Crate torsh_profiler 

Performance profiling for ToRSh

This crate provides comprehensive performance profiling capabilities for the ToRSh deep learning framework, including CPU, GPU, memory, and system profiling.

Refactored Modular Structure

The profiler has been successfully refactored from a massive 9,517-line monolithic file into a clean, maintainable modular structure:

• core: Core profiling types, event management, and profiler implementation
• platforms: Platform-specific profiling (CPU, GPU, system)
• analysis: Performance analysis and optimization recommendations
• export: Export and reporting functionality with multiple format support
• distributed: Distributed profiling coordination

Usage Examples

Basic Profiling

use torsh_profiler::{start_profiling, stop_profiling, profile_scope};

// Start global profiling
start_profiling();

{
    profile_scope!("computation");
    // Your code here
}

stop_profiling();

Advanced Profiling with Metrics

use torsh_profiler::{MetricsScope, export_global_events, ExportFormat};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    {
        let mut scope = MetricsScope::new("training_step");
        scope.set_operation_count(1000);
        scope.set_flops(50000);
        scope.set_bytes_transferred(4096);
        // Training code here
    }

    // Export results
    export_global_events(ExportFormat::ChromeTrace, "profile.json")?;
    Ok(())
}

Platform-Specific Profiling

use torsh_profiler::{UnifiedProfiler, CudaProfiler, MemoryProfiler};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut profiler = UnifiedProfiler::with_auto_detection();
    profiler.start_all()?;

    // Your GPU/CPU workload

    profiler.stop_all()?;
    Ok(())
}

Re-exports

pub use core::add_global_event;

pub use core::add_global_event as add_event;

pub use core::clear_global_events;

pub use core::get_global_stats;

pub use core::global_profiler;

pub use core::profile_function_with_category;

pub use core::start_profiling;

pub use core::stop_profiling;

pub use core::MetricsScope;

pub use core::Profiler;

pub use core::ScopeGuard;

pub use export::available_format_names;

pub use export::export_chrome_trace_format;

pub use export::export_csv_format;

pub use export::export_events;

pub use export::export_global_events;

pub use export::export_json_format;

pub use export::parse_format;

pub use export::ExportFormat;

pub use prometheus::PrometheusExporter;

pub use prometheus::PrometheusExporterBuilder;

pub use grafana::Dashboard as GrafanaDashboard;

pub use grafana::DashboardTemplates;

pub use grafana::GrafanaDashboardGenerator;

pub use grafana::GridPos;

pub use grafana::Panel;

pub use grafana::Target;

pub use cloudwatch::CloudWatchConfig;

pub use cloudwatch::CloudWatchPublisher;

pub use cloudwatch::CloudWatchPublisherBuilder;

pub use cloudwatch::Dimension;

pub use cloudwatch::MetricDatum;

pub use cloudwatch::StatisticSet;

pub use cloudwatch::Unit as CloudWatchUnit;

pub use streaming::create_high_performance_streaming_engine;

pub use streaming::create_low_latency_streaming_engine;

pub use streaming::create_streaming_engine;

pub use streaming::AdaptiveBitrateConfig;

pub use streaming::AdaptiveRateController;

pub use streaming::AdjustmentReason;

pub use streaming::AdvancedFeatures;

pub use streaming::BitrateAdjustment;

pub use streaming::BufferedEvent;

pub use streaming::CompressionAlgorithm;

pub use streaming::CompressionConfig;

pub use streaming::CompressionManager;

pub use streaming::ConnectionManager;

pub use streaming::ControlMessage;

pub use streaming::EnhancedStreamingEngine;

pub use streaming::EventBuffer;

pub use streaming::EventPriority;

pub use streaming::ProtocolConfig;

pub use streaming::QualityConfig;

pub use streaming::QualityLevel;

pub use streaming::QualityMetricsThreshold;

pub use streaming::SSEConnection;

pub use streaming::StreamConnection;

