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
//! Where/Select operation - Conditional selection
//! Pure WGSL implementation

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

/// Where/select conditional shader.
pub static WGSL_WHERE_OP: std::sync::LazyLock<String> =
    std::sync::LazyLock::new(|| include_str!("../shaders/tensor/where_op_f64.wgsl").to_string());

/// Conditional selection: output[i] = condition[i] ? x[i] : y[i].
pub struct Where {
    condition: Tensor,
    x: Tensor,
    y: Tensor,
}

impl Where {
    /// Create a where/select operation with condition, true-branch, and false-branch tensors.
    #[must_use]
    pub fn new(condition: Tensor, x: Tensor, y: Tensor) -> Self {
        Self { condition, x, y }
    }

    fn wgsl_shader() -> &'static str {
        {
            static S: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
                include_str!("../shaders/tensor/where_select_f64.wgsl").to_string()
            });
            &S
        }
    }

    /// Execute the conditional selection on GPU.
    /// # 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.condition.device();
        let size = self.condition.len();
        let output_buffer = device.create_buffer_f32(size)?;

        let bind_group_layout =
            device
                .device
                .create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
                    label: Some("Where BGL"),
                    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: true },
                                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::Storage { read_only: true },
                                has_dynamic_offset: false,
                                min_binding_size: None,
                            },
                            count: None,
                        },
                        wgpu::BindGroupLayoutEntry {
                            binding: 3,
                            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,
                        },
                    ],
                });

        let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
            label: Some("Where BG"),
            layout: &bind_group_layout,
            entries: &[
                wgpu::BindGroupEntry {
                    binding: 0,
                    resource: self.condition.buffer().as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 1,
                    resource: self.x.buffer().as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 2,
                    resource: self.y.buffer().as_entire_binding(),
                },
                wgpu::BindGroupEntry {
                    binding: 3,
                    resource: output_buffer.as_entire_binding(),
                },
            ],
        });

        let shader = device.compile_shader(Self::wgsl_shader(), Some("Where"));
        let pipeline_layout =
            device
                .device
                .create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
                    label: Some("Where PL"),
                    bind_group_layouts: &[&bind_group_layout],
                    immediate_size: 0,
                });

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

        let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
            label: Some("Where Encoder"),
        });

        {
            let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
                label: Some("Where 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::ElementWise);
            let workgroups = (size as u32).div_ceil(optimal_wg_size);
            pass.dispatch_workgroups(workgroups, 1, 1);
        }

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

        Ok(Tensor::from_buffer(
            output_buffer,
            self.condition.shape().to_vec(),
            device.clone(),
        ))
    }
}

impl Tensor {
    /// Select elements from `x` or `y` based on condition (1.0 = x, 0.0 = y).
    /// # Errors
    /// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
    /// readback fails (e.g. device lost or out of memory).
    pub fn where_select(condition: Self, x: Self, y: Self) -> Result<Self> {
        Where::new(condition, x, y).execute()
    }
}

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

    #[tokio::test]
    async fn test_where_basic() {
        let device = crate::device::test_pool::get_test_device().await;

        let condition = Tensor::from_vec_on(vec![1.0, 0.0, 1.0], vec![3], device.clone())
            .await
            .unwrap();
        let x = Tensor::from_vec_on(vec![10.0, 20.0, 30.0], vec![3], device.clone())
            .await
            .unwrap();
        let y = Tensor::from_vec_on(vec![100.0, 200.0, 300.0], vec![3], device)
            .await
            .unwrap();

        let result = Tensor::where_select(condition, x, y)
            .unwrap()
            .to_vec()
            .unwrap();

        assert!((result[0] - 10.0).abs() < 1e-5); // condition true -> x
        assert!((result[1] - 200.0).abs() < 1e-5); // condition false -> y
        assert!((result[2] - 30.0).abs() < 1e-5); // condition true -> x
    }
}