feat(ps1): model-space TMD/HMD RAM scanner (#674)#674
Merged
Conversation
|
Warning Review limit reached
More reviews will be available in 41 seconds. Learn how PR review limits work. Your organization has run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After more reviews become available, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available. Please see our Fair Usage Limits Policy for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (5)
📝 WalkthroughWalkthroughThis PR adds model-space RAM scanner support to the PS1 geometry extraction pipeline. It introduces PSX TMD and HMD structure scanners that detect and reconstruct meshes directly from emulator RAM, extends the capture pipeline to track these model-space outputs, updates statistics and telemetry to report scanner activity, and integrates all components into the existing rendering/capture framework. ChangesPS1 Model-Space Mesh Capture Integration
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related issues
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1ce3daff8c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+429
to
+433
| // We don't know prim-payload sizes ahead of time. Conservatively reserve 8 bytes | ||
| // per prim (smallest packet) — used purely for dedupe range, not for parsing. | ||
| const auto r = resolveOffset(h.primOff, flags, tmdStart, ramSize, size_t(h.nPrim) * 8u); | ||
| if (r.ok) | ||
| maxEnd = std::max(maxEnd, r.base + size_t(h.nPrim) * 8u); |
There was a problem hiding this comment.
computeTmdSpan assumes 8 bytes per primitive, but many supported packets are larger (for example ilen=5/6/8), so span can end before the real primitive stream ends. Because this same span is used for contentHash and for scan skipping, two different TMD blobs with identical prefixes can be deduped as one, and the scanner can resume inside the same blob instead of advancing past it.
Useful? React with 👍 / 👎.
Comment on lines
+39
to
+40
| if (stats.ramTmdMeshes > 0 || stats.ramHmdMeshes > 0) | ||
| stats.primarySource = Gp0CaptureSource::RamModelMesh; |
There was a problem hiding this comment.
The HMD scanner currently returns only candidate counts and explicitly emits no meshes, but endGpuCapturePass treats any nonzero ramHmdMeshes as proof of model-space mesh capture and forces primarySource to ram_model_mesh. With QTMESH_PS1_HMD_SCANNER=1, this can mislabel captures (including false positives) and hide the actual GP0 source attribution.
Useful? React with 👍 / 👎.
fernandotonon
added a commit
that referenced
this pull request
May 25, 2026
Adds two format-aware runtime scanners that find Sony SDK mesh structures in PS1 system RAM and emit them as model-space `ReconstructedMesh` parts, bypassing the inverse-projection screen-space path that produces "blob of triangles" output on retail captures. `PsxTmdRamScanner` sweeps RAM at 4-byte stride for the `0x00000041` TMD magic. For each candidate it validates flags, numObj, object-header offset ranges and primitive counts, then walks the vertex/normal pools and primitive packets (mono/Gouraud triangles & quads, textured + no-light variants) and emits a fully-formed `CapturedModelMesh` via the new `EmuHooks::onModelMesh` hook. Material names for textured packets reuse `MeshReconstructor::textureMaterialName` so the existing capture texture-decode pipeline (#421) binds them with no extra wiring. `PsxHmdRamScanner` is a v1 stub (opt-in via `QTMESH_PS1_HMD_SCANNER=1`) that surfaces magic-bytes candidates for diagnostics while we collect a known-good HMD asset to validate the primitive walk against. `CaptureBuffer::addModelMesh` dedupes by FNV-1a 64-bit hash of the TMD byte span so the same asset encountered every frame only stores once. `MeshReconstructor::buildParts` appends `snapshot.modelMeshes` to the parts list — the dedupe pass in `reconstructDeduped` collapses byte-identical TMDs to one unique mesh + N instances, same as for screen-space matrix-grouped parts. `MeshReconstructionStats::modelMeshVertices` count flows into `gteInversePercent` as trusted vertices, so the quality indicator flips from 0% to ~100% on TMD-using games. UI / Sentry: `Gp0CaptureStats::ramTmdMeshes`/`ramHmdMeshes` track per-frame counts; `Gp0CaptureSource::RamModelMesh` (label `ram_model_mesh`) is the new primary source when any TMD/HMD surfaced; status bar appends `tmd N / hmd N`; new `ps1.rip.capture.modelmesh` breadcrumb; `ps1.rip.matrix.stats` adds `model_meshes=N`. Disable per-format via `QTMESH_PS1_TMD_SCANNER=0`. Tests (8 new, all green): finds single TMD at offset, rejects garbage headers, dedupes identical copies, rejects zero-vertex objects, handles random entropy without false positives, emits correct PS1Rip_tpage_*_clut_* material names for textured prims; HMD scanner counts plausible candidates and emits nothing by default. 67 PS1 tests across 17 suites pass with the new code. Co-authored-by: Cursor <cursoragent@cursor.com>
fernandotonon
added a commit
that referenced
this pull request
May 25, 2026
