tenzr::profiler::Profiler¶

Process-wide collector of profile events with Chrome Trace export.

#include <trace.h>

Public Functions¶

Public Functions Documentation¶

function Instance¶

static Profiler & Instance()

Return:

• The process-wide profiler instance.
• Reference to the singleton Profiler.

Get the process-wide profiler.

function Start¶

void Start()

Begin recording events.

Enable recording and emit the default process/thread metadata events.

function Stop¶

void Stop()

Stop recording events.

Disable recording; existing events are retained.

function Reset¶

void Reset()

Clear all recorded events and reset the clock.

Clear recorded events and restart the trace clock.

function IsEnabled¶

inline bool IsEnabled() const

Return: true while recording is enabled.

function RecordEvent¶

void RecordEvent(
    TraceEvent event
)

Parameters:

• event Event to record.
• event Event to record.

Append a pre-built trace event. Append a raw trace event.

function RecordInstant¶

void RecordInstant(
    std::string name,
    std::string cat ="general",
    char scope ='t'
)

Parameters:

• name Event name.
• cat Event category.
• scope Instant scope ('g', 'p', or 't').
• name Event name.
• cat Event category.
• scope Instant scope (g, p, or t).

Record a zero-duration instant event. Record an instantaneous event.

function RecordCounter¶

void RecordCounter(
    std::string name,
    int64_t value,
    std::string cat ="memory"
)

Parameters:

• name Counter name.
• value Counter value.
• cat Event category.
• name Counter name.
• value Counter value.
• cat Event category.

Record a counter value. Record a numeric counter sample.

function RecordMetadata¶

void RecordMetadata(
    std::string name,
    std::string value
)

Parameters:

• name Metadata key.
• value Metadata value.
• name Metadata key.
• value Metadata value.

Record a metadata name/value pair. Record a metadata event.

function RecordStreamKernel¶

void RecordStreamKernel(
    std::string name,
    int64_t stream_id,
    int64_t dur_us,
    std::string cat ="gpu_kernel",
    std::unordered_map< std::string, std::string > args ={}
)

Parameters:

• name Kernel name.
• stream_id Stream identifier.
• dur_us Duration in microseconds.
• cat Event category.
• args Additional event arguments.
• name Kernel name.
• stream_id Source stream.
• dur_us Kernel duration in microseconds.
• cat Event category.
• args Additional event arguments.

Record a device-kernel execution. Record a device kernel duration on a stream.

function RecordMemoryTransfer¶

void RecordMemoryTransfer(
    std::string name,
    size_t bytes,
    bool h2d,
    int64_t dur_us
)

Parameters:

• name Transfer name.
• bytes Number of bytes transferred.
• h2d true for host-to-device, false for device-to-host.
• dur_us Duration in microseconds.
• name Transfer name.
• bytes Number of bytes moved.
• h2d Direction: true for host-to-device.
• dur_us Transfer duration in microseconds.

Record a host/device memory transfer. Record a host-device transfer, including derived bandwidth when the duration is known.

function RecordCollectiveComm¶

void RecordCollectiveComm(
    std::string name,
    size_t bytes,
    int64_t dur_us
)

Parameters:

• name Operation name.
• bytes Number of bytes communicated.
• dur_us Duration in microseconds.
• name Collective name.
• bytes Number of bytes communicated.
• dur_us Duration in microseconds.

Record a collective-communication operation. Record a collective communication operation.

function ExportChromeTrace¶

bool ExportChromeTrace(
    const std::string & path
) const

Parameters:

• path Output file path.
• path Destination file path.

Return:

• true on success.
• true on success.

Write recorded events as a Chrome Trace JSON file. Write recorded events to a Chrome/Perfetto trace JSON file.

function AnalyzeTrace¶

ProfileSummary AnalyzeTrace() const

Return:

• Aggregated statistics for the recorded events.
• The computed profile summary.

Aggregate recorded events into a duration summary.

function EventCount¶

size_t EventCount() const

Return:

• Number of recorded events.
• Event count.

Number of recorded events.