libbb/hash_md5_sha: use common ctx and code for md5 and sha1/256

function                                             old     new   delta
sha256_process_block64                               421     433     +12
md5_crypt                                            578     587      +9
md5_begin                                             43      50      +7
md5_hash                                              99      97      -2
sha1_end                                              85      82      -3
md5_end                                               36      31      -5
common64_end                                          93      86      -7
sha1_hash                                             97       -     -97

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

1 files changed, 416 insertions, 482 deletions

diff --git a/libbb/hash_md5_sha.c b/libbb/hash_md5_sha.c
index b07ba55f7..f5f875a64 100644
--- a/libbb/hash_md5_sha.c
+++ b/libbb/hash_md5_sha.c
@@ -32,16 +32,38 @@ static ALWAYS_INLINE uint64_t rotr64(uint64_t x, unsigned n)
 }
 
 
-typedef struct common64_ctx_t {
-	char wbuffer[64]; /* NB: always correctly aligned for uint64_t */
-	uint64_t total64;
-} common64_ctx_t;
+/* Feed data through a temporary buffer.
+ * The internal buffer remembers previous data until it has 64
+ * bytes worth to pass on.
+ */
+static void FAST_FUNC common64_hash(md5_ctx_t *ctx, const void *buffer, size_t len)
+{
+	unsigned bufpos = ctx->total64 & 63;
 
-typedef void FAST_FUNC process_block64_func(void*);
+	ctx->total64 += len;
+
+	while (1) {
+		unsigned remaining = 64 - bufpos;
+		if (remaining > len)
+			remaining = len;
+		/* Copy data into aligned buffer */
+		memcpy(ctx->wbuffer + bufpos, buffer, remaining);
+		len -= remaining;
+		buffer = (const char *)buffer + remaining;
+		bufpos += remaining;
+		/* clever way to do "if (bufpos != 64) break; ... ; bufpos = 0;" */
+		bufpos -= 64;
+		if (bufpos != 0)
+			break;
+		/* Buffer is filled up, process it */
+		ctx->process_block(ctx);
+		/*bufpos = 0; - already is */
+	}
+}
 
-static void FAST_FUNC common64_end(void *vctx, process_block64_func process_block64, int swap_needed)
+/* Process the remaining bytes in the buffer */
+static void FAST_FUNC common64_end(md5_ctx_t *ctx, int swap_needed)
 {
-	common64_ctx_t *ctx = vctx;
 	unsigned bufpos = ctx->total64 & 63;
 	/* Pad the buffer to the next 64-byte boundary with 0x80,0,0,0... */
 	ctx->wbuffer[bufpos++] = 0x80;
@@ -59,7 +81,7 @@ static void FAST_FUNC common64_end(void *vctx, process_block64_func process_bloc
 			/* wbuffer is suitably aligned for this */
 			*(uint64_t *) (&ctx->wbuffer[64 - 8]) = t;
 		}
-		process_block64(ctx);
+		ctx->process_block(ctx);
 		if (remaining >= 8)
 			break;
 		bufpos = 0;
@@ -68,6 +90,391 @@ static void FAST_FUNC common64_end(void *vctx, process_block64_func process_bloc
 
 
 /*
+ * Compute MD5 checksum of strings according to the
+ * definition of MD5 in RFC 1321 from April 1992.
+ *
+ * Written by Ulrich Drepper <drepper@gnu.ai.mit.edu>, 1995.
+ *
+ * Copyright (C) 1995-1999 Free Software Foundation, Inc.
+ * Copyright (C) 2001 Manuel Novoa III
+ * Copyright (C) 2003 Glenn L. McGrath
+ * Copyright (C) 2003 Erik Andersen
+ *
+ * Licensed under GPLv2 or later, see file LICENSE in this source tree.
+ */
+
+/* 0: fastest, 3: smallest */
+#if CONFIG_MD5_SIZE_VS_SPEED < 0
+# define MD5_SIZE_VS_SPEED 0
+#elif CONFIG_MD5_SIZE_VS_SPEED > 3
+# define MD5_SIZE_VS_SPEED 3
+#else
+# define MD5_SIZE_VS_SPEED CONFIG_MD5_SIZE_VS_SPEED
+#endif
+
+/* These are the four functions used in the four steps of the MD5 algorithm
+ * and defined in the RFC 1321.  The first function is a little bit optimized
+ * (as found in Colin Plumbs public domain implementation).
+ * #define FF(b, c, d) ((b & c) | (~b & d))
+ */
+#undef FF
+#undef FG
+#undef FH
+#undef FI
+#define FF(b, c, d) (d ^ (b & (c ^ d)))
+#define FG(b, c, d) FF(d, b, c)
+#define FH(b, c, d) (b ^ c ^ d)
+#define FI(b, c, d) (c ^ (b | ~d))
+
+/* Hash a single block, 64 bytes long and 4-byte aligned */
+static void FAST_FUNC md5_process_block64(md5_ctx_t *ctx)
+{
+#if MD5_SIZE_VS_SPEED > 0
+	/* Before we start, one word to the strange constants.
