Brought forward from 0.9.8 - 64 bit warning fixes and fussy compiler fixes.
[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(__powerpc) || defined(__ppc__) || defined(__powerpc64__)
210 #   define ROTATE(a,n)  ({ register unsigned int ret;   \
211                                 asm (                   \
212                                 "rlwinm %0,%1,%2,0,31"  \
213                                 : "=r"(ret)             \
214                                 : "r"(a), "I"(n));      \
215                            ret;                         \
216                         })
217 #  endif
218 # endif
219 #endif /* PEDANTIC */
220
221 #if HASH_LONG_LOG2==2   /* Engage only if sizeof(HASH_LONG)== 4 */
222 /* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */
223 #ifdef ROTATE
224 /* 5 instructions with rotate instruction, else 9 */
225 #define REVERSE_FETCH32(a,l)    (                                       \
226                 l=*(const HASH_LONG *)(a),                              \
227                 ((ROTATE(l,8)&0x00FF00FF)|(ROTATE((l&0x00FF00FF),24)))  \
228                                 )
229 #else
230 /* 6 instructions with rotate instruction, else 8 */
231 #define REVERSE_FETCH32(a,l)    (                               \
232                 l=*(const HASH_LONG *)(a),                      \
233                 l=(((l>>8)&0x00FF00FF)|((l&0x00FF00FF)<<8)),    \
234                 ROTATE(l,16)                                    \
235                                 )
236 /*
237  * Originally the middle line started with l=(((l&0xFF00FF00)>>8)|...
238  * It's rewritten as above for two reasons:
239  *      - RISCs aren't good at long constants and have to explicitely
240  *        compose 'em with several (well, usually 2) instructions in a
241  *        register before performing the actual operation and (as you
242  *        already realized:-) having same constant should inspire the
243  *        compiler to permanently allocate the only register for it;
244  *      - most modern CPUs have two ALUs, but usually only one has
245  *        circuitry for shifts:-( this minor tweak inspires compiler
246  *        to schedule shift instructions in a better way...
247  *
248  *                              <appro@fy.chalmers.se>
249  */
250 #endif
251 #endif
252
253 #ifndef ROTATE
254 #define ROTATE(a,n)     (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
255 #endif
256
257 /*
258  * Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED
259  * and HASH_BLOCK_HOST_ORDER ought to be the same if input data
260  * and host are of the same "endianess". It's possible to mask
261  * this with blank #define HASH_BLOCK_DATA_ORDER though...
262  *
263  *                              <appro@fy.chalmers.se>
264  */
265 #if defined(B_ENDIAN)
266 #  if defined(DATA_ORDER_IS_BIG_ENDIAN)
267 #    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
268 #      define HASH_BLOCK_DATA_ORDER_ALIGNED     HASH_BLOCK_HOST_ORDER
269 #    endif
270 #  endif
271 #elif defined(L_ENDIAN)
272 #  if defined(DATA_ORDER_IS_LITTLE_ENDIAN)
273 #    if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
274 #      define HASH_BLOCK_DATA_ORDER_ALIGNED     HASH_BLOCK_HOST_ORDER
275 #    endif
276 #  endif
277 #endif
278
279 #if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
280 #ifndef HASH_BLOCK_DATA_ORDER
281 #error "HASH_BLOCK_DATA_ORDER must be defined!"
282 #endif
283 #endif
284
285 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
286
287 #ifndef PEDANTIC
288 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
289 #  if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
290     /*
291      * This gives ~30-40% performance improvement in SHA-256 compiled
292      * with gcc [on P4]. Well, first macro to be frank. We can pull
293      * this trick on x86* platforms only, because these CPUs can fetch
294      * unaligned data without raising an exception.
