Struct ReductionConfig 

pub struct ReductionConfig {
    pub num_slots: usize,
    pub use_cooperative: bool,
    pub use_software_barrier: bool,
    pub shared_mem_bytes: usize,
}

Configuration for reduction operations.

Fields

num_slots: usize

Number of reduction slots (for parallel accumulation).

Multiple slots reduce atomic contention by spreading updates across several memory locations. The final result is computed by combining all slots on the host.

use_cooperative: bool

Use cooperative groups for grid-wide synchronization.

Requires compute capability 6.0+ (Pascal or newer). When disabled, falls back to software barriers or multi-launch.

use_software_barrier: bool

Use software barrier when cooperative groups unavailable.

Software barriers use atomic counters in global memory. This works on all devices but has higher latency.

shared_mem_bytes: usize

Shared memory size per block for reduction (bytes).

Should be at least block_size * sizeof(T) for full reduction. Default: 0 (auto-calculate based on block size).

Implementations

impl ReductionConfig

pub fn new() -> ReductionConfig

Create a new reduction config with default settings.

pub fn with_slots(self, num_slots: usize) -> ReductionConfig

Set the number of accumulation slots.

pub fn with_cooperative(self, enabled: bool) -> ReductionConfig

Enable or disable cooperative groups.

pub fn with_software_barrier(self, enabled: bool) -> ReductionConfig

Enable or disable software barrier fallback.

pub fn with_shared_mem(self, bytes: usize) -> ReductionConfig

Set explicit shared memory size.

Trait Implementations

impl Clone for ReductionConfig

fn clone(&self) -> ReductionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ReductionConfig

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

Formats the value using the given formatter.

impl Default for ReductionConfig

fn default() -> ReductionConfig

Returns the “default value” for a type.