xref: /titanic_41/usr/src/common/crypto/sha1/sha1.c (revision 2954adb080121c02025bb94537c313e6f51c9fd7)
1 /*
2  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  */
5 
6 /*
7  * The basic framework for this code came from the reference
8  * implementation for MD5.  That implementation is Copyright (C)
9  * 1991-2, RSA Data Security, Inc. Created 1991. All rights reserved.
10  *
11  * License to copy and use this software is granted provided that it
12  * is identified as the "RSA Data Security, Inc. MD5 Message-Digest
13  * Algorithm" in all material mentioning or referencing this software
14  * or this function.
15  *
16  * License is also granted to make and use derivative works provided
17  * that such works are identified as "derived from the RSA Data
18  * Security, Inc. MD5 Message-Digest Algorithm" in all material
19  * mentioning or referencing the derived work.
20  *
21  * RSA Data Security, Inc. makes no representations concerning either
22  * the merchantability of this software or the suitability of this
23  * software for any particular purpose. It is provided "as is"
24  * without express or implied warranty of any kind.
25  *
26  * These notices must be retained in any copies of any part of this
27  * documentation and/or software.
28  *
29  * NOTE: Cleaned-up and optimized, version of SHA1, based on the FIPS 180-1
30  * standard, available at http://www.itl.nist.gov/fipspubs/fip180-1.htm
31  * Not as fast as one would like -- further optimizations are encouraged
32  * and appreciated.
33  */
34 
35 #if !defined(_KERNEL) && !defined(_BOOT)
36 #include <stdint.h>
37 #include <strings.h>
38 #include <stdlib.h>
39 #include <errno.h>
40 #include <sys/systeminfo.h>
41 #endif  /* !_KERNEL && !_BOOT */
42 
43 #include <sys/types.h>
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysmacros.h>
47 #include <sys/sha1.h>
48 #include <sys/sha1_consts.h>
49 
50 #ifdef _LITTLE_ENDIAN
51 #include <sys/byteorder.h>
52 #define	HAVE_HTONL
53 #endif
54 
55 #ifdef	_BOOT
56 #define	bcopy(_s, _d, _l)	((void) memcpy((_d), (_s), (_l)))
57 #define	bzero(_m, _l)		((void) memset((_m), 0, (_l)))
58 #endif
59 
60 static void Encode(uint8_t *, const uint32_t *, size_t);
61 
62 #if	defined(__sparc)
63 
64 #define	SHA1_TRANSFORM(ctx, in) \
65 	SHA1Transform((ctx)->state[0], (ctx)->state[1], (ctx)->state[2], \
66 		(ctx)->state[3], (ctx)->state[4], (ctx), (in))
67 
68 static void SHA1Transform(uint32_t, uint32_t, uint32_t, uint32_t, uint32_t,
69     SHA1_CTX *, const uint8_t *);
70 
71 #elif	defined(__amd64)
72 
73 #define	SHA1_TRANSFORM(ctx, in) sha1_block_data_order((ctx), (in), 1)
74 #define	SHA1_TRANSFORM_BLOCKS(ctx, in, num) sha1_block_data_order((ctx), \
75 		(in), (num))
76 
77 void sha1_block_data_order(SHA1_CTX *ctx, const void *inpp, size_t num_blocks);
78 
79 #else
80 
81 #define	SHA1_TRANSFORM(ctx, in) SHA1Transform((ctx), (in))
82 
83 static void SHA1Transform(SHA1_CTX *, const uint8_t *);
84 
85 #endif
86 
87 
88 static uint8_t PADDING[64] = { 0x80, /* all zeros */ };
89 
90 /*
91  * F, G, and H are the basic SHA1 functions.
92  */
93 #define	F(b, c, d)	(((b) & (c)) | ((~b) & (d)))
94 #define	G(b, c, d)	((b) ^ (c) ^ (d))
95 #define	H(b, c, d)	(((b) & (c)) | (((b)|(c)) & (d)))
96 
97 /*
98  * ROTATE_LEFT rotates x left n bits.
99  */
100 
101 #if	defined(__GNUC__) && defined(_LP64)
102 static __inline__ uint64_t
103 ROTATE_LEFT(uint64_t value, uint32_t n)
104 {
105 	uint32_t t32;
106 
107 	t32 = (uint32_t)value;
108 	return ((t32 << n) | (t32 >> (32 - n)));
109 }
110 
111 #else
112 
113 #define	ROTATE_LEFT(x, n)	\
114 	(((x) << (n)) | ((x) >> ((sizeof (x) * NBBY)-(n))))
115 
116 #endif
117 
118 
119 /*
120  * SHA1Init()
121  *
122  * purpose: initializes the sha1 context and begins and sha1 digest operation
123  *   input: SHA1_CTX *	: the context to initializes.
124  *  output: void
125  */
126 
127 void
128 SHA1Init(SHA1_CTX *ctx)
129 {
130 	ctx->count[0] = ctx->count[1] = 0;
131 
132 	/*
133 	 * load magic initialization constants. Tell lint
134 	 * that these constants are unsigned by using U.
135 	 */
136 
137 	ctx->state[0] = 0x67452301U;
138 	ctx->state[1] = 0xefcdab89U;
139 	ctx->state[2] = 0x98badcfeU;
140 	ctx->state[3] = 0x10325476U;
141 	ctx->state[4] = 0xc3d2e1f0U;
142 }
143 
144 #ifdef VIS_SHA1
145 #ifdef _KERNEL
146 
147 #include <sys/regset.h>
148 #include <sys/vis.h>
149 #include <sys/fpu/fpusystm.h>
150 
151 /* the alignment for block stores to save fp registers */
152 #define	VIS_ALIGN	(64)
153 
154 extern int sha1_savefp(kfpu_t *, int);
155 extern void sha1_restorefp(kfpu_t *);
156 
157 uint32_t	vis_sha1_svfp_threshold = 128;
158 
159 #endif /* _KERNEL */
160 
161 /*
162  * VIS SHA-1 consts.