295      */
296 #   define HOST_c2l(c,l)        ({ unsigned int r=*((const unsigned int *)(c)); \
297                                    asm ("bswapl %0":"=r"(r):"0"(r));    \
298                                    (c)+=4; (l)=r;                       })
299 #   define HOST_l2c(l,c)        ({ unsigned int r=(l);                  \
300                                    asm ("bswapl %0":"=r"(r):"0"(r));    \
301                                    *((unsigned int *)(c))=r; (c)+=4; r; })
302 #  endif
303 # endif
304 #endif
305
306 #ifndef HOST_c2l
307 #define HOST_c2l(c,l)   (l =(((unsigned long)(*((c)++)))<<24),          \
308                          l|=(((unsigned long)(*((c)++)))<<16),          \
309                          l|=(((unsigned long)(*((c)++)))<< 8),          \
310                          l|=(((unsigned long)(*((c)++)))    ),          \
311                          l)
312 #endif
313 #define HOST_p_c2l(c,l,n)       {                                       \
314                         switch (n) {                                    \
315                         case 0: l =((unsigned long)(*((c)++)))<<24;     \
316                         case 1: l|=((unsigned long)(*((c)++)))<<16;     \
317                         case 2: l|=((unsigned long)(*((c)++)))<< 8;     \
318                         case 3: l|=((unsigned long)(*((c)++)));         \
319                                 } }
320 #define HOST_p_c2l_p(c,l,sc,len) {                                      \
321                         switch (sc) {                                   \
322                         case 0: l =((unsigned long)(*((c)++)))<<24;     \
323                                 if (--len == 0) break;                  \
324                         case 1: l|=((unsigned long)(*((c)++)))<<16;     \
325                                 if (--len == 0) break;                  \
326                         case 2: l|=((unsigned long)(*((c)++)))<< 8;     \
327                                 } }
328 /* NOTE the pointer is not incremented at the end of this */
329 #define HOST_c2l_p(c,l,n)       {                                       \
330                         l=0; (c)+=n;                                    \
331                         switch (n) {                                    \
332                         case 3: l =((unsigned long)(*(--(c))))<< 8;     \
333                         case 2: l|=((unsigned long)(*(--(c))))<<16;     \
334                         case 1: l|=((unsigned long)(*(--(c))))<<24;     \
335                                 } }
336 #ifndef HOST_l2c
337 #define HOST_l2c(l,c)   (*((c)++)=(unsigned char)(((l)>>24)&0xff),      \
338                          *((c)++)=(unsigned char)(((l)>>16)&0xff),      \
339                          *((c)++)=(unsigned char)(((l)>> 8)&0xff),      \
340                          *((c)++)=(unsigned char)(((l)    )&0xff),      \
341                          l)
342 #endif
343
344 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
345
346 #if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
347 # ifndef B_ENDIAN
348    /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
349 #  define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l)
350 #  define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l)
351 # endif
352 #endif
353
354 #ifndef HOST_c2l
355 #define HOST_c2l(c,l)   (l =(((unsigned long)(*((c)++)))    ),          \
356                          l|=(((unsigned long)(*((c)++)))<< 8),          \
357                          l|=(((unsigned long)(*((c)++)))<<16),          \
358                          l|=(((unsigned long)(*((c)++)))<<24),          \
359                          l)
360 #endif
361 #define HOST_p_c2l(c,l,n)       {                                       \
362                         switch (n) {                                    \
363                         case 0: l =((unsigned long)(*((c)++)));         \
364                         case 1: l|=((unsigned long)(*((c)++)))<< 8;     \
365                         case 2: l|=((unsigned long)(*((c)++)))<<16;     \
366                         case 3: l|=((unsigned long)(*((c)++)))<<24;     \
367                                 } }
368 #define HOST_p_c2l_p(c,l,sc,len) {                                      \
369                         switch (sc) {                                   \
370                         case 0: l =((unsigned long)(*((c)++)));         \
