[clang] [llvm] [AMDGPU] Add global_load_tr for GFX12 (PR #77772)

[Matt Arsenault via cfe-commits]

================
@@ -18178,6 +18178,51 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
     llvm::Function *F = CGM.getIntrinsic(IID, {ArgTy});
     return Builder.CreateCall(F, {Addr, Val, ZeroI32, ZeroI32, ZeroI1});
   }
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_v2i32:
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_i32:
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v8i16:
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v8f16:
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v4i16:
+  case AMDGPU::BI__builtin_amdgcn_global_load_tr_b128_v4f16: {
+
+    Intrinsic::ID IID;
+    llvm::Type *ArgTy;
+    switch (BuiltinID) {
+    case AMDGPU::BI__builtin_amdgcn_global_load_tr_b64_v2i32:
+      ArgTy = llvm::FixedVectorType::get(
+          llvm::Type::getInt32Ty(getLLVMContext()), 2);
+      IID = Intrinsic::amdgcn_global_load_tr_b64;
----------------
arsenm wrote:

Why is there a different intrinsic for each type? Why not just add one intrinsic, overloaded on the pointer type and value type?

https://github.com/llvm/llvm-project/pull/77772