Cast std::max args to Dtype

[eelstork]

This is to avoid ambiguous literals. Consistent with similar idioms in Caffe source. Fixes errors with MSVC compiler.

[seanbell]

You can achieve the same effect more cleanly with std::max<Dtype>(a, b) instead of std::max(Dtype(a), Dtype(a)).

[eelstork]

@seanbell I am aware of this (@willyd and others fixed it this way). Is it cleaner, or only more concise?

[seanbell]

I personally think it's more clear/clean, since max<T> is saying "use the max function for type T." I started using it after I ran into the problem that abs can sometimes lead to unintentional casting to integers (I needed fabs), but abs<T> would either do the right thing or generate an error (depending on which abs definition is included). We're talking about a different function (max), but that's why I like explicitly specifying which template is used via function<T>.

Anyway, that's just my opinion. Both versions work.

[longjon]

I agree with @seanbell here... in addition, it's best to avoid C-style casting (I can see an argument being made for literals, but we are not only casting literals here).

[shelhamer]

Agreed with @seanbell and @longjon. Please revise for merge. Thanks for the the compatibility check.

[eelstork]

Updated to use suggested form.

[ronghanghu]

Thanks @eelstork !