Repository navigation
LEGO Horizon Adventures (PPSA14632) #6952
THEMANnt
started this conversation in
Game bring-up
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Existing thread
Title
LEGO Horizon Adventures (PPSA14632)
Content version tested
01.004.000
AnyPS5 commit
main d92a8da + PR #6277 (fe6d2d2) + unpublished general guarded spill-recovery prototype, local runtime commit 8b0edd9a56eda837bc84bd4f7165617814a19a93. This is a modified local build; PR #6277 alone did not clear the blocker.
System tested
Windows 11 Pro (build 26200); AMD Ryzen 7 7800X3D; NVIDIA GeForce RTX 5070 Ti, driver 32.0.16.1692; ~31 GiB usable RAM. Release build, bundled MinGW GCC 15.2.
Furthest point
In game
What is shown
Tested 2026-10-16. Reached the opening cinematic and first gameplay scene. Player and camera movement were confirmed using a five-second left-stick input. The gameplay counter showed approximately 1.3–1.6 FPS. Character/body parts were missing, and captured colors appeared washed out; the rendering cause has not been established.
The run lasted 583 seconds and was deliberately stopped after movement verification. No sampler runtime-value error, lane-spill mismatch, fatal error, driver failure, or presentation failure was logged during this run. Audio, combat, save reload, later areas, and long-term stability remain unverified. This is early bring-up, not a playable/stable compatibility result.
Current blocker
Unmodified main d92a8da reaches the menu, New Game save-slot selection and difficulty selection, then aborts while loading the opening level, before gameplay:
Related work: PR #6277. All four captured compute requests still fail with that PR alone. The additional unpublished prototype validates recovered spill values against the GPU lane read; only libSceAgcDriver.prx differed from the baseline runtime. No title-specific metadata override was used. The prototype is not an upstream fix or a submitted PR.
A synthetic counterexample and technical follow-up are on PR #6277; implementation discussion stays there. Remaining bring-up blockers after the local prototype are very low gameplay frame rate and incomplete rendering. No cause or fix for those is established.
AI-assisted investigation and report: yes.
How it was run
Selected Performance graphics mode, New Game / first empty save slot, and default Scout difficulty. Launched the converted title from its runtime directory and supplied menu/movement input through the documented controller-file interface. The runtime used the baseline system libraries except for the locally rebuilt libSceAgcDriver.prx described above. No runtime metadata override. Testing was stopped manually after the first gameplay movement check.
All reactions