AMD x86 Avoid Address Generation Interlocks, Always Inline Functions if Called from One Site

AMD Athlon™ Processor x86 Code Optimization 22007E/0 — November 1999

Always Inline Functions if Called from One Site

A function should always be inlined if it can be established that it is called from just one site in the code. For the C language, determination of this characteristic is made easier if functions are explicitly declared static unless they require external linkage. This case occurs quite frequently, as functionality that could be concentrated in a single large function is split across multiple small functions for improved maintainability and readability.

Always Inline Functions with Fewer than 25 Machine Instructions

In addition, functions that create fewer than 25 machine instructions once inlined should always be inlined because it is likely that the function call overhead is close to or more than the time spent executing the function body. For large functions, the benefits of reduced function call overhead gives diminishing returns. Therefore, a function that results in the insertion of more than 500 machine instructions at the call site should probably not be inlined. Some larger functions might consist of multiple, relatively short paths that are negatively affected by function overhead. In such a case, it can be advantageous to inline larger functions. Profiling information is the best guide in determining whether to inline such large functions.

Avoid Address Generation Interlocks

Loads and stores are scheduled by the AMD Athlon processor to access the data cache in program order. Newer loads and stores with their addresses calculated can be blocked by older loads and stores whose addresses are not yet calculated – this is known as an address generation interlock. Therefore, it is advantageous to schedule loads and stores that can calculate their addresses quickly, ahead of loads and stores that require the resolution of a long dependency chain in order to generate their addresses. Consider the following code examples.

72 Avoid Address Generation Interlocks