[XPTI][INFRA] Sample E2E data collection timing test for XPTI (#13045)

XPTI has unit tests that time the cost of each individual framework
action, but an E2E timing test isn't available. This PR adds a new
sample collector that shows how data can be pulled from the SYCL runtime
using XPTI and provides timing information for the callback handler
costs/event.

Allows:
 1. Zero cost application with XPTI_TRACE_ENABLE=0
 2. Zero cost callback handlers when run in calibration mode
3. Full E2E test when run with "--format none" which gives the average
cost of callback handlers/event

---------

Signed-off-by: Vasanth Tovinkere <[email protected]>

Author: tovinkere

sycl/doc/design/SYCLInstrumentationUsingXPTI.md

@@ -6,6 +6,7 @@ project (xptifw VERSION "${XPTI_VERSION}" LANGUAGES CXX)
 set(CMAKE_CXX_STANDARD 17)
 
 set(XPTIFW_DIR ${CMAKE_CURRENT_LIST_DIR})
+set(SAMPLES_DIR ${CMAKE_CURRENT_LIST_DIR}/samples)
 
 # The XPTI framework requires the includes from
 # the proxy implementation of XPTI
@@ -68,6 +69,7 @@ endif()
 if (XPTI_BUILD_SAMPLES)
   add_subdirectory(samples/basic_collector)
   add_subdirectory(samples/syclpi_collector)
+  add_subdirectory(samples/sycl_perf_collector)
   add_subdirectory(basic_test)
 endif()
 

xptifw/CMakeLists.txt

@@ -1,6 +1,7 @@
 file(GLOB SOURCES *.cpp *.hpp)
 include_directories(${XPTIFW_DIR}/include)
 include_directories(${XPTI_DIR}/include)
+include_directories(${SAMPLES_DIR}/include)
 
 add_executable(XPTIFWBasicTests ${SOURCES})
 target_link_libraries(XPTIFWBasicTests PRIVATE xptifw ${CMAKE_DL_LIBS})

xptifw/basic_test/CMakeLists.txt

@@ -8,6 +8,7 @@
 #pragma once
 
 #include "xpti/xpti_trace_framework.hpp"
+#include "xpti_helpers.hpp"
 
 #include <chrono>
 #include <iomanip>
@@ -26,12 +27,6 @@ namespace test {
 namespace utils {
 enum class OptionType { Boolean, Integer, Float, String, Range };
 
-// We are using C++ 11, hence we cannot use
-// std::variant or std::any
-using table_row_t = std::map<int, long double>;
-using table_t = std::map<int, table_row_t>;
-using titles_t = std::vector<std::string>;
-
 class ScopedTimer {
 public:
   using time_unit_t =
@@ -301,56 +296,6 @@ class CommandLineParser {
   std::string MAppName;
 };
 
-class TableModel {
-public:
-  using row_titles_t = std::map<int, std::string>;
-
-  TableModel() {}
-
-  void setHeaders(titles_t &Titles) { MColumnTitles = Titles; }
-
-  table_row_t &addRow(int Row, std::string &RowName) {
-    if (MRowTitles.count(Row)) {
-      std::cout << "Warning: Row title already specified!\n";
-    }
-    MRowTitles[Row] = RowName;
-    return MTable[Row];
-  }
-
-  table_row_t &addRow(int Row, const char *RowName) {
-    if (MRowTitles.count(Row)) {
-      std::cout << "Warning: Row title already specified!\n";
-    }
-    MRowTitles[Row] = RowName;
-    return MTable[Row];
-  }
-
-  table_row_t &operator[](int Row) { return MTable[Row]; }
-
-  void print() {
-    std::cout << std::setw(14) << " ";
-    for (auto &Title : MColumnTitles) {
-      std::cout << std::setw(14) << Title; // Column headers
-    }
-    std::cout << "\n";
-
-    for (auto &Row : MTable) {
-      std::cout << std::setw(14) << MRowTitles[Row.first];
-      for (auto &Data : Row.second) {
-        std::cout << std::fixed << std::setw(14) << std::setprecision(0)
-                  << Data.second;
-      }
-      std::cout << "\n";
-    }
-    std::cout << "\n";
-  }
-
-private:
-  titles_t MColumnTitles;
-  row_titles_t MRowTitles;
-  table_t MTable;
-};
-
 class RangeDecoder {
 public:
   RangeDecoder(std::string &RangeStr) : MRange(RangeStr) {
@@ -436,16 +381,16 @@ class TestCorrectness {
 private:
   void runStringTableTests();
   void runStringTableTestThreads(int RunNo, int NThreads,
-                                 test::utils::TableModel &Table);
+                                 xpti::utils::TableModel &Table);
   void runTracepointTests();
   void runTracepointTestThreads(int RunNo, int nt,
-                                test::utils::TableModel &Table);
+                                xpti::utils::TableModel &Table);
   void runNotificationTests();
   void runNotificationTestThreads(int RunNo, int NThreads,
-                                  test::utils::TableModel &Table);
+                                  xpti::utils::TableModel &Table);
 
