1/*
2 * x86_64/AVX/AES-NI assembler implementation of Camellia
3 *
4 * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 */
12
13/*
14 * Version licensed under 2-clause BSD License is available at:
15 *	http://koti.mbnet.fi/axh/crypto/camellia-BSD-1.2.0-aesni1.tar.xz
16 */
17
18#include <linux/linkage.h>
19#include <asm/frame.h>
20#include <asm/nospec-branch.h>
21
22#define CAMELLIA_TABLE_BYTE_LEN 272
23
24/* struct camellia_ctx: */
25#define key_table 0
26#define key_length CAMELLIA_TABLE_BYTE_LEN
27
28/* register macros */
29#define CTX %rdi
30
31/**********************************************************************
32  16-way camellia
33 **********************************************************************/
34#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
35	vpand x, mask4bit, tmp0; \
36	vpandn x, mask4bit, x; \
37	vpsrld $4, x, x; \
38	\
39	vpshufb tmp0, lo_t, tmp0; \
40	vpshufb x, hi_t, x; \
41	vpxor tmp0, x, x;
42
43/*
44 * IN:
45 *   x0..x7: byte-sliced AB state
46 *   mem_cd: register pointer storing CD state
47 *   key: index for key material
48 * OUT:
49 *   x0..x7: new byte-sliced CD state
50 */
51#define roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
52		  t7, mem_cd, key) \
53	/* \
54	 * S-function with AES subbytes \
55	 */ \
56	vmovdqa .Linv_shift_row, t4; \
57	vbroadcastss .L0f0f0f0f, t7; \
58	vmovdqa .Lpre_tf_lo_s1, t0; \
59	vmovdqa .Lpre_tf_hi_s1, t1; \
60	\
61	/* AES inverse shift rows */ \
62	vpshufb t4, x0, x0; \
63	vpshufb t4, x7, x7; \
64	vpshufb t4, x1, x1; \
65	vpshufb t4, x4, x4; \
66	vpshufb t4, x2, x2; \
67	vpshufb t4, x5, x5; \
68	vpshufb t4, x3, x3; \
69	vpshufb t4, x6, x6; \
70	\
71	/* prefilter sboxes 1, 2 and 3 */ \
72	vmovdqa .Lpre_tf_lo_s4, t2; \
73	vmovdqa .Lpre_tf_hi_s4, t3; \
74	filter_8bit(x0, t0, t1, t7, t6); \
75	filter_8bit(x7, t0, t1, t7, t6); \
76	filter_8bit(x1, t0, t1, t7, t6); \
77	filter_8bit(x4, t0, t1, t7, t6); \
78	filter_8bit(x2, t0, t1, t7, t6); \
79	filter_8bit(x5, t0, t1, t7, t6); \
80	\
81	/* prefilter sbox 4 */ \
82	vpxor t4, t4, t4; \
83	filter_8bit(x3, t2, t3, t7, t6); \
84	filter_8bit(x6, t2, t3, t7, t6); \
85	\
86	/* AES subbytes + AES shift rows */ \
87	vmovdqa .Lpost_tf_lo_s1, t0; \
88	vmovdqa .Lpost_tf_hi_s1, t1; \
89	vaesenclast t4, x0, x0; \
90	vaesenclast t4, x7, x7; \
91	vaesenclast t4, x1, x1; \
92	vaesenclast t4, x4, x4; \
93	vaesenclast t4, x2, x2; \
94	vaesenclast t4, x5, x5; \
95	vaesenclast t4, x3, x3; \
96	vaesenclast t4, x6, x6; \
97	\
98	/* postfilter sboxes 1 and 4 */ \
99	vmovdqa .Lpost_tf_lo_s3, t2; \
100	vmovdqa .Lpost_tf_hi_s3, t3; \
101	filter_8bit(x0, t0, t1, t7, t6); \
102	filter_8bit(x7, t0, t1, t7, t6); \
103	filter_8bit(x3, t0, t1, t7, t6); \
104	filter_8bit(x6, t0, t1, t7, t6); \
105	\
106	/* postfilter sbox 3 */ \
107	vmovdqa .Lpost_tf_lo_s2, t4; \
108	vmovdqa .Lpost_tf_hi_s2, t5; \
109	filter_8bit(x2, t2, t3, t7, t6); \
110	filter_8bit(x5, t2, t3, t7, t6); \
111	\
112	vpxor t6, t6, t6; \
113	vmovq key, t0; \
114	\
115	/* postfilter sbox 2 */ \
116	filter_8bit(x1, t4, t5, t7, t2); \
117	filter_8bit(x4, t4, t5, t7, t2); \
118	\
119	vpsrldq $5, t0, t5; \
120	vpsrldq $1, t0, t1; \
121	vpsrldq $2, t0, t2; \
122	vpsrldq $3, t0, t3; \
123	vpsrldq $4, t0, t4; \
124	vpshufb t6, t0, t0; \
125	vpshufb t6, t1, t1; \
126	vpshufb t6, t2, t2; \
127	vpshufb t6, t3, t3; \
128	vpshufb t6, t4, t4; \
129	vpsrldq $2, t5, t7; \
130	vpshufb t6, t7, t7; \
131	\
132	/* \
133	 * P-function \
134	 */ \
135	vpxor x5, x0, x0; \
136	vpxor x6, x1, x1; \
137	vpxor x7, x2, x2; \
138	vpxor x4, x3, x3; \
139	\
140	vpxor x2, x4, x4; \
141	vpxor x3, x5, x5; \
142	vpxor x0, x6, x6; \
143	vpxor x1, x7, x7; \
144	\
145	vpxor x7, x0, x0; \
146	vpxor x4, x1, x1; \
147	vpxor x5, x2, x2; \
148	vpxor x6, x3, x3; \
149	\
150	vpxor x3, x4, x4; \
151	vpxor x0, x5, x5; \
152	vpxor x1, x6, x6; \
153	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
154	\
155	/* \
156	 * Add key material and result to CD (x becomes new CD) \
157	 */ \
158	\
159	vpxor t3, x4, x4; \
160	vpxor 0 * 16(mem_cd), x4, x4; \
161	\
162	vpxor t2, x5, x5; \
163	vpxor 1 * 16(mem_cd), x5, x5; \
164	\
165	vpsrldq $1, t5, t3; \
166	vpshufb t6, t5, t5; \
167	vpshufb t6, t3, t6; \
168	\
169	vpxor t1, x6, x6; \
170	vpxor 2 * 16(mem_cd), x6, x6; \
171	\
172	vpxor t0, x7, x7; \
173	vpxor 3 * 16(mem_cd), x7, x7; \
174	\
175	vpxor t7, x0, x0; \
176	vpxor 4 * 16(mem_cd), x0, x0; \
177	\
178	vpxor t6, x1, x1; \
179	vpxor 5 * 16(mem_cd), x1, x1; \
180	\
181	vpxor t5, x2, x2; \
182	vpxor 6 * 16(mem_cd), x2, x2; \
183	\
184	vpxor t4, x3, x3; \
185	vpxor 7 * 16(mem_cd), x3, x3;
186
187/*
188 * Size optimization... with inlined roundsm16, binary would be over 5 times
189 * larger and would only be 0.5% faster (on sandy-bridge).
