[PATCH] D14577: Cull non-standard variants of ARM architectures (NFC)

[A. Skrobov via llvm-commits]

tyomitch added a comment.

In http://reviews.llvm.org/D14577#287461, @rengolin wrote:

> In http://reviews.llvm.org/D14577#287449, @tyomitch wrote:
>
> > The motivation is to still accept them in -march and triples (as aliases), while cleaning them out of the rest of the source code.
> >
> > The user interface stays unaffected.
>
>
> No, it doesn't. You're removing more than you're putting back.
>
> Even if all current tests continue to pass, it doesn't mean that the change has no functional changes, just that the tests weren't picking all the changes up.
>
> For example, you're changing v7hl into v7-a, which in itself is "correct" but can have unpredictable side effects on corner cases. Those functions are called all over the place for different reasons, some of them only relevant to the internal logic of the function, like depending on the sub-arch, choose different options. These places are used on Linux, bare-metal, Darwin and Windows, with different results for each one.


I don't see what you mean.
I'm removing **all** uses of AK_ARMV7HL as part of the patch; there were only two, and in both cases it was treated identically to AK_ARMV7A.
How can, therefore, my change introduce **any** change in functionality?


http://reviews.llvm.org/D14577