pub use streaming::StreamingConfig;

pub use streaming::StreamingProtocol;

pub use streaming::StreamingStats;

pub use streaming::StreamingStatsSnapshot;

pub use streaming::TcpConnection;

pub use streaming::UdpConnection;

pub use streaming::WebSocketConnection;

pub use streaming::WebSocketMessage;

pub use alerts::create_alert_manager_with_config;

pub use alerts::get_alert_manager;

pub use alerts::AlertConfig;

pub use alerts::AlertManager;

pub use alerts::NotificationChannel;

pub use attributes::get_registry;

pub use attributes::with_profiling;

pub use attributes::AsyncProfiler;

pub use attributes::AttributeRegistry;

pub use attributes::ConditionalProfiler;

pub use attributes::ProfileAttribute;

pub use attributes::ProfiledFunction;

pub use attributes::ProfiledStruct;

pub use chrome_trace::create_chrome_event;

pub use chrome_trace::export;

pub use chrome_trace::export_to_writer;

pub use chrome_trace::phases;

pub use chrome_trace::scopes;

pub use ci_cd::CiCdConfig;

pub use ci_cd::CiCdIntegration;

pub use ci_cd::CiCdPlatform;

pub use cpu::CpuProfiler;

pub use cpu::ProfileScope;

pub use cuda::get_cuda_device_properties;

pub use cuda::get_cuda_memory_stats;

pub use cuda::CudaEvent;

pub use cuda::CudaMemoryStats;

pub use cuda::CudaProfiler;

pub use cuda::CudaSynchronizationStats;

pub use cuda::NvtxRange;

pub use custom_export::CsvColumn;

pub use custom_export::CsvFormatter;

pub use custom_export::CustomExportFormat;

pub use custom_export::CustomExporter;

pub use custom_export::ExportSchema;

pub use dashboard::alerts::create_alert_manager;

pub use dashboard::create_dashboard;

pub use dashboard::create_dashboard_with_config;

pub use dashboard::export_dashboard_html;

pub use dashboard::generate_3d_landscape;

pub use dashboard::generate_performance_heatmap;

pub use dashboard::Dashboard;

pub use dashboard::DashboardAlert;

pub use dashboard::DashboardAlertSeverity;

pub use dashboard::DashboardConfig;

pub use dashboard::DashboardData;

pub use dashboard::HeatmapCell;

pub use dashboard::MemoryMetrics;

pub use dashboard::OperationSummary;

pub use dashboard::PerformanceHeatmap;

pub use dashboard::PerformanceLandscape;

pub use dashboard::PerformanceMetrics;

pub use dashboard::PerformancePoint3D;

pub use dashboard::SystemMetrics;

pub use dashboard::VisualizationColorScheme;

pub use dashboard::VisualizationConfig;

pub use dashboard::WebSocketConfig;

pub use scirs2_integration::AdvancedProfilingConfig;

pub use scirs2_integration::BenchmarkResults;

pub use scirs2_integration::HistogramStats;

pub use scirs2_integration::MetricsSummary;

pub use scirs2_integration::PerformanceAnalysis;

pub use scirs2_integration::PerformanceTargets;

pub use scirs2_integration::SamplingStrategy;

pub use scirs2_integration::ScirS2EnhancedProfiler;

pub use scirs2_integration::ScirS2ProfilingData;

pub use scirs2_integration::ValidationLevel;

pub use instruments::create_instruments_profiler;

pub use instruments::create_instruments_profiler_with_config;

pub use instruments::export_instruments_json;

pub use instruments::get_instruments_statistics;

pub use instruments::AllocationType;

pub use instruments::EnergyComponent;

pub use instruments::InstrumentsConfig;

pub use instruments::InstrumentsExportData;

pub use instruments::InstrumentsProfiler;

pub use instruments::InstrumentsStats;

pub use instruments::SignpostInterval;

pub use macros::ProfileResult;

pub use memory::FragmentationAnalysis;

