3ed16f380fa03d0abea6796bc91f4e6f7a69e6bf
[openssl.git] / crypto / md32_common.h
1 /* crypto/md32_common.h */
2 /* ====================================================================
3  * Copyright (c) 1999-2002 The OpenSSL Project.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  *
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer. 
11  *
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in
14  *    the documentation and/or other materials provided with the
15  *    distribution.
16  *
17  * 3. All advertising materials mentioning features or use of this
18  *    software must display the following acknowledgment:
19  *    "This product includes software developed by the OpenSSL Project
20  *    for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
21  *
22  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
23  *    endorse or promote products derived from this software without
24  *    prior written permission. For written permission, please contact
25  *    licensing@OpenSSL.org.
26  *
27  * 5. Products derived from this software may not be called "OpenSSL"
28  *    nor may "OpenSSL" appear in their names without prior written
29  *    permission of the OpenSSL Project.
30  *
31  * 6. Redistributions of any form whatsoever must retain the following
32  *    acknowledgment:
33  *    "This product includes software developed by the OpenSSL Project
34  *    for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
35  *
36  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
37  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
39  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
40  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
41  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
42  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
43  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
45  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
46  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
47  * OF THE POSSIBILITY OF SUCH DAMAGE.
48  * ====================================================================
49  *
50  * This product includes cryptographic software written by Eric Young
51  * (eay@cryptsoft.com).  This product includes software written by Tim
52  * Hudson (tjh@cryptsoft.com).
53  *
54  */
55
56 /*
57  * This is a generic 32 bit "collector" for message digest algorithms.
58  * Whenever needed it collects input character stream into chunks of
59  * 32 bit values and invokes a block function that performs actual hash
60  * calculations.
61  *
62  * Porting guide.
63  *
64  * Obligatory macros:
65  *
66  * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
67  *      this macro defines byte order of input stream.
68  * HASH_CBLOCK
69  *      size of a unit chunk HASH_BLOCK operates on.
70  * HASH_LONG
71  *      has to be at lest 32 bit wide, if it's wider, then
72  *      HASH_LONG_LOG2 *has to* be defined along
73  * HASH_CTX
74  *      context structure that at least contains following
75  *      members:
76  *              typedef struct {
77  *                      ...
78  *                      HASH_LONG       Nl,Nh;
79  *                      HASH_LONG       data[HASH_LBLOCK];
80  *                      unsigned int    num;
81  *                      ...
82  *                      } HASH_CTX;
83  * HASH_UPDATE
84  *      name of "Update" function, implemented here.
85  * HASH_TRANSFORM
86  *      name of "Transform" function, implemented here.
87  * HASH_FINAL
88  *      name of "Final" function, implemented here.
89  * HASH_BLOCK_HOST_ORDER
90  *      name of "block" function treating *aligned* input message
91  *      in host byte order, implemented externally.
92  * HASH_BLOCK_DATA_ORDER
93  *      name of "block" function treating *unaligned* input message
94  *      in original (data) byte order, implemented externally (it
95  *      actually is optional if data and host are of the same
96  *      "endianess").
97  * HASH_MAKE_STRING
98  *      macro convering context variables to an ASCII hash string.
99  *
100  * Optional macros:
101  *
102  * B_ENDIAN or L_ENDIAN
103  *      defines host byte-order.
104  * HASH_LONG_LOG2
105  *      defaults to 2 if not states otherwise.
106  * HASH_LBLOCK
107  *      assumed to be HASH_CBLOCK/4 if not stated otherwise.
108  * HASH_BLOCK_DATA_ORDER_ALIGNED
109  *      alternative "block" function capable of treating
110  *      aligned input message in original (data) order,
111  *      implemented externally.
