Add Adreno Neural Fusion (ANF) support for Snapdragon Gen 6 — GameHub, Winlator and Droidspaces

[lejan-sudo]

I would like to propose investigating support for Qualcomm's new Adreno Neural Fusion (ANF) as an additional upscaling / frame-generation backend for OptiScaler.

ANF is Qualcomm's new AI-based rendering technology introduced with the Snapdragon 8 Elite Extreme Gen 6. It provides both:

• Neural Super Resolution (SR)
• Neural Frame Generation (FG)

Unlike a traditional CPU-side or external AI solution, ANF runs these workloads as GPU compute dispatches directly on the Adreno GPU.

This could be particularly interesting for the growing Snapdragon/Adreno gaming ecosystem.

-Why OptiScaler?

OptiScaler already acts as a middleware layer between the game's upscaling API and the actual upscaling/FG implementation:

Game → OptiScaler → Upscaler / FG backend

ANF could potentially be implemented as another output backend, in a similar conceptual way to FSR, XeSS or DLSS.

-Snapdragon / Adreno use cases

The most interesting applications would be:

• Windows games running through Winlator
• Games running through GameHub
• Windows games running through Droidspaces
• Linux/ARM64 gaming environments using Snapdragon Adreno GPUs
• Native ARM64 games using Adreno GPUs

This could be especially valuable because Snapdragon devices are becoming powerful enough to run demanding PC games, but rendering at native resolution can still be very expensive from a power and thermal perspective.

For example:

1080p output
→ render internally at 540p/720p
→ ANF Super Resolution
→ 1080p output

And potentially:

30 FPS real frames
→ ANF Frame Generation
→ 60 FPS displayed frames

This could significantly improve performance-per-watt on Snapdragon devices.

Important question

I am not suggesting that ANF support is necessarily straightforward.

The main question is whether OptiScaler could interface with the Adreno Neural Fusion SDK/API and expose ANF as an OptiScaler output backend.

It would be interesting to investigate:

1. Whether ANF can be initialized independently of a native Snapdragon game integration.
2. Whether the ANF SDK can be called from a Windows game running through Winlator/Wine.
3. Whether it can work through Vulkan or another API exposed by Turnip.
4. Whether ANF can be accessed from Linux/ARM64 environments such as Droidspaces.
5. Whether the required neural models/runtime can be packaged with OptiScaler.
6. Whether the required motion vectors, depth buffer and jitter information can be obtained from OptiScaler's existing input interfaces.
7. Whether ANF Frame Generation could potentially be integrated with OptiScaler's existing FG/HUDFix infrastructure.

Why this could be important

Snapdragon 8 Elite Extreme Gen 6 introduces a new Adreno architecture specifically designed around neural rendering, and ANF is supported starting with this generation.

This could give OptiScaler users on Snapdragon/Adreno hardware access to Qualcomm's native neural rendering hardware instead of relying exclusively on generic FSR/XeSS solutions.

It would also make OptiScaler much more relevant to the emerging ARM64 PC gaming ecosystem.

I would be very interested in seeing ANF support investigated, even if the first implementation has to be experimental or limited to Super Resolution before Frame Generation.

Would this be technically feasible within OptiScaler's current architecture?

There is already an existing discussion about integrating alternative trained models into the OptiScaler rendering pipeline (#526). ANF could be a particularly interesting real-world implementation of this concept because it is designed specifically for Qualcomm's Adreno neural hardware.

---Additional use case: Android gaming handhelds---

I also think ANF support could be particularly interesting for Android gaming handhelds such as AYN Odin and AYANEO Pocket devices.
These devices are increasingly being used to run Windows PC games through Winlator, GameHub and similar solutions, and many of them use Qualcomm Snapdragon/Adreno GPUs.
Supporting ANF through OptiScaler could therefore benefit not only smartphones, but also:

• AYN Odin handhelds
• AYANEO Pocket handhelds
• Other Snapdragon/Adreno Android gaming consoles
• Future handhelds using Snapdragon 8 Elite Extreme Gen 6 / Adreno A850
  This could make OptiScaler an interesting common rendering layer across the growing Snapdragon ARM64 gaming ecosystem, rather than limiting ANF support to smartphones.

In particular, a Snapdragon handheld running Windows games through Wine/Winlator could potentially use ANF as a hardware-native Qualcomm SR/FG solution, instead of relying exclusively on FSR/XeSS.