GraphCapture doesn't recognise CBs with globally allocated addresses

[derdeljanTT]

Describe the bug
Graph capture doesn't recognise that CBs with globally allocated addresses will share space with the corresponding sharded input / output operand and no additional bytes will be allocated.

To Reproduce
Go to branch derdeljan/graph-capture-cbs-repro, build tt-metal and run ./build_Debug/test/ttnn/unit_tests_ttnn --gtest_filter=MlirInterfaceTests/EltwiseUnaryOpInterfaceTestFixture.MlirInterfaceTest/0 (on a lab machine). Repro test definition is in the file tests/ttnn/unit_tests/gtests/test_graph_capture_repro.cpp.

We can see from the logs:

Always | INFO     | CB In0 has the same address as the sharded input 0
Always | INFO     | CB Out0 has the same address as the sharded output 0

that the two CBs being allocated in unary sharded program factory have the sam address as the corresponding operand, however graph capture reports as if the new space will be allocated:

circular buffers:
393216
393216

Full trace:

                 Always | INFO     | Trace: [
    {
        "connections": [
            1,
            19
        ],
        "counter": 0,
        "node_type": "capture_start",
        "params": {}
    },
    {
        "connections": [
            3,
            4,
            17
        ],
        "counter": 1,
        "node_type": "function_start",
        "params": {
            "inputs": "1",
            "name": "ttnn::relu"
        }
    },
    {
        "connections": [
            1
        ],
        "counter": 2,
        "node_type": "tensor",
        "params": {
            "shape": "ttnn.Shape([3, 1, 1024, 1024])",
            "tensor_id": "0"
        }
    },
    {
        "connections": [
            2,
            5
        ],
        "counter": 3,
        "node_type": "buffer",
        "params": {
            "layout": "HEIGHT_SHARDED",
            "size": "6291456",
            "type": "L1"
        }
    },
    {
        "connections": [
            6,
            16
        ],
        "counter": 4,
        "node_type": "function_start",
        "params": {
            "inputs": "8",
            "name": "ttnn::prim::unary"
        }
    },
    {
        "connections": [
            4
        ],
        "counter": 5,
        "node_type": "tensor",
        "params": {
            "shape": "ttnn.Shape([3, 1, 1024, 1024])",
            "tensor_id": "1"
        }
    },
    {
        "connections": [
            7,
            12,
            13,
            14,
            15
        ],
        "counter": 6,
        "node_type": "function_start",
        "params": {
            "inputs": "2",
            "name": "UnaryDeviceOperation"
        }
    },
    {
        "connections": [
            8,
            9,
            10
        ],
        "counter": 7,
        "node_type": "function_start",
        "params": {
            "inputs": "5",
            "name": "tt::tt_metal::create_device_tensor"
        }
    },
    {
        "connections": [
            11,
            11,
            11,
            18
        ],
        "counter": 8,
        "node_type": "buffer",
        "params": {
            "layout": "HEIGHT_SHARDED",
            "size": "6291456",
            "type": "L1"
        }
    },
    {
        "connections": [
            8
        ],
        "counter": 9,
        "node_type": "buffer_allocate",
        "params": {
            "address": "1866087022",
            "layout": "HEIGHT_SHARDED",
            "size": "6291456",
            "type": "L1"
        }
    },
    {
        "connections": [
            11
        ],
        "counter": 10,
        "node_type": "function_end",
        "params": {
            "name": "tt::tt_metal::create_device_tensor"
        }
    },
    {
        "connections": [],
        "counter": 11,
        "node_type": "tensor",
        "params": {
            "shape": "ttnn.Shape([3, 1, 1024, 1024])",
            "tensor_id": "2"
        }
    },
    {
        "connections": [
            6
        ],
        "counter": 12,
        "node_type": "circular_buffer_deallocate_all",
        "params": {}
    },
    {
        "connections": [],
        "counter": 13,
        "node_type": "circular_buffer_allocate",
        "params": {
            "address": "0",
            "core_range_set": "{[(x=0,y=0) - (x=3,y=3)]}",
            "size": "393216"
        }
    },
    {
        "connections": [],
        "counter": 14,
        "node_type": "circular_buffer_allocate",
        "params": {
            "address": "0",
            "core_range_set": "{[(x=0,y=0) - (x=3,y=3)]}",
            "size": "393216"
        }
    },
    {
        "connections": [
            11
        ],
        "counter": 15,
        "node_type": "function_end",
        "params": {
            "name": "UnaryDeviceOperation"
        }
    },
    {
        "connections": [
            11
        ],
        "counter": 16,
        "node_type": "function_end",
        "params": {
            "name": "ttnn::prim::unary"
        }
    },
    {
        "connections": [
            18,
            20
        ],
        "counter": 17,
        "node_type": "function_end",
        "params": {
            "name": "ttnn::relu"
        }
    },
    {
        "connections": [],
        "counter": 18,
        "node_type": "tensor",
        "params": {
            "shape": "ttnn.Shape([3, 1, 1024, 1024])",
            "tensor_id": "3"
        }
    },
    {
        "connections": [
            8
        ],
        "counter": 19,
        "node_type": "buffer_deallocate",
        "params": {
            "layout": "HEIGHT_SHARDED",
            "size": "0",
            "type": "L1"
        }
    },
    {
        "connections": [],
        "counter": 20,
        "node_type": "capture_end",
        "params": {}
    }
]

Expected behavior
Graph capture should recognise that no new space will be allocated for the above-mentioned CBs.

[mbezuljTT]

#14521 merged, closing this issue