video/gcm: fix cellGcmGetTiledPitchSize ABI and flip/vblank handler dispatch

[sp00nznet]

Summary

Two GCM/RSX bugs found while bringing up You Don't Know Jack (BLUS30569) under the recompiler. Both prevented the title's RSX command stream from being issued correctly. With these fixes the title's own recompiled code reaches the FIFO and issues real NV4097_CLEAR_SURFACE commands (visible clears on screen via the D3D12 backend).

1. cellGcmGetTiledPitchSize had the wrong ABI

The real export is uint32_t cellGcmGetTiledPitchSize(uint32_t size) — one argument, with the aligned tiled pitch returned in r3. The previous 2-arg form treated r4 as an out-pointer and wrote the pitch through it.

At the title's call site r4 still held the caller's CellVideoOutResolution* (left over from the preceding cellVideoOutGetResolution), so the write clobbered width/height with the pitch value. width became 0, and every downstream tile / display-buffer setup computed from garbage.

2. Flip/vblank handler callbacks dispatched through a clobbered OPD

Handlers were dispatched by re-resolving the handler OPD at tick time. The title's handler OPDs (0x530D70/78/80) have their code word clobbered in guest memory by a later lifted store, so re-resolving yields the TOC base instead of the real entry point.

Fix: capture the resolved {code, toc} at registration time and dispatch through a new ppu_guest_call_ct() (call-by-resolved-code/toc) in the PPU loader, which runs the real lifted handler regardless of the later OPD corruption.

Testing

• ps3recomp_runtime builds clean (Release).
• ydkj_boot.exe links clean end-to-end against the new cellGcmGetTiledPitchSize signature.
• At runtime the title issues real CLEAR_SURFACE from its own recompiled code.

🤖 Generated with Claude Code