cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

sm4-aesni-avx2-asm_64.S (15690B)


      1// SPDX-License-Identifier: GPL-2.0-or-later
      2/*
      3 * SM4 Cipher Algorithm, AES-NI/AVX2 optimized.
      4 * as specified in
      5 * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
      6 *
      7 * Copyright (C) 2018 Markku-Juhani O. Saarinen <mjos@iki.fi>
      8 * Copyright (C) 2020 Jussi Kivilinna <jussi.kivilinna@iki.fi>
      9 * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
     10 */
     11
     12/* Based on SM4 AES-NI work by libgcrypt and Markku-Juhani O. Saarinen at:
     13 *  https://github.com/mjosaarinen/sm4ni
     14 */
     15
     16#include <linux/linkage.h>
     17#include <asm/frame.h>
     18
     19#define rRIP         (%rip)
     20
     21/* vector registers */
     22#define RX0          %ymm0
     23#define RX1          %ymm1
     24#define MASK_4BIT    %ymm2
     25#define RTMP0        %ymm3
     26#define RTMP1        %ymm4
     27#define RTMP2        %ymm5
     28#define RTMP3        %ymm6
     29#define RTMP4        %ymm7
     30
     31#define RA0          %ymm8
     32#define RA1          %ymm9
     33#define RA2          %ymm10
     34#define RA3          %ymm11
     35
     36#define RB0          %ymm12
     37#define RB1          %ymm13
     38#define RB2          %ymm14
     39#define RB3          %ymm15
     40
     41#define RNOT         %ymm0
     42#define RBSWAP       %ymm1
     43
     44#define RX0x         %xmm0
     45#define RX1x         %xmm1
     46#define MASK_4BITx   %xmm2
     47
     48#define RNOTx        %xmm0
     49#define RBSWAPx      %xmm1
     50
     51#define RTMP0x       %xmm3
     52#define RTMP1x       %xmm4
     53#define RTMP2x       %xmm5
     54#define RTMP3x       %xmm6
     55#define RTMP4x       %xmm7
     56
     57
     58/* helper macros */
     59
     60/* Transpose four 32-bit words between 128-bit vector lanes. */
     61#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
     62	vpunpckhdq x1, x0, t2;                \
     63	vpunpckldq x1, x0, x0;                \
     64	                                      \
     65	vpunpckldq x3, x2, t1;                \
     66	vpunpckhdq x3, x2, x2;                \
     67	                                      \
     68	vpunpckhqdq t1, x0, x1;               \
     69	vpunpcklqdq t1, x0, x0;               \
     70	                                      \
     71	vpunpckhqdq x2, t2, x3;               \
     72	vpunpcklqdq x2, t2, x2;
     73
     74/* post-SubByte transform. */
     75#define transform_pre(x, lo_t, hi_t, mask4bit, tmp0) \
     76	vpand x, mask4bit, tmp0;                     \
     77	vpandn x, mask4bit, x;                       \
     78	vpsrld $4, x, x;                             \
     79	                                             \
     80	vpshufb tmp0, lo_t, tmp0;                    \
     81	vpshufb x, hi_t, x;                          \
     82	vpxor tmp0, x, x;
     83
     84/* post-SubByte transform. Note: x has been XOR'ed with mask4bit by
     85 * 'vaeslastenc' instruction. */
     86#define transform_post(x, lo_t, hi_t, mask4bit, tmp0) \
     87	vpandn mask4bit, x, tmp0;                     \
     88	vpsrld $4, x, x;                              \
     89	vpand x, mask4bit, x;                         \
     90	                                              \
     91	vpshufb tmp0, lo_t, tmp0;                     \
     92	vpshufb x, hi_t, x;                           \
     93	vpxor tmp0, x, x;
     94
     95
     96.section	.rodata.cst16, "aM", @progbits, 16
     97.align 16
     98
     99/*
    100 * Following four affine transform look-up tables are from work by
    101 * Markku-Juhani O. Saarinen, at https://github.com/mjosaarinen/sm4ni
    102 *
    103 * These allow exposing SM4 S-Box from AES SubByte.
