arm64 disassembler (#2127)

* make ClrMdV2Disassembler abstract, introduce Arm and Intel disassemblers

* Capstone PoC

* Implement TryGetReferencedAddress for relative branches (#2107)

* Resolve indirect addresses in disassembly (#2118)

Co-authored-by: Jan Vorlicek <[email protected]>

* Initial version of the Arm64 instruction formatter (#2119)

* Added other arm64 constant form extraction plus other changes (#2123)

Co-authored-by: Jan Vorlicek <[email protected]>
Co-authored-by: Jan Vorlicek <[email protected]>

Author: adamsitnik

samples/BenchmarkDotNet.Samples/BenchmarkDotNet.Samples.csproj

@@ -24,4 +24,4 @@
     <ProjectReference Include="..\..\src\BenchmarkDotNet\BenchmarkDotNet.csproj" />
     <ProjectReference Include="..\..\src\BenchmarkDotNet.Diagnostics.Windows\BenchmarkDotNet.Diagnostics.Windows.csproj" />
   </ItemGroup>
-</Project>
+</Project>

samples/BenchmarkDotNet.Samples/IntroDisassembly.cs

@@ -1,9 +1,10 @@
 using BenchmarkDotNet.Attributes;
+using BenchmarkDotNet.Diagnosers;
 using System.Linq;
 
 namespace BenchmarkDotNet.Samples
 {
-    [DisassemblyDiagnoser]
+    [DisassemblyDiagnoser(printInstructionAddresses: true, syntax: DisassemblySyntax.Masm)]
     public class IntroDisassembly
     {
         private int[] field = Enumerable.Range(0, 100).ToArray();

src/BenchmarkDotNet.Disassembler.x64/BenchmarkDotNet.Disassembler.x64.csproj

@@ -8,6 +8,7 @@
     <RuntimeIdentifier>win7-x64</RuntimeIdentifier>
     <PlatformTarget>x64</PlatformTarget>
     <SuppressNETCoreSdkPreviewMessage>True</SuppressNETCoreSdkPreviewMessage>
+    <DefineConstants>$(DefineConstants);CLRMDV1</DefineConstants>
   </PropertyGroup>
   <PropertyGroup>
     <OutputPath>..\BenchmarkDotNet\Disassemblers</OutputPath>

src/BenchmarkDotNet.Disassembler.x64/ClrMdV1Disassembler.cs

@@ -25,7 +25,7 @@ internal static DisassemblyResult AttachAndDisassemble(Settings settings)
 
                 ConfigureSymbols(dataTarget);
 
-                var state = new State(runtime);
+                var state = new State(runtime, settings.TargetFrameworkMoniker);
 
                 if (settings.Filters.Length > 0)
                 {
@@ -169,21 +169,23 @@ private static IEnumerable<Asm> Decode(ulong startAddress, uint size, State stat
             {
                 decoder.Decode(out var instruction);
 
-                TryTranslateAddressToName(instruction, state, depth, currentMethod);
+                TryTranslateAddressToName(instruction, state, depth, currentMethod, out ulong referencedAddress);
 
                 yield return new Asm
                 {
                     InstructionPointer = instruction.IP,
-                    Instruction = instruction
+                    InstructionLength = instruction.Length,
+                    IntelInstruction = instruction,
+                    ReferencedAddress = (referencedAddress > ushort.MaxValue) ? referencedAddress : null,
                 };
             }
         }
 
-        private static void TryTranslateAddressToName(Instruction instruction, State state, int depth, ClrMethod currentMethod)
+        private static void TryTranslateAddressToName(Instruction instruction, State state, int depth, ClrMethod currentMethod, out ulong address)
         {
             var runtime = state.Runtime;
 
-            if (!TryGetReferencedAddress(instruction, (uint)runtime.PointerSize, out ulong address))
+            if (!TryGetReferencedAddress(instruction, (uint)runtime.PointerSize, out address))
                 return;
 
             if (state.AddressToNameMapping.ContainsKey(address))

src/BenchmarkDotNet.Disassembler.x64/DataContracts.cs

@@ -24,7 +24,14 @@ public class Sharp : SourceCode
 
     public class Asm : SourceCode
     {
-        public Instruction Instruction { get; set; }
+        public int InstructionLength { get; set; }
+        public ulong? ReferencedAddress { get; set; }
+        public bool IsReferencedAddressIndirect { get; set; }
+
+        public Instruction? IntelInstruction { get; set; }
+#if !CLRMDV1
+        public Gee.External.Capstone.Arm64.Arm64Instruction Arm64Instruction { get; set; }
+#endif
     }
 
     public class MonoCode : SourceCode
@@ -101,14 +108,16 @@ public static class DisassemblerConstants
 
     internal class Settings
     {
-        internal Settings(int processId, string typeName, string methodName, bool printSource, int maxDepth, string resultsPath, string[] filters)
+        internal Settings(int processId, string typeName, string methodName, bool printSource, int maxDepth, string resultsPath, string syntax, string tfm, string[] filters)
         {
             ProcessId = processId;
             TypeName = typeName;
             MethodName = methodName;
             PrintSource = printSource;
             MaxDepth = methodName == DisassemblerConstants.DisassemblerEntryMethodName && maxDepth != int.MaxValue ? maxDepth + 1 : maxDepth;
             ResultsPath = resultsPath;
+            Syntax = syntax;
+            TargetFrameworkMoniker = tfm;
             Filters = filters;
         }
 
