Fix another endianness issue in i128 trans

[nagisa]

Apparently LLVMArbitraryPrecisionInteger demands integers to be in low-endian 64-bytes, rather than host-endian 64-bytes. This is weird, and obviously, not documented. And rustc now works a teeny bit more on big endians.

r? @eddyb

[nagisa]

Interestingly this is something that wouldn’t be eventually caught by CI if it wasn’t fully built on a big-endian machine.

[nagisa]

Probably fixes #39331.

[eddyb]

@bors r+

[bors]

📌 Commit b8036b6 has been approved by eddyb

[cuviper]

Apparently LLVMArbitraryPrecisionInteger demands integers to be in low-endian 64-bytes, rather than host-endian 64-bytes. This is weird, and obviously, not documented.

FWIW, I looked into LLVMConstIntOfArbitraryPrecision, which constructs an APInt. That just copies the given words as-is in APInt::initFromArray. Then in usage, bits are indexed by simple division in APInt::whichWord, which makes it little-endian in effect.

staticunsignedwhichWord(unsigned bitPosition) {
return bitPosition / APINT_BITS_PER_WORD;
}