Struct TensorChunkConfig 

pub struct TensorChunkConfig {
    pub strategy: ChunkingStrategy,
    pub auto_tune: bool,
    pub min_chunk_size: usize,
    pub max_chunk_size: usize,
}

Tensor-specific chunking configuration

Provides high-level configuration for tensor operations with automatic parameter selection based on hardware capabilities.

Fields

strategy: ChunkingStrategy

Chunking strategy

auto_tune: bool

Enable automatic tuning based on runtime profiling

min_chunk_size: usize

Minimum chunk size (prevents over-chunking for small tensors)

max_chunk_size: usize

Maximum chunk size (prevents cache thrashing)

Implementations

impl TensorChunkConfig

pub fn compute_intensive() -> Self

Create a compute-intensive configuration

Optimized for:

• Matrix multiplication (GEMM operations)
• Convolution operations
• FFT transformations

Performance

Expected 15-30% speedup over naive chunking through:

• CPU core utilization optimization
• Instruction-level parallelism
• Reduced synchronization overhead

pub fn memory_intensive() -> Self

Create a memory-intensive configuration

Optimized for:

• Large tensor copies
• Broadcasting operations
• Tensor reshaping

Performance

Expected 20-40% speedup through:

• Memory bandwidth optimization
• NUMA-aware memory access
• Prefetching optimization

pub fn cache_friendly() -> Self

Create a cache-friendly configuration

Optimized for:

• Reduction operations (sum, mean, max)
• Cumulative operations (cumsum, cumprod)
• Scan operations

Performance

Expected 25-50% speedup through:

• L2/L3 cache size awareness
• Cache line alignment
• Reduced cache misses

pub fn custom( chunk_size: usize, alignment: usize, prefetch_distance: usize, auto_tune: bool, ) -> Self

Create a custom configuration

pub fn compute_chunk_size( &self, tensor_size: usize, element_size: usize, ) -> usize

Apply this configuration to compute optimal chunk size for given tensor size

Arguments

• tensor_size - Total number of elements in the tensor
• element_size - Size of each element in bytes (e.g., 4 for f32)

Returns

Optimal chunk size in elements

Trait Implementations

impl Clone for TensorChunkConfig

fn clone(&self) -> TensorChunkConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TensorChunkConfig

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

Formats the value using the given formatter.