@@ -118,6 +127,8 @@ internal Settings(int processId, string typeName, string methodName, bool printS
         internal bool PrintSource { get; }
         internal int MaxDepth { get; }
         internal string[] Filters;
+        internal string Syntax { get; }
+        internal string TargetFrameworkMoniker { get; }
         internal string ResultsPath { get; }
 
         internal static Settings FromArgs(string[] args)
@@ -128,24 +139,54 @@ internal static Settings FromArgs(string[] args)
                 printSource: bool.Parse(args[3]),
                 maxDepth: int.Parse(args[4]),
                 resultsPath: args[5],
-                filters: args.Skip(6).ToArray()
+                syntax: args[6],
+                tfm: args[7],
+                filters: args.Skip(8).ToArray()
             );
     }
 
     internal class State
     {
-        internal State(ClrRuntime runtime)
+        internal State(ClrRuntime runtime, string targetFrameworkMoniker)
         {
             Runtime = runtime;
             Todo = new Queue<MethodInfo>();
             HandledMethods = new HashSet<ClrMethod>(new ClrMethodComparer());
             AddressToNameMapping = new Dictionary<ulong, string>();
+            RuntimeVersion = ParseVersion(targetFrameworkMoniker);
         }
 
         internal ClrRuntime Runtime { get; }
+        internal string TargetFrameworkMoniker { get; }
         internal Queue<MethodInfo> Todo { get; }
         internal HashSet<ClrMethod> HandledMethods { get; }
         internal Dictionary<ulong, string> AddressToNameMapping { get; }
+        internal Version RuntimeVersion { get; }
+
+        internal static Version ParseVersion(string targetFrameworkMoniker)
+        {
+            int firstDigit = -1, lastDigit = -1;
+            for (int i = 0; i < targetFrameworkMoniker.Length; i++)
+            {
+                if (char.IsDigit(targetFrameworkMoniker[i]))
+                {
+                    if (firstDigit == -1)
+                        firstDigit = i;
+
+                    lastDigit = i;
+                }
+                else if (targetFrameworkMoniker[i] == '-')
+                {
+                    break; // it can be platform specific like net7.0-windows8
+                }
+            }
+
+            string versionToParse = targetFrameworkMoniker.Substring(firstDigit, lastDigit - firstDigit + 1);
+            if (!versionToParse.Contains(".")) // Full .NET Framework (net48 etc)
+                versionToParse = string.Join(".", versionToParse.ToCharArray());
+
+            return Version.Parse(versionToParse);
+        }
 
         private sealed class ClrMethodComparer : IEqualityComparer<ClrMethod>
         {

src/BenchmarkDotNet.Disassembler.x86/BenchmarkDotNet.Disassembler.x86.csproj

@@ -8,6 +8,7 @@
     <RuntimeIdentifier>win7-x86</RuntimeIdentifier>
     <PlatformTarget>x86</PlatformTarget>
     <SuppressNETCoreSdkPreviewMessage>True</SuppressNETCoreSdkPreviewMessage>
+    <DefineConstants>$(DefineConstants);CLRMDV1</DefineConstants>
   </PropertyGroup>
   <PropertyGroup>
     <OutputPath>..\BenchmarkDotNet\Disassemblers</OutputPath>

src/BenchmarkDotNet/Attributes/DisassemblyDiagnoserAttribute.cs

@@ -8,6 +8,7 @@ namespace BenchmarkDotNet.Attributes
     public class DisassemblyDiagnoserAttribute : Attribute, IConfigSource
     {
         /// <param name="maxDepth">Includes called methods to given level. 1 by default, indexed from 1. To print just the benchmark set it to 0.</param>
+        /// <param name="syntax">The disassembly syntax. MASM is the default.</param>
         /// <param name="printSource">C#|F#|VB source code will be printed. False by default.</param>
         /// <param name="printInstructionAddresses">Print instruction addresses. False by default</param>
         /// <param name="exportGithubMarkdown">Exports to GitHub markdown. True by default.</param>
@@ -17,6 +18,7 @@ public class DisassemblyDiagnoserAttribute : Attribute, IConfigSource
         /// <param name="filters">Glob patterns applied to full method signatures by the the disassembler.</param>
         public DisassemblyDiagnoserAttribute(
             int maxDepth = 1,
+            DisassemblySyntax syntax = DisassemblySyntax.Masm,
             bool printSource = false,
             bool printInstructionAddresses = false,
             bool exportGithubMarkdown = true,
@@ -29,6 +31,7 @@ public DisassemblyDiagnoserAttribute(
                 new DisassemblyDiagnoser(
                     new DisassemblyDiagnoserConfig(
                         maxDepth: maxDepth,
+                        syntax: syntax,
                         filters: filters,
                         printSource: printSource,
                         printInstructionAddresses: printInstructionAddresses,

src/BenchmarkDotNet/BenchmarkDotNet.csproj

@@ -4,7 +4,7 @@
     <AssemblyTitle>BenchmarkDotNet</AssemblyTitle>
     <TargetFrameworks>netstandard2.0;net6.0</TargetFrameworks>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <NoWarn>$(NoWarn);1701;1702;1705;1591;3005;NU1702;CS3001;CS3003</NoWarn>
+    <NoWarn>$(NoWarn);1701;1702;1705;1591;3005;NU1702;CS3001;CS3003;CS8002</NoWarn>
     <AssemblyName>BenchmarkDotNet</AssemblyName>
     <PackageId>BenchmarkDotNet</PackageId>
     <RootNamespace>BenchmarkDotNet</RootNamespace>
@@ -17,6 +17,7 @@
   </ItemGroup>
   <ItemGroup>
     <PackageReference Include="CommandLineParser" Version="2.4.3" />
+    <PackageReference Include="Gee.External.Capstone" Version="2.2.0" />
     <PackageReference Include="Iced" Version="1.17.0" />
     <PackageReference Include="Microsoft.Diagnostics.Runtime" Version="2.2.332302" />
     <PackageReference Include="Perfolizer" Version="0.2.1" />