xref: /freebsd/sys/contrib/openzfs/module/icp/algs/skein/skein_block.c (revision 3078531de10dcae44b253a35125c949ff4235284)
1 /*
2  * Implementation of the Skein block functions.
3  * Source code author: Doug Whiting, 2008.
4  * This algorithm and source code is released to the public domain.
5  * Compile-time switches:
6  *  SKEIN_USE_ASM  -- set bits (256/512/1024) to select which
7  *                    versions use ASM code for block processing
8  *                    [default: use C for all block sizes]
9  */
10 /* Copyright 2013 Doug Whiting. This code is released to the public domain. */
11 
12 #include <sys/skein.h>
13 #include "skein_impl.h"
14 #include <sys/isa_defs.h>	/* for _ILP32 */
15 
16 #ifndef	SKEIN_USE_ASM
17 #define	SKEIN_USE_ASM	(0)	/* default is all C code (no ASM) */
18 #endif
19 
20 #ifndef	SKEIN_LOOP
21 /*
22  * The low-level checksum routines use a lot of stack space. On systems where
23  * small stacks frame are enforced (like 32-bit kernel builds), do not unroll
24  * checksum calculations to save stack space.
25  *
26  * Even with no loops unrolled, we still can exceed the 1k stack frame limit
27  * in Skein1024_Process_Block() (it hits 1272 bytes on ARM32).  We can
28  * safely ignore it though, since that the checksum functions will be called
29  * from a worker thread that won't be using much stack.  That's why we have
30  * the #pragma here to ignore the warning.
31  */
32 #if defined(_ILP32) || defined(__powerpc)	/* Assume small stack */
33 #pragma GCC diagnostic ignored "-Wframe-larger-than="
34 /*
35  * We're running on 32-bit, don't unroll loops to save stack frame space
36  *
37  * Due to the ways the calculations on SKEIN_LOOP are done in
38  * Skein_*_Process_Block(), a value of 111 disables unrolling loops
39  * in any of those functions.
40  */
41 #define	SKEIN_LOOP 111
42 #else
43 /* We're compiling with large stacks */
44 #define	SKEIN_LOOP 001		/* default: unroll 256 and 512, but not 1024 */
45 #endif
46 #endif
47 
48 /* some useful definitions for code here */
49 #define	BLK_BITS	(WCNT*64)
50 #define	KW_TWK_BASE	(0)
51 #define	KW_KEY_BASE	(3)
52 #define	ks		(kw + KW_KEY_BASE)
53 #define	ts		(kw + KW_TWK_BASE)
54 
55 /* no debugging in Illumos version */
56 #define	DebugSaveTweak(ctx)
57 
58 /* Skein_256 */
59 #if	!(SKEIN_USE_ASM & 256)
60 void
61 Skein_256_Process_Block(Skein_256_Ctxt_t *ctx, const uint8_t *blkPtr,
62     size_t blkCnt, size_t byteCntAdd)
63 {
64 	enum {
65 		WCNT = SKEIN_256_STATE_WORDS
66 	};
67 #undef  RCNT
68 #define	RCNT  (SKEIN_256_ROUNDS_TOTAL / 8)
69 
70 #ifdef	SKEIN_LOOP		/* configure how much to unroll the loop */
71 #define	SKEIN_UNROLL_256 (((SKEIN_LOOP) / 100) % 10)
72 #else
73 #define	SKEIN_UNROLL_256 (0)
74 #endif
75 
76 #if	SKEIN_UNROLL_256
77 #if	(RCNT % SKEIN_UNROLL_256)
78 #error "Invalid SKEIN_UNROLL_256"	/* sanity check on unroll count */
79 #endif
80 	size_t r;
81 	/* key schedule words : chaining vars + tweak + "rotation" */
82 	uint64_t kw[WCNT + 4 + RCNT * 2];
83 #else
84 	uint64_t kw[WCNT + 4];	/* key schedule words : chaining vars + tweak */
85 #endif
86 	/* local copy of context vars, for speed */
87 	uint64_t X0, X1, X2, X3;
88 	uint64_t w[WCNT];		/* local copy of input block */