112  *
113  * MD5 example:
114  *
115  *      #define DATA_ORDER_IS_LITTLE_ENDIAN
116  *
117  *      #define HASH_LONG               MD5_LONG
118  *      #define HASH_LONG_LOG2          MD5_LONG_LOG2
119  *      #define HASH_CTX                MD5_CTX
120  *      #define HASH_CBLOCK             MD5_CBLOCK
121  *      #define HASH_LBLOCK             MD5_LBLOCK
122  *      #define HASH_UPDATE             MD5_Update
123  *      #define HASH_TRANSFORM          MD5_Transform
124  *      #define HASH_FINAL              MD5_Final
125  *      #define HASH_BLOCK_HOST_ORDER   md5_block_host_order
126  *      #define HASH_BLOCK_DATA_ORDER   md5_block_data_order
127  *
128  *                                      <appro@fy.chalmers.se>
129  */
130
131 #if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
132 #error "DATA_ORDER must be defined!"
133 #endif
134
135 #ifndef HASH_CBLOCK
136 #error "HASH_CBLOCK must be defined!"
137 #endif
138 #ifndef HASH_LONG
139 #error "HASH_LONG must be defined!"
140 #endif
141 #ifndef HASH_CTX
142 #error "HASH_CTX must be defined!"
143 #endif
144
145 #ifndef HASH_UPDATE
146 #error "HASH_UPDATE must be defined!"
147 #endif
148 #ifndef HASH_TRANSFORM
149 #error "HASH_TRANSFORM must be defined!"
150 #endif
151 #ifndef HASH_FINAL
152 #error "HASH_FINAL must be defined!"
153 #endif
154
155 #ifndef HASH_BLOCK_HOST_ORDER
156 #error "HASH_BLOCK_HOST_ORDER must be defined!"
157 #endif
158
159 #if 0
160 /*
161  * Moved below as it's required only if HASH_BLOCK_DATA_ORDER_ALIGNED
162  * isn't defined.
163  */
164 #ifndef HASH_BLOCK_DATA_ORDER
165 #error "HASH_BLOCK_DATA_ORDER must be defined!"
166 #endif
167 #endif
168
169 #ifndef HASH_LBLOCK
170 #define HASH_LBLOCK     (HASH_CBLOCK/4)
171 #endif
172
173 #ifndef HASH_LONG_LOG2
174 #define HASH_LONG_LOG2  2
175 #endif
176
177 /*
178  * Engage compiler specific rotate intrinsic function if available.
179  */
180 #undef ROTATE
181 #ifndef PEDANTIC
182 # if defined(_MSC_VER) || defined(__ICC)
183 #  define ROTATE(a,n)   _lrotl(a,n)
184 # elif defined(__MWERKS__)
185 #  if defined(__POWERPC__)
186 #   define ROTATE(a,n)  __rlwinm(a,n,0,31)
187 #  elif defined(__MC68K__)
188     /* Motorola specific tweak. <appro@fy.chalmers.se> */
189 #   define ROTATE(a,n)  ( n<24 ? __rol(a,n) : __ror(a,32-n) )
190 #  else
191 #   define ROTATE(a,n)  __rol(a,n)
192 #  endif
193 # elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
194   /*
195    * Some GNU C inline assembler templates. Note that these are
196    * rotates by *constant* number of bits! But that's exactly
197    * what we need here...
