All 19 CLR & Runtime questions
More in CLR & Runtime
What is boxing and unboxing, and what performance cost does it incur? How does the CLR handle exceptions at a low level, and what is the performance implication of throwing exceptions frequently? How does the CLR ensure type safety and memory safety, and what categories of bugs does this prevent compared to unmanaged languages like C++? How does Span<T> improve performance for buffer and array manipulation compared to traditional array slicing?
Advertisement
Back to results for “CLR & Runtime”
.NET › CLR & Runtime

How does the CLR's Dynamic Profile-Guided Optimization (Dynamic PGO) improve JIT-compiled code quality at runtime?

Advanced

Answer

Dynamic PGO (enabled by default since .NET 8) instruments Tier 0 code with lightweight counters tracking actual runtime behavior (which branches are taken, which types appear at a call site), then uses this real, observed profile data when generating the final optimized Tier 1 code -- producing better-optimized native code tailored to the application's actual runtime behavior than static, compile-time-only optimization heuristics could achieve, since real execution patterns often differ from what static analysis alone can predict.
<PropertyGroup>
  <TieredPGO>true</TieredPGO>  <!-- default: true in .NET 8+ -->
</PropertyGroup>

// The JIT observes, e.g., that a virtual call site almost always resolves
// to the same concrete type, and generates optimized code assuming that case
Real-world example A benchmark comparing .NET 6 (without Dynamic PGO) to .NET 8 (with it enabled by default) on the same polymorphism-heavy business logic shows measurable throughput improvement purely from the JIT's better runtime-informed optimization decisions.

Common follow-ups: What's the overhead of the instrumentation phase before Tier 1 recompilation?;How does this interact with Native AOT, which has no runtime profiling phase?

Diagnostics & Performance;CLR & Runtime
All 19 CLR & Runtime questions