Redesign time correlation in oneprof, improve DPC++ kernel demangling

Author: anton-v-gorshkov

VERSION

@@ -1 +1 @@
-0.42.1
+0.42.4

build_utils/CMakeLists.txt

@@ -166,6 +166,7 @@ macro(FindIGALibrary TARGET)
         "IGA is not found. "
         "You may need to install Intel(R) Processor Graphics Driver to fix this issue.")
     else()
+      list(GET IGA_DLL_PATH 0 IGA_DLL_PATH)
       message(STATUS
         "IGA is found at ${IGA_DLL_PATH}")
       find_library(IGA_LIB_PATH

loader/loader.cc

@@ -63,12 +63,10 @@ static bool CheckBitness(HANDLE parent, HANDLE child) {
 int main(int argc, char* argv[]) {
   std::string library_file_name = GetLibFileName();
   std::string executable_path = utils::GetExecutablePath();
-  std::string library_file_path = executable_path + library_file_name;
 
+  std::string library_file_path = executable_path + library_file_name;
   if (!IsFileExists(library_file_path.c_str())) {
-    std::cout << "[ERROR] Failed to find " <<
-      library_file_name << " near the loader" << std::endl;
-    return 0;
+    library_file_path = library_file_name;
   }
 
   SharedLibrary* lib = SharedLibrary::Create(library_file_path);

tests/run.py

@@ -35,6 +35,7 @@
          ["sysmon", "-p", "-l", "-d"],
          ["onetrace",
           "-c", "-h", "-d", "-v", "-t", "-s",
+          "--demangle",
           "--kernels-per-tile",
           "--chrome-call-logging",
           "--chrome-device-timeline",
@@ -44,6 +45,7 @@
           "cl", "ze", "omp"],
          ["cl_tracer",
           "-c", "-h", "-d", "-v", "-t", "-s",
+          "--demangle",
           "--chrome-call-logging",
           "--chrome-device-timeline",
           "--chrome-kernel-timeline",
@@ -52,6 +54,7 @@
           "gpu", "dpc", "omp"],
          ["ze_tracer",
           "-c", "-h", "-d", "-v", "-t", "-s",
+          "--demangle",
           "--kernels-per-tile",
           "--chrome-call-logging",
           "--chrome-device-timeline",

tests/tools/cl_tracer.py

@@ -51,7 +51,7 @@ def run(path, option):
     if option == "gpu":
       command = [file_name_prefix + "cl_tracer" + file_extention,\
         "-h", "-d", "-t", app_file, "gpu", "1024", "1"]
-    elif option == "-v" or option == "--conditional-collection":
+    elif option == "-v" or option == "--demangle" or option == "--conditional-collection":
       command = [file_name_prefix + "cl_tracer" + file_extention,\
         "-d", option, app_file, "cpu", "1024", "1"]
     else:
@@ -102,6 +102,8 @@ def main(option):
     option = "-s"
   if len(sys.argv) > 1 and sys.argv[1] == "-v":
     option = "-v"
+  if len(sys.argv) > 1 and sys.argv[1] == "--demangle":
+    option = "--demangle"
   if len(sys.argv) > 1 and sys.argv[1] == "--chrome-call-logging":
     option = "--chrome-call-logging"
   if len(sys.argv) > 1 and sys.argv[1] == "--chrome-device-timeline":

tests/tools/onetrace.py

@@ -35,7 +35,8 @@ def run(path, option):
     app_folder = utils.get_sample_executable_path("omp_gemm")
     app_file = os.path.join(app_folder, "omp_gemm")
     command = ["./onetrace", "-h", "-d", "-t", app_file, "gpu", "1024", "1"]
-  elif option == "-v" or option == "--kernels-per-tile" or option == "--conditional-collection":
+  elif option == "-v" or option == "--demangle" or\
+       option == "--kernels-per-tile" or option == "--conditional-collection":
     app_folder = utils.get_sample_executable_path("dpc_gemm")
     app_file = os.path.join(app_folder, "dpc_gemm")
     command = ["./onetrace", "-d", option, app_file, "gpu", "1024", "1"]
@@ -88,6 +89,8 @@ def main(option):
     option = "-s"
   if len(sys.argv) > 1 and sys.argv[1] == "-v":
     option = "-v"
+  if len(sys.argv) > 1 and sys.argv[1] == "--demangle":
+    option = "--demangle"
   if len(sys.argv) > 1 and sys.argv[1] == "--kernels-per-tile":
     option = "--kernels-per-tile"
   if len(sys.argv) > 1 and sys.argv[1] == "--chrome-call-logging":

