https://bugs.winehq.org/show_bug.cgi?id=40978
--- Comment #2 from Henri Verbeet hverbeet@gmail.com --- (In reply to Ruslan Kabatsayev from comment #0)
I've tracked the problem down to lack of support of GL_EXT_vertex_array_bgra on i915. I can reproduce it on my Haswell GPU by using MESA_EXTENSION_OVERRIDE=-GL_EXT_vertex_array_bgra. (FWIW, i915 supports GL_ARB_vertex_program, which might be used as a substitute. Although it's not completely HW accelerated, but should be better than pure SW implementation.)
It's certainly possible that the fallback path broke somewhere along the way, since EXT_vertex_array_bgra is so common these days. That would of course be a legitimate regression. What's odd about this is that I was under the impression that i915 supported EXT_vertex_array_bgra, although it's been long enough since I last touched i915 hardware that I could simply be wrong about that.