primal-barracuda 0.4.0

barraCuda — BARrier-free Rust Abstracted Cross-platform Unified Dimensional Algebra. Vendor-agnostic GPU compute via WGSL.
Documentation
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Avg Pool 1D - Temporal average pooling - Pure WGSL
//!
//! Deep Debt Principles:
//! - Self-knowledge: Operation knows its own requirements (kernel size, stride)
//! - Zero hardcoding: All parameters passed at runtime
//! - Modern idiomatic Rust: Safe, zero unsafe code
//! - Complete implementation: Production-ready, no mocks
//! - Hardware-agnostic: Pure WGSL for universal compute

use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;

/// Avg Pool 1D operation - Temporal average pooling
///
/// Applies 1D average pooling over an input signal (batch, channels, length).
pub struct AvgPool1D {
    input: Tensor,
    kernel_size: usize,
    stride: usize,
}

impl AvgPool1D {
    /// Create a new `AvgPool1D` operation
    #[must_use]
    pub fn new(input: Tensor, kernel_size: usize, stride: usize) -> Self {
        Self {
            input,
            kernel_size,
            stride,
        }
    }

    /// Get the WGSL shader source
    fn wgsl_shader() -> &'static str {
        {
            static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
                include_str!("../shaders/pooling/avg_pool1d_f64.wgsl").to_string()
            });
            std::sync::LazyLock::force(&SHADER).as_str()
        }
    }

    /// Execute the avg pool 1D operation
    ///
    /// # Errors
    ///
    /// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
    /// readback fails (e.g. device lost or out of memory).
    pub fn execute(self) -> Result<Tensor> {
        let device = self.input.device();
        let shape = self.input.shape();

        // Validate input shape (batch, channels, length)
        if shape.len() != 3 {
            return Err(BarracudaError::invalid_shape(
                vec![0, 0, 0], // Expected: 3D tensor
                shape.to_vec(),
            ));
        }

        let batch_size = shape[0];
        let channels = shape[1];
        let input_size = shape[2];

        // Calculate output size
        let output_size = (input_size - self.kernel_size) / self.stride + 1;
        let total_elements = batch_size * channels * output_size;

        // Create buffers
        // Access input buffer directly (zero-copy)
        let input_buffer = self.input.buffer();

        let output_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
            label: Some("AvgPool1D Output"),
            size: (total_elements * std::mem::size_of::<f32>()) as u64,
            usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
            mapped_at_creation: false,
        });

        // Create uniform buffer for parameters
        #[repr(C)]
        #[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
        struct Params {
            input_size: u32,
            output_size: u32,
            channels: u32,
            batch_size: u32,
            kernel_size: u32,
            stride: u32,
        }

        let params = Params {
            input_size: input_size as u32,
            output_size: output_size as u32,
            channels: channels as u32,
            batch_size: batch_size as u32,
            kernel_size: self.kernel_size as u32,
            stride: self.stride as u32,
        };

        let params_buffer = device
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("AvgPool1D Params"),
                contents: bytemuck::cast_slice(&[params]),
                usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
            });

        // Create bind group layout
        let bind_group_layout =
            device
                .device
                .create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
                    label: Some("AvgPool1D Bind Group Layout"),
                    entries: &[
                        wgpu::BindGroupLayoutEntry {
                            binding: 0,
                            visibility: wgpu::ShaderStages::COMPUTE,
                            ty: wgpu::BindingType::Buffer {
                                ty: wgpu::BufferBindingType::Storage { read_only: true },
                                has_dynamic_offset: false,
                                min_binding_size: None,
                            },
                            count: None,
                        },
                        wgpu::BindGroupLayoutEntry {
                            binding: 1,
                            visibility: wgpu::ShaderStages::COMPUTE,
                            ty: wgpu::BindingType::Buffer {
                                ty: wgpu::BufferBindingType::Storage { read_only: false },
                                has_dynamic_offset: false,
                                min_binding_size: None,
                            },
                            count: None,
                        },
                        wgpu::BindGroupLayoutEntry {
                            binding: 2,
                            visibility: wgpu::ShaderStages::COMPUTE,
                            ty: wgpu::BindingType::Buffer {
                                ty: wgpu::BufferBindingType::Uniform,
                                has_dynamic_offset: false,
                                min_binding_size: None,
                            },
                            count: None,
                        },
                    ],
                });

        // Create bind group
        let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
            label: Some("AvgPool1D Bind Group"),
            layout: &bind_group_layout,
            entries: &[
                wgpu::BindGroupEntry {
                    binding: 0,
                    resource: input_buffer.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 1,
                    resource: output_buffer.as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 2,
                    resource: params_buffer.as_entire_binding(),
                },
            ],
        });

