Struct ReduceKernel 

pub struct ReduceKernel { /* private fields */ }

Reduction kernel for parallel reduction operations

Implementations

impl ReduceKernel

pub fn new(operation: ReduceOp) -> Self

Create a new reduction kernel

pub fn sum() -> Self

Create a sum reduction kernel

pub fn min() -> Self

Create a min reduction kernel

pub fn max() -> Self

Create a max reduction kernel

pub fn mean() -> Self

Create a mean reduction kernel

pub fn with_workgroup_size(self, size: u32) -> Self

Set the workgroup size

pub fn execute_u8(&self, _device: &GpuDevice, input: &[u8]) -> Result<Vec<u8>>

Execute the reduction operation on u8 data (CPU fallback).

Output encoding

Operation	Output format
Sum	8-byte little-endian u64
Min / Max	1 byte
Mean	4-byte little-endian f32
MinMax	2 bytes [min, max]
CountNonZero	8-byte little-endian u64
Histogram	256 × 4-byte little-endian u32 counts

Arguments

• _device - GPU device (CPU fallback: unused)
• input - Input data buffer

Errors

Returns an error only on internal logic failures (currently infallible).

pub fn execute_f32( &self, _device: &GpuDevice, input: &[f32], ) -> Result<Vec<f32>>

Execute the reduction operation on f32 data (CPU fallback).

Output encoding

Operation	Output (Vec<f32>)
Sum	[total_sum]
Min / Max	[value]
Mean	[mean]
MinMax	[min, max]
CountNonZero	[count as f32]
Histogram	empty (not meaningful for f32)

Arguments

• _device - GPU device (CPU fallback: unused)
• input - Input data buffer

Errors

Returns an error only on internal logic failures (currently infallible).

pub fn operation(&self) -> ReduceOp

Get the operation type

pub fn workgroup_size(&self) -> u32

Get the workgroup size

pub fn passes_required(&self, input_size: usize) -> u32

Calculate the number of reduction passes needed

pub fn estimate_flops(input_size: usize, operation: ReduceOp) -> u64

Estimate FLOPS for the reduction