…s mesh distinction Addresses two code review findings on #674: 1) PsxTmdRamScanner::computeTmdSpan no longer assumes 8 bytes per primitive. The walker now records its actual end byte position via PrimContext::maxPrimStreamEnd, and computeTmdSpan uses that directly. Two TMDs with identical 8-byte-per-prim prefixes but different full payloads now produce distinct contentHash values, so they are no longer dedupe-merged into one, and the scanner advance correctly skips past the full blob instead of resuming inside it. 2) HMD scanner v1 returns candidate counts, not emitted meshes. Split `Gp0CaptureStats::ramHmdMeshes` (actual onModelMesh emissions, 0 until the v2 walker lands) from new `ramHmdCandidates` (plausible magic-byte hits). Only emitted meshes promote primarySource to RamModelMesh — bare candidate counts no longer hide GP0 source attribution behind a misleading `ram_model_mesh` label. Status bar + Sentry breadcrumbs surface both fields with the new `hmd_cand` suffix so testers can still confirm the magic-bytes path works on HMD-using titles without misattributing the capture. Adds PsxTmdRamScannerTest.ContentHashCoversFullPrimitivePayload and PsxHmdRamScannerTest.HmdCandidatesDoNotPromoteModelMeshSource. Co-authored-by: Cursor <cursoragent@cursor.com>
fernandotonon
force-pushed
the
feat/ps1-tmd-hmd-scanner-674
branch
from
1ce3daf to
54d84b9
Compare
Adds two format-aware runtime scanners that find Sony SDK mesh structures in PS1 system RAM and emit them as model-space `ReconstructedMesh` parts, bypassing the inverse-projection screen-space path that produces "blob of triangles" output on retail captures. `PsxTmdRamScanner` sweeps RAM at 4-byte stride for the `0x00000041` TMD magic. For each candidate it validates flags, numObj, object-header offset ranges and primitive counts, then walks the vertex/normal pools and primitive packets (mono/Gouraud triangles & quads, textured + no-light variants) and emits a fully-formed `CapturedModelMesh` via the new `EmuHooks::onModelMesh` hook. Material names for textured packets reuse `MeshReconstructor::textureMaterialName` so the existing capture texture-decode pipeline (#421) binds them with no extra wiring. `PsxHmdRamScanner` is a v1 stub (opt-in via `QTMESH_PS1_HMD_SCANNER=1`) that surfaces magic-bytes candidates for diagnostics while we collect a known-good HMD asset to validate the primitive walk against. `CaptureBuffer::addModelMesh` dedupes by FNV-1a 64-bit hash of the TMD byte span so the same asset encountered every frame only stores once. `MeshReconstructor::buildParts` appends `snapshot.modelMeshes` to the parts list — the dedupe pass in `reconstructDeduped` collapses byte-identical TMDs to one unique mesh + N instances, same as for screen-space matrix-grouped parts. `MeshReconstructionStats::modelMeshVertices` count flows into `gteInversePercent` as trusted vertices, so the quality indicator flips from 0% to ~100% on TMD-using games. UI / Sentry: `Gp0CaptureStats::ramTmdMeshes`/`ramHmdMeshes` track per-frame counts; `Gp0CaptureSource::RamModelMesh` (label `ram_model_mesh`) is the new primary source when any TMD/HMD surfaced; status bar appends `tmd N / hmd N`; new `ps1.rip.capture.modelmesh` breadcrumb; `ps1.rip.matrix.stats` adds `model_meshes=N`. Disable per-format via `QTMESH_PS1_TMD_SCANNER=0`. Tests (8 new, all green): finds single TMD at offset, rejects garbage headers, dedupes identical copies, rejects zero-vertex objects, handles random entropy without false positives, emits correct PS1Rip_tpage_*_clut_* material names for textured prims; HMD scanner counts plausible candidates and emits nothing by default. 67 PS1 tests across 17 suites pass with the new code. Co-authored-by: Cursor <cursoragent@cursor.com>
…s mesh distinction Addresses two code review findings on #674: 1) PsxTmdRamScanner::computeTmdSpan no longer assumes 8 bytes per primitive. The walker now records its actual end byte position via PrimContext::maxPrimStreamEnd, and computeTmdSpan uses that directly. Two TMDs with identical 8-byte-per-prim prefixes but different full payloads now produce distinct contentHash values, so they are no longer dedupe-merged into one, and the scanner advance correctly skips past the full blob instead of resuming inside it. 2) HMD scanner v1 returns candidate counts, not emitted meshes. Split `Gp0CaptureStats::ramHmdMeshes` (actual onModelMesh emissions, 0 until the v2 walker lands) from new `ramHmdCandidates` (plausible magic-byte hits). Only emitted meshes promote primarySource to RamModelMesh — bare candidate counts no longer hide GP0 source attribution behind a misleading `ram_model_mesh` label. Status bar + Sentry breadcrumbs surface both fields with the new `hmd_cand` suffix so testers can still confirm the magic-bytes path works on HMD-using titles without misattributing the capture. Adds PsxTmdRamScannerTest.ContentHashCoversFullPrimitivePayload and PsxHmdRamScannerTest.HmdCandidatesDoNotPromoteModelMeshSource. Co-authored-by: Cursor <cursoragent@cursor.com>
fernandotonon
force-pushed
the
feat/ps1-tmd-hmd-scanner-674
branch
from
54d84b9 to
291bfce
Compare
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (1)
src/PS1/runtime/MeshReconstructor.cpp (1)
227-241: ⚡ Quick winUnify duplicated bounds-update logic to avoid drift.
