Profiling SerenityOS

Build and runtime profiling tools for performance analysis

SerenityOS provides multiple profiling approaches for analyzing both build performance and runtime behavior.

Build Profiling

There are three ways to get information about compile times:

1. Using time ninja install for overall timing
2. Reading Ninja’s log files for per-file compile times
3. Enabling GCC/Clang flags for detailed compiler pass breakdowns

Ninja Log Analysis

Ninja produces a log file with per-cpp-file compilation times, useful for identifying slow-to-compile files.

Prerequisites

• ninjatracing - Python script to convert Ninja logs
• python3 (the script expects python to be available)
• Make ninjatracing executable and available in your PATH

Usage

Generate build trace

Step 1: Clean the build

Clean build artifacts and ccache to ensure every file is compiled:

ninja clean
ccache --clear

Step 2: Build with Ninja

ninja

Step 3: Convert log to JSON

The log is written to .ninja_log in the current directory:

ninjatracing .ninja_log > trace.json

Step 4: Visualize

Drag and drop trace.json onto Speedscope or any compatible flamegraph visualizer.

Ninja trace analysis helps identify which files take the most time, making them good targets for optimization.

Compiler Time Reports

GCC -ftime-report

Adding the -ftime-report flag to GCC outputs a breakdown for each compiled file.

Enable GCC time reporting

Edit CMakeLists.txt in the Serenity root directory and add around line 220:

add_compile_options(-ftime-report)

Optionally add -ftime-report-details for more detail.

Example GCC output

Time variable                                   usr           sys          wall           GGC
 phase setup                        :   0.00 (  0%)   0.00 (  0%)   0.01 (  0%)  1326k (  2%)
 phase parsing                      :   0.57 ( 61%)   0.19 ( 83%)   1.63 ( 65%)    59M ( 74%)
 phase lang. deferred               :   0.10 ( 11%)   0.03 ( 13%)   0.30 ( 12%)  8761k ( 11%)
 phase opt and generate             :   0.23 ( 25%)   0.01 (  4%)   0.48 ( 19%)    10M ( 13%)
 |name lookup                       :   0.11 ( 12%)   0.01 (  4%)   0.25 ( 10%)  2004k (  2%)
 |overload resolution               :   0.08 (  9%)   0.00 (  0%)   0.26 ( 10%)  7900k ( 10%)
 template instantiation             :   0.27 ( 29%)   0.08 ( 35%)   0.89 ( 36%)    25M ( 32%)
 TOTAL                              :   0.93          0.23          2.50           79M

Compiler time reports are most useful if you understand compiler internals. For general use, Ninja log analysis is recommended.

Clang -ftime-report

Clang also supports -ftime-report flag with similar functionality.

Runtime Profiling

SerenityOS includes a built-in profiler for analyzing application and system performance.

The profile Command

The profile command records profiling information that can be viewed with ProfileViewer.

Basic Usage

Profile a specific process

# Profile a running process by PID
profile -p 42

# Profile a command
profile echo "Hello friends!"

System-wide profiling

# Enable whole-system profiling (requires superuser)
profile -ae

# ...do work...

# Stop profiling
profile -ad

Results are saved to /sys/kernel/profile.

Event-specific profiling

Track specific event types with the -t flag:

# Profile syscalls
profile -t syscall -- echo "Hello friends!"

# Profile page faults
profile -t page_fault -- ./my-app

# Profile context switches
profile -t context_switch -- ./my-app

Available event types:

• sample - Sampling events
• context_switch - Context switches
• page_fault - Page faults
• syscall - System calls
• read - Read operations
• kmalloc - Kernel memory allocations
• kfree - Kernel memory frees

Options

Option	Description
-p PID	Target specific process ID
-a	Profile all processes (super-user only)
-e	Enable profiling
-d	Disable profiling
-f	Free the profiling buffer
-w	Enable profiling and wait for user input to disable
-t event_type	Track specific event type

ProfileViewer

ProfileViewer is a GUI application for viewing profiling data produced by profile.

Launch it from the Applications menu or command line:

Profiler

It provides:

• Flame graphs of execution time
• Call tree analysis
• Timeline views
• Per-function statistics

Performance Tips

For accurate profiling results:

• Use release builds (-DCMAKE_BUILD_TYPE=Release)
• Disable debug macros that add logging overhead
• Profile on real hardware when possible, as VM performance varies

On Linux, QEMU is significantly faster with KVM. Ensure /dev/kvm exists and is readable/writable by your user.

See Also

• Debugging - Debugging tools and techniques
• Testing - Running and debugging tests
• strace - System call tracer