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

camellia-aesni-avx2-asm_64.S (28746B)


      1/* SPDX-License-Identifier: GPL-2.0-or-later */
      2/*
      3 * x86_64/AVX2/AES-NI assembler implementation of Camellia
      4 *
      5 * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
      6 */
      7
      8#include <linux/linkage.h>
      9#include <asm/frame.h>
     10
     11#define CAMELLIA_TABLE_BYTE_LEN 272
     12
     13/* struct camellia_ctx: */
     14#define key_table 0
     15#define key_length CAMELLIA_TABLE_BYTE_LEN
     16
     17/* register macros */
     18#define CTX %rdi
     19#define RIO %r8
     20
     21/**********************************************************************
     22  helper macros
     23 **********************************************************************/
     24#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
     25	vpand x, mask4bit, tmp0; \
     26	vpandn x, mask4bit, x; \
     27	vpsrld $4, x, x; \
     28	\
     29	vpshufb tmp0, lo_t, tmp0; \
     30	vpshufb x, hi_t, x; \
     31	vpxor tmp0, x, x;
     32
     33#define ymm0_x xmm0
     34#define ymm1_x xmm1
     35#define ymm2_x xmm2
     36#define ymm3_x xmm3
     37#define ymm4_x xmm4
     38#define ymm5_x xmm5
     39#define ymm6_x xmm6
     40#define ymm7_x xmm7
     41#define ymm8_x xmm8
     42#define ymm9_x xmm9
     43#define ymm10_x xmm10
     44#define ymm11_x xmm11
     45#define ymm12_x xmm12
     46#define ymm13_x xmm13
     47#define ymm14_x xmm14
     48#define ymm15_x xmm15
     49
     50/**********************************************************************
     51  32-way camellia
     52 **********************************************************************/
     53
     54/*
     55 * IN:
     56 *   x0..x7: byte-sliced AB state
     57 *   mem_cd: register pointer storing CD state
     58 *   key: index for key material
     59 * OUT:
     60 *   x0..x7: new byte-sliced CD state
     61 */
     62#define roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
     63		  t7, mem_cd, key) \
     64	/* \
     65	 * S-function with AES subbytes \
     66	 */ \
     67	vbroadcasti128 .Linv_shift_row, t4; \
     68	vpbroadcastd .L0f0f0f0f, t7; \
     69	vbroadcasti128 .Lpre_tf_lo_s1, t5; \
     70	vbroadcasti128 .Lpre_tf_hi_s1, t6; \
     71	vbroadcasti128 .Lpre_tf_lo_s4, t2; \
     72	vbroadcasti128 .Lpre_tf_hi_s4, t3; \
     73	\
     74	/* AES inverse shift rows */ \
     75	vpshufb t4, x0, x0; \
     76	vpshufb t4, x7, x7; \
     77	vpshufb t4, x3, x3; \
     78	vpshufb t4, x6, x6; \
     79	vpshufb t4, x2, x2; \
     80	vpshufb t4, x5, x5; \
     81	vpshufb t4, x1, x1; \
     82	vpshufb t4, x4, x4; \
     83	\
     84	/* prefilter sboxes 1, 2 and 3 */ \
     85	/* prefilter sbox 4 */ \
     86	filter_8bit(x0, t5, t6, t7, t4); \
     87	filter_8bit(x7, t5, t6, t7, t4); \
     88	vextracti128 $1, x0, t0##_x; \
     89	vextracti128 $1, x7, t1##_x; \
     90	filter_8bit(x3, t2, t3, t7, t4); \
     91	filter_8bit(x6, t2, t3, t7, t4); \
     92	vextracti128 $1, x3, t3##_x; \
     93	vextracti128 $1, x6, t2##_x; \
     94	filter_8bit(x2, t5, t6, t7, t4); \
     95	filter_8bit(x5, t5, t6, t7, t4); \
     96	filter_8bit(x1, t5, t6, t7, t4); \
     97	filter_8bit(x4, t5, t6, t7, t4); \
     98	\
     99	vpxor t4##_x, t4##_x, t4##_x; \
    100	\
    101	/* AES subbytes + AES shift rows */ \
    102	vextracti128 $1, x2, t6##_x; \
    103	vextracti128 $1, x5, t5##_x; \
    104	vaesenclast t4##_x, x0##_x, x0##_x; \
    105	vaesenclast t4##_x, t0##_x, t0##_x; \
    106	vinserti128 $1, t0##_x, x0, x0; \
    107	vaesenclast t4##_x, x7##_x, x7##_x; \
    108	vaesenclast t4##_x, t1##_x, t1##_x; \
