[clang] [win][clang] Align scalar deleting destructors with MSABI (PR #139566)

[Mariya Podchishchaeva via cfe-commits]

================
@@ -194,9 +217,31 @@ I::~I() { call_in_dtor(); }
 // CHECK-MSABI32-LABEL: define {{.*}} @"??_GI@@UAEPAXI at Z"(
 // CHECK-MSABI-NOT: call{{ }}
 // CHECK-MSABI: load i32
-// CHECK-MSABI: icmp eq i32 {{.*}}, 0
-// CHECK-MSABI: br i1
+// CHECK-MSABI-NEXT: and i32 %[[IMP_PARAM:.*]], 4
+// CHECK-MSABI-NEXT: icmp eq i32 {{.*}}, 0
+// CHECK-MSABI-NEXT: br i1 %[[CHCK]], label %dtor.entry_cont, label %dtor.call_dtor
+//
+// CHECK-MSABI: dtor.call_dtor:
+// CHECK-MSABI64-NEXT: call void @"??1I@@UEAA at XZ"({{.*}})
+// CHECK-MSABI32-NEXT: call x86_thiscallcc void @"??1I@@UAE at XZ"({{.*}})
+// CHECK-MSABI-NEXT: br label %dtor.entry_cont
+//
+// CHECK-MSABI: dtor.entry_cont:
+// CHECK-MSABI-NEXT: and i32 %[[IMP_PARAM]], 1
+// CHECK-MSABI-NEXT: icmp eq i32 %{{.*}}, 0
+// CHECK-MSABI-NEXT: br i1 %{{.*}}, label %dtor.continue, label %dtor.call_delete
+//
+// CHECK-MSABI: dtor.call_delete:
+// CHECK-MSABI-NEXT: %[[THIRDBIT1:.*]] = and i32 %[[IMP_PARAM]], 4
+// CHECK-MSABI-NEXT: %[[CHCK2:.*]] = icmp eq i32 %[[THIRDBIT1]], 0
+// CHECK-MSABI-NEXT: br i1 %[[CHCK2]], label %dtor.call_class_delete, label %dtor.call_glob_delete
+//
+// CHECK-MSABI-LABEL: dtor.call_glob_delete:
+// CHECK-MSABI64: call void @"??3 at YAXPEAX_KW4align_val_t@std@@@Z"(ptr noundef %{{.*}}, i64 noundef 96, i64 noundef 32)
+// CHECK-MSABI32: call void @"??3 at YAXPAXIW4align_val_t@std@@@Z"(ptr noundef %{{.*}}, i32 noundef 64, i32 noundef 32)
+// CHECK-MSABI-NEXT: br label %[[RETURN:.*]]
 //
+// CHECK_MSABI: dtor.call_class_delete:
----------------
Fznamznon wrote:

Nice, thanks. I probably hallucinated. I wasn't able to break this case while testing object files built with clang and msvc either.

https://github.com/llvm/llvm-project/pull/139566