nRF Profiler

The nRF Profiler provides an interface for logging and visualizing data for performance measurements, while the system is running. You can use the module to profile Application Event Manager events or custom events. The output is provided using RTT and can be visualized in a custom Python backend.

See the Profiler sample for an example of how to use the nRF Profiler.

Configuration

Since Application Event Manager events are converted to nRF Profiler events, the nRF Profiler can be configured to profile custom events or Application Event Manager events, or both.

Configuring for use with custom events

To use the nRF Profiler for custom events, complete the following steps:

  1. Enable the CONFIG_NRF_PROFILER Kconfig option. This option adds the nRF Profiler source code to the application.

  2. Call nrf_profiler_init() during the application start to initialize the nRF Profiler.

  3. Profile custom events, as described in the following section.

Profiling custom events

To profile custom events, complete the following steps:

  1. Register the custom events using nrf_profiler_register_event_type(). The following code example shows how to register event types:

    static const char * const data_names[] = {"value1", "value2", "value3", "value4", "string"};
    static const enum nrf_profiler_arg data_types[] = {NRF_PROFILER_ARG_U32, NRF_PROFILER_ARG_S32,
                NRF_PROFILER_ARG_S16, NRF_PROFILER_ARG_U8,
                NRF_PROFILER_ARG_STRING};
    
    no_data_event_id = nrf_profiler_register_event_type("no_data_event", NULL,
                NULL, 0);
    data_event_id = nrf_profiler_register_event_type("data_event", data_names,
                data_types, 5);
    
  2. Add a structure for sending information about event occurrences:

    1. Add the following mandatory functions to the structure:

    2. Add one or more of the following optional functions in-between the mandatory functions, depending on the data format:

  3. Wrap the calls in one function that you then call to profile event occurrences. The following code example shows a function for profiling an event with data:

    static void profile_data_event(uint32_t val1, int32_t val2, int16_t val3,
                uint8_t val4, const char *string)
    {
      struct log_event_buf buf;
    
      nrf_profiler_log_start(&buf);
      /* Profiling data connected with an event */
      nrf_profiler_log_encode_uint32(&buf, val1);
      nrf_profiler_log_encode_int32(&buf, val2);
      nrf_profiler_log_encode_int16(&buf, val3);
      nrf_profiler_log_encode_uint8(&buf, val4);
      nrf_profiler_log_encode_string(&buf, string);
      nrf_profiler_log_send(&buf, data_event_id);
    }
    

    Note

    The data_event_id and the data that is profiled with the event must be consistent with the registered event type. The data for every data field must be provided in the correct order.

Configuration for use with Application Event Manager

If you are using the Application Event Manager, in order to use the nRF Profiler follow the steps in Implementing profiling for Application Event Manager events and Configuration on the Application Event Manager profiler tracer documentation page.

Enabling supported backend

The nRF Profiler supports a custom backend that is based around Python scripts to visualize the output data. The backend communicates with the host using RTT.

To save profiling data, the scripts use CSV files for event occurrences and JSON files for event descriptions.

Available scripts

The scripts can be found under scripts/nrf_profiler/ in the nRF Connect SDK folder structure. The following script files are available:

  • data_collector.py - This script connects to the device using RTT, receives profiling data, and saves it to files. When running the script from the command line, provide the time for collecting data (in seconds) and the dataset name. For example:

    python3 data_collector.py 5 test1
    

    In this command, 5 is the time value for collecting data and test1 is the dataset name.

  • plot_from_files.py - This script plots events from the dataset that is provided as the command-line argument. For example:

    python3 plot_from_files.py test1
    

    In this command, test1 is the dataset name.

  • real_time_plot.py - This script connects to the device using RTT, plots data in real-time, and saves the data. When running the script from the command line, provide the dataset name. For example:

    python3 real_time_plot.py test1
    
  • merge_data.py - This script combines data from test_p and test_c datasets into one dataset test_merged. It also provides clock drift compensation based on the synchronization events: sync_event_p and sync_event_c. This enables you to observe times between events for the two connected devices. As command-line arguments, provide names of events used for synchronization for a Peripheral (sync_event_p) and a Central (sync_event_c), as well as names of datasets for the Peripheral (test_p), the Central (test_c), and the merge result (test_merged). For example:

    python3 merge_data.py test_p sync_event_p test_c sync_event_c test_merged
    

Running the backend

To enable and run the custom backend, complete the following steps:

  1. Connect device to the computer.

  2. Complete the configuration steps for use with either custom events or Application Event Manager, as described in the Configuration section.

  3. From the list of Available scripts, choose the Python script that you want to use for event visualization, analysis, and calculating statistics.

  4. Run the script from the command line using its related command.

Backend visualization

When you run either the plot_from_files.py or the real_time_plot.py script, the profiled events are visualized in a GUI window.

The visual output can look like the following diagram:

Example of nRF Profiler backend script visualization

Example of nRF Profiler backend script visualization

In this diagram:

  • Red dots indicate the submissions for the related events.

  • Blue vertical rectangles indicate the processing of the events, with their width corresponding to the processing time length.

  • The Start/Stop button can be used to pause or resume the real-time plot translation.

The GUI also supports the following actions:

  • Scrolling on the diagram using the mouse wheel zooms it in or out.

    • When plotting is paused, scrolling zooms to the cursor location.

    • When plotting in running, scrolling zooms to the right edge of the plot.

  • Middle mouse button can be used to mark an event submission or processing for tracking, and to display the event data.

