[Buildroot] [PATCH 1/1] board/pine64/pinecube: new board

[Giulio Benetti]

Hi Yann,

> Il giorno 15 ago 2022, alle ore 16:53, Yann E. MORIN <yann.morin.1998 at free.fr> ha scritto:
> 
> Giulio, Jan, All,
> 
> On 2022-08-14 13:56 +0200, Giulio Benetti spake thusly:
>> On 14/08/22 11:22, Jan Havran wrote:
>>>>> +++ b/configs/pinecube_defconfig
>>>>> +BR2_ARM_FPU_NEON_VFPV4=y
>>>> This ^^^ should be removed since NEON doesn't support complete FPU
>>>> standard API, so it's better to keep the default for BR2_cortex_a7
>>> I will fix everything and post V2, but one thing I do not understand much is
>>> this one.
>> Here there is a very good explanation on why not adding by default
>> BR2_ARM_FPU_NEON_VFPV4:
>> https://lists.buildroot.org/pipermail/buildroot/2022-January/634728.html
> 
> I even tink that, ultimately, we should just remove FPU options that
> also enable NEON, like BR2_ARM_FPU_NEON_VFPV4, because they do not make
> sense.

I was wondering if sending a patchset for this, so I’m going to.

Thank you
Best regards
Giulio

> 
> Using NEON for floating-point arithmetics should probably left to a
> per-pacakge setting, where and when t makes sense for that package (e.g.
> math-intensive packages that can cope with incorrect results (like
> real-time audio stuff that favour an occasional glitch rather than a
> packet loss).
> 
> [--SNIP--]
>> But yes, if VFPV4 is supported then yes, it makes sense to enable it
>> instead of keeping the default. So since this board is S3 based and
>> it supports VFPV4 then you can enable:
>> BR2_ARM_FPU_VFPV4
>> 
>> and same can go with H3 based boards I think.
>> 
>> @Yann, what do you think about enabling VFPV4 on boards with SoCs
>> that support VFPV4? (not NEON/VFPV4 of course)
> 
> Usually, defconfigs are very limited, and what we guarantee is very
> basic: a booting kernel with basic drivers, and busybox' login and
> shell. That does not use much floating-point instructions, so that would
> not benefit much from using a "better" FPU.
> 
> I previously rejected the big series that bulk-converted all defconfigs
> in one go, because that was not tested on real hardware.
> 
> But for a defconfig that was actually tested against real hardware, I
> think it is OK that it uses a "better" FPU. So, if the pinecube has a
> VFPv4, BR2_ARM_FPU_VFPV4 or BR2_ARM_FPU_VFPV4_D16 is OK.
> 
> Also nte that BR2_ARM_FPU_VFPV4 will make use of all the 32 flaoting
> point registers, i.e. the full range, while BR2_ARM_FPU_VFPV4_D16 will
> only make use of 16 registers. Not all VFPv4 implementation have 32
> registers, so carefully enable the appropriate one (_D16 is safe to use
> in either case, though).
> 
> Regards,
> Yann E. MORIN.
> 
> -- 
> .-----------------.--------------------.------------------.--------------------.
> |  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
> | +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
> | +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
> | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
> '------------------------------^-------^------------------^--------------------'
> _______________________________________________
> buildroot mailing list
> buildroot at buildroot.org
> https://lists.buildroot.org/mailman/listinfo/buildroot