[PATCH] [ARM] fix some ld/st instruction with wirteback

[Renato Golin]

On 15 January 2014 12:54, Jiangning Liu <liujiangning1 at gmail.com> wrote:

> For llvm.arm.neon.vld3.v1i64 plus 3*64 offset writeback, we can use vld1
> with writeback to implement the same functionality, because ARM instruction
> vld1 accept 64-bit vector list.
>

I understand that this is a different issue, but do we have to do this?
Can't we just emit vld/vst 2/3/4?


// We use a VST1 for v1i64 even if the pseudo says vld2/3/4, so
> // check for that explicitly too. Horribly hacky, but temporary.
>

I'd at least like to have a "FIXME" on that "temporary" comment, so that we
can find it easier, later.


I introduced two new function isVLDfixed and isVSTfixed to cover all _fixed
> ISDNode.
>

On those functions, a switch would be a bit less verbose and clearer.

As a quick fix, the patch is ok (with the comments above), but would be
good to at least have an idea of how much work is involved in emitting
vld/vst 2/3/4.

cheers,
--renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140115/817f6bda/attachment.html>