    109	vinserti128 $1, t1##_x, x7, x7; \
    110	vaesenclast t4##_x, x3##_x, x3##_x; \
    111	vaesenclast t4##_x, t3##_x, t3##_x; \
    112	vinserti128 $1, t3##_x, x3, x3; \
    113	vaesenclast t4##_x, x6##_x, x6##_x; \
    114	vaesenclast t4##_x, t2##_x, t2##_x; \
    115	vinserti128 $1, t2##_x, x6, x6; \
    116	vextracti128 $1, x1, t3##_x; \
    117	vextracti128 $1, x4, t2##_x; \
    118	vbroadcasti128 .Lpost_tf_lo_s1, t0; \
    119	vbroadcasti128 .Lpost_tf_hi_s1, t1; \
    120	vaesenclast t4##_x, x2##_x, x2##_x; \
    121	vaesenclast t4##_x, t6##_x, t6##_x; \
    122	vinserti128 $1, t6##_x, x2, x2; \
    123	vaesenclast t4##_x, x5##_x, x5##_x; \
    124	vaesenclast t4##_x, t5##_x, t5##_x; \
    125	vinserti128 $1, t5##_x, x5, x5; \
    126	vaesenclast t4##_x, x1##_x, x1##_x; \
    127	vaesenclast t4##_x, t3##_x, t3##_x; \
    128	vinserti128 $1, t3##_x, x1, x1; \
    129	vaesenclast t4##_x, x4##_x, x4##_x; \
    130	vaesenclast t4##_x, t2##_x, t2##_x; \
    131	vinserti128 $1, t2##_x, x4, x4; \
    132	\
    133	/* postfilter sboxes 1 and 4 */ \
    134	vbroadcasti128 .Lpost_tf_lo_s3, t2; \
    135	vbroadcasti128 .Lpost_tf_hi_s3, t3; \
    136	filter_8bit(x0, t0, t1, t7, t6); \
    137	filter_8bit(x7, t0, t1, t7, t6); \
    138	filter_8bit(x3, t0, t1, t7, t6); \
    139	filter_8bit(x6, t0, t1, t7, t6); \
    140	\
    141	/* postfilter sbox 3 */ \
    142	vbroadcasti128 .Lpost_tf_lo_s2, t4; \
    143	vbroadcasti128 .Lpost_tf_hi_s2, t5; \
    144	filter_8bit(x2, t2, t3, t7, t6); \
    145	filter_8bit(x5, t2, t3, t7, t6); \
    146	\
    147	vpbroadcastq key, t0; /* higher 64-bit duplicate ignored */ \
    148	\
    149	/* postfilter sbox 2 */ \
    150	filter_8bit(x1, t4, t5, t7, t2); \
    151	filter_8bit(x4, t4, t5, t7, t2); \
    152	vpxor t7, t7, t7; \
    153	\
    154	vpsrldq $1, t0, t1; \
    155	vpsrldq $2, t0, t2; \
    156	vpshufb t7, t1, t1; \
    157	vpsrldq $3, t0, t3; \
    158	\
    159	/* P-function */ \
    160	vpxor x5, x0, x0; \
    161	vpxor x6, x1, x1; \
    162	vpxor x7, x2, x2; \
    163	vpxor x4, x3, x3; \
    164	\
    165	vpshufb t7, t2, t2; \
    166	vpsrldq $4, t0, t4; \
    167	vpshufb t7, t3, t3; \
    168	vpsrldq $5, t0, t5; \
    169	vpshufb t7, t4, t4; \
    170	\
    171	vpxor x2, x4, x4; \
    172	vpxor x3, x5, x5; \
    173	vpxor x0, x6, x6; \
    174	vpxor x1, x7, x7; \
    175	\
    176	vpsrldq $6, t0, t6; \
    177	vpshufb t7, t5, t5; \
    178	vpshufb t7, t6, t6; \
    179	\
    180	vpxor x7, x0, x0; \
    181	vpxor x4, x1, x1; \
    182	vpxor x5, x2, x2; \
    183	vpxor x6, x3, x3; \
    184	\
    185	vpxor x3, x4, x4; \
    186	vpxor x0, x5, x5; \
    187	vpxor x1, x6, x6; \
    188	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
    189	\
    190	/* Add key material and result to CD (x becomes new CD) */ \
    191	\
    192	vpxor t6, x1, x1; \
    193	vpxor 5 * 32(mem_cd), x1, x1; \
    194	\
    195	vpsrldq $7, t0, t6; \
    196	vpshufb t7, t0, t0; \
    197	vpshufb t7, t6, t7; \
    198	\
    199	vpxor t7, x0, x0; \
    200	vpxor 4 * 32(mem_cd), x0, x0; \
    201	\
    202	vpxor t5, x2, x2; \
    203	vpxor 6 * 32(mem_cd), x2, x2; \
    204	\
    205	vpxor t4, x3, x3; \
    206	vpxor 7 * 32(mem_cd), x3, x3; \
    207	\
    208	vpxor t3, x4, x4; \
    209	vpxor 0 * 32(mem_cd), x4, x4; \
    210	\
    211	vpxor t2, x5, x5; \
    212	vpxor 1 * 32(mem_cd), x5, x5; \
    213	\
    214	vpxor t1, x6, x6; \
    215	vpxor 2 * 32(mem_cd), x6, x6; \
    216	\
    217	vpxor t0, x7, x7; \
    218	vpxor 3 * 32(mem_cd), x7, x7;
    219
    220/*
    221 * Size optimization... with inlined roundsm32 binary would be over 5 times
    222 * larger and would only marginally faster.