89 #ifdef	SKEIN_DEBUG
90 	/* use for debugging (help compiler put Xn in registers) */
91 	const uint64_t *Xptr[4];
92 	Xptr[0] = &X0;
93 	Xptr[1] = &X1;
94 	Xptr[2] = &X2;
95 	Xptr[3] = &X3;
96 #endif
97 	Skein_assert(blkCnt != 0);	/* never call with blkCnt == 0! */
98 	ts[0] = ctx->h.T[0];
99 	ts[1] = ctx->h.T[1];
100 	do {
101 		/*
102 		 * this implementation only supports 2**64 input bytes
103 		 * (no carry out here)
104 		 */
105 		ts[0] += byteCntAdd;	/* update processed length */
106 
107 		/* precompute the key schedule for this block */
108 		ks[0] = ctx->X[0];
109 		ks[1] = ctx->X[1];
110 		ks[2] = ctx->X[2];
111 		ks[3] = ctx->X[3];
112 		ks[4] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^ SKEIN_KS_PARITY;
113 
114 		ts[2] = ts[0] ^ ts[1];
115 
116 		/* get input block in little-endian format */
117 		Skein_Get64_LSB_First(w, blkPtr, WCNT);
118 		DebugSaveTweak(ctx);
119 		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
120 
121 		X0 = w[0] + ks[0];	/* do the first full key injection */
122 		X1 = w[1] + ks[1] + ts[0];
123 		X2 = w[2] + ks[2] + ts[1];
124 		X3 = w[3] + ks[3];
125 
126 		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL,
127 		    Xptr);	/* show starting state values */
128 
129 		blkPtr += SKEIN_256_BLOCK_BYTES;
130 
131 		/* run the rounds */
132 
133 #define	Round256(p0, p1, p2, p3, ROT, rNum)                          \
134 	X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0; \
135 	X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2; \
136 
137 #if	SKEIN_UNROLL_256 == 0
138 #define	R256(p0, p1, p2, p3, ROT, rNum)		/* fully unrolled */	\
139 	Round256(p0, p1, p2, p3, ROT, rNum)		\
140 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
141 
142 #define	I256(R)								\
143 	X0 += ks[((R) + 1) % 5]; /* inject the key schedule value */ \
144 	X1 += ks[((R) + 2) % 5] + ts[((R) + 1) % 3];			\
145 	X2 += ks[((R) + 3) % 5] + ts[((R) + 2) % 3];			\
146 	X3 += ks[((R) + 4) % 5] + (R) + 1;			\
147 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
148 #else				/* looping version */
149 #define	R256(p0, p1, p2, p3, ROT, rNum)                             \
150 	Round256(p0, p1, p2, p3, ROT, rNum)                             \
151 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
152 
153 #define	I256(R)								\
154 	X0 += ks[r + (R) + 0];	/* inject the key schedule value */	\
155 	X1 += ks[r + (R) + 1] + ts[r + (R) + 0];			\
156 	X2 += ks[r + (R) + 2] + ts[r + (R) + 1];			\
157 	X3 += ks[r + (R) + 3] + r + (R);				\
158 	ks[r + (R) + 4] = ks[r + (R) - 1];   /* rotate key schedule */	\
159 	ts[r + (R) + 2] = ts[r + (R) - 1];			\
160 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
161 
162 		/* loop through it */
163 		for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_256)
164 #endif
165 		{
166 #define	R256_8_rounds(R)                         \
167 	R256(0, 1, 2, 3, R_256_0, 8 * (R) + 1);  \
168 	R256(0, 3, 2, 1, R_256_1, 8 * (R) + 2);  \
169 	R256(0, 1, 2, 3, R_256_2, 8 * (R) + 3);  \
170 	R256(0, 3, 2, 1, R_256_3, 8 * (R) + 4);  \
171 	I256(2 * (R));                           \
172 	R256(0, 1, 2, 3, R_256_4, 8 * (R) + 5);  \
173 	R256(0, 3, 2, 1, R_256_5, 8 * (R) + 6);  \
