OpenGL Performance Candidates

OpenGL Performance Candidates

Lifecycle: Performance - Unproven
Reviewed source: 322986f4c5502b3fa887a9605ac8cd88e9b2c579
State: The source paths are verified. Their performance value and compatibility remain unproven because no OpenGL benchmark was run.

The OpenGL backend already has important optimizations: native S3TC uploads, module/program caches, persistent program binaries, glMultiDrawArrays(), buffer orphaning for inline arrays, and partial texture-parameter caching. Those are not new candidates.

This directory covers work that remains around depth generation, cross-context synchronization, uniform uploads, surface readback, geometry streaming, surface-to-texture handling, state application, queries, PVIDEO, and cold shader compilation.

Candidate directory

Candidate	Mechanism type	Main work under test	Technical reference
Native depth variant	Numerical/early-depth experiment	Custom fragment-depth writes and possible late fragment rejection	Open full page →
Asynchronous display handoff	Host synchronization architecture	CPU waits between render and display contexts	Open full page →
Completed display reuse	Retained-result shortcut	Repeated display composition when all output inputs are unchanged	Open full page →
Uniform epoch tracking	Prepared-state shortcut	Rebuilding and uploading unchanged active uniforms	Open full page →
GPU-downscaled readback	Transfer-reduction optimization	Downloading scaled pixels that the CPU immediately discards	Open full page →
Persistent mapped geometry streaming	Streaming-buffer architecture	Repeated orphan/subdata uploads and opaque driver recycling	Open full page →
Copy-image surface transfer	Exact copy subset	FBO setup, clear, shader draw, and state disturbance for compatible copies	Open full page →
Direct surface binding	Copy-bypass shortcut	Materializing another texture when the surface itself can be sampled safely	Open full page →
Sampler objects	State separation/cache	Reapplying sampling state to texture objects	Open full page →
Render-state shadow	State-call suppression	Reissuing unchanged blend/depth/stencil/raster state every draw	Open full page →
Deferred pooled occlusion queries	Query-lifetime optimization	Per-draw query allocation, blocking result retrieval, and deletion	Open full page →
PVIDEO GPU conversion	Resource reuse/GPU conversion	Repeated RGBA allocation, CPU conversion, and full texture redefinition	Open full page →
Cold shader stall avoidance	Asynchronous/alternate execution	Synchronous compile/link/validation on first-seen states	Open full page →

Classification

flowchart TD
    A[OpenGL source-backed idea] --> B{Already implemented?}
    B -->|Yes| C[Document under Implemented; do not duplicate]
    B -->|No| D{Runtime evidence exists?}
    D -->|No| E[Performance - Unproven]
    D -->|Yes, not in main| F[Performance - Researched]
    F -->|Lands in main| G[Performance - Implemented]

Loading

Shared evaluation rule

A successful experiment must prove all of the following:

1. The intended route was reached often enough to matter.
2. The expensive work actually disappeared rather than moving elsewhere.
3. Guest-visible output, ordering, reports, and lifecycle behavior remained within the declared contract.
4. An uninstrumented Release build improved fixed guest progress or tail latency on a representative workload.
5. The reference route remained available for unsupported capabilities or failed invariants.

Suggested order

The sharply bounded items should be tested first:

1. Uniform epoch tracking.
2. GPU-downscaled surface readback.
3. Copy-image surface transfer.
4. PVIDEO texture reuse and GPU conversion.
5. Query pooling/deferred retrieval.

The higher-risk architectural work follows after route measurements:

1. Native depth variant.
2. Asynchronous display handoff and completed-output reuse.
3. Persistent mapped streaming.
4. Direct surface binding.
5. Render-state shadow.
6. Cold shader stall avoidance.

Related

• Performance - Unproven
• OpenGL Renderer
• OpenGL vs Vulkan
• Performance Methodology
• Debugging Lag & Hot-Path Analysis
• Debugging Tools & Commands