test: aarch64: Skip Jit uni reorder if 4d matrix and zero point is defined

[Ryo-not-rio]

Description

Jit:uni reorders for certain 4d matrices are returning incorrect results when the src zero point != 0. A guard has been added to skip jit uni reorders for any 4d matrices where src zero point != 0 so the simple:any implementation will be used. A new testcase has also been added to benchdnn.

Reproduction:

./tests/benchdnn/benchdnn --reorder  --sdt=u8 --ddt=f32 --stag=adbc --attr-zero-points=src:common:1 1x32x128x33

General

• Do all unit and benchdnn tests (make test and make test_benchdnn_*) pass locally for each commit?
• Have you formatted the code using clang-format?

Bug fixes

• Have you included information on how to reproduce the issue (either in a github issue or in this PR)?
• Have you added relevant regression tests?

[dzarukin]

Since this implementation is a copy-paste of existing one for x64, I suggest moving the check either into prb_init or into tr::kernel_t::desc_init.

As for the check itself, is it only ndims=4? Or any ndims is affected by this limitation?

If the latter, then existing testing coverage should be enough to trigger the error and probably might not be needed.

[Ryo-not-rio]

Thank you for the review, I have moved the check to prb_init.

As far as I can tell, the issue seems to be only ndims=4. I have run test_benchdnn_modeC_reorder_ci_cpu which passes so I would assume that the other ndims are okay

[Shreyas-fuj]

Hi @Ryo-not-rio, @theComputeKid, @dzarukin
Can you please hold this PR as we have found the bug fix for this and will be raising a PR shortly so that the above mentioned cases passes on jit:uni itself instead of falling back to simple:any.
Thanks.

[theComputeKid]

@Shreyas-fuj Thanks for looking into it. Much appreciated.