NativeAOT: IlcOptimizationPreference & IlcInstructionSet (#1994)

adamsitnik · web-flow · commit 1c935dc71ea0 · 2022-04-17T13:35:28.000+02:00
* allow the users to specify IlcOptimizationPreference, use "Speed" as default

* allow the users to specify Instruction Set

* specify platform in explicit way

* provide default Instruction Set based on host process
diff --git a/src/BenchmarkDotNet/Toolchains/NativeAot/Generator.cs b/src/BenchmarkDotNet/Toolchains/NativeAot/Generator.cs
@@ -4,6 +4,8 @@
 using System.IO;
 using System.Linq;
 using System.Text;
+using BenchmarkDotNet.Environments;
+using BenchmarkDotNet.Extensions;
 using BenchmarkDotNet.Loggers;
 using BenchmarkDotNet.Portability;
 using BenchmarkDotNet.Running;
@@ -26,7 +28,8 @@ internal Generator(string ilCompilerVersion,
             string runtimeFrameworkVersion, string targetFrameworkMoniker, string cliPath,
             string runtimeIdentifier, IReadOnlyDictionary<string, string> feeds, bool useNuGetClearTag,
             bool useTempFolderForRestore, string packagesRestorePath,
-            bool rootAllApplicationAssemblies, bool ilcGenerateCompleteTypeMetadata, bool ilcGenerateStackTraceData)
+            bool rootAllApplicationAssemblies, bool ilcGenerateCompleteTypeMetadata, bool ilcGenerateStackTraceData,
+            string ilcOptimizationPreference, string ilcInstructionSet)
             : base(targetFrameworkMoniker, cliPath, GetPackagesDirectoryPath(useTempFolderForRestore, packagesRestorePath), runtimeFrameworkVersion)
         {
             this.ilCompilerVersion = ilCompilerVersion;
@@ -38,6 +41,8 @@ internal Generator(string ilCompilerVersion,
             this.rootAllApplicationAssemblies = rootAllApplicationAssemblies;
             this.ilcGenerateCompleteTypeMetadata = ilcGenerateCompleteTypeMetadata;
             this.ilcGenerateStackTraceData = ilcGenerateStackTraceData;
+            this.ilcOptimizationPreference = ilcOptimizationPreference;
+            this.ilcInstructionSet = ilcInstructionSet;
         }
 
         private readonly string ilCompilerVersion;
@@ -50,6 +55,8 @@ internal Generator(string ilCompilerVersion,
         private readonly bool rootAllApplicationAssemblies;
         private readonly bool ilcGenerateCompleteTypeMetadata;
         private readonly bool ilcGenerateStackTraceData;
+        private readonly string ilcOptimizationPreference;
+        private readonly string ilcInstructionSet;
 
         private bool IsNuGet => feeds.ContainsKey(NativeAotNuGetFeed) && !string.IsNullOrWhiteSpace(ilCompilerVersion);
 
@@ -127,11 +134,13 @@ private string GenerateProjectForNuGetBuild(BuildPartition buildPartition, Artif
     <AssemblyName>{artifactsPaths.ProgramName}</AssemblyName>
     <AssemblyTitle>{artifactsPaths.ProgramName}</AssemblyTitle>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+    <PlatformTarget>{buildPartition.Platform.ToConfig()}</PlatformTarget>
     <TreatWarningsAsErrors>False</TreatWarningsAsErrors>
     <DebugSymbols>false</DebugSymbols>
     <UseSharedCompilation>false</UseSharedCompilation>
     <Deterministic>true</Deterministic>
     <RunAnalyzers>false</RunAnalyzers>
+    <IlcOptimizationPreference>{ilcOptimizationPreference}</IlcOptimizationPreference>
     {GetTrimmingSettings()}
     <IlcGenerateCompleteTypeMetadata>{ilcGenerateCompleteTypeMetadata}</IlcGenerateCompleteTypeMetadata>
     <IlcGenerateStackTraceData>{ilcGenerateStackTraceData}</IlcGenerateStackTraceData>
@@ -149,6 +158,9 @@ private string GenerateProjectForNuGetBuild(BuildPartition buildPartition, Artif
   <ItemGroup>
     {string.Join(Environment.NewLine, GetRdXmlFiles(buildPartition.RepresentativeBenchmarkCase.Descriptor.Type, logger).Select(file => $"<RdXmlFile Include=\"{file}\" />"))}
   </ItemGroup>
+  <ItemGroup>
+    {GetInstructionSetSettings(buildPartition)}
+  </ItemGroup>
 </Project>";
 
