tenstorrent / tt-mlir

Tenstorrent MLIR compiler
https://tenstorrent.github.io/tt-mlir/
Apache License 2.0
78 stars 13 forks source link

Uplift third_party/tt-metal to 2024-11-15 (42035c1711 + cherry-pick fix) #1295

Closed kmabeeTT closed 6 days ago

kmabeeTT commented 1 week ago

This gets us another week (Nov8 -> Nov15) of tt-metal. Using stable tt-metal 42035c1711 from tt-metal main plus a cherry pick fix for an assertion fix (not yet on tt-metal main, but expected to go in soon).

CI Runs:

tt-mlir (Passing): link tt-forge-fe (Running) : link

Would merge if approved and if tt-forge-fe is passing.

kmabeeTT commented 1 week ago

Will have to temporarily hold off merging this, since there is a regression in tt-forge-fe hit by CI.

https://github.com/tenstorrent/tt-forge-fe/actions/runs/11863453098/job/33064951475#step:11:25859

E       RuntimeError: TT_FATAL @ /__w/tt-forge-fe/tt-forge-fe/third_party/tt-mlir/third_party/tt-metal/src/tt-metal/ttnn/cpp/ttnn/operations/data_movement/pad/pad.cpp:41: rank == 4
E       info:
E       Tensor rank is not 4
E       backtrace:

Edit: Bisected to tt-metal commit here, https://github.com/tenstorrent/tt-metal/commit/3112241537b629fbd25bc28c9aa64ffab86abb79, will open ticket.

Edit2: Metal ticket tenstorrent/tt-metal/issues/15167 solved and fix merged to tt-metal now.

kmabeeTT commented 6 days ago

I propose to merge this after Nick lands https://github.com/tenstorrent/tt-mlir/pull/1309 since it passes CI for tt-mlir and tt-forge-fe now when cherry-picking tt-metal fix for previous mentioned issue (added tt-mlir test to catch it yesterday in seperate PR) and because latest tt-metal exposes other failures. Link to passing CI:

tt-forge-fe: https://github.com/tenstorrent/tt-forge-fe/actions/runs/11906743486 tt-mlir: https://github.com/tenstorrent/tt-mlir/actions/runs/11906770146