https://bugs.winehq.org/show_bug.cgi?id=53737
--- Comment #11 from Paul Gofman pgofman@codeweavers.com --- So I am not sure that attribution of the regression to commit dfd5f109fb4ebad859bf3ce3960b3b2b2ad1341d is fully correct (although of course it increased VM usage a bit).
If possible, it would be interesting to further bisect the regression after that blamed commit by doing the bisect starting from the next after dfd5f109fb4ebad859bf3ce3960b3b2b2ad1341d and applying the revert on each step. It can be done by putting the revert as a patch and doing 'git apply revert.patch' at each step, then 'git reset --hard' after build.
Yet I think the most interesting is why it is so close to VM exhaustion even before dfd5f109fb4ebad859bf3ce3960b3b2b2ad1341d, the major issue (if there is indeed a specific one and not just borderline VRAM usage by the game plus usual pulseaudio etc. overhead) is probably elsewhere. That needs some debugging.