I did some more extensive tests and it looks like this shift business is quite a mess[*]
But this MR only exposes existent info to the application, so it seems fine to me. It did rebase cleanly on top of current master for me.
[*]: Both Nvidia and AMD Windows return questionable shifts for a few particular pixel formats. AMD also reports 128-bit float formats with RGBA ordering, which would break the test, but we don't support those so we don't care.
Attaching the modified test that I used locally [0001-opengl32-tests-Hack-tests-for-_SHIFT_ARB.txt](/uploads/924ece652979cbe7e3bd57b59e9768c7/0001-opengl32-tests-Hack-tests-for-_SHIFT_ARB.txt)