Merge pull request #1226 from SpiNNakerManchester/custom_energy_monit…

…oring

Custom energy monitoring

c_common/models/chip_power_monitor/src/chip_power_monitor.c

@@ -51,6 +51,11 @@ struct sample_params {
     uint32_t frequency;
 };
 
+struct recording {
+    uint32_t time;
+    uint32_t core_counters[NUM_CPUS];
+};
+
 //! \brief The recording channel we use.
 //!
 //! Only one recording channel is used by this application.
@@ -73,7 +78,7 @@ static uint32_t time;
 static uint32_t timer = 0;
 
 //! Where we aggregate the sample activity counts.
-static uint32_t core_counters[NUM_CPUS];
+static struct recording recording;
 //! How many samples have we done so far within this aggregate step?
 static uint32_t sample_count;
 //! The number of samples to aggregate per recording entry.
@@ -105,15 +110,16 @@ static inline uint32_t get_random_busy(void) {
 //! \brief Synchronously records the current contents of the core_counters to
 //! the recording region.
 static inline void record_aggregate_sample(void) {
+    recording.time = time;
     recording_record(
-            RECORDING_CHANNEL_ID, core_counters, sizeof(core_counters));
+            RECORDING_CHANNEL_ID, &recording, sizeof(recording));
 }
 
 //! \brief Resets the state of the core_counters and the sample_count variables
 //! to zero.
 static inline void reset_core_counters(void) {
     for (uint32_t i = 0 ; i < NUM_CPUS ; i++) {
-        core_counters[i] = 0;
+        recording.core_counters[i] = 0;
     }
     sample_count = 0;
 }
@@ -154,7 +160,7 @@ static inline void count_core_states(void) {
 
     for (uint32_t i = 0, j = 1 ; i < NUM_CPUS ; i++, j <<= 1) {
         if (!(sample & j)) {
-            core_counters[i]++;
+            recording.core_counters[i]++;
         }
     }
 }
@@ -175,10 +181,8 @@ static void sample_in_slot(UNUSED uint unused0, UNUSED uint unused1) {
 
         recording_finalise();
 
-        // Subtract 1 from the time so this tick gets done again on the next
-        // run
-        time--;
-
+        // Invert the time calculation so that any time read is correct
+        time = (time * sample_frequency) / timer;
         simulation_ready_to_read();
 
         return;

spinn_front_end_common/data/fec_data_view.py

@@ -83,6 +83,7 @@ class _FecDataModel(object):
         "_database_file_path",
         "_database_socket_addresses",
         "_ds_database_path",
+        "_energy_checkpoints",
         "_executable_targets",
         "_executable_types",
         "_first_machine_time_step",
@@ -190,6 +191,7 @@ def _soft_reset(self) -> None:
         self._first_machine_time_step = 0
         self._run_step: Optional[int] = None
         self._n_run_steps: Optional[int] = None
+        self._energy_checkpoints: List[int] = []
 
     def _clear_notification_protocol(self) -> None:
         if self._notification_protocol:
@@ -1355,3 +1357,28 @@ def iterate_live_output_devices(cls) -> Iterable[LiveOutputDevice]:
         :rtype: iterable(LiveOutputDevice)
         """
         return iter(cls.__fec_data._live_output_devices)
+
+    @classmethod
+    def add_energy_checkpoint(cls, checkpoint_ms: int):
+        """
+        Add an energy checkpoint.
+
+        :param checkpoint: The checkpoint to be added in milliseconds
+        """
+        cls.__fec_data._energy_checkpoints.append(checkpoint_ms)
+
+    @classmethod
+    def iterate_energy_checkpoints(cls) -> Iterable[int]:
+        """
+        Iterate over energy checkpoints.
+
+        :rtype: iterable(int)
+        """
+        return iter(cls.__fec_data._energy_checkpoints)
+
+    @classmethod
+    def clear_energy_checkpoints(cls) -> None:
+        """
+        Clear all energy checkpoints.
+        """
+        cls.__fec_data._energy_checkpoints.clear()

spinn_front_end_common/interface/abstract_spinnaker_base.py

@@ -557,8 +557,6 @@ def __run(self, run_time: Optional[float], sync_time: float):
                     "Only binaries that use the simulation interface can be"
                     " run more than once")
 