This block duplicates min/max bounds code already used elsewhere; extracting a shared helper keeps future stats changes consistent.
♻️ Proposed refactor
namespace { +void expandBounds(MeshReconstructionStats &stats, const ReconstructedVertex &v) +{ + if (!stats.hasBounds()) { + stats.boundsMinX = stats.boundsMaxX = v.px; + stats.boundsMinY = stats.boundsMaxY = v.py; + stats.boundsMinZ = stats.boundsMaxZ = v.pz; + return; + } + stats.boundsMinX = std::min(stats.boundsMinX, v.px); + stats.boundsMaxX = std::max(stats.boundsMaxX, v.px); + stats.boundsMinY = std::min(stats.boundsMinY, v.py); + stats.boundsMaxY = std::max(stats.boundsMaxY, v.py); + stats.boundsMinZ = std::min(stats.boundsMinZ, v.pz); + stats.boundsMaxZ = std::max(stats.boundsMaxZ, v.pz); +} + void accumulateVertexStats(const ReconstructedVertex &v, MeshReconstructionStats &stats, bool usedGte) { ++stats.totalVertices; @@ - if (!stats.hasBounds()) { - stats.boundsMinX = stats.boundsMaxX = v.px; - stats.boundsMinY = stats.boundsMaxY = v.py; - stats.boundsMinZ = stats.boundsMaxZ = v.pz; - return; - } - stats.boundsMinX = std::min(stats.boundsMinX, v.px); - stats.boundsMaxX = std::max(stats.boundsMaxX, v.px); - stats.boundsMinY = std::min(stats.boundsMinY, v.py); - stats.boundsMaxY = std::max(stats.boundsMaxY, v.py); - stats.boundsMinZ = std::min(stats.boundsMinZ, v.pz); - stats.boundsMaxZ = std::max(stats.boundsMaxZ, v.pz); + expandBounds(stats, v); } @@ - if (!statsOut->hasBounds()) { - statsOut->boundsMinX = statsOut->boundsMaxX = v.px; - statsOut->boundsMinY = statsOut->boundsMaxY = v.py; - statsOut->boundsMinZ = statsOut->boundsMaxZ = v.pz; - } else { - statsOut->boundsMinX = std::min(statsOut->boundsMinX, v.px); - statsOut->boundsMaxX = std::max(statsOut->boundsMaxX, v.px); - statsOut->boundsMinY = std::min(statsOut->boundsMinY, v.py); - statsOut->boundsMaxY = std::max(statsOut->boundsMaxY, v.py); - statsOut->boundsMinZ = std::min(statsOut->boundsMinZ, v.pz); - statsOut->boundsMaxZ = std::max(statsOut->boundsMaxZ, v.pz); - } + expandBounds(*statsOut, v);🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/PS1/runtime/MeshReconstructor.cpp` around lines 227 - 241, This duplicated min/max bounds logic should be centralized: extract a helper (e.g., MeshStats::updateBoundsWithVertex or MeshReconstructor::updateBoundsForVertex) and replace the duplicated block in MeshReconstructor::… (the place with the if (!statsOut->hasBounds()) check) with a call to that helper; the helper should take the stats object and a vertex (v) and implement the hasBounds() check plus min/max updates so all callers (including accumulateVertexStats) share the exact same bounds-update behavior.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/PS1/PS1_RIP_DESIGN.md`:
- Around line 78-85: The Markdown table for "Per-core capability" lacks
surrounding blank lines causing markdownlint MD058; open
src/PS1/PS1_RIP_DESIGN.md, locate the table that starts with "**Per-core
capability (`#662`, `#674`):**" and ends after the CI / no-disc smoke row, and add a
single blank line immediately before the table and a single blank line
immediately after the table so the table is separated from surrounding
paragraphs.
