Camellia update. Quoting camellia.c:

/*
 * This release balances code size and performance. In particular key
 * schedule setup is fully unrolled, because doing so *significantly*
 * reduces amount of instructions per setup round and code increase is
 * justifiable. In block functions on the other hand only inner loops
 * are unrolled, as full unroll gives only nominal performance boost,
 * while code size grows 4 or 7 times. Also, unlike previous versions
 * this one "encourages" compiler to keep intermediate variables in
 * registers, which should give better "all round" results, in other
 * words reasonable performance even with not so modern compilers.
 */

1 files changed, 59 insertions, 189 deletions

diff --git a/crypto/camellia/cmll_cbc.c b/crypto/camellia/cmll_cbc.c
index 4141a7b59b..545e6c1f41 100644
--- a/crypto/camellia/cmll_cbc.c
+++ b/crypto/camellia/cmll_cbc.c
@@ -55,219 +55,89 @@
 # endif
 #endif
 #include <assert.h>
-#include <stdio.h>
 #include <string.h>
 
 #include <openssl/camellia.h>
 #include "cmll_locl.h"
-
 void Camellia_cbc_encrypt(const unsigned char *in, unsigned char *out,
-		     const unsigned long length, const CAMELLIA_KEY *key,
-		     unsigned char *ivec, const int enc) {
+	const unsigned long length, const CAMELLIA_KEY *key,
+	unsigned char *ivec, const int enc) 
+	{
 
 	unsigned long n;
 	unsigned long len = length;
+	unsigned char tmp[CAMELLIA_BLOCK_SIZE];
 	const unsigned char *iv = ivec;
-	union {	u32 t32[CAMELLIA_BLOCK_SIZE/sizeof(u32)];
-		u8  t8 [CAMELLIA_BLOCK_SIZE]; } tmp;
-	const union { long one; char little; } camellia_endian = {1};
-
 
 	assert(in && out && key && ivec);
 	assert((CAMELLIA_ENCRYPT == enc)||(CAMELLIA_DECRYPT == enc));
 