pub use memory::LeakDetectionResults;

pub use memory::MemoryBlock;

pub use memory::MemoryEvent;

pub use memory::MemoryEventType;

pub use memory::MemoryLeak;

pub use memory::MemoryProfiler;

pub use memory::MemoryStats;

pub use memory::MemoryTimeline;

pub use memory::SystemMemoryInfo;

pub use memory_optimization::create_memory_optimizer;

pub use memory_optimization::create_memory_optimizer_for_low_memory;

pub use memory_optimization::create_memory_optimizer_with_aggressive_settings;

pub use memory_optimization::AdaptivePoolManager;

pub use memory_optimization::AdvancedMemoryOptimizer;

pub use memory_optimization::MemoryOptimizationConfig;

pub use memory_optimization::MemoryOptimizationStats;

pub use memory_optimization::MemorySnapshot;

pub use memory_optimization::MemoryStrategies;

pub use memory_optimization::MemoryUsagePredictor;

pub use memory_optimization::OptimizationExportData;

pub use memory_optimization::OptimizationStatsSummary;

pub use core::metrics::CorrelationAnalysis;

pub use core::metrics::CorrelationStrength;

pub use core::metrics::CorrelationSummary;

pub use core::metrics::CorrelationType;

pub use core::metrics::MemoryCorrelation;

pub use core::metrics::OperationCorrelation;

pub use core::metrics::PerformanceCorrelation;

pub use core::metrics::TemporalCorrelation;

pub use core::events::*;

pub use core::metrics::*;

pub use export::dashboard::*;

pub use export::formats::*;

pub use export::reporting::*;

pub use platforms::cpu::*;

pub use platforms::gpu::*;

pub use platforms::system::*;

pub use analysis::ml_analysis::*;

pub use analysis::optimization::*;

pub use analysis::regression::*;

pub use distributed::profiling::*;

Modules

advanced_visualization

Advanced Visualization Export

alerts

Alert system for performance monitoring

amd

AMD Tools Integration

analysis

Performance analysis and optimization Performance analysis and optimization recommendations

attributes

Attribute-based profiling support

chrome_trace

Chrome tracing format export

ci_cd

CI/CD integration for performance profiling

cloud_providers

Cloud Provider Integrations for ToRSh Profiler

cloudwatch

AWS CloudWatch metrics integration for torsh-profiler

core

Core profiling functionality Core profiling types and utilities

cpu

CPU profiling

cross_platform

Cross-platform Profiling Support

cuda

CUDA profiling

custom_export

Custom export formats for profiling data

custom_tools

Custom tool APIs for profiler integration

dashboard

Real-time Performance Dashboard

distributed

Distributed profiling coordination Distributed profiling coordination

export

Export and reporting capabilities Export and reporting functionality

grafana

Grafana dashboard integration for torsh-profiler

instruments

Apple Instruments profiling integration

integrated_profiler

Integrated Profiler System

kubernetes

Kubernetes Operator for Cloud-Native Profiling

macros

Convenient macros for profiling operations

memory

Memory profiling

memory_optimization

Advanced Memory Optimization Features

ml_analysis

Machine Learning-based performance analysis

nsight

NVIDIA Nsight profiling integration

online_learning

Online Learning Module for Real-time Performance Analysis

optimization

Performance optimizations and overhead minimization

platforms

Platform-specific profiling implementations Platform-specific profiling implementations

