Roctracer perfetto flow fixes (#267)

* testing label updates

- automatically add "gpu", "roctracer", "rocm-smi", and "rocprofiler" test labels when appropriate

* Bump version to v1.9.1

* roctracer and config updates

- fix perfetto::Flow
  - use roctracer correlation ID instead of critical trace correlation ID
- renamed ambiguous _cid, _parent_cid, _corr_id variables to _crit_cid, _parent_crit_cid, _roct_cid
- use atomic_{mutex,lock} instead of STL mutex/lock
- support for individual perfetto annotations for HIP API args
- OMNITRACE_PERFETTO_COMPACT_ROCTRACER_ANNOTATIONS option for controlling compact vs. individual perfetto annotations for HIP API args

* Update timemory submodule

- argparser updates
  - help prints to std::cout by default now
  - supports setting custom ostream

* cmake formatting

* config::get_setting_value updates

- config::get_setting_value returns std::optional instead of std::pair<bool, Tp>

[ROCm/rocprofiler-systems commit: 279a8e0952]

projects/rocprofiler-systems/source/lib/core/config.cpp

@@ -406,11 +406,7 @@ configure_settings(bool _init)
         "durations are needed, see OMNITRACE_TRACE_PERIODS.",
         0.0, "trace", "profile", "perfetto", "timemory");
 
-    auto _clock_s =
-        config::get_setting_value<std::string>("OMNITRACE_TRACE_PERIOD_CLOCK_ID").second;
-
     auto _clock_choices = std::vector<std::string>{};
-
     for(const auto& itr : constraint::get_valid_clock_ids())
     {
         _clock_choices.emplace_back(
@@ -696,6 +692,18 @@ configure_settings(bool _init)
                              "feature may dramatically reduce the size of the trace",
                              true, "perfetto", "data", "debugging", "advanced");
 
+    OMNITRACE_CONFIG_SETTING(
+        bool, "OMNITRACE_PERFETTO_COMPACT_ROCTRACER_ANNOTATIONS",
+        "When PERFETTO_ANNOTATIONS, USE_ROCTRACER, and ROCTRACER_HIP_API are all "
+        "enabled, enabling this option will result in the arg information for HIP API "
+        "calls to all be within one annotation (e.g., args=\"stream=0x0, dst=0x1F, "
+        "sizeBytes=64, src=0x08, kind=1\"). When disabled, each parameter will be an "
+        "individual annotation (e.g. stream, dst, sizeBytes, etc.). The benefit of the "
+        "former is that it is faster to serialize and consumes less file space; the "
+        "benefit of the latter is that it becomes much easier to find slices in the "
+        "trace with the same value",
+        false, "perfetto", "data", "debugging", "roctracer", "rocm", "advanced");
+
     OMNITRACE_CONFIG_SETTING(
         uint64_t, "OMNITRACE_THREAD_POOL_SIZE",
         "Max number of threads for processing background tasks",
@@ -1095,7 +1103,7 @@ configure_mode_settings(const std::shared_ptr<settings>& _config)
         }
         else
         {
-            bool _changed = get_setting_value<bool>(_name).second != _v;
+            bool _changed = get_setting_value<bool>(_name).value_or(!_v) != _v;
             OMNITRACE_BASIC_VERBOSE(
                 1 && _changed,
                 "[configure_mode_settings] Overriding %s to %s in %s mode...\n",
@@ -1105,7 +1113,7 @@ configure_mode_settings(const std::shared_ptr<settings>& _config)
     };
 
     auto _use_causal = get_setting_value<bool>("OMNITRACE_USE_CAUSAL");
-    if(_use_causal.first && _use_causal.second) set_env("OMNITRACE_MODE", "causal", 1);
+    if(_use_causal && *_use_causal) set_env("OMNITRACE_MODE", "causal", 1);
 
     if(get_mode() == Mode::Coverage)
     {