ARM64-SVE: Delay free all ops within conditional select

src/coreclr/jit/lsraarm64.cpp

@@ -1944,7 +1944,8 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
         GenTreeHWIntrinsic* embOp2Node = intrin.op2->AsHWIntrinsic();
         size_t              numArgs    = embOp2Node->GetOperandCount();
         const HWIntrinsic   intrinEmb(embOp2Node);
-        numArgs = embOp2Node->GetOperandCount();
+        numArgs              = embOp2Node->GetOperandCount();
+        GenTree* prefUseNode = nullptr;
 
         if (HWIntrinsicInfo::IsFmaIntrinsic(intrinEmb.id))
         {
@@ -1961,44 +1962,15 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
             unsigned resultOpNum =
                 embOp2Node->GetResultOpNumForRmwIntrinsic(user, intrinEmb.op1, intrinEmb.op2, intrinEmb.op3);
 
-            GenTree* emitOp1 = intrinEmb.op1;
-            GenTree* emitOp2 = intrinEmb.op2;
-            GenTree* emitOp3 = intrinEmb.op3;
-
-            if (resultOpNum == 2)
-            {
-                // op2 = op1 + (op2 * op3)
-                std::swap(emitOp1, emitOp3);
-                std::swap(emitOp1, emitOp2);
-                // op1 = (op1 * op2) + op3
-            }
-            else if (resultOpNum == 3)
+            if (resultOpNum == 0)
             {
-                // op3 = op1 + (op2 * op3)
-                std::swap(emitOp1, emitOp3);
-                // op1 = (op1 * op2) + op3
+                prefUseNode = embOp2Node->Op(1);
             }
             else
             {
-                // op1 = op1 + (op2 * op3)
-                // Nothing needs to be done
+                assert(resultOpNum >= 1 && resultOpNum <= 3);
+                prefUseNode = embOp2Node->Op(resultOpNum);
             }
-
-            GenTree* ops[] = {intrinEmb.op1, intrinEmb.op2, intrinEmb.op3};
-            for (GenTree* op : ops)
-            {
-                if (op == emitOp1)
-                {
-                    tgtPrefUse = BuildUse(op);
-                    srcCount++;
-                }
-                else if (op == emitOp2 || op == emitOp3)
-                {
-                    srcCount += BuildDelayFreeUses(op, emitOp1);
-                }
-            }
-
-            srcCount += BuildDelayFreeUses(intrin.op3, emitOp1);
         }
         else
         {
@@ -2045,22 +2017,32 @@ int LinearScan::BuildHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree, int* pDstCou
             {
                 prefUseOpNum = 2;
             }
-            GenTree* prefUseNode = embOp2Node->Op(prefUseOpNum);
-            for (size_t argNum = 1; argNum <= numArgs; argNum++)
+            prefUseNode = embOp2Node->Op(prefUseOpNum);
+        }
+
+        for (size_t argNum = 1; argNum <= numArgs; argNum++)
+        {
+            GenTree* node = embOp2Node->Op(argNum);
+
+            if (node == prefUseNode)
             {
-                if (argNum == prefUseOpNum)
-                {
-                    tgtPrefUse = BuildUse(prefUseNode);
-                    srcCount += 1;
-                }
-                else
-                {
-                    srcCount += BuildDelayFreeUses(embOp2Node->Op(argNum), prefUseNode);
-                }
+                tgtPrefUse = BuildUse(node);
+                srcCount++;
             }
+            else
+            {
+                RefPosition* useRefPosition = nullptr;
 
-            srcCount += BuildDelayFreeUses(intrin.op3, prefUseNode);
+                int uses = BuildDelayFreeUses(node, nullptr, RBM_NONE, &useRefPosition);
+                srcCount += uses;
+
+                // It is a hard requirement that these are not allocated to the same register as the destination,
+                // so verify no optimizations kicked in to skip setting the delay-free.
+                assert((useRefPosition != nullptr && useRefPosition->delayRegFree) || (uses == 0));
+            }
         }
+
+        srcCount += BuildDelayFreeUses(intrin.op3, prefUseNode);
     }
     else if (intrin.op2 != nullptr)
     {

