Mobile-DDGI: Lightweight Probe-Based Global Illumination via Adaptive Budget Allocation

Abstract

While Dynamic Diffuse Global Illumination (DDGI) and its extensions achieve efficient real-time global illumination on desktop hardware, their computational overhead remains a prohibitive bottleneck for resource-constrained mobile platforms. To address this limitation, we introduce an adaptive, probe-based framework that aggressively prioritizes updates within the camera frustum. Central to our approach is a novel probe importance metric that optimally distributes a strictly constrained mobile ray budget to the most perceptually significant probes. Evaluated across four distinct mobile ray tracing pipelines, our method significantly reduces the overall ray budget while robustly adapting to rapid radiometric and geometric changes without compromising visual fidelity.

Key Contributions

Camera-attached multiple probe volumes
Per-frame probe update budget
Probe importance metric

Core Idea

Mobile-DDGI combines three ideas:

Camera-attached cascaded probe volumes
Tracking window-based logical probe movement
Adaptive probe update budget allocation

The renderer maintains multiple DDGI probe volumes around the camera. Each volume has a different probe spacing, allowing dense indirect lighting near the viewer and coarser coverage farther away.

Only a limited number of probes are updated every frame. The update budget is assigned to probes that are expected to have the greatest visual impact.

Probe Volume Structure

Mobile-DDGI uses camera-attached cascaded DDGI volumes.

Property	Description
Volume type	Camera-attached cascaded probe volumes
Probe grid size	`16 x 8 x 16` probes per cascade
Cascade spacing	Each cascade doubles the spacing of the previous one
Probe movement	Tracking window / infinite scrolling-style logical index update

This structure keeps high-resolution probe coverage near the camera while avoiding full-scene dense probe allocation.

Adaptive Probe Update Algorithm

Mobile-DDGI updates probes in the following order:

Step	Operation	Description
1	Update newly active probes	Probes newly entering the active region due to camera movement are updated first.
2	Compute probe importance	Each probe receives an importance value based on usage, distance, and view-center proximity.
3	Sort probes by priority	Probes are sorted according to the computed importance value.
4	Update high-priority probes	A fixed portion of the remaining budget is assigned to visually important probes.
5	Round-robin update	The residual budget is used to update the remaining probes sequentially.

In the evaluated configuration, the update budget is set to 200 probes per 16 x 8 x 16 cascade. For four cascades, this means that only 800 probes are updated per frame out of 8,192 total probes.

Probe Importance Metric

Each probe is assigned an importance score $\rho$ :

\rho = w_{\mathrm{ref}} \cdot w_{\mathrm{dist}} \cdot w_{\mathrm{center}}

Term	Meaning	Purpose
$w_{\mathrm{ref}}$	Probe usage weight	Prioritizes probes contributing to the current indirect-lighting computation.
$w_{\mathrm{dist}}$	Distance weight	Assigns greater importance to probes closer to the camera.
$w_{\mathrm{center}}$	View-center weight	Assigns greater importance to probes closer to the screen center.

Probe Usage Weight

w_{\mathrm{ref}} = \begin{cases} 1, & \text{if probe is referenced}, \\[4pt] w_{\mathrm{ref}}^{\mathrm{min}}, & \text{otherwise}. \end{cases}

The weight $w_{\mathrm{ref}}$ prevents unused probes from being completely ignored while assigning higher priority to probes that directly contribute to shading.

Distance Weight

w_{\mathrm{dist}} = \begin{cases} 1, & \text{if } \lVert \mathbf{p}-\mathbf{c} \rVert < k, \\[4pt] \exp\left( -\left( \lVert \mathbf{p}-\mathbf{c} \rVert-k \right) \right), & \text{otherwise}. \end{cases}

where:

Symbol	Description
$\mathbf{p}$	Probe position
$\mathbf{c}$	Camera position
$k$	Distance threshold

Probes near the camera are more likely to affect the final image and are therefore updated more aggressively.

View-Center Weight

The projected screen-space position of probe is defined as

\mathbf{d} = \left( \frac{X_{\mathrm{clip}}}{W_{\mathrm{clip}}}, \frac{Y_{\mathrm{clip}}}{W_{\mathrm{clip}}} \right).

Its normalized distance from the center of the screen is

n_d = \operatorname{clamp} \left( \lVert \mathbf{d} \rVert, 0, 1 \right).

The view-center weight is then computed as

w_{\mathrm{center}} = \max \left( 1- \operatorname{smoothstep} \left( 0.7, 1, n_d \right), w_{\mathrm{center}}^{\mathrm{min}} \right).

The weight $w_{\mathrm{center}}$ assigns higher priority to probes projected near the center of the user’s view. This formulation also supports future gaze-aware or foveated-rendering extensions.

Probe State Machine

Mobile-DDGI uses a probe state machine to avoid unnecessary probe updates.

State	Description
`FRESHMAN`	Newly introduced probe that needs initialization.
`CANDIDATE`	Probe selected as a potential update target.
`UPDATE`	Probe selected for ray tracing and DDGI data update.
`INACTIVE`	Probe that is not actively updated in the current frame.

Rendering Pipeline

The Mobile-DDGI rendering pipeline is organized as follows:

Step	Stage	Description
1	Geometry pass	Render scene geometry and G-buffer data.
2	Check probe reference	Determine which probes are referenced during indirect lighting.
3	Priority sort	Sort probes using the adaptive importance metric.
4	Probe trace	Trace rays for selected probes.
5	Probe classification	Classify probes based on activity and update state.
6	Probe update	Update probe irradiance and distance data.
7	Indirect lighting pass	Evaluate indirect lighting from DDGI probe data.
8	Lighting pass	Combine direct lighting, indirect lighting, and material shading.

Summary

Mobile-DDGI reduces DDGI update cost on mobile GPUs by allocating a strict per-frame probe update budget to the most important probes.

The algorithm prioritizes:

Newly active probes caused by camera movement
Probes referenced by current indirect lighting
Probes close to the camera
Probes close to the center of the user’s view
Remaining probes through round-robin updates

This makes probe-based global illumination more suitable for mobile ray tracing environments with limited ray tracing throughput and strict performance constraints.