In `@src/PS1/runtime/Gp0HookDispatch.cpp`:
- Line 428: The primary source is promoted before model-mesh counters are
populated: move or repeat the call so
promoteModelMeshSource(pickPrimarySource(...)) runs after stats.ramTmdMeshes and
stats.ramHmdMeshes are computed; specifically, ensure stats.ramTmdMeshes and
stats.ramHmdMeshes are assigned (the code that sets them around the block that
currently assigns at lines ~527-530) before calling
pickPrimarySource/promoteModelMeshSource (functions: pickPrimarySource,
promoteModelMeshSource) so RamModelMesh can be considered when choosing
stats.primarySource; alternatively, call promoteModelMeshSource again after the
counters are set to update stats.primarySource.
In `@src/PS1/runtime/PsxModelRamScanCommon.h`:
- Around line 45-46: The FNV-1a 64-bit offset basis constant is incorrect:
replace the current kFnvOffset value with the canonical 64-bit offset basis
14695981039346656037ULL (0xcbf29ce484222325) in PsxModelRamScanCommon.h so that
constexpr uint64_t kFnvOffset and the existing constexpr uint64_t kFnvPrime use
the correct standard values.
---
Nitpick comments:
In `@src/PS1/runtime/MeshReconstructor.cpp`:
- Around line 227-241: This duplicated min/max bounds logic should be
centralized: extract a helper (e.g., MeshStats::updateBoundsWithVertex or
MeshReconstructor::updateBoundsForVertex) and replace the duplicated block in
MeshReconstructor::… (the place with the if (!statsOut->hasBounds()) check) with
a call to that helper; the helper should take the stats object and a vertex (v)
and implement the hasBounds() check plus min/max updates so all callers
(including accumulateVertexStats) share the exact same bounds-update behavior.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 3fa80718-3cf3-4fe4-8561-bcbcdfabf62e
📒 Files selected for processing (29)
src/PS1/CMakeLists.txtsrc/PS1/PS1_RIP_DESIGN.mdsrc/PS1/runtime/CaptureBuffer.cppsrc/PS1/runtime/CaptureBuffer.hsrc/PS1/runtime/CaptureSnapshot.cppsrc/PS1/runtime/CaptureSnapshot.hsrc/PS1/runtime/CapturedModelMesh.hsrc/PS1/runtime/EmuHooks.hsrc/PS1/runtime/Gp0CaptureStats.cppsrc/PS1/runtime/Gp0CaptureStats.hsrc/PS1/runtime/Gp0HookDispatch.cppsrc/PS1/runtime/MeshReconstructionStats.cppsrc/PS1/runtime/MeshReconstructionStats.hsrc/PS1/runtime/MeshReconstructor.cppsrc/PS1/runtime/MeshReconstructor.hsrc/PS1/runtime/PS1RipManager.cppsrc/PS1/runtime/PS1RipSessionWindow.cppsrc/PS1/runtime/PS1RipWorker.cppsrc/PS1/runtime/PsxHmdRamScanner.cppsrc/PS1/runtime/PsxHmdRamScanner.hsrc/PS1/runtime/PsxHmdRamScanner_test.cppsrc/PS1/runtime/PsxModelRamScanCommon.hsrc/PS1/runtime/PsxTmdRamScanner.cppsrc/PS1/runtime/PsxTmdRamScanner.hsrc/PS1/runtime/PsxTmdRamScanner_test.cppsrc/PS1/runtime/ReconstructedMesh.hsrc/PS1/runtime/RipperHooks.cppsrc/PS1/runtime/RipperHooks.htests/CMakeLists.txt
…, shared bounds helper Addresses three more code review findings on #674: 1) PsxModelRamScanCommon::fnv1a64 used a malformed 64-bit offset basis (1469598103934665603ULL — one digit short of canonical). Replaced with the standard FNV-1a 64-bit constants (0xcbf29ce484222325 / 0x100000001b3). Pre-existing contentHash values were never persisted across runs (in-session dedupe only) so the change is safe. Added Fnv1a64MatchesCanonicalReference test. 2) Gp0HookDispatch::captureFromSystemRam was calling promoteModelMeshSource while stats.ramTmdMeshes / ramHmdMeshes were still 0 — the inner merged-RAM scan can't see model meshes because the outer wrapper populates those counters AFTER this returns. Removed the no-op promotion from captureFromSystemRam and added a single promoteModelMeshSource call inside captureFrameFromSystemRam, AFTER stats.ramTmdMeshes is assigned. RamModelMesh can now actually win. Added CaptureFramePromotesPrimarySourceWhenTmdEmitted test. 3) MeshReconstructor.cpp had two near-identical bounds-update blocks (accumulateVertexStats and the snapshot.modelMeshes loop). Extracted `expandBounds(stats, vertex)` so both share the same hasBounds() anchor + min/max logic — keeps future stats tweaks in lock-step. Also added blank lines around the per-core capability table in PS1_RIP_DESIGN.md so it stops tripping markdownlint MD058. Co-authored-by: Cursor <cursoragent@cursor.com>
