JIT: Copy loop memory dependence recursively on hoisting (#116068)

When hoisting we were cloning the full tree but only copying memory
dependence of the root node. Ensure we copy it for the full tree.

Author: jakobbotsch

src/coreclr/jit/optimizer.cpp

@@ -3674,30 +3674,7 @@ void Compiler::optPerformHoistExpr(GenTree* origExpr, BasicBlock* exprBb, FlowGr
 
     preheader->CopyFlags(exprBb, BBF_COPY_PROPAGATE);
 
-    Statement* hoistStmt = gtNewStmt(hoist);
-
-    // Simply append the statement at the end of the preHead's list.
-    Statement* firstStmt = preheader->firstStmt();
-    if (firstStmt != nullptr)
-    {
-        /* append after last statement */
-
-        Statement* lastStmt = preheader->lastStmt();
-        assert(lastStmt->GetNextStmt() == nullptr);
-
-        lastStmt->SetNextStmt(hoistStmt);
-        hoistStmt->SetPrevStmt(lastStmt);
-        firstStmt->SetPrevStmt(hoistStmt);
-    }
-    else
-    {
-        /* Empty pre-header - store the single statement in the block */
-
-        preheader->bbStmtList = hoistStmt;
-        hoistStmt->SetPrevStmt(hoistStmt);
-    }
-
-    hoistStmt->SetNextStmt(nullptr);
+    fgInsertStmtAtEnd(preheader, fgNewStmtFromTree(hoist));
 
 #ifdef DEBUG
     if (verbose)
@@ -3708,12 +3685,6 @@ void Compiler::optPerformHoistExpr(GenTree* origExpr, BasicBlock* exprBb, FlowGr
     }
 #endif
 
-    if (fgNodeThreading == NodeThreading::AllTrees)
-    {
-        gtSetStmtInfo(hoistStmt);
-        fgSetStmtSeq(hoistStmt);
-    }
-
 #ifdef DEBUG
     if (m_nodeTestData != nullptr)
     {
@@ -4304,22 +4275,41 @@ void Compiler::optRecordLoopMemoryDependence(GenTree* tree, BasicBlock* block, V
 }
 
 //------------------------------------------------------------------------
-// optCopyLoopMemoryDependence: record that tree's loop memory dependence
+// optCopyLoopMemoryDependence: Recursively record that tree's loop memory dependence
 //   is the same as some other tree.
 //
 // Arguments:
 //   fromTree -- tree to copy dependence from
 //   toTree -- tree in question
 //
+// Remarks:
+//   This requires 'toTree' to be in its own statement
+//
 void Compiler::optCopyLoopMemoryDependence(GenTree* fromTree, GenTree* toTree)
 {
+    assert(fromTree->OperGet() == toTree->OperGet());
+
     NodeToLoopMemoryBlockMap* const map      = GetNodeToLoopMemoryBlockMap();
     BasicBlock*                     mapBlock = nullptr;
 
     if (map->Lookup(fromTree, &mapBlock))
     {
         map->Set(toTree, mapBlock);
     }
+
+    GenTreeOperandIterator fromIterCur = fromTree->OperandsBegin();
+    GenTreeOperandIterator fromIterEnd = fromTree->OperandsEnd();
+    GenTreeOperandIterator toIterCur   = toTree->OperandsBegin();
+    GenTreeOperandIterator toIterEnd   = toTree->OperandsEnd();
+
+    while (fromIterCur != fromIterEnd)
+    {
+        optCopyLoopMemoryDependence(*fromIterCur, *toIterCur);
+        ++fromIterCur;
+        ++toIterCur;
+    }
+
+    assert(toIterCur == toIterEnd);
 }
 
 //------------------------------------------------------------------------

src/tests/JIT/Regression/JitBlue/Runtime_115109/Runtime_115109.cs

@@ -0,0 +1,46 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+// Generated by Fuzzlyn v2.8 on 2025-04-27 22:46:08
+// Run on X86 Windows
+// Seed: 2255917678885586044-vectort,vector128,vector256,x86aes,x86avx,x86avx2,x86avx512bw,x86avx512bwvl,x86avx512cd,x86avx512cdvl,x86avx512dq,x86avx512dqvl,x86avx512f,x86avx512fvl,x86bmi1,x86bmi2,x86fma,x86lzcnt,x86pclmulqdq,x86popcnt,x86sse,x86sse2,x86sse3,x86sse41,x86sse42,x86ssse3,x86x86base
+// Reduced from 118.3 KiB to 1.0 KiB in 00:17:25
+// Debug: Outputs <0, 0, 0, 0, 0, 0, 0, 0>
+// Release: Outputs <1, 0, 0, 0, 0, 0, 0, 0>
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+using Xunit;
+
+public class Runtime_115109
+{
+    static int s_5 = 101;
+    static Vector<int>[] s_10;
+    static byte s_21;
+    
+    [Fact]
+    public static int TestEntryPoint()
+    {
+        s_10 = [Vector128.CreateScalar(0).AsVector()];
+        bool vr5 = 0 <= M10();
+        return s_10[0][0];
+    }
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    private static byte M10()
+    {
+        byte lvar2 = 3;
+        do
+        {
+            s_5 = 100;
+            int lvar4 = 0;
+            do
+            {
+                s_10[0] = Vector128.CreateScalar(s_5).AsVector();
+            }
+            while (++lvar4 < 2);
+        }
+        while (--lvar2 > 1);
+        return s_21;
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_115109/Runtime_115109.csproj

@@ -0,0 +1,8 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>