174 	R256(0, 1, 2, 3, R_256_6, 8 * (R) + 7);  \
175 	R256(0, 3, 2, 1, R_256_7, 8 * (R) + 8);  \
176 	I256(2 * (R) + 1);
177 
178 			R256_8_rounds(0);
179 
180 #define	R256_Unroll_R(NN) \
181 	((SKEIN_UNROLL_256 == 0 && SKEIN_256_ROUNDS_TOTAL / 8 > (NN)) || \
182 	(SKEIN_UNROLL_256 > (NN)))
183 
184 #if	R256_Unroll_R(1)
185 			R256_8_rounds(1);
186 #endif
187 #if	R256_Unroll_R(2)
188 			R256_8_rounds(2);
189 #endif
190 #if	R256_Unroll_R(3)
191 			R256_8_rounds(3);
192 #endif
193 #if	R256_Unroll_R(4)
194 			R256_8_rounds(4);
195 #endif
196 #if	R256_Unroll_R(5)
197 			R256_8_rounds(5);
198 #endif
199 #if	R256_Unroll_R(6)
200 			R256_8_rounds(6);
201 #endif
202 #if	R256_Unroll_R(7)
203 			R256_8_rounds(7);
204 #endif
205 #if	R256_Unroll_R(8)
206 			R256_8_rounds(8);
207 #endif
208 #if	R256_Unroll_R(9)
209 			R256_8_rounds(9);
210 #endif
211 #if	R256_Unroll_R(10)
212 			R256_8_rounds(10);
213 #endif
214 #if	R256_Unroll_R(11)
215 			R256_8_rounds(11);
216 #endif
217 #if	R256_Unroll_R(12)
218 			R256_8_rounds(12);
219 #endif
220 #if	R256_Unroll_R(13)
221 			R256_8_rounds(13);
222 #endif
223 #if	R256_Unroll_R(14)
224 			R256_8_rounds(14);
225 #endif
226 #if	(SKEIN_UNROLL_256 > 14)
227 #error  "need more unrolling in Skein_256_Process_Block"
228 #endif
229 		}
230 		/*
231 		 * do the final "feedforward" xor, update context chaining vars
232 		 */
233 		ctx->X[0] = X0 ^ w[0];
234 		ctx->X[1] = X1 ^ w[1];
235 		ctx->X[2] = X2 ^ w[2];
236 		ctx->X[3] = X3 ^ w[3];
237 
238 		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
239 
240 		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
241 	} while (--blkCnt);
242 	ctx->h.T[0] = ts[0];
243 	ctx->h.T[1] = ts[1];
244 }
245 
246 #if	defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
247 size_t
248 Skein_256_Process_Block_CodeSize(void)
249 {
250 	return ((uint8_t *)Skein_256_Process_Block_CodeSize) -
251 	    ((uint8_t *)Skein_256_Process_Block);
252 }
253 
254 uint_t
255 Skein_256_Unroll_Cnt(void)
256 {
257 	return (SKEIN_UNROLL_256);
258 }
259 #endif
260 #endif
261 
262 /* Skein_512 */
263 #if	!(SKEIN_USE_ASM & 512)
264 void
265 Skein_512_Process_Block(Skein_512_Ctxt_t *ctx, const uint8_t *blkPtr,
266     size_t blkCnt, size_t byteCntAdd)
267 {
268 	enum {
269 		WCNT = SKEIN_512_STATE_WORDS
270 	};
271 #undef  RCNT
272 #define	RCNT  (SKEIN_512_ROUNDS_TOTAL / 8)
273 
274 #ifdef	SKEIN_LOOP		/* configure how much to unroll the loop */
275 #define	SKEIN_UNROLL_512 (((SKEIN_LOOP) / 10) % 10)
276 #else
277 #define	SKEIN_UNROLL_512 (0)
278 #endif
279 
280 #if	SKEIN_UNROLL_512
281 #if	(RCNT % SKEIN_UNROLL_512)
282 #error "Invalid SKEIN_UNROLL_512"	/* sanity check on unroll count */
283 #endif
284 	size_t r;
285 	/* key schedule words : chaining vars + tweak + "rotation" */
286 	uint64_t kw[WCNT + 4 + RCNT * 2];
287 #else
288 	uint64_t kw[WCNT + 4];	/* key schedule words : chaining vars + tweak */
289 #endif
290 	/* local copy of vars, for speed */
291 	uint64_t X0, X1, X2, X3, X4, X5, X6, X7;
292 	uint64_t w[WCNT];		/* local copy of input block */
293 #ifdef	SKEIN_DEBUG
294 	/* use for debugging (help compiler put Xn in registers) */
295 	const uint64_t *Xptr[8];
296 	Xptr[0] = &X0;
