[PATCH] Add Cortex-A9 scheduling classes for vldm/vstm instructions that access more than 32 bytes

Renato Golin renato.golin at linaro.org
Wed Aug 14 14:35:06 PDT 2013


On 14 August 2013 22:31, Silviu Baranga <Silviu.Baranga at arm.com> wrote:

> I think the test would be easily breakable if I would add those checks.
> The vldm/vstm instructions were generated from fills/spills, so they might
> go away in the future or get replaced by some other code sequence.
>

So, I thought about that, and this is primarily why I think you *should*
put the checks there.

My point is that, on the day it gets out, this test will be irrelevant to
the feature in question, and will be wasting test time for nothing.

If you leave enough comments on the test to that regard, and put the CHECKs
in place, whoever changes LLVM so that the vldm/vstm instructions won't be
generated, will have the conscious decision to remove the test altogether.

cheers,
--renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130814/52c650d4/attachment.html>


More information about the llvm-commits mailing list