[Mlir-commits] [mlir] [mlir][arith][transforms] Fix f4E2M1FN to f32 cast (PR #160121)

[Jorn Tuyls]

jtuyls wrote:

@kuhar @krzysz00 There was one more issue with the -0.0 (f4) value, it was casted to -0.5 f32. I added a fix and test for that as well now: https://github.com/llvm/llvm-project/pull/160121/files#diff-4a23dbc7eb7e7f10e00c4e31ebb2f21359a4cceffc542169a975e8cc9ceba332R412

Could you check again and help merge when it looks good?

https://github.com/llvm/llvm-project/pull/160121