297 	Xptr[1] = &X1;
298 	Xptr[2] = &X2;
299 	Xptr[3] = &X3;
300 	Xptr[4] = &X4;
301 	Xptr[5] = &X5;
302 	Xptr[6] = &X6;
303 	Xptr[7] = &X7;
304 #endif
305 
306 	Skein_assert(blkCnt != 0);	/* never call with blkCnt == 0! */
307 	ts[0] = ctx->h.T[0];
308 	ts[1] = ctx->h.T[1];
309 	do {
310 		/*
311 		 * this implementation only supports 2**64 input bytes
312 		 * (no carry out here)
313 		 */
314 		ts[0] += byteCntAdd;	/* update processed length */
315 
316 		/* precompute the key schedule for this block */
317 		ks[0] = ctx->X[0];
318 		ks[1] = ctx->X[1];
319 		ks[2] = ctx->X[2];
320 		ks[3] = ctx->X[3];
321 		ks[4] = ctx->X[4];
322 		ks[5] = ctx->X[5];
323 		ks[6] = ctx->X[6];
324 		ks[7] = ctx->X[7];
325 		ks[8] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^
326 		    ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^ SKEIN_KS_PARITY;
327 
328 		ts[2] = ts[0] ^ ts[1];
329 
330 		/* get input block in little-endian format */
331 		Skein_Get64_LSB_First(w, blkPtr, WCNT);
332 		DebugSaveTweak(ctx);
333 		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
334 
335 		X0 = w[0] + ks[0];	/* do the first full key injection */
336 		X1 = w[1] + ks[1];
337 		X2 = w[2] + ks[2];
338 		X3 = w[3] + ks[3];
339 		X4 = w[4] + ks[4];
340 		X5 = w[5] + ks[5] + ts[0];
341 		X6 = w[6] + ks[6] + ts[1];
342 		X7 = w[7] + ks[7];
343 
344 		blkPtr += SKEIN_512_BLOCK_BYTES;
345 
346 		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL,
347 		    Xptr);
348 		/* run the rounds */
349 #define	Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)		\
350 	X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0;\
351 	X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2;\
352 	X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4;\
353 	X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6;
354 
355 #if	SKEIN_UNROLL_512 == 0
356 #define	R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)	/* unrolled */	\
357 	Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)		\
358 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rNum, Xptr);
359 
360 #define	I512(R)								\
361 	X0 += ks[((R) + 1) % 9];	/* inject the key schedule value */\
362 	X1 += ks[((R) + 2) % 9];					\
363 	X2 += ks[((R) + 3) % 9];					\
364 	X3 += ks[((R) + 4) % 9];					\
365 	X4 += ks[((R) + 5) % 9];					\
366 	X5 += ks[((R) + 6) % 9] + ts[((R) + 1) % 3];			\
367 	X6 += ks[((R) + 7) % 9] + ts[((R) + 2) % 3];			\
368 	X7 += ks[((R) + 8) % 9] + (R) + 1;				\
369 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
370 #else				/* looping version */
371 #define	R512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)			\
372 	Round512(p0, p1, p2, p3, p4, p5, p6, p7, ROT, rNum)		\
373 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rNum, Xptr);
374 
375 #define	I512(R)								\
376 	X0 += ks[r + (R) + 0];	/* inject the key schedule value */	\
377 	X1 += ks[r + (R) + 1];						\
378 	X2 += ks[r + (R) + 2];						\
379 	X3 += ks[r + (R) + 3];						\
380 	X4 += ks[r + (R) + 4];						\
381 	X5 += ks[r + (R) + 5] + ts[r + (R) + 0];			\
382 	X6 += ks[r + (R) + 6] + ts[r + (R) + 1];			\
383 	X7 += ks[r + (R) + 7] + r + (R);				\
384 	ks[r + (R)+8] = ks[r + (R) - 1];	/* rotate key schedule */\
