[Mlir-commits] [mlir] Revert "[mlir][tensor] Forward concat insert_slice destination into DPS provider" (PR #190143)

[Dhruv Chauhan]

dchauhan-arm wrote:

My preference would be to keep this PR limited to the revert so we can get the problematic change out cleanly. A follow up patch will reintroduce the optimization in a safer form, and I can add some mlir-opt regression test there if I can reduce the repro to a normal pass pipeline. If that is not possible, I’ll still try and add the narrowest stable regression test that still covers this issue. Would that be acceptable? 

https://github.com/llvm/llvm-project/pull/190143