tenzr::profiler::BottleneckReport¶

Heuristic analysis of the dominant performance bottleneck.

#include <trace.h>

Public Functions¶

Public Attributes¶

Public Functions Documentation¶

function Format¶

std::string Format() const

Return:

• Multi-line human-readable report.
• Formatted report text.

Render the identified bottleneck and recommendations.

Public Attributes Documentation¶

variable primary_bottleneck¶

std::string primary_bottleneck;

variable compute_to_comm_ratio¶

double compute_to_comm_ratio {0.0};

variable memory_transfer_pct¶

double memory_transfer_pct {0.0};

variable collective_comm_pct¶

double collective_comm_pct {0.0};

Share of time spent on collectives.

variable total_device_kernel_us¶

int64_t total_device_kernel_us {0};

variable total_memory_transfer_us¶

int64_t total_memory_transfer_us {0};

variable total_collective_comm_us¶

int64_t total_collective_comm_us {0};

Total collective-communication time.

variable recommendations¶

std::vector< std::string > recommendations;