Edit

Profile-Guided Optimization for desktop apps

Profile-Guided Optimization (PGO) is a compiler optimization technique that uses runtime profiling data to produce more efficient native code. When you use PGO, the compiler rearranges code based on how your app actually runs, resulting in better instruction cache utilization, smaller code size, and faster execution.

Important

Desktop Windows App SDK apps use the standard MSVC PGO workflow. Unlike UWP apps, desktop apps do not require manually copying pgort140.dll into the app package — the PGO runtime library is available through the standard C/C++ runtime.

How PGO works

PGO is a three-step process:

  1. Instrument — Build your app with instrumentation enabled. The compiler inserts probes that record how your code executes.
  2. Profile — Run the instrumented app and exercise your most important scenarios. The profiling data is saved to .pgc files.
  3. Optimize — Rebuild your app using the profiling data. The compiler optimizes code layout, inlining, and branching based on actual runtime behavior.

Step-by-step guide

1. Configure for Release

Set your solution configuration to Release (or another configuration that produces optimized code). PGO has no benefit when applied to debug builds.

2. Enable Whole Program Optimization

In your project properties, go to C/C++ > Optimization and verify that Whole Program Optimization (/GL) is enabled.

3. Instrument the build

Go to Linker > Optimization and set Link Time Code Generation to Profile Guided Optimization - Instrument (/LTCG:PGInstrument).

Build your solution. A .pgd file is generated alongside your build output.

4. Run your app and collect profile data

Run your app and exercise the scenarios you want to optimize — startup, navigation, data processing, or any performance-critical workflow. While the app runs, use the pgosweep.exe tool (included in the MSVC toolset) to collect profiling data:

pgosweep.exe MyApp.exe MyApp!Scenario1.pgc

You can collect multiple .pgc files for different scenarios.

5. Merge profile data (optional)

By default, all .pgc files placed alongside the .pgd file are merged automatically during the optimize step. To weight certain scenarios differently, use the pgomgr.exe tool:

pgomgr.exe -merge:3 MyApp!CoreScenario.pgc MyApp.pgd

This gives the CoreScenario run three times the priority of other runs.

6. Optimize the build

Go back to Linker > Optimization and set Link Time Code Generation to Profile Guided Optimization - Optimize (/LTCG:PGOptimize).

Rebuild the solution. The linker automatically merges all .pgc files and produces an optimized binary based on the profiling data.

Tips for effective PGO

  • Profile realistic scenarios. The quality of PGO optimization depends on how representative your profiling runs are. Exercise the code paths your users will actually hit.
  • Combine multiple runs. Collect profiling data from several different usage scenarios to cover a broad range of code paths.
  • Re-profile when code changes significantly. Profile data becomes stale when you make major changes to your codebase. Re-instrument and re-profile periodically.
  • Measure the impact. Compare startup time, throughput, and memory usage before and after PGO to verify the optimization is worthwhile.

Related content

  • Last updated on