[llvm] [AMDGPU] Update code object metadata for kernarg preload (PR #134666)

[Matt Arsenault via llvm-commits]

================
@@ -161,14 +231,27 @@ struct AMDGPUFunctionArgInfo {
   ArgDescriptor WorkItemIDZ;
 
   // Map the index of preloaded kernel arguments to its descriptor.
-  SmallDenseMap<int, KernArgPreloadDescriptor> PreloadKernArgs{};
+  SmallDenseMap<int, KernArgPreload::KernArgPreloadDescriptor>
+      PreloadKernArgs{};
+  // Map hidden argument to the index of it's descriptor.
+  SmallDenseMap<KernArgPreload::HiddenArg, int> PreloadHiddenArgsIndexMap{};
   // The first user SGPR allocated for kernarg preloading.
   Register FirstKernArgPreloadReg;
 
   std::tuple<const ArgDescriptor *, const TargetRegisterClass *, LLT>
   getPreloadedValue(PreloadedValue Value) const;
 
   static AMDGPUFunctionArgInfo fixedABILayout();
+
+  // Returns preload argument descriptors for an IR argument index. Isel may
+  // split IR arguments into multiple parts, the return vector holds all parts
+  // associated with an IR argument in the kernel signature.
+  SmallVector<const KernArgPreload::KernArgPreloadDescriptor *, 4>
+  getPreloadDescriptorsForArgIdx(unsigned ArgIdx) const;
+
+  // Returns the hidden arguments `KernArgPreloadDescriptor` if it is preloaded.
+  std::optional<const KernArgPreload::KernArgPreloadDescriptor *>
----------------
arsenm wrote:

Avoid optional of pointer

https://github.com/llvm/llvm-project/pull/134666