+	   They are defined in RFC 1321 as
+	   T[i] = (int)(4294967296.0 * fabs(sin(i))), i=1..64
+	 */
+	static const uint32_t C_array[] = {
+		/* round 1 */
+		0xd76aa478, 0xe8c7b756, 0x242070db, 0xc1bdceee,
+		0xf57c0faf, 0x4787c62a, 0xa8304613, 0xfd469501,
+		0x698098d8, 0x8b44f7af, 0xffff5bb1, 0x895cd7be,
+		0x6b901122, 0xfd987193, 0xa679438e, 0x49b40821,
+		/* round 2 */
+		0xf61e2562, 0xc040b340, 0x265e5a51, 0xe9b6c7aa,
+		0xd62f105d, 0x02441453, 0xd8a1e681, 0xe7d3fbc8,
+		0x21e1cde6, 0xc33707d6, 0xf4d50d87, 0x455a14ed,
+		0xa9e3e905, 0xfcefa3f8, 0x676f02d9, 0x8d2a4c8a,
+		/* round 3 */
+		0xfffa3942, 0x8771f681, 0x6d9d6122, 0xfde5380c,
+		0xa4beea44, 0x4bdecfa9, 0xf6bb4b60, 0xbebfbc70,
+		0x289b7ec6, 0xeaa127fa, 0xd4ef3085, 0x4881d05,
+		0xd9d4d039, 0xe6db99e5, 0x1fa27cf8, 0xc4ac5665,
+		/* round 4 */
+		0xf4292244, 0x432aff97, 0xab9423a7, 0xfc93a039,
+		0x655b59c3, 0x8f0ccc92, 0xffeff47d, 0x85845dd1,
+		0x6fa87e4f, 0xfe2ce6e0, 0xa3014314, 0x4e0811a1,
+		0xf7537e82, 0xbd3af235, 0x2ad7d2bb, 0xeb86d391
+	};
+	static const char P_array[] ALIGN1 = {
+# if MD5_SIZE_VS_SPEED > 1
+		0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,	/* 1 */
+# endif
+		1, 6, 11, 0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12,	/* 2 */
+		5, 8, 11, 14, 1, 4, 7, 10, 13, 0, 3, 6, 9, 12, 15, 2,	/* 3 */
+		0, 7, 14, 5, 12, 3, 10, 1, 8, 15, 6, 13, 4, 11, 2, 9	/* 4 */
+	};
+#endif
+	uint32_t *words = (void*) ctx->wbuffer;
+	uint32_t A = ctx->hash[0];
+	uint32_t B = ctx->hash[1];
+	uint32_t C = ctx->hash[2];
+	uint32_t D = ctx->hash[3];
+
+#if MD5_SIZE_VS_SPEED >= 2  /* 2 or 3 */
+
+	static const char S_array[] ALIGN1 = {
+		7, 12, 17, 22,
+		5, 9, 14, 20,
+		4, 11, 16, 23,
+		6, 10, 15, 21
+	};
+	const uint32_t *pc;
+	const char *pp;
+	const char *ps;
+	int i;
+	uint32_t temp;
+
+# if BB_BIG_ENDIAN
+	for (i = 0; i < 16; i++)
+		words[i] = SWAP_LE32(words[i]);
+# endif
+
+# if MD5_SIZE_VS_SPEED == 3
+	pc = C_array;
+	pp = P_array;
+	ps = S_array - 4;
+
+	for (i = 0; i < 64; i++) {
+		if ((i & 0x0f) == 0)
+			ps += 4;
+		temp = A;
+		switch (i >> 4) {
+		case 0:
+			temp += FF(B, C, D);
+			break;
+		case 1:
+			temp += FG(B, C, D);
+			break;
+		case 2:
+			temp += FH(B, C, D);
+			break;
+		case 3:
+			temp += FI(B, C, D);
+		}
+		temp += words[(int) (*pp++)] + *pc++;
+		temp = rotl32(temp, ps[i & 3]);
+		temp += B;
+		A = D;
+		D = C;
+		C = B;
+		B = temp;
+	}
+# else  /* MD5_SIZE_VS_SPEED == 2 */
+	pc = C_array;
+	pp = P_array;
+	ps = S_array;
+
+	for (i = 0; i < 16; i++) {
+		temp = A + FF(B, C, D) + words[(int) (*pp++)] + *pc++;
+		temp = rotl32(temp, ps[i & 3]);
+		temp += B;
+		A = D;
+		D = C;
+		C = B;
+		B = temp;
+	}
+	ps += 4;
+	for (i = 0; i < 16; i++) {
+		temp = A + FG(B, C, D) + words[(int) (*pp++)] + *pc++;
+		temp = rotl32(temp, ps[i & 3]);
+		temp += B;
+		A = D;
+		D = C;
+		C = B;
+		B = temp;
+	}
+	ps += 4;
+	for (i = 0; i < 16; i++) {
+		temp = A + FH(B, C, D) + words[(int) (*pp++)] + *pc++;
+		temp = rotl32(temp, ps[i & 3]);
+		temp += B;
+		A = D;
+		D = C;
+		C = B;
+		B = temp;
+	}
+	ps += 4;
+	for (i = 0; i < 16; i++) {
+		temp = A + FI(B, C, D) + words[(int) (*pp++)] + *pc++;
