Maxpool Config Support Needed for YOLOX

[ankitmcw]

Maxpool variant in YoloX Dark5 submodule:

1. Maxpool variant ((1, 384, 20, 20), (5, 5), (2, 2), (1, 1)) getting low pcc(0.04816194863771736)
2. Maxpool variant ((1, 384, 20, 20), (9, 9), (4, 4), (1, 1)) hangs
3. Maxpool variant ((1, 384, 20, 20), (13, 13), (6, 6), (1, 1)) throws error

Statically allocated circular buffers in program 10 clash with L1 buffers on core range [(x=0,y=0) - (x=11,y=7)]. L1 buffer allocated at 702464 and static circular buffer region ends at 846656

To reproduce, checkout to branch here and run

cd $TT_METAL_HOME
pytest tests/ttnn/unit_tests/operations/test_max_pool2d.py::test_run_max_pool_yolox

Variant ((1, 384, 20, 20), (9, 9), (4, 4), (1, 1)) is currently commented in branch due to hang issue.

[punithsekar]

In latest main, we are currently encountering the following issue across all three test cases.

>       output = ttl.tensor.max_pool2d_v2(
            haloed_act,
            reader_indices,
            in_n,
            in_h,
            in_w,
            window_h,
            window_w,
            stride_h,
            stride_w,
            pad_h,
            pad_w,
            output_mem_config=self.output_sharded_memory_config,
        )
E       RuntimeError: TT_FATAL @ ../tt_eager/tt_dnn/op_library/pool/max_pool.cpp:30: is_pow2
E       info:
E       Row size (nchannels * bytes = 768) should be power of 2 (false).
E       backtrace:
E        --- tt::tt_metal::MaxPool::validate(std::__1::vector<tt::tt_metal::Tensor, std::__1::allocator<tt::tt_metal::Tensor>> const&) const