385 	ts[r + (R)+2] = ts[r + (R) - 1];				\
386 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
387 
388 		/* loop through it */
389 		for (r = 1; r < 2 * RCNT; r += 2 * SKEIN_UNROLL_512)
390 #endif				/* end of looped code definitions */
391 		{
392 #define	R512_8_rounds(R)	/* do 8 full rounds */			\
393 	R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_0, 8 * (R) + 1);		\
394 	R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_1, 8 * (R) + 2);		\
395 	R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_2, 8 * (R) + 3);		\
396 	R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_3, 8 * (R) + 4);		\
397 	I512(2 * (R));							\
398 	R512(0, 1, 2, 3, 4, 5, 6, 7, R_512_4, 8 * (R) + 5);		\
399 	R512(2, 1, 4, 7, 6, 5, 0, 3, R_512_5, 8 * (R) + 6);		\
400 	R512(4, 1, 6, 3, 0, 5, 2, 7, R_512_6, 8 * (R) + 7);		\
401 	R512(6, 1, 0, 7, 2, 5, 4, 3, R_512_7, 8 * (R) + 8);		\
402 	I512(2*(R) + 1);		/* and key injection */
403 
404 			R512_8_rounds(0);
405 
406 #define	R512_Unroll_R(NN) \
407 	((SKEIN_UNROLL_512 == 0 && SKEIN_512_ROUNDS_TOTAL / 8 > (NN)) || \
408 	(SKEIN_UNROLL_512 > (NN)))
409 
410 #if	R512_Unroll_R(1)
411 			R512_8_rounds(1);
412 #endif
413 #if	R512_Unroll_R(2)
414 			R512_8_rounds(2);
415 #endif
416 #if	R512_Unroll_R(3)
417 			R512_8_rounds(3);
418 #endif
419 #if	R512_Unroll_R(4)
420 			R512_8_rounds(4);
421 #endif
422 #if	R512_Unroll_R(5)
423 			R512_8_rounds(5);
424 #endif
425 #if	R512_Unroll_R(6)
426 			R512_8_rounds(6);
427 #endif
428 #if	R512_Unroll_R(7)
429 			R512_8_rounds(7);
430 #endif
431 #if	R512_Unroll_R(8)
432 			R512_8_rounds(8);
433 #endif
434 #if	R512_Unroll_R(9)
435 			R512_8_rounds(9);
436 #endif
437 #if	R512_Unroll_R(10)
438 			R512_8_rounds(10);
439 #endif
440 #if	R512_Unroll_R(11)
441 			R512_8_rounds(11);
442 #endif
443 #if	R512_Unroll_R(12)
444 			R512_8_rounds(12);
445 #endif
446 #if	R512_Unroll_R(13)
447 			R512_8_rounds(13);
448 #endif
449 #if	R512_Unroll_R(14)
450 			R512_8_rounds(14);
451 #endif
452 #if	(SKEIN_UNROLL_512 > 14)
453 #error "need more unrolling in Skein_512_Process_Block"
454 #endif
455 		}
456 
457 		/*
458 		 * do the final "feedforward" xor, update context chaining vars
459 		 */
460 		ctx->X[0] = X0 ^ w[0];
461 		ctx->X[1] = X1 ^ w[1];
462 		ctx->X[2] = X2 ^ w[2];
463 		ctx->X[3] = X3 ^ w[3];
464 		ctx->X[4] = X4 ^ w[4];
465 		ctx->X[5] = X5 ^ w[5];
466 		ctx->X[6] = X6 ^ w[6];
467 		ctx->X[7] = X7 ^ w[7];
468 		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
469 
470 		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
471 	} while (--blkCnt);
472 	ctx->h.T[0] = ts[0];
473 	ctx->h.T[1] = ts[1];
474 }
475 
476 #if	defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
477 size_t
478 Skein_512_Process_Block_CodeSize(void)
479 {
480 	return ((uint8_t *)Skein_512_Process_Block_CodeSize) -
481 	    ((uint8_t *)Skein_512_Process_Block);
482 }
483 
484 uint_t
485 Skein_512_Unroll_Cnt(void)
486 {
487 	return (SKEIN_UNROLL_512);
488 }
489 #endif
490 #endif
491 
492 /*  Skein1024 */
493 #if	!(SKEIN_USE_ASM & 1024)
494 void
495 Skein1024_Process_Block(Skein1024_Ctxt_t *ctx, const uint8_t *blkPtr,
496     size_t blkCnt, size_t byteCntAdd)
497 {
