Skip to content

IBL Optimization Strategies

This document outlines architectural ideas for reducing the performance impact of environment map updates, specifically the long GPU stalls during Prefiltered Specular Map generation.

1. Problem Analysis

Current metrics show that generating the Prefiltered Specular Map for a 4K environment takes ~350ms on integrated graphics. This causes a complete freeze of the rendering loop.

2. Proposed Solutions

A. Progressive IBL (Temporal Amortization)

Instead of calculating all mipmap levels in a single loop, process them incrementally. - Mechanism: Use a state machine in the AsyncLoader or App loop. - Workflow: Frame N: Load HDR -> Frame N+1: Mip 0 -> Frame N+2: Mip 1 ... - Result: Zero freezes. The user sees the environment "sharpen" or "blur" over a few frames.

B. Adaptive Sample Counting

Lower mip levels represent rougher surfaces where high-frequency details are already lost. - Idea: Use 1024 samples for Mip 0, but scale down to 512, 256, 128 for higher mips. - Technique: Pass u_sample_count as a uniform to the compute shader.

C. Resolution Capping

Directly processing a 4K texture for IBL is often overkill as reflexes are rarely pixel-perfect. - Rule: Cap the Prefiltered Specular Map resolution to 1024x512 regardless of the source HDR size. - Benefit: Quadrant reduction in complexity compared to 4K.

D. Tiled Dispatching

For very large mips, even a single glDispatchCompute can exceed a reasonable frame budget. - Idea: Split a single mipmap level into tiles (e.g., 4x4 grid). - Mechanism: Dispatch only a subset of tiles per frame until the level is complete.

E. Instant Placeholder (The "Fast Path")

Avoid waiting for any compute shader to show the new environment. - Workflow: 1. Load HDR to RAM. 2. Upload to VRAM. 3. glCopyImageSubData HDR to SpecMap Mip 0 (Instant). 4. Gradually replace mips as they are computed. - Result: Visual feedback is instantaneous.

3. Implementation Roadmap

  1. Implement Resolution Capping (Easiest, high ROI).
  2. Implement Progressive Mip Generation (Best user experience).
  3. Implement Adaptive Samples (Pure GPU optimization).