+		temp = rotl32(temp, ps[i & 3]);
+		temp += B;
+		A = D;
+		D = C;
+		C = B;
+		B = temp;
+	}
+# endif
+	/* Add checksum to the starting values */
+	ctx->hash[0] += A;
+	ctx->hash[1] += B;
+	ctx->hash[2] += C;
+	ctx->hash[3] += D;
+
+#else  /* MD5_SIZE_VS_SPEED == 0 or 1 */
+
+	uint32_t A_save = A;
+	uint32_t B_save = B;
+	uint32_t C_save = C;
+	uint32_t D_save = D;
+# if MD5_SIZE_VS_SPEED == 1
+	const uint32_t *pc;
+	const char *pp;
+	int i;
+# endif
+
+	/* First round: using the given function, the context and a constant
+	   the next context is computed.  Because the algorithm's processing
+	   unit is a 32-bit word and it is determined to work on words in
+	   little endian byte order we perhaps have to change the byte order
+	   before the computation.  To reduce the work for the next steps
+	   we save swapped words in WORDS array.  */
+# undef OP
+# define OP(a, b, c, d, s, T) \
+	do { \
+		a += FF(b, c, d) + (*words IF_BIG_ENDIAN(= SWAP_LE32(*words))) + T; \
+		words++; \
+		a = rotl32(a, s); \
+		a += b; \
+	} while (0)
+
+	/* Round 1 */
+# if MD5_SIZE_VS_SPEED == 1
+	pc = C_array;
+	for (i = 0; i < 4; i++) {
+		OP(A, B, C, D, 7, *pc++);
+		OP(D, A, B, C, 12, *pc++);
+		OP(C, D, A, B, 17, *pc++);
+		OP(B, C, D, A, 22, *pc++);
+	}
+# else
+	OP(A, B, C, D, 7, 0xd76aa478);
+	OP(D, A, B, C, 12, 0xe8c7b756);
+	OP(C, D, A, B, 17, 0x242070db);
+	OP(B, C, D, A, 22, 0xc1bdceee);
+	OP(A, B, C, D, 7, 0xf57c0faf);
+	OP(D, A, B, C, 12, 0x4787c62a);
+	OP(C, D, A, B, 17, 0xa8304613);
+	OP(B, C, D, A, 22, 0xfd469501);
+	OP(A, B, C, D, 7, 0x698098d8);
+	OP(D, A, B, C, 12, 0x8b44f7af);
+	OP(C, D, A, B, 17, 0xffff5bb1);
+	OP(B, C, D, A, 22, 0x895cd7be);
+	OP(A, B, C, D, 7, 0x6b901122);
+	OP(D, A, B, C, 12, 0xfd987193);
+	OP(C, D, A, B, 17, 0xa679438e);
+	OP(B, C, D, A, 22, 0x49b40821);
+# endif
+	words -= 16;
+
+	/* For the second to fourth round we have the possibly swapped words
+	   in WORDS.  Redefine the macro to take an additional first
+	   argument specifying the function to use.  */
+# undef OP
+# define OP(f, a, b, c, d, k, s, T) \
+	do { \
+		a += f(b, c, d) + words[k] + T; \
+		a = rotl32(a, s); \
+		a += b; \
+	} while (0)
+
+	/* Round 2 */
+# if MD5_SIZE_VS_SPEED == 1
+	pp = P_array;
+	for (i = 0; i < 4; i++) {
+		OP(FG, A, B, C, D, (int) (*pp++), 5, *pc++);
+		OP(FG, D, A, B, C, (int) (*pp++), 9, *pc++);
+		OP(FG, C, D, A, B, (int) (*pp++), 14, *pc++);
+		OP(FG, B, C, D, A, (int) (*pp++), 20, *pc++);
+	}
+# else
+	OP(FG, A, B, C, D, 1, 5, 0xf61e2562);
+	OP(FG, D, A, B, C, 6, 9, 0xc040b340);
+	OP(FG, C, D, A, B, 11, 14, 0x265e5a51);
+	OP(FG, B, C, D, A, 0, 20, 0xe9b6c7aa);
+	OP(FG, A, B, C, D, 5, 5, 0xd62f105d);
+	OP(FG, D, A, B, C, 10, 9, 0x02441453);
+	OP(FG, C, D, A, B, 15, 14, 0xd8a1e681);
+	OP(FG, B, C, D, A, 4, 20, 0xe7d3fbc8);
+	OP(FG, A, B, C, D, 9, 5, 0x21e1cde6);
+	OP(FG, D, A, B, C, 14, 9, 0xc33707d6);
+	OP(FG, C, D, A, B, 3, 14, 0xf4d50d87);
+	OP(FG, B, C, D, A, 8, 20, 0x455a14ed);
+	OP(FG, A, B, C, D, 13, 5, 0xa9e3e905);
+	OP(FG, D, A, B, C, 2, 9, 0xfcefa3f8);
+	OP(FG, C, D, A, B, 7, 14, 0x676f02d9);
+	OP(FG, B, C, D, A, 12, 20, 0x8d2a4c8a);
+# endif
+