498 	/* do it in C, always looping (unrolled is bigger AND slower!) */
499 	enum {
500 		WCNT = SKEIN1024_STATE_WORDS
501 	};
502 #undef  RCNT
503 #define	RCNT  (SKEIN1024_ROUNDS_TOTAL/8)
504 
505 #ifdef	SKEIN_LOOP		/* configure how much to unroll the loop */
506 #define	SKEIN_UNROLL_1024 ((SKEIN_LOOP)%10)
507 #else
508 #define	SKEIN_UNROLL_1024 (0)
509 #endif
510 
511 #if	(SKEIN_UNROLL_1024 != 0)
512 #if	(RCNT % SKEIN_UNROLL_1024)
513 #error "Invalid SKEIN_UNROLL_1024"	/* sanity check on unroll count */
514 #endif
515 	size_t r;
516 	/* key schedule words : chaining vars + tweak + "rotation" */
517 	uint64_t kw[WCNT + 4 + RCNT * 2];
518 #else
519 	uint64_t kw[WCNT + 4];	/* key schedule words : chaining vars + tweak */
520 #endif
521 
522 	/* local copy of vars, for speed */
523 	uint64_t X00, X01, X02, X03, X04, X05, X06, X07, X08, X09, X10, X11,
524 	    X12, X13, X14, X15;
525 	uint64_t w[WCNT];		/* local copy of input block */
526 #ifdef	SKEIN_DEBUG
527 	/* use for debugging (help compiler put Xn in registers) */
528 	const uint64_t *Xptr[16];
529 	Xptr[0] = &X00;
530 	Xptr[1] = &X01;
531 	Xptr[2] = &X02;
532 	Xptr[3] = &X03;
533 	Xptr[4] = &X04;
534 	Xptr[5] = &X05;
535 	Xptr[6] = &X06;
536 	Xptr[7] = &X07;
537 	Xptr[8] = &X08;
538 	Xptr[9] = &X09;
539 	Xptr[10] = &X10;
540 	Xptr[11] = &X11;
541 	Xptr[12] = &X12;
542 	Xptr[13] = &X13;
543 	Xptr[14] = &X14;
544 	Xptr[15] = &X15;
545 #endif
546 
547 	Skein_assert(blkCnt != 0);	/* never call with blkCnt == 0! */
548 	ts[0] = ctx->h.T[0];
549 	ts[1] = ctx->h.T[1];
550 	do {
551 		/*
552 		 * this implementation only supports 2**64 input bytes
553 		 * (no carry out here)
554 		 */
555 		ts[0] += byteCntAdd;	/* update processed length */
556 
557 		/* precompute the key schedule for this block */
558 		ks[0] = ctx->X[0];
559 		ks[1] = ctx->X[1];
560 		ks[2] = ctx->X[2];
561 		ks[3] = ctx->X[3];
562 		ks[4] = ctx->X[4];
563 		ks[5] = ctx->X[5];
564 		ks[6] = ctx->X[6];
565 		ks[7] = ctx->X[7];
566 		ks[8] = ctx->X[8];
567 		ks[9] = ctx->X[9];
568 		ks[10] = ctx->X[10];
569 		ks[11] = ctx->X[11];
570 		ks[12] = ctx->X[12];
571 		ks[13] = ctx->X[13];
572 		ks[14] = ctx->X[14];
573 		ks[15] = ctx->X[15];
574 		ks[16] = ks[0] ^ ks[1] ^ ks[2] ^ ks[3] ^
575 		    ks[4] ^ ks[5] ^ ks[6] ^ ks[7] ^
576 		    ks[8] ^ ks[9] ^ ks[10] ^ ks[11] ^
577 		    ks[12] ^ ks[13] ^ ks[14] ^ ks[15] ^ SKEIN_KS_PARITY;
578 
579 		ts[2] = ts[0] ^ ts[1];
580 
581 		/* get input block in little-endian format */
582 		Skein_Get64_LSB_First(w, blkPtr, WCNT);
583 		DebugSaveTweak(ctx);
584 		Skein_Show_Block(BLK_BITS, &ctx->h, ctx->X, blkPtr, w, ks, ts);
585 
586 		X00 = w[0] + ks[0];	/* do the first full key injection */
587 		X01 = w[1] + ks[1];
588 		X02 = w[2] + ks[2];
589 		X03 = w[3] + ks[3];
590 		X04 = w[4] + ks[4];
591 		X05 = w[5] + ks[5];
592 		X06 = w[6] + ks[6];
593 		X07 = w[7] + ks[7];
594 		X08 = w[8] + ks[8];
595 		X09 = w[9] + ks[9];
596 		X10 = w[10] + ks[10];
597 		X11 = w[11] + ks[11];
598 		X12 = w[12] + ks[12];
599 		X13 = w[13] + ks[13] + ts[0];
600 		X14 = w[14] + ks[14] + ts[1];
601 		X15 = w[15] + ks[15];
602 
603 		Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INITIAL,
604 		    Xptr);
605 
606 #define	Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC,	\
