X-Moto

X-Moto 0.6.3 runs at 1920x1080 with the right picture and sound. Our package batches the game’s OpenGL drawing (a frame went from 18,500 GL calls to 165), which took the benchmark replay from 44.5 to about 160 fps; its played sessions ran at the 60 fps vsync cap with clean sound (“flawless”, the player). It is one of the games where game-side batching paid most.

Packages and patches

xmoto, xmoto-data (0.6.3+repack-1+ps3.N, packages/xmoto), replacing Debian’s.

Controllers (0005). Controller N (the N-th SDL game controller) plays player N, no binding needed, at the same time as the keys: right trigger or R1 accelerate, left trigger or L1 brake, d-pad or left stick lean, Cross turns around, Triangle restarts the level, Square restarts from the checkpoint, Start pauses. The table is padBindings in Input.cpp; Options > Controls > Controller lists it. Pads plugged in later take the first free player; controller events are always on, whatever “Enable joysticks” says. Menus take the d-pad, left stick, Cross and Circle.

packages/xmoto/test/glbatch-test.sh checks the GL layer on the PC’s llvmpipe: 400 frames of a random scene, the same pixels with and without the stream buffer, no GL errors (make test ONLY=xmoto-glbatch, needs make xmoto-work).

How it performs

• Bench (batch 2026-10-18-batch16, kernel .433, x3): base 160.7 steady, 1% low 106.7, no gameplay stalls, longest 43 ms, 51 MB free, 37 MB of VRAM, 148 underruns; stream (0004) 162.1 (+1%), 106.3, 0.4 s, 380 ms. In batch 2026-10-18-mesa-416-hot-spots (one run each) base 160.4 against stream 163.1, CPU 298 against 282 s. So the stream buffer gives about 1% and less CPU, and the hitches move from run to run: no default change yet (the investigation).
• On the SSD root with ps3rsx’s shadow_fast (batch 2026-10-19-ssdroot2-shadow-fast): 159.0 against 158.4, no effect. The microphone hears the game’s music and engine in all 8 clips, with 200-218 ALSA underruns a run.
• Memory: 49 MB Rss, nothing swapped, 0.3 MB free in its heap (nothing to cut).
• Played (2026-10-23 and 10-24, three sessions up to 26 minutes): 60 fps median, 1% low 34-44, 0.3-1.3 s of stutter a minute.

The benchmark (recipe) plays the “credits” replay, which ships inside xmoto.bin where --replay doesn’t read it, so assets/xmoto-credits.rpl is it extracted. --timedemo --benchmark draws every frame, prints Average framerate and exits; it sets no swap interval itself, so vblank_mode=0. The replay runs in real time (about 144 s), so two passes: the first is the warm-up. Since the steady state, X-Moto’s fps value is the game’s average over the whole replay, not the HUD median (--hud xmoto keeps the old meaning).

Known problems

• The engine sound is quiet under the music, not missing.

Investigations

• 2026-10-17 game-side batching: 44.5 to 166.5 fps (0001-0002).
• 2026-10-18 the batches copied at every draw (0004).