    104 */
    105
    106/* pre-SubByte affine transform, from SM4 field to AES field. */
    107.Lpre_tf_lo_s:
    108	.quad 0x9197E2E474720701, 0xC7C1B4B222245157
    109.Lpre_tf_hi_s:
    110	.quad 0xE240AB09EB49A200, 0xF052B91BF95BB012
    111
    112/* post-SubByte affine transform, from AES field to SM4 field. */
    113.Lpost_tf_lo_s:
    114	.quad 0x5B67F2CEA19D0834, 0xEDD14478172BBE82
    115.Lpost_tf_hi_s:
    116	.quad 0xAE7201DD73AFDC00, 0x11CDBE62CC1063BF
    117
    118/* For isolating SubBytes from AESENCLAST, inverse shift row */
    119.Linv_shift_row:
    120	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
    121	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
    122
    123/* Inverse shift row + Rotate left by 8 bits on 32-bit words with vpshufb */
    124.Linv_shift_row_rol_8:
    125	.byte 0x07, 0x00, 0x0d, 0x0a, 0x0b, 0x04, 0x01, 0x0e
    126	.byte 0x0f, 0x08, 0x05, 0x02, 0x03, 0x0c, 0x09, 0x06
    127
    128/* Inverse shift row + Rotate left by 16 bits on 32-bit words with vpshufb */
    129.Linv_shift_row_rol_16:
    130	.byte 0x0a, 0x07, 0x00, 0x0d, 0x0e, 0x0b, 0x04, 0x01
    131	.byte 0x02, 0x0f, 0x08, 0x05, 0x06, 0x03, 0x0c, 0x09
    132
    133/* Inverse shift row + Rotate left by 24 bits on 32-bit words with vpshufb */
    134.Linv_shift_row_rol_24:
    135	.byte 0x0d, 0x0a, 0x07, 0x00, 0x01, 0x0e, 0x0b, 0x04
    136	.byte 0x05, 0x02, 0x0f, 0x08, 0x09, 0x06, 0x03, 0x0c
    137
    138/* For CTR-mode IV byteswap */
    139.Lbswap128_mask:
    140	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
    141
    142/* For input word byte-swap */
    143.Lbswap32_mask:
    144	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
    145
    146.align 4
    147/* 4-bit mask */
    148.L0f0f0f0f:
    149	.long 0x0f0f0f0f
    150
    151/* 12 bytes, only for padding */
    152.Lpadding_deadbeef:
    153	.long 0xdeadbeef, 0xdeadbeef, 0xdeadbeef
    154
    155.text
    156.align 16
    157
    158.align 8
    159SYM_FUNC_START_LOCAL(__sm4_crypt_blk16)
    160	/* input:
    161	 *	%rdi: round key array, CTX
    162	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
    163	 *						plaintext blocks
    164	 * output:
    165	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
    166	 * 						ciphertext blocks
    167	 */
    168	FRAME_BEGIN
    169
    170	vbroadcasti128 .Lbswap32_mask rRIP, RTMP2;
    171	vpshufb RTMP2, RA0, RA0;
    172	vpshufb RTMP2, RA1, RA1;
    173	vpshufb RTMP2, RA2, RA2;
    174	vpshufb RTMP2, RA3, RA3;
    175	vpshufb RTMP2, RB0, RB0;
    176	vpshufb RTMP2, RB1, RB1;
    177	vpshufb RTMP2, RB2, RB2;
    178	vpshufb RTMP2, RB3, RB3;
    179
    180	vpbroadcastd .L0f0f0f0f rRIP, MASK_4BIT;
    181	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
    182	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
    183
    184#define ROUND(round, s0, s1, s2, s3, r0, r1, r2, r3)                \
    185	vpbroadcastd (4*(round))(%rdi), RX0;                        \
    186	vbroadcasti128 .Lpre_tf_lo_s rRIP, RTMP4;                   \
    187	vbroadcasti128 .Lpre_tf_hi_s rRIP, RTMP1;                   \
