Re: [lwip-devel] Is there any interest in an inlined IP checksum compile-time option?

[address@hidden]

bill wrote:
> I’ve found (on just one target) that inlining the IP checksum yields 
> from 1.25 to 1.5% more throughput transmitting packets. Since the 
> compiler already touches the 16 (u16_t) words and can sum in a 
> register variable, the additional code isn’t much more compared to the 
> function call plus htons that is used currently (the inline sum works 
> out to be in network order which may be true only on a little endian 
> processor). The offset word is always 0 and can be skipped. I did this 
> inlining is only in the ip_output_if function. I know it’s not much, 
> but it’s wasted processor time nonetheless.

Did you mean inlining inet_chksum_pseudo()? Because LWIP_CHKSUM() can 
already be aligned as it is a define. Of course, the IP code could use 
that define instead of using the function that directly calls this 
(inet_chksum())! The problem with inlining inet_chksum_pseudo() would be 
that it has to be in a header file to do that. And after all, I think 
creating an optimized LWIP_CHKSUM routine in assembly has more impact 
than inlining this code...

Simon