190 */
191.align 8
192roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd:
193	roundsm16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
194		  %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, %xmm15,
195		  %rcx, (%r9));
196	ret;
197ENDPROC(roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
198
199.align 8
200roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab:
201	roundsm16(%xmm4, %xmm5, %xmm6, %xmm7, %xmm0, %xmm1, %xmm2, %xmm3,
202		  %xmm12, %xmm13, %xmm14, %xmm15, %xmm8, %xmm9, %xmm10, %xmm11,
203		  %rax, (%r9));
204	ret;
205ENDPROC(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
206
207/*
208 * IN/OUT:
209 *  x0..x7: byte-sliced AB state preloaded
210 *  mem_ab: byte-sliced AB state in memory
211 *  mem_cb: byte-sliced CD state in memory
212 */
213#define two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
214		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
215	leaq (key_table + (i) * 8)(CTX), %r9; \
216	call roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
217	\
218	vmovdqu x4, 0 * 16(mem_cd); \
219	vmovdqu x5, 1 * 16(mem_cd); \
220	vmovdqu x6, 2 * 16(mem_cd); \
221	vmovdqu x7, 3 * 16(mem_cd); \
222	vmovdqu x0, 4 * 16(mem_cd); \
223	vmovdqu x1, 5 * 16(mem_cd); \
224	vmovdqu x2, 6 * 16(mem_cd); \
225	vmovdqu x3, 7 * 16(mem_cd); \
226	\
227	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
228	call roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
229	\
230	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
231
232#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
233
234#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
235	/* Store new AB state */ \
236	vmovdqu x0, 0 * 16(mem_ab); \
237	vmovdqu x1, 1 * 16(mem_ab); \
238	vmovdqu x2, 2 * 16(mem_ab); \
239	vmovdqu x3, 3 * 16(mem_ab); \
240	vmovdqu x4, 4 * 16(mem_ab); \
241	vmovdqu x5, 5 * 16(mem_ab); \
242	vmovdqu x6, 6 * 16(mem_ab); \
243	vmovdqu x7, 7 * 16(mem_ab);
244
245#define enc_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
246		      y6, y7, mem_ab, mem_cd, i) \
247	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
248		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
249	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
250		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
251	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
252		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
253
254#define dec_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
255		      y6, y7, mem_ab, mem_cd, i) \
256	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
257		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
258	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
259		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
260	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
261		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
262
263/*
264 * IN:
265 *  v0..3: byte-sliced 32-bit integers
266 * OUT:
267 *  v0..3: (IN <<< 1)
268 */
269#define rol32_1_16(v0, v1, v2, v3, t0, t1, t2, zero) \
270	vpcmpgtb v0, zero, t0; \
271	vpaddb v0, v0, v0; \
272	vpabsb t0, t0; \
273	\
274	vpcmpgtb v1, zero, t1; \
275	vpaddb v1, v1, v1; \
276	vpabsb t1, t1; \
277	\
278	vpcmpgtb v2, zero, t2; \
279	vpaddb v2, v2, v2; \
280	vpabsb t2, t2; \
281	\
282	vpor t0, v1, v1; \
283	\
284	vpcmpgtb v3, zero, t0; \
285	vpaddb v3, v3, v3; \
286	vpabsb t0, t0; \
287	\
288	vpor t1, v2, v2; \
289	vpor t2, v3, v3; \
290	vpor t0, v0, v0;
291
292/*
293 * IN:
294 *   r: byte-sliced AB state in memory
295 *   l: byte-sliced CD state in memory
296 * OUT:
297 *   x0..x7: new byte-sliced CD state
298 */
299#define fls16(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
300	      tt1, tt2, tt3, kll, klr, krl, krr) \
301	/* \
302	 * t0 = kll; \
303	 * t0 &= ll; \
304	 * lr ^= rol32(t0, 1); \
305	 */ \
306	vpxor tt0, tt0, tt0; \
307	vmovd kll, t0; \
308	vpshufb tt0, t0, t3; \
309	vpsrldq $1, t0, t0; \
310	vpshufb tt0, t0, t2; \
311	vpsrldq $1, t0, t0; \
312	vpshufb tt0, t0, t1; \
313	vpsrldq $1, t0, t0; \
314	vpshufb tt0, t0, t0; \
315	\
316	vpand l0, t0, t0; \
317	vpand l1, t1, t1; \
318	vpand l2, t2, t2; \
319	vpand l3, t3, t3; \
320	\
321	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
322	\
323	vpxor l4, t0, l4; \
324	vmovdqu l4, 4 * 16(l); \
325	vpxor l5, t1, l5; \
326	vmovdqu l5, 5 * 16(l); \
327	vpxor l6, t2, l6; \
328	vmovdqu l6, 6 * 16(l); \
329	vpxor l7, t3, l7; \
330	vmovdqu l7, 7 * 16(l); \
331	\
332	/* \
333	 * t2 = krr; \
334	 * t2 |= rr; \
335	 * rl ^= t2; \
336	 */ \
337	\
338	vmovd krr, t0; \
339	vpshufb tt0, t0, t3; \
340	vpsrldq $1, t0, t0; \
341	vpshufb tt0, t0, t2; \
342	vpsrldq $1, t0, t0; \
343	vpshufb tt0, t0, t1; \
344	vpsrldq $1, t0, t0; \
345	vpshufb tt0, t0, t0; \
