[llvm] [AMDGPU] Restrict promote alloca on pointers across address spaces (PR #119762)

[Sumanth Gundapaneni via llvm-commits]

================
@@ -198,3 +167,39 @@ entry:
   %tmp = load ptr addrspace(3), ptr addrspace(5) %alloca, align 8
   ret ptr addrspace(3) %tmp
 }
+
+; Will not vectorize because we are doing a load/store of a pointer across
+; address spaces of varying pointer sizes.
+define ptr @alloca_load_store_ptr64_full_ivec(ptr %arg) {
+; CHECK-LABEL: define ptr @alloca_load_store_ptr64_full_ivec
+; CHECK-SAME: (ptr [[ARG:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[ALLOCA:%.*]] = alloca [8 x i8], align 8, addrspace(5)
+; CHECK-NEXT:    store ptr [[ARG]], ptr addrspace(5) [[ALLOCA]], align 8
+; CHECK-NEXT:    [[TMP:%.*]] = load ptr, ptr addrspace(5) [[ALLOCA]], align 8
+; CHECK-NEXT:    ret ptr [[TMP]]
+;
+entry:
+  %alloca = alloca [8 x i8], align 8, addrspace(5)
+  store ptr %arg, ptr addrspace(5) %alloca, align 8
+  %tmp = load ptr, ptr addrspace(5) %alloca, align 8
+  ret ptr %tmp
+}
+
+; Will not vectorize because we are doing a load/store of a pointer across
+; address spaces of varying pointer sizes.
----------------
sgundapa wrote:

Should the amdgpu promote alloca pass do this transformation or should this pass expect the IR to be in a state where pointer storages across address space are done after using an addrspacecast .

https://github.com/llvm/llvm-project/pull/119762