[PATCH] ARMv7A Cleanup

[Renato Golin]

Hi Bernie!

Thanks for the review!

For a bit of context, my intention is not to fix any bug or behaviour, but
to make the representation of cores more complete across all functions in
Clang.

There are a number of opportunities to fix stuff, de-duplicate functions,
merge classes, and make it easier for people to use "-mcpu=cortex-a8"
without defining "-target arm" and get all the defaults relevant. But
that's for later, right now, I just want to get the names right. ;)


On 6 September 2013 11:45, Bernie Ogden <bogden at arm.com> wrote:

> 1) That NEON test probably should be checking for startswith("8") as well
> as
> "7". But you could leave that for me if you want to confine this to v7A
> cleanup.
>

This is a question for Tim. I agree that, to follow the rules, we should,
but the AArch64 back-end might not be ready yet to generate NEON for the
whole class of applications. If it is, I can easily add it.


2) It seems to be rather the done thing, but does getDefaultDefinitions do
> anything that setting VFP/NEON as subtarget features for the relevant CPUs
> in tabglegen wouldn't do? If not, can we kill it?
>

This is something that crossed my mind, too. I find that function almost
pointless, but I don't know what the rest of the code in the Driver does,
nor how Clang uses TableGen for that kind of thing.

The issue here, I think, it's that's not for codegen or the back-end, but
for Clang itself to generate the correct IR, pass on the correct flags
down, etc.

Clang experts can chime in.



> 3) Does test/Driver/arm-mfpu.c help with some of the testing?
>

Yes, good point. I'll complement arm-mfup and arm-cortex-cpus, should be
enough.

Thanks!
--renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130906/f72e7885/attachment.html>