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

[Vitaly Buka via llvm-commits]

================
@@ -443,8 +443,13 @@ static const MemoryMapParams Linux_I386_MemoryMapParams = {
 static const MemoryMapParams Linux_X86_64_MemoryMapParams = {
     0,              // AndMask (not used)
     0x500000000000, // XorMask
+#if LLVM_MSAN_SHADOW_OFFSET_2MB == 1
----------------
vitalybuka wrote:

For LLVM/ part's we should not use preprocessor, and just use compiler flags.

If it's very niche use case copt<> is OK, no need to expose it to frontend.

However `-march` support would be nice?
If you decide to do `-march`, please do in future patches. Here copt<> is OK.

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