https://bugs.winehq.org/show_bug.cgi?id=46336
--- Comment #10 from Paul Gofman gofmanp@gmail.com --- This one, Bug #34266, Bug #36089, bug #30061 are about the same issue, with the most complete explanation in Bug #34266.
Just to avoid any confusion in testing, the solution from here or Bug #36089 (both avoiding inf result from 1.0 / x) is absolutely not an universal hack for all the similar bugs (while 1.0 / x is probably rather common place to get inf). To make a universal hack, one needs to check and workaround NaN results for literally every operation generated in GLSL.