    223 */
    224.align 8
    225SYM_FUNC_START_LOCAL(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
    226	roundsm32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    227		  %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
    228		  %rcx, (%r9));
    229	RET;
    230SYM_FUNC_END(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
    231
    232.align 8
    233SYM_FUNC_START_LOCAL(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
    234	roundsm32(%ymm4, %ymm5, %ymm6, %ymm7, %ymm0, %ymm1, %ymm2, %ymm3,
    235		  %ymm12, %ymm13, %ymm14, %ymm15, %ymm8, %ymm9, %ymm10, %ymm11,
    236		  %rax, (%r9));
    237	RET;
    238SYM_FUNC_END(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
    239
    240/*
    241 * IN/OUT:
    242 *  x0..x7: byte-sliced AB state preloaded
    243 *  mem_ab: byte-sliced AB state in memory
    244 *  mem_cb: byte-sliced CD state in memory
    245 */
    246#define two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    247		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
    248	leaq (key_table + (i) * 8)(CTX), %r9; \
    249	call roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
    250	\
    251	vmovdqu x0, 4 * 32(mem_cd); \
    252	vmovdqu x1, 5 * 32(mem_cd); \
    253	vmovdqu x2, 6 * 32(mem_cd); \
    254	vmovdqu x3, 7 * 32(mem_cd); \
    255	vmovdqu x4, 0 * 32(mem_cd); \
    256	vmovdqu x5, 1 * 32(mem_cd); \
    257	vmovdqu x6, 2 * 32(mem_cd); \
    258	vmovdqu x7, 3 * 32(mem_cd); \
    259	\
    260	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
    261	call roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
    262	\
    263	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
    264
    265#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
    266
    267#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
    268	/* Store new AB state */ \
    269	vmovdqu x4, 4 * 32(mem_ab); \
    270	vmovdqu x5, 5 * 32(mem_ab); \
    271	vmovdqu x6, 6 * 32(mem_ab); \
    272	vmovdqu x7, 7 * 32(mem_ab); \
    273	vmovdqu x0, 0 * 32(mem_ab); \
    274	vmovdqu x1, 1 * 32(mem_ab); \
    275	vmovdqu x2, 2 * 32(mem_ab); \
    276	vmovdqu x3, 3 * 32(mem_ab);
    277
    278#define enc_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    279		      y6, y7, mem_ab, mem_cd, i) \
    280	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    281		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
    282	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    283		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
    284	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    285		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
    286
    287#define dec_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    288		      y6, y7, mem_ab, mem_cd, i) \
    289	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    290		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
    291	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    292		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
    293	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    294		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
    295
    296/*
    297 * IN:
    298 *  v0..3: byte-sliced 32-bit integers
    299 * OUT:
    300 *  v0..3: (IN <<< 1)
    301 */
    302#define rol32_1_32(v0, v1, v2, v3, t0, t1, t2, zero) \
    303	vpcmpgtb v0, zero, t0; \
    304	vpaddb v0, v0, v0; \
    305	vpabsb t0, t0; \
    306	\
    307	vpcmpgtb v1, zero, t1; \
    308	vpaddb v1, v1, v1; \
    309	vpabsb t1, t1; \
    310	\
    311	vpcmpgtb v2, zero, t2; \
    312	vpaddb v2, v2, v2; \
    313	vpabsb t2, t2; \
    314	\
    315	vpor t0, v1, v1; \
    316	\
    317	vpcmpgtb v3, zero, t0; \
    318	vpaddb v3, v3, v3; \
    319	vpabsb t0, t0; \
    320	\
    321	vpor t1, v2, v2; \
    322	vpor t2, v3, v3; \
    323	vpor t0, v0, v0;
    324
    325/*
    326 * IN:
    327 *   r: byte-sliced AB state in memory
    328 *   l: byte-sliced CD state in memory
    329 * OUT:
    330 *   x0..x7: new byte-sliced CD state
    331 */
    332#define fls32(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
    333	      tt1, tt2, tt3, kll, klr, krl, krr) \
    334	/* \
    335	 * t0 = kll; \
    336	 * t0 &= ll; \
    337	 * lr ^= rol32(t0, 1); \
    338	 */ \
    339	vpbroadcastd kll, t0; /* only lowest 32-bit used */ \
    340	vpxor tt0, tt0, tt0; \
    341	vpshufb tt0, t0, t3; \
    342	vpsrldq $1, t0, t0; \
    343	vpshufb tt0, t0, t2; \
    344	vpsrldq $1, t0, t0; \
    345	vpshufb tt0, t0, t1; \
    346	vpsrldq $1, t0, t0; \
    347	vpshufb tt0, t0, t0; \
    348	\
    349	vpand l0, t0, t0; \
    350	vpand l1, t1, t1; \
