[dpdk-dev] [PATCH v2 1/1] net: fix aliasing issue in checksum computation

[Olivier Matz]

On Mon, Oct 18, 2021 at 09:35:41AM +0200, Morten Brørup wrote:
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Georg Sauthoff
> > Sent: Sunday, 17 October 2021 22.37
> 
> +Ferruh, as delegate to v1 in Patchwork.
> 
> > 
> > That means a superfluous cast is removed and aliasing through a uint8_t
> > pointer is eliminated. NB: The C standard specifies that a unsigned
> > char
> > pointer may alias while the C standard doesn't include such requirement
> > for uint8_t pointers.
> > 
> > Also simplified the loop since a modern C compiler can speed up (i.e.
> > auto-vectorize) it in a similar way. For example, GCC auto-vectorizes
> > it
> > for Haswell using AVX registers while halving the number of
> > instructions
> > in the generated code.
> > 
> > Signed-off-by: Georg Sauthoff <mail at gms.tf>
> > ---
> >  lib/net/rte_ip.h | 27 ++++++++-------------------
> >  1 file changed, 8 insertions(+), 19 deletions(-)
> > 
> > diff --git a/lib/net/rte_ip.h b/lib/net/rte_ip.h
> > index 05948b69b7..1b8c6519a9 100644
> > --- a/lib/net/rte_ip.h
> > +++ b/lib/net/rte_ip.h
> > @@ -141,29 +141,18 @@ rte_ipv4_hdr_len(const struct rte_ipv4_hdr
> > *ipv4_hdr)
> >  static inline uint32_t
> >  __rte_raw_cksum(const void *buf, size_t len, uint32_t sum)
> >  {
> > -	/* workaround gcc strict-aliasing warning */
> > -	uintptr_t ptr = (uintptr_t)buf;
> > +	/* extend strict-aliasing rules */
> >  	typedef uint16_t __attribute__((__may_alias__)) u16_p;
> > -	const u16_p *u16_buf = (const u16_p *)ptr;
> > -
> > -	while (len >= (sizeof(*u16_buf) * 4)) {
> > -		sum += u16_buf[0];
> > -		sum += u16_buf[1];
> > -		sum += u16_buf[2];
> > -		sum += u16_buf[3];
> > -		len -= sizeof(*u16_buf) * 4;
> > -		u16_buf += 4;
> > -	}
> > -	while (len >= sizeof(*u16_buf)) {
> > +	const u16_p *u16_buf = (const u16_p *)buf;
> > +	const u16_p *end = u16_buf + len / sizeof(*u16_buf);
> > +
> > +	for (; u16_buf != end; ++u16_buf)
> >  		sum += *u16_buf;
> > -		len -= sizeof(*u16_buf);
> > -		u16_buf += 1;
> > -	}
> > 
> > -	/* if length is in odd bytes */
> > -	if (len == 1) {
> > +	/* if length is odd, keeping it byte order independent */
> > +	if (unlikely(len % 2)) {
> >  		uint16_t left = 0;
> > -		*(uint8_t *)&left = *(const uint8_t *)u16_buf;
> > +		*(unsigned char*)&left = *(const unsigned char *)end;
> >  		sum += left;
> >  	}
> > 
> > --
> > 2.31.1
> > 
> 
> Great work documenting your thoughts behind this patch, Georg! I, for one, didn't know about the aliasing difference between uint8_t and unsigned char. :-)
> 
> After taking a good look at v2 and the Godbolt reference to confirm the claimed benefits, there can be no doubts about this patch.

+1, thanks for the good documentation

> Reviewed-by: Morten Brørup <mb at smartsharesystems.com>

Acked-by: Olivier Matz <olivier.matz at 6wind.com>