With the version of llvm we are on, i128 is only aligned to 8 bytes. We were using i128 to set the alignment of aggregate types. This was leading the the wrong alignment. Instead use f128 which is actually aligned to 16 bytes on our current version of llvm
With the version of llvm we are on,
i128
is only aligned to 8 bytes. We were usingi128
to set the alignment of aggregate types. This was leading the the wrong alignment. Instead usef128
which is actually aligned to 16 bytes on our current version of llvm