346	\
347	vpor 4 * 16(r), t0, t0; \
348	vpor 5 * 16(r), t1, t1; \
349	vpor 6 * 16(r), t2, t2; \
350	vpor 7 * 16(r), t3, t3; \
351	\
352	vpxor 0 * 16(r), t0, t0; \
353	vpxor 1 * 16(r), t1, t1; \
354	vpxor 2 * 16(r), t2, t2; \
355	vpxor 3 * 16(r), t3, t3; \
356	vmovdqu t0, 0 * 16(r); \
357	vmovdqu t1, 1 * 16(r); \
358	vmovdqu t2, 2 * 16(r); \
359	vmovdqu t3, 3 * 16(r); \
360	\
361	/* \
362	 * t2 = krl; \
363	 * t2 &= rl; \
364	 * rr ^= rol32(t2, 1); \
365	 */ \
366	vmovd krl, t0; \
367	vpshufb tt0, t0, t3; \
368	vpsrldq $1, t0, t0; \
369	vpshufb tt0, t0, t2; \
370	vpsrldq $1, t0, t0; \
371	vpshufb tt0, t0, t1; \
372	vpsrldq $1, t0, t0; \
373	vpshufb tt0, t0, t0; \
374	\
375	vpand 0 * 16(r), t0, t0; \
376	vpand 1 * 16(r), t1, t1; \
377	vpand 2 * 16(r), t2, t2; \
378	vpand 3 * 16(r), t3, t3; \
379	\
380	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
381	\
382	vpxor 4 * 16(r), t0, t0; \
383	vpxor 5 * 16(r), t1, t1; \
384	vpxor 6 * 16(r), t2, t2; \
385	vpxor 7 * 16(r), t3, t3; \
386	vmovdqu t0, 4 * 16(r); \
387	vmovdqu t1, 5 * 16(r); \
388	vmovdqu t2, 6 * 16(r); \
389	vmovdqu t3, 7 * 16(r); \
390	\
391	/* \
392	 * t0 = klr; \
393	 * t0 |= lr; \
394	 * ll ^= t0; \
395	 */ \
396	\
397	vmovd klr, t0; \
398	vpshufb tt0, t0, t3; \
399	vpsrldq $1, t0, t0; \
400	vpshufb tt0, t0, t2; \
401	vpsrldq $1, t0, t0; \
402	vpshufb tt0, t0, t1; \
403	vpsrldq $1, t0, t0; \
404	vpshufb tt0, t0, t0; \
405	\
406	vpor l4, t0, t0; \
407	vpor l5, t1, t1; \
408	vpor l6, t2, t2; \
409	vpor l7, t3, t3; \
410	\
411	vpxor l0, t0, l0; \
412	vmovdqu l0, 0 * 16(l); \
413	vpxor l1, t1, l1; \
414	vmovdqu l1, 1 * 16(l); \
415	vpxor l2, t2, l2; \
416	vmovdqu l2, 2 * 16(l); \
417	vpxor l3, t3, l3; \
418	vmovdqu l3, 3 * 16(l);
419
420#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
421	vpunpckhdq x1, x0, t2; \
422	vpunpckldq x1, x0, x0; \
423	\
424	vpunpckldq x3, x2, t1; \
425	vpunpckhdq x3, x2, x2; \
426	\
427	vpunpckhqdq t1, x0, x1; \
428	vpunpcklqdq t1, x0, x0; \
429	\
430	vpunpckhqdq x2, t2, x3; \
431	vpunpcklqdq x2, t2, x2;
432
433#define byteslice_16x16b(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, a3, \
434			 b3, c3, d3, st0, st1) \
435	vmovdqu d2, st0; \
436	vmovdqu d3, st1; \
437	transpose_4x4(a0, a1, a2, a3, d2, d3); \
438	transpose_4x4(b0, b1, b2, b3, d2, d3); \
439	vmovdqu st0, d2; \
440	vmovdqu st1, d3; \
441	\
442	vmovdqu a0, st0; \
443	vmovdqu a1, st1; \
444	transpose_4x4(c0, c1, c2, c3, a0, a1); \
445	transpose_4x4(d0, d1, d2, d3, a0, a1); \
446	\
447	vmovdqu .Lshufb_16x16b, a0; \
448	vmovdqu st1, a1; \
449	vpshufb a0, a2, a2; \
450	vpshufb a0, a3, a3; \
451	vpshufb a0, b0, b0; \
452	vpshufb a0, b1, b1; \
453	vpshufb a0, b2, b2; \
454	vpshufb a0, b3, b3; \
455	vpshufb a0, a1, a1; \
456	vpshufb a0, c0, c0; \
457	vpshufb a0, c1, c1; \
458	vpshufb a0, c2, c2; \
459	vpshufb a0, c3, c3; \
460	vpshufb a0, d0, d0; \
461	vpshufb a0, d1, d1; \
462	vpshufb a0, d2, d2; \
463	vpshufb a0, d3, d3; \
464	vmovdqu d3, st1; \
465	vmovdqu st0, d3; \
466	vpshufb a0, d3, a0; \
467	vmovdqu d2, st0; \
468	\
469	transpose_4x4(a0, b0, c0, d0, d2, d3); \
470	transpose_4x4(a1, b1, c1, d1, d2, d3); \
471	vmovdqu st0, d2; \
472	vmovdqu st1, d3; \
473	\
474	vmovdqu b0, st0; \
475	vmovdqu b1, st1; \
476	transpose_4x4(a2, b2, c2, d2, b0, b1); \
477	transpose_4x4(a3, b3, c3, d3, b0, b1); \
478	vmovdqu st0, b0; \
479	vmovdqu st1, b1; \
480	/* does not adjust output bytes inside vectors */
481
482/* load blocks to registers and apply pre-whitening */
483#define inpack16_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
484		     y6, y7, rio, key) \
485	vmovq key, x0; \
486	vpshufb .Lpack_bswap, x0, x0; \
487	\
488	vpxor 0 * 16(rio), x0, y7; \
489	vpxor 1 * 16(rio), x0, y6; \
490	vpxor 2 * 16(rio), x0, y5; \
491	vpxor 3 * 16(rio), x0, y4; \
492	vpxor 4 * 16(rio), x0, y3; \
493	vpxor 5 * 16(rio), x0, y2; \
494	vpxor 6 * 16(rio), x0, y1; \
495	vpxor 7 * 16(rio), x0, y0; \
496	vpxor 8 * 16(rio), x0, x7; \
497	vpxor 9 * 16(rio), x0, x6; \
498	vpxor 10 * 16(rio), x0, x5; \
499	vpxor 11 * 16(rio), x0, x4; \
500	vpxor 12 * 16(rio), x0, x3; \
501	vpxor 13 * 16(rio), x0, x2; \
502	vpxor 14 * 16(rio), x0, x1; \
503	vpxor 15 * 16(rio), x0, x0;
504
505/* byteslice pre-whitened blocks and store to temporary memory */
506#define inpack16_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
507		      y6, y7, mem_ab, mem_cd) \
508	byteslice_16x16b(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
509			 y5, y6, y7, (mem_ab), (mem_cd)); \
510	\
511	vmovdqu x0, 0 * 16(mem_ab); \
512	vmovdqu x1, 1 * 16(mem_ab); \
513	vmovdqu x2, 2 * 16(mem_ab); \
514	vmovdqu x3, 3 * 16(mem_ab); \
515	vmovdqu x4, 4 * 16(mem_ab); \
516	vmovdqu x5, 5 * 16(mem_ab); \
