Struct ComputePipeline 

pub struct ComputePipeline<T: Pod> { /* private fields */ }

GPU compute pipeline for chaining operations.

This struct provides a high-level API for building and executing GPU compute pipelines that chain multiple operations together without intermediate CPU transfers.

Implementations

impl<T: Pod + Zeroable> ComputePipeline<T>

pub fn new( context: &GpuContext, input: GpuBuffer<T>, width: u32, height: u32, ) -> GpuResult<Self>

Create a new compute pipeline from a GPU buffer.

pub fn from_data( context: &GpuContext, data: &[T], width: u32, height: u32, ) -> GpuResult<Self>

Create a pipeline from data.

pub fn buffer(&self) -> &GpuBuffer<T>

Get the current buffer.

pub fn dimensions(&self) -> (u32, u32)

Get the current dimensions.

pub fn element_wise( self, op: ElementWiseOp, other: &GpuBuffer<T>, ) -> GpuResult<Self>

Apply element-wise operation with another buffer.

pub fn unary(self, op: UnaryOp) -> GpuResult<Self>

Apply unary operation.

pub fn scalar(self, op: ScalarOp) -> GpuResult<Self>

Apply scalar operation.

pub fn gaussian_blur(self, sigma: f32) -> GpuResult<Self>

Apply Gaussian blur.

pub fn resize( self, new_width: u32, new_height: u32, method: ResamplingMethod, ) -> GpuResult<Self>

Resize the raster.

pub fn add(self, value: f32) -> GpuResult<Self>

Add a constant value.

pub fn multiply(self, value: f32) -> GpuResult<Self>

Multiply by a constant value.

pub fn clamp(self, min: f32, max: f32) -> GpuResult<Self>

Clamp values to a range.

pub fn threshold( self, threshold: f32, above: f32, below: f32, ) -> GpuResult<Self>

Apply threshold.

pub fn abs(self) -> GpuResult<Self>

Apply absolute value.

pub fn sqrt(self) -> GpuResult<Self>

Apply square root.

pub fn log(self) -> GpuResult<Self>

Apply natural logarithm.

pub fn exp(self) -> GpuResult<Self>

Apply exponential.

pub async fn statistics(&self) -> GpuResult<Statistics>

Compute statistics on current buffer.

This method converts the buffer to f32 internally for statistics computation, supporting all GPU data types.

pub async fn statistics_with_conversion( &self, src_type: GpuDataType, params: &ConversionParams, ) -> GpuResult<Statistics>

Compute statistics on current buffer with explicit type conversion.

Use this method when you know the source data type for optimal conversion.

pub async fn histogram( &self, num_bins: u32, min_value: f32, max_value: f32, ) -> GpuResult<Vec<u32>>

Compute histogram on current buffer.

pub async fn reduce(&self, op: ReductionOp) -> GpuResult<T>

where T: Copy,

Compute reduction (sum, min, max, etc.).

pub fn finish(self) -> GpuBuffer<T>

Get the result buffer (consumes the pipeline).

pub async fn read(self) -> GpuResult<Vec<T>>

Read the result to CPU memory asynchronously.

pub fn read_blocking(self) -> GpuResult<Vec<T>>

Read the result to CPU memory synchronously.

pub fn convert_to_f32( self, src_type: GpuDataType, params: &ConversionParams, ) -> GpuResult<ComputePipeline<f32>>

Convert the current buffer to f32 with specified conversion parameters.

This creates a new pipeline with f32 data type.

Errors

Returns an error if conversion fails.

pub fn linear_transform(self, scale: f32, offset: f32) -> GpuResult<Self>

Apply linear transformation: output = input * scale + offset.

This is a convenience method for common scaling operations.

pub fn normalize_range( self, current_min: f32, current_max: f32, new_min: f32, new_max: f32, ) -> GpuResult<Self>

Normalize values to a specific range.

Maps the current value range to [new_min, new_max].

impl ComputePipeline<f32>

Specialized implementation for f32 pipelines with full conversion support.

pub fn convert_to_type<U: Pod + Zeroable>( self, dst_type: GpuDataType, params: &ConversionParams, ) -> GpuResult<ComputePipeline<U>>

Convert f32 buffer to another data type.

Errors

Returns an error if conversion fails.

pub fn from_u8_normalized( context: &GpuContext, data: &[u8], width: u32, height: u32, ) -> GpuResult<Self>

Create a pipeline from u8 data with automatic normalization to [0, 1].

Errors

Returns an error if buffer creation fails.

pub fn from_u16_normalized( context: &GpuContext, data: &[u16], width: u32, height: u32, ) -> GpuResult<Self>

Create a pipeline from u16 data with automatic normalization to [0, 1].

Errors

Returns an error if buffer creation fails.

pub fn scale_offset(self, scale: f32, offset: f32) -> GpuResult<Self>

Apply scale and offset transformation optimized for type conversion.

This uses GPU compute for efficient transformation.