    351	vpand l2, t2, t2; \
    352	vpand l3, t3, t3; \
    353	\
    354	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
    355	\
    356	vpxor l4, t0, l4; \
    357	vpbroadcastd krr, t0; /* only lowest 32-bit used */ \
    358	vmovdqu l4, 4 * 32(l); \
    359	vpxor l5, t1, l5; \
    360	vmovdqu l5, 5 * 32(l); \
    361	vpxor l6, t2, l6; \
    362	vmovdqu l6, 6 * 32(l); \
    363	vpxor l7, t3, l7; \
    364	vmovdqu l7, 7 * 32(l); \
    365	\
    366	/* \
    367	 * t2 = krr; \
    368	 * t2 |= rr; \
    369	 * rl ^= t2; \
    370	 */ \
    371	\
    372	vpshufb tt0, t0, t3; \
    373	vpsrldq $1, t0, t0; \
    374	vpshufb tt0, t0, t2; \
    375	vpsrldq $1, t0, t0; \
    376	vpshufb tt0, t0, t1; \
    377	vpsrldq $1, t0, t0; \
    378	vpshufb tt0, t0, t0; \
    379	\
    380	vpor 4 * 32(r), t0, t0; \
    381	vpor 5 * 32(r), t1, t1; \
    382	vpor 6 * 32(r), t2, t2; \
    383	vpor 7 * 32(r), t3, t3; \
    384	\
    385	vpxor 0 * 32(r), t0, t0; \
    386	vpxor 1 * 32(r), t1, t1; \
    387	vpxor 2 * 32(r), t2, t2; \
    388	vpxor 3 * 32(r), t3, t3; \
    389	vmovdqu t0, 0 * 32(r); \
    390	vpbroadcastd krl, t0; /* only lowest 32-bit used */ \
    391	vmovdqu t1, 1 * 32(r); \
    392	vmovdqu t2, 2 * 32(r); \
    393	vmovdqu t3, 3 * 32(r); \
    394	\
    395	/* \
    396	 * t2 = krl; \
    397	 * t2 &= rl; \
    398	 * rr ^= rol32(t2, 1); \
    399	 */ \
    400	vpshufb tt0, t0, t3; \
    401	vpsrldq $1, t0, t0; \
    402	vpshufb tt0, t0, t2; \
    403	vpsrldq $1, t0, t0; \
    404	vpshufb tt0, t0, t1; \
    405	vpsrldq $1, t0, t0; \
    406	vpshufb tt0, t0, t0; \
    407	\
    408	vpand 0 * 32(r), t0, t0; \
    409	vpand 1 * 32(r), t1, t1; \
    410	vpand 2 * 32(r), t2, t2; \
    411	vpand 3 * 32(r), t3, t3; \
    412	\
    413	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
    414	\
    415	vpxor 4 * 32(r), t0, t0; \
    416	vpxor 5 * 32(r), t1, t1; \
    417	vpxor 6 * 32(r), t2, t2; \
    418	vpxor 7 * 32(r), t3, t3; \
    419	vmovdqu t0, 4 * 32(r); \
    420	vpbroadcastd klr, t0; /* only lowest 32-bit used */ \
    421	vmovdqu t1, 5 * 32(r); \
    422	vmovdqu t2, 6 * 32(r); \
    423	vmovdqu t3, 7 * 32(r); \
    424	\
    425	/* \
    426	 * t0 = klr; \
    427	 * t0 |= lr; \
    428	 * ll ^= t0; \
    429	 */ \
    430	\
    431	vpshufb tt0, t0, t3; \
    432	vpsrldq $1, t0, t0; \
    433	vpshufb tt0, t0, t2; \
    434	vpsrldq $1, t0, t0; \
    435	vpshufb tt0, t0, t1; \
    436	vpsrldq $1, t0, t0; \
    437	vpshufb tt0, t0, t0; \
    438	\
    439	vpor l4, t0, t0; \
    440	vpor l5, t1, t1; \
    441	vpor l6, t2, t2; \
    442	vpor l7, t3, t3; \
    443	\
    444	vpxor l0, t0, l0; \
    445	vmovdqu l0, 0 * 32(l); \
    446	vpxor l1, t1, l1; \
    447	vmovdqu l1, 1 * 32(l); \
    448	vpxor l2, t2, l2; \
    449	vmovdqu l2, 2 * 32(l); \
    450	vpxor l3, t3, l3; \
    451	vmovdqu l3, 3 * 32(l);
    452
    453#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
    454	vpunpckhdq x1, x0, t2; \
    455	vpunpckldq x1, x0, x0; \
    456	\
    457	vpunpckldq x3, x2, t1; \
    458	vpunpckhdq x3, x2, x2; \
    459	\
    460	vpunpckhqdq t1, x0, x1; \
    461	vpunpcklqdq t1, x0, x0; \
    462	\
    463	vpunpckhqdq x2, t2, x3; \
    464	vpunpcklqdq x2, t2, x2;
    465
    466#define byteslice_16x16b_fast(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, \
    467			      a3, b3, c3, d3, st0, st1) \
    468	vmovdqu d2, st0; \
    469	vmovdqu d3, st1; \
    470	transpose_4x4(a0, a1, a2, a3, d2, d3); \
    471	transpose_4x4(b0, b1, b2, b3, d2, d3); \
    472	vmovdqu st0, d2; \
    473	vmovdqu st1, d3; \
    474	\
    475	vmovdqu a0, st0; \
    476	vmovdqu a1, st1; \
    477	transpose_4x4(c0, c1, c2, c3, a0, a1); \
    478	transpose_4x4(d0, d1, d2, d3, a0, a1); \
    479	\
    480	vbroadcasti128 .Lshufb_16x16b, a0; \
    481	vmovdqu st1, a1; \
    482	vpshufb a0, a2, a2; \
    483	vpshufb a0, a3, a3; \
    484	vpshufb a0, b0, b0; \
    485	vpshufb a0, b1, b1; \
    486	vpshufb a0, b2, b2; \
    487	vpshufb a0, b3, b3; \
    488	vpshufb a0, a1, a1; \
    489	vpshufb a0, c0, c0; \
    490	vpshufb a0, c1, c1; \
    491	vpshufb a0, c2, c2; \
    492	vpshufb a0, c3, c3; \
    493	vpshufb a0, d0, d0; \
    494	vpshufb a0, d1, d1; \
    495	vpshufb a0, d2, d2; \
    496	vpshufb a0, d3, d3; \
    497	vmovdqu d3, st1; \
    498	vmovdqu st0, d3; \
    499	vpshufb a0, d3, a0; \
    500	vmovdqu d2, st0; \
    501	\
    502	transpose_4x4(a0, b0, c0, d0, d2, d3); \
    503	transpose_4x4(a1, b1, c1, d1, d2, d3); \
    504	vmovdqu st0, d2; \
    505	vmovdqu st1, d3; \
    506	\
    507	vmovdqu b0, st0; \
    508	vmovdqu b1, st1; \
    509	transpose_4x4(a2, b2, c2, d2, b0, b1); \