+	/* Round 3 */
+# if MD5_SIZE_VS_SPEED == 1
+	for (i = 0; i < 4; i++) {
+		OP(FH, A, B, C, D, (int) (*pp++), 4, *pc++);
+		OP(FH, D, A, B, C, (int) (*pp++), 11, *pc++);
+		OP(FH, C, D, A, B, (int) (*pp++), 16, *pc++);
+		OP(FH, B, C, D, A, (int) (*pp++), 23, *pc++);
+	}
+# else
+	OP(FH, A, B, C, D, 5, 4, 0xfffa3942);
+	OP(FH, D, A, B, C, 8, 11, 0x8771f681);
+	OP(FH, C, D, A, B, 11, 16, 0x6d9d6122);
+	OP(FH, B, C, D, A, 14, 23, 0xfde5380c);
+	OP(FH, A, B, C, D, 1, 4, 0xa4beea44);
+	OP(FH, D, A, B, C, 4, 11, 0x4bdecfa9);
+	OP(FH, C, D, A, B, 7, 16, 0xf6bb4b60);
+	OP(FH, B, C, D, A, 10, 23, 0xbebfbc70);
+	OP(FH, A, B, C, D, 13, 4, 0x289b7ec6);
+	OP(FH, D, A, B, C, 0, 11, 0xeaa127fa);
+	OP(FH, C, D, A, B, 3, 16, 0xd4ef3085);
+	OP(FH, B, C, D, A, 6, 23, 0x04881d05);
+	OP(FH, A, B, C, D, 9, 4, 0xd9d4d039);
+	OP(FH, D, A, B, C, 12, 11, 0xe6db99e5);
+	OP(FH, C, D, A, B, 15, 16, 0x1fa27cf8);
+	OP(FH, B, C, D, A, 2, 23, 0xc4ac5665);
+# endif
+
+	/* Round 4 */
+# if MD5_SIZE_VS_SPEED == 1
+	for (i = 0; i < 4; i++) {
+		OP(FI, A, B, C, D, (int) (*pp++), 6, *pc++);
+		OP(FI, D, A, B, C, (int) (*pp++), 10, *pc++);
+		OP(FI, C, D, A, B, (int) (*pp++), 15, *pc++);
+		OP(FI, B, C, D, A, (int) (*pp++), 21, *pc++);
+	}
+# else
+	OP(FI, A, B, C, D, 0, 6, 0xf4292244);
+	OP(FI, D, A, B, C, 7, 10, 0x432aff97);
+	OP(FI, C, D, A, B, 14, 15, 0xab9423a7);
+	OP(FI, B, C, D, A, 5, 21, 0xfc93a039);
+	OP(FI, A, B, C, D, 12, 6, 0x655b59c3);
+	OP(FI, D, A, B, C, 3, 10, 0x8f0ccc92);
+	OP(FI, C, D, A, B, 10, 15, 0xffeff47d);
+	OP(FI, B, C, D, A, 1, 21, 0x85845dd1);
+	OP(FI, A, B, C, D, 8, 6, 0x6fa87e4f);
+	OP(FI, D, A, B, C, 15, 10, 0xfe2ce6e0);
+	OP(FI, C, D, A, B, 6, 15, 0xa3014314);
+	OP(FI, B, C, D, A, 13, 21, 0x4e0811a1);
+	OP(FI, A, B, C, D, 4, 6, 0xf7537e82);
+	OP(FI, D, A, B, C, 11, 10, 0xbd3af235);
+	OP(FI, C, D, A, B, 2, 15, 0x2ad7d2bb);
+	OP(FI, B, C, D, A, 9, 21, 0xeb86d391);
+# undef OP
+# endif
+	/* Add checksum to the starting values */
+	ctx->hash[0] = A_save + A;
+	ctx->hash[1] = B_save + B;
+	ctx->hash[2] = C_save + C;
+	ctx->hash[3] = D_save + D;
+#endif
+}
+#undef FF
+#undef FG
+#undef FH
+#undef FI
+
+/* Initialize structure containing state of computation.
+ * (RFC 1321, 3.3: Step 3)
+ */
+void FAST_FUNC md5_begin(md5_ctx_t *ctx)
+{
+	ctx->hash[0] = 0x67452301;
+	ctx->hash[1] = 0xefcdab89;
+	ctx->hash[2] = 0x98badcfe;
+	ctx->hash[3] = 0x10325476;
+	ctx->total64 = 0;
+	ctx->process_block = md5_process_block64;
+}
+
+/* Used also for sha1 and sha256 */
+void FAST_FUNC md5_hash(md5_ctx_t *ctx, const void *buffer, size_t len)
+{
+	common64_hash(ctx, buffer, len);
+}
+
+/* Process the remaining bytes in the buffer and put result from CTX
+ * in first 16 bytes following RESBUF.  The result is always in little
+ * endian byte order, so that a byte-wise output yields to the wanted
+ * ASCII representation of the message digest.
+ */
+void FAST_FUNC md5_end(md5_ctx_t *ctx, void *resbuf)
+{
+	/* MD5 stores total in LE, need to swap on BE arches: */
+	common64_end(ctx, /*swap_needed:*/ BB_BIG_ENDIAN);
+
+	/* The MD5 result is in little endian byte order.
+	 * We (ab)use the fact that A-D are consecutive in memory.
