Fix side effect handling for certain TernaryLogic cases (dotnet#114848)

There were a few cases of TernaryLogic with NOT/AND and NOT/OR that
swapped operands without considering side effects. Add some new logic to
hoist those side effects if necessary.

Fixes dotnet#114324

Author: BruceForstall

src/coreclr/jit/hwintrinsicxarch.cpp

@@ -4405,6 +4405,65 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic        intrinsic,
                         }
                     }
 
+                    // Some normalization cases require us to swap the operands, which might require
+                    // spilling side effects. Check that here.
+                    //
+                    // cast in switch clause is needed for old gcc
+                    switch ((TernaryLogicOperKind)info.oper1)
+                    {
+                        case TernaryLogicOperKind::Not:
+                        {
+                            assert(info.oper1Use != TernaryLogicUseFlags::None);
+
+                            bool needSideEffectSpill = false;
+
+                            if (info.oper2 == TernaryLogicOperKind::And)
+                            {
+                                assert(info.oper2Use != TernaryLogicUseFlags::None);
+
+                                if ((control == static_cast<uint8_t>(~0xCC & 0xF0)) || // ~B & A
+                                    (control == static_cast<uint8_t>(~0xAA & 0xF0)) || // ~C & A
+                                    (control == static_cast<uint8_t>(~0xAA & 0xCC)))   // ~C & B
+                                {
+                                    // We're normalizing to ~B & C, so we need another swap
+                                    std::swap(val2, val3);
+                                    needSideEffectSpill = (control == static_cast<uint8_t>(~0xAA & 0xCC)); // ~C & B
+                                }
+                            }
+                            else if (info.oper2 == TernaryLogicOperKind::Or)
+                            {
+                                assert(info.oper2Use != TernaryLogicUseFlags::None);
+
+                                if ((control == static_cast<uint8_t>(~0xCC | 0xF0)) || // ~B | A
+                                    (control == static_cast<uint8_t>(~0xAA | 0xF0)) || // ~C | A
+                                    (control == static_cast<uint8_t>(~0xAA | 0xCC)))   // ~C | B
+                                {
+                                    // We're normalizing to ~B | C, so we need another swap
+                                    std::swap(val2, val3);
+                                    needSideEffectSpill = (control == static_cast<uint8_t>(~0xAA | 0xCC)); // ~C | B
+                                }
+                            }
+
+                            if (needSideEffectSpill)
+                            {
+                                // Side-effect cases:
+                                // ~B op A ; order before swap C A B
+                                //    op1 & op2 already set to be spilled; no further spilling necessary
+                                // ~C op A ; order before swap B A C
+                                //    op1 already set to be spilled; no further spilling necessary
+                                // ~C op B ; order before swap A B C
+                                //    nothing already set to be spilled; op1 & op2 need to be spilled
+
+                                spillOp1 = true;
+                                spillOp2 = true;
+                            }
+                            break;
+                        }
+
+                        default:
+                            break;
+                    }
+
                     if (spillOp1)
                     {
                         impSpillSideEffect(true, stackState.esStackDepth -
@@ -4529,8 +4588,7 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic        intrinsic,
                                     (control == static_cast<uint8_t>(~0xAA & 0xF0)) || // ~C & A
                                     (control == static_cast<uint8_t>(~0xAA & 0xCC)))   // ~C & B
                                 {
-                                    // We're normalizing to ~B & C, so we need another swap
-                                    std::swap(*val2, *val3);
+                                    // We already normalized to ~B & C above.
                                 }
                                 else
                                 {
@@ -4556,8 +4614,7 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic        intrinsic,
                                     (control == static_cast<uint8_t>(~0xAA | 0xF0)) || // ~C | A
                                     (control == static_cast<uint8_t>(~0xAA | 0xCC)))   // ~C | B
                                 {
-                                    // We're normalizing to ~B & C, so we need another swap
-                                    std::swap(*val2, *val3);
+                                    // We already normalized to ~B | C above.
                                 }
                                 else
                                 {

src/tests/JIT/Regression/JitBlue/Runtime_114324/Runtime_114324.cs

@@ -0,0 +1,51 @@
+// 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 v2.5 on 2025-04-06 16:30:45
+// Run on X64 Windows
+// Seed: 13621799076737484256-vectort,vector128,vector256,x86aes,x86avx,x86avx2,x86avx512bw,x86avx512bwvl,x86avx512cd,x86avx512cdvl,x86avx512dq,x86avx512dqvl,x86avx512f,x86avx512fvl,x86avx512fx64,x86bmi1,x86bmi1x64,x86bmi2,x86bmi2x64,x86fma,x86lzcnt,x86lzcntx64,x86pclmulqdq,x86popcnt,x86popcntx64,x86sse,x86ssex64,x86sse2,x86sse2x64,x86sse3,x86sse41,x86sse41x64,x86sse42,x86sse42x64,x86ssse3,x86x86base
+// Reduced from 20.1 KiB to 0.7 KiB in 00:01:13
+// Debug: Runs successfully
+// Release: Throws 'System.NullReferenceException'
+using System;
+using System.Numerics;
+using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.X86;
+using Xunit;
+
+public class Runtime_114324
+{
+    public static ulong[] s_3;
+
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        if (Avx512F.VL.IsSupported)
+        {
+            var vr1 = Vector128.Create<int>(1);
+            Vector128<int> vr2 = Avx512F.VL.TernaryLogic(
+                    vr1,
+                    Sse2.ShiftLeftLogical128BitLane(
+                        Vector128.Create<int>((int)Sse42.Crc32(M2(), 0)),
+                        0),
+                    Avx512F.VL.ConvertToVector128Int32(Vector256.Create<ulong>(s_3[0])),
+                    221);
+            System.Console.WriteLine(vr2);
+            Assert.Equal(Vector128.Create(-1,-1,-1,-1), vr2);
+        }
+    }
+
+    public static uint M2()
+    {
+        s_3 = new ulong[]
+        {
+            0
+        };
+        return M3();
+    }
+
+    public static byte M3()
+    {
+        return default(byte);
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_114324/Runtime_114324.csproj

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