198    *                                    <appro@fy.chalmers.se>
199    */
200 #  if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
201 #   define ROTATE(a,n)  ({ register unsigned int ret;   \
202                                 asm (                   \
203                                 "roll %1,%0"            \
204                                 : "=r"(ret)             \
205                                 : "I"(n), "0"(a)        \
206                                 : "cc");                \
207                            ret;                         \
208                         })
209 #  elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \
210         defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__)
211 #   define ROTATE(a,n)  ({ register unsigned int ret;   \
212                                 asm (                   \
213                                 "rlwinm %0,%1,%2,0,31"  \
214                                 : "=r"(ret)             \
215                                 : "r"(a), "I"(n));      \
216                            ret;                         \
217                         })
218 #  endif
219 # endif
220 #endif /* PEDANTIC */
221
222 #if HASH_LONG_LOG2==2   /* Engage only if sizeof(HASH_LONG)== 4 */
223 /* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */
224 #ifdef ROTATE
225 /* 5 instructions with rotate instruction, else 9 */
226 #define REVERSE_FETCH32(a,l)    (                                       \
227                 l=*(const HASH_LONG *)(a),                              \
228                 ((ROTATE(l,8)&0x00FF00FF)|(ROTATE((l&0x00FF00FF),24)))  \
229                                 )
230 #else
231 /* 6 instructions with rotate instruction, else 8 */
232 #define REVERSE_FETCH32(a,l)    (                               \
233                 l=*(const HASH_LONG *)(a),                      \
234                 l=(((l>>8)&0x00FF00FF)|((l&0x00FF00FF)<<8)),    \
235                 ROTATE(l,16)                                    \
236                                 )
237 /*
238  * Originally the middle line started with l=(((l&0xFF00FF00)>>8)|...
239  * It's rewritten as above for two reasons:
240  *      - RISCs aren't good at long constants and have to explicitely
241  *        compose 'em with several (well, usually 2) instructions in a
242  *        register before performing the actual operation and (as you
243  *        already realized:-) having same constant should inspire the
244  *        compiler to permanently allocate the only register for it;
245  *      - most modern CPUs have two ALUs, but usually only one has
246  *        circuitry for shifts:-( this minor tweak inspires compiler
247  *        to schedule shift instructions in a better way...
248  *
249  *                              <appro@fy.chalmers.se>
250  */
251 #endif
252 #endif
253
254 #ifndef ROTATE
255 #define ROTATE(a,n)     (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
256 #endif
257
258 /*
259  * Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED
260  * and HASH_BLOCK_HOST_ORDER ought to be the same if input data
261  * and host are of the same "endianess". It's possible to mask
262  * this with blank #define HASH_BLOCK_DATA_ORDER though...
263  *
264  *                              <appro@fy.chalmers.se>
265  */
266 #if defined(B_ENDIAN)
267 #  if defined(DATA_ORDER_IS_BIG_ENDIAN)
268 #    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
269 #      define HASH_BLOCK_DATA_ORDER_ALIGNED     HASH_BLOCK_HOST_ORDER
270 #    endif
271 #  endif
272 #elif defined(L_ENDIAN)
273 #  if defined(DATA_ORDER_IS_LITTLE_ENDIAN)
274 #    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
275 #      define HASH_BLOCK_DATA_ORDER_ALIGNED     HASH_BLOCK_HOST_ORDER
276 #    endif
277 #  endif
278 #endif
279
280 #if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
281 #ifndef HASH_BLOCK_DATA_ORDER
282 #error "HASH_BLOCK_DATA_ORDER must be defined!"
283 #endif
284 #endif
285
286 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
287
288 #ifndef PEDANTIC
289 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
290 #  if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \
291       (defined(__x86_64) || defined(__x86_64__))
292 #   if !defined(B_ENDIAN)
293     /*
294      * This gives ~30-40% performance improvement in SHA-256 compiled
295      * with gcc [on P4]. Well, first macro to be frank. We can pull
296      * this trick on x86* platforms only, because these CPUs can fetch
297      * unaligned data without raising an exception.
