Skip to content

test(saros): publish current-main startup integration and remaining GPU fault - #73

Draft
JJenkx wants to merge 4 commits into
mainfrom
validation/saros-d92-startup
Draft

JJenkx wants to merge 4 commits into
mainfrom
validation/saros-d92-startup

Conversation

@JJenkx

@JJenkx JJenkx commented Oct 15, 2026 •

Copy link
Copy Markdown
Owner

What

Saros PPSA07631 v01.000.000 aborts on current main d92a8dad at sceAudioPropagationSourceGetAudioPath: index 0 out of range (0 audio paths). This draft publishes a focused startup integration of audio path state #56, AGC FS table #6160, and 32-bit atomic alignment #6094, preserving the component authors.

The integration remains unstable. A cold startup passed the former exception and rendered the Soltari intro logo at 240 seconds, reaching the planned 300-second stop. A warm run then failed around 100 seconds with GPU virtual-memory faults and Vulkan device loss. The kernel reset recovered the GPU. Title-menu progression and gameplay are unverified.

The component gaps still apply: active acoustic path rendering is unsupported; the FS table currently points to return-only placeholder routines; the console's atomic alignment mode remains unknown. These observations do not validate those assumptions.

Tested

Linux / Ryzen 7950X / Radeon RX 6650 XT, RADV Mesa 26.2.4. Public head fe380fe0c6905c8407ac73e72cf79eaa09567eb0 retains the official SPIRV-Tools pin. The native test head 05113b1e3f4b7813ad4a6e43e6a0e2d30f624a53 differs only by the common private validator-table teardown repair gitlink.

  • Full Linux Release project build passed on the test head.
  • Nine selected CTests passed with zero skips on Radeon (1.68 s) and Lavapipe (1.73 s): shader memory, unaligned atomic, four existing buffer-atomic cases, FS-table calls, scalar calls and audio propagation.
  • Same converted title tree, pipeline 1, no input: main plus the common teardown repair aborted at 70.25 s on the path getter (exit 134). The candidate cold run completed at the 300-second timeout (exit 124, 302.70 s elapsed); its warm run aborted at 100.28 s (exit 134), before the scheduled Cross input. Both restored the temporary kernel setting to 64 MiB and left no game process running.
  • The warm log reports A GPUVM fault has been detected, a read at 0xffff80363dda3000, and vkQueueSubmit recorder: Vulkan result -4. The game log identifies an instruction-fetch fault. The kernel log for the same Saros PID (1216766) records TCP texture-cache read faults before the instruction-fetch reports; the first recorded read is at 0x9d3d880000. This ordering does not identify the first failing guest instruction or prove a lifetime defect. No causal attribution to one component has been established.
  • Conventions, whitespace and project-source comparison passed. An independent build with the public vendor pin, Windows, title gameplay and performance are unverified.

The older Saros stack's ordered-append, runtime image-table, table-call and AJM ratio-ramp dependencies have merged into main. This branch omits its old private shader probes and superseded mouse workaround. No proprietary title assets, screenshots, shader requests or raw logs are committed.

Checklist

  • Draft integration for review and reproduction
  • Original component authors retained
  • Depends on: none (listed component implementations included)
  • Unblocks: none; GPU fault remains open
  • GPUs tested: RX 6650 XT / RADV and Lavapipe on Linux, scope above
  • AI-assisted: yes

JJenkx and others added 4 commits October 15, 2026 02:24
…atomics

The RDNA2 ISA (8.1.7) ignores the two low address bits of dword or wider
non-formatted accesses in alignment mode DWORD. MXGP 26 (PPSA30839) sets
bits with buffer_atomic_or at odd byte offsets, which recorded an
unaligned BDA fault and dropped the atomic. A 32-bit buffer atomic now
acts on the dword that holds its address; 64-bit atomics off an 8-byte
boundary still fault.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants