[compiler-rt] [compiler-rt] Add big endian support to __aeabi_(idivmod|uidivmod|uldivmod) (PR #126277)

[Simon Tatham via llvm-commits]

statham-arm wrote:

Can you explain some more in the commit message, please? All it says is that it's "making things work", but not what was wrong with them, or why, or what values go where.

The 32-bit integer division functions `__aeabi_idivmod` and `__aeabi_uidivmod` shouldn't change their ABI depending on endianness. They're defined to deliver the quotient in r0 and the remainder in r1, whether they're big- or little-endian. [(source)](https://github.com/ARM-software/abi-aa/blob/main/rtabi32/rtabi32.rst#integer-32-32-32-division-functions) So _why_ is this change swapping the result registers around? Which one do you think is which? Are they coming back from `__divmodsi4` in opposite order, for some reason?

Also, all these changes are conditional on the macro `_YUGA_BIG_ENDIAN`. What is that? Where's it defined? The ACLE macro name is `__ARM_BIG_ENDIAN`. Does this other macro indicate a different change to the ABI?

https://github.com/llvm/llvm-project/pull/126277