Performance Optimization

Relevant source files

• docs/CHANGELOG.txt
• imgui.cpp
• imgui.h
• imgui_demo.cpp
• imgui_draw.cpp
• imgui_internal.h
• imgui_tables.cpp
• imgui_widgets.cpp

Purpose and Scope

This document covers performance optimization techniques and considerations for Dear ImGui applications. It explains how the immediate-mode architecture affects performance, identifies common bottlenecks, and provides practical optimization strategies at the widget, rendering, and system levels.

For integration and backend setup, see Getting Started. For rendering architecture details, see Draw System and Rendering Pipeline. For font-specific optimizations, see Font System.

---

Performance Philosophy

Dear ImGui is designed with performance as a core principle. The immediate-mode paradigm requires UI to be regenerated every frame, so the library is architected to minimize overhead:

• No malloc/free in typical idle frames - Memory is pre-allocated and reused
• O(1) or O(N) algorithms preferred - Searches and scans are minimized
• Debug build performance matters - Not just release builds are optimized
• Zero-cost when not used - Collapsed windows and clipped items have minimal cost

Sources: imgui.cpp220-224

---

Frame Performance Pipeline

Understanding where time is spent during a frame helps identify optimization opportunities:

Frame Execution Flow with Key Functions

CPU Cost Distribution (typical application):

• Widget submission code: 30-50%
• Layout and item logic: 20-30%
• Draw command generation: 15-25%
• Backend rendering overhead: 5-15%

Key Functions to Profile:

• ImGui::NewFrame() - Input processing and per-frame setup
• ImGui::Begin() / ImGui::End() - Window management
• ButtonBehavior() / InputTextEx() - Widget interaction logic
• ImDrawList::AddRect() / AddText() - Draw command generation
• ImGui::Render() / ImGui::EndFrame() - Finalization
• Backend RenderDrawData() - GPU submission

Sources: imgui.cpp5270-5388 imgui.cpp5432-5511 imgui_internal.h3397-3450

---

Measurement and Profiling

Built-in Metrics

Use ImGui::ShowMetricsWindow() to display real-time performance metrics:

The metrics window shows:

• Active windows count
• Active widgets count
• Vertices/indices generated
• Draw calls required
• Memory allocations

Sources: imgui.cpp98-101 imgui_demo.cpp408

Frame Time Measurement

Sources: imgui.h2491-2563

External Profilers

Dear ImGui integrates well with standard profilers:

• Visual Studio Profiler
• Instruments (macOS)
• Tracy Profiler
• Optick

Key functions to profile:

• ImGui::NewFrame()
• ImGui::Render()
• Individual widget functions (if hot)
• Backend RenderDrawData() implementation

Sources: imgui.cpp5270-5388

---

Widget-Level Optimizations

Window SkipItems Optimization

When a window is collapsed, minimized, or fully clipped, window->SkipItems is set to true, causing widgets to early-exit with minimal overhead:

SkipItems Flow in Begin() and Widgets

Implementation Pattern (from actual widgets):

Performance Impact:

• Collapsed window with 100 widgets: < 0.01ms (vs ~1ms if fully rendered)
• Clipped window: Similar early exit behavior
• No memory allocations in SkipItems path

Sources: imgui.cpp6809-6865 imgui_widgets.cpp161-163 imgui_widgets.cpp554-669

Text Rendering Optimization

TextEx() implements automatic coarse clipping for long text content to avoid processing text outside the visible viewport:

Long Text Clipping Logic

Implementation Details (from TextEx()):

Performance Impact:

• 10,000 line text with 20 visible lines: 500x faster (2ms → 0.004ms)
• Uses memchr() for fast newline scanning
• Only calls CalcTextSize() for visible lines
• No difference in visual output

Note: Flag ImGuiTextFlags_NoWidthForLargeClippedText can skip width calculation for clipped lines, saving additional time.

Sources: imgui_widgets.cpp160-266

---

Draw Call and Rendering Optimization

Draw Command Batching

Dear ImGui automatically batches primitives into minimal draw calls. Draw commands are only split when necessary:

Minimize draw calls by:

• Using atlas textures (fonts are automatically atlased)
• Avoiding ImGui::PushClipRect() when not needed
• Minimizing custom draw callbacks
• Using fewer unique textures

Sources: imgui_draw.cpp504-665 imgui.h3018-3044

Vertex Buffer Usage

Dear ImGui generates indexed triangle lists with shared vertices:

Primitive Type	Vertices	Indices	Efficiency
Rectangle (filled)	4	6	Optimal
Rectangle (outline)	8	24	Good
Circle/Arc	N+2	3*N	Variable
Text character	4	6	Optimal

