Eliminate redundant memcpy of IV material. Performance improvement varies

from platform to platform and can be as large as 20%.

crypto/aes/aes_cbc.c

@@ -65,7 +65,7 @@ void AES_cbc_encrypt(const unsigned char *in, unsigned char *out,
 
 	unsigned long n;
 	unsigned long len = length;
-	unsigned char tmp[AES_BLOCK_SIZE];
+	unsigned char tmp[AES_BLOCK_SIZE], *iv = ivec;
 
 	assert(in && out && key && ivec);
 	assert((AES_ENCRYPT == enc)||(AES_DECRYPT == enc));
@@ -73,22 +73,40 @@ void AES_cbc_encrypt(const unsigned char *in, unsigned char *out,
 	if (AES_ENCRYPT == enc) {
 		while (len >= AES_BLOCK_SIZE) {
 			for(n=0; n < AES_BLOCK_SIZE; ++n)
-				tmp[n] = in[n] ^ ivec[n];
-			AES_encrypt(tmp, out, key);
-			memcpy(ivec, out, AES_BLOCK_SIZE);
+				out[n] = in[n] ^ iv[n];
+			AES_encrypt(out, out, key);
+			iv = out;
 			len -= AES_BLOCK_SIZE;
 			in += AES_BLOCK_SIZE;
 			out += AES_BLOCK_SIZE;
 		}
 		if (len) {
 			for(n=0; n < len; ++n)
-				tmp[n] = in[n] ^ ivec[n];
+				tmp[n] = in[n] ^ iv[n];
 			for(n=len; n < AES_BLOCK_SIZE; ++n)
-				tmp[n] = ivec[n];
+				tmp[n] = iv[n];
 			AES_encrypt(tmp, tmp, key);
 			memcpy(out, tmp, AES_BLOCK_SIZE);
-			memcpy(ivec, tmp, AES_BLOCK_SIZE);
+			iv = out;
 		}			
+		memcpy(ivec,iv,AES_BLOCK_SIZE);
+	} else if (in != out) {
+		while (len >= AES_BLOCK_SIZE) {
+			AES_decrypt(in, out, key);
+			for(n=0; n < AES_BLOCK_SIZE; ++n)
+				out[n] ^= iv[n];
+			iv = in;
+			len -= AES_BLOCK_SIZE;
+			in  += AES_BLOCK_SIZE;
+			out += AES_BLOCK_SIZE;
+		}
+		if (len) {
+			AES_decrypt(in,tmp,key);
+			for(n=0; n < len; ++n)
+				out[n] = tmp[n] ^ iv[n];
+			iv = in;
+		}
+		memcpy(ivec,iv,AES_BLOCK_SIZE);
 	} else {
 		while (len >= AES_BLOCK_SIZE) {
 			memcpy(tmp, in, AES_BLOCK_SIZE);