Ensure to use float reserved register for TYP_SIMD12 (#108629)

* Use Float reserved registers * add test case * Use consistent internal register requirement for TYP_SIMD12 * Also do in loongarch and risc * Use Extract to retrieve internal register for SIMD12
dotnet · Oct 10, 2024 · cb256e2 · cb256e2
1 parent f41e95c
commit cb256e2
Show file tree

Hide file tree

Showing 7 changed files with 119 additions and 4 deletions.
diff --git a/src/coreclr/jit/emitarm64.cpp b/src/coreclr/jit/emitarm64.cpp
@@ -17424,7 +17424,7 @@ void emitter::emitStoreSimd12ToLclOffset(unsigned varNum, unsigned offset, regNu
     else
     {
         // Extract upper 4-bytes from data
-        regNumber tmpReg = codeGen->internalRegisters.GetSingle(tmpRegProvider);
+        regNumber tmpReg = codeGen->internalRegisters.Extract(tmpRegProvider);
         emitIns_R_R_I(INS_mov, EA_4BYTE, tmpReg, dataReg, 2);
 
         // 4-byte write

diff --git a/src/coreclr/jit/lsraarm64.cpp b/src/coreclr/jit/lsraarm64.cpp
@@ -645,7 +645,7 @@ int LinearScan::BuildNode(GenTree* tree)
             {
                 // We need an internal register different from targetReg in which 'tree' produces its result
                 // because both targetReg and internal reg will be in use at the same time.
-                buildInternalFloatRegisterDefForNode(tree, allSIMDRegs());
+                buildInternalIntRegisterDefForNode(tree);
                 setInternalRegsDelayFree = true;
                 buildInternalRegisterUses();
             }

diff --git a/src/coreclr/jit/lsrabuild.cpp b/src/coreclr/jit/lsrabuild.cpp
@@ -4137,9 +4137,15 @@ int LinearScan::BuildStoreLoc(GenTreeLclVarCommon* storeLoc)
 #ifdef FEATURE_SIMD
     if (varTypeIsSIMD(storeLoc) && !op1->IsVectorZero() && (storeLoc->TypeGet() == TYP_SIMD12))
     {
+#ifdef TARGET_ARM64
+        // Need an additional register to extract upper 4 bytes of Vector3,
+        // it has to be float for x86.
+        buildInternalIntRegisterDefForNode(storeLoc);
+#else
         // Need an additional register to extract upper 4 bytes of Vector3,
         // it has to be float for x86.
         buildInternalFloatRegisterDefForNode(storeLoc, allSIMDRegs());
+#endif // TARGET_ARM64
     }
 #endif // FEATURE_SIMD
 

diff --git a/src/coreclr/jit/lsraloongarch64.cpp b/src/coreclr/jit/lsraloongarch64.cpp
@@ -93,7 +93,7 @@ int LinearScan::BuildNode(GenTree* tree)
             {
                 // We need an internal register different from targetReg in which 'tree' produces its result
                 // because both targetReg and internal reg will be in use at the same time.
-                buildInternalFloatRegisterDefForNode(tree, allSIMDRegs());
+                buildInternalIntRegisterDefForNode(tree);
                 setInternalRegsDelayFree = true;
                 buildInternalRegisterUses();
             }

diff --git a/src/coreclr/jit/lsrariscv64.cpp b/src/coreclr/jit/lsrariscv64.cpp
@@ -94,7 +94,7 @@ int LinearScan::BuildNode(GenTree* tree)
             {
                 // We need an internal register different from targetReg in which 'tree' produces its result
                 // because both targetReg and internal reg will be in use at the same time.
-                buildInternalFloatRegisterDefForNode(tree, allSIMDRegs());
+                buildInternalIntRegisterDefForNode(tree);
                 setInternalRegsDelayFree = true;
                 buildInternalRegisterUses();
             }

