Skip to content

experiment: test faster linear texture uploads for Pokopia - #18

Closed
Framerr wants to merge 4 commits into
nx2from
experiment/pokopia-linear-texture-upload
Closed

Framerr wants to merge 4 commits into
nx2from
experiment/pokopia-linear-texture-upload

Conversation

@Framerr

@Framerr Framerr commented Nov 8, 2026

Copy link
Copy Markdown
Owner

Isolated performance experiment

Port notsapinho/Eglantine commit 6e73445ccb, adapting the UnswizzleLinearImage declaration to our baseline's util.h layout. Read pitch-linear textures directly from guest memory into staging (avoid intermediate duplicate copy).

No photo DMA readback fix is in this branch. This is a stand-alone performance A/B test against nx2, not a candidate for direct merge without user testing. Build via read-only fork-windows-check workflow uploads a temporary Windows artifact and creates no release.

Check Pokopia FPS, shader stalls, photos, scene appearance, and loading. User confirmed notsapinho build 24 reaches steady 60 FPS, but that is a combined result of multiple commits; this patch's effect is not yet verified in Frameropia.

Related #15. Do not merge or release yet.

Framerr commented Nov 8, 2026

Copy link
Copy Markdown
Owner Author

Reporter validation — no measurable benefit (2026-11-08)

Tested Windows artifact from run 42427749053. Reporter observed no significant performance improvement and all saved photos corrupt. Importantly, this experiment did not include photo DMA readback fix #17, so photo corruption cannot be attributed to the texture upload change alone. With no observed benefit, do not merge; close the experiment, keep branch for reference. No release.

@Framerr Framerr closed this Nov 8, 2026
Framerr added a commit that referenced this pull request Nov 8, 2026
Ship the validated Pokopia DMA readback fix, new transparent Frameropia F logo for README and Windows app icon, redesigned README and honest v0.0.3 release notes. Exclude rejected experiments (#16, #18, #19, #20). Preserve existing user data and executable names.
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.

1 participant