From 0ea7869e1685ec80904463bcd4c93ebce5d5c59c Mon Sep 17 00:00:00 2001
From: EgorBo <egorbo@gmail.com>
Date: Wed, 12 Feb 2025 01:03:51 +0100
Subject: [PATCH] Merge store/loads for return buffers

---
 src/coreclr/jit/lower.cpp | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/src/coreclr/jit/lower.cpp b/src/coreclr/jit/lower.cpp
index 34c8bd16b310c0..75365be8385399 100644
--- a/src/coreclr/jit/lower.cpp
+++ b/src/coreclr/jit/lower.cpp
@@ -9439,9 +9439,16 @@ void Lowering::LowerStoreIndirCoalescing(GenTreeIndir* ind)
         //
         // IND<byte> is always fine (and all IND<X> created here from such)
         // IND<simd> is not required to be atomic per our Memory Model
-        const bool allowsNonAtomic =
+        bool allowsNonAtomic =
             ((ind->gtFlags & GTF_IND_ALLOW_NON_ATOMIC) != 0) && ((prevInd->gtFlags & GTF_IND_ALLOW_NON_ATOMIC) != 0);
 
+        if (!allowsNonAtomic && currData.baseAddr->OperIs(GT_LCL_VAR) &&
+            (currData.baseAddr->AsLclVar()->GetLclNum() == comp->info.compRetBuffArg))
+        {
+            // RetBuf is a private stack memory, so we don't need to worry about atomicity.
+            allowsNonAtomic = true;
+        }
+
         if (!allowsNonAtomic && (genTypeSize(ind) > 1) && !varTypeIsSIMD(ind))
         {
             // TODO-CQ: if we see that the target is a local memory (non address exposed)