Perf_Mon: Update OP profiler and post processing script

[mo-tenstorrent]

Follow the suggested headers from the spreadsheet.

https://tenstorrent.sharepoint.com/:x:/r/sites/Jasmina/_layouts/15/Doc.aspx?sourcedoc=%7B10AF97D8-38F7-4A56-9B2F-1F78FE7F92C0%7D&file=reports%20-%20brainstorming.xlsx&action=default&mobileredirect=true

• [ ] Input/Output expansion to separate columns (Post processing Only)

• [ ] New fields (OP profiler and OP infra):

1. Opcode: Modify OP infra to set name as OP code
2. OP Type: Add a OP_profiler setter and pybind to be set by OP infra and python
3. Opcode Call Count: Add a OP_profiler setter to be set by OP infra
4. OP Cache Setting: Add a OP_profiler setter to be set by OP infra
5. OP Cache Hit: Add a OP_profiler setter to be set by OP infra
6. OP Cache Hash: Add a OP_profiler setter to be set by OP infra

• [ ] Update Docs with definition on headers and a small sample CSV as a screenshot

[mo-tenstorrent]

@jvasilje @TT-billteng @tt-aho @arakhmati

Feel free to comment

[TT-billteng]

Should OP Cache Hash be part of the metadata field? Fallback ops won't have a hash right?

[mo-tenstorrent]

Good call, not a fan of NAs or dashes in the field?

[TT-billteng]

Good call, not a fan of NAs or dashes in the field?

Fallback OPs don't have an OP cache hash so that's why I'm wondering

[arakhmati]

Yeah, I don't think Op Cache Hash should be part of the metadata. Especially, because eventually I think the hash could become an integer instead of a string

[mo-tenstorrent]

Test files that ran and passed against https://github.com/tenstorrent-metal/tt-metal/pull/1615

profile_log_ops_stable_diff_1pass.tgz profile_log_ops_small_mock.tgz profile_log_ops_small_mock.csv profile_log_ops_full_bert.csv profile_log_ops_stable_diff_1pass.csv profile_log_ops_full_bert.tgz