517	vmovdqu x6, 6 * 16(mem_ab); \
518	vmovdqu x7, 7 * 16(mem_ab); \
519	vmovdqu y0, 0 * 16(mem_cd); \
520	vmovdqu y1, 1 * 16(mem_cd); \
521	vmovdqu y2, 2 * 16(mem_cd); \
522	vmovdqu y3, 3 * 16(mem_cd); \
523	vmovdqu y4, 4 * 16(mem_cd); \
524	vmovdqu y5, 5 * 16(mem_cd); \
525	vmovdqu y6, 6 * 16(mem_cd); \
526	vmovdqu y7, 7 * 16(mem_cd);
527
528/* de-byteslice, apply post-whitening and store blocks */
529#define outunpack16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
530		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
531	byteslice_16x16b(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, y3, \
532			 y7, x3, x7, stack_tmp0, stack_tmp1); \
533	\
534	vmovdqu x0, stack_tmp0; \
535	\
536	vmovq key, x0; \
537	vpshufb .Lpack_bswap, x0, x0; \
538	\
539	vpxor x0, y7, y7; \
540	vpxor x0, y6, y6; \
541	vpxor x0, y5, y5; \
542	vpxor x0, y4, y4; \
543	vpxor x0, y3, y3; \
544	vpxor x0, y2, y2; \
545	vpxor x0, y1, y1; \
546	vpxor x0, y0, y0; \
547	vpxor x0, x7, x7; \
548	vpxor x0, x6, x6; \
549	vpxor x0, x5, x5; \
550	vpxor x0, x4, x4; \
551	vpxor x0, x3, x3; \
552	vpxor x0, x2, x2; \
553	vpxor x0, x1, x1; \
554	vpxor stack_tmp0, x0, x0;
555
556#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
557		     y6, y7, rio) \
558	vmovdqu x0, 0 * 16(rio); \
559	vmovdqu x1, 1 * 16(rio); \
560	vmovdqu x2, 2 * 16(rio); \
561	vmovdqu x3, 3 * 16(rio); \
562	vmovdqu x4, 4 * 16(rio); \
563	vmovdqu x5, 5 * 16(rio); \
564	vmovdqu x6, 6 * 16(rio); \
565	vmovdqu x7, 7 * 16(rio); \
566	vmovdqu y0, 8 * 16(rio); \
567	vmovdqu y1, 9 * 16(rio); \
568	vmovdqu y2, 10 * 16(rio); \
569	vmovdqu y3, 11 * 16(rio); \
570	vmovdqu y4, 12 * 16(rio); \
571	vmovdqu y5, 13 * 16(rio); \
572	vmovdqu y6, 14 * 16(rio); \
573	vmovdqu y7, 15 * 16(rio);
574
575
576/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
577.section	.rodata.cst16, "aM", @progbits, 16
578.align 16
579
580#define SHUFB_BYTES(idx) \
581	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
582
583.Lshufb_16x16b:
584	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3);
585
586.Lpack_bswap:
587	.long 0x00010203
588	.long 0x04050607
589	.long 0x80808080
590	.long 0x80808080
591
592/* For CTR-mode IV byteswap */
593.Lbswap128_mask:
594	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
595
596/* For XTS mode IV generation */
597.Lxts_gf128mul_and_shl1_mask:
598	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
599
600/*
601 * pre-SubByte transform
602 *
603 * pre-lookup for sbox1, sbox2, sbox3:
604 *   swap_bitendianness(
605 *       isom_map_camellia_to_aes(
606 *           camellia_f(
607 *               swap_bitendianess(in)
608 *           )
609 *       )
610 *   )
611 *
612 * (note: '⊕ 0xc5' inside camellia_f())
613 */
614.Lpre_tf_lo_s1:
615	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
616	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
617.Lpre_tf_hi_s1:
618	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
619	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
620
621/*
622 * pre-SubByte transform
623 *
624 * pre-lookup for sbox4:
625 *   swap_bitendianness(
626 *       isom_map_camellia_to_aes(
627 *           camellia_f(
628 *               swap_bitendianess(in <<< 1)
629 *           )
630 *       )
631 *   )
632 *
633 * (note: '⊕ 0xc5' inside camellia_f())
634 */
635.Lpre_tf_lo_s4:
636	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
637	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
638.Lpre_tf_hi_s4:
639	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
640	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
641
642/*
643 * post-SubByte transform
644 *
645 * post-lookup for sbox1, sbox4:
646 *  swap_bitendianness(
647 *      camellia_h(
648 *          isom_map_aes_to_camellia(
649 *              swap_bitendianness(
650 *                  aes_inverse_affine_transform(in)
651 *              )
652 *          )
653 *      )
654 *  )
655 *
656 * (note: '⊕ 0x6e' inside camellia_h())
657 */
658.Lpost_tf_lo_s1:
659	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
660	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
661.Lpost_tf_hi_s1:
662	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
663	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
664
665/*
666 * post-SubByte transform
667 *
668 * post-lookup for sbox2:
669 *  swap_bitendianness(
670 *      camellia_h(
671 *          isom_map_aes_to_camellia(
672 *              swap_bitendianness(
673 *                  aes_inverse_affine_transform(in)
674 *              )
675 *          )
676 *      )
677 *  ) <<< 1
678 *
679 * (note: '⊕ 0x6e' inside camellia_h())
680 */
681.Lpost_tf_lo_s2:
682	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
