use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
#[repr(C)]
#[derive(Copy, Clone, Debug, bytemuck::Pod, bytemuck::Zeroable)]
struct TensorDotParams {
output_size: u32,
contraction_size: u32,
a_outer_size: u32,
b_outer_size: u32,
a_outer_stride: u32,
b_outer_stride: u32,
a_contract_stride: u32,
b_contract_stride: u32,
}
pub struct TensorDot {
tensor_a: Tensor,
tensor_b: Tensor,
axes_a: Vec<usize>,
axes_b: Vec<usize>,
}
impl TensorDot {
pub fn new(
tensor_a: Tensor,
tensor_b: Tensor,
axes_a: Vec<usize>,
axes_b: Vec<usize>,
) -> Result<Self> {
if axes_a.len() != axes_b.len() {
return Err(BarracudaError::invalid_op(
"tensor_dot",
"Contraction axes must have same length",
));
}
let shape_a = tensor_a.shape();
let shape_b = tensor_b.shape();
for &axis in &axes_a {
if axis >= shape_a.len() {
return Err(BarracudaError::invalid_op(
"tensor_dot",
format!(
"Axis {} out of range for tensor A (rank {})",
axis,
shape_a.len()
),
));
}
}
for &axis in &axes_b {
if axis >= shape_b.len() {
return Err(BarracudaError::invalid_op(
"tensor_dot",
format!(
"Axis {} out of range for tensor B (rank {})",
axis,
shape_b.len()
),
));
}
}
for (i, (&axis_a, &axis_b)) in axes_a.iter().zip(axes_b.iter()).enumerate() {
if shape_a[axis_a] != shape_b[axis_b] {
return Err(BarracudaError::invalid_op(
"tensor_dot",
format!(
"Contraction dimension {} mismatch: {} != {}",
i, shape_a[axis_a], shape_b[axis_b]
),
));
}
}
Ok(Self {
tensor_a,
tensor_b,
axes_a,
axes_b,
})
}
fn wgsl_shader() -> &'static str {
{
static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
include_str!("../shaders/misc/tensor_dot_f64.wgsl").to_string()
});
std::sync::LazyLock::force(&SHADER).as_str()
}
}
pub fn execute(self) -> Result<Tensor> {
let device = self.tensor_a.device();
let shape_a = self.tensor_a.shape();
let shape_b = self.tensor_b.shape();
let contraction_size: usize = self.axes_a.iter().map(|&axis| shape_a[axis]).product();
let mut a_outer_dims = Vec::new();
let mut b_outer_dims = Vec::new();
for (i, &dim) in shape_a.iter().enumerate() {
if !self.axes_a.contains(&i) {
a_outer_dims.push(dim);
}
}
for (i, &dim) in shape_b.iter().enumerate() {
if !self.axes_b.contains(&i) {
b_outer_dims.push(dim);
}
}
let a_outer_size: usize = a_outer_dims.iter().product();
let b_outer_size: usize = b_outer_dims.iter().product();
let output_size = a_outer_size * b_outer_size;
let a_outer_stride = contraction_size;
let b_outer_stride = 1;
let a_contract_stride = 1;
let b_contract_stride = b_outer_size;
let output_buffer = device.create_buffer_f32(output_size)?;
let params = TensorDotParams {
output_size: output_size as u32,
contraction_size: contraction_size as u32,
a_outer_size: a_outer_size as u32,
b_outer_size: b_outer_size as u32,
a_outer_stride: a_outer_stride as u32,
b_outer_stride: b_outer_stride as u32,
a_contract_stride: a_contract_stride as u32,
b_contract_stride: b_contract_stride as u32,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("TensorDot Params"),
contents: bytemuck::bytes_of(¶ms),
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
});
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("TensorDot 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: 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("TensorDot Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: params_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.tensor_a.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: self.tensor_b.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: output_buffer.as_entire_binding(),
},
],
});
let shader = device.compile_shader(Self::wgsl_shader(), Some("TensorDot"));
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("TensorDot Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("TensorDot 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("TensorDot Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("TensorDot Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
let caps = DeviceCapabilities::from_device(device);
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::MatMul);
let workgroups = (output_size as u32).div_ceil(optimal_wg_size);
pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
let mut output_shape = a_outer_dims;
output_shape.extend(b_outer_dims);
Ok(Tensor::from_buffer(
output_buffer,
output_shape,
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_tensor_dot_basic() {
let device = crate::device::test_pool::get_test_device().await;
let a = Tensor::from_vec_on(vec![1.0, 2.0, 3.0], vec![3], device.clone())
.await
.unwrap();
let b = Tensor::from_vec_on(vec![4.0, 5.0, 6.0], vec![3], device.clone())
.await
.unwrap();
let result = TensorDot::new(a, b, vec![0], vec![0])
.unwrap()
.execute()
.unwrap();
let output = result.to_vec().unwrap();
assert_eq!(output.len(), 1);
assert!((output[0] - 32.0).abs() < 1e-4);
}
}