[llvm] [ASAN] Make asan pass idempotent. (PR #99439)

[via llvm-commits]

================
@@ -0,0 +1,56 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 5
+; This test checks in the second run, function is not instrumented again.
+; RUN: opt < %s -passes=asan,asan -S | FileCheck %s
----------------
skc7 wrote:

We found this issue when a HIP program is first compiled to bitcode with -fsanitize=address, then the bitcode is compiled to object file with -fsanitize=address again. This lead to calculation of shadow memory twice due to double instrumentation.

As you pointed out, ConstructorKind::None can lead to asan.module_ctor to be not present in the module. Do you suggest any other way to check if module is instrumented?

https://github.com/llvm/llvm-project/pull/99439