[Mlir-commits] [mlir] [mlir][scf] Relax requirements for loops fusion (PR #79187)

[llvmlistbot at llvm.org]

fabrizio-indirli wrote:

Thank you for having a look @Hardcode84,
I believe you're correct, I had missed those potential aliasing cases.
I have removed that part of the patch (the one handling the 2nd case) and I've modified the LIT tests.
Some of the test cases that I've made are a bit "artificial" (e.g. now the 1st loop reads from the output tensor to avoid the aliasing issue, but this is probably uncommon in real scenarios) but they should pass

https://github.com/llvm/llvm-project/pull/79187