[dpdk-dev] [PATCH v3 5/7] examples/l3fwd: add neon support for l3fwd

[Sekhar, Ashwin]

On Thu, 2017-05-11 at 18:01 +0800, Jianbo Liu wrote:
> On 11 May 2017 at 17:49, Sekhar, Ashwin <Ashwin.Sekhar at cavium.com>
> wrote:
> > 
> > Hi Jianbo,
> > 
> > Thanks for v3. Small compilation error. See inline comment.
> > Otherwise
> > it looks fine.
> > 
> > On Thu, 2017-05-11 at 17:25 +0800, Jianbo Liu wrote:
> > > 
> > > Use ARM NEON intrinsics to accelerate l3 fowarding.
> > > 
> > > Signed-off-by: Jianbo Liu <jianbo.liu at linaro.org>
> > > ---
> > [...]
> > 
> > > 
> > > +/**
> > > + * Process one packet:
> > > + * Update source and destination MAC addresses in the ethernet
> > > header.
> > > + * Perform RFC1812 checks and updates for IPV4 packets.
> > > + */
> > > +static inline void
> > > +process_packet(struct rte_mbuf *pkt, uint16_t *dst_port)
> > > +{
> > > +     struct ether_hdr *eth_hdr;
> > > +     uint32x4_t te, ve;
> > > +
> > > +     eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
> > > +
> > > +     te = vld1q_u32((uint32_t *)eth_hdr);
> > > +     ve = vreinterpretq_u32_s32(val_eth[dst_port[0]]);
> > > +
> > > +
> > > +     rfc1812_process((struct ipv4_hdr *)(eth_hdr + 1), dst_port,
> > > +                     pkt->packet_type);
> > > +
> > > +     ve = vcopyq_lane_u32(ve, 3, te, 3);
> > Compilation error here. This should be vcopyq_laneq_u32 (Extra q
> > after
> > lane)
> No vcopyq_laneq_u32 in arm_neon.h of my environment. I thought it's a
> typo so I changed.
> 
> my gcc version 4.8.5 20150623 (Red Hat 4.8.5-11) (GCC).
> What about yours?
> 
I am using GCC 7.1. No error with this version.

Also to cross check I tried the following versions as well which all
gave compilation errors.
 * gcc (Ubuntu/Linaro 4.9.2-10ubuntu13) 4.9.2
 * gcc 5.3
 * GCC 6.3

So looks like vcopyq_laneq_u32 is not supported in GCC versions < 7.
We can add a wrapper for the same in
./lib/librte_eal/common/include/arch/arm/rte_vect.h for gcc versions <
7.

But I think we can defer this activity. Because I have some other
patches, which moves around the definition of GCC_VERSION, and adds
wrappers for some unsupported instrinsics. Please see below.
http://dpdk.org/dev/patchwork/patch/24161/
http://dpdk.org/dev/patchwork/patch/24162/

I think we can add the vcopyq_laneq_u32 change and the wrapper for the
same after the above patches are merged.

And FYI - Documentation for the vcopyq_laneq_u32 can be found in below
document.
http://infocenter.arm.com/help/topic/com.arm.doc.ihi0073a/IHI0
073A_arm_neon_intrinsics_ref.pdf

> > 
> > > 
> > > +     vst1q_u32((uint32_t *)eth_hdr, ve);
> > > +}
> > > +
> > [...]