+	 */
+#if BB_BIG_ENDIAN
+	ctx->hash[0] = SWAP_LE32(ctx->hash[0]);
+	ctx->hash[1] = SWAP_LE32(ctx->hash[1]);
+	ctx->hash[2] = SWAP_LE32(ctx->hash[2]);
+	ctx->hash[3] = SWAP_LE32(ctx->hash[3]);
+#endif
+	memcpy(resbuf, ctx->hash, sizeof(ctx->hash[0]) * 4);
+}
+
+
+/*
  * Based on shasum from http://www.netsw.org/crypto/hash/
  * Majorly hacked up to use Dr Brian Gladman's sha1 code
  *
@@ -396,51 +803,6 @@ void FAST_FUNC sha512_begin(sha512_ctx_t *ctx)
 	/*ctx->total64[0] = ctx->total64[1] = 0; - already done */
 }
 
-
-/* Used also for sha256 */
-void FAST_FUNC sha1_hash(sha1_ctx_t *ctx, const void *buffer, size_t len)
-{
-	unsigned bufpos = ctx->total64 & 63;
-	unsigned remaining;
-
-	ctx->total64 += len;
-#if 0
-	remaining = 64 - bufpos;
-
-	/* Hash whole blocks */
-	while (len >= remaining) {
-		memcpy(ctx->wbuffer + bufpos, buffer, remaining);
-		buffer = (const char *)buffer + remaining;
-		len -= remaining;
-		remaining = 64;
-		bufpos = 0;
-		ctx->process_block(ctx);
-	}
-
-	/* Save last, partial blosk */
-	memcpy(ctx->wbuffer + bufpos, buffer, len);
-#else
-	/* Tiny bit smaller code */
-	while (1) {
-		remaining = 64 - bufpos;
-		if (remaining > len)
-			remaining = len;
-		/* Copy data into aligned buffer */
-		memcpy(ctx->wbuffer + bufpos, buffer, remaining);
-		len -= remaining;
-		buffer = (const char *)buffer + remaining;
-		bufpos += remaining;
-		/* clever way to do "if (bufpos != 64) break; ... ; bufpos = 0;" */
-		bufpos -= 64;
-		if (bufpos != 0)
-			break;
-		/* Buffer is filled up, process it */
-		ctx->process_block(ctx);
-		/*bufpos = 0; - already is */
-	}
-#endif
-}
-
 void FAST_FUNC sha512_hash(sha512_ctx_t *ctx, const void *buffer, size_t len)
 {
 	unsigned bufpos = ctx->total64[0] & 127;
@@ -488,14 +850,13 @@ void FAST_FUNC sha512_hash(sha512_ctx_t *ctx, const void *buffer, size_t len)
 #endif
 }
 
-
 /* Used also for sha256 */
 void FAST_FUNC sha1_end(sha1_ctx_t *ctx, void *resbuf)
 {
 	unsigned hash_size;
 
 	/* SHA stores total in BE, need to swap on LE arches: */
-	common64_end(ctx, (process_block64_func*) ctx->process_block, /*swap_needed:*/ BB_LITTLE_ENDIAN);
+	common64_end(ctx, /*swap_needed:*/ BB_LITTLE_ENDIAN);
 
 	hash_size = (ctx->process_block == sha1_process_block64) ? 5 : 8;
 	/* This way we do not impose alignment constraints on resbuf: */
@@ -540,430 +901,3 @@ void FAST_FUNC sha512_end(sha512_ctx_t *ctx, void *resbuf)
 	}
 	memcpy(resbuf, ctx->hash, sizeof(ctx->hash));
 }
-
-
-/*
- * Compute MD5 checksum of strings according to the
- * definition of MD5 in RFC 1321 from April 1992.
- *
- * Written by Ulrich Drepper <drepper@gnu.ai.mit.edu>, 1995.
- *
- * Copyright (C) 1995-1999 Free Software Foundation, Inc.
- * Copyright (C) 2001 Manuel Novoa III
- * Copyright (C) 2003 Glenn L. McGrath
- * Copyright (C) 2003 Erik Andersen
- *
- * Licensed under GPLv2 or later, see file LICENSE in this source tree.
- */
-
-/* 0: fastest, 3: smallest */
-#if CONFIG_MD5_SIZE_VS_SPEED < 0
-# define MD5_SIZE_VS_SPEED 0
-#elif CONFIG_MD5_SIZE_VS_SPEED > 3
-# define MD5_SIZE_VS_SPEED 3
-#else
-# define MD5_SIZE_VS_SPEED CONFIG_MD5_SIZE_VS_SPEED
-#endif
-
-/* Initialize structure containing state of computation.
- * (RFC 1321, 3.3: Step 3)
- */
-void FAST_FUNC md5_begin(md5_ctx_t *ctx)
-{
-	ctx->A = 0x67452301;
-	ctx->B = 0xefcdab89;
-	ctx->C = 0x98badcfe;
-	ctx->D = 0x10325476;
-	ctx->total64 = 0;
-}
-
-/* These are the four functions used in the four steps of the MD5 algorithm
- * and defined in the RFC 1321.  The first function is a little bit optimized
- * (as found in Colin Plumbs public domain implementation).
- * #define FF(b, c, d) ((b & c) | (~b & d))
- */
-#undef FF
-#undef FG
-#undef FH
-#undef FI
-#define FF(b, c, d) (d ^ (b & (c ^ d)))
-#define FG(b, c, d) FF(d, b, c)
-#define FH(b, c, d) (b ^ c ^ d)
-#define FI(b, c, d) (c ^ (b | ~d))
-
-/* Hash a single block, 64 bytes long and 4-byte aligned */
-static void FAST_FUNC md5_process_block64(md5_ctx_t *ctx)
-{
-#if MD5_SIZE_VS_SPEED > 0
-	/* Before we start, one word to the strange constants.
