Merge branch 'main' into pytest-v2

cmake/BuildSettings.cmake

@@ -86,7 +86,7 @@ endif()
 #
 add_flag_if_avail(
     "-W" "-Wall" "-Wno-unknown-pragmas" "-Wno-unused-function" "-Wno-ignored-attributes"
-    "-Wno-attributes" "-Wno-missing-field-initializers")
+    "-Wno-attributes" "-Wno-missing-field-initializers" "-Wno-interference-size")
 
 if(OMNITRACE_BUILD_DEBUG)
     add_flag_if_avail("-g3" "-fno-omit-frame-pointer")

cmake/Formatting.cmake

@@ -64,6 +64,8 @@ if(OMNITRACE_CLANG_FORMAT_EXE
     file(GLOB_RECURSE examples ${PROJECT_SOURCE_DIR}/examples/*.cpp
          ${PROJECT_SOURCE_DIR}/examples/*.c ${PROJECT_SOURCE_DIR}/examples/*.hpp
          ${PROJECT_SOURCE_DIR}/examples/*.h)
+    file(GLOB_RECURSE tests_source ${PROJECT_SOURCE_DIR}/tests/source/*.cpp
+         ${PROJECT_SOURCE_DIR}/tests/source/*.hpp)
     file(GLOB_RECURSE external ${PROJECT_SOURCE_DIR}/examples/lulesh/external/kokkos/*)
     file(
         GLOB_RECURSE
@@ -86,6 +88,7 @@ if(OMNITRACE_CLANG_FORMAT_EXE
         add_custom_target(
             format-omnitrace-source
             ${OMNITRACE_CLANG_FORMAT_EXE} -i ${sources} ${headers} ${examples}
+            ${tests_source}
             COMMENT "[omnitrace] Running C++ formatter ${OMNITRACE_CLANG_FORMAT_EXE}...")
     endif()
 

examples/fork/fork.cpp

@@ -5,6 +5,8 @@
 #include <cstdio>
 #include <cstdlib>
 #include <cstring>
+#include <pthread.h>
+#include <set>
 #include <string>
 #include <sys/wait.h>
 #include <thread>
@@ -24,71 +26,111 @@ print_info(const char* _name)
 int
 run(const char* _name, int nchildren)
 {
-    auto _threads = std::vector<std::thread>{};
+    auto _barrier  = pthread_barrier_t{};
+    auto _threads  = std::vector<std::thread>{};
+    auto _children = std::vector<pid_t>{};
+    _children.resize(nchildren, 0);
+    pthread_barrier_init(&_barrier, nullptr, nchildren + 1);
     for(int i = 0; i < nchildren; ++i)
     {
         omnitrace_user_push_region("launch_child");
-        auto _run = [i, _name]() {
-            pid_t _pid = fork();
-            if(_pid == 0)
+        auto _run = [&_barrier, &_children, i, _name](uint64_t _nsec) {
+            pthread_barrier_wait(&_barrier);
+            _children.at(i) = fork();
+            if(_children.at(i) == 0)
             {
                 // child code
                 print_info(_name);
+                printf("[%s][%i] child job starting...\n", _name, getpid());
                 auto _sleep = [=]() {
-                    std::this_thread::sleep_for(std::chrono::seconds{ i + 1 });
+                    omnitrace_user_push_region("child_process_child_thread");
+                    std::this_thread::sleep_for(std::chrono::seconds{ _nsec });
+                    omnitrace_user_pop_region("child_process_child_thread");
                 };
+                omnitrace_user_push_region("child_process");
                 std::thread{ _sleep }.join();
+                omnitrace_user_push_region("child_process");
+                printf("[%s][%i] child job complete\n", _name, getpid());
                 exit(EXIT_SUCCESS);
             }
+            else
+            {
+                pthread_barrier_wait(&_barrier);
+            }
         };
-        _threads.emplace_back(_run);
+        _threads.emplace_back(_run, i + 1);
         omnitrace_user_pop_region("launch_child");
     }
 
+    // all child threads should start executing their fork once this returns
+    pthread_barrier_wait(&_barrier);
+    // wait for the threads to successfully fork
+    pthread_barrier_wait(&_barrier);
+
     omnitrace_user_push_region("wait_for_children");
 
