Re: [PATCH 0/6] Correct and improve the ARM tunings

[Adrian Bunk <bunk@stusta.de>]

On Wed, Apr 03, 2019 at 01:48:04PM -0700, Andre McCurdy wrote:
>...
> If we let the user pass an arbitrary string to -march then we lose the
> ability to determine that (for example) it's safe for a
> "armv7athf-vfpv3" machine to pull from a "armv7athf-vfpv3d16" package
> feed.

For ARM <= v7 this is possible, except for bugs like the mentioned 
around thumb.

With ARMv8 you end up duplicating everything gcc knows about 
architecture and feature flags for a cpu, and about all the
interdependencies between features flags.

It will also have to tell whether a cpu contains the optional
support for the old ARMv7 32bit ABI, or only for the ILP32 variant
of the ARMv8 ABI.

> Whether or not anyone in the real world actually maintains a generic
> package feed and pulls from it from multiple machine specific builds
> (verses setting up separate package feeds for each DEFAULTTUNE they
> care about) would be an interesting question...

It would be good to look at what exactly has to be supported here,
and the complexity involved.

The simplest "generic" would be armv8.0 with no feature flags.

If a generic "is tune A a superset of tune B" is required,
you end up duplicating a lot of hardware information from gcc.

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed