[llvm] [AMDGPU] Sink uniform buffer address offsets into soffset (PR #160939)

[Krzysztof Drewniak via llvm-commits]

================
@@ -328,9 +328,6 @@ bool AMDGPUDAGToDAGISel::matchLoadD16FromBuildVector(SDNode *N) const {
 }
 
 void AMDGPUDAGToDAGISel::PreprocessISelDAG() {
-  if (!Subtarget->d16PreservesUnusedBits())
----------------
krzysz00 wrote:

?

https://github.com/llvm/llvm-project/pull/160939