683	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
684.Lpost_tf_hi_s2:
685	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
686	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
687
688/*
689 * post-SubByte transform
690 *
691 * post-lookup for sbox3:
692 *  swap_bitendianness(
693 *      camellia_h(
694 *          isom_map_aes_to_camellia(
695 *              swap_bitendianness(
696 *                  aes_inverse_affine_transform(in)
697 *              )
698 *          )
699 *      )
700 *  ) >>> 1
701 *
702 * (note: '⊕ 0x6e' inside camellia_h())
703 */
704.Lpost_tf_lo_s3:
705	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
706	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
707.Lpost_tf_hi_s3:
708	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
709	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
710
711/* For isolating SubBytes from AESENCLAST, inverse shift row */
712.Linv_shift_row:
713	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
714	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
715
716/* 4-bit mask */
717.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
718.align 4
719.L0f0f0f0f:
720	.long 0x0f0f0f0f
721
722.text
723
724.align 8
725__camellia_enc_blk16:
726	/* input:
727	 *	%rdi: ctx, CTX
728	 *	%rax: temporary storage, 256 bytes
729	 *	%xmm0..%xmm15: 16 plaintext blocks
730	 * output:
731	 *	%xmm0..%xmm15: 16 encrypted blocks, order swapped:
732	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
733	 */
734	FRAME_BEGIN
735
736	leaq 8 * 16(%rax), %rcx;
737
738	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
739		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
740		      %xmm15, %rax, %rcx);
741
742	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
743		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
744		     %xmm15, %rax, %rcx, 0);
745
746	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
747	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
748	      %xmm15,
749	      ((key_table + (8) * 8) + 0)(CTX),
750	      ((key_table + (8) * 8) + 4)(CTX),
751	      ((key_table + (8) * 8) + 8)(CTX),
752	      ((key_table + (8) * 8) + 12)(CTX));
753
754	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
755		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
756		     %xmm15, %rax, %rcx, 8);
757
758	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
759	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
760	      %xmm15,
761	      ((key_table + (16) * 8) + 0)(CTX),
762	      ((key_table + (16) * 8) + 4)(CTX),
763	      ((key_table + (16) * 8) + 8)(CTX),
764	      ((key_table + (16) * 8) + 12)(CTX));
765
766	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
767		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
768		     %xmm15, %rax, %rcx, 16);
769
770	movl $24, %r8d;
771	cmpl $16, key_length(CTX);
772	jne .Lenc_max32;
773
774.Lenc_done:
775	/* load CD for output */
776	vmovdqu 0 * 16(%rcx), %xmm8;
777	vmovdqu 1 * 16(%rcx), %xmm9;
778	vmovdqu 2 * 16(%rcx), %xmm10;
779	vmovdqu 3 * 16(%rcx), %xmm11;
780	vmovdqu 4 * 16(%rcx), %xmm12;
781	vmovdqu 5 * 16(%rcx), %xmm13;
782	vmovdqu 6 * 16(%rcx), %xmm14;
783	vmovdqu 7 * 16(%rcx), %xmm15;
784
785	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
786		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
787		    %xmm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 16(%rax));
788
789	FRAME_END
790	ret;
791
792.align 8
793.Lenc_max32:
794	movl $32, %r8d;
795
796	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
797	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
798	      %xmm15,
799	      ((key_table + (24) * 8) + 0)(CTX),
800	      ((key_table + (24) * 8) + 4)(CTX),
801	      ((key_table + (24) * 8) + 8)(CTX),
802	      ((key_table + (24) * 8) + 12)(CTX));
803
804	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
805		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
806		     %xmm15, %rax, %rcx, 24);
807
808	jmp .Lenc_done;
809ENDPROC(__camellia_enc_blk16)
810
811.align 8
812__camellia_dec_blk16:
813	/* input:
814	 *	%rdi: ctx, CTX
815	 *	%rax: temporary storage, 256 bytes
816	 *	%r8d: 24 for 16 byte key, 32 for larger
817	 *	%xmm0..%xmm15: 16 encrypted blocks
818	 * output:
819	 *	%xmm0..%xmm15: 16 plaintext blocks, order swapped:
820	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
821	 */
822	FRAME_BEGIN
823
824	leaq 8 * 16(%rax), %rcx;
825
826	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
827		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
828		      %xmm15, %rax, %rcx);
829
830	cmpl $32, %r8d;
831	je .Ldec_max32;
832
833.Ldec_max24:
834	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
835		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
