Repository navigation
[BUG] When dealing with Irreversible matrix, mx.linalg.invwill cause core dumped. #2106
Description
Describe the bug
When dealing with Irreversible matrix, mx.linalg.invwill cause core dumped.
To Reproduce
Include code snippet
import mlx.core as mx
import numpy as np
a = mx.array([[4,1],[4,1]],dtype=mx.float32)
print(a)
b = mx.linalg.inv(a)
print(b)
Expected behavior
It should throw a error.
Desktop (please complete the following information):
- MacOS 15.1.1
- Version 0.24.1
Activity
awni commentedon Apr 22, 2025
This is a tricky one to fix, but it would be good to work towards being able to catch errors from another thread.
I think one thing we could think about is to have an error flag on the cpu command encoder and set that flag when an op fails. Then have the main thread check the flag at certain points. But we also need to be careful to leave the eval in a valid state to make this work.
Redempt1onzzZZ commentedon Apr 23, 2025
I notice that in
/mlx/mlx/backend/cpu/inverse.cpp, in the funcgeneral_inv, current src code simply judge the info!=0, maybe adding an extra judgement if (info > 0) can solve this problem.My understanding of the source code of MLX is still limited, so it might be a wrong idea. I attempted to modify the source code for some development, but encountered a build issue and am still trying to solve it
mihirneal commentedon Apr 23, 2025
It's weird, i'm not getting the core dump error but this instead, which i think is pretty explicit that it is failing to invert an invertible matrix.
libc++abi: terminating due to uncaught exception of type std::runtime_error: [Inverse::eval_cpu] LU factorization failed with error code 2zcbenz commentedon Mar 18, 2026
Closing as a duplicate of #2670.