     int   _status   = 0;
     pid_t _wait_pid = 0;
     // parent waits for all the child processes
-    while((_wait_pid = wait(&_status)) > 0)
+    for(auto& itr : _children)
     {
-        printf("[%s][%i] returned from wait with pid = %i :: ", _name, getpid(),
-               _wait_pid);
-        if(WIFEXITED(_status))
-        {
-            printf("exited, status=%d\n", WEXITSTATUS(_status));
-        }
-        else if(WIFSIGNALED(_status))
-        {
-            printf("killed by signal %d\n", WTERMSIG(_status));
-        }
-        else if(WIFSTOPPED(_status))
+        while(itr == 0)
+        {}
+        printf("[%s][%i] performing waitpid(%i, ...)\n", _name, getpid(), itr);
+        while((_wait_pid = waitpid(itr, &_status, WUNTRACED | WNOHANG)) <= 0)
         {
-            printf("stopped by signal %d\n", WSTOPSIG(_status));
-        }
-        else if(WIFCONTINUED(_status))
-        {
-            printf("continued\n");
-        }
-        else
-        {
-            printf("unknown\n");
+            if(_wait_pid == 0) continue;
+
+            printf("[%s][%i] returned from waitpid(%i) with pid = %i (status = %i) :: ",
+                   _name, getpid(), itr, _wait_pid, _status);
+            if(WIFEXITED(_status))
+            {
+                printf("exited, status=%d\n", WEXITSTATUS(_status));
+            }
+            else if(WIFSIGNALED(_status))
+            {
+                printf("killed by signal %d\n", WTERMSIG(_status));
+            }
+            else if(WIFSTOPPED(_status))
+            {
+                printf("stopped by signal %d\n", WSTOPSIG(_status));
+            }
+            else if(WIFCONTINUED(_status))
+            {
+                printf("continued\n");
+            }
+            else
+            {
+                printf("unknown\n");
+            }
         }
     }
 
+    printf("[%s][%i] joining threads ...\n", _name, getpid());
     for(auto& itr : _threads)
         itr.join();
 
     omnitrace_user_pop_region("wait_for_children");
+
+    printf("[%s][%i] returning (error code: %i) ...\n", _name, getpid(), _status);
     return _status;
 }
 
 int
 main(int argc, char** argv)
 {
-    int _n = 4;
-    if(argc > 1) _n = std::stoi(argv[1]);
+    int _nfork = 4;
+    int _nrep  = 1;
+    if(argc > 1) _nfork = std::stoi(argv[1]);
+    if(argc > 2) _nrep = std::stoi(argv[2]);
 
     print_info(argv[0]);
-    return run(argv[0], _n);
+    for(int i = 0; i < _nrep; ++i)
+    {
+        auto _ec = run(argv[0], _nfork);
+        if(_ec != 0) return _ec;
+    }
+
+    printf("[%s][%i] job complete\n", argv[0], getpid());
+    return EXIT_SUCCESS;
 }

source/bin/omnitrace-sample/impl.cpp

@@ -140,12 +140,6 @@ get_initial_environment()
     auto _mode = get_env<std::string>("OMNITRACE_MODE", "sampling", false);
 
     update_env(_env, "OMNITRACE_USE_SAMPLING", (_mode != "causal"));
-    update_env(_env, "OMNITRACE_CRITICAL_TRACE", false);
-    update_env(_env, "OMNITRACE_USE_PROCESS_SAMPLING", false);
-
-    // update_env(_env, "OMNITRACE_USE_PID", false);
-    // update_env(_env, "OMNITRACE_TIME_OUTPUT", false);
-    // update_env(_env, "OMNITRACE_OUTPUT_PATH", "omnitrace-output/%tag%/%launch_time%");
 
 #if defined(OMNITRACE_USE_ROCTRACER) || defined(OMNITRACE_USE_ROCPROFILER)
     update_env(_env, "HSA_TOOLS_LIB", _dl_libpath);

source/bin/tests/CMakeLists.txt

@@ -1,6 +1,7 @@
 set(OMNITRACE_ABORT_FAIL_REGEX
-    "### ERROR ###|address of faulting memory reference|exiting with non-zero exit code|terminate called after throwing an instance|calling abort.. in |Exit code: [1-9]"
-    CACHE INTERNAL "Regex to catch abnormal exits when a PASS_REGULAR_EXPRESSION is set")
+    "### ERROR ###|unknown-hash=|address of faulting memory reference|exiting with non-zero exit code|terminate called after throwing an instance|calling abort.. in |Exit code: [1-9]"
+    CACHE INTERNAL "Regex to catch abnormal exits when a PASS_REGULAR_EXPRESSION is set"
+          FORCE)
 
 # adds a ctest for executable
 function(OMNITRACE_ADD_BIN_TEST)

source/lib/common/defines.h.in

@@ -46,6 +46,17 @@
 #define OMNITRACE_HIP_VERSION_MAJOR @OMNITRACE_HIP_VERSION_MAJOR@
 #define OMNITRACE_HIP_VERSION_MINOR @OMNITRACE_HIP_VERSION_MINOR@
 #define OMNITRACE_HIP_VERSION_PATCH @OMNITRACE_HIP_VERSION_PATCH@
+
+// these can be set via defining the variable in CMake, e.g.:
+//      cmake -D OMNITRACE_CACHELINE_SIZE=N /path/to/source
+// if not defined when configuring cmake, these values fall back to
+// default values set in core/containers/aligned_static_vector.hpp.
+// the OMNITRACE_CACHELINE_SIZE_MIN is used to ensure portability
+#cmakedefine OMNITRACE_CACHELINE_SIZE @OMNITRACE_CACHELINE_SIZE@
+#cmakedefine OMNITRACE_CACHELINE_SIZE_MIN @OMNITRACE_CACHELINE_SIZE_MIN@
+
+// misc definitions which can be configured by cmake to override the defaults
+#cmakedefine OMNITRACE_ROCM_MAX_COUNTERS @OMNITRACE_ROCM_MAX_COUNTERS@
 // clang-format on
 
