[PATCH] D61491: AMDGPU: Be explicit about whether the high-word in SI_PC_ADD_REL_OFFSET is 0

[Nicolai Hähnle via Phabricator via llvm-commits]

nhaehnle added a comment.

In D61491#1533471 <https://reviews.llvm.org/D61491#1533471>, @arsenm wrote:

> Are you saying we should remove shouldEmitFixup and just always use buildPCRelGlobalAddress?


The current trunk always uses `buildPCRelGlobalAddress`, and that makes sense to me. `shouldEmitFixup` is part of the logic that determines what kind of PC-relative global address is used. There seem to be three modes right now:

1. 64-bit PC-relative to the symbol
2. 32-bit PC-relative to the symbol
3. 64-bit PC-relative to the GOT, loading the final symbol address from the GOT.

The logic deciding between these cases seems like it could benefit from being streamlined and simplified, but I wouldn't remove it outright as the distinction can be helpful.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61491/new/

https://reviews.llvm.org/D61491