371                                 if (--len == 0) break;                  \
372                         case 1: l|=((unsigned long)(*((c)++)))<< 8;     \
373                                 if (--len == 0) break;                  \
374                         case 2: l|=((unsigned long)(*((c)++)))<<16;     \
375                                 } }
376 /* NOTE the pointer is not incremented at the end of this */
377 #define HOST_c2l_p(c,l,n)       {                                       \
378                         l=0; (c)+=n;                                    \
379                         switch (n) {                                    \
380                         case 3: l =((unsigned long)(*(--(c))))<<16;     \
381                         case 2: l|=((unsigned long)(*(--(c))))<< 8;     \
382                         case 1: l|=((unsigned long)(*(--(c))));         \
383                                 } }
384 #ifndef HOST_l2c
385 #define HOST_l2c(l,c)   (*((c)++)=(unsigned char)(((l)    )&0xff),      \
386                          *((c)++)=(unsigned char)(((l)>> 8)&0xff),      \
387                          *((c)++)=(unsigned char)(((l)>>16)&0xff),      \
388                          *((c)++)=(unsigned char)(((l)>>24)&0xff),      \
389                          l)
390 #endif
391
392 #endif
393
394 /*
395  * Time for some action:-)
396  */
397
398 int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len)
399         {
400         const unsigned char *data=data_;
401         register HASH_LONG * p;
402         register HASH_LONG l;
403         size_t sw,sc,ew,ec;
404
405         if (len==0) return 1;
406
407         l=(c->Nl+(((HASH_LONG)len)<<3))&0xffffffffUL;
408         /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
409          * Wei Dai <weidai@eskimo.com> for pointing it out. */
410         if (l < c->Nl) /* overflow */
411                 c->Nh++;
412         c->Nh+=(len>>29);       /* might cause compiler warning on 16-bit */
413         c->Nl=l;
414
415         if (c->num != 0)
416                 {
417                 p=c->data;
418                 sw=c->num>>2;
419                 sc=c->num&0x03;
420
421                 if ((c->num+len) >= HASH_CBLOCK)
422                         {
423                         l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l;
424                         for (; sw<HASH_LBLOCK; sw++)
425                                 {
426                                 HOST_c2l(data,l); p[sw]=l;
427                                 }
428                         HASH_BLOCK_HOST_ORDER (c,p,1);
429                         len-=(HASH_CBLOCK-c->num);
430                         c->num=0;
431                         /* drop through and do the rest */
432                         }
433                 else
434                         {
435                         c->num+=(unsigned int)len;
436                         if ((sc+len) < 4) /* ugly, add char's to a word */
437                                 {
438                                 l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l;
439                                 }
440                         else
441                                 {
442                                 ew=(c->num>>2);
443                                 ec=(c->num&0x03);
444                                 if (sc)
445                                         l=p[sw];
446                                 HOST_p_c2l(data,l,sc);
447                                 p[sw++]=l;
448                                 for (; sw < ew; sw++)
449                                         {
450                                         HOST_c2l(data,l); p[sw]=l;
451                                         }
452                                 if (ec)
453                                         {
454                                         HOST_c2l_p(data,l,ec); p[sw]=l;
455                                         }
456                                 }
457                         return 1;
458                         }
459                 }
460
461         sw=len/HASH_CBLOCK;
462         if (sw > 0)
463                 {
464 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
465                 /*
466                  * Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined
467                  * only if sizeof(HASH_LONG)==4.
