The host layout behave very differently in that case and we have tests
that check the exact results.

This series depends on secret agreement APIs added in GnuTLS 3.8.2.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=49543

It's not set anywhere.

This requires moving the stack switch to Wow64PrepareForException.

Based on a patch by Paul Gofman.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55927

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55950

This should help understand some virgl/ChromeOS-specific WineD3D
behavior better.

Based on a patch by Michael Müller.

This should more closely match the algorithm used by native, discovered through
extensive manual testing.

This does not include any automated tests, both because I could not get winmm to
recognize a second plugged joystick, and (as I eventually discovered while
testing) the number of tests that would be necessary to conclusively prove this
algorithm turned out to be impractical.