[Mlir-commits] [mlir] [mlir][TilingInterface] Add scf::tileUsingSCFForallOp method to tile using the interface to generate `scf::forall`. (PR #67083)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

@nicolasvasilache I adapted the implementation here from `linalg::tileToForAll`. Just a few questions while I add more tests

1) The `scf.forall` op allows two modes, the `num_threads` mode and the `tile_size` mode. I actually not sure there is a difference apart from loop bounds here, or am I missing something? Currently I am only using the `tile_size` mode.
2) The linalg operation has an `omitTileOffsetBoundsCheck` here https://github.com/llvm/llvm-project/blob/985362e2f38f0e1d70a3067851ae072ac11ffb33/mlir/lib/Dialect/Linalg/Transforms/Tiling.cpp#L322 . I am not sure why that is needed. Is there a difference for the `tile_size` mode and the `num_threads` mode here?


https://github.com/llvm/llvm-project/pull/67083