Make Dolphin FIFO captures a deep GX parity oracle - #1
Draft
ai-tdd-labs wants to merge 3 commits into
Draft
Conversation
Carry initial TMEM through DFF-to-DOLT conversion and export deterministic per-draw GX state, texture, matrix, vertex, post-clip, TEV, and pixel observations. Snapshot state at the true draw boundary so delayed sink delivery cannot attribute successor state to the previous draw. Constraint: DFF remains the hardware-shaped authority; JSONL is derived analyzer output. Rejected: Compare screenshots alone | cannot localize the first pipeline divergence. Confidence: high Scope-risk: moderate Directive: Preserve the DFF-parser self-check before using GXRuntime as an oracle. Tested: 7 CTest targets; 90-frame AC DFF with 20,511 and 22,328-draw exact analyzer parity; Aurora pixel replay Not-tested: Interactive input-driven scenes beyond the no-input attract loop
Extend compact replay evidence with world, clip, UV, TLUT, and rejection observations while explicitly rounding binary32 transforms so compiler FMA contraction cannot invent a geometry divergence. Keep optional pixel replay at canonical EFB dimensions independent of Retina backing scale.\n\nConstraint: DFF and GXRuntime must derive identical evidence from identical FIFO state on different compilers and display scales.\nRejected: Widen float tolerances | it hides real geometry errors and leaves host-dependent evidence.\nConfidence: high\nScope-risk: moderate\nDirective: Preserve explicit binary32 operation ordering in compact geometry analysis.\nTested: Headless and Aurora dolgx_replay builds; 7/7 GXRuntime tests; 35,393 AC draws match the DFF analyzer trace exactly.\nNot-tested: Full canonical-EFB pixel replay under every supported host GPU.
Mid-frame DFF captures do not carry a full MEM1 snapshot, so Mode-A replay must distinguish captured zero bytes from allocator zero-fill. Track validated MEM_UPDATE coverage at draw-command boundaries and expose texture hashes only when the complete source span is authoritative. Constraint: Rendering and replay digests must remain unchanged; only parity evidence authority is gated. Rejected: Treat all-zero hashes as unknown | genuine captured textures can legally contain only zero bytes. Confidence: high Scope-risk: narrow Directive: Preserve command-boundary provenance because sink draw callbacks are intentionally delayed. Tested: headless full build; frontend ctest 7/7; Aurora dolgx_replay build; null-null-hash subprocess regression. Not-tested: live GPU pixel replay of a partial-window DFF.
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.
What changed
Preserves initial TMEM in DFF-to-DOLT conversion and adds deterministic per-draw JSONL export for GX state, textures, matrices, vertex payloads, post-clip samples, TEV state, and frame pixels. Draw state is snapshotted at the command boundary so delayed sink completion cannot observe successor state.
Why
Pixel-only comparison cannot localize the first graphics-pipeline divergence, and missing initial TMEM can corrupt early CI textures and palettes.
Verified