-	   They are defined in RFC 1321 as
-	   T[i] = (int)(4294967296.0 * fabs(sin(i))), i=1..64
-	 */
-	static const uint32_t C_array[] = {
-		/* round 1 */
-		0xd76aa478, 0xe8c7b756, 0x242070db, 0xc1bdceee,
-		0xf57c0faf, 0x4787c62a, 0xa8304613, 0xfd469501,
-		0x698098d8, 0x8b44f7af, 0xffff5bb1, 0x895cd7be,
-		0x6b901122, 0xfd987193, 0xa679438e, 0x49b40821,
-		/* round 2 */
-		0xf61e2562, 0xc040b340, 0x265e5a51, 0xe9b6c7aa,
-		0xd62f105d, 0x02441453, 0xd8a1e681, 0xe7d3fbc8,
-		0x21e1cde6, 0xc33707d6, 0xf4d50d87, 0x455a14ed,
-		0xa9e3e905, 0xfcefa3f8, 0x676f02d9, 0x8d2a4c8a,
-		/* round 3 */
-		0xfffa3942, 0x8771f681, 0x6d9d6122, 0xfde5380c,
-		0xa4beea44, 0x4bdecfa9, 0xf6bb4b60, 0xbebfbc70,
-		0x289b7ec6, 0xeaa127fa, 0xd4ef3085, 0x4881d05,
-		0xd9d4d039, 0xe6db99e5, 0x1fa27cf8, 0xc4ac5665,
-		/* round 4 */
-		0xf4292244, 0x432aff97, 0xab9423a7, 0xfc93a039,
-		0x655b59c3, 0x8f0ccc92, 0xffeff47d, 0x85845dd1,
-		0x6fa87e4f, 0xfe2ce6e0, 0xa3014314, 0x4e0811a1,
-		0xf7537e82, 0xbd3af235, 0x2ad7d2bb, 0xeb86d391
-	};
-	static const char P_array[] ALIGN1 = {
-# if MD5_SIZE_VS_SPEED > 1
-		0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,	/* 1 */
-# endif
-		1, 6, 11, 0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12,	/* 2 */
-		5, 8, 11, 14, 1, 4, 7, 10, 13, 0, 3, 6, 9, 12, 15, 2,	/* 3 */
-		0, 7, 14, 5, 12, 3, 10, 1, 8, 15, 6, 13, 4, 11, 2, 9	/* 4 */
-	};
-#endif
-	uint32_t *words = (void*) ctx->wbuffer;
-	uint32_t A = ctx->A;
-	uint32_t B = ctx->B;
-	uint32_t C = ctx->C;
-	uint32_t D = ctx->D;
-
-#if MD5_SIZE_VS_SPEED >= 2  /* 2 or 3 */
-
-	static const char S_array[] ALIGN1 = {
-		7, 12, 17, 22,
-		5, 9, 14, 20,
-		4, 11, 16, 23,
-		6, 10, 15, 21
-	};
-	const uint32_t *pc;
-	const char *pp;
-	const char *ps;
-	int i;
-	uint32_t temp;
-
-# if BB_BIG_ENDIAN
-	for (i = 0; i < 16; i++)
-		words[i] = SWAP_LE32(words[i]);
-# endif
-
-# if MD5_SIZE_VS_SPEED == 3
-	pc = C_array;
-	pp = P_array;
-	ps = S_array - 4;
-
-	for (i = 0; i < 64; i++) {
-		if ((i & 0x0f) == 0)
-			ps += 4;
-		temp = A;
-		switch (i >> 4) {
-		case 0:
-			temp += FF(B, C, D);
-			break;
-		case 1:
-			temp += FG(B, C, D);
-			break;
-		case 2:
-			temp += FH(B, C, D);
-			break;
-		case 3:
-			temp += FI(B, C, D);
-		}
-		temp += words[(int) (*pp++)] + *pc++;
-		temp = rotl32(temp, ps[i & 3]);
-		temp += B;
-		A = D;
-		D = C;
-		C = B;
-		B = temp;
-	}
-# else  /* MD5_SIZE_VS_SPEED == 2 */
-	pc = C_array;
-	pp = P_array;
-	ps = S_array;
-
-	for (i = 0; i < 16; i++) {
-		temp = A + FF(B, C, D) + words[(int) (*pp++)] + *pc++;
-		temp = rotl32(temp, ps[i & 3]);
-		temp += B;
-		A = D;
-		D = C;
-		C = B;
-		B = temp;
-	}
-	ps += 4;
-	for (i = 0; i < 16; i++) {
-		temp = A + FG(B, C, D) + words[(int) (*pp++)] + *pc++;
-		temp = rotl32(temp, ps[i & 3]);
-		temp += B;
-		A = D;
-		D = C;
-		C = B;
-		B = temp;
-	}
-	ps += 4;
-	for (i = 0; i < 16; i++) {
-		temp = A + FH(B, C, D) + words[(int) (*pp++)] + *pc++;
-		temp = rotl32(temp, ps[i & 3]);
-		temp += B;
-		A = D;
-		D = C;
-		C = B;
-		B = temp;
-	}
-	ps += 4;
-	for (i = 0; i < 16; i++) {
-		temp = A + FI(B, C, D) + words[(int) (*pp++)] + *pc++;
-		temp = rotl32(temp, ps[i & 3]);
-		temp += B;
-		A = D;
-		D = C;
-		C = B;
-		B = temp;
-	}
-# endif
-	/* Add checksum to the starting values */
-	ctx->A += A;
-	ctx->B += B;
-	ctx->C += C;
-	ctx->D += D;
-
-#else  /* MD5_SIZE_VS_SPEED == 0 or 1 */
-
-	uint32_t A_save = A;
-	uint32_t B_save = B;
-	uint32_t C_save = C;
-	uint32_t D_save = D;
-# if MD5_SIZE_VS_SPEED == 1
-	const uint32_t *pc;
-	const char *pp;