Optimization tips:

• Prefer filled primitives over outlines when possible
• Reduce circle segment counts: ImGui::GetStyle().CircleTessellationMaxError
• Anti-aliasing adds extra vertices - disable if not needed

Sources: imgui_draw.cpp1275-1540

---

Memory Management Optimization

Memory Allocation Strategy

Dear ImGui aims for zero malloc/free in idle frames by pre-allocating and reusing memory pools:

Memory Pools and Allocation Patterns

Key Data Structures and Allocation Patterns:

Structure	Location	Allocation Pattern	Lifecycle
ImGuiContext	GImGui global	Once at CreateContext()	Until DestroyContext()
ImGuiWindow	g.Windows[]	Once per window name, pooled	Persists across frames
ImDrawList	window->DrawList	One per window	Persists, buffers grow
ImDrawCmd	drawlist->CmdBuffer	Reused vector capacity	Cleared per frame
ImDrawVert	drawlist->VtxBuffer	Grows, keeps capacity	Cleared per frame
ImGuiTable	g.Tables pool	Pooled by ID via ImPool<>	Reused across frames
ImGuiStorage	window->StateStorage	Grows as needed	Persists
ImGuiWindowTempData	window->DC	Stack member	Per-frame state
ImGuiTableTempData	g.TablesTempData[]	Temporary vector	Stacked per nested table

Critical Functions for Memory Management:

Memory Allocation Locations:

1. Startup Phase (CreateContext):

   • ImGuiContext allocation
   • Initial g.Windows vector capacity
   • Default font atlas

2. First Frame (NewFrame → Render):

   • First window creation → ImGuiWindow allocated
   • First draw commands → VtxBuffer/IdxBuffer grow
   • First table → ImGuiTable allocated from pool

3. Steady State (Subsequent Frames):

   • No allocations if UI remains stable
   • Allocations only on:
     • New windows created
     • Buffer growth (more vertices than previous max)
     • New tables or columns

4. Shutdown (DestroyContext):

   • Free all pools and buffers

Monitoring Memory Usage:

Sources: imgui.cpp4131-4153 imgui_internal.h2252-2387 imgui_internal.h2813-2989 imgui_draw.cpp446-458

Reducing Allocations

Sources: imgui.h2096-2166

---

Large List Optimization with ImGuiListClipper

For lists with thousands of items, ImGuiListClipper automatically calculates which items are visible and skips rendering the rest:

ImGuiListClipper Usage Pattern

Performance Impact by List Size

Total Items	Without Clipper	With Clipper	Speedup
100 items	0.5 ms	0.15 ms	3x
1,000 items	5 ms	0.15 ms	33x
10,000 items	50 ms	0.15 ms	333x
100,000 items	500 ms	0.15 ms	3333x

Assumes ~20 items visible on screen

ImGuiListClipper Internal Flow

Advanced Usage

Table Integration

Key Clipper Members:

• DisplayStart / DisplayEnd - Range of visible items to render
• ItemsCount - Total number of items
• ItemsHeight - Calculated height per item
• StartPosY / StartSeekOffsetY - Internal seek position tracking

Requirements and Limitations:

• Items must have uniform height by default
• For variable height, use IncludeItemsByIndex() to force-include specific items
• First Step() call measures the first item to determine height
• Compatible with BeginTable() / TableNextRow() automatically
• Not compatible with Columns() (legacy API)

Sources: imgui.h2393-2446 imgui.cpp8962-9123 imgui_demo.cpp2740-2850

---

Table Performance Optimization

Column Freezing and Scrolling

Frozen columns and rows have minimal overhead:

Automatic optimizations:

• Cells outside viewport are clipped
• Compatible with ImGuiListClipper for row clipping
• Frozen cells cached for reuse

Sources: imgui_tables.cpp49-71 imgui_tables.cpp743-877

Table Column Visibility

Disable unused columns:

Sources: imgui_tables.cpp1066-1179

---

Font and Text Rendering Performance

Dynamic Font System (v1.92+)

The dynamic font system (when backend supports ImGuiBackendFlags_RendererHasTextures) provides:

• On-demand glyph loading (no upfront cost for unused characters)
• Automatic atlas resizing
• Multiple font sizes without pre-baking

Performance characteristics:

• First use of new glyph: ~0.1ms (one-time cost)
• Subsequent uses: Zero additional cost
• Atlas updates: Handled incrementally

Sources: imgui_draw.cpp16-23 imgui.h35

Text Rendering Cost

Text Length	Typical Frame Cost	Optimization
< 100 chars	< 0.01 ms	None needed
100-1000 chars	0.01-0.1 ms	Consider clipping
1000-10000 chars	0.1-1 ms	Use clipper or manual clipping
> 10000 chars	> 1 ms	Use InputTextMultiline with ImGuiInputTextFlags_ReadOnly

Font Atlas Considerations

Pre-v1.92 (or without modern backend):

V1.92+ with modern backend:

Sources: imgui_draw.cpp2686-2949 docs/FONTS.md74-85

---

Common Performance Anti-Patterns

Anti-Pattern 1: Ignoring Begin() Return Value

Problem: Expensive logic executes even when window is collapsed or invisible.

