libmcl-sys 0.1.2-rc3

use libmcl_sys::*;
use rand::Rng;
use std::ffi::{c_void, CString};
use std::mem::size_of;

fn gemm_seq(a: &Vec<i32>, b: &Vec<i32>, c: &mut Vec<i32>, n: usize) {
    for i in 0..n {
        for j in 0..n {
            for k in 0..n {
                c[i * n + j] += a[i * n + k] * b[k * n + j]
            }
        }
    }
}

fn gemm_mcl(
    a: &mut Vec<i32>,
    b: &mut Vec<i32>,
    c: &mut Vec<i32>,
    n: &mut usize,
    reps: usize,
    sync: &bool,
) {
    let mut hdls: Vec<*mut mcl_handle> = Vec::new();
    unsafe {
        for i in 0..reps {
            let size: u64 = c.len() as u64;
            let mut pes: [u64; 3] = [*n as u64, *n as u64, 1];
            let mut les: [u64; 3] = [1; 3];
            let kernel_path = CString::new("tests/gemmN.cl").unwrap();
            let kernel_name = CString::new("gemmN").unwrap();

            let empty = CString::new("").unwrap();
            mcl_prg_load(kernel_path.into_raw(), empty.into_raw(), MCL_PRG_SRC.into());
            hdls.push(mcl_task_create());

            // Get a raw void ptr to our data to pass it to mcl C inteface
            let a_ptr: *mut c_void = a.as_mut_ptr() as *mut c_void;
            let b_ptr: *mut c_void = b.as_mut_ptr() as *mut c_void;
            let c_ptr: *mut c_void = c.as_mut_ptr() as *mut c_void;
            let n_ptr: *mut c_void = n as *mut _ as *mut c_void;
            let pes_ptr: *mut u64 = &mut pes as *mut _ as *mut u64;
            let les_ptr: *mut u64 = &mut les as *mut _ as *mut u64;

            assert_eq!(mcl_task_set_kernel(hdls[i], kernel_name.into_raw(), 4), 0);
            assert_eq!(
                mcl_task_set_arg(
                    hdls[i],
                    0,
                    a_ptr,
                    size * size_of::<i32>() as u64,
                    (MCL_ARG_BUFFER | MCL_ARG_INPUT).into()
                ),
                0
            );
            assert_eq!(
                mcl_task_set_arg(
                    hdls[i],
                    1,
                    b_ptr,
                    size * size_of::<i32>() as u64,
                    (MCL_ARG_BUFFER | MCL_ARG_INPUT).into()
                ),
                0
            );
            assert_eq!(
                mcl_task_set_arg(
                    hdls[i],
                    2,
                    n_ptr,
                    size_of::<usize>() as u64,
                    (MCL_ARG_SCALAR | MCL_ARG_INPUT).into()
                ),
                0
            );
            assert_eq!(
                mcl_task_set_arg(
                    hdls[i],
                    3,
                    c_ptr,
                    size * size_of::<i32>() as u64,
                    (MCL_ARG_BUFFER | MCL_ARG_OUTPUT).into()
                ),
                0
            );
            assert_eq!(mcl_exec(hdls[i], pes_ptr, les_ptr, MCL_TASK_GPU.into()), 0);

            if *sync {
                assert_eq!(mcl_wait(hdls[i]), 0);
            }
        }

        if !*sync {
            for i in 0..reps {
                assert_eq!(mcl_wait(hdls[i]), 0);
            }
        }

        for i in 0..reps {
            assert_eq!(mcl_hdl_free(hdls[i]), 0);
        }
    }
}

#[test]
fn gemm() {
    unsafe {
        let workers = 2;
        let mut n = 128;
        let nn = n * n;
        let reps = 100;
        let sync = false;

        assert_eq!(mcl_init(workers, MCL_NULL.into()), 0);

        let mut rng = rand::thread_rng();

        // Generate a and b matrices of size NxN and initialize with random numbers in [0, 100)
        let mut a: Vec<i32> = (0..nn).map(|_| rng.gen_range(0..100)).collect();
        let mut b: Vec<i32> = (0..nn).map(|_| rng.gen_range(0..100)).collect();

        // Allocate the c matrix that will hold the results.
        let mut c: Vec<i32> = vec![0; nn];
        let mut c_seq: Vec<i32> = vec![0; nn];

        gemm_seq(&a, &b, &mut c_seq, n);

        println!("Async mcl gemm");
        gemm_mcl(&mut a, &mut b, &mut c, &mut n, reps, &sync);
        assert_eq!(c_seq, c);

        let mut c = vec![0; nn];
        let sync = true;

        println!("Sync mcl gemm");
        gemm_mcl(&mut a, &mut b, &mut c, &mut n, reps, &sync);
        assert_eq!(c_seq, c);

        assert_eq!(mcl_finit(), 0);
    }
}