rebase

Author: pavelsavara

coreclr-ep-profiler-design.md

@@ -22,8 +22,6 @@
         <!-- single thread -->
         <FeaturePortableTimer Condition="'$(WasmEnableThreads)' != 'true'">false</FeaturePortableTimer>
         <FeaturePortableThreadPool Condition="'$(WasmEnableThreads)' != 'true'">false</FeaturePortableThreadPool>
-        <!-- until we implement ST event pipe for CoreCLR -->
-        <FeaturePerfTracing>false</FeaturePerfTracing>
         <ProfilingSupportedBuild>false</ProfilingSupportedBuild>
     </PropertyGroup>
 

src/coreclr/clr.featuredefines.props

@@ -24,10 +24,8 @@ if(CLR_CMAKE_TARGET_TIZEN_LINUX)
 endif()
 
 if(NOT DEFINED FEATURE_EVENT_TRACE)
-  if (NOT CLR_CMAKE_TARGET_BROWSER)
-    # To actually disable FEATURE_EVENT_TRACE, also change clr.featuredefines.props
-    set(FEATURE_EVENT_TRACE 1)
-  endif()
+  # To actually disable FEATURE_EVENT_TRACE, also change clr.featuredefines.props
+  set(FEATURE_EVENT_TRACE 1)
 endif(NOT DEFINED FEATURE_EVENT_TRACE)
 
 if(NOT DEFINED FEATURE_EVENTSOURCE_XPLAT)
@@ -37,9 +35,11 @@ if(NOT DEFINED FEATURE_EVENTSOURCE_XPLAT)
   endif()
 endif(NOT DEFINED FEATURE_EVENTSOURCE_XPLAT)
 
-if(NOT DEFINED FEATURE_PERFTRACING AND FEATURE_EVENT_TRACE)
-  set(FEATURE_PERFTRACING 1)
-endif(NOT DEFINED FEATURE_PERFTRACING AND FEATURE_EVENT_TRACE)
+if(NOT DEFINED FEATURE_PERFTRACING)
+  if(FEATURE_EVENT_TRACE)
+    set(FEATURE_PERFTRACING 1)
+  endif()
+endif(NOT DEFINED FEATURE_PERFTRACING)
 
 if(NOT DEFINED FEATURE_DBGIPC)
   if(CLR_CMAKE_TARGET_UNIX)

src/coreclr/clrfeatures.cmake

@@ -610,6 +610,8 @@ RETAIL_CONFIG_DWORD_INFO(INTERNAL_EventPipeCircularMB, W("EventPipeCircularMB"),
 RETAIL_CONFIG_DWORD_INFO(INTERNAL_EventPipeProcNumbers, W("EventPipeProcNumbers"), 0, "Enable/disable capturing processor numbers in EventPipe event headers")
 RETAIL_CONFIG_DWORD_INFO(INTERNAL_EventPipeOutputStreaming, W("EventPipeOutputStreaming"), 1, "Enable/disable streaming for trace file set in DOTNET_EventPipeOutputPath.  Non-zero values enable streaming.")
 RETAIL_CONFIG_DWORD_INFO(INTERNAL_EventPipeEnableStackwalk, W("EventPipeEnableStackwalk"), 1, "Set to 0 to disable collecting stacks for EventPipe events.")
+RETAIL_CONFIG_STRING_INFO(INTERNAL_WasmPerformanceInstrumentation, W("WasmPerformanceInstrumentation"), "Configuration for WASM performance instrumentation profiler.")
+RETAIL_CONFIG_DWORD_INFO(INTERNAL_WasmPerformanceInstrumentationInterval, W("WasmPerformanceInstrumentationInterval"), 10, "Desired sample interval in milliseconds for WASM performance instrumentation profiler. 0 means sample every samplepoint.")
 
 #ifdef FEATURE_AUTO_TRACE
 RETAIL_CONFIG_DWORD_INFO_EX(INTERNAL_AutoTrace_N_Tracers, W("AutoTrace_N_Tracers"), 0, "", CLRConfig::LookupOptions::ParseIntegerAsBase10)

src/coreclr/inc/clrconfigvalues.h

@@ -121,13 +121,20 @@ enum EtwGCSettingFlags
 #define ETWFireEvent(EventName) FireEtw##EventName(GetClrInstanceId())
 
