Tom Westerhout

I've created a gist for you: https://gist.github.com/twesterhout/2ed231fc037e3013341a0c8efee178a7 If you download all the files and then run `nix build`, it should try to build and test ArrayFire.

@umar456 I've updated the gist with more instructions how to reproduce the failures. A few other things: * Here's a backtrace from `gdb`: ``` Thread 9 "solve_dense_cpu" received signal SIGSEGV,...

I'm not having any luck with `-fsanitize=address` because that setup has memory overcommit disabled. Will try `valgrind` now...

I've also managed to reproduce the issue on a Sandybridge machine: the exact same code, and the segfault also happens in `zgemv_n_SANDYBRIDGE`.

And some output from valgrind if it helps: ``` [ RUN ] Solve/3.LeastSquaresUnderDeterminedMultipleOfTwoLarge ==39914== Thread 17: ==39914== Invalid read of size 16 ==39914== at 0x76CD798: zgemv_n_HASWELL (in /nix/store/nz7x22bdcln7vjws178lx8abrz0cw0v1-openblas-0.3.21/lib/libopenblasp-r0.3.21.so) ==39914== by...

Interestingly, I don't get a segfault via valgrind either, but it does show two invalid reads in `gemv`.

@umar456 I'm still struggling with this issue. I've tried reproducing the issue with plain C and OpenBLAS, but to no avail. Are there any BLAS settings that ArrayFire modifies? E.g....

I tested with Intel MKL 2023 yesterday, and all tests passed without issues. I'm going to try the fix suggested by @martin-frbg in https://github.com/xianyi/OpenBLAS/issues/4013. Should be relatively easy to accomplish...