-	if(((size_t)in|(size_t)out|(size_t)ivec) % sizeof(u32) == 0)
+	if (CAMELLIA_ENCRYPT == enc) 
 		{
-		if (CAMELLIA_ENCRYPT == enc)
+		while (len >= CAMELLIA_BLOCK_SIZE) 
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				XOR4WORD2((u32 *)out,
-					(u32 *)in, (u32 *)iv);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				key->enc(key->rd_key, (u32 *)out);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				iv = out;
-				len -= CAMELLIA_BLOCK_SIZE;
-				in += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				for(n=0; n < len; ++n)
-					out[n] = in[n] ^ iv[n];
-				for(n=len; n < CAMELLIA_BLOCK_SIZE; ++n)
-					out[n] = iv[n];
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				key->enc(key->rd_key, (u32 *)out);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				iv = out;
-				}
-			memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+			for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
+				out[n] = in[n] ^ iv[n];
+			Camellia_encrypt(out, out, key);
+			iv = out;
+			len -= CAMELLIA_BLOCK_SIZE;
+			in += CAMELLIA_BLOCK_SIZE;
+			out += CAMELLIA_BLOCK_SIZE;
 			}
-		else if (in != out)
+		if (len) 
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				memcpy(out,in,CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				key->dec(key->rd_key,(u32 *)out);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				XOR4WORD((u32 *)out, (u32 *)iv);
-				iv = in;
-				len -= CAMELLIA_BLOCK_SIZE;
-				in  += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->dec(key->rd_key, tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				for(n=0; n < len; ++n)
-					out[n] = tmp.t8[n] ^ iv[n];
-				iv = in;
-				}
-			memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+			for(n=0; n < len; ++n)
+				out[n] = in[n] ^ iv[n];
+			for(n=len; n < CAMELLIA_BLOCK_SIZE; ++n)
+				out[n] = iv[n];
+			Camellia_encrypt(out, out, key);
+			iv = out;
 			}
-		else /* in == out */
+		memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+		} 
+	else if (in != out) 
+		{
+		while (len >= CAMELLIA_BLOCK_SIZE) 
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				key->dec(key->rd_key, (u32 *)out);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				XOR4WORD((u32 *)out, (u32 *)ivec);
-				memcpy(ivec, tmp.t8, CAMELLIA_BLOCK_SIZE);
-				len -= CAMELLIA_BLOCK_SIZE;
-				in += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				key->dec(key->rd_key,(u32 *)out);
-				if (camellia_endian.little)
-					SWAP4WORD((u32 *)out);
-				for(n=0; n < len; ++n)
-					out[n] ^= ivec[n];
-				for(n=len; n < CAMELLIA_BLOCK_SIZE; ++n)
-					out[n] = tmp.t8[n];
-				memcpy(ivec, tmp.t8, CAMELLIA_BLOCK_SIZE);
-				}
+			Camellia_decrypt(in, out, key);
+			for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
+				out[n] ^= iv[n];
+			iv = in;
+			len -= CAMELLIA_BLOCK_SIZE;
+			in  += CAMELLIA_BLOCK_SIZE;
+			out += CAMELLIA_BLOCK_SIZE;
 			}
-		}
-	else /* no aligned */
-		{
-		if (CAMELLIA_ENCRYPT == enc)
+		if (len) 
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
-					tmp.t8[n] = in[n] ^ iv[n];
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->enc(key->rd_key, tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				memcpy(out, tmp.t8, CAMELLIA_BLOCK_SIZE);
-				iv = out;
-				len -= CAMELLIA_BLOCK_SIZE;
-				in += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				for(n=0; n < len; ++n)
-					tmp.t8[n] = in[n] ^ iv[n];
-				for(n=len; n < CAMELLIA_BLOCK_SIZE; ++n)
-					tmp.t8[n] = iv[n];
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->enc(key->rd_key, tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				memcpy(out, tmp.t8, CAMELLIA_BLOCK_SIZE);
-				iv = out;
-				}
-			memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+			Camellia_decrypt(in,tmp,key);
+			for(n=0; n < len; ++n)
+				out[n] = tmp[n] ^ iv[n];
+			iv = in;
 			}
-		else if (in != out)
+		memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+		} 
+	else 
+		{
+		while (len >= CAMELLIA_BLOCK_SIZE) 
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				memcpy(tmp.t8,in,CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->dec(key->rd_key,tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
-					out[n] = tmp.t8[n] ^ iv[n];
-				iv = in;
-				len -= CAMELLIA_BLOCK_SIZE;
-				in  += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->dec(key->rd_key, tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				for(n=0; n < len; ++n)
-					out[n] = tmp.t8[n] ^ iv[n];
-				iv = in;
-				}
-			memcpy(ivec,iv,CAMELLIA_BLOCK_SIZE);
+			memcpy(tmp, in, CAMELLIA_BLOCK_SIZE);
+			Camellia_decrypt(in, out, key);
+			for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
+				out[n] ^= ivec[n];
+			memcpy(ivec, tmp, CAMELLIA_BLOCK_SIZE);
+			len -= CAMELLIA_BLOCK_SIZE;
+			in += CAMELLIA_BLOCK_SIZE;
+			out += CAMELLIA_BLOCK_SIZE;
 			}
-		else
+		if (len)
 			{
-			while (len >= CAMELLIA_BLOCK_SIZE)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->dec(key->rd_key, tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				for(n=0; n < CAMELLIA_BLOCK_SIZE; ++n)
-					tmp.t8[n] ^= ivec[n];
-				memcpy(ivec, in, CAMELLIA_BLOCK_SIZE);
-				memcpy(out, tmp.t8, CAMELLIA_BLOCK_SIZE);
-				len -= CAMELLIA_BLOCK_SIZE;
-				in += CAMELLIA_BLOCK_SIZE;
-				out += CAMELLIA_BLOCK_SIZE;
-				}
-			if (len)
-				{
-				memcpy(tmp.t8, in, CAMELLIA_BLOCK_SIZE);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				key->dec(key->rd_key,tmp.t32);
-				if (camellia_endian.little)
-					SWAP4WORD(tmp.t32);
-				for(n=0; n < len; ++n)
-					tmp.t8[n] ^= ivec[n];
-				memcpy(ivec, in, CAMELLIA_BLOCK_SIZE);
-				memcpy(out,tmp.t8,len);
-				}
+			memcpy(tmp, in, CAMELLIA_BLOCK_SIZE);
+			Camellia_decrypt(tmp, out, key);
+			for(n=0; n < len; ++n)
+				out[n] ^= ivec[n];
+			for(n=len; n < CAMELLIA_BLOCK_SIZE; ++n)
+				out[n] = tmp[n];
+			memcpy(ivec, tmp, CAMELLIA_BLOCK_SIZE);
 			}
 		}
-}
+	}