tests/tools/ze_tracer.py

@@ -30,7 +30,8 @@ def run(path, option):
     app_folder = utils.get_sample_executable_path("omp_gemm")
     app_file = os.path.join(app_folder, "omp_gemm")
     command = ["./ze_tracer", "-h", "-d", "-t", app_file, "gpu", "1024", "1"]
-  elif option == "-v" or option == "--kernels-per-tile" or option == "--conditional-collection":
+  elif option == "-v" or option == "--demangle" or\
+       option == "--kernels-per-tile" or option == "--conditional-collection":
     app_folder = utils.get_sample_executable_path("ze_gemm")
     app_file = os.path.join(app_folder, "ze_gemm")
     command = ["./ze_tracer", "-d", option, app_file, "1024", "1"]
@@ -81,6 +82,8 @@ def main(option):
     option = "-s"
   if len(sys.argv) > 1 and sys.argv[1] == "-v":
     option = "-v"
+  if len(sys.argv) > 1 and sys.argv[1] == "--demangle":
+    option = "--demangle"
   if len(sys.argv) > 1 and sys.argv[1] == "--kernels-per-tile":
     option = "--kernels-per-tile"
   if len(sys.argv) > 1 and sys.argv[1] == "--chrome-call-logging":

tools/cl_tracer/README.md

@@ -16,6 +16,7 @@ Options:
 --chrome-kernel-timeline       Dump device activities to JSON file per kernel name
 --chrome-device-stages         Dump device activities by stages to JSON file
 --verbose [-v]                 Enable verbose mode to show more kernel information
+--demangle                     Demangle DPC++ kernel names
 --tid                          Print thread ID into host API trace
 --pid                          Print process ID into host API and device activity trace
 --output [-o] <filename>       Print console logs into the file

tools/cl_tracer/cl_api_callbacks.h

@@ -2931,7 +2931,8 @@ static void clEnqueueTaskOnEnter(
   stream << " commandQueue = " << *(params->commandQueue);
   stream << " kernel = " << *(params->kernel);
   if (*(params->kernel) != nullptr) {
-    std::string kernel_name = utils::cl::GetKernelName(*(params->kernel));
+    std::string kernel_name = utils::cl::GetKernelName(
+        *(params->kernel), collector->Demangle());
     if (!kernel_name.empty()) {
       stream << " (" << kernel_name << ")";
     }
@@ -3582,7 +3583,8 @@ static void clEnqueueNDRangeKernelOnEnter(
   stream << " commandQueue = " << *(params->commandQueue);
   stream << " kernel = " << *(params->kernel);
   if (*(params->kernel) != nullptr) {
-    std::string kernel_name = utils::cl::GetKernelName(*(params->kernel));
+    std::string kernel_name = utils::cl::GetKernelName(
+        *(params->kernel), collector->Demangle());
     if (!kernel_name.empty()) {
       stream << " (" << kernel_name << ")";
     }
@@ -5966,6 +5968,9 @@ static void clCreateKernelOnEnter(
     stream << " kernelName = \"\"";
   } else {
     stream << " kernelName = \"" << *(params->kernelName) << "\"";
+    if (collector->Demangle()) {
+      stream << " (" << utils::Demangle(*(params->kernelName)) << ")";
+    }
   }
   stream << " errcodeRet = " << *(params->errcodeRet);
   stream << std::endl;

tools/cl_tracer/cl_api_collector.h

@@ -60,7 +60,7 @@ class ClApiCollector {
   static ClApiCollector* Create(
       cl_device_id device,
       Correlator* correlator,
-      ApiCollectorOptions options = {false, false, false},
+      ApiCollectorOptions options,
       OnClFunctionFinishCallback callback = nullptr,
       void* callback_data = nullptr) {
     PTI_ASSERT(device != nullptr);
@@ -115,6 +115,10 @@ class ClApiCollector {
     return options_.need_pid;
   }
 
+  bool Demangle() const {
+    return options_.demangle;
+  }
+
   void Log(const std::string& text) {
     PTI_ASSERT(correlator_ != nullptr);
     correlator_->Log(text);
@@ -332,7 +336,7 @@ class ClApiCollector {
   ClApiTracer* tracer_ = nullptr;
 
   Correlator* correlator_ = nullptr;
-  ApiCollectorOptions options_ = {false, false, false};
+  ApiCollectorOptions options_;
   cl_device_type device_type_ = CL_DEVICE_TYPE_ALL;
 
   OnClFunctionFinishCallback callback_ = nullptr;