Struct GpuEngine 

pub struct GpuEngine {
    pub device: Device,
    pub queue: Queue,
    /* private fields */
}

GPU compute engine for aggregations

Fields

device: Device

GPU device handle (public for benchmarking)

queue: Queue

GPU command queue (public for benchmarking)

Implementations

impl GpuEngine

pub async fn new() -> Result<Self>

Initialize GPU engine

Errors

Returns error if GPU initialization fails (no GPU available, driver issues, etc.)

pub async fn sum_i32(&self, data: &Int32Array) -> Result<i32>

Execute SUM aggregation on GPU

Arguments

• data - Input array (Int32 or Float32)

Returns

Sum of all elements

Errors

Returns error if GPU execution fails

pub async fn sum_f32(&self, data: &Float32Array) -> Result<f32>

Execute SUM aggregation on GPU (f32)

Errors

Returns error if GPU execution fails

pub async fn count(&self, data: &dyn Array) -> Result<usize>

Execute COUNT aggregation on GPU

Errors

Returns error if GPU execution fails

pub async fn min_i32(&self, data: &Int32Array) -> Result<i32>

Execute MIN aggregation on GPU

Errors

Returns error if GPU execution fails

pub async fn max_i32(&self, data: &Int32Array) -> Result<i32>

Execute MAX aggregation on GPU

Errors

Returns error if GPU execution fails

pub async fn avg_f32(&self, data: &Float32Array) -> Result<f32>

Execute AVG aggregation on GPU (reuses sum + count)

Errors

Returns error if GPU execution fails

pub async fn fused_filter_sum( &self, data: &Int32Array, filter_threshold: i32, filter_op: &str, ) -> Result<i32>

Execute fused filter+sum aggregation on GPU (JIT-compiled kernel)

Toyota Way: Muda elimination - fuses filter and sum in single pass, eliminating intermediate buffer write.

Arguments

• data - Input array (Int32)
• filter_threshold - Filter threshold value (e.g., WHERE value > 1000)
• filter_op - Filter operator (“gt”, “lt”, “eq”, “gte”, “lte”, “ne”)

Returns

Sum of filtered elements

Errors

Returns error if GPU execution fails

Example

// Equivalent to: SELECT SUM(value) FROM data WHERE value > 1000
let result = engine.fused_filter_sum(&data, 1000, "gt").await?;