[lld] [lld][AMDGPU] Support R_AMDGPU_ABS32_(LO|HI) relocations (PR #191597)

[Matt Arsenault via llvm-commits]

================
@@ -181,6 +183,8 @@ RelExpr AMDGPU::getRelExpr(RelType type, const Symbol &s,
   switch (type) {
   case R_AMDGPU_ABS32:
   case R_AMDGPU_ABS64:
+  case R_AMDGPU_ABS32_LO:
----------------
arsenm wrote:

I've wondered if we can replace these with the regular ABS64 relocations, and then just use expressions to extract the lo/hi half for the 32-bit operands 

https://github.com/llvm/llvm-project/pull/191597