Pros and cons of 'inline'

Question

First of all, I would like to state the facts I know about 'inline', so that you don't bother to restate them.

1. An inline function is a special kind of function whose definition must be available in every translation unit in which the function is used.
2. It is a hint to the compiler (which it is free to ignore) to omit the function call, and expand the body instead of the call.
3. The only pro I know of is that (2.) may make the code faster.
4. The only con I know if is that (1.) increases coupling which is bad.

Now let's consider templates. If I have a template library, I need to provide the definitions of the function templates in every translation unit, right? Let's forget about the controversial 'export' for a while, since it doesn't really solve the problem anyway. So, I come to a conclusion that there is no reason not to make a template function inline because the only con of inline I know of is there a priori.

Please correct me if I am wrong. Thanks in advance.

Answer 1

The only pro I know of is that (2.) may make the code faster.

May being the operative word. Inlined functions may make certain code paths faster, yes.

But an inlined function puts additional pressure on the instruction cache on most modern CPUs. If your function is too large to fit in the L1 instruction cache, it may actually run slower than performing a function call (for which the CPU can optimize by prefetching the function and its return site).

Inlining a function may also put undue pressure on the L2 cache - if an inlined function is used an unusually large number of times, the extra code size will increase the likelihood of cache misses, leading to long delays and pipeline stalls as the CPU twiddles its thumbs waiting for the memory bus to do something.

inline is far from being a silver bullet. Compilers with aggressive optimization will ignore the inline hint completely, as they will instead choose functions to inline based on heuristics such as code size or the presence of branches, irrespective of the presence or absence of the inline keyword.

The only con I know if is that (1.) increases coupling which is bad.

This is something I've never heard. "Coupling" is a concept I've only heard used when describing the high-level relationships of code. It's more an issue of maintainability and generality of code. inline is an issue of low-level code generation.

As to templates, again, an aggressively-optimizing compiler will inline if its heuristics show an advantage to doing so.

There is, however, a link-level issue to consider: You may need to declare a function or template inline (or static, depending on the situation) in order to eliminate duplicate symbols at link time or restrict symbol visibility. This is, of course, not an optimization.

In summary, don't bother using the inline keyword except when specifically required.

Answer 2

I think you shouldn't bother with writing inline or not before template function definitions. I think most compilers will inline small functions anyway, whether you ask them or not. Some compilers are even able to do this at link time for functions defined in only one translation unit.

Imho, the only point to the inline keyword is to be able to define non-template functions inside headers, for those compilers who don't inline at link time (or when you want a "header only" library).