use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
const SHADER_F64: &str = include_str!("../shaders/dropout/spatial_dropout_f64.wgsl");
static SHADER_F32: std::sync::LazyLock<String> =
std::sync::LazyLock::new(|| SHADER_F64.to_string());
#[repr(C)]
#[derive(Copy, Clone, Debug, bytemuck::Pod, bytemuck::Zeroable)]
struct SpatialDropoutParams {
batch_size: u32,
channels: u32,
height: u32,
width: u32,
drop_prob: f32,
training: u32,
_padding: u32,
_padding2: u32,
}
pub struct SpatialDropout {
input: Tensor,
mask: Tensor,
drop_prob: f32,
training: bool,
}
impl SpatialDropout {
pub fn new(input: Tensor, mask: Tensor, drop_prob: f32, training: bool) -> Result<Self> {
let input_shape = input.shape();
if input_shape.len() != 4 {
return Err(BarracudaError::invalid_op(
"spatial_dropout",
"input must be 4D tensor [B, C, H, W]",
));
}
let mask_shape = mask.shape();
if mask_shape.len() != 2
|| mask_shape[0] != input_shape[0]
|| mask_shape[1] != input_shape[1]
{
return Err(BarracudaError::invalid_op(
"spatial_dropout",
"mask must be 2D tensor [B, C] matching input batch and channels",
));
}
if !(0.0..1.0).contains(&drop_prob) {
return Err(BarracudaError::invalid_op(
"spatial_dropout",
"drop_prob must be in range [0.0, 1.0)",
));
}
Ok(Self {
input,
mask,
drop_prob,
training,
})
}
fn wgsl_shader() -> &'static str {
&SHADER_F32
}
pub fn execute(self) -> Result<Tensor> {
let device = self.input.device();
let shape = self.input.shape();
let batch_size = shape[0];
let channels = shape[1];
let height = shape[2];
let width = shape[3];
let output_size = batch_size * channels * height * width;
let output_buffer = device.create_buffer_f32(output_size)?;
let params = SpatialDropoutParams {
batch_size: batch_size as u32,
channels: channels as u32,
height: height as u32,
width: width as u32,
drop_prob: self.drop_prob,
training: u32::from(self.training),
_padding: 0,
_padding2: 0,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("spatial_dropout_params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM,
});
let shader = device.compile_shader(Self::wgsl_shader(), Some("spatial_dropout_shader"));
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("spatial_dropout_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 pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("spatial_dropout_pipeline_layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("spatial_dropout_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("spatial_dropout_bind_group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.input.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.mask.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 mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("spatial_dropout_encoder"),
});
{
let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("spatial_dropout_pass"),
timestamp_writes: None,
});
compute_pass.set_pipeline(&pipeline);
compute_pass.set_bind_group(0, Some(&bind_group), &[]);
let caps = DeviceCapabilities::from_device(device);
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::ElementWise);
let workgroups_x = (width as u32).div_ceil(optimal_wg_size);
let workgroups_y = (height as u32).div_ceil(optimal_wg_size);
let workgroups_z = batch_size * channels;
compute_pass.dispatch_workgroups(workgroups_x, workgroups_y, workgroups_z as u32);
}
device.submit_commands(Some(encoder.finish()));
Ok(Tensor::from_buffer(
output_buffer,
vec![batch_size, channels, height, width],
device.clone(),
))
}
}
impl Tensor {
pub fn spatial_dropout(self, mask: Self, drop_prob: f32, training: bool) -> Result<Self> {
SpatialDropout::new(self, mask, drop_prob, training)?.execute()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_spatial_dropout_basic() {
let device = crate::device::test_pool::get_test_device().await;
let input = Tensor::from_vec_on(vec![1.0; 3 * 4 * 4], vec![1, 3, 4, 4], device.clone())
.await
.unwrap();
let mask = Tensor::from_vec_on(vec![1.0; 3], vec![1, 3], device.clone())
.await
.unwrap();
let output = input.spatial_dropout(mask, 0.5, true).unwrap();
let result = output.to_vec().unwrap();
assert_eq!(output.shape(), &[1, 3, 4, 4]);
assert_eq!(result.len(), 48);
assert!(result.iter().all(|&x| x.is_finite()));
}
}