298      */
299 #   define HOST_c2l(c,l)        ({ unsigned int r=*((const unsigned int *)(c)); \
300                                    asm ("bswapl %0":"=r"(r):"0"(r));    \
301                                    (c)+=4; (l)=r;                       })
302 #   define HOST_l2c(l,c)        ({ unsigned int r=(l);                  \
303                                    asm ("bswapl %0":"=r"(r):"0"(r));    \
304                                    *((unsigned int *)(c))=r; (c)+=4; r; })
305 #   endif
306 #  endif
307 # endif
308 #endif
309
310 #ifndef HOST_c2l
311 #define HOST_c2l(c,l)   (l =(((unsigned long)(*((c)++)))<<24),          \
312                          l|=(((unsigned long)(*((c)++)))<<16),          \
313                          l|=(((unsigned long)(*((c)++)))<< 8),          \
314                          l|=(((unsigned long)(*((c)++)))    ),          \
315                          l)
316 #endif
317 #define HOST_p_c2l(c,l,n)       {                                       \
318                         switch (n) {                                    \
319                         case 0: l =((unsigned long)(*((c)++)))<<24;     \
320                         case 1: l|=((unsigned long)(*((c)++)))<<16;     \
321                         case 2: l|=((unsigned long)(*((c)++)))<< 8;     \
322                         case 3: l|=((unsigned long)(*((c)++)));         \
323                                 } }
324 #define HOST_p_c2l_p(c,l,sc,len) {                                      \
325                         switch (sc) {                                   \
326                         case 0: l =((unsigned long)(*((c)++)))<<24;     \
327                                 if (--len == 0) break;                  \
328                         case 1: l|=((unsigned long)(*((c)++)))<<16;     \
329                                 if (--len == 0) break;                  \
330                         case 2: l|=((unsigned long)(*((c)++)))<< 8;     \
331                                 } }
332 /* NOTE the pointer is not incremented at the end of this */
333 #define HOST_c2l_p(c,l,n)       {                                       \
334                         l=0; (c)+=n;                                    \
335                         switch (n) {                                    \
336                         case 3: l =((unsigned long)(*(--(c))))<< 8;     \
337                         case 2: l|=((unsigned long)(*(--(c))))<<16;     \
338                         case 1: l|=((unsigned long)(*(--(c))))<<24;     \
339                                 } }
340 #ifndef HOST_l2c
341 #define HOST_l2c(l,c)   (*((c)++)=(unsigned char)(((l)>>24)&0xff),      \
342                          *((c)++)=(unsigned char)(((l)>>16)&0xff),      \
343                          *((c)++)=(unsigned char)(((l)>> 8)&0xff),      \
344                          *((c)++)=(unsigned char)(((l)    )&0xff),      \
345                          l)
346 #endif
347
348 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
349
350 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
351 # ifndef B_ENDIAN
352    /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
353 #  define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l)
354 #  define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l)
355 # endif
356 #endif
357
358 #ifndef HOST_c2l
359 #define HOST_c2l(c,l)   (l =(((unsigned long)(*((c)++)))    ),          \
360                          l|=(((unsigned long)(*((c)++)))<< 8),          \
361                          l|=(((unsigned long)(*((c)++)))<<16),          \
362                          l|=(((unsigned long)(*((c)++)))<<24),          \
363                          l)
364 #endif
365 #define HOST_p_c2l(c,l,n)       {                                       \
366                         switch (n) {                                    \
367                         case 0: l =((unsigned long)(*((c)++)));         \
368                         case 1: l|=((unsigned long)(*((c)++)))<< 8;     \
369                         case 2: l|=((unsigned long)(*((c)++)))<<16;     \
370                         case 3: l|=((unsigned long)(*((c)++)))<<24;     \
371                                 } }
372 #define HOST_p_c2l_p(c,l,sc,len) {                                      \
373                         switch (sc) {                                   \
374                         case 0: l =((unsigned long)(*((c)++)));         \
375                                 if (--len == 0) break;                  \
376                         case 1: l|=((unsigned long)(*((c)++)))<< 8;     \
377                                 if (--len == 0) break;                  \
378                         case 2: l|=((unsigned long)(*((c)++)))<<16;     \
379                                 } }
380 /* NOTE the pointer is not incremented at the end of this */
381 #define HOST_c2l_p(c,l,n)       {                                       \
382                         l=0; (c)+=n;                                    \
383                         switch (n) {                                    \
384                         case 3: l =((unsigned long)(*(--(c))))<<16;     \
385                         case 2: l|=((unsigned long)(*(--(c))))<< 8;     \
386                         case 1: l|=((unsigned long)(*(--(c))));         \
387                                 } }
388 #ifndef HOST_l2c
389 #define HOST_l2c(l,c)   (*((c)++)=(unsigned char)(((l)    )&0xff),      \
390                          *((c)++)=(unsigned char)(((l)>> 8)&0xff),      \
391                          *((c)++)=(unsigned char)(((l)>>16)&0xff),      \
392                          *((c)++)=(unsigned char)(((l)>>24)&0xff),      \
393                          l)
394 #endif
395
396 #endif
397
398 /*
399  * Time for some action:-)
400  */
401
402 int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len)
403         {
404         const unsigned char *data=data_;