    510	transpose_4x4(a3, b3, c3, d3, b0, b1); \
    511	vmovdqu st0, b0; \
    512	vmovdqu st1, b1; \
    513	/* does not adjust output bytes inside vectors */
    514
    515/* load blocks to registers and apply pre-whitening */
    516#define inpack32_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    517		     y6, y7, rio, key) \
    518	vpbroadcastq key, x0; \
    519	vpshufb .Lpack_bswap, x0, x0; \
    520	\
    521	vpxor 0 * 32(rio), x0, y7; \
    522	vpxor 1 * 32(rio), x0, y6; \
    523	vpxor 2 * 32(rio), x0, y5; \
    524	vpxor 3 * 32(rio), x0, y4; \
    525	vpxor 4 * 32(rio), x0, y3; \
    526	vpxor 5 * 32(rio), x0, y2; \
    527	vpxor 6 * 32(rio), x0, y1; \
    528	vpxor 7 * 32(rio), x0, y0; \
    529	vpxor 8 * 32(rio), x0, x7; \
    530	vpxor 9 * 32(rio), x0, x6; \
    531	vpxor 10 * 32(rio), x0, x5; \
    532	vpxor 11 * 32(rio), x0, x4; \
    533	vpxor 12 * 32(rio), x0, x3; \
    534	vpxor 13 * 32(rio), x0, x2; \
    535	vpxor 14 * 32(rio), x0, x1; \
    536	vpxor 15 * 32(rio), x0, x0;
    537
    538/* byteslice pre-whitened blocks and store to temporary memory */
    539#define inpack32_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    540		      y6, y7, mem_ab, mem_cd) \
    541	byteslice_16x16b_fast(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, \
    542			      y4, y5, y6, y7, (mem_ab), (mem_cd)); \
    543	\
    544	vmovdqu x0, 0 * 32(mem_ab); \
    545	vmovdqu x1, 1 * 32(mem_ab); \
    546	vmovdqu x2, 2 * 32(mem_ab); \
    547	vmovdqu x3, 3 * 32(mem_ab); \
    548	vmovdqu x4, 4 * 32(mem_ab); \
    549	vmovdqu x5, 5 * 32(mem_ab); \
    550	vmovdqu x6, 6 * 32(mem_ab); \
    551	vmovdqu x7, 7 * 32(mem_ab); \
    552	vmovdqu y0, 0 * 32(mem_cd); \
    553	vmovdqu y1, 1 * 32(mem_cd); \
    554	vmovdqu y2, 2 * 32(mem_cd); \
    555	vmovdqu y3, 3 * 32(mem_cd); \
    556	vmovdqu y4, 4 * 32(mem_cd); \
    557	vmovdqu y5, 5 * 32(mem_cd); \
    558	vmovdqu y6, 6 * 32(mem_cd); \
    559	vmovdqu y7, 7 * 32(mem_cd);
    560
    561/* de-byteslice, apply post-whitening and store blocks */
    562#define outunpack32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
    563		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
    564	byteslice_16x16b_fast(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, \
    565			      y3, y7, x3, x7, stack_tmp0, stack_tmp1); \
    566	\
    567	vmovdqu x0, stack_tmp0; \
    568	\
    569	vpbroadcastq key, x0; \
    570	vpshufb .Lpack_bswap, x0, x0; \
    571	\
    572	vpxor x0, y7, y7; \
    573	vpxor x0, y6, y6; \
    574	vpxor x0, y5, y5; \
    575	vpxor x0, y4, y4; \
    576	vpxor x0, y3, y3; \
    577	vpxor x0, y2, y2; \
    578	vpxor x0, y1, y1; \
    579	vpxor x0, y0, y0; \
    580	vpxor x0, x7, x7; \
    581	vpxor x0, x6, x6; \
    582	vpxor x0, x5, x5; \
    583	vpxor x0, x4, x4; \
    584	vpxor x0, x3, x3; \
    585	vpxor x0, x2, x2; \
    586	vpxor x0, x1, x1; \
    587	vpxor stack_tmp0, x0, x0;
    588
    589#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
    590		     y6, y7, rio) \
    591	vmovdqu x0, 0 * 32(rio); \
    592	vmovdqu x1, 1 * 32(rio); \
    593	vmovdqu x2, 2 * 32(rio); \
    594	vmovdqu x3, 3 * 32(rio); \
    595	vmovdqu x4, 4 * 32(rio); \
    596	vmovdqu x5, 5 * 32(rio); \
    597	vmovdqu x6, 6 * 32(rio); \
    598	vmovdqu x7, 7 * 32(rio); \
    599	vmovdqu y0, 8 * 32(rio); \
    600	vmovdqu y1, 9 * 32(rio); \
    601	vmovdqu y2, 10 * 32(rio); \
    602	vmovdqu y3, 11 * 32(rio); \
    603	vmovdqu y4, 12 * 32(rio); \
    604	vmovdqu y5, 13 * 32(rio); \
    605	vmovdqu y6, 14 * 32(rio); \
    606	vmovdqu y7, 15 * 32(rio);
    607
    608
    609.section	.rodata.cst32.shufb_16x16b, "aM", @progbits, 32
    610.align 32
    611#define SHUFB_BYTES(idx) \
    612	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
    613.Lshufb_16x16b:
    614	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
    615	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
    616
    617.section	.rodata.cst32.pack_bswap, "aM", @progbits, 32
    618.align 32
    619.Lpack_bswap:
    620	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
    621	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
    622
    623/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
    624.section	.rodata.cst16, "aM", @progbits, 16
    625.align 16
    626
    627/*
    628 * pre-SubByte transform
    629 *
    630 * pre-lookup for sbox1, sbox2, sbox3:
    631 *   swap_bitendianness(
    632 *       isom_map_camellia_to_aes(
    633 *           camellia_f(
    634 *               swap_bitendianess(in)
    635 *           )
    636 *       )
    637 *   )
    638 *
    639 * (note: '⊕ 0xc5' inside camellia_f())