        // Create compute pipeline
        let shader_module = device.compile_shader(Self::wgsl_shader(), Some("Shader"));

        let pipeline_layout =
            device
                .device
                .create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
                    label: Some("AvgPool1D Pipeline Layout"),
                    bind_group_layouts: &[&bind_group_layout],
                    immediate_size: 0,
                });

        let compute_pipeline =
            device
                .device
                .create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
                    label: Some("AvgPool1D Pipeline"),
                    layout: Some(&pipeline_layout),
                    module: &shader_module,
                    entry_point: Some("main"),
                    cache: None,
                    compilation_options: Default::default(),
                });

        // Execute compute shader
        let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
            label: Some("AvgPool1D Encoder"),
        });

        {
            let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
                label: Some("AvgPool1D Pass"),
                timestamp_writes: None,
            });
            compute_pass.set_pipeline(&compute_pipeline);
            compute_pass.set_bind_group(0, Some(&bind_group), &[]);
            // Deep Debt Evolution: Capability-based dispatch
            let caps = DeviceCapabilities::from_device(device);
            let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::Convolution);
            let workgroups = (total_elements as u32).div_ceil(optimal_wg_size);
            compute_pass.dispatch_workgroups(workgroups, 1, 1);
        }

        device.submit_commands(Some(encoder.finish()));

        // Read back results
        let output_data = crate::utils::read_buffer(device, &output_buffer, total_elements)?;

        Ok(Tensor::new(
            output_data,
            vec![batch_size, channels, output_size],
            device.clone(),
        ))
    }
}

impl Tensor {
    /// Apply 1D average pooling over the tensor
    ///
    /// # Arguments
    ///
    /// * `kernel_size` - Size of the pooling window
    /// * `stride` - Stride of the pooling operation
    ///
    /// # Returns
    ///
    /// Pooled tensor with shape (batch, channels, `output_length`)
    ///
    /// # Errors
    ///
    /// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
    /// readback fails (e.g. device lost or out of memory).
    pub fn avg_pool1d_wgsl(self, kernel_size: usize, stride: usize) -> Result<Self> {
        AvgPool1D::new(self, kernel_size, stride).execute()
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[tokio::test]
    async fn test_avg_pool1d_basic() {
        let device = crate::device::test_pool::get_test_device().await;
        // Input: [1, 1, 4] - single batch, single channel, 4 elements
        let data = vec![1.0, 2.0, 3.0, 4.0];
        let input = Tensor::new(data, vec![1, 1, 4], device);

        let output = input.avg_pool1d_wgsl(2, 2).unwrap();

        assert_eq!(output.shape(), &[1, 1, 2]);
        let result = output.to_vec().unwrap();
        // Avg of [1,2]=1.5, [3,4]=3.5
        assert_eq!(result[0], 1.5);
        assert_eq!(result[1], 3.5);
    }

    #[tokio::test]
    async fn test_avg_pool1d_multi_channel() {
        let device = crate::device::test_pool::get_test_device().await;
        // Input: [1, 2, 4] - single batch, 2 channels, 4 elements each
        let data = vec![
            1.0, 2.0, 3.0, 4.0, // Channel 0
            4.0, 3.0, 2.0, 1.0, // Channel 1
        ];
        let input = Tensor::new(data, vec![1, 2, 4], device);

        let output = input.avg_pool1d_wgsl(2, 2).unwrap();

        assert_eq!(output.shape(), &[1, 2, 2]);
        let result = output.to_vec().unwrap();
        // Channel 0: avg([1,2])=1.5, avg([3,4])=3.5
        // Channel 1: avg([4,3])=3.5, avg([2,1])=1.5
        assert_eq!(result[0], 1.5);
        assert_eq!(result[1], 3.5);
        assert_eq!(result[2], 3.5);
        assert_eq!(result[3], 1.5);
    }

    #[tokio::test]
    async fn test_avg_pool1d_stride_one() {
        let device = crate::device::test_pool::get_test_device().await;
        // Input: [1, 1, 4] - overlapping windows
        let data = vec![2.0, 4.0, 6.0, 8.0];
        let input = Tensor::new(data, vec![1, 1, 4], device);

        let output = input.avg_pool1d_wgsl(3, 1).unwrap();

        assert_eq!(output.shape(), &[1, 1, 2]);
        let result = output.to_vec().unwrap();
        // avg([2,4,6])=4.0, avg([4,6,8])=6.0
        assert_eq!(result[0], 4.0);
        assert_eq!(result[1], 6.0);
    }
}