diff --git a/README.md b/README.md
index 11cc1d5..2daae57 100644
--- a/README.md
+++ b/README.md
@@ -20,9 +20,9 @@ make install
 ```c
 #include <simplecrypto.h>
 ```
-2. Call functions. Don't forget to `free` the returned digest.
+2. Call functions.
 ```c
-uint8_t* md5(const uint8_t *data, size_t data_len);
+uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]);
 
 TEADAT* tea_encrypt_qq(const TEA t[4], const TEADAT* src);
 TEADAT* tea_encrypt(const TEA t[4], const uint32_t sumtable[0x10], const TEADAT* src);
diff --git a/md5.c b/md5.c
index 68c093d..c509001 100644
--- a/md5.c
+++ b/md5.c
@@ -49,7 +49,11 @@ static uint32_t to_uint32(const uint8_t *bytes) {
 }
 
 uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]) {
-	uint32_t w[16];
+	#ifdef WORDS_BIGENDIAN
+		uint32_t w[16];
+	#else
+		uint32_t* w;
+	#endif
 
 	// These vars will contain the hash
 	// Initialize variables - simple count in nibbles:
@@ -85,8 +89,12 @@ uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]) {
 	//for each 512-bit chunk of message:
 	for(offset=0; offset<new_len; offset += (512/8)) {
 		// break chunk into sixteen 32-bit words w[j], 0 ≤ j ≤ 15
-		for (i = 0; i < 16; i++)
-			w[i] = to_uint32(msg + offset + i*4);
+		#ifdef WORDS_BIGENDIAN
+			for (i = 0; i < 16; i++)
+				w[i] = to_uint32(msg + offset + i*4);
+		#else
+			w = (uint32_t*)(msg + offset);
+		#endif
 
 		// Initialize hash value for this chunk:
 		a = h0;
@@ -95,28 +103,41 @@ uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]) {
 		d = h3;
 
 		// Main loop:
-		for(i = 0; i<64; i++) {
-
-			if (i < 16) {
-				f = (b & c) | ((~b) & d);
-				g = i;
-			} else if (i < 32) {
-				f = (d & b) | ((~d) & c);
-				g = (5*i + 1) % 16;
-			} else if (i < 48) {
-				f = b ^ c ^ d;
-				g = (3*i + 5) % 16;          
-			} else {
-				f = c ^ (b | (~d));
-				g = (7*i) % 16;
-			}
-
+		for(i = 0; i < 16; i++) {
+			f = (b & c) | ((~b) & d);
+			g = i;
 			temp = d;
 			d = c;
 			c = b;
-			b = b + LEFTROTATE((a + f + k[i] + w[g]), r[i]);
+			b += LEFTROTATE((a + f + k[i] + w[g]), r[i]);
+			a = temp;
+		}
+		for(i = 16; i < 32; i++) {
+			f = (d & b) | ((~d) & c);
+			g = (5*i + 1) % 16;
+			temp = d;
+			d = c;
+			c = b;
+			b += LEFTROTATE((a + f + k[i] + w[g]), r[i]);
+			a = temp;
+		}
+		for(i = 32; i < 48; i++) {
+			f = b ^ c ^ d;
+			g = (3*i + 5) % 16;
+			temp = d;
+			d = c;
+			c = b;
+			b += LEFTROTATE((a + f + k[i] + w[g]), r[i]);
+			a = temp;
+		}
+		for(i = 48; i < 64; i++) {
+			f = c ^ (b | (~d));
+			g = (7*i) % 16;
+			temp = d;
+			d = c;
+			c = b;
+			b += LEFTROTATE((a + f + k[i] + w[g]), r[i]);
 			a = temp;
-
 		}
 
 		// Add this chunk's hash to result so far:
@@ -124,7 +145,6 @@ uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]) {
 		h1 += b;
 		h2 += c;
 		h3 += d;
-
 	}
 
 	// cleanup
@@ -139,11 +159,12 @@ uint8_t* md5(const uint8_t *data, size_t data_len, uint8_t digest[16]) {
 }
 
 #ifdef TEST_SIMPLE_CRYPTO
+#include <stdio.h>
 int main(int argc, char **argv) {
 	char *msg;
 	size_t len;
 	int i;
-	uint8_t* result;
+	uint8_t result[16];
 
 	if (argc < 2) {
 		printf("usage: %s 'string'\n", argv[0]);