    640 */
    641.Lpre_tf_lo_s1:
    642	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
    643	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
    644.Lpre_tf_hi_s1:
    645	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
    646	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
    647
    648/*
    649 * pre-SubByte transform
    650 *
    651 * pre-lookup for sbox4:
    652 *   swap_bitendianness(
    653 *       isom_map_camellia_to_aes(
    654 *           camellia_f(
    655 *               swap_bitendianess(in <<< 1)
    656 *           )
    657 *       )
    658 *   )
    659 *
    660 * (note: '⊕ 0xc5' inside camellia_f())
    661 */
    662.Lpre_tf_lo_s4:
    663	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
    664	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
    665.Lpre_tf_hi_s4:
    666	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
    667	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
    668
    669/*
    670 * post-SubByte transform
    671 *
    672 * post-lookup for sbox1, sbox4:
    673 *  swap_bitendianness(
    674 *      camellia_h(
    675 *          isom_map_aes_to_camellia(
    676 *              swap_bitendianness(
    677 *                  aes_inverse_affine_transform(in)
    678 *              )
    679 *          )
    680 *      )
    681 *  )
    682 *
    683 * (note: '⊕ 0x6e' inside camellia_h())
    684 */
    685.Lpost_tf_lo_s1:
    686	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
    687	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
    688.Lpost_tf_hi_s1:
    689	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
    690	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
    691
    692/*
    693 * post-SubByte transform
    694 *
    695 * post-lookup for sbox2:
    696 *  swap_bitendianness(
    697 *      camellia_h(
    698 *          isom_map_aes_to_camellia(
    699 *              swap_bitendianness(
    700 *                  aes_inverse_affine_transform(in)
    701 *              )
    702 *          )
    703 *      )
    704 *  ) <<< 1
    705 *
    706 * (note: '⊕ 0x6e' inside camellia_h())
    707 */
    708.Lpost_tf_lo_s2:
    709	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
    710	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
    711.Lpost_tf_hi_s2:
    712	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
    713	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
    714
    715/*
    716 * post-SubByte transform
    717 *
    718 * post-lookup for sbox3:
    719 *  swap_bitendianness(
    720 *      camellia_h(
    721 *          isom_map_aes_to_camellia(
    722 *              swap_bitendianness(
    723 *                  aes_inverse_affine_transform(in)
    724 *              )
    725 *          )
    726 *      )
    727 *  ) >>> 1
    728 *
    729 * (note: '⊕ 0x6e' inside camellia_h())
    730 */
    731.Lpost_tf_lo_s3:
    732	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
    733	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
    734.Lpost_tf_hi_s3:
    735	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
    736	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
    737
    738/* For isolating SubBytes from AESENCLAST, inverse shift row */
    739.Linv_shift_row:
    740	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
    741	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
    742
    743.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
    744.align 4
    745/* 4-bit mask */
    746.L0f0f0f0f:
    747	.long 0x0f0f0f0f
    748
    749.text
    750
    751.align 8
    752SYM_FUNC_START_LOCAL(__camellia_enc_blk32)
    753	/* input:
    754	 *	%rdi: ctx, CTX
    755	 *	%rax: temporary storage, 512 bytes
    756	 *	%ymm0..%ymm15: 32 plaintext blocks
    757	 * output:
    758	 *	%ymm0..%ymm15: 32 encrypted blocks, order swapped:
    759	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
    760	 */
    761	FRAME_BEGIN
    762
    763	leaq 8 * 32(%rax), %rcx;
    764
    765	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    766		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    767		      %ymm15, %rax, %rcx);
    768
    769	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    770		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    771		     %ymm15, %rax, %rcx, 0);
    772
    773	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    774	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    775	      %ymm15,
    776	      ((key_table + (8) * 8) + 0)(CTX),
    777	      ((key_table + (8) * 8) + 4)(CTX),
