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

[Krzysztof Drewniak via llvm-commits]

================
@@ -341,8 +338,23 @@ void AMDGPUDAGToDAGISel::PreprocessISelDAG() {
 
     switch (N->getOpcode()) {
     case ISD::BUILD_VECTOR:
-      // TODO: Match load d16 from shl (extload:i16), 16
-      MadeChange |= matchLoadD16FromBuildVector(N);
+      // D16 optimization requires subtarget support
----------------
krzysz00 wrote:

So I claim that voffset => soffset should either go there or be an IR pass after uniformity analysis like you suggested.

(Making it an IR pass also means you can do s_buffer_load promotion there)

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