         private string GenerateProjectForLocalBuild(BuildPartition buildPartition, ArtifactsPaths artifactsPaths, ILogger logger) => $@"
@@ -162,11 +174,13 @@ private string GenerateProjectForLocalBuild(BuildPartition buildPartition, Artif
     <AssemblyName>{artifactsPaths.ProgramName}</AssemblyName>
     <AssemblyTitle>{artifactsPaths.ProgramName}</AssemblyTitle>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+    <PlatformTarget>{buildPartition.Platform.ToConfig()}</PlatformTarget>
     <TreatWarningsAsErrors>False</TreatWarningsAsErrors>
     <DebugSymbols>false</DebugSymbols>
     <UseSharedCompilation>false</UseSharedCompilation>
     <Deterministic>true</Deterministic>
     <RunAnalyzers>false</RunAnalyzers>
+    <IlcOptimizationPreference>{ilcOptimizationPreference}</IlcOptimizationPreference>
     {GetTrimmingSettings()}
     <IlcGenerateCompleteTypeMetadata>{ilcGenerateCompleteTypeMetadata}</IlcGenerateCompleteTypeMetadata>
     <IlcGenerateStackTraceData>{ilcGenerateStackTraceData}</IlcGenerateStackTraceData>
@@ -184,6 +198,9 @@ private string GenerateProjectForLocalBuild(BuildPartition buildPartition, Artif
   <ItemGroup>
     {string.Join(Environment.NewLine, GetRdXmlFiles(buildPartition.RepresentativeBenchmarkCase.Descriptor.Type, logger).Select(file => $"<RdXmlFile Include=\"{file}\" />"))}
   </ItemGroup>
+  <ItemGroup>
+    {GetInstructionSetSettings(buildPartition)}
+  </ItemGroup>
 </Project>";
 
         private string GetTrimmingSettings()
@@ -192,6 +209,14 @@ private string GetTrimmingSettings()
                 // TrimMode is set in explicit way as for older versions it might have different default value
                 : "<TrimMode>link</TrimMode><TrimmerDefaultAction>link</TrimmerDefaultAction>";
 