163  */
164 static uint64_t VIS[] = {
165 	0x8000000080000000ULL,
166 	0x0002000200020002ULL,
167 	0x5a8279996ed9eba1ULL,
168 	0x8f1bbcdcca62c1d6ULL,
169 	0x012389ab456789abULL};
170 
171 extern void SHA1TransformVIS(uint64_t *, uint32_t *, uint32_t *, uint64_t *);
172 
173 
174 /*
175  * SHA1Update()
176  *
177  * purpose: continues an sha1 digest operation, using the message block
178  *          to update the context.
179  *   input: SHA1_CTX *	: the context to update
180  *          void *	: the message block
181  *          size_t    : the length of the message block in bytes
182  *  output: void
183  */
184 
185 void
186 SHA1Update(SHA1_CTX *ctx, const void *inptr, size_t input_len)
187 {
188 	uint32_t i, buf_index, buf_len;
189 	uint64_t X0[40], input64[8];
190 	const uint8_t *input = inptr;
191 #ifdef _KERNEL
192 	int usevis = 0;
193 #else
194 	int usevis = 1;
195 #endif /* _KERNEL */
196 
197 	/* check for noop */
198 	if (input_len == 0)
199 		return;
200 
201 	/* compute number of bytes mod 64 */
202 	buf_index = (ctx->count[1] >> 3) & 0x3F;
203 
204 	/* update number of bits */
205 	if ((ctx->count[1] += (input_len << 3)) < (input_len << 3))
206 		ctx->count[0]++;
207 
208 	ctx->count[0] += (input_len >> 29);
209 
210 	buf_len = 64 - buf_index;
211 
212 	/* transform as many times as possible */
213 	i = 0;
214 	if (input_len >= buf_len) {
215 #ifdef _KERNEL
216 		kfpu_t *fpu;
217 		if (fpu_exists) {
218 			uint8_t fpua[sizeof (kfpu_t) + GSR_SIZE + VIS_ALIGN];
219 			uint32_t len = (input_len + buf_index) & ~0x3f;
220 			int svfp_ok;
221 
222 			fpu = (kfpu_t *)P2ROUNDUP((uintptr_t)fpua, 64);
223 			svfp_ok = ((len >= vis_sha1_svfp_threshold) ? 1 : 0);
224 			usevis = fpu_exists && sha1_savefp(fpu, svfp_ok);
225 		} else {
226 			usevis = 0;
227 		}
228 #endif /* _KERNEL */
229 
230 		/*
231 		 * general optimization:
232 		 *
233 		 * only do initial bcopy() and SHA1Transform() if
234 		 * buf_index != 0.  if buf_index == 0, we're just
235 		 * wasting our time doing the bcopy() since there
236 		 * wasn't any data left over from a previous call to
237 		 * SHA1Update().
238 		 */
239 
240 		if (buf_index) {
241 			bcopy(input, &ctx->buf_un.buf8[buf_index], buf_len);
242 			if (usevis) {
243 				SHA1TransformVIS(X0,
244 				    ctx->buf_un.buf32,
245 				    &ctx->state[0], VIS);
246 			} else {
247 				SHA1_TRANSFORM(ctx, ctx->buf_un.buf8);
248 			}
249 			i = buf_len;
250 		}
251 
252 		/*
253 		 * VIS SHA-1: uses the VIS 1.0 instructions to accelerate
254 		 * SHA-1 processing. This is achieved by "offloading" the
255 		 * computation of the message schedule (MS) to the VIS units.
256 		 * This allows the VIS computation of the message schedule
257 		 * to be performed in parallel with the standard integer
258 		 * processing of the remainder of the SHA-1 computation.
259 		 * performance by up to around 1.37X, compared to an optimized
260 		 * integer-only implementation.
261 		 *
262 		 * The VIS implementation of SHA1Transform has a different API
263 		 * to the standard integer version:
264 		 *
265 		 * void SHA1TransformVIS(
266 		 *	 uint64_t *, // Pointer to MS for ith block
267 		 *	 uint32_t *, // Pointer to ith block of message data
268 		 *	 uint32_t *, // Pointer to SHA state i.e ctx->state
269 		 *	 uint64_t *, // Pointer to various VIS constants
270 		 * )
271 		 *
272 		 * Note: the message data must by 4-byte aligned.
273 		 *
274 		 * Function requires VIS 1.0 support.
275 		 *
276 		 * Handling is provided to deal with arbitrary byte alingment
277 		 * of the input data but the performance gains are reduced
278 		 * for alignments other than 4-bytes.
279 		 */
280 		if (usevis) {
281 			if (!IS_P2ALIGNED(&input[i], sizeof (uint32_t))) {
282 				/*
283 				 * Main processing loop - input misaligned
284 				 */
285 				for (; i + 63 < input_len; i += 64) {
286 					bcopy(&input[i], input64, 64);
287 					SHA1TransformVIS(X0,
288 					    (uint32_t *)input64,
289 					    &ctx->state[0], VIS);
290 				}
291 			} else {
292 				/*
293 				 * Main processing loop - input 8-byte aligned
294 				 */
295 				for (; i + 63 < input_len; i += 64) {
296 					SHA1TransformVIS(X0,
297 					    /* LINTED E_BAD_PTR_CAST_ALIGN */
298 					    (uint32_t *)&input[i], /* CSTYLED */
299 					    &ctx->state[0], VIS);
300 				}
301 
302 			}
303 #ifdef _KERNEL
304 			sha1_restorefp(fpu);
305 #endif /* _KERNEL */
306 		} else {
307 			for (; i + 63 < input_len; i += 64) {
308 				SHA1_TRANSFORM(ctx, &input[i]);
309 			}
310 		}
311 
312 		/*
313 		 * general optimization:
314 		 *
315 		 * if i and input_len are the same, return now instead
316 		 * of calling bcopy(), since the bcopy() in this case
317 		 * will be an expensive nop.
