I'll assume that by "C++ file" you mean ".CPP" file.  The .H file is just as
much C++.

Inline function bodies have to be visible to the compiler in every
translation unit where they're used.  The most common way to ensure that is
to simply put the inline definitions in the same header file where they're
declared.

If a function has been declared as inline but the definition is not visible
to the compiler, the compiler simply cannot continue:  since the function is
inline, the compiler can't generate a call to an out of line funciton,
neither can it make something up (of course).

If a function is used in only one .CPP file, that function's definition can
of course be placed in that .CPP file.  The disadvantage to doing that, of
course, is that if you subsequently want to use that function in another
.CPP file, you'll have to move the definition.  The wrong solution is to put
duplicate definitions of the function into two files.  That's just asking
for trouble down the line when someone modifies one copy of the function and
not another.  It's also a violation of what's known as the "One Definition
Rule".

-cd

A function, member of a class or not, should be written as inline when
it is small and called frequently and so we can avoid the time cost of
continuous function calls with minimal impact on the produced binary size.

Now VS 2003 unfortunately creates everything as inlined, basically
ignoring the optimisation abilities that C++ provides making it to be
like VC# 2003.

Under this view, any such large inlined function that is called more
than once, produces significant binary size impact in comparison to the
non-inlined definition, and thus should be made a regular definition
(that is non-inlined) by hand.