[llvm] [AMDGPU] Fix LDS access via flat pointer argument in amdgpu-sw-lower-lds (PR #209842)

[via llvm-commits]

================
@@ -790,6 +792,80 @@ void AMDGPUSwLowerLDS::translateLDSMemoryOperationsToGlobalMemory(
   }
 }
 
+void AMDGPUSwLowerLDS::lowerFlatToLocalRoundTrips(Function *Func) {
----------------
skc7 wrote:

I removed the separate all-functions fixup and integrated this into the existing SW LDS lowering flow. The module-level scan is still needed to know which callee arguments can carry LDS-derived flat pointers, but the actual per-function rewrite is now isolated in `getFlatPtrForRoundTripLDSAccess` / l`owerNonKernelLDSFlatArgAccesses`.

https://github.com/llvm/llvm-project/pull/209842