Implement Topdown metric support

This commit adds support for the topdown metrics (fe-bound, be-bound
etc.) as seen in perf

CMakeLists.txt

@@ -170,6 +170,9 @@ set(SOURCE_FILES
     src/perf/counter/userspace/reader.cpp
 
     src/perf/counter/userspace/writer.cpp
+    src/perf/counter/topdown/reader.cpp
+
+    src/perf/counter/topdown/writer.cpp
     src/perf/counter/group/writer.cpp
 
     src/perf/sample/writer.cpp

include/lo2s/config.hpp

@@ -96,6 +96,8 @@ struct Config
     bool use_nec;
     std::chrono::microseconds nec_read_interval;
     std::chrono::milliseconds nec_check_interval;
+  //topdown
+  bool topdown;
 };
 
 const Config& config();

include/lo2s/measurement_scope.hpp

@@ -31,6 +31,7 @@ enum class MeasurementScopeType
     USERSPACE_METRIC,
     BIO,
     SYSCALL,
+    TOPDOWN,
     UNKNOWN
 };
 
@@ -71,6 +72,11 @@ struct MeasurementScope
         return { MeasurementScopeType::SYSCALL, s };
     }
 
+    static MeasurementScope topdown(ExecutionScope s)
+    {
+        return { MeasurementScopeType::SYSCALL, s };
+    }
+
     friend bool operator==(const MeasurementScope& lhs, const MeasurementScope& rhs)
     {
         return (lhs.scope == rhs.scope) && lhs.type == rhs.type;
@@ -101,6 +107,8 @@ struct MeasurementScope
             return fmt::format("block layer I/O events for {}", scope.name());
         case MeasurementScopeType::SYSCALL:
             return fmt::format("syscall events for {}", scope.name());
+        case MeasurementScopeType::TOPDOWN:
+            return fmt::format("topdown events for {}", scope.name());
         default:
             throw new std::runtime_error("Unknown ExecutionScopeType!");
         }

include/lo2s/monitor/scope_monitor.hpp

@@ -25,6 +25,7 @@
 #include <lo2s/monitor/poll_monitor.hpp>
 
 #include <lo2s/perf/counter/group/writer.hpp>
+#include <lo2s/perf/counter/topdown/writer.hpp>
 #include <lo2s/perf/counter/userspace/writer.hpp>
 
 #include <lo2s/perf/sample/writer.hpp>
@@ -74,6 +75,8 @@ class ScopeMonitor : public PollMonitor
     std::unique_ptr<perf::sample::Writer> sample_writer_;
     std::unique_ptr<perf::counter::group::Writer> group_counter_writer_;
     std::unique_ptr<perf::counter::userspace::Writer> userspace_counter_writer_;
+
+    std::unique_ptr<perf::counter::topdown::Writer> topdown_writer_;
 };
 } // namespace monitor
 } // namespace lo2s

include/lo2s/perf/counter/topdown/reader.hpp

@@ -0,0 +1,72 @@
+/*
+ * This file is part of the lo2s software.
+ * Linux OTF2 sampling
+ *
+ * Copyright (c) 2021,
+ *    Technische Universitaet Dresden, Germany
+ *
+ * lo2s is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * lo2s is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with lo2s.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <lo2s/execution_scope.hpp>
+#include <lo2s/perf/counter/counter_collection.hpp>
+#include <lo2s/trace/trace.hpp>
+
+#include <cstdint>
+
+#include <vector>
+
+#include <cstdlib>
+namespace lo2s
+{
+namespace perf
+{
+namespace counter
+{
+namespace topdown
+{
+struct TopdownEvent
+{
+    uint64_t nr;
+    uint64_t slots;
+    uint64_t retiring;
+    uint64_t bad_spec;
+    uint64_t fe_bound;
+    uint64_t be_bound;
+};
+
+template <class T>
+class Reader
+{
+public:
+    Reader(ExecutionScope scope);
+
+    void read();
+
+    int fd()
+    {
+        return timer_fd_;
+    }
+
+protected:
+    int leader_fd_;
+    std::vector<int> counter_fds_;
+    int timer_fd_;
+};
+} // namespace topdown
+} // namespace counter
+} // namespace perf
+} // namespace lo2s

include/lo2s/perf/counter/topdown/writer.hpp

@@ -0,0 +1,54 @@
+/*
+ * This file is part of the lo2s software.
+ * Linux OTF2 sampling
+ *
+ * Copyright (c) 2018,
+ *    Technische Universitaet Dresden, Germany
+ *
+ * lo2s is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * lo2s is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with lo2s.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <lo2s/perf/counter/counter_collection.hpp>
+#include <lo2s/perf/counter/metric_writer.hpp>
+#include <lo2s/perf/counter/topdown/reader.hpp>
+#include <lo2s/perf/time/converter.hpp>
+#include <lo2s/trace/trace.hpp>
+
+namespace lo2s
+{
+namespace perf
+{
+namespace counter
+{
+namespace topdown
+{
+class Writer : public Reader<Writer>
+{
+public:
+    Writer(ExecutionScope scope, trace::Trace& trace);
+
+    bool handle(TopdownEvent* ev);
+
+private:
+    time::Converter time_converter_;
+    otf2::writer::local& writer_;
+    otf2::definition::metric_instance metric_instance_;
+    otf2::event::metric metric_event_;
+};
+} // namespace topdown
+} // namespace counter
+} // namespace perf
+} // namespace lo2s

