How to avoid memory leak in a simple looping?

[3togo]

The following program will crash after running for 21 times in my computer.

How to avoid it?

use arrayfire::*;
fn main() {
    let mut y = randu!(1000, 1000, 100);
    for i in 0..100 {
        let x = randu!(1000, 1000, 100);
        y = y + x;
        println!("i={}", i);
        device_gc();
    }
    println!("max(y)={:#?}", max_all(&y));
}

   Compiling crash v0.1.0 (/work/gitee/j_viewr/egs/crash)
    Finished dev [unoptimized + debuginfo] target(s) in 0.32s
     Running `target/debug/crash`
i=0
i=1
i=2
i=3
i=4
i=5
i=6
i=7
i=8
i=9
i=10
i=11
i=12
i=13
i=14
i=15
i=16
i=17
i=18
i=19
i=20
i=21
thread 'main' panicked at 'Error message: System or Device ran out of memory
Last error: In function virtual void* cuda::Allocator::nativeAlloc(size_t)
In file src/backend/cuda/memory.cpp:157
CUDA Error (2): out of memory

 0# 0x00007FE39F5BC8F3 in /opt/arrayfire/lib/libafcuda.so.3
 1# 0x00007FE39FE277EE in /opt/arrayfire/lib/libafcuda.so.3
 2# 0x00007FE3A027B4B1 in /opt/arrayfire/lib/libafcuda.so.3
 3# 0x00007FE39F5BE920 in /opt/arrayfire/lib/libafcuda.so.3
 4# 0x00007FE39EF4A3A4 in /opt/arrayfire/lib/libafcuda.so.3
 5# 0x00007FE39EF4A506 in /opt/arrayfire/lib/libafcuda.so.3
 6# 0x00007FE3A0237B00 in /opt/arrayfire/lib/libafcuda.so.3
 7# 0x00007FE3A021B2D3 in /opt/arrayfire/lib/libafcuda.so.3
 8# 0x00007FE39FB132B5 in /opt/arrayfire/lib/libafcuda.so.3
 9# 0x00007FE39FBB0B05 in /opt/arrayfire/lib/libafcuda.so.3
10# 0x00007FE39FBA4671 in /opt/arrayfire/lib/libafcuda.so.3
11# af_add in /opt/arrayfire/lib/libafcuda.so.3
12# af_add in /opt/arrayfire/lib/libaf.so.3
13# 0x00005632DF68AA26 in target/debug/crash
14# 0x00005632DF68AB79 in target/debug/crash
15# 0x00005632DF68B08F in target/debug/crash
16#', /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/error.rs:37:14
stack backtrace:
   0: rust_begin_unwind
             at /rustc/a55dd71d5fb0ec5a6a3a9e8c27b2127ba491ce52/library/std/src/panicking.rs:584:5
   1: core::panicking::panic_fmt
             at /rustc/a55dd71d5fb0ec5a6a3a9e8c27b2127ba491ce52/library/core/src/panicking.rs:142:14
   2: arrayfire::core::error::handle_error_general
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/error.rs:37:14
   3: arrayfire::core::error::Callback::call
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/error.rs:29:9
   4: arrayfire::core::error::HANDLE_ERROR
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/error.rs:95:5
   5: arrayfire::core::arith::add_helper
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/arith.rs:410:17
   6: arrayfire::core::arith::add
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/arith.rs:468:22
   7: arrayfire::core::arith::<impl core::ops::arith::Add<arrayfire::core::array::Array<B>> for arrayfire::core::array::Array<A>>::add
             at /home/eli/.cargo/registry/src/mirrors.ustc.edu.cn-12df342d903acd47/arrayfire-3.8.0/src/core/arith.rs:785:17
   8: crash::main
             at ./src/main.rs:6:13
   9: core::ops::function::FnOnce::call_once
             at /rustc/a55dd71d5fb0ec5a6a3a9e8c27b2127ba491ce52/library/core/src/ops/function.rs:248:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

[3togo]

I have tried drop(), device_gc() and arrayfire::sync() but in vain.

[3togo]

The following modification will fix the problem.

use arrayfire::*;

fn main() {
    set_backend(Backend::OPENCL);
    let mut y = randu!(1000, 1000, 100);
    for i in 0..1000 {
        let x = randu!(1000, 1000, 100);
        let z = x + &y;
        y = z.copy();
        println!("i={}", i);
    }
    println!("max(y)={:#?}", max_all(&y));
}

[9prady9]

Sounds like a bug because the code you shared should never hit memory limit because x is created inside the loop and there exists only one array outside of loop that is being modified. I would expect three buffers to be alive during each iteration: 1y, 1x and 1output buffer created during the binary operation. The 1output and 1x should be reused. Even if some reason more buffers are used, it still shouldn't lead to OOM. What is the total device memory of your GPU - also what is your GPU ?