Impact: Can waste 1-10ms per frame for expensive UI logic in collapsed windows.

Sources: imgui.cpp6809-6865 imgui.h652-657

Anti-Pattern 2: Per-Frame Allocations

Problem: Allocating memory every frame causes performance spikes and fragmentation.

Sources: imgui.cpp220-224

Anti-Pattern 3: String Concatenation in Loops

Problem: Repeated string operations cause O(N²) reallocations.

Impact: 1000 lines can go from 50ms to 0.5ms with proper buffering.

Sources: imgui.h2167-2180

Anti-Pattern 4: Per-Frame Texture Upload

Problem: Creating/destroying GPU textures every frame is extremely expensive.

Impact: Texture creation can cost 10-100ms depending on size and API. Caching provides 100-1000x speedup.

Sources: imgui.h1068-1083

Anti-Pattern 5: Not Using ImGuiListClipper

Problem: Rendering thousands of items even when only 20 are visible.

Sources: imgui.h2393-2446 imgui.cpp8962-9123

Anti-Pattern 6: Excessive PushID/PopID

Problem: ID stack operations add overhead when called thousands of times unnecessarily.

Sources: imgui.cpp8514-8576

---

Anti-Pattern Summary Table

Anti-Pattern	Cost	Fix	Speedup
Ignoring Begin() return	1-10 ms	Check return value	10-100x
Per-frame allocations	0.1-1 ms	Use static variables	10-100x
String concatenation in loops	10-50 ms	Use ImGuiTextBuffer	100x
Per-frame texture creation	10-100 ms	Cache textures	100-1000x
Not using clipper for large lists	10-500 ms	Use ImGuiListClipper	100-1000x
Excessive PushID for non-interactive	0.5-2 ms	Only ID interactive widgets	2-5x

---

Debug and Profiling Tools

Metrics Window Breakdown

Key sections:

Section	Information	Optimization Use
Dear ImGui	Frame rate, active windows	Overall health check
Windows	Individual window stats	Find heavy windows
DrawLists	Vertices, indices, commands	Identify draw call issues
Memory	Allocations, pool sizes	Memory leak detection
Viewports	Multi-viewport stats	Platform window overhead

Style Visualizer

Adjust visual settings that affect performance:

• CircleTessellationMaxError: Higher = fewer vertices
• CurveTessellationTol: Higher = fewer vertices
• AntiAlias* flags: Disable for performance gain

Sources: imgui_demo.cpp408-409

Custom Performance Markers

Sources: imgui.cpp13098-13201

---

Platform-Specific Considerations

High DPI Displays

High DPI rendering increases vertex counts proportionally:

Sources: imgui.h2521-2528 docs/FAQ.md90-93

Mobile Platforms

Additional considerations for mobile:

• Reduce circle/arc tessellation further
• Minimize texture switching (single atlas preferred)
• Disable anti-aliasing for low-power devices
• Use simpler widget variants (e.g., Selectable instead of Button with complex content)

Sources: imgui_internal.h267

---

Optimization Decision Tree

Sources: Multiple sections above

---

Summary: Performance Budget Guidelines

Typical performance targets (desktop, 60 FPS):

Component	Budget	Notes
Dear ImGui total	< 2-3 ms	Including draw generation
Widget submission	< 1-2 ms	Application code + ImGui logic
Draw command generation	< 0.5 ms	ImDrawList operations
Backend rendering	< 0.5 ms	GPU upload + draw calls

When to optimize:

• Dear ImGui consistently uses > 3ms per frame
• Frame time spikes above 16ms (60 FPS)
• Profiler shows ImGui as top CPU user

What usually works:

• Add ImGuiListClipper for lists > 100 items
• Check Begin() return values
• Cache expensive computations
• Minimize unique textures

What rarely helps:

• Micro-optimizing individual widget calls
• Complex custom drawing instead of built-in widgets
• Premature optimization without profiling

Sources: imgui.cpp220-224 imgui_demo.cpp9700-9800