Publish Chimera GFX source

2026-09-03 03:27:14 +02:00
commit a6037502d7
828 changed files with 100454 additions and 0 deletions
@@ -0,0 +1,74 @@
+# Phase 1.0P VideoOut submit analysis
+
+Phase 1.0P is an offline postmortem of the consumed Phase-1.0O run. It adds no
+target source or artifact and performs no device action. The exact runtime
+evidence remains the M artifact (1,845,208 bytes, SHA-256
+`c99a0856309a357ad2667d89b4924e4063ad214cae09c8a419457b0732f583cd`)
+and the ignored O trace (18,031 bytes, SHA-256
+`3d0b8811ae11f5cac2c2d331e252e1789a10588cab0e6045828a6b6af0fe1eb6`).
+
+## Proven call boundary
+
+The exact linked SDL overlay source has SHA-256
+`9949a280fed40241746788a8c001280455bc629e0566a88f82f16a8634d43025`.
+Its linker map has SHA-256
+`6768ffc7267b9b362c3b953571e5dfeed1e70004a3e19f41ef500c3f26272719`
+and locates `PS5_VideoInit` at `0xfdc20` and
+`PS5_PresentEarlyDiagnosticFrame` at `0xfe560`.
+
+Disassembly of the exact M ELF proves that the indirect call at `0xfe7bc`
+loads the VideoOut handle into `edi`, zero into `esi`, one into `edx`, and
+zero into `ecx`. The corresponding GOT slot is `0x1779b0`. The code tests
+`eax` and, on failure, reads `errno` before emitting D07. The O trace therefore
+binds the following facts across source, map, disassembly and runtime:
+
+- `sceVideoOutOpen` returned positive handle `0x4e100100`;
+- `sceVideoOutRegisterBuffers2(handle, 0, 0, buffers, 2, attr, 0, NULL)`
+  returned `0`;
+- the diagnostic copied pixels into registered buffer zero;
+- `sceVideoOutSubmitFlip(handle, 0, 1, 0)` returned `-1`;
+- the immediately saved `errno` was `0`;
+- no flip-event wait followed the failed submit.
+
+The unmodified public SDL source at commit
+`0baf4ac49382b537ba449901b5b6d0d189bb1fbb` also starts `frame_id` at zero,
+uses `idx = frame_id % 2`, and calls the same tuple on its first normal frame.
+The earlier diagnostic-versus-normal first-index mismatch has therefore been
+eliminated and cannot explain the O result.
+
+## Evidence limits
+
+SDK v0.41 commit `d2e2e585740362976a39fdd5ccf390f199a7bc37` exports symbol stubs,
+but supplies no public VideoOut prototypes, structures, constants or return
+semantics. The SDL fork is the only local public source for the prototype,
+the 32-byte buffer descriptor, the opaque 80-byte attribute object and the
+format constant `0x8000000022000000`. Buffer registration returning zero is
+useful runtime evidence, but it does not independently validate those opaque
+semantics.
+
+| Question | Classification | Reason |
+|---|---|---|
+| exact failure site | PROVEN | source/map/disassembly plus D07 |
+| raw submit return | PROVEN | D07 is `-1` |
+| useful errno identity | ABSENT | saved value is `0` |
+| frame-zero difference from normal path | REJECTED_CURRENT_CAUSE | both paths now use zero |
+| buffer-zero validity | UNPROVEN | consistent with source, not independently specified |
+| flip-mode value `1` | UNPROVEN | no local public semantic definition |
+| frame-ID zero validity | UNPROVEN | used by SDL, not independently specified |
+| buffer attribute/layout semantics | UNPROVEN | opaque fork-local declaration |
+| flip-master or active-app ownership | UNPROVEN | export names are not callable ABI evidence |
+| visible presentation or cleanup | UNPROVEN | no successful submit or terminal proof |
+
+Calling `sceVideoOutSetFlipMaster`, a status query, or changing a submit
+argument based only on an export name would be an ABI guess. Phase 1.0P does
+not authorize or recommend such a change. Its decision is
+`VIDEOOUT_SUBMIT_FAILURE_SITE_PROVEN_ROOT_CAUSE_UNRESOLVED`.
+
+## Safe continuation gate
+
+Before another target artifact is designed, acceptable public evidence must
+define the relevant VideoOut signature, arguments, structure layouts, return
+codes and required ownership/state. If that evidence cannot be obtained, the
+hardware path remains fail-closed. A later artifact also needs a corrected
+terminal protocol, separately audited and separately authorized before any
+device action.