[llvm-commits] [patch] Invalid alignments for NEON loads

[Tim Northover]

Hi again,

Right, there are now two patches (superseding the original):

1. extload.diff makes LLVM use the correct NEON loads for extloadv*
patterns. Before, we would use vld1.32 for everything from a 16-bit
load to a 64-bit load (my own fault, embarrassingly). With this patch
we'll select the correct version (I believe).

2. vld1-align.diff: as before this brings assembly printing in line
with object emission. Unfortunately, with the change above I can no
longer find a way to trigger this code-path so there's no regression
test. Wang, I've CCed you in the vague hope you'll remember what made
you put the initial binary change in. I still think the change is
clearly correct, but it's probably less pressing now.

Can I commit them?

Tim.

On Thu, Aug 9, 2012 at 9:01 PM, Tim Northover <t.p.northover at gmail.com> wrote:
>> Ok to commit?
>
> Anyone thinking of reviewing this, it might be an idea to hold off
> until tomorrow. I've realised (well, had it pointed out to me) that
> the instruction being emitted is wrong in some cases. The original
> change is probably sound regardless of whether it's triggered here in
> the end, but the regression test may not be valid after everything's
> done properly.
>
> Until tomorrow...
>
> Tim.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: extload.diff
Type: application/octet-stream
Size: 12240 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120810/0c1480cb/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: vld1-align.diff
Type: application/octet-stream
Size: 2429 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120810/0c1480cb/attachment-0001.obj>