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds two format-aware runtime scanners that find Sony SDK mesh structures in PS1 system RAM and emit them as model-space
ReconstructedMeshparts, bypassing the inverse-projection screen-space path. On TMD-using games this is what turns the "blob of triangles" output into recognizable geometry without waiting on #676 (forked mednafen with in-core GTE hook).PsxTmdRamScanner— sweeps RAM (4-byte stride) for the0x00000041TMD magic. Validates flags / numObj / offsets / counts, walks vertex pools + primitive packets (mono/Gouraud tri & quad, textured + no-light variants), and emits via the newEmuHooks::onModelMeshhook. Material names for textured packets use the standardPS1Rip_tpage_XXXX_clut_YYYY_stN_dmNtemplate so the existing capture texture-decode pipeline (PS1: CLUT-aware texture decoding (4bpp/8bpp/15bpp → RGBA) #421) binds them with no extra wiring. Supports bothflag=0(on-disk relative offsets) andflag=1(KSEG0 absolute pointers).PsxHmdRamScanner— v1 stub for the0x00000050HMD magic. Opt-in viaQTMESH_PS1_HMD_SCANNER=1; surfaces candidate counts for diagnostics. Full primitive walk lands once a known-good HMD asset is available to validate against (HMD's hierarchical layout is a documented false-positive source).CaptureBuffer::addModelMeshdedupes by FNV-1a 64-bit content hash so the same asset encountered every frame only stores once.MeshReconstructor::buildPartsappendssnapshot.modelMeshesto the parts list; the existingMeshTopologyHashdedupe inreconstructDedupedcollapses byte-identical TMDs into one unique mesh + N instance transforms.MeshReconstructionStats::modelMeshVerticesflows intogteInversePercentas trusted vertices, so the quality indicator flips from 0% → ~100% on TMD-using games.tmd N / hmd N; newGp0CaptureSource::RamModelMesh(labelram_model_mesh) wins the primary-source label whenever any TMD/HMD surfaced; newps1.rip.capture.modelmeshSentry breadcrumb;ps1.rip.matrix.statsaddsmodel_meshes=N.Builds on #662 (live GP0 FIFO bridge); rebase against master once #662 is merged.
Coverage
0x41)0x50)Test plan
PsxTmdRamScannerTest.FindsSingleTmdAtOffset— hand-crafted 1-object 1-tri TMD at byte 0x10000 of 2 MiB zeroed RAM → 1 mesh, 3 verts, finite world-space positions.PsxTmdRamScannerTest.RejectsGarbageHeader— wrong magic / oversized numObj / bad flags / OOB pointers → 0 meshes.PsxTmdRamScannerTest.DedupesIdenticalCopies— two byte-identical TMDs at different offsets → 1 mesh (content-hash dedupe).PsxTmdRamScannerTest.RejectsZeroVertexObject— degenerate object header → 0 meshes.PsxTmdRamScannerTest.HandlesPureRandomEntropyWithoutFalsePositives— 64 KiB of seeded random bytes → 0 false positives.PsxTmdRamScannerTest.EmitsCorrectMaterialNameForTexturedPrim—0x24textured-tri packet →PS1Rip_tpage_1234_clut_abcd_...material binding.PsxHmdRamScannerTest.CountsPlausibleHmdCandidates— magic + plausible map length → counted; OOB map length → rejected.PsxHmdRamScannerTest.EmitsNothingByDefault— env-gated stub returns 0 meshes by default.tmd Nshows non-zero in the status bar and source flips toram_model_mesh.tmd 0(as expected — custom engines fall through to the screen-space path).Out of scope (queued follow-ups)
mednafen_psx_libretrowith in-core RTPS/RTPT callback for ground-truth model-space recovery on any game.Made with Cursor
Summary by CodeRabbit
New Features
Improvements
Documentation