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
//! ACOS - Inverse cosine operation - 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
//! - ✅ Capability-based dispatch (vendor-optimized workgroups)

use crate::device::DeviceCapabilities;
use crate::error::Result;
use crate::tensor::Tensor;

/// Element-wise inverse cosine: acos(x), valid for x ∈ [-1, 1].
pub struct Acos {
    input: Tensor,
}

impl Acos {
    /// Create an acos operation.
    #[must_use]
    pub fn new(input: Tensor) -> Self {
        Self { input }
    }

    fn wgsl_shader() -> &'static str {
        static SHADER: std::sync::LazyLock<String> =
            std::sync::LazyLock::new(|| include_str!("../shaders/math/acos_f64.wgsl").to_string());
        &SHADER
    }

    /// Execute acos 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.input.device();
        let size: usize = self.input.shape().iter().product();

        let output_buffer = device.create_buffer_f32(size)?;

        #[repr(C)]
        #[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
        struct Metadata {
            size: u32,
        }

        let metadata = Metadata { size: size as u32 };
        let metadata_buffer = device
            .device
            .create_buffer_init(&wgpu::util::BufferInitDescriptor {
                label: Some("ACOS Metadata"),
                contents: bytemuck::cast_slice(&[metadata]),
                usage: wgpu::BufferUsages::UNIFORM,
            });

        let bind_group_layout =
            device
                .device
                .create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
                    label: Some("ACOS 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,
                        },
                    ],
                });

        let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
            label: Some("ACOS 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: metadata_buffer.as_entire_binding(),
                },
            ],
        });

        let shader = device.compile_shader(Self::wgsl_shader(), Some("ACOS"));

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

        let pipeline = device
            .device
            .create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
                label: Some("ACOS 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("ACOS Encoder"),
        });

        {
            let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
                label: Some("ACOS Pass"),
                timestamp_writes: None,
            });
            pass.set_pipeline(&pipeline);
            pass.set_bind_group(0, Some(&bind_group), &[]);
            // Dispatch using standard 1D shader workgroup size (256)
            let caps = DeviceCapabilities::from_device(device);
            let workgroups = caps.dispatch_1d(size as u32);
            pass.dispatch_workgroups(workgroups, 1, 1);
        }

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

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

impl Tensor {
    /// Compute element-wise acos(x).
    ///
    /// # Errors
    ///
    /// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
    /// readback fails (e.g. device lost or out of memory).
    pub fn acos(self) -> Result<Self> {
        Acos::new(self).execute()
    }
}

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

    #[tokio::test]
    async fn test_acos() {
        let device = crate::device::test_pool::get_test_device().await;
        // acos(x) is defined for x ∈ [-1, 1]
        let data = vec![-1.0, -0.5, 0.0, 0.5, 1.0];
        let input = Tensor::new(data, vec![5], device);
        let output = input.acos().unwrap();
        let result = output.to_vec().unwrap();
        assert!(result.iter().all(|&x| x.is_finite()));
        // acos(0) ≈ π/2, acos(1) = 0, acos(-1) ≈ π
        assert!((result[2] - std::f32::consts::FRAC_PI_2).abs() < 0.01);
        assert!((result[4]).abs() < 0.01);
    }
}