[clang] [llvm] AMDGPU: Add builtin/intrinsic global_(load|store)_b128 (PR #172090)

[Matt Arsenault via cfe-commits]

================
@@ -906,6 +906,31 @@ def int_amdgcn_bitop3 :
                         [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>, llvm_i32_ty],
                         [IntrNoMem, IntrSpeculatable, ImmArg<ArgIndex<3>>]>;
 
+class AMDGPUGlobalStore : Intrinsic <
+  [],
+  [global_ptr_ty,          // Base global pointer to store to
+   llvm_v4i32_ty,          // Data to store
+   llvm_metadata_ty],      // Scope
+  [ IntrWriteMem, WriteOnly<ArgIndex<0>>, NoCapture<ArgIndex<0>>,
+    IntrWillReturn, IntrNoCallback, IntrNoFree ],
+  "",
+  [SDNPMemOperand, SDNPMayStore]
+>;
+
+def int_amdgcn_global_store_b128 : AMDGPUGlobalStore;
+
+class AMDGPUGlobalLoad : Intrinsic <
+  [llvm_v4i32_ty],
+  [global_ptr_ty,          // Base global pointer to load from
+   llvm_metadata_ty],      // Scope
+  [ IntrReadMem, ReadOnly<ArgIndex<0>>, NoCapture<ArgIndex<0>>, IntrWillReturn,
----------------
arsenm wrote:

Can we mark this with NonNull on the ptr arg? 

https://github.com/llvm/llvm-project/pull/172090