    188	vmovdqa RX0, RX1;                                           \
    189	vpxor s1, RX0, RX0;                                         \
    190	vpxor s2, RX0, RX0;                                         \
    191	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
    192	vbroadcasti128 .Lpost_tf_lo_s rRIP, RTMP2;                  \
    193	vbroadcasti128 .Lpost_tf_hi_s rRIP, RTMP3;                  \
    194	vpxor r1, RX1, RX1;                                         \
    195	vpxor r2, RX1, RX1;                                         \
    196	vpxor r3, RX1, RX1; /* r1 ^ r2 ^ r3 ^ rk */                 \
    197	                                                            \
    198	/* sbox, non-linear part */                                 \
    199	transform_pre(RX0, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
    200	transform_pre(RX1, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
    201	vextracti128 $1, RX0, RTMP4x;                               \
    202	vextracti128 $1, RX1, RTMP0x;                               \
    203	vaesenclast MASK_4BITx, RX0x, RX0x;                         \
    204	vaesenclast MASK_4BITx, RTMP4x, RTMP4x;                     \
    205	vaesenclast MASK_4BITx, RX1x, RX1x;                         \
    206	vaesenclast MASK_4BITx, RTMP0x, RTMP0x;                     \
    207	vinserti128 $1, RTMP4x, RX0, RX0;                           \
    208	vbroadcasti128 .Linv_shift_row rRIP, RTMP4;                 \
    209	vinserti128 $1, RTMP0x, RX1, RX1;                           \
    210	transform_post(RX0, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
    211	transform_post(RX1, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
    212	                                                            \
    213	/* linear part */                                           \
    214	vpshufb RTMP4, RX0, RTMP0;                                  \
    215	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
    216	vpshufb RTMP4, RX1, RTMP2;                                  \
    217	vbroadcasti128 .Linv_shift_row_rol_8 rRIP, RTMP4;           \
    218	vpxor RTMP2, r0, r0; /* r0 ^ x */                           \
    219	vpshufb RTMP4, RX0, RTMP1;                                  \
    220	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
    221	vpshufb RTMP4, RX1, RTMP3;                                  \
    222	vbroadcasti128 .Linv_shift_row_rol_16 rRIP, RTMP4;          \
    223	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) */               \
    224	vpshufb RTMP4, RX0, RTMP1;                                  \
    225	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
    226	vpshufb RTMP4, RX1, RTMP3;                                  \
    227	vbroadcasti128 .Linv_shift_row_rol_24 rRIP, RTMP4;          \
    228	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) ^ rol(x,16) */   \
    229	vpshufb RTMP4, RX0, RTMP1;                                  \
    230	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