    778	      ((key_table + (8) * 8) + 8)(CTX),
    779	      ((key_table + (8) * 8) + 12)(CTX));
    780
    781	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    782		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    783		     %ymm15, %rax, %rcx, 8);
    784
    785	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    786	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    787	      %ymm15,
    788	      ((key_table + (16) * 8) + 0)(CTX),
    789	      ((key_table + (16) * 8) + 4)(CTX),
    790	      ((key_table + (16) * 8) + 8)(CTX),
    791	      ((key_table + (16) * 8) + 12)(CTX));
    792
    793	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    794		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    795		     %ymm15, %rax, %rcx, 16);
    796
    797	movl $24, %r8d;
    798	cmpl $16, key_length(CTX);
    799	jne .Lenc_max32;
    800
    801.Lenc_done:
    802	/* load CD for output */
    803	vmovdqu 0 * 32(%rcx), %ymm8;
    804	vmovdqu 1 * 32(%rcx), %ymm9;
    805	vmovdqu 2 * 32(%rcx), %ymm10;
    806	vmovdqu 3 * 32(%rcx), %ymm11;
    807	vmovdqu 4 * 32(%rcx), %ymm12;
    808	vmovdqu 5 * 32(%rcx), %ymm13;
    809	vmovdqu 6 * 32(%rcx), %ymm14;
    810	vmovdqu 7 * 32(%rcx), %ymm15;
    811
    812	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    813		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    814		    %ymm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 32(%rax));
    815
    816	FRAME_END
    817	RET;
    818
    819.align 8
    820.Lenc_max32:
    821	movl $32, %r8d;
    822
    823	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    824	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    825	      %ymm15,
    826	      ((key_table + (24) * 8) + 0)(CTX),
    827	      ((key_table + (24) * 8) + 4)(CTX),
    828	      ((key_table + (24) * 8) + 8)(CTX),
    829	      ((key_table + (24) * 8) + 12)(CTX));
    830
    831	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    832		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    833		     %ymm15, %rax, %rcx, 24);
    834
    835	jmp .Lenc_done;
    836SYM_FUNC_END(__camellia_enc_blk32)
    837
    838.align 8
    839SYM_FUNC_START_LOCAL(__camellia_dec_blk32)
    840	/* input:
    841	 *	%rdi: ctx, CTX
    842	 *	%rax: temporary storage, 512 bytes
    843	 *	%r8d: 24 for 16 byte key, 32 for larger
    844	 *	%ymm0..%ymm15: 16 encrypted blocks
    845	 * output:
    846	 *	%ymm0..%ymm15: 16 plaintext blocks, order swapped:
    847	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
    848	 */
    849	FRAME_BEGIN
    850
    851	leaq 8 * 32(%rax), %rcx;
    852
    853	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    854		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    855		      %ymm15, %rax, %rcx);
    856
    857	cmpl $32, %r8d;
    858	je .Ldec_max32;
    859
    860.Ldec_max24:
    861	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    862		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    863		     %ymm15, %rax, %rcx, 16);
    864
    865	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    866	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    867	      %ymm15,
    868	      ((key_table + (16) * 8) + 8)(CTX),
    869	      ((key_table + (16) * 8) + 12)(CTX),
    870	      ((key_table + (16) * 8) + 0)(CTX),
    871	      ((key_table + (16) * 8) + 4)(CTX));
    872
    873	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    874		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    875		     %ymm15, %rax, %rcx, 8);
    876
    877	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    878	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    879	      %ymm15,
    880	      ((key_table + (8) * 8) + 8)(CTX),
    881	      ((key_table + (8) * 8) + 12)(CTX),
    882	      ((key_table + (8) * 8) + 0)(CTX),
    883	      ((key_table + (8) * 8) + 4)(CTX));
    884
    885	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    886		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    887		     %ymm15, %rax, %rcx, 0);
    888
    889	/* load CD for output */
    890	vmovdqu 0 * 32(%rcx), %ymm8;
    891	vmovdqu 1 * 32(%rcx), %ymm9;
    892	vmovdqu 2 * 32(%rcx), %ymm10;
    893	vmovdqu 3 * 32(%rcx), %ymm11;
    894	vmovdqu 4 * 32(%rcx), %ymm12;
    895	vmovdqu 5 * 32(%rcx), %ymm13;
    896	vmovdqu 6 * 32(%rcx), %ymm14;
    897	vmovdqu 7 * 32(%rcx), %ymm15;
    898
    899	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    900		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    901		    %ymm15, (key_table)(CTX), (%rax), 1 * 32(%rax));
    902
    903	FRAME_END
    904	RET;
    905
    906.align 8
