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
//! Pdist - Pure WGSL
//!
//! Deep Debt Principles:
//! - Self-knowledge: Operation knows its computation
//! - Zero hardcoding: Hardware-agnostic implementation
//! - 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;

/// Pairwise Distance (all pairs) operation
pub struct Pdist {
    input: Tensor,
    p: f32,
    epsilon: f32,
}

impl Pdist {
    /// Create a new pdist operation
    ///
    /// # Errors
    ///
    /// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
    /// readback fails (e.g. device lost or out of memory).
    pub fn new(input: Tensor, p: Option<f32>, epsilon: Option<f32>) -> Result<Self> {
        let input_shape = input.shape();
        if input_shape.len() < 2 {
            return Err(BarracudaError::invalid_op(
                "pdist",
                "input must be at least 2D",
            ));
        }

        Ok(Self {
            input,
            p: p.unwrap_or(2.0),
            epsilon: epsilon.unwrap_or(1e-8),
        })
    }

    /// Get the WGSL shader source
    fn wgsl_shader() -> &'static str {
        {
            static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
                include_str!("../shaders/misc/pdist_f64.wgsl").to_string()
            });
            std::sync::LazyLock::force(&SHADER).as_str()
        }
    }

    /// Execute the pdist 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 input_shape = self.input.shape();
        let num_vectors = input_shape[0];
        let dim = input_shape[1..].iter().product::<usize>();

        // Output is condensed distance matrix: n*(n-1)/2 pairs
        let num_pairs = num_vectors * (num_vectors - 1) / 2;
        let output_buffer = device.create_buffer_f32(num_pairs)?;

        // Create uniform buffer for parameters
        #[repr(C)]
        #[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
        struct Params {
            num_vectors: u32,
            dim: u32,
            p: f32,
            epsilon: f32,
        }

        let params = Params {
            num_vectors: num_vectors as u32,
            dim: dim as u32,
            p: self.p,
            epsilon: self.epsilon,
        };

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

        // Compile shader
        let shader_module = device.compile_shader(Self::wgsl_shader(), Some("Pdist Shader"));

        // Create bind group layout
        let bind_group_layout =
            device
                .device
                .create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
                    label: Some("Pdist 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("Pdist Bind Group"),
            layout: &bind_group_layout,
            entries: &[
                wgpu::BindGroupEntry {
                    binding: 0,
                    resource: self.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 pipeline
        let pipeline_layout =
            device
                .device
                .create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
                    label: Some("Pdist Pipeline Layout"),
                    bind_group_layouts: &[&bind_group_layout],
                    immediate_size: 0,
                });

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

        // Encode and execute
        let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
            label: Some("Pdist Encoder"),
        });

        {
            let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
                label: Some("Pdist Pass"),
                timestamp_writes: None,
            });

            pass.set_pipeline(&pipeline);
            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::MatMul);
            let workgroups_x = (num_vectors as u32).div_ceil(optimal_wg_size);
            let workgroups_y = (num_vectors as u32).div_ceil(optimal_wg_size);
            pass.dispatch_workgroups(workgroups_x, workgroups_y, 1);
        }

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

        // Create output tensor
        Ok(Tensor::from_buffer(
            output_buffer,
            vec![num_pairs],
            device.clone(),
        ))
    }
}

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

    #[tokio::test]
    async fn test_pdist_basic() {
        let device = crate::device::test_pool::get_test_device().await;
        let num_vectors = 3;
        let dim = 2;

        let input = Tensor::from_vec_on(
            vec![0.0, 0.0, 1.0, 0.0, 0.0, 1.0],
            vec![num_vectors, dim],
            device.clone(),
        )
        .await
        .unwrap();

        let output = Pdist::new(input, None, None).unwrap().execute().unwrap();

        // 3 choose 2 = 3 pairs
        assert_eq!(output.shape(), &[3]);
    }
}