use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
#[repr(C)]
#[derive(Copy, Clone, Debug, bytemuck::Pod, bytemuck::Zeroable)]
struct DiceLossParams {
size: u32,
smooth_val: f32,
_padding: [u32; 2],
}
pub struct DiceLoss {
predicted: Tensor,
target: Tensor,
smooth: f32,
}
impl DiceLoss {
pub fn new(predicted: Tensor, target: Tensor, smooth: f32) -> Result<Self> {
if predicted.shape() != target.shape() {
return Err(BarracudaError::shape_mismatch(
predicted.shape().to_vec(),
target.shape().to_vec(),
));
}
Ok(Self {
predicted,
target,
smooth,
})
}
fn wgsl_shader() -> &'static str {
static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
include_str!("../shaders/loss/dice_loss_f64.wgsl").to_string()
});
std::sync::LazyLock::force(&SHADER).as_str()
}
pub fn execute(self) -> Result<Tensor> {
let device = self.predicted.device();
let size = self.predicted.len();
let output_buffer = device.create_buffer_f32(1)?;
let params = DiceLossParams {
size: size as u32,
smooth_val: self.smooth,
_padding: [0, 0],
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Dice Loss Params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM,
});
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Dice Loss 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: 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: false },
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::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Dice Loss Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.predicted.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.target.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: params_buffer.as_entire_binding(),
},
],
});
let shader_module = device.compile_shader(Self::wgsl_shader(), Some("Dice Loss Shader"));
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Dice Loss Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Dice Loss Pipeline"),
layout: Some(&pipeline_layout),
module: &shader_module,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("Dice Loss Encoder"),
});
{
let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Dice Loss Pass"),
timestamp_writes: None,
});
compute_pass.set_pipeline(&pipeline);
compute_pass.set_bind_group(0, Some(&bind_group), &[]);
use crate::device::{DeviceCapabilities, WorkloadType};
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);
compute_pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
Ok(Tensor::from_buffer(output_buffer, vec![1], device.clone()))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_dice_loss_perfect_overlap() {
let device = crate::device::test_pool::get_test_device().await;
let pred = Tensor::from_vec_on(vec![1.0, 1.0, 0.0, 0.0], vec![4], device.clone())
.await
.unwrap();
let target = Tensor::from_vec_on(vec![1.0, 1.0, 0.0, 0.0], vec![4], device)
.await
.unwrap();
let loss = DiceLoss::new(pred, target, 1.0).unwrap().execute().unwrap();
let result = loss.to_vec().unwrap();
assert!(result[0] < 0.1, "Perfect overlap should have loss ≈ 0");
}
#[tokio::test]
async fn test_dice_loss_no_overlap() {
let device = crate::device::test_pool::get_test_device().await;
let pred = Tensor::from_vec_on(vec![1.0, 1.0, 0.0, 0.0], vec![4], device.clone())
.await
.unwrap();
let target = Tensor::from_vec_on(vec![0.0, 0.0, 1.0, 1.0], vec![4], device)
.await
.unwrap();
let loss = DiceLoss::new(pred, target, 1.0).unwrap().execute().unwrap();
let result = loss.to_vec().unwrap();
assert!(result[0] > 0.5, "No overlap should have high loss");
}
#[tokio::test]
async fn test_dice_loss_partial_overlap() {
let device = crate::device::test_pool::get_test_device().await;
let pred = Tensor::from_vec_on(vec![0.8, 0.6, 0.2, 0.1], vec![4], device.clone())
.await
.unwrap();
let target = Tensor::from_vec_on(vec![1.0, 0.0, 0.0, 1.0], vec![4], device)
.await
.unwrap();
let loss = DiceLoss::new(pred, target, 1.0).unwrap().execute().unwrap();
let result = loss.to_vec().unwrap();
assert!(result[0] > 0.0 && result[0] < 1.0);
assert!(result[0].is_finite());
}
}