r374288 - Recommit "[Clang] Pragma vectorize_width() implies vectorize(enable)"

[Hans Wennborg via cfe-commits]

> On Mon, Oct 21, 2019 at 9:18 AM Hans Wennborg via cfe-commits <cfe-commits at lists.llvm.org> wrote:
>>
>> As expected, this broke the Chromium build again (but it seems only at
>> -Oz this time).
>
>
> To be clear, is the desired behavior that the vectorization is performed even at -Oz? I suspect that it is, but I just want to confirm.

I'm not the author, so I'm not sure. It could also be that the desired
behaviour was "if vectorizing, use 4-element vectors".

But from what I understand, the desired behaviour of the pragma is
exactly that, to turn force enable vectorization.

Actually, reading
http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations,
I don't find support for this new behaviour, but maybe that's written
somewhere else.