BUG: OpenBLAS error

[TamirEliav]

Describe the issue:

Hi
First, thanks for the great tool! it does a remarkable job!

I have a very long recording (118 hours, 3.14TB), with 128 ch (4-shank probe configuration).
I tried to sort a piece of it (5h), but it returned error during "Computing drift correction".

First, I get the following warning message about OpenBLAS in the terminal:

OpenBLAS warning: precompiled NUM_THREADS exceeded, adding auxiliary array for thread metadata.
To avoid this warning, please rebuild your copy of OpenBLAS with a larger NUM_THREADS setting
or set the environment variable OPENBLAS_NUM_THREADS to 24 or lower

and then the program crash, with the following output:

BLAS : Bad memory unallocation! : 562 000005C3F5780000
BLAS : Bad memory unallocation! : 562 000005C405780000
QObject::killTimer: Timers cannot be stopped from another thread
QObject::~QObject: Timers cannot be stopped from another thread

It happens both when running from GUI or code.

When I try shorter durations (<2h), I manage to run it, and the results looks great.

It shouldn't be a problem of hardware limitation, because I use a very strong workstation with 1TB RAM, 20GB GPU memory, 56 cores (or 112 logical processors), and data is located in ssd drive. From the warning message I suspect the opposite, i.e. that I have too many cores and KS is trying to invoke more threads than OpenBLAS is configured to support. As suggsted in the warning, I tried to change the number of threads from the code as follows:

os.environ["OPENBLAS_NUM_THREADS"] = "16"
os.environ["OMP_NUM_THREADS"] = "16"
os.environ["NUM_THREADS"] = "16"

but none solved the problem.
The other suggested option was to rebuild OpenBLAS, but before I try this I wanted to ask here for help, maybe there is an easier solution.

Here are the log file and the terminal output:
kilosort4.log
terminal_error.txt

Thanks! Tamir

Reproduce the bug:

Error message:

Version information:

KS: 4.0.24
CUDA: 11.8
window 11 Version 10.0.26100 Build 26100
Processor Intel(R) Xeon(R) w9-3495X, 1896 Mhz, 56 Core(s), 112 Logical Processor(s)