libsystemd-network: Speed up checksum computation using 64 bit integers

Improve the checksum computation by using 64 bit integers instead of the 16 bit
integers in the existing implementation. This change speeds up the computation
with approximately 78% both on 64 bit and 32 bit systems.

Please see RFC 1071 for details.

2 files changed, 40 insertions, 13 deletions

diff --git a/src/libsystemd-network/dhcp-internal.h b/src/libsystemd-network/dhcp-internal.h
index 324c2a8d7a..2188a7f89a 100644
--- a/src/libsystemd-network/dhcp-internal.h
+++ b/src/libsystemd-network/dhcp-internal.h
@@ -48,7 +48,7 @@ int dhcp_option_parse(DHCPMessage *message, size_t len,
 int dhcp_message_init(DHCPMessage *message, uint8_t op, uint32_t xid, uint8_t type,
                       uint8_t **opt, size_t *optlen);
 
-uint16_t dhcp_packet_checksum(void *buf, int len);
+uint16_t dhcp_packet_checksum(void *buf, size_t len);
 
 void dhcp_packet_append_ip_headers(DHCPPacket *packet, be32_t source_addr,
                                    uint16_t source, be32_t destination_addr,
diff --git a/src/libsystemd-network/dhcp-packet.c b/src/libsystemd-network/dhcp-packet.c
index 102ed096af..fba9c46ef2 100644
--- a/src/libsystemd-network/dhcp-packet.c
+++ b/src/libsystemd-network/dhcp-packet.c
@@ -59,21 +59,48 @@ int dhcp_message_init(DHCPMessage *message, uint8_t op, uint32_t xid,
         return 0;
 }
 
-uint16_t dhcp_packet_checksum(void *buf, int len) {
-        uint32_t sum;
-        uint16_t *check;
-        int i;
-        uint8_t *odd;
+uint16_t dhcp_packet_checksum(void *buf, size_t len) {
+        uint64_t *buf_64 = buf;
+        uint64_t *end_64 = (uint64_t*)buf + (len / sizeof(uint64_t));
+        uint32_t *buf_32;
+        uint16_t *buf_16;
+        uint8_t *buf_8;
+        uint64_t sum = 0;
+
+        while (buf_64 < end_64) {
+                sum += *buf_64;
+                if (sum < *buf_64)
+                        sum++;
+
+                buf_64 ++;
+        }
+
+        buf_32 = (uint32_t*)buf_64;
+
+        if (len & sizeof(uint32_t)) {
+                sum += *buf_32;
+                if (sum < *buf_32)
+                        sum++;
+
+                buf_32 ++;
+        }
+
+        buf_16 = (uint16_t*)buf_32;
 
-        sum = 0;
-        check = buf;
+        if (len & sizeof(uint16_t)) {
+                sum += *buf_16;
+                if (sum < *buf_16)
+                        sum ++;
+
+                buf_16 ++;
+        }
 
-        for (i = 0; i < len / 2 ; i++)
-                sum += check[i];
+        buf_8 = (uint8_t*)buf_16;
 
-        if (len & 0x01) {
-                odd = buf;
-                sum += odd[len - 1];
+        if (len & sizeof(uint8_t)) {
+                sum += *buf_8;
+                if (sum < *buf_8)
+                        sum++;
         }
 
         while (sum >> 16)