[Mlir-commits] [mlir] [mlir][SCF] Deprecate `linalg::tileToForallOp` and `linalg::tileToForallOpUsingTileSizes` (PR #91878)

[Nicolas Vasilache]

nicolasvasilache wrote:

> Can you provide some information about what transformations need the num threads information that cannot be derived when needed (like just getting the number of iterations of the loop)

Not immediately as this is out of my immediate working memory but TL;DR:
- logic to reinfer number of threads from IR and in particular 1-trip count loop etc is extremely brittle as soon as things don't trivially divide. IIRC it may be quite brittle even in cases that things divide but do not quote me on this
- tile dynamic sized ops by dynamic adaptive tile sizes and things become even more hairy
- rediscorvering a static number of threads after we threw the information away is not worth it

https://github.com/llvm/llvm-project/pull/91878