@@ -152,12 +173,11 @@ int main(int argc, char **argv) {
 	msg = argv[1];
 
 	len = strlen(msg);
-	result = md5((uint8_t*)msg, len);
+	md5((uint8_t*)msg, len, result);
 	// display result
 	for (i = 0; i < 16; i++)
 		printf("%2.2x", result[i]);
-	puts("");
-	free(result);
+	putchar('\n');
 
 	return 0;
 }
diff --git a/tea.c b/tea.c
index 2ad8ef6..27ea46b 100644
--- a/tea.c
+++ b/tea.c
@@ -29,49 +29,7 @@ const static uint32_t qqsumtable[0x10] = {
 };
 
 TEADAT* tea_encrypt_qq(const TEA t[4], const TEADAT* src) {
-	int64_t lens = src->len;
-	int64_t fill = 10 - (lens+1)%8;
-	int64_t dstlen = fill+lens+7;
-	uint8_t* dstdat = (uint8_t*)malloc(dstlen);
-	((uint32_t*)dstdat)[0] = rand();
-	((uint32_t*)dstdat)[1] = rand();
-	((uint32_t*)dstdat)[2] = rand();
-	dstdat[0] = (fill-3)|0xF8; // 存储pad长度
-	memcpy(dstdat+fill, src->data, lens);
-
-	uint64_t iv1 = 0, iv2 = 0, holder;
-	for(int64_t i = 0; i < dstlen/8; i++) {
-		#ifdef WORDS_BIGENDIAN
-			uint64_t block = ((uint64_t*)dstdat)[i];
-		#else
-			uint64_t block = __builtin_bswap64(((uint64_t*)dstdat)[i]);
-		#endif
-		holder = block ^ iv1;
-
-		iv1 = holder;
-		uint32_t v1 = holder;
-		iv1 >>= 32;
-		uint32_t v0 = iv1;
-		for (int i = 0; i < 0x10; i++) {
-			v0 += (v1 + qqsumtable[i]) ^ ((v1 << 4) + t[0]) ^ ((v1 >> 5) + t[1]);
-			v1 += (v0 + qqsumtable[i]) ^ ((v0 << 4) + t[2]) ^ ((v0 >> 5) + t[3]);
-		}
-		iv1 = ((uint64_t)v0<<32) | (uint64_t)v1;
-
-		iv1 = iv1 ^ iv2;
-		iv2 = holder;
-		#ifdef WORDS_BIGENDIAN
-			((uint64_t*)dstdat)[i] = iv1;
-		#else
-			((uint64_t*)dstdat)[i] = __builtin_bswap64(iv1);
-		#endif
-	}
-
-	TEADAT* dst = (TEADAT*)malloc(sizeof(TEADAT));
-	dst->len = dstlen;
-	dst->data = dstdat;
-	dst->ptr = dstdat;
-	return dst;
+	return tea_encrypt(t, qqsumtable, src);
 }
 
 TEADAT* tea_encrypt(const TEA t[4], const uint32_t sumtable[0x10], const TEADAT* src) {
@@ -159,45 +117,7 @@ TEADAT* tea_encrypt_native_endian(const TEA t[4], const uint32_t sumtable[0x10],
 }
 
 TEADAT* tea_decrypt_qq(const TEA t[4], const TEADAT* src) {
-	if (src->len < 16 || (src->len)%8 != 0) {
-		return NULL;
-	}
-	uint8_t* dstdat = (uint8_t*)malloc(src->len);
-
-	uint64_t iv1, iv2 = 0, holder = 0;
-	for(int64_t i = 0; i < src->len/8; i++) {
-		#ifdef WORDS_BIGENDIAN
-			iv1 = ((uint64_t*)(src->data))[i];
-		#else
-			iv1 = __builtin_bswap64(((uint64_t*)(src->data))[i]);
-		#endif
-
-		iv2 ^= iv1;
-
-		uint32_t v1 = iv2;
-		iv2 >>= 32;
-		uint32_t v0 = iv2;
-		for (int i = 0x0f; i >= 0; i--) {
-			v1 -= (v0 + qqsumtable[i]) ^ ((v0 << 4) + t[2]) ^ ((v0 >> 5) + t[3]);
-			v0 -= (v1 + qqsumtable[i]) ^ ((v1 << 4) + t[0]) ^ ((v1 >> 5) + t[1]);
-		}
-		iv2 = ((uint64_t)v0<<32) | (uint64_t)v1;
-
-		#ifdef WORDS_BIGENDIAN
-			((uint64_t*)dstdat)[i] = iv2^holder;
-		#else
-			((uint64_t*)dstdat)[i] = __builtin_bswap64(iv2^holder);
-		#endif
-
-		holder = iv1;
-	}
-
-	TEADAT* dst = (TEADAT*)malloc(sizeof(TEADAT));
-	int start = (dstdat[0]&7)+3;
-	dst->len = src->len-7-start;
-	dst->data = dstdat+start;
-	dst->ptr = dstdat;
-	return dst;
+	return tea_decrypt(t, qqsumtable, src);
 }
 
 TEADAT* tea_decrypt(const TEA t[4], const uint32_t sumtable[0x10], const TEADAT* src) {