607 	pD, pE, pF, ROT, rNum)						\
608 	X##p0 += X##p1; X##p1 = RotL_64(X##p1, ROT##_0); X##p1 ^= X##p0;\
609 	X##p2 += X##p3; X##p3 = RotL_64(X##p3, ROT##_1); X##p3 ^= X##p2;\
610 	X##p4 += X##p5; X##p5 = RotL_64(X##p5, ROT##_2); X##p5 ^= X##p4;\
611 	X##p6 += X##p7; X##p7 = RotL_64(X##p7, ROT##_3); X##p7 ^= X##p6;\
612 	X##p8 += X##p9; X##p9 = RotL_64(X##p9, ROT##_4); X##p9 ^= X##p8;\
613 	X##pA += X##pB; X##pB = RotL_64(X##pB, ROT##_5); X##pB ^= X##pA;\
614 	X##pC += X##pD; X##pD = RotL_64(X##pD, ROT##_6); X##pD ^= X##pC;\
615 	X##pE += X##pF; X##pF = RotL_64(X##pF, ROT##_7); X##pF ^= X##pE;
616 
617 #if	SKEIN_UNROLL_1024 == 0
618 #define	R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD,	\
619 	pE, pF, ROT, rn)						\
620 	Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC,	\
621 	pD, pE, pF, ROT, rn)						\
622 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, rn, Xptr);
623 
624 #define	I1024(R)							\
625 	X00 += ks[((R) + 1) % 17];	/* inject the key schedule value */\
626 	X01 += ks[((R) + 2) % 17];					\
627 	X02 += ks[((R) + 3) % 17];					\
628 	X03 += ks[((R) + 4) % 17];					\
629 	X04 += ks[((R) + 5) % 17];					\
630 	X05 += ks[((R) + 6) % 17];					\
631 	X06 += ks[((R) + 7) % 17];					\
632 	X07 += ks[((R) + 8) % 17];					\
633 	X08 += ks[((R) + 9) % 17];					\
634 	X09 += ks[((R) + 10) % 17];					\
635 	X10 += ks[((R) + 11) % 17];					\
636 	X11 += ks[((R) + 12) % 17];					\
637 	X12 += ks[((R) + 13) % 17];					\
638 	X13 += ks[((R) + 14) % 17] + ts[((R) + 1) % 3];			\
639 	X14 += ks[((R) + 15) % 17] + ts[((R) + 2) % 3];			\
640 	X15 += ks[((R) + 16) % 17] + (R) +1;				\
641 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
642 #else				/* looping version */
643 #define	R1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC, pD,	\
644 	pE, pF, ROT, rn)						\
645 	Round1024(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pA, pB, pC,	\
646 	pD, pE, pF, ROT, rn)						\
647 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, 4 * (r - 1) + rn, Xptr);
648 
649 #define	I1024(R)							\
650 	X00 += ks[r + (R) + 0];	/* inject the key schedule value */	\
651 	X01 += ks[r + (R) + 1];						\
652 	X02 += ks[r + (R) + 2];						\
653 	X03 += ks[r + (R) + 3];						\
654 	X04 += ks[r + (R) + 4];						\
655 	X05 += ks[r + (R) + 5];						\
656 	X06 += ks[r + (R) + 6];						\
657 	X07 += ks[r + (R) + 7];						\
658 	X08 += ks[r + (R) + 8];						\
659 	X09 += ks[r + (R) + 9];						\
660 	X10 += ks[r + (R) + 10];					\
661 	X11 += ks[r + (R) + 11];					\
662 	X12 += ks[r + (R) + 12];					\
663 	X13 += ks[r + (R) + 13] + ts[r + (R) + 0];			\
664 	X14 += ks[r + (R) + 14] + ts[r + (R) + 1];			\
665 	X15 += ks[r + (R) + 15] +  r + (R);				\
666 	ks[r + (R) + 16] = ks[r + (R) - 1];	/* rotate key schedule */\
667 	ts[r + (R) + 2] = ts[r + (R) - 1];				\
668 	Skein_Show_R_Ptr(BLK_BITS, &ctx->h, SKEIN_RND_KEY_INJECT, Xptr);
669 
670 		/* loop through it */
671 		for (r = 1; r <= 2 * RCNT; r += 2 * SKEIN_UNROLL_1024)
672 #endif
673 		{
674 #define	R1024_8_rounds(R)	/* do 8 full rounds */			\
675 	R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13,	\
676 	    14, 15, R1024_0, 8 * (R) + 1);				\
