balign 32 614.Ltail_avx: 615 vpxor %xmm10,%xmm15,%xmm15 616.Ltail_no_xor_avx: 617 vpunpckhqdq %xmm15,%xmm15,%xmm8 618 vpxor %xmm0,%xmm3,%xmm3 619 vpclmulqdq $0x00,%xmm6,%xmm15,%xmm0 620 vpxor %xmm15,%xmm8,%xmm8 621 vpxor %xmm1,%xmm4,%xmm4 622 vpclmulqdq $0x11,%xmm6,%xmm15,%xmm1 623 vpxor %xmm2,%xmm5,%xmm5 624 vpclmulqdq $0x00,%xmm7,%xmm8,%xmm2 625 626 vmovdqu (%r10),%xmm12 627 628 vpxor %xmm0,%xmm3,%xmm10 629 vpxor %xmm1,%xmm4,%xmm11 630 vpxor %xmm2,%xmm5,%xmm5 631 632 vpxor %xmm10,%xmm5,%xmm5 633 vpxor %xmm11,%xmm5,%xmm5 634 vpslldq $8,%xmm5,%xmm9 635 vpsrldq $8,%xmm5,%xmm5 636 vpxor %xmm9,%xmm10,%xmm10 637 vpxor %xmm5,%xmm11,%xmm11 638 639 vpclmulqdq $0x10,%xmm12,%xmm10,%xmm9 640 vpalignr $8,%xmm10,%xmm10,%xmm10 641 vpxor %xmm9,%xmm10,%xmm10 642 643 vpclmulqdq $0x10,%xmm12,%xmm10,%xmm9 644 vpalignr $8,%xmm10,%xmm10,%xmm10 645 vpxor %xmm11,%xmm10,%xmm10 646 vpxor %xmm9,%xmm10,%xmm10 647 648 cmpq $0,%rcx 649 jne .Lshort_avx 650 651 vpshufb %xmm13,%xmm10,%xmm10 652 vmovdqu %xmm10,(%rdi) 653 vzeroupper 654 RET 655.cfi_endproc 656SET_SIZE(gcm_ghash_avx) 657 658#endif /* !_WIN32 || _KERNEL */ 659 660SECTION_STATIC 661.balign 64 662.Lbswap_mask: 663.byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 664.L0x1c2_polynomial: 665.byte 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2 666.L7_mask: 667.long 7,0,7,0 668.L7_mask_poly: 669.long 7,0,450,0 670.balign 64 671SET_OBJ(.Lrem_4bit) 672.Lrem_4bit: 673.long 0,0,0,471859200,0,943718400,0,610271232 674.long 0,1887436800,0,1822425088,0,1220542464,0,1423966208 675.long 0,3774873600,0,4246732800,0,3644850176,0,3311403008 676.long 0,2441084928,0,2376073216,0,2847932416,0,3051356160 677SET_OBJ(.Lrem_8bit) 678.Lrem_8bit: 679.value 0x0000,0x01C2,0x0384,0x0246,0x0708,0x06CA,0x048C,0x054E 680.value 0x0E10,0x0FD2,0x0D94,0x0C56,0x0918,0x08DA,0x0A9C,0x0B5E 681.value 0x1C20,0x1DE2,0x1FA4,0x1E66,0x1B28,0x1AEA,0x18AC,0x196E 682.value 0x1230,0x13F2,0x11B4,0x1076,0x1538,0x14FA,0x16BC,0x177E 683.value 0x3840,0x3982,0x3BC4,0x3A06,0x3F48,0x3E8A,0x3CCC,0x3D0E 684.value 0x3650,0x3792,0x35D4,0x3416,0x3158,0x309A,0x32DC,0x331E 685.value 0x2460,0x25A2,0x27E4,0x2626,0x2368,0x22AA,0x20EC,0x212E 686.value 0x2A70,0x2BB2,0x29F4,0x2836,0x2D78,0x2CBA,0x2EFC,0x2F3E 687.value 0x7080,0x7142,0x7304,0x72C6,0x7788,0x764A,0x740C,0x75CE 688.value 0x7E90,0x7F52,0x7D14,0x7CD6,0x7998,0x785A,0x7A1C,0x7BDE 689.value 0x6CA0,0x6D62,0x6F24,0x6EE6,0x6BA8,0x6A6A,0x682C,0x69EE 690.value 0x62B0,0x6372,0x6134,0x60F6,0x65B8,0x647A,0x663C,0x67FE 691.value 0x48C0,0x4902,0x4B44,0x4A86,0x4FC8,0x4E0A,0x4C4C,0x4D8E 692.value 0x46D0,0x4712,0x4554,0x4496,0x41D8,0x401A,0x425C,0x439E 693.value 0x54E0,0x5522,0x5764,0x56A6,0x53E8,0x522A,0x506C,0x51AE 694.value 0x5AF0,0x5B32,0x5974,0x58B6,0x5DF8,0x5C3A,0x5E7C,0x5FBE 695.value 0xE100,0xE0C2,0xE284,0xE346,0xE608,0xE7CA,0xE58C,0xE44E 696.value 0xEF10,0xEED2,0xEC94,0xED56,0xE818,0xE9DA,0xEB9C,0xEA5E 697.value 0xFD20,0xFCE2,0xFEA4,0xFF66,0xFA28,0xFBEA,0xF9AC,0xF86E 698.value 0xF330,0xF2F2,0xF0B4,0xF176,0xF438,0xF5FA,0xF7BC,0xF67E 699.value 0xD940,0xD882,0xDAC4,0xDB06,0xDE48,0xDF8A,0xDDCC,0xDC0E 700.value 0xD750,0xD692,0xD4D4,0xD516,0xD058,0xD19A,0xD3DC,0xD21E 701.value 0xC560,0xC4A2,0xC6E4,0xC726,0xC268,0xC3AA,0xC1EC,0xC02E 702.value 0xCB70,0xCAB2,0xC8F4,0xC936,0xCC78,0xCDBA,0xCFFC,0xCE3E 703.value 0x9180,0x9042,0x9204,0x93C6,0x9688,0x974A,0x950C,0x94CE 704.value 0x9F90,0x9E52,0x9C14,0x9DD6,0x9898,0x995A,0x9B1C,0x9ADE 705.value 0x8DA0,0x8C62,0x8E24,0x8FE6,0x8AA8,0x8B6A,0x892C,0x88EE 706.value 0x83B0,0x8272,0x8034,0x81F6,0x84B8,0x857A,0x873C,0x86FE 707.value 0xA9C0,0xA802,0xAA44,0xAB86,0xAEC8,0xAF0A,0xAD4C,0xAC8E 708.value 0xA7D0,0xA612,0xA454,0xA596,0xA0D8,0xA11A,0xA35C,0xA29E 709.value 0xB5E0,0xB422,0xB664,0xB7A6,0xB2E8,0xB32A,0xB16C,0xB0AE 710.value 0xBBF0,0xBA32,0xB874,0xB9B6,0xBCF8,0xBD3A,0xBF7C,0xBEBE 711 712.byte 71,72,65,83,72,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0 713.balign 64 714 715/* Mark the stack non-executable. */ 716#if defined(__linux__) && defined(__ELF__) 717.section .note.GNU-stack,"",%progbits 718#endif 719 720#endif /* defined(__x86_64__) && defined(HAVE_AVX) && defined(HAVE_AES) ... */ 721