Prevent ttlmir-opt/translate from using llvm.so

[vwellsTT]

Found that both ttmlir-opt and ttmlir-translate are still linking to LLVM dylib, despite DISABLE_LLVM_LINK_LLVM_DYLIB flags in respective CMakeLists.txt. Changing this flag seems to remedy the problem.

[nsmithtt]

Suspiciously no fallout 🧐, we'll take it if it's true! Though I'm not sure the GH action actually tested it.

@vmilosevic, I don't think this triggered a docker rebuild just looking through the GH actions.

[vwellsTT]

Suspiciously no fallout 🧐, we'll take it if it's true! Though I'm not sure the GH action actually tested it.

Yeah, not sure I trust that either haha. Are there comprehensive manual tests I can run, if there isn't convenient way to force CI to pick up changes?

[vmilosevic]

The image tag you are looking for (ghcr.io/tenstorrent/tt-mlir/tt-mlir-ci-ubuntu-22-04:dt-dcb58a0a097d08b1e54cc8eb4e48ec854b4217bcefb8ae2f39e3280af6288a7f) was already built by previous run on same branch https://github.com/tenstorrent/tt-mlir/actions/runs/12200962107/job/34038297242
And it was again built in a run after that https://github.com/tenstorrent/tt-mlir/actions/runs/12202629226/job/34043759728
So it was already prepared when the last run was executed.
I'm not sure why it was built twice like it was deleted from GHCR between runs.
If you want to verify linking in CI we can print linker info from CI script, something like

ldd tt-mlir | grep llvm.so

[vwellsTT]

Does this mean the CI tests did indeed run on this new env? I've been using this change on my own branch to force ttmlir-translate/opt to not link to .so, so I'm fairly confident that part works--more worried it will also break something that actually needs to link to .so

[nsmithtt]

The image tag you are looking for (ghcr.io/tenstorrent/tt-mlir/tt-mlir-ci-ubuntu-22-04:dt-dcb58a0a097d08b1e54cc8eb4e48ec854b4217bcefb8ae2f39e3280af6288a7f) was already built by previous run on same branch https://github.com/tenstorrent/tt-mlir/actions/runs/12200962107/job/34038297242 And it was again built in a run after that https://github.com/tenstorrent/tt-mlir/actions/runs/12202629226/job/34043759728 So it was already prepared when the last run was executed. I'm not sure why it was built twice like it was deleted from GHCR between runs. If you want to verify linking in CI we can print linker info from CI script, something like

ldd tt-mlir | grep llvm.so

I see, OK, then I think we should be safe to land it then. We'll keep a close eye on it.