+        private string GetInstructionSetSettings(BuildPartition buildPartition)
+        {
+            string instructionSet = ilcInstructionSet ?? GetCurrentInstructionSet(buildPartition.Platform);
+            return !string.IsNullOrEmpty(instructionSet)
+                ? $@"<IlcArg Include=""--instructionset:{instructionSet}"" />"
+                : "";
+        }
+
         public IEnumerable<string> GetRdXmlFiles(Type benchmarkTarget, ILogger logger)
         {
             yield return GeneratedRdXmlFileName;
@@ -237,5 +262,56 @@ private void GenerateReflectionFile(ArtifactsPaths artifactsPaths)
             else
                 throw new InvalidOperationException($"Can't get directory of projectFilePath ('{artifactsPaths.ProjectFilePath}')");
         }
+
+        private string GetCurrentInstructionSet(Platform platform)
+            => string.Join(",", GetHostProcessInstructionSets(platform));
+
+        // based on https://github.com/dotnet/runtime/blob/ce61c09a5f6fc71d8f717d3fc4562f42171869a0/src/coreclr/tools/Common/JitInterface/CorInfoInstructionSet.cs#L727
+        private static IEnumerable<string> GetHostProcessInstructionSets(Platform platform)
+        {
+            switch (platform)
+            {
+                case Platform.X86:
+                case Platform.X64:
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.X86Base")) yield return "base";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Sse")) yield return "sse";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Sse2")) yield return "sse2";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Sse3")) yield return "sse3";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Sse41")) yield return "sse4.1";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Sse42")) yield return "sse4.2";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Avx")) yield return "avx";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Avx2")) yield return "avx2";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Aes")) yield return "aes";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Bmi1")) yield return "bmi";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Bmi2")) yield return "bmi2";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Fma")) yield return "fma";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Lzcnt")) yield return "lzcnt";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Pclmulqdq")) yield return "pclmul";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.Popcnt")) yield return "popcnt";
+                    if (GetIsSupported("System.Runtime.Intrinsics.X86.AvxVnni")) yield return "avxvnni";
+                    break;
+                case Platform.Arm64:
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.ArmBase")) yield return "base";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.AdvSimd")) yield return "neon";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Aes")) yield return "aes";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Crc32")) yield return "crc";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Dp")) yield return "dotprod";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Rdm")) yield return "rdma";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Sha1")) yield return "sha1";
+                    if (GetIsSupported("System.Runtime.Intrinsics.Arm.Sha256")) yield return "sha2";
+                    // todo: handle "lse"
+                    break;
+                default:
+                    yield break;
+            }
+        }
+
+        private static bool GetIsSupported(string typeName)
+        {
+            Type type = Type.GetType(typeName);
+            if (type == null) return false;
+
+            return (bool)type.GetProperty("IsSupported", System.Reflection.BindingFlags.Public | System.Reflection.BindingFlags.Static).GetValue(null, null);
+        }
     }
 }
diff --git a/src/BenchmarkDotNet/Toolchains/NativeAot/NativeAotToolchain.cs b/src/BenchmarkDotNet/Toolchains/NativeAot/NativeAotToolchain.cs
@@ -28,11 +28,13 @@ internal NativeAotToolchain(string displayName,
             string runtimeFrameworkVersion, string targetFrameworkMoniker, string runtimeIdentifier,
             string customDotNetCliPath, string packagesRestorePath,
             Dictionary<string, string> feeds, bool useNuGetClearTag, bool useTempFolderForRestore,
-            bool rootAllApplicationAssemblies, bool ilcGenerateCompleteTypeMetadata, bool ilcGenerateStackTraceData)
+            bool rootAllApplicationAssemblies, bool ilcGenerateCompleteTypeMetadata, bool ilcGenerateStackTraceData,
+            string ilcOptimizationPreference, string ilcInstructionSet)
             : base(displayName,
                 new Generator(ilCompilerVersion, runtimeFrameworkVersion, targetFrameworkMoniker, customDotNetCliPath,
                     runtimeIdentifier, feeds, useNuGetClearTag, useTempFolderForRestore, packagesRestorePath,
-                    rootAllApplicationAssemblies, ilcGenerateCompleteTypeMetadata, ilcGenerateStackTraceData),
+                    rootAllApplicationAssemblies, ilcGenerateCompleteTypeMetadata, ilcGenerateStackTraceData,
+                    ilcOptimizationPreference, ilcInstructionSet),
                 new DotNetCliPublisher(customDotNetCliPath, GetExtraArguments(runtimeIdentifier), GetEnvironmentVariables(ilcPath)),
                 new Executor())
         {
diff --git a/src/BenchmarkDotNet/Toolchains/NativeAot/NativeAotToolchainBuilder.cs b/src/BenchmarkDotNet/Toolchains/NativeAot/NativeAotToolchainBuilder.cs
@@ -17,6 +17,8 @@ public class NativeAotToolchainBuilder : CustomDotNetCliToolchainBuilder
         private bool rootAllApplicationAssemblies;
         private bool ilcGenerateCompleteTypeMetadata = true;
         private bool ilcGenerateStackTraceData = true;
+        private string ilcOptimizationPreference = "Speed";
+        private string ilcInstructionSet = null;
 
         private bool isIlCompilerConfigured;
 
@@ -110,6 +112,34 @@ public NativeAotToolchainBuilder IlcGenerateStackTraceData(bool value)
             return this;
         }
 
