Current global address caculation doesn't work for AMDGPU in some cases (https://reviews.llvm.org/D142569#4506212).
The root cause is the sh_addr is not substracted when caculating the address.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
What is the special case that makes this not work? I figured this was working in general since I've used it for the rpc symbol as well.
Comment Actions
It seems to depend on the section the data ends up in. Not 100% sure. If this doesn't break anyone, it's good.