Re: [dpdk-dev] [RFC PATCH] build/pkg-config: remove machine arch flag

[Ruifeng Wang <Ruifeng.Wang@arm.com>]

> -----Original Message-----
> From: Jerin Jacob <jerinjacobk@gmail.com>
> Sent: Thursday, September 16, 2021 11:43 AM
> To: David Christensen <drc@linux.vnet.ibm.com>
> Cc: Bruce Richardson <bruce.richardson@intel.com>; dpdk-dev
> <dev@dpdk.org>; Luca Boccassi <bluca@debian.org>; David Marchand
> <david.marchand@redhat.com>; ian.stokes@intel.com; Ilya Maximets
> <i.maximets@ovn.org>; jerinj@marvell.com; Ruifeng Wang
> <Ruifeng.Wang@arm.com>; Honnappa Nagarahalli
> <Honnappa.Nagarahalli@arm.com>; Ananyev, Konstantin
> <konstantin.ananyev@intel.com>; Ferruh Yigit <ferruh.yigit@intel.com>
> Subject: Re: [dpdk-dev] [RFC PATCH] build/pkg-config: remove machine arch
> flag
> 
> On Thu, Sep 16, 2021 at 2:53 AM David Christensen <drc@linux.vnet.ibm.com>
> wrote:
> >
> >
> >
> > On 9/15/21 9:28 AM, Bruce Richardson wrote:
> > > On Fri, Dec 11, 2020 at 03:51:11PM +0000, Bruce Richardson wrote:
> > >> Traditionally any apps built using DPDK had to support the same
> > >> instruction sets as supported when the DPDK SDK itself was built,
> > >> since that was "leaked" through to the end-app and DPDK headers via
> > >> RTE_MACHINE_CPUFLAG_* values. Therefore, when converting over to
> > >> meson for app build correctness the "-march=" flag was included in
> > >> the pkg-config cflags output.
> > >>
> > >> However, since the DPDK-specific CPU flags are now obsolete, and we
> > >> instead check directly for compiler-defined flags, we should no
> > >> longer need to force the exact same architecture match in all
> > >> cases. To faciliate such flexibility, a new pkg_config_machine_args
> > >> array - which defaults to the existing machine_args array has been
> > >> defined. The individual architectures - x86, arm and ppc - can
> > >> choose if and how to override this value themselves.
> > >>
> > >> For x86, since SSE4.2 is the minimum instruction-set level needed
> > >> to run DPDK, and since some header files assume that minimum level
> > >> of instruction set support, we override the "-march=" value with "-
> msse4"
> > >> for the pkg-config file. This allows end applications to set their
> > >> own "march" value while still ensuring valid DPDK compilation.
> > >>
> > >> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > >> ---
> > >>
> > >> CC: Jerin Jacob <jerinj@marvell.com>
> > >> CC: Ruifeng Wang <ruifeng.wang@arm.com>
> > >> CC: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> > >> CC: David Christensen <drc@linux.vnet.ibm.com> for Feedback
> > >> requested from ARM and PPC maintainers as to this change and what
> > >> flags, if any, need to be in the .pc file for DPDK on such
> > >> platforms. For example - is setting 'pkg_config_machine_args' to
> > >> 'machine_args' by default necessary behaviour?
> > >>
> > >
> > > Ping for further thoughts or input on this patch.
> 
> Armv8 has dedicated config(taken least cpu flags which supports all
> armv8) for this purpose config/arm/arm64_armv8_linux_gcc.
> So this change looks good to me. @Ruifeng Wang (Arm Technology China)
> Any comments?
> 

Agree. For distro build, only armv8-a is requested. The change looks fine.

Thanks,
Ruifeng
> 
> >
> > On initial inspection I don't see any additional value for PPC systems.
> >   Selection of a CPU through -mcpu implies a full ISA for that CPU,
> > including things like vector extensions, at least with respect to the
> > minimum supported POWER8 CPU (ISA 2.07) and later ISAs.  The values of
> > machine_args and pkg_config_machine_args should be identical in all
> > cases I can think of.
> >
> > If a third party OpenPOWER CPU comes along some time in the future
> > then the requirements may change.
> 
> 
> 
> 
> >
> > Dave