Struct SparseProfiler 

pub struct SparseProfiler {
    pub config: BenchmarkConfig,
    pub measurements: Vec<PerformanceMeasurement>,
    pub operation_counters: HashMap<String, usize>,
}

Comprehensive performance profiler for sparse tensor operations

The SparseProfiler is the main class for benchmarking sparse tensor operations. It provides methods for profiling format conversions, matrix operations, and memory usage patterns.

Fields

config: BenchmarkConfig

Configuration for benchmark execution

measurements: Vec<PerformanceMeasurement>

Collected performance measurements

operation_counters: HashMap<String, usize>

Operation counters for tracking

Implementations

impl SparseProfiler

pub fn new(config: BenchmarkConfig) -> Self

Create a new sparse profiler with the given configuration

pub fn benchmark_format_conversion( &mut self, dense_matrix: &Tensor, ) -> TorshResult<Vec<PerformanceMeasurement>>

Benchmark sparse format conversion operations

This method profiles the conversion between different sparse formats (COO, CSR, etc.) and measures the performance characteristics.

Arguments

• dense_matrix - Input dense matrix to convert to sparse formats

Returns

Vector of performance measurements for each format conversion

pub fn benchmark_sparse_matmul( &mut self, lhs: &dyn SparseTensor, rhs: &dyn SparseTensor, ) -> TorshResult<Vec<PerformanceMeasurement>>

Benchmark sparse matrix multiplication operations

This method profiles matrix multiplication between sparse matrices in different formats and measures performance characteristics.

Arguments

• lhs - Left-hand side sparse matrix
• rhs - Right-hand side sparse matrix

Returns

Vector of performance measurements for matrix multiplication

pub fn benchmark_dense_to_sparse( &mut self, dense_matrix: &Tensor, sparsity_threshold: f32, ) -> TorshResult<Vec<PerformanceMeasurement>>

Benchmark dense to sparse conversion operations

This method profiles the conversion from dense matrices to sparse formats with different sparsity patterns and thresholds.

Arguments

• dense_matrix - Input dense matrix
• sparsity_threshold - Threshold for considering elements as zero

Returns

Vector of performance measurements for dense-to-sparse conversion

pub fn profile_format_comparison( &mut self, dense_matrix: &Tensor, ) -> TorshResult<Vec<PerformanceMeasurement>>

Profile format comparison for a given dense matrix

This method converts a dense matrix to all supported sparse formats and compares their performance characteristics.

pub fn clear_measurements(&mut self)

Clear all collected measurements and reset counters

pub fn measurement_count(&self) -> usize

Get the total number of measurements collected

pub fn get_measurements_for_operation( &self, operation: &str, ) -> Vec<&PerformanceMeasurement>

Get measurements for a specific operation

Trait Implementations

impl Debug for SparseProfiler

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for SparseProfiler

fn default() -> Self

Returns the “default value” for a type.