EDM Packet Header Optimization: Simplify packet size storage and access (+other misc low-level optimizations)

[SeanNijjar]

Ticket

#17686

What's changed

Several low-level optimizations in EDM fabric:

• Simplify packet size storage and access

  • promote to "top-level" of packet to remove conditionality previously needed to get size info from packet
  • NOTE: packet size now specifies PAYLOAD SIZE ONLY!!! The header size must be implicitly added by fabric.
    • net this is still fine because we had to previous subtract header size when writing out to noc.

• Migrate eth_send_packet calls to new version that takes size in bytes.

  • This version avoids a number of shift operations that were present in the previously used version.

• Add and using new eth write remote reg (eth_write_remote_reg_no_txq_check) that doesn't have conditional context switch in body of function

Extra functionality: Added NOC_UNICAST_INLINE_WRITE eth packet command type to address a regression as a result of the above change (if the command type wasn't added)

Checklist

• All post commit CI passes: https://github.com/tenstorrent/tt-metal/actions/runs/13206241810
• T3K: https://github.com/tenstorrent/tt-metal/actions/runs/13206245905
• TG:https://github.com/tenstorrent/tt-metal/actions/runs/13206243531
• Blackhole Post commit CI passes (if applicable)
• Model regression CI passes (if applicable)
• Device performance regression CI passes (if applicable)
• (For models and ops writers) Full new models tests CI passes (if applicable)
• New/Existing tests provide coverage for changes

[SeanNijjar]

Wanted to call this out. It's a bit suspicious that I had to start including this after a rebase even though it appears nothing fundamentally changed here or at the caller. @tt-aho, @ayerofieiev-tt - any ideas why or objections to the new inclusion here?

[SeanNijjar]

For reference this is the error:

mesh_device = MeshDevice(2x4 grid, 8 devices), cores = {[(x=0,y=0) - (x=7,y=6)]}, initial_value = 0

    def create_global_semaphore_with_same_address(mesh_device, cores, initial_value):
        semaphore_handles = ttnn.create_global_semaphore_with_same_address(mesh_device, cores, initial_value)
>       addrs = ttnn.get_global_semaphore_address(semaphore_handles)
E       TypeError: Unable to convert function return value to a Python type! The signature was
E           (global_semaphore: ttnn._ttnn.global_semaphore.multi_device_global_semaphore) -> std::__1::vector<unsigned long, std::__1::allocator<unsigned long>>
E       
E       Did you forget to `#include <pybind11/stl.h>`? Or <pybind11/complex.h>,
E       <pybind11/functional.h>, <pybind11/chrono.h>, etc. Some automatic
E       conversions are optional and require extra headers to be included
E       when compiling your pybind11 module.

tests/ttnn/unit_tests/operations/ccl/test_ccl_common.py:35: TypeError

[tt-aho]

Think this is fine. It's needed for binding vector. Might be that this was included indirectly from a different file and that this is no longer the case now.

[tt-aho]

Think this is fine. It's needed for binding vector. Might be that this was included indirectly from a different file and that this is no longer the case now.

[SeanNijjar]

Going to close this and migrate to this other PR which has some additional changes.

Reason: There was a regression in this PR that was fixed there. Additionally this other PR heavily modifies the same files and I wanted to avoid headaches with rebasing conflicts. It's all the same type of change (optimizing EDM fabric)
#17749