[Framerr]

Purpose

Isolate the photo-capture fix from notsapinho/Eglantine commit 7dde4f15, present in build 24. The reporter tested that complete build and observed no corrupted photos and consistent 60 FPS, but build 24 includes other unrelated changes; this PR tests only the DMA readback correction.

Change

• After successful accelerated ImageToBuffer on Switch 2, flush the destination range from BufferCache to guest memory before DMA completes.
• Control: EDEN_NX2_DMA_READBACK_FLUSH=0 disables the new flush, allowing A/B comparison.
• Changes one production file: src/video_core/engines/maxwell_dma.cpp.

Isolation / safeguards

• Based on nx2, not the rejected, photo-regressing PR experiment: isolate Orchid #66 buffer readback fix for Pokopia photos #16.
• Draft: do not merge until runtime validation. No version bump, release or planned build yet.
• No performance optimizations, shader changes, app icon/README changes in this PR. Those require separate review and eventual coordinated release.

To verify

• Build only when the reporter requests a test build.
• Compare 10+ saved photos across multiple game sessions against Frameropia v0.0.2 and notsapinho build 24.
• Check geometry flicker, saving and FPS.
• Run same test with EDEN_NX2_DMA_READBACK_FLUSH=0 to confirm causality.

Related issue: #14. Keep main and nx2 intact.

[Framerr]

Reporter validation — approved for integration (2026-11-08)

Tested Windows artifact from run 42427657634: 48 of 50 Pokémon Pokopia saved photos were correct (96%), only 2 of 50 corrupted. Reporter explicitly approves this improvement.

Possible relationship between the remaining failures and shaders compiling during capture is unconfirmed; requires further investigation. This significantly improves results but is not a complete fix. Integrate only this DMA readback correction; keep the still-open photo issue tracked. No new version or release with this merge.