use crate::device::DeviceCapabilities;
use crate::error::{BarracudaError, Result};
use crate::ops::matmul::MatMul;
use crate::tensor::Tensor;
pub struct MatrixPower {
input: Tensor,
power: i32,
}
impl MatrixPower {
pub fn new(input: Tensor, power: i32) -> Result<Self> {
let shape = input.shape();
if shape.len() < 2 {
return Err(BarracudaError::invalid_op(
"matrix_power",
"Requires at least 2D tensor",
));
}
let rows = shape[shape.len() - 2];
let cols = shape[shape.len() - 1];
if rows != cols {
return Err(BarracudaError::invalid_op(
"matrix_power",
format!("Requires square matrix, got {rows}x{cols}"),
));
}
if power < 0 {
return Err(BarracudaError::invalid_op(
"matrix_power",
"Negative powers not supported (requires matrix inversion)",
));
}
Ok(Self { input, power })
}
fn wgsl_shader() -> &'static str {
static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
include_str!("../shaders/math/matrix_power_f64.wgsl").to_string()
});
&SHADER
}
pub fn execute(self) -> Result<Tensor> {
let device = self.input.device();
let shape = self.input.shape();
let size = shape[shape.len() - 1];
let matrix_size = size * size;
if self.power == 0 {
let identity_buffer = device.create_buffer_f32(matrix_size)?;
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct MatrixPowerParams {
size: u32,
_pad1: u32,
_pad2: u32,
_pad3: u32,
}
let params = MatrixPowerParams {
size: size as u32,
_pad1: 0,
_pad2: 0,
_pad3: 0,
};
let params_buffer =
device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("MatrixPower Identity Params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
});
let shader_module =
device.compile_shader(Self::wgsl_shader(), Some("MatrixPower Shader"));
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("MatrixPower Identity Bind Group Layout"),
entries: &[
wgpu::BindGroupLayoutEntry {
binding: 0,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
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("MatrixPower Identity Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: params_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: identity_buffer.as_entire_binding(),
},
],
});
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("MatrixPower Identity Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline =
device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("MatrixPower Identity Pipeline"),
layout: Some(&pipeline_layout),
module: &shader_module,
entry_point: Some("init_identity"),
cache: None,
compilation_options: Default::default(),
});
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("MatrixPower Identity Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("MatrixPower Identity Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
let caps = DeviceCapabilities::from_device(device);
let (workgroups_x, workgroups_y) = caps.dispatch_2d(size as u32, size as u32);
pass.dispatch_workgroups(workgroups_x, workgroups_y, 1);
}
device.submit_commands(Some(encoder.finish()));
let output_shape = shape.to_vec();
let output_elem_count = output_shape.iter().product::<usize>();
let output_data =
crate::utils::read_buffer(device, &identity_buffer, output_elem_count)?;
return Ok(Tensor::new(output_data, output_shape, device.clone()));
}
if self.power == 1 {
return Ok(self.input);
}
let mut n = self.power as u32;
let mut base = self.input.clone();
let identity_data: Vec<f32> = (0..matrix_size)
.map(|i| if i % (size + 1) == 0 { 1.0 } else { 0.0 })
.collect();
let mut result = Tensor::from_data(&identity_data, shape.to_vec(), device.clone())?;
while n > 0 {
if n % 2 == 1 {
result = MatMul::new(&result, &base).execute()?;
}
base = MatMul::new(&base, &base).execute()?;
n /= 2;
}
Ok(result)
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_matrix_power_basic() {
let device = crate::device::test_pool::get_test_device().await;
let matrix = Tensor::from_vec_on(vec![2.0, 0.0, 0.0, 2.0], vec![2, 2], device.clone())
.await
.unwrap();
let result = MatrixPower::new(matrix, 2).unwrap().execute().unwrap();
let output = result.to_vec().unwrap();
assert!((output[0] - 4.0).abs() < 1e-4);
assert!((output[3] - 4.0).abs() < 1e-4);
}
#[tokio::test]
async fn test_matrix_power_zero() {
let device = crate::device::test_pool::get_test_device().await;
let matrix = Tensor::from_vec_on(vec![5.0, 3.0, 2.0, 1.0], vec![2, 2], device.clone())
.await
.unwrap();
let result = MatrixPower::new(matrix, 0).unwrap().execute().unwrap();
let output = result.to_vec().unwrap();
assert!((output[0] - 1.0).abs() < 1e-5);
assert!(output[1].abs() < 1e-5);
assert!(output[2].abs() < 1e-5);
assert!((output[3] - 1.0).abs() < 1e-5);
}
#[tokio::test]
async fn test_matrix_power_one() {
let device = crate::device::test_pool::get_test_device().await;
let matrix = Tensor::from_vec_on(vec![1.0, 2.0, 3.0, 4.0], vec![2, 2], device.clone())
.await
.unwrap();
let result = MatrixPower::new(matrix.clone(), 1)
.unwrap()
.execute()
.unwrap();
let output = result.to_vec().unwrap();
let input = matrix.to_vec().unwrap();
assert_eq!(output, input);
}
}