On Wed Apr 12 11:20:44 2023 +0000, Giovanni Mascellani wrote:
As far as I can tell we're not yet taking the logarithm of a negative number, and it's not completely clear to me why we cannot do that in a test constrained to SM4+. But I am not sure it's a good idea to waste more time on this, so I'm approving.
For SM4+ it will results in nans, for older versions it will abs() input. None of that behavior is compiled in as a part of intrinsic.