Running "make check" on Windows yields lots of unwarrented unexpected failures


in total I had 177 unexpected failures. Many of the failures (but not all) happen because of
different calling conventions between Win64 and basically everything else on x86-64,
in particular the different registers used for argument passing.
I see basically the following approaches to this:

  1. Add some working target triple to the failing tests. (Like x86_64-apple-darwin or x86_64-pc-linux-gnu.)

  2. Relax the tested patterns, so that they are less picky about the registers used for
    argument passing.

  3. Basically like point 1, but add another pattern for Win64.

  4. XFAIL the test for Win64. (That would be the poorest solution.)

I could possibly provide some patches, but I would like to know if there is a preferred solution.