[Arith] Allow const folding on fp16 involving one and zero

[masahi]

Since #12515, const folding on fp16, when both lhs and rhs are constant, is disabled. This is the case even when one of the arguments are one or zero. This leads to a real problem described in #13532 (comment). So in practice, compiling winograd fp16 on GPU is currently broken.

This PR enables trivial algebraic simplification for fp16 const folding when (1) both arguments are constant and (2) one or both of the inputs are one or zero. Since we already do such simplification for any dtype when only one of the inputs are constant (for example

tvm/src/arith/const_fold.h

Lines 209 to 216 in 3a639a4

	if (fa) {
	if (fa->value == 1) return b;
	if (fa->value == 0) return a;
	}
	if (fb) {
	if (fb->value == 1) return a;
	if (fb->value == 0) return b;
	}

), I just removed all early exit return NullOpt for fp16 cases and use the code path like above to simplify 0h * 2.5h to 0h etc.

cc @vinx13 @wrongtest-intellif @Lunderberg

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @junrushao _{See #10317 for details}

_{Generated by tvm-bot}

[leandron]

LGTM, thanks @masahi