   test::utils::CommandLineParser &MParser;
-  test::utils::TableModel MTable;
+  xpti::utils::TableModel MTable;
   std::set<long> MThreads, MTests;
   long MTracepoints;
   const char *MSource = "foo.cpp";
@@ -598,13 +543,13 @@ class TestPerformance {
 private:
   void runDataStructureTests();
   void runDataStructureTestsThreads(int RunNo, int NThreads,
-                                    test::utils::TableModel &Table);
+                                    xpti::utils::TableModel &Table);
   void runInstrumentationTests();
   void runInstrumentationTestsThreads(int RunNo, int NThreads,
-                                      test::utils::TableModel &Table);
+                                      xpti::utils::TableModel &Table);
 
   test::utils::CommandLineParser &MParser;
-  test::utils::TableModel MTable;
+  xpti::utils::TableModel MTable;
   std::set<long> MThreads, MTests;
   long MTracepoints;
   long MTracepointInstances;

xptifw/basic_test/cl_processor.hpp

@@ -54,7 +54,7 @@ enum class FWColumns {
 };
 
 void TestPerformance::runDataStructureTestsThreads(
-    int RunNo, int NumThreads, test::utils::TableModel &Model) {
+    int RunNo, int NumThreads, xpti::utils::TableModel &Model) {
   xptiReset();
   uint64_t TimeInNS;
   double ElapsedTime;
@@ -402,9 +402,9 @@ void TestPerformance::runDataStructureTestsThreads(
 }
 
 void TestPerformance::runDataStructureTests() {
-  test::utils::TableModel Model;
+  xpti::utils::TableModel Model;
 
-  test::utils::titles_t Columns{"Threads",      "Str.Insert", "Str.Lookup",
+  xpti::utils::titles_t Columns{"Threads",      "Str.Insert", "Str.Lookup",
                                 "St.Ins/Lu",    "TP Create",  "TP Un-Cached",
                                 "TP FW-Cached", "TP Local",   "Notify"};
   std::cout << std::setw(Columns.size() * 15 / 2)
@@ -426,7 +426,7 @@ void TestPerformance::runDataStructureTests() {
 }
 
 void TestPerformance::runInstrumentationTestsThreads(
-    int RunNo, int NumThreads, test::utils::TableModel &Model) {
+    int RunNo, int NumThreads, xpti::utils::TableModel &Model) {
   xptiReset();
   uint64_t TimeInNS;
   double ElapsedTime;
@@ -535,9 +535,9 @@ void TestPerformance::runInstrumentationTestsThreads(
 }
 
 void TestPerformance::runInstrumentationTests() {
-  test::utils::TableModel Model;
+  xpti::utils::TableModel Model;
 
