[PATCH] ARM: I have a cunning plan (atomics)
t.p.northover at gmail.com
Thu Mar 27 04:32:22 PDT 2014
On 26 March 2014 18:06, Evan Cheng <evan.cheng at apple.com> wrote:
> I have no real objection to the plan. But can you take advantage of the opportunity to add some comments?
Of course, I'll be uploading a new version shortly.
> Is there any impact on debug info?
There was (my IRBuilders didn't pick up a DebugLoc from the original
instruction), but there shouldn't be now.
> Also, what would happen to atomic load > 64-bit? It would cause isel failure? Can we emit a nicer diagnostics?
It would have failed, I've now changed it back to the original
behaviour of falling back to a libcall. You're basically screwed if
this happens anyway, but it is theoretically possible for the platform
to provide correct semantics at least.
More information about the llvm-commits