#include <stdint.h>
#include <stddef.h>

/* Forward Declarations */
static inline uint16_t read_be16(const uint8_t *ptr);
static uint32_t sum_words(const uint8_t *data, size_t len);
static inline uint16_t fold_carries(uint32_t sum);
/* -------------------- */

/* RFC 1071: Compute ICMP checksum (16-bit one's complement sum) */
uint16_t
icmp_checksum(const void *data, size_t len)
{
	uint32_t sum;

	sum = sum_words(data, len);
	sum = fold_carries(sum);

	return (uint16_t)~sum;
}

/* Sum all 16-bit words */
static uint32_t
sum_words(const uint8_t *data, size_t len)
{
	const uint8_t *ptr = data;
	uint32_t sum = 0;
	size_t words = len >> 1;

	/* Duff's device: unroll loop by 4 */
	if (words > 0)
	{
		size_t n = (words + 3) >> 2;
		switch (words & 3) {
		case 0: do { sum += read_be16(ptr); ptr += 2;
		case 3:		 sum += read_be16(ptr); ptr += 2;
		case 2:		 sum += read_be16(ptr); ptr += 2;
		case 1:		 sum += read_be16(ptr); ptr += 2;
				} while (--n > 0);
		}
	}

	/* Handle odd byte if present */
	if (len & 1)
		sum += (uint32_t)(*ptr << 8);

	return sum;
}

/* Fold 32-bit sum to 16 bits by adding carry bits back */
static inline uint16_t
fold_carries(uint32_t sum)
{
	while (sum >> 16)
		sum = (sum & 0xFFFF) + (sum >> 16);
	return (uint16_t)sum;
}

/* Read 16-bit word in network byte order (big-endian) */
static inline uint16_t
read_be16(const uint8_t *ptr)
{
	return (uint16_t)((*ptr << 8) | ptr[1]);
}