pytorch · briancoutinho · Sep 18, 2025 · Sep 19, 2025 · Sep 19, 2025 · Sep 24, 2025
diff --git a/libkineto/include/ActivityType.h b/libkineto/include/ActivityType.h
@@ -41,6 +41,7 @@ enum class ActivityType {
   HPU_OP, // HPU host side runtime event
   XPU_RUNTIME, // host side xpu runtime events
   COLLECTIVE_COMM, // collective communication
+  GPU_PM_COUNTER, // GPU performance monitoring counter
 
   // PRIVATEUSE1 Activity types are used for custom backends.
   // The corresponding device type is `DeviceType::PrivateUse1` in PyTorch.

diff --git a/libkineto/include/KinetoDynamicPluginInterface.h b/libkineto/include/KinetoDynamicPluginInterface.h
diff --git a/libkineto/libkineto_defs.bzl b/libkineto/libkineto_defs.bzl
@@ -70,6 +70,9 @@ def get_libkineto_cpu_only_srcs(with_api = True):
         "src/init.cpp",
         "src/output_csv.cpp",
         "src/output_json.cpp",
+        "src/dynamic_plugin/PluginLoader.h",
+        "src/dynamic_plugin/PluginProfiler.h",
+        "src/dynamic_plugin/PluginTraceBuilder.h",
     ] + (get_libkineto_api_srcs() if with_api else [])
 
 def get_libkineto_public_headers():
@@ -80,6 +83,7 @@ def get_libkineto_public_headers():
         "include/ActivityType.h",
         "include/Config.h",
         "include/ClientInterface.h",
+        "include/KinetoDynamicPluginInterface.h",
         "include/GenericTraceActivity.h",
         "include/IActivityProfiler.h",
         "include/ILoggerObserver.h",

diff --git a/libkineto/src/ActivityType.cpp b/libkineto/src/ActivityType.cpp
@@ -39,6 +39,7 @@ static constexpr std::array<ActivityTypeName, activityTypeCount + 1> map{
      {"hpu_op", ActivityType::HPU_OP},
      {"xpu_runtime", ActivityType::XPU_RUNTIME},
      {"collective_comm", ActivityType::COLLECTIVE_COMM},
+     {"gpu_pm_counter", ActivityType::GPU_PM_COUNTER},
      {"privateuse1_runtime", ActivityType::PRIVATEUSE1_RUNTIME},
      {"privateuse1_driver", ActivityType::PRIVATEUSE1_DRIVER},
      {"ENUM_COUNT", ActivityType::ENUM_COUNT}}};

diff --git a/libkineto/src/CuptiActivityProfiler.cpp b/libkineto/src/CuptiActivityProfiler.cpp
@@ -1494,7 +1494,7 @@ void CuptiActivityProfiler::finalizeTrace(
   if (!process_name.empty()) {
     logger.handleDeviceInfo(
         {pid, pid, process_name, "CPU"}, captureWindowStartTime_);
-    if (!cpuOnly_ && use_default_device_info) {
+    if (use_default_device_info) {
       // Usually, GPU events use device id as pid (0-7).
       // In some cases, CPU sockets are numbered starting from 0.
       // In the worst case, 8 CPU sockets + 8 GPUs, so the max GPU ID is 15.