318 		 */
319 
320 		if (input_len == i)
321 			return;
322 
323 		buf_index = 0;
324 	}
325 
326 	/* buffer remaining input */
327 	bcopy(&input[i], &ctx->buf_un.buf8[buf_index], input_len - i);
328 }
329 
330 #else /* VIS_SHA1 */
331 
332 void
333 SHA1Update(SHA1_CTX *ctx, const void *inptr, size_t input_len)
334 {
335 	uint32_t i, buf_index, buf_len;
336 	const uint8_t *input = inptr;
337 #if defined(__amd64)
338 	uint32_t	block_count;
339 #endif	/* __amd64 */
340 
341 	/* check for noop */
342 	if (input_len == 0)
343 		return;
344 
345 	/* compute number of bytes mod 64 */
346 	buf_index = (ctx->count[1] >> 3) & 0x3F;
347 
348 	/* update number of bits */
349 	if ((ctx->count[1] += (input_len << 3)) < (input_len << 3))
350 		ctx->count[0]++;
351 
352 	ctx->count[0] += (input_len >> 29);
353 
354 	buf_len = 64 - buf_index;
355 
356 	/* transform as many times as possible */
357 	i = 0;
358 	if (input_len >= buf_len) {
359 
360 		/*
361 		 * general optimization:
362 		 *
363 		 * only do initial bcopy() and SHA1Transform() if
364 		 * buf_index != 0.  if buf_index == 0, we're just
365 		 * wasting our time doing the bcopy() since there
366 		 * wasn't any data left over from a previous call to
367 		 * SHA1Update().
368 		 */
369 
370 		if (buf_index) {
371 			bcopy(input, &ctx->buf_un.buf8[buf_index], buf_len);
372 			SHA1_TRANSFORM(ctx, ctx->buf_un.buf8);
373 			i = buf_len;
374 		}
375 
376 #if !defined(__amd64)
377 		for (; i + 63 < input_len; i += 64)
378 			SHA1_TRANSFORM(ctx, &input[i]);
379 #else
380 		block_count = (input_len - i) >> 6;
381 		if (block_count > 0) {
382 			SHA1_TRANSFORM_BLOCKS(ctx, &input[i], block_count);
383 			i += block_count << 6;
384 		}
385 #endif	/* !__amd64 */
386 
387 		/*
388 		 * general optimization:
389 		 *
390 		 * if i and input_len are the same, return now instead
391 		 * of calling bcopy(), since the bcopy() in this case
392 		 * will be an expensive nop.
393 		 */
394 
395 		if (input_len == i)
396 			return;
397 
398 		buf_index = 0;
399 	}
400 
401 	/* buffer remaining input */
402 	bcopy(&input[i], &ctx->buf_un.buf8[buf_index], input_len - i);
403 }
404 
405 #endif /* VIS_SHA1 */
406 
407 /*
408  * SHA1Final()
409  *
410  * purpose: ends an sha1 digest operation, finalizing the message digest and
411  *          zeroing the context.
412  *   input: uchar_t *	: A buffer to store the digest.
413  *			: The function actually uses void* because many
414  *			: callers pass things other than uchar_t here.
415  *          SHA1_CTX *  : the context to finalize, save, and zero
416  *  output: void
417  */
418 
419 void
420 SHA1Final(void *digest, SHA1_CTX *ctx)
421 {
422 	uint8_t		bitcount_be[sizeof (ctx->count)];
423 	uint32_t	index = (ctx->count[1] >> 3) & 0x3f;
424 
425 	/* store bit count, big endian */
426 	Encode(bitcount_be, ctx->count, sizeof (bitcount_be));
427 
428 	/* pad out to 56 mod 64 */
429 	SHA1Update(ctx, PADDING, ((index < 56) ? 56 : 120) - index);
430 
431 	/* append length (before padding) */
432 	SHA1Update(ctx, bitcount_be, sizeof (bitcount_be));
433 
434 	/* store state in digest */
435 	Encode(digest, ctx->state, sizeof (ctx->state));
436 
437 	/* zeroize sensitive information */
438 	bzero(ctx, sizeof (*ctx));
439 }
440 
441 
442 #if !defined(__amd64)
443 
444 typedef uint32_t sha1word;
445 
446 /*
447  * sparc optimization:
448  *
449  * on the sparc, we can load big endian 32-bit data easily.  note that
450  * special care must be taken to ensure the address is 32-bit aligned.
451  * in the interest of speed, we don't check to make sure, since
452  * careful programming can guarantee this for us.