836		     %xmm15, %rax, %rcx, 16);
837
838	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
839	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
840	      %xmm15,
841	      ((key_table + (16) * 8) + 8)(CTX),
842	      ((key_table + (16) * 8) + 12)(CTX),
843	      ((key_table + (16) * 8) + 0)(CTX),
844	      ((key_table + (16) * 8) + 4)(CTX));
845
846	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
847		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
848		     %xmm15, %rax, %rcx, 8);
849
850	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
851	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
852	      %xmm15,
853	      ((key_table + (8) * 8) + 8)(CTX),
854	      ((key_table + (8) * 8) + 12)(CTX),
855	      ((key_table + (8) * 8) + 0)(CTX),
856	      ((key_table + (8) * 8) + 4)(CTX));
857
858	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
859		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
860		     %xmm15, %rax, %rcx, 0);
861
862	/* load CD for output */
863	vmovdqu 0 * 16(%rcx), %xmm8;
864	vmovdqu 1 * 16(%rcx), %xmm9;
865	vmovdqu 2 * 16(%rcx), %xmm10;
866	vmovdqu 3 * 16(%rcx), %xmm11;
867	vmovdqu 4 * 16(%rcx), %xmm12;
868	vmovdqu 5 * 16(%rcx), %xmm13;
869	vmovdqu 6 * 16(%rcx), %xmm14;
870	vmovdqu 7 * 16(%rcx), %xmm15;
871
872	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
873		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
874		    %xmm15, (key_table)(CTX), (%rax), 1 * 16(%rax));
875
876	FRAME_END
877	ret;
878
879.align 8
880.Ldec_max32:
881	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
882		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
883		     %xmm15, %rax, %rcx, 24);
884
885	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
886	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
887	      %xmm15,
888	      ((key_table + (24) * 8) + 8)(CTX),
889	      ((key_table + (24) * 8) + 12)(CTX),
890	      ((key_table + (24) * 8) + 0)(CTX),
891	      ((key_table + (24) * 8) + 4)(CTX));
892
893	jmp .Ldec_max24;
894ENDPROC(__camellia_dec_blk16)
895
896ENTRY(camellia_ecb_enc_16way)
897	/* input:
898	 *	%rdi: ctx, CTX
899	 *	%rsi: dst (16 blocks)
900	 *	%rdx: src (16 blocks)
901	 */
902	 FRAME_BEGIN
903
904	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
905		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
906		     %xmm15, %rdx, (key_table)(CTX));
907
908	/* now dst can be used as temporary buffer (even in src == dst case) */
909	movq	%rsi, %rax;
910
911	call __camellia_enc_blk16;
912
913	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
914		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
915		     %xmm8, %rsi);
916
917	FRAME_END
918	ret;
919ENDPROC(camellia_ecb_enc_16way)
920
921ENTRY(camellia_ecb_dec_16way)
922	/* input:
923	 *	%rdi: ctx, CTX
924	 *	%rsi: dst (16 blocks)
925	 *	%rdx: src (16 blocks)
926	 */
927	 FRAME_BEGIN
928
929	cmpl $16, key_length(CTX);
930	movl $32, %r8d;
931	movl $24, %eax;
932	cmovel %eax, %r8d; /* max */
933
934	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
935		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
936		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
937
938	/* now dst can be used as temporary buffer (even in src == dst case) */
939	movq	%rsi, %rax;
940
941	call __camellia_dec_blk16;
942
943	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
944		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
945		     %xmm8, %rsi);
946
947	FRAME_END
948	ret;
949ENDPROC(camellia_ecb_dec_16way)
950
951ENTRY(camellia_cbc_dec_16way)
952	/* input:
953	 *	%rdi: ctx, CTX
954	 *	%rsi: dst (16 blocks)
955	 *	%rdx: src (16 blocks)
956	 */
957	FRAME_BEGIN
958
959	cmpl $16, key_length(CTX);
960	movl $32, %r8d;
961	movl $24, %eax;
962	cmovel %eax, %r8d; /* max */
963
964	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
965		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
966		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
967
968	/*
969	 * dst might still be in-use (in case dst == src), so use stack for
970	 * temporary storage.
971	 */
972	subq $(16 * 16), %rsp;
973	movq %rsp, %rax;
974
975	call __camellia_dec_blk16;
976
977	addq $(16 * 16), %rsp;
978
979	vpxor (0 * 16)(%rdx), %xmm6, %xmm6;
980	vpxor (1 * 16)(%rdx), %xmm5, %xmm5;
981	vpxor (2 * 16)(%rdx), %xmm4, %xmm4;
982	vpxor (3 * 16)(%rdx), %xmm3, %xmm3;
983	vpxor (4 * 16)(%rdx), %xmm2, %xmm2;
984	vpxor (5 * 16)(%rdx), %xmm1, %xmm1;
985	vpxor (6 * 16)(%rdx), %xmm0, %xmm0;
986	vpxor (7 * 16)(%rdx), %xmm15, %xmm15;
987	vpxor (8 * 16)(%rdx), %xmm14, %xmm14;
988	vpxor (9 * 16)(%rdx), %xmm13, %xmm13;
989	vpxor (10 * 16)(%rdx), %xmm12, %xmm12;
990	vpxor (11 * 16)(%rdx), %xmm11, %xmm11;
991	vpxor (12 * 16)(%rdx), %xmm10, %xmm10;