-        self._adjust_config(run_time)
-
         # Install the Control-C handler
         if self.__is_main_thread():
             signal.signal(signal.SIGINT, self.__signal_handler)
@@ -1756,7 +1754,7 @@ def _execute_load_application_data_specification(self) -> None:
         :rtype: dict(tuple(int,int,int),DataWritten) or DsWriteInfo
         """
         with FecTimer("Load Application data specification",
-                      TimerWork.LOADING) as timer:
+                      TimerWork.LOADING_DATA) as timer:
             if timer.skip_if_virtual_board():
                 return
             return load_application_data_specs()
@@ -1892,6 +1890,7 @@ def _do_load(self) -> None:
             self._report_memory_on_chip()
             self._report_compressed(compressed)
         self._execute_application_load_executables()
+        self._execute_router_provenance_gatherer("Load", TimerWork.LOADING)
 
         FecTimer.end_category(TimerCategory.LOADING)
 
@@ -1951,19 +1950,20 @@ def _execute_placements_provenance_gatherer(self) -> None:
                 timer.skip(str(ex))
                 return
 
-    def _execute_router_provenance_gatherer(self) -> None:
+    def _execute_router_provenance_gatherer(
+            self, prefix: str, phase: TimerWork) -> None:
         """
         Runs, times and log the RouterProvenanceGatherer if requested.
         """
         with FecTimer(
-                "Router provenance gatherer", TimerWork.EXTRACTING) as timer:
+                "Router provenance gatherer", phase) as timer:
             if timer.skip_if_cfg_false("Reports",
                                        "read_router_provenance_data"):
                 return
             if timer.skip_if_virtual_board():
                 return
             try:
-                router_provenance_gatherer()
+                router_provenance_gatherer(prefix)
             except DataNotYetAvialable as ex:
                 timer.skip(str(ex))
                 return
@@ -1991,7 +1991,6 @@ def _do_read_provenance(self) -> None:
         """
         self._execute_graph_provenance_gatherer()
         self._execute_placements_provenance_gatherer()
-        self._execute_router_provenance_gatherer()
         self._execute_profile_data_gatherer()
 
     def _report_energy(self) -> None:
@@ -2004,7 +2003,6 @@ def _report_energy(self) -> None:
             if timer.skip_if_virtual_board():
                 return
 
-            # TODO runtime is None
             power_used = compute_energy_used()
 
             energy_provenance_reporter(power_used)
@@ -2127,11 +2125,16 @@ def _do_extract_from_machine(self) -> None:
         :param run_time: the run duration in milliseconds.
         :type run_time: int or None
         """
+        self._execute_router_provenance_gatherer("Run", TimerWork.EXTRACTING)
+        for chip in FecDataView.get_machine().chips:
+            FecDataView().get_transceiver().clear_router_diagnostic_counters(
+                chip.x, chip.y)
         self._execute_extract_iobuff()
         self._execute_buffer_extractor()
         self._execute_clear_io_buf()
+        self._execute_router_provenance_gatherer(
+            "Extract", TimerWork.EXTRACTING)
 
-        # FinaliseTimingData never needed as just pushed self._ to inputs
         self._do_read_provenance()
         self._report_energy()
         self._do_provenance_reports()

spinn_front_end_common/interface/config_handler.py

@@ -72,6 +72,7 @@ def __init__(self, data_writer_cls: Optional[Type[FecDataWriter]] = None):
         self._debug_configs()
         self._previous_handler()
         self._reserve_system_vertices()
+        self._ensure_provenance_for_energy_report()
 
     def __toggle_config(self, section: str, option: str, to_false: List[str],
                         to_true: List[str]):
@@ -178,20 +179,6 @@ def _reserve_system_vertices(self):
             self._data_writer.add_sample_monitor_vertex(
                 sample_speedup_vertex(), False)
 
-    def _adjust_config(self, runtime: Optional[float]):
-        """
-        Adjust and checks the configuration based on runtime
-
-        :param runtime:
-        :type runtime: int or bool
-        :raises ConfigurationException:
-        """
-        if runtime is None:
-            if get_config_bool("Reports", "write_energy_report"):
-                set_config("Reports", "write_energy_report", "False")
-                logger.info("[Reports]write_energy_report has been set to "
-                            "False as runtime is set to forever")
-
     def _remove_excess_folders(
             self, max_kept: int, starting_directory: str,
             remove_errored_folders: Optional[bool]):
@@ -257,3 +244,8 @@ def _set_up_report_specifics(self) -> None:
             f.write("\n")
             f.write("Traceback of setup call:\n")
             traceback.print_stack(file=f)
+
+    def _ensure_provenance_for_energy_report(self):
+        if get_config_bool("Reports", "write_energy_report"):
+            set_config("Reports", "read_router_provenance_data", "True")
+            set_config("Reports", "read_placements_provenance_data", "True")