mirror of
https://github.com/libretro/parallel-ps2.git
synced 2026-11-10 03:11:38 -10:00
LibretroAdmin
5ad55b00aa
retro_run polls the pad, latches it and lets the EE start a frame, then consumes the vsync that ends that frame and presents it. Behind that vsync the EE waits at a frame gate until the next retro_run opens it, so between two retro_runs it runs nothing and no frame is drawn from an older retro_run's input. A pause opens the gate too, and the EE stops at the vsync it waited behind: a savestate taken there holds no pending frame, and the retro_run after a load runs a frame on its own input. A GS that closes opens it as well. When a drain has left a frame pending, retro_run presents it and the EE stays where it stopped until the next one. tests/latency: a synthetic BIOS clears the screen once a vsync to a colour it reads from main RAM, which the harness writes before a retro_run as the core latches the pad there, and waits between retro_runs as a frontend does while it presents; each colour must be in the picture of the retro_run it was written before. tests/mtgsvsync models the gate on the real work eventcount: each frame is scanned out from its vsync's registers, drawn from the input polled in the same call, and the producer starts nothing between calls. A producer let go at the vsync, and a gate opened before polling, both fail it.