RE: [PATCH] build: add missing arch define for Arm

["Juraj Linkeš" <juraj.linkes@pantheon.tech>]

> -----Original Message-----
> From: Ruifeng Wang <Ruifeng.Wang@arm.com>
> Sent: Friday, January 14, 2022 10:05 AM
> To: thomas@monjalon.net
> Cc: dev@dpdk.org; stable@dpdk.org; viktorin@rehivetech.com;
> bruce.richardson@intel.com; stephen@networkplumber.org; Juraj Linkeš
> <juraj.linkes@pantheon.tech>; Honnappa Nagarahalli
> <Honnappa.Nagarahalli@arm.com>; nd <nd@arm.com>; nd <nd@arm.com>
> Subject: RE: [PATCH] build: add missing arch define for Arm
> 
> > -----Original Message-----
> > From: Thomas Monjalon <thomas@monjalon.net>
> > Sent: Friday, January 14, 2022 1:33 AM
> > To: Ruifeng Wang <Ruifeng.Wang@arm.com>
> > Cc: dev@dpdk.org; stable@dpdk.org; viktorin@rehivetech.com;
> > bruce.richardson@intel.com; stephen@networkplumber.org;
> > juraj.linkes@pantheon.tech; Honnappa Nagarahalli
> > <Honnappa.Nagarahalli@arm.com>; nd <nd@arm.com>
> > Subject: Re: [PATCH] build: add missing arch define for Arm
> >
> > 17/12/2021 09:54, Ruifeng Wang:
> > > As per design document, RTE_ARCH is the name of the architecture.
> > > However, the definition was missing on Arm with meson build.
> > > It impacts applications that refers to this string.
> > >
> > > Added for Arm builds.
> > >
> > > Fixes: b1d48c41189a ("build: support ARM with meson")
> > > Cc: stable@dpdk.org
> > >
> > > Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > > ---
> > >                  ['RTE_ARCH_ARMv8_AARCH32', true],
> > > +                ['RTE_ARCH', 'arm64_aarch32'],
> >
> > Why not armv8_aarch32?
> 
> Thanks for the comments.
> Agreed. armv8_aarch32 is consistent with the RTE_ARCH_xx macro above.
> 
> >
> > [...]
> > >          dpdk_conf.set('RTE_ARCH_ARMv7', true)
> > > +        dpdk_conf.set('RTE_ARCH', 'armv7')
> > [...]
> > >      # armv8 build
> > > +    dpdk_conf.set('RTE_ARCH', 'arm64')
> >
> > Why not armv8?
> >
> > What I prefer the most in silicon industry is the naming craziness :)
> 
> While armv8 usually refers to one generation of the Arm architecture, arm64 is
> more generic for 64-bit architectures.
> And what defined for armv8 build is RTE_ARCH_ARM64. So for consistency,
> arm64 is better?
> 

Using armv8_aarch32 along with arm64 doesn't seem right. We should unite these and I think armv8 makes sense. As you mentioned arvm8 is an arm64 architecture and using the more precise identification is better in my opinion (as that gives more information). As for the consistency with RTE_ARCH_ARM64, I think the problem is that we don't have the RTE_ARCH_ARMv8 flag (which would provide the consistency, but won't be used):
The current code is, accurately, written for 64bit arm architectures (all of them).
There is currently no need to differentiate between 64bit arm architectures which is why RTE_ARCH_ARMv8 doesn't exist.
However, armv8 exists and we know how to identify it which is why I think setting RTE_ARCH to armv8 is the way to go.

So my thinking is RTE_ARCH should be set to armv8, which implies RTE_ARCH_ARMv8 which in turn implies RTE_ARCH_ARM64. We're just missing the middle part since there's no use for it now.

And to be fully consistent, we could add RTE_ARCH_ARM32 to armv7 (as a superset of RTE_ARCH_ARMv7, but that likely won't be of much use).