use gpgpru::prelude::*;
use gpgpru::{Device, GpuSlice, Result, create_buffer, kernel, read_buffer};
#[kernel("examples/hazard.slang", "fill")]
pub fn fill(data: &mut GpuSlice<f32>, data_len: u32, val: f32) {}
#[kernel("examples/hazard2.slang", "multiply_copy")]
pub fn multiply_copy(
src: &GpuSlice<f32>,
src_len: u32,
dst: &mut GpuSlice<f32>,
dst_len: u32,
multiplier: f32,
) {
}
fn main() -> Result<()> {
let device = Device::new()?;
const SIZE: usize = 4096;
let group_count = SIZE.div_ceil(256) as u32;
let mut buf_a = create_buffer(&vec![0.0f32; SIZE])
.on(Shared)
.with(&device)?;
let mut buf_b = create_buffer(&vec![0.0f32; SIZE])
.on(Shared)
.with(&device)?;
let mut buf_c = create_buffer(&vec![0.0f32; SIZE])
.on(Shared)
.with(&device)?;
let buf_a_size = buf_a.size() as u32;
let buf_b_size = buf_b.size() as u32;
let buf_c_size = buf_c.size() as u32;
fill(&mut buf_a, buf_a_size, 3.0).queue(&device, (group_count, 1, 1))?;
multiply_copy(&buf_a, buf_a_size, &mut buf_b, buf_b_size, 4.0)
.queue(&device, (group_count, 1, 1))?;
fill(&mut buf_a, buf_a_size, 100.0).queue(&device, (group_count, 1, 1))?;
multiply_copy(&buf_b, buf_b_size, &mut buf_c, buf_c_size, 2.0)
.queue(&device, (group_count, 1, 1))?;
fill(&mut buf_b, buf_b_size as u32, 500.0).queue(&device, (group_count, 1, 1))?;
device.dispatch()?;
let mut res_a = vec![0.0f32; SIZE];
let mut res_b = vec![0.0f32; SIZE];
let mut res_c = vec![0.0f32; SIZE];
read_buffer(&buf_a).into(&mut res_a).with(&device)?;
read_buffer(&buf_b).into(&mut res_b).with(&device)?;
read_buffer(&buf_c).into(&mut res_c).with(&device)?;
assert_eq!(res_a[0], 100.0, "WAW/WAR on BufA failed");
assert_eq!(res_b[0], 500.0, "WAW/WAR on BufB failed");
assert_eq!(res_c[0], 24.0, "RAW/WAR Chain on BufC failed");
Ok(())
}