-	int i;
-# endif
-
-	/* First round: using the given function, the context and a constant
-	   the next context is computed.  Because the algorithm's processing
-	   unit is a 32-bit word and it is determined to work on words in
-	   little endian byte order we perhaps have to change the byte order
-	   before the computation.  To reduce the work for the next steps
-	   we save swapped words in WORDS array.  */
-# undef OP
-# define OP(a, b, c, d, s, T) \
-	do { \
-		a += FF(b, c, d) + (*words IF_BIG_ENDIAN(= SWAP_LE32(*words))) + T; \
-		words++; \
-		a = rotl32(a, s); \
-		a += b; \
-	} while (0)
-
-	/* Round 1 */
-# if MD5_SIZE_VS_SPEED == 1
-	pc = C_array;
-	for (i = 0; i < 4; i++) {
-		OP(A, B, C, D, 7, *pc++);
-		OP(D, A, B, C, 12, *pc++);
-		OP(C, D, A, B, 17, *pc++);
-		OP(B, C, D, A, 22, *pc++);
-	}
-# else
-	OP(A, B, C, D, 7, 0xd76aa478);
-	OP(D, A, B, C, 12, 0xe8c7b756);
-	OP(C, D, A, B, 17, 0x242070db);
-	OP(B, C, D, A, 22, 0xc1bdceee);
-	OP(A, B, C, D, 7, 0xf57c0faf);
-	OP(D, A, B, C, 12, 0x4787c62a);
-	OP(C, D, A, B, 17, 0xa8304613);
-	OP(B, C, D, A, 22, 0xfd469501);
-	OP(A, B, C, D, 7, 0x698098d8);
-	OP(D, A, B, C, 12, 0x8b44f7af);
-	OP(C, D, A, B, 17, 0xffff5bb1);
-	OP(B, C, D, A, 22, 0x895cd7be);
-	OP(A, B, C, D, 7, 0x6b901122);
-	OP(D, A, B, C, 12, 0xfd987193);
-	OP(C, D, A, B, 17, 0xa679438e);
-	OP(B, C, D, A, 22, 0x49b40821);
-# endif
-	words -= 16;
-
-	/* For the second to fourth round we have the possibly swapped words
-	   in WORDS.  Redefine the macro to take an additional first
-	   argument specifying the function to use.  */
-# undef OP
-# define OP(f, a, b, c, d, k, s, T) \
-	do { \
-		a += f(b, c, d) + words[k] + T; \
-		a = rotl32(a, s); \
-		a += b; \
-	} while (0)
-
-	/* Round 2 */
-# if MD5_SIZE_VS_SPEED == 1
-	pp = P_array;
-	for (i = 0; i < 4; i++) {
-		OP(FG, A, B, C, D, (int) (*pp++), 5, *pc++);
-		OP(FG, D, A, B, C, (int) (*pp++), 9, *pc++);
-		OP(FG, C, D, A, B, (int) (*pp++), 14, *pc++);
-		OP(FG, B, C, D, A, (int) (*pp++), 20, *pc++);
-	}
-# else
-	OP(FG, A, B, C, D, 1, 5, 0xf61e2562);
-	OP(FG, D, A, B, C, 6, 9, 0xc040b340);
-	OP(FG, C, D, A, B, 11, 14, 0x265e5a51);
-	OP(FG, B, C, D, A, 0, 20, 0xe9b6c7aa);
-	OP(FG, A, B, C, D, 5, 5, 0xd62f105d);
-	OP(FG, D, A, B, C, 10, 9, 0x02441453);
-	OP(FG, C, D, A, B, 15, 14, 0xd8a1e681);
-	OP(FG, B, C, D, A, 4, 20, 0xe7d3fbc8);
-	OP(FG, A, B, C, D, 9, 5, 0x21e1cde6);
-	OP(FG, D, A, B, C, 14, 9, 0xc33707d6);
-	OP(FG, C, D, A, B, 3, 14, 0xf4d50d87);
-	OP(FG, B, C, D, A, 8, 20, 0x455a14ed);
-	OP(FG, A, B, C, D, 13, 5, 0xa9e3e905);
-	OP(FG, D, A, B, C, 2, 9, 0xfcefa3f8);
-	OP(FG, C, D, A, B, 7, 14, 0x676f02d9);
-	OP(FG, B, C, D, A, 12, 20, 0x8d2a4c8a);
-# endif
-
-	/* Round 3 */
-# if MD5_SIZE_VS_SPEED == 1
-	for (i = 0; i < 4; i++) {
-		OP(FH, A, B, C, D, (int) (*pp++), 4, *pc++);
-		OP(FH, D, A, B, C, (int) (*pp++), 11, *pc++);
-		OP(FH, C, D, A, B, (int) (*pp++), 16, *pc++);
-		OP(FH, B, C, D, A, (int) (*pp++), 23, *pc++);
-	}
-# else
-	OP(FH, A, B, C, D, 5, 4, 0xfffa3942);
-	OP(FH, D, A, B, C, 8, 11, 0x8771f681);
-	OP(FH, C, D, A, B, 11, 16, 0x6d9d6122);
-	OP(FH, B, C, D, A, 14, 23, 0xfde5380c);
-	OP(FH, A, B, C, D, 1, 4, 0xa4beea44);
-	OP(FH, D, A, B, C, 4, 11, 0x4bdecfa9);
-	OP(FH, C, D, A, B, 7, 16, 0xf6bb4b60);
-	OP(FH, B, C, D, A, 10, 23, 0xbebfbc70);
-	OP(FH, A, B, C, D, 13, 4, 0x289b7ec6);
-	OP(FH, D, A, B, C, 0, 11, 0xeaa127fa);
-	OP(FH, C, D, A, B, 3, 16, 0xd4ef3085);
-	OP(FH, B, C, D, A, 6, 23, 0x04881d05);
-	OP(FH, A, B, C, D, 9, 4, 0xd9d4d039);
-	OP(FH, D, A, B, C, 12, 11, 0xe6db99e5);
