[llvm] It is really the byte order of the target that matters (PR #107915)

[James Henderson via llvm-commits]

https://github.com/jh7370 requested changes to this pull request.

I can't believe that this PR is correct for all cases (I don't know enough about some of the tools to know whether it does for those ones, but certainly the yaml2obj/llvm-objdump ones are wrong).

Also, there are certainly cases where the host machine's endianness is significant to the test, because it impacts how numbers are represented in memory. If the program were to then print those numbers as bytes, rather than as their representation (or equally, if the program were to try to interpret an arbitrary set of bytes as a number without considering whether byte swapping was necessary), then the output would be different. I therefore cannot believe that removing the host-byteorder-* directives can be correct in all cases.

Please provide more explanation why in each test case (or family of related test cases) what you've done is correct.

https://github.com/llvm/llvm-project/pull/107915