[PATCH] D38253: [ARM] Restore the right frame pointer register in Int_eh_sjlj_longjmp

[Matthias Braun via Phabricator via llvm-commits]

MatzeB added a comment.

In https://reviews.llvm.org/D38253#883234, @mstorsjo wrote:

> FWIW, this was more or less approved by @t.p.northover on IRC:
>
> "The second one is ugly, but that describes SjLj as a whole so I doubt we'd be making the world a worse place." and "There was actually some talk about standardising FP a while ago. Doesn't seem to have gone anywhere yet though."


This is NOT how patch aproval works in llvm!! Please revert!

Just glancing at this I can see this changes the asmprinter output but not the modelling of the instruction which now changes r11 in some cases but does not model that fact with machine operands.



================
Comment at: llvm/trunk/lib/Target/ARM/ARMAsmPrinter.cpp:1207
 
+  const MachineFunction &MF = *MI->getParent()->getParent();
+  const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
----------------
There’s no need to put those statements on the hot path of this function.


Repository:
  rL LLVM

https://reviews.llvm.org/D38253