[llvm] Document FP relative offsets (PR #91031)

[Florian Mayer via llvm-commits]

https://github.com/fmayer created https://github.com/llvm/llvm-project/pull/91031

None

>From 68e3acdb3ef1ddf70cd05a7e8f1d0f68cf5d63e3 Mon Sep 17 00:00:00 2001
From: Florian Mayer <fmayer at google.com>
Date: Fri, 3 May 2024 16:42:49 -0700
Subject: [PATCH] Document FP relative offsets

---
 llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
index fa661b17c13a90..fca1824165e7e3 100644
--- a/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
@@ -1271,6 +1271,9 @@ Value *HWAddressSanitizer::getFrameRecordInfo(IRBuilder<> &IRB) {
   // FP is 0xfffffffffffFFFF0  (4 lower bits are zero)
   // We only really need ~20 lower non-zero bits (FFFF), so we mix like this:
   //       0xFFFFPPPPPPPPPPPP
+  //
+  // FP works because in AArch64FrameLowering::getFrameIndexReference, we
+  // prefer FP-relative offsets for functions compiled with HWASan.
   FP = IRB.CreateShl(FP, 44);
   return IRB.CreateOr(PC, FP);
 }