This is the patch that lowers x86 intrinsics to native IR
in order to enable optimizations.
Corresponding LLVM part: https://reviews.llvm.org/D46179
Previous D44786 version was reverted due to crashes caused by LLVM part. This revision includes a more optimal pattern for addus.
Added missing scalar intrinsics without rounding.
Whoops, that's a wrong revision. I'll revert it shortly.
Removed signed intrinsics lowering due to the pattern being too complicated - instead some minor optimizations were introduced on LLVM side.