453  */
454 
455 #if	defined(_BIG_ENDIAN)
456 #define	LOAD_BIG_32(addr)	(*(uint32_t *)(addr))
457 
458 #elif	defined(HAVE_HTONL)
459 #define	LOAD_BIG_32(addr) htonl(*((uint32_t *)(addr)))
460 
461 #else
462 /* little endian -- will work on big endian, but slowly */
463 #define	LOAD_BIG_32(addr)	\
464 	(((addr)[0] << 24) | ((addr)[1] << 16) | ((addr)[2] << 8) | (addr)[3])
465 #endif	/* _BIG_ENDIAN */
466 
467 /*
468  * SHA1Transform()
469  */
470 #if	defined(W_ARRAY)
471 #define	W(n) w[n]
472 #else	/* !defined(W_ARRAY) */
473 #define	W(n) w_ ## n
474 #endif	/* !defined(W_ARRAY) */
475 
476 
477 #if	defined(__sparc)
478 
479 /*
480  * sparc register window optimization:
481  *
482  * `a', `b', `c', `d', and `e' are passed into SHA1Transform
483  * explicitly since it increases the number of registers available to
484  * the compiler.  under this scheme, these variables can be held in
485  * %i0 - %i4, which leaves more local and out registers available.
486  *
487  * purpose: sha1 transformation -- updates the digest based on `block'
488  *   input: uint32_t	: bytes  1 -  4 of the digest
489  *          uint32_t	: bytes  5 -  8 of the digest
490  *          uint32_t	: bytes  9 - 12 of the digest
491  *          uint32_t	: bytes 12 - 16 of the digest
492  *          uint32_t	: bytes 16 - 20 of the digest
493  *          SHA1_CTX *	: the context to update
494  *          uint8_t [64]: the block to use to update the digest
495  *  output: void
496  */
497 
498 void
499 SHA1Transform(uint32_t a, uint32_t b, uint32_t c, uint32_t d, uint32_t e,
500     SHA1_CTX *ctx, const uint8_t blk[64])
501 {
502 	/*
503 	 * sparc optimization:
504 	 *
505 	 * while it is somewhat counter-intuitive, on sparc, it is
506 	 * more efficient to place all the constants used in this
507 	 * function in an array and load the values out of the array
508 	 * than to manually load the constants.  this is because
509 	 * setting a register to a 32-bit value takes two ops in most
510 	 * cases: a `sethi' and an `or', but loading a 32-bit value
511 	 * from memory only takes one `ld' (or `lduw' on v9).  while
512 	 * this increases memory usage, the compiler can find enough
513 	 * other things to do while waiting to keep the pipeline does
514 	 * not stall.  additionally, it is likely that many of these
515 	 * constants are cached so that later accesses do not even go
516 	 * out to the bus.
517 	 *
518 	 * this array is declared `static' to keep the compiler from
519 	 * having to bcopy() this array onto the stack frame of
520 	 * SHA1Transform() each time it is called -- which is
521 	 * unacceptably expensive.
522 	 *
523 	 * the `const' is to ensure that callers are good citizens and
524 	 * do not try to munge the array.  since these routines are
525 	 * going to be called from inside multithreaded kernelland,
526 	 * this is a good safety check. -- `sha1_consts' will end up in
527 	 * .rodata.
528 	 *
529 	 * unfortunately, loading from an array in this manner hurts
530 	 * performance under Intel.  So, there is a macro,
531 	 * SHA1_CONST(), used in SHA1Transform(), that either expands to
532 	 * a reference to this array, or to the actual constant,
533 	 * depending on what platform this code is compiled for.
534 	 */
535 
536 	static const uint32_t sha1_consts[] = {
537 		SHA1_CONST_0, SHA1_CONST_1, SHA1_CONST_2, SHA1_CONST_3
538 	};
539 
540 	/*
541 	 * general optimization:
542 	 *
543 	 * use individual integers instead of using an array.  this is a
544 	 * win, although the amount it wins by seems to vary quite a bit.
545 	 */
546 
547 	uint32_t	w_0, w_1, w_2,  w_3,  w_4,  w_5,  w_6,  w_7;
548 	uint32_t	w_8, w_9, w_10, w_11, w_12, w_13, w_14, w_15;
549 
550 	/*
551 	 * sparc optimization:
552 	 *
553 	 * if `block' is already aligned on a 4-byte boundary, use
554 	 * LOAD_BIG_32() directly.  otherwise, bcopy() into a
555 	 * buffer that *is* aligned on a 4-byte boundary and then do
556 	 * the LOAD_BIG_32() on that buffer.  benchmarks have shown
557 	 * that using the bcopy() is better than loading the bytes
558 	 * individually and doing the endian-swap by hand.
559 	 *
560 	 * even though it's quite tempting to assign to do:
561 	 *
562 	 * blk = bcopy(ctx->buf_un.buf32, blk, sizeof (ctx->buf_un.buf32));
563 	 *
564 	 * and only have one set of LOAD_BIG_32()'s, the compiler
565 	 * *does not* like that, so please resist the urge.