+        /// <summary>
+        /// Options related to code generation.
+        /// </summary>
+        /// <param name="value">"Speed" to favor code execution speed (default), "Size" to favor smaller code size</param>
+        [PublicAPI]
+        public NativeAotToolchainBuilder IlcOptimizationPreference(string value = "Speed")
+        {
+            ilcOptimizationPreference = value;
+
+            return this;
+        }
+
+        /// <summary>
+        /// By default, the compiler targets the minimum instruction set supported by the target OS and architecture.
+        /// This option allows targeting newer instruction sets for better performance.
+        /// The native binary will require the instruction sets to be supported by the hardware in order to run.
+        /// For example, `avx2,bmi2,fma,pclmul,popcnt,aes` will produce binary that takes advantage of instruction sets
+        /// that are typically present on current Intel and AMD processors.
+        /// </summary>
+        /// <param name="value">Specify empty string ("", not null) to use the defaults.</param>
+        [PublicAPI]
+        public NativeAotToolchainBuilder IlcInstructionSet(string value)
+        {
+            ilcInstructionSet = value;
+
+            return this;
+        }
+
         [PublicAPI]
         public override IToolchain ToToolchain()
         {
@@ -130,7 +160,9 @@ public override IToolchain ToToolchain()
                 useTempFolderForRestore: useTempFolderForRestore,
                 rootAllApplicationAssemblies: rootAllApplicationAssemblies,
                 ilcGenerateCompleteTypeMetadata: ilcGenerateCompleteTypeMetadata,
-                ilcGenerateStackTraceData: ilcGenerateStackTraceData
+                ilcGenerateStackTraceData: ilcGenerateStackTraceData,
+                ilcOptimizationPreference: ilcOptimizationPreference,
+                ilcInstructionSet: ilcInstructionSet
             );
         }
     }
diff --git a/tests/BenchmarkDotNet.IntegrationTests/NativeAotTests.cs b/tests/BenchmarkDotNet.IntegrationTests/NativeAotTests.cs
@@ -5,6 +5,7 @@
 using BenchmarkDotNet.Jobs;
 using BenchmarkDotNet.Portability;
 using BenchmarkDotNet.Tests.XUnit;
+using BenchmarkDotNet.Toolchains.NativeAot;
 using Xunit.Abstractions;
 
 namespace BenchmarkDotNet.IntegrationTests
@@ -23,21 +24,40 @@ public void LatestNativeAotVersionIsSupported()
             if (ContinuousIntegration.IsAppVeyorOnWindows()) // too time consuming for AppVeyor (1h limit)
                 return;
 
+            var toolchain = NativeAotToolchain.CreateBuilder().UseNuGet().IlcInstructionSet(IsAvx2Supported() ? "avx2" : "").ToToolchain();
+
             var config = ManualConfig.CreateEmpty()
                 .AddJob(Job.Dry
-                    .WithRuntime(NativeAotRuntime.GetCurrentVersion())); // we test against latest version for current TFM to make sure we avoid issues like #1055
+                    .WithRuntime(NativeAotRuntime.GetCurrentVersion()) // we test against latest version for current TFM to make sure we avoid issues like #1055
+                    .WithToolchain(toolchain)
+                    .WithEnvironmentVariable(NativeAotBenchmark.EnvVarKey, IsAvx2Supported().ToString().ToLower()));
 
             CanExecute<NativeAotBenchmark>(config);
         }
+
+        private bool IsAvx2Supported()
+        {
+#if NET6_0_OR_GREATER
+            return System.Runtime.Intrinsics.X86.Avx2.IsSupported;
+#else
+            return false;
+#endif
+        }
     }
 
     public class NativeAotBenchmark
     {
+        internal const string EnvVarKey = "AVX2_IsSupported";
+
         [Benchmark]
         public void Check()
         {
             if (!RuntimeInformation.IsNativeAOT)
                 throw new Exception("This is NOT NativeAOT");
+#if NET6_0_OR_GREATER
+            if (System.Runtime.Intrinsics.X86.Avx2.IsSupported != bool.Parse(Environment.GetEnvironmentVariable(EnvVarKey)))
+                throw new Exception("Unexpected Instruction Set");
+#endif
         }
     }
 }