992	vpxor (13 * 16)(%rdx), %xmm9, %xmm9;
993	vpxor (14 * 16)(%rdx), %xmm8, %xmm8;
994	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
995		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
996		     %xmm8, %rsi);
997
998	FRAME_END
999	ret;
1000ENDPROC(camellia_cbc_dec_16way)
1001
1002#define inc_le128(x, minus_one, tmp) \
1003	vpcmpeqq minus_one, x, tmp; \
1004	vpsubq minus_one, x, x; \
1005	vpslldq $8, tmp, tmp; \
1006	vpsubq tmp, x, x;
1007
1008ENTRY(camellia_ctr_16way)
1009	/* input:
1010	 *	%rdi: ctx, CTX
1011	 *	%rsi: dst (16 blocks)
1012	 *	%rdx: src (16 blocks)
1013	 *	%rcx: iv (little endian, 128bit)
1014	 */
1015	FRAME_BEGIN
1016
1017	subq $(16 * 16), %rsp;
1018	movq %rsp, %rax;
1019
1020	vmovdqa .Lbswap128_mask, %xmm14;
1021
1022	/* load IV and byteswap */
1023	vmovdqu (%rcx), %xmm0;
1024	vpshufb %xmm14, %xmm0, %xmm15;
1025	vmovdqu %xmm15, 15 * 16(%rax);
1026
1027	vpcmpeqd %xmm15, %xmm15, %xmm15;
1028	vpsrldq $8, %xmm15, %xmm15; /* low: -1, high: 0 */
1029
1030	/* construct IVs */
1031	inc_le128(%xmm0, %xmm15, %xmm13);
1032	vpshufb %xmm14, %xmm0, %xmm13;
1033	vmovdqu %xmm13, 14 * 16(%rax);
1034	inc_le128(%xmm0, %xmm15, %xmm13);
1035	vpshufb %xmm14, %xmm0, %xmm13;
1036	vmovdqu %xmm13, 13 * 16(%rax);
1037	inc_le128(%xmm0, %xmm15, %xmm13);
1038	vpshufb %xmm14, %xmm0, %xmm12;
1039	inc_le128(%xmm0, %xmm15, %xmm13);
1040	vpshufb %xmm14, %xmm0, %xmm11;
1041	inc_le128(%xmm0, %xmm15, %xmm13);
1042	vpshufb %xmm14, %xmm0, %xmm10;
1043	inc_le128(%xmm0, %xmm15, %xmm13);
1044	vpshufb %xmm14, %xmm0, %xmm9;
1045	inc_le128(%xmm0, %xmm15, %xmm13);
1046	vpshufb %xmm14, %xmm0, %xmm8;
1047	inc_le128(%xmm0, %xmm15, %xmm13);
1048	vpshufb %xmm14, %xmm0, %xmm7;
1049	inc_le128(%xmm0, %xmm15, %xmm13);
1050	vpshufb %xmm14, %xmm0, %xmm6;
1051	inc_le128(%xmm0, %xmm15, %xmm13);
1052	vpshufb %xmm14, %xmm0, %xmm5;
1053	inc_le128(%xmm0, %xmm15, %xmm13);
1054	vpshufb %xmm14, %xmm0, %xmm4;
1055	inc_le128(%xmm0, %xmm15, %xmm13);
1056	vpshufb %xmm14, %xmm0, %xmm3;
1057	inc_le128(%xmm0, %xmm15, %xmm13);
1058	vpshufb %xmm14, %xmm0, %xmm2;
1059	inc_le128(%xmm0, %xmm15, %xmm13);
1060	vpshufb %xmm14, %xmm0, %xmm1;
1061	inc_le128(%xmm0, %xmm15, %xmm13);
1062	vmovdqa %xmm0, %xmm13;
1063	vpshufb %xmm14, %xmm0, %xmm0;
1064	inc_le128(%xmm13, %xmm15, %xmm14);
1065	vmovdqu %xmm13, (%rcx);
1066
1067	/* inpack16_pre: */
1068	vmovq (key_table)(CTX), %xmm15;
1069	vpshufb .Lpack_bswap, %xmm15, %xmm15;
1070	vpxor %xmm0, %xmm15, %xmm0;
1071	vpxor %xmm1, %xmm15, %xmm1;
1072	vpxor %xmm2, %xmm15, %xmm2;
1073	vpxor %xmm3, %xmm15, %xmm3;
1074	vpxor %xmm4, %xmm15, %xmm4;
1075	vpxor %xmm5, %xmm15, %xmm5;
1076	vpxor %xmm6, %xmm15, %xmm6;
1077	vpxor %xmm7, %xmm15, %xmm7;
1078	vpxor %xmm8, %xmm15, %xmm8;
1079	vpxor %xmm9, %xmm15, %xmm9;
1080	vpxor %xmm10, %xmm15, %xmm10;
1081	vpxor %xmm11, %xmm15, %xmm11;
1082	vpxor %xmm12, %xmm15, %xmm12;
1083	vpxor 13 * 16(%rax), %xmm15, %xmm13;
1084	vpxor 14 * 16(%rax), %xmm15, %xmm14;
1085	vpxor 15 * 16(%rax), %xmm15, %xmm15;
1086
1087	call __camellia_enc_blk16;
1088
1089	addq $(16 * 16), %rsp;
1090
1091	vpxor 0 * 16(%rdx), %xmm7, %xmm7;
1092	vpxor 1 * 16(%rdx), %xmm6, %xmm6;
1093	vpxor 2 * 16(%rdx), %xmm5, %xmm5;
1094	vpxor 3 * 16(%rdx), %xmm4, %xmm4;
1095	vpxor 4 * 16(%rdx), %xmm3, %xmm3;
1096	vpxor 5 * 16(%rdx), %xmm2, %xmm2;
1097	vpxor 6 * 16(%rdx), %xmm1, %xmm1;
1098	vpxor 7 * 16(%rdx), %xmm0, %xmm0;
1099	vpxor 8 * 16(%rdx), %xmm15, %xmm15;
1100	vpxor 9 * 16(%rdx), %xmm14, %xmm14;
1101	vpxor 10 * 16(%rdx), %xmm13, %xmm13;
1102	vpxor 11 * 16(%rdx), %xmm12, %xmm12;
1103	vpxor 12 * 16(%rdx), %xmm11, %xmm11;
1104	vpxor 13 * 16(%rdx), %xmm10, %xmm10;
1105	vpxor 14 * 16(%rdx), %xmm9, %xmm9;
1106	vpxor 15 * 16(%rdx), %xmm8, %xmm8;
1107	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
1108		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
1109		     %xmm8, %rsi);
1110
1111	FRAME_END
1112	ret;
1113ENDPROC(camellia_ctr_16way)
1114
1115#define gf128mul_x_ble(iv, mask, tmp) \
1116	vpsrad $31, iv, tmp; \
1117	vpaddq iv, iv, iv; \
1118	vpshufd $0x13, tmp, tmp; \
1119	vpand mask, tmp, tmp; \
1120	vpxor tmp, iv, iv;
1121
1122.align 8
1123camellia_xts_crypt_16way:
1124	/* input:
1125	 *	%rdi: ctx, CTX
1126	 *	%rsi: dst (16 blocks)
1127	 *	%rdx: src (16 blocks)
1128	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1129	 *	%r8: index for input whitening key
1130	 *	%r9: pointer to  __camellia_enc_blk16 or __camellia_dec_blk16
1131	 */
1132	FRAME_BEGIN
1133
1134	subq $(16 * 16), %rsp;
1135	movq %rsp, %rax;
1136
1137	vmovdqa .Lxts_gf128mul_and_shl1_mask, %xmm14;
1138
1139	/* load IV */
1140	vmovdqu (%rcx), %xmm0;
1141	vpxor 0 * 16(%rdx), %xmm0, %xmm15;
1142	vmovdqu %xmm15, 15 * 16(%rax);
1143	vmovdqu %xmm0, 0 * 16(%rsi);
