use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
#[derive(Clone, Copy)]
pub enum DistanceMetric {
Euclidean = 0,
Manhattan = 1,
}
pub struct PbcDistance {
input_a: Tensor, input_b: Tensor, box_dims: Vec<f32>, metric: DistanceMetric,
}
impl PbcDistance {
pub fn new(
input_a: Tensor,
input_b: Tensor,
box_dims: Vec<f32>,
metric: DistanceMetric,
) -> Result<Self> {
let shape_a = input_a.shape();
let shape_b = input_b.shape();
if shape_a.len() != 2 || shape_b.len() != 2 {
return Err(BarracudaError::InvalidShape {
expected: vec![0, 0],
actual: shape_a.to_vec(),
});
}
let d_a = shape_a[1];
let d_b = shape_b[1];
if d_a != d_b {
return Err(BarracudaError::InvalidShape {
expected: vec![shape_b[0], d_a],
actual: vec![shape_b[0], d_b],
});
}
if box_dims.len() != d_a {
return Err(BarracudaError::Device(format!(
"Box dimensions ({}) must match particle dimensions ({})",
box_dims.len(),
d_a
)));
}
for &dim in &box_dims {
if dim <= 0.0 {
return Err(BarracudaError::Device(
"Box dimensions must be positive".to_string(),
));
}
}
Ok(Self {
input_a,
input_b,
box_dims,
metric,
})
}
pub fn execute(self) -> Result<Tensor> {
let device = self.input_a.device();
let shape_a = self.input_a.shape();
let shape_b = self.input_b.shape();
let m = shape_a[0];
let n = shape_b[0];
let d = shape_a[1];
let output_size = (m * n * std::mem::size_of::<f32>()) as u64;
let output_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("PBC Distance Output"),
size: output_size,
usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
mapped_at_creation: false,
});
let box_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("PBC Box Dimensions"),
contents: bytemuck::cast_slice(&self.box_dims),
usage: wgpu::BufferUsages::STORAGE,
});
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct Params {
m: u32,
n: u32,
d: u32,
metric: u32,
}
let params = Params {
m: m as u32,
n: n as u32,
d: d as u32,
metric: self.metric as u32,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("PBC Params"),
contents: bytemuck::bytes_of(¶ms),
usage: wgpu::BufferUsages::UNIFORM,
});
let shader = device.compile_shader(include_str!("pbc.wgsl"), Some("PBC Distance Shader"));
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("PBC 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,
},
wgpu::BindGroupLayoutEntry {
binding: 4,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("PBC PL"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("PBC Pipeline"),
layout: Some(&pipeline_layout),
module: &shader,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("PBC BG"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.input_a.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.input_b.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: box_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 4,
resource: params_buffer.as_entire_binding(),
},
],
});
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("PBC Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("PBC Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
let workgroups_x = (m as u32).div_ceil(16);
let workgroups_y = (n as u32).div_ceil(16);
pass.dispatch_workgroups(workgroups_x, workgroups_y, 1);
}
device.submit_commands(Some(encoder.finish()));
Ok(Tensor::from_buffer(
output_buffer,
vec![m, n],
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_pbc_distance_simple() {
let device = crate::device::test_pool::get_test_device().await;
let pos_a = vec![0.1, 0.2, 0.3, 0.8, 0.9, 0.7]; let pos_b = vec![0.9, 0.8, 0.7];
let tensor_a = Tensor::from_data(&pos_a, vec![2, 3], device.clone()).unwrap();
let tensor_b = Tensor::from_data(&pos_b, vec![1, 3], device).unwrap();
let box_dims = vec![1.0, 1.0, 1.0];
let pbc =
PbcDistance::new(tensor_a, tensor_b, box_dims, DistanceMetric::Euclidean).unwrap();
let result = pbc.execute().unwrap();
assert_eq!(result.shape(), &[2, 1]);
}
#[tokio::test]
async fn test_pbc_wrapping() {
let device = crate::device::test_pool::get_test_device().await;
let pos_a = vec![0.1, 0.5, 0.5]; let pos_b = vec![0.9, 0.5, 0.5];
let tensor_a = Tensor::from_data(&pos_a, vec![1, 3], device.clone()).unwrap();
let tensor_b = Tensor::from_data(&pos_b, vec![1, 3], device).unwrap();
let box_dims = vec![1.0, 1.0, 1.0];
let pbc =
PbcDistance::new(tensor_a, tensor_b, box_dims, DistanceMetric::Euclidean).unwrap();
let result = pbc.execute().unwrap();
let data = result.to_vec().unwrap();
assert!(
data[0] < 0.5,
"PBC wrapping should give shorter distance: got {}, expected < 0.5",
data[0]
);
}
#[tokio::test]
async fn test_pbc_multiple_particles() {
let device = crate::device::test_pool::get_test_device().await;
let pos = vec![0.1, 0.1, 0.1, 0.5, 0.5, 0.5, 0.9, 0.9, 0.9, 0.2, 0.8, 0.3];
let tensor = Tensor::from_data(&pos, vec![4, 3], device).unwrap();
let box_dims = vec![1.0, 1.0, 1.0];
let pbc =
PbcDistance::new(tensor.clone(), tensor, box_dims, DistanceMetric::Euclidean).unwrap();
let result = pbc.execute().unwrap();
assert_eq!(result.shape(), &[4, 4]);
let data = result.to_vec().unwrap();
assert!(data[0] < 1e-5, "Self-distance should be zero");
assert!(data[5] < 1e-5, "Self-distance should be zero");
assert!(data[10] < 1e-5, "Self-distance should be zero");
assert!(data[15] < 1e-5, "Self-distance should be zero");
}
}