diff --git a/libkineto/src/dynamic_plugin/PluginLoader.h b/libkineto/src/dynamic_plugin/PluginLoader.h
@@ -0,0 +1,139 @@
+#pragma once
+
+#include <dlfcn.h>
+#include <filesystem>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "Logger.h"
+#include "KinetoDynamicPluginInterface.h"
+#include "PluginProfiler.h"
+
+namespace libkineto {
+
+class PluginRegistry {
+public:
+  static PluginRegistry &instance() {
+    static PluginRegistry instance;
+    return instance;
+  }
+
+  int registerPluginProfiler(const KinetoPlugin_ProfilerInterface *pProfiler) {
+    if (pProfiler == nullptr) {
+      LOG(ERROR) << "Failed to register plugin profiler of nullptr";
+
+      return -1;
+    }
+
+    // Store in raw registry
+    rawPluginProfilers_.push_back(*pProfiler);
+
+    // Pass to internal registry
+    const auto &profiler = rawPluginProfilers_.back();
+    libkineto::api().registerProfilerFactory(
+        [profiler]() -> std::unique_ptr<IActivityProfiler> {
+          return std::make_unique<PluginProfiler>(profiler);
+        });
+
+    return 0;
+  }
+
+  const KinetoPlugin_Registry toCRegistry() {
+    return KinetoPlugin_Registry{
+        .unpaddedStructSize = KINETO_PLUGIN_REGISTRY_UNPADDED_STRUCT_SIZE,
+        .pRegistryHandle =
+            reinterpret_cast<KinetoPlugin_RegistryHandle *>(this),
+        .registerProfiler = cRegisterProfiler};
+  }
+
+private:
+  PluginRegistry() = default;
+  ~PluginRegistry() = default;
+  PluginRegistry(const PluginRegistry &) = delete;
+  PluginRegistry &operator=(const PluginRegistry &) = delete;
+
+  static int
+  cRegisterProfiler(KinetoPlugin_RegistryHandle *pRegistryHandle,
+                    const KinetoPlugin_ProfilerInterface *pProfiler) {
+    auto pPluginRegistry = reinterpret_cast<PluginRegistry *>(pRegistryHandle);
+    return pPluginRegistry->registerPluginProfiler(pProfiler);
+  }
+
+  std::vector<KinetoPlugin_ProfilerInterface> rawPluginProfilers_;
+};
+
+inline void loadPlugins() {
+  const char *pPluginLibDirPathEnvVar = KINETO_PLUGIN_LIB_DIR_PATH_ENV_VARIABLE;
+
+  const char *pPluginLibDirPath = getenv(pPluginLibDirPathEnvVar);
+
+  if (pPluginLibDirPath == nullptr) {
+    LOG(VERBOSE) << "Environment variable " << pPluginLibDirPathEnvVar
+                 << " not set";
+
+    return;
+  }
+
+  if (unsetenv(pPluginLibDirPathEnvVar) == -1) {
+    LOG(ERROR) << "Failed to unset environment variable "
+               << pPluginLibDirPathEnvVar << " at unsetenv() with error "
+               << strerror(errno);
+
+    return;
+  }
+
+  std::vector<std::string> libFilePaths;
+  try {
+    for (const auto &entry :
+         std::filesystem::directory_iterator(pPluginLibDirPath)) {
+      if (entry.is_regular_file() && entry.path().extension() == ".so") {
+        libFilePaths.push_back(entry.path().string());
+      }
+    }
+  } catch (const std::filesystem::filesystem_error &e) {
+    LOG(ERROR) << "Error: " << e.what();
+
+    return;
+  }
+
+  PluginRegistry &pluginRegistry = PluginRegistry::instance();
+  KinetoPlugin_Registry cPluginRegistry = pluginRegistry.toCRegistry();
+
+  for (const auto &libFilePath : libFilePaths) {
+    // Clear error state
+    dlerror();
+
+    void *pHandle = dlopen(libFilePath.c_str(), RTLD_LAZY);
+    if (pHandle == nullptr) {
+      char *pError = dlerror();
+      LOG(WARNING) << "Failed to open " << libFilePath
+                   << " at dlopen() with error " << pError;
+      continue;
+    }
+
+    int (*pfxRegister)(const KinetoPlugin_Registry *pRegistry) =
+        reinterpret_cast<int (*)(const KinetoPlugin_Registry *pRegistry)>(
+            dlsym(pHandle, "KinetoPlugin_register"));
+
+    if (pfxRegister == nullptr) {
+      char *pError = dlerror();
+      LOG(VERBOSE) << "Failed to find symbol KinetoPlugin_register() from "
+                   << libFilePath << " at dlsym() with error " << pError;
+
+      continue;
+    }
+
+    LOG(INFO) << "Found symbol KinetoPlugin_register() from " << libFilePath;
+
+    int errorCode = pfxRegister(&cPluginRegistry);
+    if (errorCode != 0) {
+      LOG(ERROR) << "Failed to register plugin profiler from " << libFilePath
+                 << " at pfxRegister() with error " << errorCode;
+    }
+  }
+
+  return;
+}
+
+} // namespace libkineto