Re: [PATCH] tune-cortexa5.inc: Allow tuning for vfpv3d16, vfpv3 and neon-vfpv4

[Mark Hatle <mark.hatle@windriver.com>]

On 11/12/15 12:43 PM, Alexandre Belloni wrote:
> On 12/11/2015 at 12:26:57 -0600, Mark Hatle wrote :
>> Seems odd to me that the package arch are the same between the generic (assumed
>> to be the default ARMv7a, which is USUALLY VFPv4-D32 in most cases.)
>>
>> If we want to limit to v3 and/or D16.  It might make sense for a different
>> package arch, otherwise there is no way to separate the feeds in a
>> multiple-machine configuration with different per-target optimizations.
>>
>> (similarly the -neon versions as well, but that is existing code)
>>
> 
> Yeah, I'm also not sure this is the correct thing but all the package
> arch for tune-armv7a* are armv7a so I've replicated that.
> 
> Also, I just realized there are no users for armv7a*-vfpv3* in oe-core,
> maybe I should leave those tunes in my soc specific BSP layer?
> 

There are a lot of tunes defined that have no direct oe-core users.  We try to
define things we think are useful and supported by the included toolchain.

External BSPs may be required to actually test it.

--Mark