include/lo2s/trace/trace.hpp

@@ -130,6 +130,7 @@ class Trace
     otf2::definition::mapping_table merge_syscall_contexts(const std::set<int64_t>& used_syscalls);
 
     otf2::writer::local& sample_writer(const ExecutionScope& scope);
+    otf2::writer::local& topdown_writer(const ExecutionScope& scope);
     otf2::writer::local& metric_writer(const MeasurementScope& scope);
     otf2::writer::local& syscall_writer(const Cpu& cpu);
     otf2::writer::local& bio_writer(BlockDevice dev);
@@ -169,6 +170,32 @@ class Trace
         return cpuid_metric_class_;
     }
 
+    otf2::definition::metric_class topdown_metric_class()
+    {
+        if (!topdown_metric_class_)
+        {
+            topdown_metric_class_ = registry_.create<otf2::definition::metric_class>(
+                otf2::common::metric_occurence::async, otf2::common::recorder_kind::abstract);
+            topdown_metric_class_->add_member(metric_member(
+                "slots", "Number of UOPS slots", otf2::common::metric_mode::absolute_point,
+                otf2::common::type::int64, "slots"));
+            topdown_metric_class_->add_member(metric_member(
+                "retiring", "Number of UOPS retired", otf2::common::metric_mode::absolute_point,
+                otf2::common::type::int64, "slots"));
+            topdown_metric_class_->add_member(metric_member(
+                "bad_spec", "Number of UOPS lost to bad speculation",
+                otf2::common::metric_mode::absolute_point, otf2::common::type::int64, "slots"));
+            topdown_metric_class_->add_member(metric_member(
+                "fe_bound", "Number of front-end bound UOPS",
+                otf2::common::metric_mode::absolute_point, otf2::common::type::int64, "slots"));
+            topdown_metric_class_->add_member(metric_member(
+                "be_bound", "Number of back-end bound UOPS",
+                otf2::common::metric_mode::absolute_point, otf2::common::type::int64, "slots"));
+        }
+
+        return topdown_metric_class_;
+    }
+
     otf2::definition::metric_member& get_event_metric_member(perf::EventDescription event)
     {
         return registry_.emplace<otf2::definition::metric_member>(
@@ -347,6 +374,7 @@ class Trace
     otf2::definition::regions_group& syscall_regions_group_;
 
     otf2::definition::detail::weak_ref<otf2::definition::metric_class> cpuid_metric_class_;
+    otf2::definition::detail::weak_ref<otf2::definition::metric_class> topdown_metric_class_;
     std::map<std::set<Cpu>, otf2::definition::detail::weak_ref<otf2::definition::metric_class>>
         perf_group_metric_classes_;
     std::map<std::set<Cpu>, otf2::definition::detail::weak_ref<otf2::definition::metric_class>>

src/config.cpp

@@ -161,6 +161,7 @@ void parse_program_options(int argc, const char** argv)
     auto& sensors_options = parser.group("sensors options");
     auto& io_options = parser.group("I/O recording options");
     auto& nec_options = parser.group("NEC SX-Aurora Tsubasa recording options");
+    auto& topdown_options = parser.group("Topdown metrics recording options");
 
     lo2s::Config config;
 
@@ -344,6 +345,8 @@ void parse_program_options(int argc, const char** argv)
         .metavar("MSEC")
         .default_value("100");
 
+    topdown_options.toggle("topdown", "Enable recording of topdown metrics");
+
     nitro::options::arguments arguments;
     try
     {
@@ -370,6 +373,7 @@ void parse_program_options(int argc, const char** argv)
     config.use_sensors = arguments.given("sensors");
     config.use_block_io = arguments.given("block-io");
     config.use_nec = arguments.given("nec");
+    config.topdown = arguments.given("topdown");
     config.command = arguments.positionals();
 
     if (arguments.given("help"))

src/monitor/scope_monitor.cpp

@@ -72,6 +72,11 @@ ScopeMonitor::ScopeMonitor(ExecutionScope scope, MainMonitor& parent, bool enabl
         add_fd(userspace_counter_writer_->fd());
     }
 
+    if(config().topdown)
+      {
+    topdown_writer_ = std::make_unique<perf::counter::topdown::Writer>(scope, parent.trace());
+    add_fd(topdown_writer_->fd());
+      }
     // note: start() can now be called
 }
 
@@ -116,6 +121,10 @@ void ScopeMonitor::monitor(int fd)
     {
         userspace_counter_writer_->read();
     }
+    if (fd == topdown_writer_->fd())
+    {
+        topdown_writer_->read();
+    }
 }
 } // namespace monitor
 } // namespace lo2s