r216249 - [test/CodeGen/ARM] Adpat test to match new codegen after r216236.

[James Molloy]

Hi Renato,

I think there are two tests that can be derived from here, and maybe
> both of them can be easily done on their own, depending on how Clang
> lowers its intrinsics.


But what value do these derived tests have? In order to change the code the
intrinsics produce (i.e. break them), you need to touch arm_neon.td. But
arm_neon.td would also be the generation source for these tests. So you
have a tautological testing mechanism.

Cheers,

James


On 26 August 2014 09:56, Renato Golin <renato.golin at linaro.org> wrote:

> On 26 August 2014 08:24, James Molloy <james at jamesmolloy.co.uk> wrote:
> > I can see that test had to be originally generated somehow, but now we
> have
> > it and we don't expect large churn in the intrinsics, does it make sense
> to
> > reimplement the autogeneration mechanism? I haven't worked at all on
> getting
> > it put back in the new framework, but if people think it's the right
> thing
> > to do I can make time to do so. I'm just not convinced that it has value.
>
> Hi James,
>
> I think there are two tests that can be derived from here, and maybe
> both of them can be easily done on their own, depending on how Clang
> lowers its intrinsics.
>
> The first, Clang based test, is more or less what that one is, but
> instead of checking for asm instructions, it emits IR and checks for
> the correct builtin call. I should hope that no ARM-specific
> optimization has occurred by that time. The second, a list of IR
> one-liners with the intrinsics generated by Clang (well, any
> front-end), and their unoptimized correspondents.
>
> To make sure we don't lose intrinsics by updating one and not the
> other, we should have a script in tree (utils) that generates them,
> possibly based on some other source in tree (table gen), for both
> Clang and LLVM counterparts.
>
> Having said that, I don't think this project is particularly critical
> to our test infrastructure, so I'd give this to a summer student this
> year or a GSoC project the next.
>
> cheers,
> --renato
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140826/947e32d6/attachment.html>