Go 1.21 update introduces Profile-Guided Optimization as a fully developed feature

The Go (Programming Language) 1.21 update (released a few months ago) has officially incorporated Profile-Guided Optimization (PGO) as a fully developed feature, following its beta release in Go 1.20. This integration comes after improvements were made to PGO, based on community feedback and solutions provided for identified limitations during its preview stage.

PGO is a technique that utilizes application behavior profiles to guide the Go compiler's optimization choices. As a Go binary is built, the compiler uses a variety of optimizations to improve the binary's performance, including constant propagation, escape analysis, inlining, and devirtualization. These optimizations help reduce runtime evaluation costs, garbage collection overhead, and facilitate further optimizations in the caller.

Incorporating compiler optimizations into Go releases is a challenging task due to the potential for overly aggressive optimizations to adversely impact performance or prolong build times. PGO addresses this problem by enabling the compiler to make more informed decisions based on profiling data, leading to the optimization of frequently used functions and a more precise selection of common cases. PGO is also referred to as Feedback-Directed Optimization (FDO) in some scenarios.

We highly recommend that you check the profile-guided optimization user guide shared by the Go team, as well as the official announcement, where you will find a fully detailed example of using PGO to improve the performance of an application, including the entire setup, profiling, evaluation, and more.