use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
const WGSL_SUB_F64: &str = include_str!("../shaders/math/elementwise_sub_f64.wgsl");
static SHADER_F32: std::sync::LazyLock<String> =
std::sync::LazyLock::new(|| WGSL_SUB_F64.to_string());
pub struct Sub {
lhs: Tensor,
rhs: Tensor,
}
impl Sub {
pub fn new(lhs: Tensor, rhs: Tensor) -> Result<Self> {
if lhs.shape() != rhs.shape() {
return Err(BarracudaError::shape_mismatch(
lhs.shape().to_vec(),
rhs.shape().to_vec(),
));
}
Ok(Self { lhs, rhs })
}
fn wgsl_shader() -> &'static str {
&SHADER_F32
}
pub fn execute(self) -> Result<Tensor> {
let device = self.lhs.device();
let size = self.lhs.len();
let output_buffer = device.create_buffer_f32(size)?;
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Sub 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: false },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Sub BG"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.lhs.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.rhs.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: output_buffer.as_entire_binding(),
},
],
});
let shader = device.compile_shader(Self::wgsl_shader(), Some("Sub"));
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Sub PL"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Sub 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("Sub Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Sub Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
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 = (size as u32).div_ceil(optimal_wg_size);
pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
Ok(Tensor::from_buffer(
output_buffer,
self.lhs.shape().to_vec(),
device.clone(),
))
}
}
impl Tensor {
pub fn sub(&self, other: &Self) -> Result<Self> {
Sub::new(self.clone(), other.clone())?.execute()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_sub_basic() {
let device = crate::device::test_pool::get_test_device().await;
let lhs = Tensor::from_vec_on(vec![10.0, 20.0, 30.0], vec![3], device.clone())
.await
.unwrap();
let rhs = Tensor::from_vec_on(vec![1.0, 2.0, 3.0], vec![3], device)
.await
.unwrap();
let result = lhs.sub(&rhs).unwrap().to_vec().unwrap();
assert!((result[0] - 9.0).abs() < 1e-5);
assert!((result[1] - 18.0).abs() < 1e-5);
assert!((result[2] - 27.0).abs() < 1e-5);
}
#[tokio::test]
async fn test_sub_edge_cases() {
let device = crate::device::test_pool::get_test_device().await;
let lhs = Tensor::from_vec_on(vec![0.0, 1e-6, -1e-6, 1.0, -1.0], vec![5], device.clone())
.await
.unwrap();
let rhs = Tensor::from_vec_on(vec![0.0, 1e-6, -1e-6, 1.0, -1.0], vec![5], device)
.await
.unwrap();
let result = lhs.sub(&rhs).unwrap().to_vec().unwrap();
assert!(result[0].abs() < 1e-12);
assert!(result[1].abs() < 1e-12);
assert!(result[2].abs() < 1e-12);
}
#[tokio::test]
async fn test_sub_boundary() {
let device = crate::device::test_pool::get_test_device().await;
let lhs = Tensor::from_vec_on(vec![f32::INFINITY, 1e10, 0.0], vec![3], device.clone())
.await
.unwrap();
let rhs = Tensor::from_vec_on(vec![100.0, 100.0, 100.0], vec![3], device)
.await
.unwrap();
let result = lhs.sub(&rhs).unwrap().to_vec().unwrap();
assert!(result[0].is_infinite() && result[0].is_sign_positive());
assert_eq!(result[2], -100.0);
}
#[tokio::test]
async fn test_sub_large_tensor() {
let device = crate::device::test_pool::get_test_device().await;
let size = 1000;
let lhs_data: Vec<f32> = (0..size).map(|i| (i as f32) * 2.0).collect();
let rhs_data: Vec<f32> = (0..size).map(|i| i as f32).collect();
let lhs = Tensor::from_vec_on(lhs_data, vec![size], device.clone())
.await
.unwrap();
let rhs = Tensor::from_vec_on(rhs_data, vec![size], device)
.await
.unwrap();
let result = lhs.sub(&rhs).unwrap().to_vec().unwrap();
for (i, &val) in result.iter().enumerate() {
assert!((val - i as f32).abs() < 1e-4);
}
}
#[tokio::test]
async fn test_sub_precision() {
let device = crate::device::test_pool::get_test_device().await;
let lhs_data = vec![5.0, 2.5, 1.0, 0.0, -1.0];
let rhs_data = vec![2.0, 1.5, 0.5, 0.0, -0.5];
let lhs = Tensor::from_vec_on(lhs_data.clone(), vec![5], device.clone())
.await
.unwrap();
let rhs = Tensor::from_vec_on(rhs_data.clone(), vec![5], device)
.await
.unwrap();
let gpu_result = lhs.sub(&rhs).unwrap().to_vec().unwrap();
let cpu_result: Vec<f32> = lhs_data
.iter()
.zip(rhs_data.iter())
.map(|(&a, &b)| a - b)
.collect();
for (i, (&gpu, &cpu)) in gpu_result.iter().zip(cpu_result.iter()).enumerate() {
assert!(
(gpu - cpu).abs() < 1e-6,
"Error at {i}: GPU={gpu}, CPU={cpu}"
);
}
}
}