Skip to content

How it evolved

A short history of the Linux port, from the first commit to the first playable release.

Date Milestone
2026-10-05 First commit: God of War recompiled with PS2Recomp, Windows first (KIexster)
2026-10-09 to 10-15 Research on Windows: native VU1, native IOP, exact game speed, controls, FMV and audio work
2026-10-14 Linux starts: build pipeline, ISO launcher, gamepad autodetect, Tk settings window, CI steps
2026-10-14 A Vulkan GS renderer is built in three phases (compute prims, textured prims, batching), with the compiled VU1 and threaded GS on by default. Result: about 20-30 FPS
2026-10-15 Debug cheats (god mode, no enemies), mute option
2026-10-31 Decision: adopt Waffle565's native runtime (OpenGL render-target GS, native VU1 and IOP, native SPU2/DMA) in place of our Vulkan renderer. Same controls, same movies, same cheats; average moves from ~21 to ~55 FPS. (#22)
2026-11-01 English FPS overlay, fullscreen toggle, render scale (1080p default), 1% low overlay
2026-11-01 AppImage builder and automated release on a version tag; v0.1.0, then v0.1.1 (fixes a first-run crash)
2026-11-01 Setup installs missing build packages itself; English-only READMEs; this manual
2026-11-02 v0.2.0: grid lines and the halo around characters fixed (desert haze, then the fog/DOF pass: high-resolution depth read, self-copy for feedback passes, extended copy clamp); repository cleanup: Linux and English only

v0.0.1 is a source snapshot of the Vulkan era, kept for history.

Why the Vulkan renderer was dropped

The Vulkan GS was a real step up from the CPU renderer, but the frame was still limited by VU1, GS and upload work (about 21 FPS on main). Waffle565's runtime removes that bottleneck by compiling the VU1 and IOP ahead of time and drawing through render targets, so it was adopted wholesale and our launcher, pad mapping, cheats and overlay were carried over. The Vulkan-era history is preserved on the adopt/upstream-runtime branch.

A correction worth keeping

An early "GPU choice" A/B test looked like a win but was a silent fallback to the CPU renderer: forcing NVIDIA offload variables breaks the GS EGL context. Always confirm [gs-hw] render-target renderer ready in the log before trusting a measurement.

What is next

See Status: other pads, longer microcode capture and the 44-50 FPS dips.