  • When plotting is paused, you can click and drag with the left mouse button to pan the plot.

  • You can click the left or right mouse button to place a vertical line at the cursor location. When two lines are present, the application measures the time between them and displays it.

Shell integration

The nRF Profiler is integrated with Zephyr’s Shell module. You can use the CONFIG_NRF_PROFILER_SHELL option to add an additional subcommand set (nrf_profiler) to the shell. The option is enabled by default.

This subcommand set contains the following commands:

list

Show a list of profiled event types. The letters “E” or “D” indicate if profiling is currently enabled or disabled for a given event type.

enable or disable

Enable or disable profiling. If called without additional arguments, the command applies to all event types. To enable or disable profiling for specific event types, pass the event type indexes (as displayed by list) as arguments.

API documentation

Header file: include/nrf_profiler.h
Source files: subsys/nrf_profiler/
group nrf_profiler

Profiler.

Defines

CONFIG_NRF_PROFILER_MAX_NUMBER_OF_APP_EVENTS

Maximum number of events.

CONFIG_NRF_PROFILER_NUMBER_OF_INTERNAL_EVENTS

Number of internal events.

NRF_PROFILER_MAX_NUMBER_OF_APPLICATION_AND_INTERNAL_EVENTS

Maximum number of events including user application events and internal events.

Enums

enum nrf_profiler_arg

Data types for profiling.

Values:

enumerator NRF_PROFILER_ARG_U8
enumerator NRF_PROFILER_ARG_S8
enumerator NRF_PROFILER_ARG_U16
enumerator NRF_PROFILER_ARG_S16
enumerator NRF_PROFILER_ARG_U32
enumerator NRF_PROFILER_ARG_S32
enumerator NRF_PROFILER_ARG_STRING
enumerator NRF_PROFILER_ARG_TIMESTAMP

Functions

int nrf_profiler_init(void)

Initialize the Profiler.

Note

This function is thread-safe, but not safe to use in interrupts.

Return values:

0 – If the operation was successful.

void nrf_profiler_term(void)

Terminate the Profiler.

const char *nrf_profiler_get_event_descr(size_t nrf_profiler_event_id)

Retrieve the description of an event type.

Parameters:
  • nrf_profiler_event_id – Event ID.

Returns:

Event description.

static inline bool is_profiling_enabled(size_t nrf_profiler_event_id)

Check if profiling is enabled for a given event type.

Parameters:
  • nrf_profiler_event_id – Event ID.

Returns:

Logical value indicating if the event type is currently profiled.

uint16_t nrf_profiler_register_event_type(const char *name, const char *const *args, const enum nrf_profiler_arg *arg_types, uint8_t arg_cnt)

Register an event type.

Note

This function is thread-safe, but not safe to use in interrupts.

Parameters:
  • name – Name of the event type.

  • args – Names of data values sent with the event.

  • arg_types – Types of data values sent with the event.

  • arg_cnt – Number of data values sent with the event.

Returns:

ID assigned to the event type.

void nrf_profiler_log_start(struct log_event_buf *buf)

Initialize a buffer for the data of an event.

Parameters:
  • buf – Pointer to the data buffer.

void nrf_profiler_log_encode_uint32(struct log_event_buf *buf, uint32_t data)

Encode and add uint32_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_int32(struct log_event_buf *buf, int32_t data)

Encode and add int32_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_uint16(struct log_event_buf *buf, uint16_t data)

Encode and add uint16_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_int16(struct log_event_buf *buf, int16_t data)

Encode and add int16_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_uint8(struct log_event_buf *buf, uint8_t data)

Encode and add uint8_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_int8(struct log_event_buf *buf, int8_t data)

Encode and add int8_t data type to a buffer.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • data – Data to add to the buffer.

void nrf_profiler_log_encode_string(struct log_event_buf *buf, const char *string)

Encode and add string to a buffer.

Maximum 255 characters can be sent (the rest is omitted).

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • string – String to add to the buffer.

void nrf_profiler_log_add_mem_address(struct log_event_buf *buf, const void *mem_address)

Encode and add the event’s address in memory to the buffer.

This information is used for event identification.

Note

The buffer must be initialized with nrf_profiler_log_start before calling this function.

Parameters:
  • buf – Pointer to the data buffer.

  • mem_address – Memory address to encode.

void nrf_profiler_log_send(struct log_event_buf *buf, uint16_t event_type_id)

Send data from the buffer to the host.

This function only sends data that is already stored in the buffer. Use nrf_profiler_log_encode_uint32, nrf_profiler_log_encode_int32, nrf_profiler_log_encode_uint16, nrf_profiler_log_encode_int16, nrf_profiler_log_encode_uint8, nrf_profiler_log_encode_int8, nrf_profiler_log_encode_string or nrf_profiler_log_add_mem_address to add data to the buffer.

Parameters:
  • event_type_id – Event type ID as assigned to the event type when it is registered.

  • buf – Pointer to the data buffer.

Variables

uint8_t nrf_profiler_num_events

Number of event types registered in the Profiler.

struct nrf_profiler_event_enabled_bm
#include <nrf_profiler.h>

Bitmask indicating event is enabled. This structure is private to nrf_profiler and should not be referred from outside.

struct log_event_buf
#include <nrf_profiler.h>

Buffer required for data that is sent with the event.

Public Members

uint8_t *payload

Pointer to the end of the payload.

uint8_t payload_start[CONFIG_NRF_PROFILER_CUSTOM_EVENT_BUF_LEN]

Array where the payload is located before it is sent.