 #define OMNITRACE_VERSION                                                                \
@@ -87,16 +98,22 @@
 #endif
 // clang-format on
 
-#if !defined(OMNITRACE_MAX_COUNTERS)
-#    define OMNITRACE_MAX_COUNTERS 25
-#endif
-
-#if !defined(OMNITRACE_ROCM_LOOK_AHEAD)
-#    define OMNITRACE_ROCM_LOOK_AHEAD 128
+// in general, we want to make sure the cache line size is not less than
+// 64 bytes (most common cacheline size for x86-64 CPUs) so unless
+// OMNITRACE_CACHELINE_SIZE was explicitly set, we set the min to 64
+// and use the max value of OMNITRACE_CACHELINE_SIZE and
+// OMNITRACE_CACHELINE_SIZE_MIN to assure that false-sharing is well
+// guarded against
+#if !defined(OMNITRACE_CACHELINE_SIZE_MIN)
+#    if defined(OMNITRACE_CACHELINE_SIZE)
+#        define OMNITRACE_CACHELINE_SIZE_MIN OMNITRACE_CACHELINE_SIZE
+#    else
+#        define OMNITRACE_CACHELINE_SIZE_MIN 64
+#    endif
 #endif
 
-#if !defined(OMNITRACE_MAX_ROCM_QUEUES)
-#    define OMNITRACE_MAX_ROCM_QUEUES OMNITRACE_MAX_THREADS
+#if !defined(OMNITRACE_ROCM_MAX_COUNTERS)
+#    define OMNITRACE_ROCM_MAX_COUNTERS 25
 #endif
 
 #define OMNITRACE_ATTRIBUTE(...)    __attribute__((__VA_ARGS__))

source/lib/core/binary/address_range.cpp

@@ -183,7 +183,7 @@ address_range::operator+=(address_range _v)
 hash_value_t
 address_range::hash() const
 {
-    return (is_range()) ? tim::get_combined_hash_id(hash_value_t{ low }, high)
+    return (is_range()) ? tim::get_hash_id(hash_value_t{ low }, high)
                         : hash_value_t{ low };
 }
 }  // namespace binary

source/lib/core/concepts.hpp

@@ -52,6 +52,13 @@ using tim::identity_t;  // NOLINT
 template <typename Tp>
 struct use_placement_new_when_generating_unique_ptr : std::false_type
 {};
+
+template <typename Tp, typename... Args>
+auto
+make_unique(Args&&... args)
+{
+    return unique_ptr_t<Tp>{ new Tp{ std::forward<Args>(args)... } };
+}
 }  // namespace omnitrace
 
 namespace tim

source/lib/core/config.cpp

@@ -293,7 +293,7 @@ configure_settings(bool _init)
     OMNITRACE_CONFIG_SETTING(
         bool, "OMNITRACE_USE_ROCM_SMI",
         "Enable sampling GPU power, temp, utilization, and memory usage", true, "backend",
-        "rocm_smi", "rocm");
+        "rocm_smi", "rocm", "process_sampling");
 
     OMNITRACE_CONFIG_SETTING(
         bool, "OMNITRACE_USE_ROCTX",
@@ -1154,7 +1154,6 @@ configure_mode_settings(const std::shared_ptr<settings>& _config)
     {
         set_default_setting_value("OMNITRACE_USE_SAMPLING", true);
         set_default_setting_value("OMNITRACE_USE_PROCESS_SAMPLING", true);
-        _set("OMNITRACE_CRITICAL_TRACE", false);
     }
 
     if(gpu::device_count() == 0)

source/lib/core/containers/CMakeLists.txt

@@ -1,7 +1,11 @@
 #
 set(containers_sources)
 
-set(containers_headers ${CMAKE_CURRENT_LIST_DIR}/stable_vector.hpp
-                       ${CMAKE_CURRENT_LIST_DIR}/static_vector.hpp)
+set(containers_headers
+    ${CMAKE_CURRENT_LIST_DIR}/aligned_static_vector.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/c_array.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/operators.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/stable_vector.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/static_vector.hpp)
 
 target_sources(omnitrace-core-library PRIVATE ${containers_sources} ${containers_headers})