405         register HASH_LONG * p;
406         register HASH_LONG l;
407         size_t sw,sc,ew,ec;
408
409         if (len==0) return 1;
410
411         l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL;
412         /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
413          * Wei Dai <weidai@eskimo.com> for pointing it out. */
414         if (l < c->Nl) /* overflow */
415                 c->Nh++;
416         c->Nh+=(len>>29);       /* might cause compiler warning on 16-bit */
417         c->Nl=l;
418
419         if (c->num != 0)
420                 {
421                 p=c->data;
422                 sw=c->num>>2;
423                 sc=c->num&0x03;
424
425                 if ((c->num+len) >= HASH_CBLOCK)
426                         {
427                         l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l;
428                         for (; sw<HASH_LBLOCK; sw++)
429                                 {
430                                 HOST_c2l(data,l); p[sw]=l;
431                                 }
432                         HASH_BLOCK_HOST_ORDER (c,p,1);
433                         len-=(HASH_CBLOCK-c->num);
434                         c->num=0;
435                         /* drop through and do the rest */
436                         }
437                 else
438                         {
439                         c->num+=(unsigned int)len;
440                         if ((sc+len) < 4) /* ugly, add char's to a word */
441                                 {
442                                 l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l;
443                                 }
444                         else
445                                 {
446                                 ew=(c->num>>2);
447                                 ec=(c->num&0x03);
448                                 if (sc)
449                                         l=p[sw];
450                                 HOST_p_c2l(data,l,sc);
451                                 p[sw++]=l;
452                                 for (; sw < ew; sw++)
453                                         {
454                                         HOST_c2l(data,l); p[sw]=l;
455                                         }
456                                 if (ec)
457                                         {
458                                         HOST_c2l_p(data,l,ec); p[sw]=l;
459                                         }
460                                 }
461                         return 1;
462                         }
463                 }
464
465         sw=len/HASH_CBLOCK;
466         if (sw > 0)
467                 {
468 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
469                 /*
470                  * Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined
471                  * only if sizeof(HASH_LONG)==4.
472                  */
473                 if ((((size_t)data)%4) == 0)
474                         {
475                         /* data is properly aligned so that we can cast it: */
476                         HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,sw);
477                         sw*=HASH_CBLOCK;
478                         data+=sw;
479                         len-=sw;
480                         }
481                 else
482 #if !defined(HASH_BLOCK_DATA_ORDER)
483                         while (sw--)
484                                 {
485                                 memcpy (p=c->data,data,HASH_CBLOCK);
486                                 HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1);
487                                 data+=HASH_CBLOCK;
488                                 len-=HASH_CBLOCK;
489                                 }
490 #endif
491 #endif
492 #if defined(HASH_BLOCK_DATA_ORDER)
493                         {
494                         HASH_BLOCK_DATA_ORDER(c,data,sw);
495                         sw*=HASH_CBLOCK;
496                         data+=sw;
497                         len-=sw;
498                         }
499 #endif
500                 }
501
502         if (len!=0)
503                 {
504                 p = c->data;
505                 c->num = len;
506                 ew=len>>2;      /* words to copy */
507                 ec=len&0x03;
508                 for (; ew; ew--,p++)
509                         {
510                         HOST_c2l(data,l); *p=l;
511                         }
512                 HOST_c2l_p(data,l,ec);
513                 *p=l;
514                 }
515         return 1;
516         }
517
518
519 void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data)
520         {
521 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
522         if ((((size_t)data)%4) == 0)
523                 /* data is properly aligned so that we can cast it: */
524                 HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,1);
525         else
526 #if !defined(HASH_BLOCK_DATA_ORDER)
527                 {
528                 memcpy (c->data,data,HASH_CBLOCK);
529                 HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1);
530                 }
531 #endif
532 #endif
533 #if defined(HASH_BLOCK_DATA_ORDER)
534         HASH_BLOCK_DATA_ORDER (c,data,1);
535 #endif
536         }
537
538
539 int HASH_FINAL (unsigned char *md, HASH_CTX *c)
540         {
541         register HASH_LONG *p;
542         register unsigned long l;
543         register int i,j;
544         static const unsigned char end[4]={0x80,0x00,0x00,0x00};
545         const unsigned char *cp=end;
546
547         /* c->num should definitly have room for at least one more byte. */
548         p=c->data;
549         i=c->num>>2;
550         j=c->num&0x03;
551
552 #if 0
553         /* purify often complains about the following line as an
554          * Uninitialized Memory Read.  While this can be true, the
555          * following p_c2l macro will reset l when that case is true.