    231	vpslld $2, RTMP0, RTMP1;                                    \
    232	vpsrld $30, RTMP0, RTMP0;                                   \
    233	vpxor RTMP0, s0, s0;                                        \
    234	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
    235	vpxor RTMP1, s0, s0;                                        \
    236	vpshufb RTMP4, RX1, RTMP3;                                  \
    237	vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,24) */               \
    238	vpslld $2, RTMP2, RTMP3;                                    \
    239	vpsrld $30, RTMP2, RTMP2;                                   \
    240	vpxor RTMP2, r0, r0;                                        \
    241	/* r0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
    242	vpxor RTMP3, r0, r0;
    243
    244	leaq (32*4)(%rdi), %rax;
    245.align 16
    246.Lroundloop_blk8:
    247	ROUND(0, RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3);
    248	ROUND(1, RA1, RA2, RA3, RA0, RB1, RB2, RB3, RB0);
    249	ROUND(2, RA2, RA3, RA0, RA1, RB2, RB3, RB0, RB1);
    250	ROUND(3, RA3, RA0, RA1, RA2, RB3, RB0, RB1, RB2);
    251	leaq (4*4)(%rdi), %rdi;
    252	cmpq %rax, %rdi;
    253	jne .Lroundloop_blk8;
    254
    255#undef ROUND
    256
    257	vbroadcasti128 .Lbswap128_mask rRIP, RTMP2;
    258
    259	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
    260	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
    261	vpshufb RTMP2, RA0, RA0;
    262	vpshufb RTMP2, RA1, RA1;
    263	vpshufb RTMP2, RA2, RA2;
    264	vpshufb RTMP2, RA3, RA3;
    265	vpshufb RTMP2, RB0, RB0;
    266	vpshufb RTMP2, RB1, RB1;
    267	vpshufb RTMP2, RB2, RB2;
    268	vpshufb RTMP2, RB3, RB3;
    269
    270	FRAME_END
    271	RET;
    272SYM_FUNC_END(__sm4_crypt_blk16)
    273
    274#define inc_le128(x, minus_one, tmp) \
    275	vpcmpeqq minus_one, x, tmp;  \
    276	vpsubq minus_one, x, x;      \
    277	vpslldq $8, tmp, tmp;        \
    278	vpsubq tmp, x, x;
    279
    280/*
    281 * void sm4_aesni_avx2_ctr_enc_blk16(const u32 *rk, u8 *dst,
    282 *                                   const u8 *src, u8 *iv)
    283 */
    284.align 8
    285SYM_FUNC_START(sm4_aesni_avx2_ctr_enc_blk16)
    286	/* input:
    287	 *	%rdi: round key array, CTX
    288	 *	%rsi: dst (16 blocks)
    289	 *	%rdx: src (16 blocks)
    290	 *	%rcx: iv (big endian, 128bit)
    291	 */
    292	FRAME_BEGIN
    293
    294	movq 8(%rcx), %rax;
    295	bswapq %rax;
    296
    297	vzeroupper;
    298
    299	vbroadcasti128 .Lbswap128_mask rRIP, RTMP3;
    300	vpcmpeqd RNOT, RNOT, RNOT;
    301	vpsrldq $8, RNOT, RNOT;   /* ab: -1:0 ; cd: -1:0 */
    302	vpaddq RNOT, RNOT, RTMP2; /* ab: -2:0 ; cd: -2:0 */
    303
    304	/* load IV and byteswap */
    305	vmovdqu (%rcx), RTMP4x;
    306	vpshufb RTMP3x, RTMP4x, RTMP4x;
    307	vmovdqa RTMP4x, RTMP0x;
    308	inc_le128(RTMP4x, RNOTx, RTMP1x);
    309	vinserti128 $1, RTMP4x, RTMP0, RTMP0;
    310	vpshufb RTMP3, RTMP0, RA0; /* +1 ; +0 */
    311
    312	/* check need for handling 64-bit overflow and carry */
    313	cmpq $(0xffffffffffffffff - 16), %rax;