 #define ETW_TRACING_INITIALIZED(RegHandle) (TRUE)
+#if defined(HOST_BROWSER) || defined(HOST_WASI)
+#define ETW_EVENT_ENABLED(Context, EventDescriptor) (EventPipeHelper::IsEnabled(Context, EventDescriptor.Level, EventDescriptor.Keyword))
+#define ETW_CATEGORY_ENABLED(Context, Level, Keyword) (EventPipeHelper::IsEnabled(Context, Level, Keyword))
+#define ETW_TRACING_ENABLED(Context, EventDescriptor) (EventEnabled##EventDescriptor())
+#define ETW_TRACING_CATEGORY_ENABLED(Context, Level, Keyword) (EventPipeHelper::IsEnabled(Context, Level, Keyword))
+#else // HOST_BROWSER || HOST_WASI
 #define ETW_EVENT_ENABLED(Context, EventDescriptor) (EventPipeHelper::IsEnabled(Context, EventDescriptor.Level, EventDescriptor.Keyword) || \
         (XplatEventLogger::IsKeywordEnabled(Context, EventDescriptor.Level, EventDescriptor.Keyword)))
 #define ETW_CATEGORY_ENABLED(Context, Level, Keyword) (EventPipeHelper::IsEnabled(Context, Level, Keyword) || \
         (XplatEventLogger::IsKeywordEnabled(Context, Level, Keyword)))
 #define ETW_TRACING_ENABLED(Context, EventDescriptor) (EventEnabled##EventDescriptor())
 #define ETW_TRACING_CATEGORY_ENABLED(Context, Level, Keyword) (EventPipeHelper::IsEnabled(Context, Level, Keyword) || \
         (XplatEventLogger::IsKeywordEnabled(Context, Level, Keyword)))
+#endif // HOST_BROWSER || HOST_WASI
 #define ETW_PROVIDER_ENABLED(ProviderSymbol) (TRUE)
 #else //defined(FEATURE_PERFTRACING)
 #define ETW_INLINE
@@ -416,7 +423,7 @@ class XplatEventLoggerConfiguration
 };
 #endif // defined(FEATURE_PERFTRACING) || defined(FEATURE_EVENTSOURCE_XPLAT)
 
-#if defined(HOST_UNIX) && (defined(FEATURE_EVENT_TRACE) || defined(FEATURE_EVENTSOURCE_XPLAT))
+#if defined(HOST_UNIX) && !defined(HOST_BROWSER) && !defined(HOST_WASI) && (defined(FEATURE_EVENT_TRACE) || defined(FEATURE_EVENTSOURCE_XPLAT))
 
 class XplatEventLoggerController
 {
@@ -557,7 +564,7 @@ class XplatEventLogger
 };
 
 
-#endif  // defined(HOST_UNIX) && (defined(FEATURE_EVENT_TRACE) || defined(FEATURE_EVENTSOURCE_XPLAT))
+#endif  // defined(HOST_UNIX) && !defined(HOST_BROWSER) && !defined(HOST_WASI) && (defined(FEATURE_EVENT_TRACE) || defined(FEATURE_EVENTSOURCE_XPLAT))
 
 #if defined(FEATURE_EVENT_TRACE)
 

src/coreclr/inc/eventtracebase.h

@@ -45,6 +45,9 @@ static const char *g_stackTypeString[] = { "I4", "I8", "R4", "R8", "O ", "VT", "
 
 const char* CorInfoHelperToName(CorInfoHelpFunc helper);
 
+bool InterpCompiler::s_samplingProfilerEnabled = false;
+bool InterpCompiler::s_browserProfilerEnabled = false;
+
 #if MEASURE_MEM_ALLOC
 #include <minipal/mutex.h>
 
@@ -2104,6 +2107,12 @@ InterpCompiler::InterpCompiler(COMP_HANDLE compHnd,
     DWORD jitFlagsSize = m_compHnd->getJitFlags(&m_corJitFlags, sizeof(m_corJitFlags));
     assert(jitFlagsSize == sizeof(m_corJitFlags));
 
+    m_emitSamplingProfiler = s_samplingProfilerEnabled
+        && InterpConfig.WasmPerformanceInstrumentation().contains(compHnd, m_methodHnd, m_classHnd, &m_methodInfo->args);
+
+    m_emitBrowserProfiler = s_browserProfilerEnabled
+        && InterpConfig.WasmPerformanceInstrumentation().contains(compHnd, m_methodHnd, m_classHnd, &m_methodInfo->args);
+
 #ifdef DEBUG
     m_methodName = ::PrintMethodName(compHnd, m_classHnd, m_methodHnd, &m_methodInfo->args,
                             /* includeAssembly */ false,
@@ -2809,7 +2818,11 @@ void InterpCompiler::EmitBranch(InterpOpcode opcode, int32_t ilOffset)
 
     // Backwards branch, emit safepoint
     if (ilOffset < 0)
+    {
         AddIns(INTOP_SAFEPOINT);
+        if (m_emitSamplingProfiler)
+            AddIns(INTOP_PROF_SAMPLEPOINT);
+    }
 
