[PATCH] D61095: [AArch64][Windows] Compute function length correctly in unwind tables.

[Reid Kleckner via Phabricator via llvm-commits]

rnk added a comment.

In D61095#1477913 <https://reviews.llvm.org/D61095#1477913>, @efriedma wrote:

> Well, I guess there are sort of two potential approaches to emitting it "later".  One, we could emit something like ".word (endfunc-beginfunc)/4+0x10800000", as long as we have some conservative estimate we could use to prove that we don't need to split the unwind data.  The other would be to use a dedicated, relaxable fragment, which could grow to accommodate splitting the unwind data if necessary.  The dedicated fragment is probably a bit more bulletproof, and more consistent with the way we handle relaxation in other contexts, but it would require a lot of refactoring to implement.


Oh, yes, that would be the usual way of doing things, build a symbolic MCExpr for it. With the inline line table encoding, we couldn't do that because the label differences get encoded with a variable length encoding. Anyway, a FIXME is fine.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61095/new/

https://reviews.llvm.org/D61095