use crate::device::DeviceCapabilities;
use crate::error::Result;
use crate::tensor::Tensor;
pub struct Take {
input: Tensor,
indices: Vec<u32>,
}
impl Take {
pub fn new(input: Tensor, indices: Vec<u32>) -> Result<Self> {
let input_size = input.shape().iter().product::<usize>();
if indices.iter().any(|&idx| idx as usize >= input_size) {
return Err(crate::error::BarracudaError::InvalidInput {
message: format!(
"Index out of bounds: input_size={input_size}, indices={indices:?}"
),
});
}
Ok(Self { input, indices })
}
fn wgsl_shader() -> &'static str {
static S: std::sync::LazyLock<String> = std::sync::LazyLock::new(|| {
include_str!("../shaders/tensor/take_f64.wgsl").to_string()
});
&S
}
pub fn execute(self) -> Result<Tensor> {
let device = self.input.device();
let output_size = self.indices.len();
if output_size == 0 {
return Ok(Tensor::new(vec![], vec![0], device.clone()));
}
let input_buffer = self.input.buffer();
let indices_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Take Indices"),
contents: bytemuck::cast_slice(&self.indices),
usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_DST,
});
let output_byte_size = (output_size * std::mem::size_of::<f32>()).max(32);
let output_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("Take Output Buffer"),
size: output_byte_size as u64,
usage: wgpu::BufferUsages::STORAGE
| wgpu::BufferUsages::COPY_DST
| wgpu::BufferUsages::COPY_SRC,
mapped_at_creation: false,
});
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct Params {
output_size: u32,
input_size: u32,
_pad1: u32,
_pad2: u32,
}
let input_size = self.input.shape().iter().product::<usize>();
let params = Params {
output_size: output_size as u32,
input_size: input_size as u32,
_pad1: 0,
_pad2: 0,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Take Params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
});
let shader_module = device.compile_shader(Self::wgsl_shader(), Some("Take Shader"));
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Take 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("Take Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: params_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: input_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: indices_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: output_buffer.as_entire_binding(),
},
],
});
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Take Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let compute_pipeline =
device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Take 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("Take Encoder"),
});
{
let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Take Pass"),
timestamp_writes: None,
});
compute_pass.set_pipeline(&compute_pipeline);
compute_pass.set_bind_group(0, Some(&bind_group), &[]);
let caps = DeviceCapabilities::from_device(device);
let workgroups = caps.dispatch_1d(output_size as u32);
compute_pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
let output_data = crate::utils::read_buffer(device, &output_buffer, output_size)?;
Ok(Tensor::new(output_data, vec![output_size], device.clone()))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_take_basic() {
let device = crate::device::test_pool::get_test_device().await;
let input = Tensor::from_data(&[10.0, 20.0, 30.0, 40.0], vec![4], device).unwrap();
let result = Take::new(input, vec![0, 2, 1]).unwrap().execute().unwrap();
let output = result.to_vec().unwrap();
assert_eq!(output.len(), 3);
assert_eq!(output[0], 10.0);
assert_eq!(output[1], 30.0);
assert_eq!(output[2], 20.0);
}
#[tokio::test]
async fn test_take_repeated() {
let device = crate::device::test_pool::get_test_device().await;
let input = Tensor::from_data(&[1.0, 2.0, 3.0], vec![3], device).unwrap();
let result = Take::new(input, vec![0, 0, 1, 1, 2])
.unwrap()
.execute()
.unwrap();
let output = result.to_vec().unwrap();
assert_eq!(output.len(), 5);
assert_eq!(output[0], 1.0);
assert_eq!(output[1], 1.0);
assert_eq!(output[2], 2.0);
assert_eq!(output[3], 2.0);
assert_eq!(output[4], 3.0);
}
#[tokio::test]
async fn test_take_large() {
let device = crate::device::test_pool::get_test_device().await;
let data: Vec<f32> = (0..1000).map(|i| i as f32).collect();
let input = Tensor::from_data(&data, vec![1000], device).unwrap();
let indices: Vec<u32> = (0..100).map(|i| (i * 10) as u32).collect();
let result = Take::new(input, indices).unwrap().execute().unwrap();
let output = result.to_vec().unwrap();
assert_eq!(output.len(), 100);
for (i, &val) in output.iter().enumerate() {
assert_eq!(val, (i * 10) as f32);
}
}
#[tokio::test]
async fn test_take_empty() {
let device = crate::device::test_pool::get_test_device().await;
let input = Tensor::from_data(&[1.0, 2.0, 3.0], vec![3], device).unwrap();
let result = Take::new(input, vec![]).unwrap().execute().unwrap();
let output = result.to_vec().unwrap();
assert_eq!(output.len(), 0);
}
#[tokio::test]
async fn test_take_invalid() {
let device = crate::device::test_pool::get_test_device().await;
let input = Tensor::from_data(&[1.0, 2.0, 3.0], vec![3], device).unwrap();
assert!(Take::new(input, vec![0, 5]).is_err());
}
}