diff --git a/src/tests/JIT/Regression/JitBlue/Runtime_108609/Runtime_108609.cs b/src/tests/JIT/Regression/JitBlue/Runtime_108609/Runtime_108609.cs
@@ -0,0 +1,101 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+// Found by Antigen
+// Reduced from 50.05 KB to 4.88 KB.
+// Assertion failed 'genExactlyOneBit(availableSet)' in 'TestClass:Method0():this' during 'Generate code' (IL size 782; hash 0x46e9aa75; FullOpts)
+//
+//    File: /Users/runner/work/1/s/src/coreclr/jit/codegencommon.cpp Line: 133
+
+
+using System;
+using System.Collections.Generic;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+using System.Runtime.Intrinsics.Arm;
+using System.Numerics;
+using Xunit;
+
+public class Runtime_108609
+{
+    public struct S1
+    {
+        public Vector3 v3_0;
+        public short short_1;
+    }
+    static bool s_bool_4 = true;
+    static int s_int_10 = 5;
+    static sbyte s_sbyte_12 = -5;
+    static Vector64<int> s_v64_int_23 = Vector64<int>.Zero;
+    static Vector64<double> s_v64_double_28 = Vector64.Create(-4.956521739130435);
+    static Vector128<sbyte> s_v128_sbyte_30 = Vector128<sbyte>.Zero;
+    static Vector128<short> s_v128_short_31 = Vector128.Create(0, 2, -5, 0, 1, 0, 2, 2);
+    static Vector128<ulong> s_v128_ulong_36 = Vector128<ulong>.Zero;
+    static Vector128<double> s_v128_double_38 = Vector128.Create(1.0769230769230769, 72.01666666666667);
+    static Vector<ulong> s_v_ulong_46 = Vector.Create((ulong)1);
+    static Vector<float> s_v_float_47 = Vector<float>.Zero;
+    static S1 s_s1_52 = new S1();
+    char char_56 = 'K';
+    int int_60 = -1;
+    Vector64<int> v64_int_73 = Vector64.Create(2147483646, -1);
+    Vector64<double> v64_double_78 = Vector64.Create(-0.9420289855072463);
+    Vector128<sbyte> v128_sbyte_80 = Vector128.Create(-2, -2, -1, 0, 5, -1, -1, -1, 1, 2, -1, 0, -1, 72, -5, 0);
+    Vector128<short> v128_short_81 = Vector128.CreateScalar((short)-1);
+    Vector128<ulong> v128_ulong_86 = Vector128<ulong>.AllBitsSet;
+    Vector<ulong> v_ulong_96 = Vector.Create((ulong)1);
+    Vector<float> v_float_97 = Vector.Create(2f);
+    Vector3 v3_100 = Vector3.Create(0.09090909f, 2.4f, -0.94f);
+    S1 s1_102 = new S1();
+    static int s_loopInvariant = 4;
+    private static List<string> toPrint = new List<string>();
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    private decimal LeafMethod3()
+    {
+        return 19;
+    }
+
+    private Vector<float> Method2(decimal p_decimal_130, decimal p_decimal_131, sbyte p_sbyte_132, out Vector128<ulong> p_v128_ulong_133, S1 p_s1_134)
+    {
+        p_v128_ulong_133 = s_v128_ulong_36;
+        return Vector<float>.Zero;
+    }
+
+    private Vector128<sbyte> Method3(sbyte p_sbyte_152, Vector<ulong> p_v_ulong_153, S1 p_s1_154, uint p_uint_155, S1 p_s1_156, decimal p_decimal_157, Vector64<double> p_v64_double_158)
+    {
+        unchecked
+        {
+            return s_v128_sbyte_30 = s_v128_sbyte_30;
+        }
+    }
+
+    private void Method0()
+    {
+        unchecked
+        {
+            S1 s1_218 = new S1();
+            if (!s_bool_4)
+            {
+                int __loopvar0 = s_loopInvariant;
+                for (int __loopSecondaryVar0_0 = 15 - 4; s_int_10 < Vector64.GetElement(s_v64_int_23 - Vector64<int>.Zero * (15 + 4) | (s_v64_int_23 += AdvSimd.Max(Vector64<int>.Zero, v64_int_73)), 15 & 4); __loopvar0--, __loopSecondaryVar0_0++, s_s1_52.short_1 = Vector128.Sum(AdvSimd.ShiftLogicalRoundedSaturate(Vector128<short>.AllBitsSet, v128_short_81) & Vector128.AsInt16(s_v128_double_38) - (15 | 4) * (v128_short_81 *= s_v128_short_31)))
+                {
+                    v3_100 *= s1_218.v3_0 - (s1_218.v3_0 = s1_218.v3_0 = s1_218.v3_0 *= s1_218.v3_0) * Vector3.RadiansToDegrees(Vector3.Zero + s_s1_52.v3_0 + v3_100 * v3_100) - Vector3.Sin(s1_218.v3_0 *= v3_100 * (s1_218.v3_0 *= s1_218.v3_0) - v3_100 * s_s1_52.v3_0);
+                }
+            }
+            v_float_97 = Method2(15 % 4, LeafMethod3(), s_sbyte_12, out v128_ulong_86, s1_218);
+            v128_sbyte_80 = Method3(15 % 4, s_v_ulong_46 *= s_v_ulong_46 += v_ulong_96 = s_v_ulong_46 ^ v_ulong_96 * (15 - 4), s1_218, 15 & 4, s1_102, LeafMethod3(), v64_double_78 * s_v64_double_28 - (s_v64_double_28 = v64_double_78) | Vector64.GreaterThan(v64_double_78, v64_double_78) | (s_v64_double_28 = Vector64<double>.Zero + v64_double_78) | (s_v64_double_28 = v64_double_78) | Vector64.GreaterThan(v64_double_78, v64_double_78) | (s_v64_double_28 = Vector64<double>.Zero + v64_double_78) | s_v64_double_28);
+            return;
+        }
+    }
+
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        Antigen();
+    }
+
+    private static int Antigen()
+    {
+        new Runtime_108609().Method0();
+        return string.Join(Environment.NewLine, toPrint).GetHashCode();
+    }
+}
diff --git a/src/tests/JIT/Regression/JitBlue/Runtime_108609/Runtime_108609.csproj b/src/tests/JIT/Regression/JitBlue/Runtime_108609/Runtime_108609.csproj
@@ -0,0 +1,8 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>