[PATCH] D18046: [X86] Providing correct unwind info in function epilogue

[Violeta Vukobrat via llvm-commits]

violetav added a comment.

To avoid having to calculate CFA on entry/end in the late pass, we could remember information about entry/end offset (and register) of a MBB in frame lowering and just use that info in the pass later. Possible problems could be BB merging, splitting and creating in later passes. Reid, is this what you had in mind? I am not sure if you meant to leave the insertion of CFI instructions in epilogue in the frame lowering, or to use the information available during frame lowering in order to simplify the implemented pass (but leave the insertion of CFI instructions in the late pass). Also, is storing information in side tables already used somewhere in the code?

We could store certain information about instructions: how they affect the offset/register. This information could be written down when the instruction is created, and then used in the late pass to generate and insert CFI instructions. That could make it easier to do optimizations that affect the CFI between X86FrameLowering and this pass.

I will start investigating these ideas, to see what kind of problems may arise. Do you have any thoughts on them?

Keno, could you explain your proposal a bit further, maybe provide an example?


Repository:
  rL LLVM

http://reviews.llvm.org/D18046