1144
1145	/* construct IVs */
1146	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1147	vpxor 1 * 16(%rdx), %xmm0, %xmm15;
1148	vmovdqu %xmm15, 14 * 16(%rax);
1149	vmovdqu %xmm0, 1 * 16(%rsi);
1150
1151	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1152	vpxor 2 * 16(%rdx), %xmm0, %xmm13;
1153	vmovdqu %xmm0, 2 * 16(%rsi);
1154
1155	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1156	vpxor 3 * 16(%rdx), %xmm0, %xmm12;
1157	vmovdqu %xmm0, 3 * 16(%rsi);
1158
1159	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1160	vpxor 4 * 16(%rdx), %xmm0, %xmm11;
1161	vmovdqu %xmm0, 4 * 16(%rsi);
1162
1163	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1164	vpxor 5 * 16(%rdx), %xmm0, %xmm10;
1165	vmovdqu %xmm0, 5 * 16(%rsi);
1166
1167	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1168	vpxor 6 * 16(%rdx), %xmm0, %xmm9;
1169	vmovdqu %xmm0, 6 * 16(%rsi);
1170
1171	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1172	vpxor 7 * 16(%rdx), %xmm0, %xmm8;
1173	vmovdqu %xmm0, 7 * 16(%rsi);
1174
1175	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1176	vpxor 8 * 16(%rdx), %xmm0, %xmm7;
1177	vmovdqu %xmm0, 8 * 16(%rsi);
1178
1179	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1180	vpxor 9 * 16(%rdx), %xmm0, %xmm6;
1181	vmovdqu %xmm0, 9 * 16(%rsi);
1182
1183	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1184	vpxor 10 * 16(%rdx), %xmm0, %xmm5;
1185	vmovdqu %xmm0, 10 * 16(%rsi);
1186
1187	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1188	vpxor 11 * 16(%rdx), %xmm0, %xmm4;
1189	vmovdqu %xmm0, 11 * 16(%rsi);
1190
1191	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1192	vpxor 12 * 16(%rdx), %xmm0, %xmm3;
1193	vmovdqu %xmm0, 12 * 16(%rsi);
1194
1195	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1196	vpxor 13 * 16(%rdx), %xmm0, %xmm2;
1197	vmovdqu %xmm0, 13 * 16(%rsi);
1198
1199	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1200	vpxor 14 * 16(%rdx), %xmm0, %xmm1;
1201	vmovdqu %xmm0, 14 * 16(%rsi);
1202
1203	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1204	vpxor 15 * 16(%rdx), %xmm0, %xmm15;
1205	vmovdqu %xmm15, 0 * 16(%rax);
1206	vmovdqu %xmm0, 15 * 16(%rsi);
1207
1208	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
1209	vmovdqu %xmm0, (%rcx);
1210
1211	/* inpack16_pre: */
1212	vmovq (key_table)(CTX, %r8, 8), %xmm15;
1213	vpshufb .Lpack_bswap, %xmm15, %xmm15;
1214	vpxor 0 * 16(%rax), %xmm15, %xmm0;
1215	vpxor %xmm1, %xmm15, %xmm1;
1216	vpxor %xmm2, %xmm15, %xmm2;
1217	vpxor %xmm3, %xmm15, %xmm3;
1218	vpxor %xmm4, %xmm15, %xmm4;
1219	vpxor %xmm5, %xmm15, %xmm5;
1220	vpxor %xmm6, %xmm15, %xmm6;
1221	vpxor %xmm7, %xmm15, %xmm7;
1222	vpxor %xmm8, %xmm15, %xmm8;
1223	vpxor %xmm9, %xmm15, %xmm9;
1224	vpxor %xmm10, %xmm15, %xmm10;
1225	vpxor %xmm11, %xmm15, %xmm11;
1226	vpxor %xmm12, %xmm15, %xmm12;
1227	vpxor %xmm13, %xmm15, %xmm13;
1228	vpxor 14 * 16(%rax), %xmm15, %xmm14;
1229	vpxor 15 * 16(%rax), %xmm15, %xmm15;
1230
1231	CALL_NOSPEC %r9;
1232
1233	addq $(16 * 16), %rsp;
1234
1235	vpxor 0 * 16(%rsi), %xmm7, %xmm7;
1236	vpxor 1 * 16(%rsi), %xmm6, %xmm6;
1237	vpxor 2 * 16(%rsi), %xmm5, %xmm5;
1238	vpxor 3 * 16(%rsi), %xmm4, %xmm4;
1239	vpxor 4 * 16(%rsi), %xmm3, %xmm3;
1240	vpxor 5 * 16(%rsi), %xmm2, %xmm2;
1241	vpxor 6 * 16(%rsi), %xmm1, %xmm1;
1242	vpxor 7 * 16(%rsi), %xmm0, %xmm0;
1243	vpxor 8 * 16(%rsi), %xmm15, %xmm15;
1244	vpxor 9 * 16(%rsi), %xmm14, %xmm14;
1245	vpxor 10 * 16(%rsi), %xmm13, %xmm13;
1246	vpxor 11 * 16(%rsi), %xmm12, %xmm12;
1247	vpxor 12 * 16(%rsi), %xmm11, %xmm11;
1248	vpxor 13 * 16(%rsi), %xmm10, %xmm10;
1249	vpxor 14 * 16(%rsi), %xmm9, %xmm9;
1250	vpxor 15 * 16(%rsi), %xmm8, %xmm8;
1251	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
1252		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
1253		     %xmm8, %rsi);
1254
1255	FRAME_END
1256	ret;
1257ENDPROC(camellia_xts_crypt_16way)
1258
1259ENTRY(camellia_xts_enc_16way)
1260	/* input:
1261	 *	%rdi: ctx, CTX
1262	 *	%rsi: dst (16 blocks)
1263	 *	%rdx: src (16 blocks)
1264	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1265	 */
1266	xorl %r8d, %r8d; /* input whitening key, 0 for enc */
1267
1268	leaq __camellia_enc_blk16, %r9;
1269
1270	jmp camellia_xts_crypt_16way;
1271ENDPROC(camellia_xts_enc_16way)
1272
1273ENTRY(camellia_xts_dec_16way)
1274	/* input:
1275	 *	%rdi: ctx, CTX
1276	 *	%rsi: dst (16 blocks)
1277	 *	%rdx: src (16 blocks)
1278	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
1279	 */
1280
1281	cmpl $16, key_length(CTX);
1282	movl $32, %r8d;
1283	movl $24, %eax;
1284	cmovel %eax, %r8d;  /* input whitening key, last for dec */
1285
1286	leaq __camellia_dec_blk16, %r9;
1287
1288	jmp camellia_xts_crypt_16way;
1289ENDPROC(camellia_xts_dec_16way)
1290