[llvm] [HEXAGON] [MachinePipeliner] Fix the DAG in case of dependent phis. (PR #135925)

[Ryotaro Kasuga via llvm-commits]

================

----------------
kasuga-fj wrote:

I tried running this test in my local environment with the `-debug-only=pipeliner` option, and the output says: `Unable to analyzeLoop, can NOT pipeline Loop`. This means that the pipeliner bails out before reaching `updatePhiDependences`, so this test doesn't ensure that a circuit is not created. I don't know much about Hexagon, but it seems that the input to the pipeliner doesn't form a hardware loop for some reason (you can use `-stop-before=pipeliner` to check what input is being fed into the pipeliner). In my opinion, it would be easier to use .mir instead of .ll and run the test with the `-run-pass=pipeliner` option to avoid this kind of issue.


https://github.com/llvm/llvm-project/pull/135925