Skip to content

Run Windows games on Apple Silicon — free, open source, with an open compatibility database that tells you what actually works.

engine r12: note the DXVK Direct3D 9 failure shared with r11 and its workaround (highball#198)

Gauthier Piarrettef2a4a65a44fd
diff --git a/spike/engines/x64-crossover26.3-r12.json b/spike/engines/x64-crossover26.3-r12.json
index b95f90b..e1d1ae1 100644
--- a/spike/engines/x64-crossover26.3-r12.json
+++ b/spike/engines/x64-crossover26.3-r12.json
@@ -112,6 +112,7 @@
"minMacOS": "14.0",
"notes": [
"r12 (2026-09-24): r11 plus patch 0012, athei's x87sidecar cooperative attach with a loader check. The check alone changes every 32-bit launch: get_alternate_wineloader() used to name an i386-unix loader that does not exist in this wow64-only build, and env.c then relaunched the program through start.exe /exec, so every 32-bit game ran as a child of a detached start.exe. Measured 2026-09-24: two processes on r11, one on this build. With ROSETTA_X87_PATH pointing at athei's x87sidecar (not shipped yet) an i386 program is accelerated: Half-Life 2 timedemo 131 to 174 fps on wined3d, M1 Pro, macOS 27.0, measured on the 6788a14 trial build with WINEARCH=wow64; on this build the attach was verified with a test program only. Every other component is byte-identical to r11. Smoked 2026-09-24 on a clone of re-w11: The Last Flame (DXMT) and Five Nights at Freddy's (Wine's own Direct3D) draw, the 32-bit game runs as one process.",
+ "Known on r11 and r12 (highball#198, 2026-09-24): a Direct3D 9 game on the DXVK mode gets a window that never draws, every memory mapping fails with VK_ERROR_OUT_OF_HOST_MEMORY. Highball's build of MoltenVK is the difference, at 1.4.1 and 1.4.2 alike, while Khronos's own 1.4.2 build draws on this same engine; the two Highball patches do not run on that path. Until the library is rebuilt, use Wine's own Direct3D for Direct3D 9 games on this engine, which draws and was the faster mode for Half-Life 2 here.",
"r11 (2026-09-20): r10 with patches 0010 and 0011. 0010 (Proton 164af86d): the Media Foundation source reader lets the video processor's allocator create sample textures, so a game's media player gets a real shared handle from DXMT instead of null (The Last Flame's menu video now plays, highball#99; 3Shain/dxmt#135). 0011: on macOS %gs:0x68, where MSVC-built code such as Unity's Mono reads GetLastError() directly, is kept in step with TEB->LastErrorValue, so a Unity game's error paths no longer see garbage (The Last Flame's Start threw IOException Unknown error on a missing save). DXMT_ALLOW_CROSS_PROCESS_SWAPCHAIN=1 is in the base environment: CEF launcher windows paint on Wine 11 with it. Known: winedmo.so still links a Homebrew FFmpeg 9 the engine does not ship, so Wine's MF sources fall back to GStreamer; unchanged from r10.",
"r10 (2026-09-15): r9 with MoltenVK 1.4.2 (July 2026) in place of 1.4.1, same two Highball patches. 1.4.2 fixes heap placement checks, imported texture residency, cached texture invalidation and an occlusion query crash, all candidates for Red Dead Redemption 2's trees popping in and out on Apple Silicon (highball#93). A candidate engine, not the default: pick it under Advanced. Everything else is byte-identical to r9.",
"r9 (2026-09-15): r8 with the d3dmetal-tsshim component rebuilt so its exports carry Microsoft's d3d12 ordinals (D3D12CreateDevice at 101), which D3DMetal mirrors. The earlier shim numbered them from 1, so a game that binds d3d12 imports by ordinal (Path of Exile 2 delay-loads D3D12CreateDevice as ordinal 101) died at startup with 'ordinal 101 out of range' and Wine's exception 0xc06d007f (highball-db#53). The Wine build and every other component are byte-identical to r8, so existing downloads are reused; only the 46 KB shim tarball is new.",