566 	 */
567 
568 	if ((uintptr_t)blk & 0x3) {		/* not 4-byte aligned? */
569 		bcopy(blk, ctx->buf_un.buf32,  sizeof (ctx->buf_un.buf32));
570 		w_15 = LOAD_BIG_32(ctx->buf_un.buf32 + 15);
571 		w_14 = LOAD_BIG_32(ctx->buf_un.buf32 + 14);
572 		w_13 = LOAD_BIG_32(ctx->buf_un.buf32 + 13);
573 		w_12 = LOAD_BIG_32(ctx->buf_un.buf32 + 12);
574 		w_11 = LOAD_BIG_32(ctx->buf_un.buf32 + 11);
575 		w_10 = LOAD_BIG_32(ctx->buf_un.buf32 + 10);
576 		w_9  = LOAD_BIG_32(ctx->buf_un.buf32 +  9);
577 		w_8  = LOAD_BIG_32(ctx->buf_un.buf32 +  8);
578 		w_7  = LOAD_BIG_32(ctx->buf_un.buf32 +  7);
579 		w_6  = LOAD_BIG_32(ctx->buf_un.buf32 +  6);
580 		w_5  = LOAD_BIG_32(ctx->buf_un.buf32 +  5);
581 		w_4  = LOAD_BIG_32(ctx->buf_un.buf32 +  4);
582 		w_3  = LOAD_BIG_32(ctx->buf_un.buf32 +  3);
583 		w_2  = LOAD_BIG_32(ctx->buf_un.buf32 +  2);
584 		w_1  = LOAD_BIG_32(ctx->buf_un.buf32 +  1);
585 		w_0  = LOAD_BIG_32(ctx->buf_un.buf32 +  0);
586 	} else {
587 		/* LINTED E_BAD_PTR_CAST_ALIGN */
588 		w_15 = LOAD_BIG_32(blk + 60);
589 		/* LINTED E_BAD_PTR_CAST_ALIGN */
590 		w_14 = LOAD_BIG_32(blk + 56);
591 		/* LINTED E_BAD_PTR_CAST_ALIGN */
592 		w_13 = LOAD_BIG_32(blk + 52);
593 		/* LINTED E_BAD_PTR_CAST_ALIGN */
594 		w_12 = LOAD_BIG_32(blk + 48);
595 		/* LINTED E_BAD_PTR_CAST_ALIGN */
596 		w_11 = LOAD_BIG_32(blk + 44);
597 		/* LINTED E_BAD_PTR_CAST_ALIGN */
598 		w_10 = LOAD_BIG_32(blk + 40);
599 		/* LINTED E_BAD_PTR_CAST_ALIGN */
600 		w_9  = LOAD_BIG_32(blk + 36);
601 		/* LINTED E_BAD_PTR_CAST_ALIGN */
602 		w_8  = LOAD_BIG_32(blk + 32);
603 		/* LINTED E_BAD_PTR_CAST_ALIGN */
604 		w_7  = LOAD_BIG_32(blk + 28);
605 		/* LINTED E_BAD_PTR_CAST_ALIGN */
606 		w_6  = LOAD_BIG_32(blk + 24);
607 		/* LINTED E_BAD_PTR_CAST_ALIGN */
608 		w_5  = LOAD_BIG_32(blk + 20);
609 		/* LINTED E_BAD_PTR_CAST_ALIGN */
610 		w_4  = LOAD_BIG_32(blk + 16);
611 		/* LINTED E_BAD_PTR_CAST_ALIGN */
612 		w_3  = LOAD_BIG_32(blk + 12);
613 		/* LINTED E_BAD_PTR_CAST_ALIGN */
614 		w_2  = LOAD_BIG_32(blk +  8);
615 		/* LINTED E_BAD_PTR_CAST_ALIGN */
616 		w_1  = LOAD_BIG_32(blk +  4);
617 		/* LINTED E_BAD_PTR_CAST_ALIGN */
618 		w_0  = LOAD_BIG_32(blk +  0);
619 	}
620 #else	/* !defined(__sparc) */
621 
622 void /* CSTYLED */
623 SHA1Transform(SHA1_CTX *ctx, const uint8_t blk[64])
624 {
625 	/* CSTYLED */
626 	sha1word a = ctx->state[0];
627 	sha1word b = ctx->state[1];
628 	sha1word c = ctx->state[2];
629 	sha1word d = ctx->state[3];
630 	sha1word e = ctx->state[4];
631 
632 #if	defined(W_ARRAY)
633 	sha1word	w[16];
634 #else	/* !defined(W_ARRAY) */
635 	sha1word	w_0, w_1, w_2,  w_3,  w_4,  w_5,  w_6,  w_7;
636 	sha1word	w_8, w_9, w_10, w_11, w_12, w_13, w_14, w_15;
637 #endif	/* !defined(W_ARRAY) */
638 
639 	W(0)  = LOAD_BIG_32((void *)(blk +  0));
640 	W(1)  = LOAD_BIG_32((void *)(blk +  4));
641 	W(2)  = LOAD_BIG_32((void *)(blk +  8));
642 	W(3)  = LOAD_BIG_32((void *)(blk + 12));
643 	W(4)  = LOAD_BIG_32((void *)(blk + 16));
644 	W(5)  = LOAD_BIG_32((void *)(blk + 20));
645 	W(6)  = LOAD_BIG_32((void *)(blk + 24));
646 	W(7)  = LOAD_BIG_32((void *)(blk + 28));
647 	W(8)  = LOAD_BIG_32((void *)(blk + 32));
648 	W(9)  = LOAD_BIG_32((void *)(blk + 36));
649 	W(10) = LOAD_BIG_32((void *)(blk + 40));
650 	W(11) = LOAD_BIG_32((void *)(blk + 44));
651 	W(12) = LOAD_BIG_32((void *)(blk + 48));
652 	W(13) = LOAD_BIG_32((void *)(blk + 52));
653 	W(14) = LOAD_BIG_32((void *)(blk + 56));
654 	W(15) = LOAD_BIG_32((void *)(blk + 60));
655 
656 #endif	/* !defined(__sparc) */
657 
658 	/*
659 	 * general optimization:
660 	 *
661 	 * even though this approach is described in the standard as
662 	 * being slower algorithmically, it is 30-40% faster than the
663 	 * "faster" version under SPARC, because this version has more
664 	 * of the constraints specified at compile-time and uses fewer
665 	 * variables (and therefore has better register utilization)
666 	 * than its "speedier" brother.  (i've tried both, trust me)
667 	 *
668 	 * for either method given in the spec, there is an "assignment"
669 	 * phase where the following takes place:
670 	 *
671 	 *	tmp = (main_computation);
672 	 *	e = d; d = c; c = rotate_left(b, 30); b = a; a = tmp;
673 	 *
674 	 * we can make the algorithm go faster by not doing this work,
675 	 * but just pretending that `d' is now `e', etc. this works
676 	 * really well and obviates the need for a temporary variable.