    314	ja .Lhandle_ctr_carry;
    315
    316	/* construct IVs */
    317	vpsubq RTMP2, RTMP0, RTMP0; /* +3 ; +2 */
    318	vpshufb RTMP3, RTMP0, RA1;
    319	vpsubq RTMP2, RTMP0, RTMP0; /* +5 ; +4 */
    320	vpshufb RTMP3, RTMP0, RA2;
    321	vpsubq RTMP2, RTMP0, RTMP0; /* +7 ; +6 */
    322	vpshufb RTMP3, RTMP0, RA3;
    323	vpsubq RTMP2, RTMP0, RTMP0; /* +9 ; +8 */
    324	vpshufb RTMP3, RTMP0, RB0;
    325	vpsubq RTMP2, RTMP0, RTMP0; /* +11 ; +10 */
    326	vpshufb RTMP3, RTMP0, RB1;
    327	vpsubq RTMP2, RTMP0, RTMP0; /* +13 ; +12 */
    328	vpshufb RTMP3, RTMP0, RB2;
    329	vpsubq RTMP2, RTMP0, RTMP0; /* +15 ; +14 */
    330	vpshufb RTMP3, RTMP0, RB3;
    331	vpsubq RTMP2, RTMP0, RTMP0; /* +16 */
    332	vpshufb RTMP3x, RTMP0x, RTMP0x;
    333
    334	jmp .Lctr_carry_done;
    335
    336.Lhandle_ctr_carry:
    337	/* construct IVs */
    338	inc_le128(RTMP0, RNOT, RTMP1);
    339	inc_le128(RTMP0, RNOT, RTMP1);
    340	vpshufb RTMP3, RTMP0, RA1; /* +3 ; +2 */
    341	inc_le128(RTMP0, RNOT, RTMP1);
    342	inc_le128(RTMP0, RNOT, RTMP1);
    343	vpshufb RTMP3, RTMP0, RA2; /* +5 ; +4 */
    344	inc_le128(RTMP0, RNOT, RTMP1);
    345	inc_le128(RTMP0, RNOT, RTMP1);
    346	vpshufb RTMP3, RTMP0, RA3; /* +7 ; +6 */
    347	inc_le128(RTMP0, RNOT, RTMP1);
    348	inc_le128(RTMP0, RNOT, RTMP1);
    349	vpshufb RTMP3, RTMP0, RB0; /* +9 ; +8 */
    350	inc_le128(RTMP0, RNOT, RTMP1);
    351	inc_le128(RTMP0, RNOT, RTMP1);
    352	vpshufb RTMP3, RTMP0, RB1; /* +11 ; +10 */
    353	inc_le128(RTMP0, RNOT, RTMP1);
    354	inc_le128(RTMP0, RNOT, RTMP1);
    355	vpshufb RTMP3, RTMP0, RB2; /* +13 ; +12 */
    356	inc_le128(RTMP0, RNOT, RTMP1);
    357	inc_le128(RTMP0, RNOT, RTMP1);
    358	vpshufb RTMP3, RTMP0, RB3; /* +15 ; +14 */
    359	inc_le128(RTMP0, RNOT, RTMP1);
    360	vextracti128 $1, RTMP0, RTMP0x;
    361	vpshufb RTMP3x, RTMP0x, RTMP0x; /* +16 */
    362
    363.align 4
    364.Lctr_carry_done:
    365	/* store new IV */
    366	vmovdqu RTMP0x, (%rcx);
    367
    368	call __sm4_crypt_blk16;
    369
    370	vpxor (0 * 32)(%rdx), RA0, RA0;
    371	vpxor (1 * 32)(%rdx), RA1, RA1;
    372	vpxor (2 * 32)(%rdx), RA2, RA2;
    373	vpxor (3 * 32)(%rdx), RA3, RA3;
    374	vpxor (4 * 32)(%rdx), RB0, RB0;
    375	vpxor (5 * 32)(%rdx), RB1, RB1;
    376	vpxor (6 * 32)(%rdx), RB2, RB2;
    377	vpxor (7 * 32)(%rdx), RB3, RB3;
    378
    379	vmovdqu RA0, (0 * 32)(%rsi);
    380	vmovdqu RA1, (1 * 32)(%rsi);
    381	vmovdqu RA2, (2 * 32)(%rsi);
    382	vmovdqu RA3, (3 * 32)(%rsi);
    383	vmovdqu RB0, (4 * 32)(%rsi);
    384	vmovdqu RB1, (5 * 32)(%rsi);
    385	vmovdqu RB2, (6 * 32)(%rsi);
    386	vmovdqu RB3, (7 * 32)(%rsi);
    387
    388	vzeroall;
    389	FRAME_END
    390	RET;
    391SYM_FUNC_END(sm4_aesni_avx2_ctr_enc_blk16)
    392
    393/*
    394 * void sm4_aesni_avx2_cbc_dec_blk16(const u32 *rk, u8 *dst,
    395 *                                   const u8 *src, u8 *iv)
    396 */
    397.align 8
    398SYM_FUNC_START(sm4_aesni_avx2_cbc_dec_blk16)
    399	/* input:
    400	 *	%rdi: round key array, CTX
    401	 *	%rsi: dst (16 blocks)