src/tests/JIT/Regression/JitBlue/Runtime_106864/Runtime_106864.cs

@@ -0,0 +1,40 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using Xunit;
+
+// Generated by Fuzzlyn v2.3 on 2024-08-23 09:12:06
+// Run on Arm64 Windows
+// Seed: 9639718980642677114-vectort,vector64,vector128,armsve
+// Reduced from 52.6 KiB to 0.4 KiB in 00:00:26
+// Hits JIT assert in Release:
+// Assertion failed 'targetReg != embMaskOp2Reg' in 'Program:Main(Fuzzlyn.ExecutionServer.IRuntime)' during 'Generate code' (IL size 32; hash 0xade6b36b; FullOpts)
+//
+//     File: C:\dev\dotnet\runtime2\src\coreclr\jit\hwintrinsiccodegenarm64.cpp Line: 818
+//
+using System;
+using System.Numerics;
+using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.Arm;
+
+public class C1
+{
+    public Vector<short> F1;
+}
+
+public class Runtime_106864
+{
+    public static C1 s_2 = new C1();
+
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        if (Sve.IsSupported)
+        {
+            C1 vr2 = s_2;
+            var vr3 = vr2.F1;
+            var vr4 = vr2.F1;
+            vr2.F1 = Sve.Max(vr3, vr4);
+        }
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_106864/Runtime_106864.csproj

@@ -0,0 +1,9 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+    <NoWarn>$(NoWarn),SYSLIB5003</NoWarn>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>

src/tests/JIT/Regression/JitBlue/Runtime_106867/Runtime_106867.cs

@@ -0,0 +1,54 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using Xunit;
+
+// Generated by Fuzzlyn v2.3 on 2024-08-23 10:10:06
+// Run on Arm64 Windows
+// Seed: 13584223539078280353-vectort,vector64,vector128,armsve
+// Reduced from 87.4 KiB to 0.8 KiB in 00:00:52
+// Hits JIT assert in Release:
+// Assertion failed 'secondId->idReg1() != secondId->idReg4()' in 'S0:M6(ubyte,double):this' during 'Emit code' (IL size 81; hash 0x596acd7c; FullOpts)
+//
+//     File: C:\dev\dotnet\runtime2\src\coreclr\jit\emitarm64sve.cpp Line: 18601
+//
+using System;
+using System.Numerics;
+using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.Arm;
+
+public struct S0
+{
+    public void M6(byte arg0, double arg1)
+    {
+        var vr0 = Vector128.CreateScalar(119.12962f).AsVector();
+        var vr3 = Runtime_106867.s_2;
+        var vr4 = Vector128.CreateScalar(1f).AsVector();
+        var vr5 = Runtime_106867.s_2;
+        var vr2 = Sve.FusedMultiplySubtractNegated(vr3, vr4, vr5);
+        if ((Sve.ConditionalExtractLastActiveElement(vr0, 0, vr2) < 0))
+        {
+            this = this;
+        }
+    }
+}
+
+public class Runtime_106867
+{
+    public static Vector<float> s_2;
+    public static double[] s_5 = new double[]
+    {
+        0
+    };
+    public static byte s_16;
+
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        if (Sve.IsSupported)
+        {
+            var vr6 = s_5[0];
+            new S0().M6(s_16, vr6);
+        }
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_106867/Runtime_106867.csproj

@@ -0,0 +1,9 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+    <NoWarn>$(NoWarn),SYSLIB5003</NoWarn>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>

src/tests/JIT/Regression/JitBlue/Runtime_106867/Runtime_106867_1.cs

@@ -0,0 +1,42 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using Xunit;
+using System.Runtime.CompilerServices;
+
+// Generated by Fuzzlyn v2.3 on 2024-08-23 10:12:51
+// Run on Arm64 Windows
+// Seed: 4576767951799510057-vectort,vector64,vector128,armsve
+// Reduced from 32.2 KiB to 0.5 KiB in 00:00:25
+// Hits JIT assert in Release:
+// Assertion failed 'secondId->idReg1() != secondId->idReg3()' in 'Program:Main(Fuzzlyn.ExecutionServer.IRuntime)' during 'Emit code' (IL size 55; hash 0xade6b36b; FullOpts)
+//
+//     File: C:\dev\dotnet\runtime2\src\coreclr\jit\emitarm64sve.cpp Line: 18600
+//
+using System;
+using System.Numerics;
+using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.Arm;
+
+public class Runtime_106866_2
+{
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        if (Sve.IsSupported)
+        {
+            Vector<short> vr4 = default(Vector<short>);
+            vr4 = Sve.MultiplyAdd(vr4, vr4, vr4);
+            var vr5 = (short)0;
+            var vr6 = Vector128.CreateScalar(vr5).AsVector();
+            var vr7 = (short)0;
+            var vr8 = Sve.ConditionalExtractLastActiveElement(vr6, vr7, vr4);
+            Consume(vr8);
+        }
+    }
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    static void Consume<T>(T val)
+    {
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_106867/Runtime_106867_1.csproj

@@ -0,0 +1,9 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+    <NoWarn>$(NoWarn),SYSLIB5003</NoWarn>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>