677 	 * however, we still explicitly perform the rotate action,
678 	 * since it is cheaper on SPARC to do it once than to have to
679 	 * do it over and over again.
680 	 */
681 
682 	/* round 1 */
683 	e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(0) + SHA1_CONST(0); /* 0 */
684 	b = ROTATE_LEFT(b, 30);
685 
686 	d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(1) + SHA1_CONST(0); /* 1 */
687 	a = ROTATE_LEFT(a, 30);
688 
689 	c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(2) + SHA1_CONST(0); /* 2 */
690 	e = ROTATE_LEFT(e, 30);
691 
692 	b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(3) + SHA1_CONST(0); /* 3 */
693 	d = ROTATE_LEFT(d, 30);
694 
695 	a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(4) + SHA1_CONST(0); /* 4 */
696 	c = ROTATE_LEFT(c, 30);
697 
698 	e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(5) + SHA1_CONST(0); /* 5 */
699 	b = ROTATE_LEFT(b, 30);
700 
701 	d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(6) + SHA1_CONST(0); /* 6 */
702 	a = ROTATE_LEFT(a, 30);
703 
704 	c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(7) + SHA1_CONST(0); /* 7 */
705 	e = ROTATE_LEFT(e, 30);
706 
707 	b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(8) + SHA1_CONST(0); /* 8 */
708 	d = ROTATE_LEFT(d, 30);
709 
710 	a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(9) + SHA1_CONST(0); /* 9 */
711 	c = ROTATE_LEFT(c, 30);
712 
713 	e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(10) + SHA1_CONST(0); /* 10 */
714 	b = ROTATE_LEFT(b, 30);
715 
716 	d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(11) + SHA1_CONST(0); /* 11 */
717 	a = ROTATE_LEFT(a, 30);
718 
719 	c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(12) + SHA1_CONST(0); /* 12 */
720 	e = ROTATE_LEFT(e, 30);
721 
722 	b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(13) + SHA1_CONST(0); /* 13 */
723 	d = ROTATE_LEFT(d, 30);
724 
725 	a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(14) + SHA1_CONST(0); /* 14 */
726 	c = ROTATE_LEFT(c, 30);
727 
728 	e = ROTATE_LEFT(a, 5) + F(b, c, d) + e + W(15) + SHA1_CONST(0); /* 15 */
729 	b = ROTATE_LEFT(b, 30);
730 
731 	W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1);		/* 16 */
732 	d = ROTATE_LEFT(e, 5) + F(a, b, c) + d + W(0) + SHA1_CONST(0);
733 	a = ROTATE_LEFT(a, 30);
734 
735 	W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1);		/* 17 */
736 	c = ROTATE_LEFT(d, 5) + F(e, a, b) + c + W(1) + SHA1_CONST(0);
737 	e = ROTATE_LEFT(e, 30);
738 
739 	W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1);	/* 18 */
740 	b = ROTATE_LEFT(c, 5) + F(d, e, a) + b + W(2) + SHA1_CONST(0);
741 	d = ROTATE_LEFT(d, 30);
742 
743 	W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1);		/* 19 */
744 	a = ROTATE_LEFT(b, 5) + F(c, d, e) + a + W(3) + SHA1_CONST(0);
745 	c = ROTATE_LEFT(c, 30);
746 
747 	/* round 2 */
748 	W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1);		/* 20 */
749 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(4) + SHA1_CONST(1);
750 	b = ROTATE_LEFT(b, 30);
751 
752 	W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1);		/* 21 */
753 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(5) + SHA1_CONST(1);
754 	a = ROTATE_LEFT(a, 30);
755 
756 	W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1);		/* 22 */
757 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(6) + SHA1_CONST(1);
758 	e = ROTATE_LEFT(e, 30);
759 
760 	W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1);		/* 23 */
761 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(7) + SHA1_CONST(1);
762 	d = ROTATE_LEFT(d, 30);
763 
764 	W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1);		/* 24 */
765 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(8) + SHA1_CONST(1);
766 	c = ROTATE_LEFT(c, 30);
767 
768 	W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1);		/* 25 */
769 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(9) + SHA1_CONST(1);
770 	b = ROTATE_LEFT(b, 30);
771 
772 	W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1);	/* 26 */
773 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(10) + SHA1_CONST(1);
774 	a = ROTATE_LEFT(a, 30);
775 
776 	W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1);	/* 27 */
777 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(11) + SHA1_CONST(1);
778 	e = ROTATE_LEFT(e, 30);
779 
780 	W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1);	/* 28 */
781 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(12) + SHA1_CONST(1);
782 	d = ROTATE_LEFT(d, 30);
783 
784 	W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1);	/* 29 */
785 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(13) + SHA1_CONST(1);
786 	c = ROTATE_LEFT(c, 30);
787 
788 	W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1);	/* 30 */
789 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(14) + SHA1_CONST(1);
790 	b = ROTATE_LEFT(b, 30);
791 
792 	W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1);	/* 31 */
793 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(15) + SHA1_CONST(1);
794 	a = ROTATE_LEFT(a, 30);
795 
