[llvm] r262440 - revert r262424 because there's a *clang test* for AArch64 that checks -O3 asm output

[Renato Golin via llvm-commits]

On 2 March 2016 at 15:54, Sanjay Patel <spatel at rotateright.com> wrote:
> Thanks for replying. I was going to ping the AArch64 experts about this.
> There were 7 tests that broke with my patch in:
> clang/test/CodeGen/aarch64-neon-misc.c

Right, this makes things easier, since we only need to change one test for now.


> This is at least the 3rd time I've had a clang test fail on me after making
> an LLVM optimizer change. So I admit that I wanted to just XFAIL the whole
> thing, but I wouldn't do that. :)

Please, don't. Bad valid tests are better than no tests. If they were
useless, however, I'd just remove.


> So unless something has changed, I think we need to split each test in
> aarch64-neon-misc.c (and other files like it) into pieces:
> 1. A clang test for the C-level intrinsics -> LLVM intrinsics and/or native IR.
> 2. A codegen test for LLVM intrinsic/IR -> asm.

I absolutely agree, but I don't want this to become a witch hunt. I
think we (ARM folks) really need to get together and come up with a
plan.

That's what PR26815 is all about. I've copied some people, feel free
to copy even more. We need to prioritise this once and for all.

cheers,
--renato