power

Power profiling capabilities for energy-efficient performance monitoring

prelude

Prelude module for convenient imports

prometheus

Prometheus metrics integration for torsh-profiler

regression

Performance regression detection system

reporting

Comprehensive reporting system for performance profiling

scirs2_integration

SCIRS2 Integration for Advanced Profiling

streaming

Enhanced Real-time Streaming Capabilities

tensorboard

TensorBoard export functionality

thermal

Thermal analysis system for performance profiling

vtune

Intel VTune profiling integration

workload_characterization

Workload Characterization

Macros

benchmark_scirs2

Macro for advanced benchmarking with SciRS2

collect_scirs2_metrics

Macro for advanced metrics collection

cuda_nvtx_range

Macro for NVTX range profiling

profile_alloc

Profile memory allocation with tracking

profile_async

Profile async operations

profile_attribute

Attribute-like macro for profiling functions

profile_block

Profile a block of code with automatic naming

profile_closure

Profile a closure with optional name and category

profile_compare

Benchmark and profile comparison between different implementations

profile_cuda

Profile CUDA operations

profile_current_function

Profile the current function automatically

profile_function

profile_if

Conditionally profile based on a feature flag or condition

profile_loop

Profile loop iterations with automatic batching

profile_metrics

profile_sampled

Profile with sampling (only profile every N calls)

profile_scirs2_comprehensive

Enhanced macro for comprehensive metrics profiling

profile_scirs2_sampling

Convenient macros for SCIRS2-enhanced profiling

profile_scirs2_validated

Advanced profiling macro with validation

profile_scope

Macros for convenient scope profiling

profile_tensor_op

Profile tensor operations with automatic FLOPS counting

profile_thread_local

Profile with thread-local storage for reduced overhead

profile_with_metadata

Profile with custom metadata

profile_with_overhead

Profile with automatic overhead measurement

profiled_fn

Helper macro for creating profiled function wrappers

profiling_scope

Create a profiling scope with automatic cleanup

Structs

AnomalyAnalysis

Anomaly analysis result structure

BottleneckAnalysis

Bottleneck analysis results

BottleneckEvent

A performance bottleneck event

EfficiencyIssue

Efficiency issue

MemoryAnomaly

Memory anomaly data structure

MemoryHotspot

Memory hotspot information

OverheadStats

Overhead statistics for profiling operations

PatternAnalysis

Pattern analysis result structure

PerformanceAnomaly

Performance anomaly data structure

PerformancePattern

Performance pattern data structure

ProfileEvent

Core profiling event structure

ThreadContentionEvent

Thread contention event

UnifiedProfiler

Enhanced unified profiler combining all platform profilers with simplified API

Enums

BottleneckSeverity

Severity of a bottleneck

EfficiencyIssueType

Type of efficiency issue

Constants

VERSION

VERSION_MAJOR

VERSION_MINOR

VERSION_PATCH

Functions

analyze_global_correlations

Analyze global correlations with proper implementation

are_global_stack_traces_enabled

Check if global stack traces are enabled

create_basic_profiler

Create a basic profiler for development

create_production_profiler

Create a profiler optimized for production use

create_unified_profiler

Create a unified profiler with automatic platform detection

detect_global_anomalies

Detect global anomalies in profiling data (stub implementation)

detect_global_patterns

Detect global patterns in profiling data (stub implementation)

export_duration_histogram

Export duration histogram (stub implementation)

export_global_anomaly_analysis

Export global anomaly analysis (stub implementation)

export_global_correlation_analysis

Export global correlation analysis (stub implementation)

export_global_csv

export_global_custom

Export using a custom format

export_global_json

export_global_pattern_analysis

Export global pattern analysis (stub implementation)

export_global_tensorboard

export_global_trace

Enhanced global export functions with multiple format support

export_memory_scatter_plot

Export memory scatter plot (stub implementation)

export_operation_frequency_chart

Export operation frequency chart (stub implementation)

export_performance_trend_chart

Export performance trend chart (stub implementation)

get_global_custom_export_formats

Get available custom export format names

get_global_overhead_stats

is_global_overhead_tracking_enabled

register_global_custom_export_format

Register a custom export format globally

reset_global_overhead_stats

set_global_overhead_tracking_enabled

Enhanced overhead tracking

set_global_stack_traces_enabled

Set global stack traces enabled with enhanced functionality

Type Aliases

TorshResult

Convenience type alias for Results in this crate