-	OP(FH, C, D, A, B, 15, 16, 0x1fa27cf8);
-	OP(FH, B, C, D, A, 2, 23, 0xc4ac5665);
-# endif
-
-	/* Round 4 */
-# if MD5_SIZE_VS_SPEED == 1
-	for (i = 0; i < 4; i++) {
-		OP(FI, A, B, C, D, (int) (*pp++), 6, *pc++);
-		OP(FI, D, A, B, C, (int) (*pp++), 10, *pc++);
-		OP(FI, C, D, A, B, (int) (*pp++), 15, *pc++);
-		OP(FI, B, C, D, A, (int) (*pp++), 21, *pc++);
-	}
-# else
-	OP(FI, A, B, C, D, 0, 6, 0xf4292244);
-	OP(FI, D, A, B, C, 7, 10, 0x432aff97);
-	OP(FI, C, D, A, B, 14, 15, 0xab9423a7);
-	OP(FI, B, C, D, A, 5, 21, 0xfc93a039);
-	OP(FI, A, B, C, D, 12, 6, 0x655b59c3);
-	OP(FI, D, A, B, C, 3, 10, 0x8f0ccc92);
-	OP(FI, C, D, A, B, 10, 15, 0xffeff47d);
-	OP(FI, B, C, D, A, 1, 21, 0x85845dd1);
-	OP(FI, A, B, C, D, 8, 6, 0x6fa87e4f);
-	OP(FI, D, A, B, C, 15, 10, 0xfe2ce6e0);
-	OP(FI, C, D, A, B, 6, 15, 0xa3014314);
-	OP(FI, B, C, D, A, 13, 21, 0x4e0811a1);
-	OP(FI, A, B, C, D, 4, 6, 0xf7537e82);
-	OP(FI, D, A, B, C, 11, 10, 0xbd3af235);
-	OP(FI, C, D, A, B, 2, 15, 0x2ad7d2bb);
-	OP(FI, B, C, D, A, 9, 21, 0xeb86d391);
-# undef OP
-# endif
-	/* Add checksum to the starting values */
-	ctx->A = A_save + A;
-	ctx->B = B_save + B;
-	ctx->C = C_save + C;
-	ctx->D = D_save + D;
-#endif
-}
-#undef FF
-#undef FG
-#undef FH
-#undef FI
-
-/* Feed data through a temporary buffer to call md5_hash_aligned_block()
- * with chunks of data that are 4-byte aligned and a multiple of 64 bytes.
- * This function's internal buffer remembers previous data until it has 64
- * bytes worth to pass on.  Call md5_end() to flush this buffer. */
-void FAST_FUNC md5_hash(md5_ctx_t *ctx, const void *buffer, size_t len)
-{
-	unsigned bufpos = ctx->total64 & 63;
-	unsigned remaining;
-
-	/* RFC 1321 specifies the possible length of the file up to 2^64 bits.
-	 * Here we only track the number of bytes.  */
-	ctx->total64 += len;
-#if 0
-	remaining = 64 - bufpos;
-
-	/* Hash whole blocks */
-	while (len >= remaining) {
-		memcpy(ctx->wbuffer + bufpos, buffer, remaining);
-		buffer = (const char *)buffer + remaining;
-		len -= remaining;
-		remaining = 64;
-		bufpos = 0;
-		md5_process_block64(ctx);
-	}
-
-	/* Save last, partial blosk */
-	memcpy(ctx->wbuffer + bufpos, buffer, len);
-#else
-	/* Tiny bit smaller code */
-	while (1) {
-		remaining = 64 - bufpos;
-		if (remaining > len)
-			remaining = len;
-		/* Copy data into aligned buffer */
-		memcpy(ctx->wbuffer + bufpos, buffer, remaining);
-		len -= remaining;
-		buffer = (const char *)buffer + remaining;
-		bufpos += remaining;
-		/* clever way to do "if (bufpos != 64) break; ... ; bufpos = 0;" */
-		bufpos -= 64;
-		if (bufpos != 0)
-			break;
-		/* Buffer is filled up, process it */
-		md5_process_block64(ctx);
-		/*bufpos = 0; - already is */
-	}
-#endif
-}
-
-/* Process the remaining bytes in the buffer and put result from CTX
- * in first 16 bytes following RESBUF.  The result is always in little
- * endian byte order, so that a byte-wise output yields to the wanted
- * ASCII representation of the message digest.
- */
-void FAST_FUNC md5_end(md5_ctx_t *ctx, void *resbuf)
-{
-	/* MD5 stores total in LE, need to swap on BE arches: */
-	common64_end(ctx, (process_block64_func*) md5_process_block64, /*swap_needed:*/ BB_BIG_ENDIAN);
-
-	/* The MD5 result is in little endian byte order.
-	 * We (ab)use the fact that A-D are consecutive in memory.
-	 */
-#if BB_BIG_ENDIAN
-	ctx->A = SWAP_LE32(ctx->A);
-	ctx->B = SWAP_LE32(ctx->B);
-	ctx->C = SWAP_LE32(ctx->C);
-	ctx->D = SWAP_LE32(ctx->D);
-#endif
-	memcpy(resbuf, &ctx->A, sizeof(ctx->A) * 4);
-}