796 	W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1);		/* 32 */
797 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(0) + SHA1_CONST(1);
798 	e = ROTATE_LEFT(e, 30);
799 
800 	W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1);		/* 33 */
801 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(1) + SHA1_CONST(1);
802 	d = ROTATE_LEFT(d, 30);
803 
804 	W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1);	/* 34 */
805 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(2) + SHA1_CONST(1);
806 	c = ROTATE_LEFT(c, 30);
807 
808 	W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1);		/* 35 */
809 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(3) + SHA1_CONST(1);
810 	b = ROTATE_LEFT(b, 30);
811 
812 	W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1);		/* 36 */
813 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(4) + SHA1_CONST(1);
814 	a = ROTATE_LEFT(a, 30);
815 
816 	W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1);		/* 37 */
817 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(5) + SHA1_CONST(1);
818 	e = ROTATE_LEFT(e, 30);
819 
820 	W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1);		/* 38 */
821 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(6) + SHA1_CONST(1);
822 	d = ROTATE_LEFT(d, 30);
823 
824 	W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1);		/* 39 */
825 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(7) + SHA1_CONST(1);
826 	c = ROTATE_LEFT(c, 30);
827 
828 	/* round 3 */
829 	W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1);		/* 40 */
830 	e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(8) + SHA1_CONST(2);
831 	b = ROTATE_LEFT(b, 30);
832 
833 	W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1);		/* 41 */
834 	d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(9) + SHA1_CONST(2);
835 	a = ROTATE_LEFT(a, 30);
836 
837 	W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1);	/* 42 */
838 	c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(10) + SHA1_CONST(2);
839 	e = ROTATE_LEFT(e, 30);
840 
841 	W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1);	/* 43 */
842 	b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(11) + SHA1_CONST(2);
843 	d = ROTATE_LEFT(d, 30);
844 
845 	W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1);	/* 44 */
846 	a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(12) + SHA1_CONST(2);
847 	c = ROTATE_LEFT(c, 30);
848 
849 	W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1);	/* 45 */
850 	e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(13) + SHA1_CONST(2);
851 	b = ROTATE_LEFT(b, 30);
852 
853 	W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1);	/* 46 */
854 	d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(14) + SHA1_CONST(2);
855 	a = ROTATE_LEFT(a, 30);
856 
857 	W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1);	/* 47 */
858 	c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(15) + SHA1_CONST(2);
859 	e = ROTATE_LEFT(e, 30);
860 
861 	W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1);		/* 48 */
862 	b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(0) + SHA1_CONST(2);
863 	d = ROTATE_LEFT(d, 30);
864 
865 	W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1);		/* 49 */
866 	a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(1) + SHA1_CONST(2);
867 	c = ROTATE_LEFT(c, 30);
868 
869 	W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1);	/* 50 */
870 	e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(2) + SHA1_CONST(2);
871 	b = ROTATE_LEFT(b, 30);
872 
873 	W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1);		/* 51 */
874 	d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(3) + SHA1_CONST(2);
875 	a = ROTATE_LEFT(a, 30);
876 
877 	W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1);		/* 52 */
878 	c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(4) + SHA1_CONST(2);
879 	e = ROTATE_LEFT(e, 30);
880 
881 	W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1);		/* 53 */
882 	b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(5) + SHA1_CONST(2);
883 	d = ROTATE_LEFT(d, 30);
884 
885 	W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1);		/* 54 */
886 	a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(6) + SHA1_CONST(2);
887 	c = ROTATE_LEFT(c, 30);
888 
889 	W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1);		/* 55 */
890 	e = ROTATE_LEFT(a, 5) + H(b, c, d) + e + W(7) + SHA1_CONST(2);
891 	b = ROTATE_LEFT(b, 30);
892 
893 	W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1);		/* 56 */
894 	d = ROTATE_LEFT(e, 5) + H(a, b, c) + d + W(8) + SHA1_CONST(2);
895 	a = ROTATE_LEFT(a, 30);
896 
897 	W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1);		/* 57 */
898 	c = ROTATE_LEFT(d, 5) + H(e, a, b) + c + W(9) + SHA1_CONST(2);
899 	e = ROTATE_LEFT(e, 30);
900 
901 	W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1);	/* 58 */
902 	b = ROTATE_LEFT(c, 5) + H(d, e, a) + b + W(10) + SHA1_CONST(2);
903 	d = ROTATE_LEFT(d, 30);
904 
905 	W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1);	/* 59 */
906 	a = ROTATE_LEFT(b, 5) + H(c, d, e) + a + W(11) + SHA1_CONST(2);
