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 RandomErasingParams {
batch_size: u32,
channels: u32,
height: u32,
width: u32,
erase_value: f32,
_pad0: u32,
_pad1: u32,
_pad2: u32,
_pad3: u32,
_pad4: u32,
_pad5: u32,
_pad6: u32,
}
pub struct RandomErasing {
input: Tensor,
erase_boxes: Tensor,
erase_value: f32,
}
impl RandomErasing {
pub fn new(input: Tensor, erase_boxes: Tensor, erase_value: f32) -> Result<Self> {
let erase_shape = erase_boxes.shape();
if erase_shape.len() != 2 || erase_shape[1] != 4 {
return Err(BarracudaError::invalid_op(
"RandomErasing",
format!("erase_boxes must be 2D [batch_size, 4], got shape {erase_shape:?}"),
));
}
Ok(Self {
input,
erase_boxes,
erase_value,
})
}
fn wgsl_shader() -> &'static str {
static SHADER: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
include_str!("../shaders/math/random_erasing_f64.wgsl").to_string()
});
&SHADER
}
pub fn execute(self) -> Result<Tensor> {
let device = self.input.device();
let input_shape = self.input.shape();
if input_shape.len() != 4 {
return Err(BarracudaError::invalid_op(
"RandomErasing",
format!(
"input must be 4D [batch, channels, height, width], got shape {input_shape:?}"
),
));
}
let batch_size = input_shape[0];
let channels = input_shape[1];
let height = input_shape[2];
let width = input_shape[3];
if self.erase_boxes.shape()[0] != batch_size {
return Err(BarracudaError::invalid_op(
"RandomErasing",
format!(
"erase_boxes batch size {} must match input batch size {}",
self.erase_boxes.shape()[0],
batch_size
),
));
}
let output_size = batch_size * channels * height * width;
let output_buffer = device.create_buffer_f32(output_size)?;
let params = RandomErasingParams {
batch_size: batch_size as u32,
channels: channels as u32,
height: height as u32,
width: width as u32,
erase_value: self.erase_value,
_pad0: 0,
_pad1: 0,
_pad2: 0,
_pad3: 0,
_pad4: 0,
_pad5: 0,
_pad6: 0,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("RandomErasing Params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM,
});
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("RandomErasing 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("RandomErasing Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.input.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.erase_boxes.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 = device.compile_shader(Self::wgsl_shader(), Some("RandomErasing"));
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("RandomErasing Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("RandomErasing 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("RandomErasing Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("RandomErasing Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
let caps = DeviceCapabilities::from_device(device);
let (wg_x, wg_y, wg_z) = caps.optimal_workgroup_size_3d(WorkloadType::Convolution);
let workgroups_x = (width as u32).div_ceil(wg_x);
let workgroups_y = (height as u32).div_ceil(wg_y);
let workgroups_z = ((batch_size * channels) as u32).div_ceil(wg_z);
pass.dispatch_workgroups(workgroups_x, workgroups_y, workgroups_z);
}
device.submit_commands(Some(encoder.finish()));
Ok(Tensor::from_buffer(
output_buffer,
input_shape.to_vec(),
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_random_erasing_basic() {
let device = crate::device::test_pool::get_test_device().await;
let batch_size = 2;
let channels = 3;
let height = 32;
let width = 32;
let input = Tensor::from_vec_on(
vec![1.0; batch_size * channels * height * width],
vec![batch_size, channels, height, width],
device.clone(),
)
.await
.unwrap();
let erase_boxes = Tensor::from_vec_on(
vec![5.0, 5.0, 10.0, 10.0, 10.0, 10.0, 8.0, 8.0], vec![batch_size, 4],
device.clone(),
)
.await
.unwrap();
let result = RandomErasing::new(input, erase_boxes, 0.0)
.unwrap()
.execute()
.unwrap();
assert_eq!(result.shape(), &[batch_size, channels, height, width]);
}
}