[compiler-rt] [llvm] [msan] ability to make MSan up to 20x faster on AMD CPUs (-DLLVM_MSAN_SHADOW_OFFSET_2MB=1) (PR #171993)

[Thurston Dang via llvm-commits]

================
@@ -350,6 +350,12 @@ if(NOT COMPILER_RT_HAS_FUNC_SYMBOL)
   add_definitions(-D__func__=__FUNCTION__)
 endif()
 
+if (LLVM_MSAN_SHADOW_OFFSET_2MB)
----------------
thurstond wrote:

Two alternatives to consider for the definition:
- name it something like MSan Zen workaround, to make it obvious what the purpose is
- make the MSan shadow offset a parameter (e.g., -DLLVM_MSAN_SHADOW_OFFSET=2048576), so that if some other platform ends up needing a different offset, it won't be necessary to make a new flag

https://github.com/llvm/llvm-project/pull/171993