     InterpBasicBlock *pTargetBB = m_ppOffsetToBB[target];
     if (pTargetBB == NULL)
@@ -5648,6 +5661,9 @@ void InterpCompiler::EmitRet(CORINFO_METHOD_INFO* methodInfo)
         return;
     }
 
+    if (m_emitBrowserProfiler)
+        AddIns(INTOP_PROF_LEAVE);
+
     if (m_methodInfo->args.isAsyncCall())
     {
         // We're doing a standard return. Set the continuation return to NULL.
@@ -8107,6 +8123,13 @@ void InterpCompiler::GenerateCode(CORINFO_METHOD_INFO* methodInfo)
     // Safepoint at each method entry. This could be done as part of a call, rather than
     // adding an opcode.
     AddIns(INTOP_SAFEPOINT);
+    if (m_emitSamplingProfiler)
+        AddIns(INTOP_PROF_SAMPLEPOINT);
+    if (m_emitBrowserProfiler)
+    {
+        AddIns(INTOP_PROF_ENTER);
+        m_pLastNewIns->data[0] = GetMethodDataItemIndex(m_methodHnd);
+    }
 
     if (m_continuationArgIndex != -1)
     {

src/coreclr/interpreter/compiler.cpp

@@ -627,6 +627,8 @@ class InterpCompiler
     COMP_HANDLE m_compHnd;
     CORINFO_METHOD_INFO* m_methodInfo;
     CORJIT_FLAGS m_corJitFlags;
+    bool m_emitSamplingProfiler;
+    bool m_emitBrowserProfiler;
 
     void DeclarePointerIsClass(CORINFO_CLASS_HANDLE clsHnd)
     {
@@ -1106,6 +1108,9 @@ class InterpCompiler
 
     int32_t* GetCode(int32_t *pCodeSize);
 
+    static bool s_samplingProfilerEnabled;
+    static bool s_browserProfilerEnabled;
+
 #if MEASURE_MEM_ALLOC
     // Memory statistics for profiling.
     using InterpMemStats = MemStats<InterpMemKindTraits>;

src/coreclr/interpreter/compiler.h

@@ -29,6 +29,14 @@ extern "C" INTERP_API void jitStartup(ICorJitHost* jitHost)
     InterpCompiler::initMemStats();
 #endif
 
+    // Enable profiling instrumentation if DOTNET_WasmPerformanceInstrumentation is set.
+    // This must happen before any managed code is compiled so all methods get samplepoints.
+    if (!InterpConfig.WasmPerformanceInstrumentation().isEmpty())
+    {
+        InterpCompiler::s_samplingProfilerEnabled = true;
+        InterpCompiler::s_browserProfilerEnabled = true;
+    }
+
     g_interpInitialized = true;
 }
 /*****************************************************************************/

src/coreclr/interpreter/eeinterp.cpp

@@ -75,6 +75,9 @@ OPDEF(INTOP_LDLOCA, "ldloca", 3, 1, 0, InterpOpInt)
 OPDEF(INTOP_SWITCH, "switch", 0, 0, 1, InterpOpSwitch)
 
 OPDEF(INTOP_SAFEPOINT, "safepoint", 1, 0, 0, InterpOpNoArgs)
+OPDEF(INTOP_PROF_SAMPLEPOINT, "prof.samplepoint", 1, 0, 0, InterpOpNoArgs)
+OPDEF(INTOP_PROF_ENTER, "prof.enter", 2, 0, 0, InterpOpMethodHandle)
+OPDEF(INTOP_PROF_LEAVE, "prof.leave", 1, 0, 0, InterpOpNoArgs)
 OPDEF(INTOP_BR, "br", 2, 0, 0, InterpOpBranch)
 
 OPDEF(INTOP_BRFALSE_I4, "brfalse.i4", 3, 0, 1, InterpOpBranch)

src/coreclr/interpreter/inc/intops.def

@@ -36,6 +36,7 @@ RELEASE_CONFIG_INTEGER(InterpMode, "InterpMode", 0); // Interpreter mode, one of
 // 3: use interpreter for everything, the full interpreter-only mode, no fallbacks to R2R or JIT whatsoever. Implies DOTNET_ReadyToRun=0, DOTNET_EnableHWIntrinsic=0
 
 RELEASE_CONFIG_INTEGER(DisplayMemStats, "JitMemStats", 0); // Display interpreter memory usage statistics (0=off, 1=summary, 2=detailed per-method)
+RELEASE_CONFIG_METHODSET(WasmPerformanceInstrumentation, "WasmPerformanceInstrumentation") // Method filter for WASM performance instrumentation profiler. Uses standard MethodSet pattern format.
 
 #undef CONFIG_STRING
 #undef RELEASE_CONFIG_STRING