ALL ZEROES + MESSES UP DEVICE | 4x8 | 32x4096x128 | i32->i32 | Strix

[Abhishek-Varma]

Repro:

aie_compilation_flags = [
            "--iree-amdaie-num-rows=4",
            "--iree-amdaie-num-cols=8",
        ]
self.register(
                Matmul(
                    32,
                    4096,
                    128,
                    "i32",
                    "i32",
                    test_params=TestParams(
                        tile_pipeline="pack-peel-4-level-tiling",
                        run_on_target=["npu4"],
                        aie_compilation_flags=aie_compilation_flags,
                    ),
                )
            )

Issue: Results in ALL ZEROES + messes up with the device driver.

To check how it messes up the driver :-
0. Run xrt-smi examine - you should see the device present.

1. Run 32 x 32 x 32 first - it will work.
2. Then try the above 32 x 4096 x 128 - it'll give zeroes.
3. Now again try 32 x 32 x 32 - it won't work.
4. Now when you try xrt-smi examine it'd show 0 devices present.

Unable to upload IR but with 5e6574d61d15ef130c915c001be2d991b8324836 as the Top-Of-Main commit, it can be generated.