677 	R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05,	\
678 	    08, 01, R1024_1, 8 * (R) + 2);				\
679 	R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11,	\
680 	    10, 09, R1024_2, 8 * (R) + 3);				\
681 	R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03,	\
682 	    12, 07, R1024_3, 8 * (R) + 4);				\
683 	I1024(2 * (R));							\
684 	R1024(00, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13,	\
685 	    14, 15, R1024_4, 8 * (R) + 5);				\
686 	R1024(00, 09, 02, 13, 06, 11, 04, 15, 10, 07, 12, 03, 14, 05,	\
687 	    08, 01, R1024_5, 8 * (R) + 6);				\
688 	R1024(00, 07, 02, 05, 04, 03, 06, 01, 12, 15, 14, 13, 08, 11,	\
689 	    10, 09, R1024_6, 8 * (R) + 7);				\
690 	R1024(00, 15, 02, 11, 06, 13, 04, 09, 14, 01, 08, 05, 10, 03,	\
691 	    12, 07, R1024_7, 8 * (R) + 8);				\
692 	I1024(2 * (R) + 1);
693 
694 			R1024_8_rounds(0);
695 
696 #define	R1024_Unroll_R(NN)						\
697 	((SKEIN_UNROLL_1024 == 0 && SKEIN1024_ROUNDS_TOTAL/8 > (NN)) ||	\
698 	(SKEIN_UNROLL_1024 > (NN)))
699 
700 #if	R1024_Unroll_R(1)
701 			R1024_8_rounds(1);
702 #endif
703 #if	R1024_Unroll_R(2)
704 			R1024_8_rounds(2);
705 #endif
706 #if	R1024_Unroll_R(3)
707 			R1024_8_rounds(3);
708 #endif
709 #if	R1024_Unroll_R(4)
710 			R1024_8_rounds(4);
711 #endif
712 #if	R1024_Unroll_R(5)
713 			R1024_8_rounds(5);
714 #endif
715 #if	R1024_Unroll_R(6)
716 			R1024_8_rounds(6);
717 #endif
718 #if	R1024_Unroll_R(7)
719 			R1024_8_rounds(7);
720 #endif
721 #if	R1024_Unroll_R(8)
722 			R1024_8_rounds(8);
723 #endif
724 #if	R1024_Unroll_R(9)
725 			R1024_8_rounds(9);
726 #endif
727 #if	R1024_Unroll_R(10)
728 			R1024_8_rounds(10);
729 #endif
730 #if	R1024_Unroll_R(11)
731 			R1024_8_rounds(11);
732 #endif
733 #if	R1024_Unroll_R(12)
734 			R1024_8_rounds(12);
735 #endif
736 #if	R1024_Unroll_R(13)
737 			R1024_8_rounds(13);
738 #endif
739 #if	R1024_Unroll_R(14)
740 			R1024_8_rounds(14);
741 #endif
742 #if	(SKEIN_UNROLL_1024 > 14)
743 #error  "need more unrolling in Skein_1024_Process_Block"
744 #endif
745 		}
746 		/*
747 		 * do the final "feedforward" xor, update context chaining vars
748 		 */
749 
750 		ctx->X[0] = X00 ^ w[0];
751 		ctx->X[1] = X01 ^ w[1];
752 		ctx->X[2] = X02 ^ w[2];
753 		ctx->X[3] = X03 ^ w[3];
754 		ctx->X[4] = X04 ^ w[4];
755 		ctx->X[5] = X05 ^ w[5];
756 		ctx->X[6] = X06 ^ w[6];
757 		ctx->X[7] = X07 ^ w[7];
758 		ctx->X[8] = X08 ^ w[8];
759 		ctx->X[9] = X09 ^ w[9];
760 		ctx->X[10] = X10 ^ w[10];
761 		ctx->X[11] = X11 ^ w[11];
762 		ctx->X[12] = X12 ^ w[12];
763 		ctx->X[13] = X13 ^ w[13];
764 		ctx->X[14] = X14 ^ w[14];
765 		ctx->X[15] = X15 ^ w[15];
766 
767 		Skein_Show_Round(BLK_BITS, &ctx->h, SKEIN_RND_FEED_FWD, ctx->X);
768 
769 		ts[1] &= ~SKEIN_T1_FLAG_FIRST;
770 		blkPtr += SKEIN1024_BLOCK_BYTES;
771 	} while (--blkCnt);
772 	ctx->h.T[0] = ts[0];
773 	ctx->h.T[1] = ts[1];
774 }
775 
776 #if	defined(SKEIN_CODE_SIZE) || defined(SKEIN_PERF)
777 size_t
778 Skein1024_Process_Block_CodeSize(void)
779 {
780 	return ((uint8_t *)Skein1024_Process_Block_CodeSize) -
781 	    ((uint8_t *)Skein1024_Process_Block);
782 }
783 
784 uint_t
785 Skein1024_Unroll_Cnt(void)
786 {
787 	return (SKEIN_UNROLL_1024);
788 }
789 #endif
790 #endif
791