Is there a way to check accuracy of computation results from linalg operations?

I couldn’t find any way to check if lowered linalg operations (or any other operations) produce correct results, is there any built-in system to check if computation results correct? How can I confirm that my optimization still produce the same result as original operation?

I think you are probably looking for these integration tests: llvm-project/mlir/test/Integration/Dialect/Linalg/CPU at main · llvm/llvm-project · GitHub

You can write your own tests.

1 Like