利用tile SIMD指令加速计算TCP伪头校验和

static inline
uint32_t tile_fast_tcp_seed(uint32_t saddr, uint32_t daddr, uint16_t size)
{
	uint32_t cs = 0;
	cs = __insn_sadh_u(__insn_bytex(saddr), 0);
	cs = __insn_sadh_u(cs, 0);
	cs = cs + __insn_sadh_u(__insn_sadh_u(__insn_bytex(daddr), 0), 0);
	cs = __insn_sadh_u(cs, 0);
	cs = cs + __insn_sadh_u(__insn_sadh_u((size << 16) | 0x0006, 0), 0);
	cs = __insn_sadh_u(cs, 0);
	return cs;
}