556          * This is because j&0x03 contains the number of 'valid' bytes
557          * already in p[i].  If and only if j&0x03 == 0, the UMR will
558          * occur but this is also the only time p_c2l will do
559          * l= *(cp++) instead of l|= *(cp++)
560          * Many thanks to Alex Tang <altitude@cic.net> for pickup this
561          * 'potential bug' */
562 #ifdef PURIFY
563         if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */
564 #endif
565         l=p[i];
566 #else
567         l = (j==0) ? 0 : p[i];
568 #endif
569         HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */
570
571         if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */
572                 {
573                 if (i<HASH_LBLOCK) p[i]=0;
574                 HASH_BLOCK_HOST_ORDER (c,p,1);
575                 i=0;
576                 }
577         for (; i<(HASH_LBLOCK-2); i++)
578                 p[i]=0;
579
580 #if   defined(DATA_ORDER_IS_BIG_ENDIAN)
581         p[HASH_LBLOCK-2]=c->Nh;
582         p[HASH_LBLOCK-1]=c->Nl;
583 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
584         p[HASH_LBLOCK-2]=c->Nl;
585         p[HASH_LBLOCK-1]=c->Nh;
586 #endif
587         HASH_BLOCK_HOST_ORDER (c,p,1);
588
589 #ifndef HASH_MAKE_STRING
590 #error "HASH_MAKE_STRING must be defined!"
591 #else
592         HASH_MAKE_STRING(c,md);
593 #endif
594
595         c->num=0;
596         /* clear stuff, HASH_BLOCK may be leaving some stuff on the stack
597          * but I'm not worried :-)
598         OPENSSL_cleanse((void *)c,sizeof(HASH_CTX));
599          */
600         return 1;
601         }
602
603 #ifndef MD32_REG_T
604 #define MD32_REG_T long
605 /*
606  * This comment was originaly written for MD5, which is why it
607  * discusses A-D. But it basically applies to all 32-bit digests,
608  * which is why it was moved to common header file.
609  *
610  * In case you wonder why A-D are declared as long and not
611  * as MD5_LONG. Doing so results in slight performance
612  * boost on LP64 architectures. The catch is we don't
613  * really care if 32 MSBs of a 64-bit register get polluted
614  * with eventual overflows as we *save* only 32 LSBs in
615  * *either* case. Now declaring 'em long excuses the compiler
616  * from keeping 32 MSBs zeroed resulting in 13% performance
617  * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
618  * Well, to be honest it should say that this *prevents* 
619  * performance degradation.
620  *                              <appro@fy.chalmers.se>
621  * Apparently there're LP64 compilers that generate better
622  * code if A-D are declared int. Most notably GCC-x86_64
623  * generates better code.
624  *                              <appro@fy.chalmers.se>
625  */
626 #endif