    907.Ldec_max32:
    908	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    909		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    910		     %ymm15, %rax, %rcx, 24);
    911
    912	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    913	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    914	      %ymm15,
    915	      ((key_table + (24) * 8) + 8)(CTX),
    916	      ((key_table + (24) * 8) + 12)(CTX),
    917	      ((key_table + (24) * 8) + 0)(CTX),
    918	      ((key_table + (24) * 8) + 4)(CTX));
    919
    920	jmp .Ldec_max24;
    921SYM_FUNC_END(__camellia_dec_blk32)
    922
    923SYM_FUNC_START(camellia_ecb_enc_32way)
    924	/* input:
    925	 *	%rdi: ctx, CTX
    926	 *	%rsi: dst (32 blocks)
    927	 *	%rdx: src (32 blocks)
    928	 */
    929	FRAME_BEGIN
    930
    931	vzeroupper;
    932
    933	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    934		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    935		     %ymm15, %rdx, (key_table)(CTX));
    936
    937	/* now dst can be used as temporary buffer (even in src == dst case) */
    938	movq	%rsi, %rax;
    939
    940	call __camellia_enc_blk32;
    941
    942	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
    943		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
    944		     %ymm8, %rsi);
    945
    946	vzeroupper;
    947
    948	FRAME_END
    949	RET;
    950SYM_FUNC_END(camellia_ecb_enc_32way)
    951
    952SYM_FUNC_START(camellia_ecb_dec_32way)
    953	/* input:
    954	 *	%rdi: ctx, CTX
    955	 *	%rsi: dst (32 blocks)
    956	 *	%rdx: src (32 blocks)
    957	 */
    958	FRAME_BEGIN
    959
    960	vzeroupper;
    961
    962	cmpl $16, key_length(CTX);
    963	movl $32, %r8d;
    964	movl $24, %eax;
    965	cmovel %eax, %r8d; /* max */
    966
    967	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
    968		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
    969		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
    970
    971	/* now dst can be used as temporary buffer (even in src == dst case) */
    972	movq	%rsi, %rax;
    973
    974	call __camellia_dec_blk32;
    975
    976	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
    977		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
    978		     %ymm8, %rsi);
    979
    980	vzeroupper;
    981
    982	FRAME_END
    983	RET;
    984SYM_FUNC_END(camellia_ecb_dec_32way)
    985
    986SYM_FUNC_START(camellia_cbc_dec_32way)
    987	/* input:
    988	 *	%rdi: ctx, CTX
    989	 *	%rsi: dst (32 blocks)
    990	 *	%rdx: src (32 blocks)
    991	 */
    992	FRAME_BEGIN
    993	subq $(16 * 32), %rsp;
    994
    995	vzeroupper;
    996
    997	cmpl $16, key_length(CTX);
    998	movl $32, %r8d;
    999	movl $24, %eax;
   1000	cmovel %eax, %r8d; /* max */
   1001
   1002	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
   1003		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
   1004		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
   1005
   1006	cmpq %rsi, %rdx;
   1007	je .Lcbc_dec_use_stack;
   1008
   1009	/* dst can be used as temporary storage, src is not overwritten. */
   1010	movq %rsi, %rax;
   1011	jmp .Lcbc_dec_continue;
   1012
   1013.Lcbc_dec_use_stack:
   1014	/*
   1015	 * dst still in-use (because dst == src), so use stack for temporary
   1016	 * storage.
   1017	 */
   1018	movq %rsp, %rax;
   1019
   1020.Lcbc_dec_continue:
   1021	call __camellia_dec_blk32;
   1022
   1023	vmovdqu %ymm7, (%rax);
   1024	vpxor %ymm7, %ymm7, %ymm7;
   1025	vinserti128 $1, (%rdx), %ymm7, %ymm7;
   1026	vpxor (%rax), %ymm7, %ymm7;
   1027	vpxor (0 * 32 + 16)(%rdx), %ymm6, %ymm6;
   1028	vpxor (1 * 32 + 16)(%rdx), %ymm5, %ymm5;
   1029	vpxor (2 * 32 + 16)(%rdx), %ymm4, %ymm4;
   1030	vpxor (3 * 32 + 16)(%rdx), %ymm3, %ymm3;
   1031	vpxor (4 * 32 + 16)(%rdx), %ymm2, %ymm2;
   1032	vpxor (5 * 32 + 16)(%rdx), %ymm1, %ymm1;
   1033	vpxor (6 * 32 + 16)(%rdx), %ymm0, %ymm0;
   1034	vpxor (7 * 32 + 16)(%rdx), %ymm15, %ymm15;
   1035	vpxor (8 * 32 + 16)(%rdx), %ymm14, %ymm14;
   1036	vpxor (9 * 32 + 16)(%rdx), %ymm13, %ymm13;
   1037	vpxor (10 * 32 + 16)(%rdx), %ymm12, %ymm12;
   1038	vpxor (11 * 32 + 16)(%rdx), %ymm11, %ymm11;
   1039	vpxor (12 * 32 + 16)(%rdx), %ymm10, %ymm10;
   1040	vpxor (13 * 32 + 16)(%rdx), %ymm9, %ymm9;
   1041	vpxor (14 * 32 + 16)(%rdx), %ymm8, %ymm8;
   1042	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
   1043		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
   1044		     %ymm8, %rsi);
   1045
   1046	vzeroupper;
   1047
   1048	addq $(16 * 32), %rsp;
   1049	FRAME_END
   1050	RET;
   1051SYM_FUNC_END(camellia_cbc_dec_32way)