    402	 *	%rdx: src (16 blocks)
    403	 *	%rcx: iv
    404	 */
    405	FRAME_BEGIN
    406
    407	vzeroupper;
    408
    409	vmovdqu (0 * 32)(%rdx), RA0;
    410	vmovdqu (1 * 32)(%rdx), RA1;
    411	vmovdqu (2 * 32)(%rdx), RA2;
    412	vmovdqu (3 * 32)(%rdx), RA3;
    413	vmovdqu (4 * 32)(%rdx), RB0;
    414	vmovdqu (5 * 32)(%rdx), RB1;
    415	vmovdqu (6 * 32)(%rdx), RB2;
    416	vmovdqu (7 * 32)(%rdx), RB3;
    417
    418	call __sm4_crypt_blk16;
    419
    420	vmovdqu (%rcx), RNOTx;
    421	vinserti128 $1, (%rdx), RNOT, RNOT;
    422	vpxor RNOT, RA0, RA0;
    423	vpxor (0 * 32 + 16)(%rdx), RA1, RA1;
    424	vpxor (1 * 32 + 16)(%rdx), RA2, RA2;
    425	vpxor (2 * 32 + 16)(%rdx), RA3, RA3;
    426	vpxor (3 * 32 + 16)(%rdx), RB0, RB0;
    427	vpxor (4 * 32 + 16)(%rdx), RB1, RB1;
    428	vpxor (5 * 32 + 16)(%rdx), RB2, RB2;
    429	vpxor (6 * 32 + 16)(%rdx), RB3, RB3;
    430	vmovdqu (7 * 32 + 16)(%rdx), RNOTx;
    431	vmovdqu RNOTx, (%rcx); /* store new IV */
    432
    433	vmovdqu RA0, (0 * 32)(%rsi);
    434	vmovdqu RA1, (1 * 32)(%rsi);
    435	vmovdqu RA2, (2 * 32)(%rsi);
    436	vmovdqu RA3, (3 * 32)(%rsi);
    437	vmovdqu RB0, (4 * 32)(%rsi);
    438	vmovdqu RB1, (5 * 32)(%rsi);
    439	vmovdqu RB2, (6 * 32)(%rsi);
    440	vmovdqu RB3, (7 * 32)(%rsi);
    441
    442	vzeroall;
    443	FRAME_END
    444	RET;
    445SYM_FUNC_END(sm4_aesni_avx2_cbc_dec_blk16)
    446
    447/*
    448 * void sm4_aesni_avx2_cfb_dec_blk16(const u32 *rk, u8 *dst,
    449 *                                   const u8 *src, u8 *iv)
    450 */
    451.align 8
    452SYM_FUNC_START(sm4_aesni_avx2_cfb_dec_blk16)
    453	/* input:
    454	 *	%rdi: round key array, CTX
    455	 *	%rsi: dst (16 blocks)
    456	 *	%rdx: src (16 blocks)
    457	 *	%rcx: iv
    458	 */
    459	FRAME_BEGIN
    460
    461	vzeroupper;
    462
    463	/* Load input */
    464	vmovdqu (%rcx), RNOTx;
    465	vinserti128 $1, (%rdx), RNOT, RA0;
    466	vmovdqu (0 * 32 + 16)(%rdx), RA1;
    467	vmovdqu (1 * 32 + 16)(%rdx), RA2;
    468	vmovdqu (2 * 32 + 16)(%rdx), RA3;
    469	vmovdqu (3 * 32 + 16)(%rdx), RB0;
    470	vmovdqu (4 * 32 + 16)(%rdx), RB1;
    471	vmovdqu (5 * 32 + 16)(%rdx), RB2;
    472	vmovdqu (6 * 32 + 16)(%rdx), RB3;
    473
    474	/* Update IV */
    475	vmovdqu (7 * 32 + 16)(%rdx), RNOTx;
    476	vmovdqu RNOTx, (%rcx);
    477
    478	call __sm4_crypt_blk16;
    479
    480	vpxor (0 * 32)(%rdx), RA0, RA0;
    481	vpxor (1 * 32)(%rdx), RA1, RA1;
    482	vpxor (2 * 32)(%rdx), RA2, RA2;
    483	vpxor (3 * 32)(%rdx), RA3, RA3;
    484	vpxor (4 * 32)(%rdx), RB0, RB0;
    485	vpxor (5 * 32)(%rdx), RB1, RB1;
    486	vpxor (6 * 32)(%rdx), RB2, RB2;
    487	vpxor (7 * 32)(%rdx), RB3, RB3;
    488
    489	vmovdqu RA0, (0 * 32)(%rsi);
    490	vmovdqu RA1, (1 * 32)(%rsi);
    491	vmovdqu RA2, (2 * 32)(%rsi);
    492	vmovdqu RA3, (3 * 32)(%rsi);
    493	vmovdqu RB0, (4 * 32)(%rsi);
    494	vmovdqu RB1, (5 * 32)(%rsi);
    495	vmovdqu RB2, (6 * 32)(%rsi);
    496	vmovdqu RB3, (7 * 32)(%rsi);
    497
    498	vzeroall;
    499	FRAME_END
    500	RET;
    501SYM_FUNC_END(sm4_aesni_avx2_cfb_dec_blk16)