-  test::utils::titles_t Columns{
+  xpti::utils::titles_t Columns{
       "Threads",     "TP LU+Notify(ns)", "TP Create(ns)", "Ev/s,cb=10",
       "Ev/s,cb=100", "Ev/s,cb=500",      "Ev/s,cb=1000",  "Ev/s,cb=2000"};
   std::cout << std::setw(Columns.size() * 15 / 2) << "Framework Tests\n";

xptifw/basic_test/performance_tests.cpp

@@ -88,7 +88,7 @@ enum class TPColumns {
 enum class NColumns { Threads, Notifications, PassRate };
 
 void TestCorrectness::runStringTableTestThreads(
-    int RunNo, int NumThreads, test::utils::TableModel &Model) {
+    int RunNo, int NumThreads, xpti::utils::TableModel &Model) {
   xptiReset();
   constexpr int NumStrings = 1000;
 
@@ -184,9 +184,9 @@ void TestCorrectness::runStringTableTestThreads(
 }
 
 void TestCorrectness::runStringTableTests() {
-  test::utils::TableModel Model;
+  xpti::utils::TableModel Model;
 
-  test::utils::titles_t Columns{"Threads", "Insert", "Lookup", "Duplicate",
+  xpti::utils::titles_t Columns{"Threads", "Insert", "Lookup", "Duplicate",
                                 "Pass rate"};
   std::cout << std::setw(25) << "String Table Tests\n";
   Model.setHeaders(Columns);
@@ -207,7 +207,7 @@ void TestCorrectness::runStringTableTests() {
 }
 
 void TestCorrectness::runTracepointTestThreads(int RunNo, int NumThreads,
-                                               test::utils::TableModel &Model) {
+                                               xpti::utils::TableModel &Model) {
   xptiReset();
   constexpr int TracepointCount = 1000;
 
@@ -348,9 +348,9 @@ void TestCorrectness::runTracepointTestThreads(int RunNo, int NumThreads,
 }
 
 void TestCorrectness::runTracepointTests() {
-  test::utils::TableModel Model;
+  xpti::utils::TableModel Model;
 
-  test::utils::titles_t Columns{"Threads",   "Create",  "Lookup",
+  xpti::utils::titles_t Columns{"Threads",   "Create",  "Lookup",
                                 "Duplicate", "Payload", "Pass rate"};
   std::cout << std::setw(25) << "Tracepoint Tests\n";
   Model.setHeaders(Columns);
@@ -371,7 +371,7 @@ void TestCorrectness::runTracepointTests() {
 }
 
 void TestCorrectness::runNotificationTestThreads(
-    int RunNo, int NumThreads, test::utils::TableModel &Model) {
+    int RunNo, int NumThreads, xpti::utils::TableModel &Model) {
   xptiReset();
   int TPCount = 30, CallbackCount = TPCount * 30;
   std::vector<xpti::payload_t *> Payloads;
@@ -502,9 +502,9 @@ void TestCorrectness::runNotificationTestThreads(
 }
 
 void TestCorrectness::runNotificationTests() {
-  test::utils::TableModel Model;
+  xpti::utils::TableModel Model;
 
-  test::utils::titles_t Columns{"Threads", "Notify", "Pass rate"};
+  xpti::utils::titles_t Columns{"Threads", "Notify", "Pass rate"};
   std::cout << std::setw(25) << "Notification Tests\n";
   Model.setHeaders(Columns);
 

xptifw/basic_test/semantic_tests.cpp

@@ -19,7 +19,6 @@
 
 static uint8_t GStreamID = 0;
 std::mutex GIOMutex;
-xpti::ThreadID GThreadIDEnum;
 
 // The lone callback function we are going to use to demonstrate how to attach
 // the collector to the running executable
@@ -111,9 +110,10 @@ XPTI_CALLBACK_API void tpCallback(uint16_t TraceType,
                                   xpti::trace_event_data_t *Event,
                                   uint64_t Instance, const void *UserData) {
   auto Payload = xptiQueryPayload(Event);
-  xpti::timer::tick_t Time = xpti::timer::rdtsc();
-  auto TID = xpti::timer::getThreadID();
-  uint32_t CPU = GThreadIDEnum.enumID(TID);
+  xpti::utils::timer::measurement_t M;
+  uint64_t Time = M.clock();
+  auto TID = M.thread();
+  uint32_t CPU = M.cpu();
   std::string Name;
 
   if (Payload->name_sid() != xpti::invalid_id) {