468                  */
469                 if ((((size_t)data)%4) == 0)
470                         {
471                         /* data is properly aligned so that we can cast it: */
472                         HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,sw);
473                         sw*=HASH_CBLOCK;
474                         data+=sw;
475                         len-=sw;
476                         }
477                 else
478 #if !defined(HASH_BLOCK_DATA_ORDER)
479                         while (sw--)
480                                 {
481                                 memcpy (p=c->data,data,HASH_CBLOCK);
482                                 HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1);
483                                 data+=HASH_CBLOCK;
484                                 len-=HASH_CBLOCK;
485                                 }
486 #endif
487 #endif
488 #if defined(HASH_BLOCK_DATA_ORDER)
489                         {
490                         HASH_BLOCK_DATA_ORDER(c,data,sw);
491                         sw*=HASH_CBLOCK;
492                         data+=sw;
493                         len-=sw;
494                         }
495 #endif
496                 }
497
498         if (len!=0)
499                 {
500                 p = c->data;
501                 c->num = len;
502                 ew=len>>2;      /* words to copy */
503                 ec=len&0x03;
504                 for (; ew; ew--,p++)
505                         {
506                         HOST_c2l(data,l); *p=l;
507                         }
508                 HOST_c2l_p(data,l,ec);
509                 *p=l;
510                 }
511         return 1;
512         }
513
514
515 void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data)
516         {
517 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
518         if ((((size_t)data)%4) == 0)
519                 /* data is properly aligned so that we can cast it: */
520                 HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,1);
521         else
522 #if !defined(HASH_BLOCK_DATA_ORDER)
523                 {
524                 memcpy (c->data,data,HASH_CBLOCK);
525                 HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1);
526                 }
527 #endif
528 #endif
529 #if defined(HASH_BLOCK_DATA_ORDER)
530         HASH_BLOCK_DATA_ORDER (c,data,1);
531 #endif
532         }
533
534
535 int HASH_FINAL (unsigned char *md, HASH_CTX *c)
536         {
537         register HASH_LONG *p;
538         register unsigned long l;
539         register int i,j;
540         static const unsigned char end[4]={0x80,0x00,0x00,0x00};
541         const unsigned char *cp=end;
542
543         /* c->num should definitly have room for at least one more byte. */
544         p=c->data;
545         i=c->num>>2;
546         j=c->num&0x03;
547
548 #if 0
549         /* purify often complains about the following line as an
550          * Uninitialized Memory Read.  While this can be true, the
551          * following p_c2l macro will reset l when that case is true.
552          * This is because j&0x03 contains the number of 'valid' bytes
553          * already in p[i].  If and only if j&0x03 == 0, the UMR will
554          * occur but this is also the only time p_c2l will do
555          * l= *(cp++) instead of l|= *(cp++)
556          * Many thanks to Alex Tang <altitude@cic.net> for pickup this
557          * 'potential bug' */
558 #ifdef PURIFY
559         if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */
560 #endif
561         l=p[i];
562 #else
563         l = (j==0) ? 0 : p[i];
564 #endif
565         HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */
566
567         if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */
568                 {
569                 if (i<HASH_LBLOCK) p[i]=0;
570                 HASH_BLOCK_HOST_ORDER (c,p,1);
571                 i=0;
572                 }
573         for (; i<(HASH_LBLOCK-2); i++)
574                 p[i]=0;
575
576 #if   defined(DATA_ORDER_IS_BIG_ENDIAN)
577         p[HASH_LBLOCK-2]=c->Nh;
578         p[HASH_LBLOCK-1]=c->Nl;
579 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
580         p[HASH_LBLOCK-2]=c->Nl;
581         p[HASH_LBLOCK-1]=c->Nh;
582 #endif
583         HASH_BLOCK_HOST_ORDER (c,p,1);
584
585 #ifndef HASH_MAKE_STRING
586 #error "HASH_MAKE_STRING must be defined!"
587 #else
588         HASH_MAKE_STRING(c,md);
589 #endif
590
591         c->num=0;
592         /* clear stuff, HASH_BLOCK may be leaving some stuff on the stack
593          * but I'm not worried :-)
594         OPENSSL_cleanse((void *)c,sizeof(HASH_CTX));
595          */
596         return 1;
597         }
598
599 #ifndef MD32_REG_T
600 #define MD32_REG_T long
601 /*
602  * This comment was originaly written for MD5, which is why it
603  * discusses A-D. But it basically applies to all 32-bit digests,
604  * which is why it was moved to common header file.
605  *
606  * In case you wonder why A-D are declared as long and not
607  * as MD5_LONG. Doing so results in slight performance
608  * boost on LP64 architectures. The catch is we don't
609  * really care if 32 MSBs of a 64-bit register get polluted
610  * with eventual overflows as we *save* only 32 LSBs in
611  * *either* case. Now declaring 'em long excuses the compiler
612  * from keeping 32 MSBs zeroed resulting in 13% performance
613  * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
614  * Well, to be honest it should say that this *prevents* 
615  * performance degradation.
616  *                              <appro@fy.chalmers.se>
617  * Apparently there're LP64 compilers that generate better
618  * code if A-D are declared int. Most notably GCC-x86_64
619  * generates better code.
620  *                              <appro@fy.chalmers.se>
621  */
622 #endif