907 	c = ROTATE_LEFT(c, 30);
908 
909 	/* round 4 */
910 	W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1);	/* 60 */
911 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(12) + SHA1_CONST(3);
912 	b = ROTATE_LEFT(b, 30);
913 
914 	W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1);	/* 61 */
915 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(13) + SHA1_CONST(3);
916 	a = ROTATE_LEFT(a, 30);
917 
918 	W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1);	/* 62 */
919 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(14) + SHA1_CONST(3);
920 	e = ROTATE_LEFT(e, 30);
921 
922 	W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1);	/* 63 */
923 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(15) + SHA1_CONST(3);
924 	d = ROTATE_LEFT(d, 30);
925 
926 	W(0) = ROTATE_LEFT((W(13) ^ W(8) ^ W(2) ^ W(0)), 1);		/* 64 */
927 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(0) + SHA1_CONST(3);
928 	c = ROTATE_LEFT(c, 30);
929 
930 	W(1) = ROTATE_LEFT((W(14) ^ W(9) ^ W(3) ^ W(1)), 1);		/* 65 */
931 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(1) + SHA1_CONST(3);
932 	b = ROTATE_LEFT(b, 30);
933 
934 	W(2) = ROTATE_LEFT((W(15) ^ W(10) ^ W(4) ^ W(2)), 1);	/* 66 */
935 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(2) + SHA1_CONST(3);
936 	a = ROTATE_LEFT(a, 30);
937 
938 	W(3) = ROTATE_LEFT((W(0) ^ W(11) ^ W(5) ^ W(3)), 1);		/* 67 */
939 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(3) + SHA1_CONST(3);
940 	e = ROTATE_LEFT(e, 30);
941 
942 	W(4) = ROTATE_LEFT((W(1) ^ W(12) ^ W(6) ^ W(4)), 1);		/* 68 */
943 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(4) + SHA1_CONST(3);
944 	d = ROTATE_LEFT(d, 30);
945 
946 	W(5) = ROTATE_LEFT((W(2) ^ W(13) ^ W(7) ^ W(5)), 1);		/* 69 */
947 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(5) + SHA1_CONST(3);
948 	c = ROTATE_LEFT(c, 30);
949 
950 	W(6) = ROTATE_LEFT((W(3) ^ W(14) ^ W(8) ^ W(6)), 1);		/* 70 */
951 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(6) + SHA1_CONST(3);
952 	b = ROTATE_LEFT(b, 30);
953 
954 	W(7) = ROTATE_LEFT((W(4) ^ W(15) ^ W(9) ^ W(7)), 1);		/* 71 */
955 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(7) + SHA1_CONST(3);
956 	a = ROTATE_LEFT(a, 30);
957 
958 	W(8) = ROTATE_LEFT((W(5) ^ W(0) ^ W(10) ^ W(8)), 1);		/* 72 */
959 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(8) + SHA1_CONST(3);
960 	e = ROTATE_LEFT(e, 30);
961 
962 	W(9) = ROTATE_LEFT((W(6) ^ W(1) ^ W(11) ^ W(9)), 1);		/* 73 */
963 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(9) + SHA1_CONST(3);
964 	d = ROTATE_LEFT(d, 30);
965 
966 	W(10) = ROTATE_LEFT((W(7) ^ W(2) ^ W(12) ^ W(10)), 1);	/* 74 */
967 	a = ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(10) + SHA1_CONST(3);
968 	c = ROTATE_LEFT(c, 30);
969 
970 	W(11) = ROTATE_LEFT((W(8) ^ W(3) ^ W(13) ^ W(11)), 1);	/* 75 */
971 	e = ROTATE_LEFT(a, 5) + G(b, c, d) + e + W(11) + SHA1_CONST(3);
972 	b = ROTATE_LEFT(b, 30);
973 
974 	W(12) = ROTATE_LEFT((W(9) ^ W(4) ^ W(14) ^ W(12)), 1);	/* 76 */
975 	d = ROTATE_LEFT(e, 5) + G(a, b, c) + d + W(12) + SHA1_CONST(3);
976 	a = ROTATE_LEFT(a, 30);
977 
978 	W(13) = ROTATE_LEFT((W(10) ^ W(5) ^ W(15) ^ W(13)), 1);	/* 77 */
979 	c = ROTATE_LEFT(d, 5) + G(e, a, b) + c + W(13) + SHA1_CONST(3);
980 	e = ROTATE_LEFT(e, 30);
981 
982 	W(14) = ROTATE_LEFT((W(11) ^ W(6) ^ W(0) ^ W(14)), 1);	/* 78 */
983 	b = ROTATE_LEFT(c, 5) + G(d, e, a) + b + W(14) + SHA1_CONST(3);
984 	d = ROTATE_LEFT(d, 30);
985 
986 	W(15) = ROTATE_LEFT((W(12) ^ W(7) ^ W(1) ^ W(15)), 1);	/* 79 */
987 
988 	ctx->state[0] += ROTATE_LEFT(b, 5) + G(c, d, e) + a + W(15) +
989 	    SHA1_CONST(3);
990 	ctx->state[1] += b;
991 	ctx->state[2] += ROTATE_LEFT(c, 30);
992 	ctx->state[3] += d;
993 	ctx->state[4] += e;
994 
995 	/* zeroize sensitive information */
996 	W(0) = W(1) = W(2) = W(3) = W(4) = W(5) = W(6) = W(7) = W(8) = 0;
997 	W(9) = W(10) = W(11) = W(12) = W(13) = W(14) = W(15) = 0;
998 }
999 #endif	/* !__amd64 */
1000 
1001 
1002 /*
1003  * Encode()
1004  *
1005  * purpose: to convert a list of numbers from little endian to big endian
1006  *   input: uint8_t *	: place to store the converted big endian numbers
1007  *	    uint32_t *	: place to get numbers to convert from
1008  *          size_t	: the length of the input in bytes
1009  *  output: void
1010  */
1011 
1012 static void
1013 Encode(uint8_t *_RESTRICT_KYWD output, const uint32_t *_RESTRICT_KYWD input,
1014     size_t len)
1015 {
1016 	size_t		i, j;
1017 
1018 #if	defined(__sparc)
1019 	if (IS_P2ALIGNED(output, sizeof (uint32_t))) {
1020 		for (i = 0, j = 0; j < len; i++, j += 4) {
1021 			/* LINTED E_BAD_PTR_CAST_ALIGN */
1022 			*((uint32_t *)(output + j)) = input[i];
1023 		}
1024 	} else {
1025 #endif	/* little endian -- will work on big endian, but slowly */
1026 		for (i = 0, j = 0; j < len; i++, j += 4) {
1027 			output[j]	= (input[i] >> 24) & 0xff;
1028 			output[j + 1]	= (input[i] >> 16) & 0xff;
1029 			output[j + 2]	= (input[i] >>  8) & 0xff;
1030 			output[j + 3]	= input[i] & 0xff;
1031 		}
1032 #if	defined(__sparc)
1033 	}
1034 #endif
1035 }
1036