Cool! Didn't mean this to sound negative, I think the post is pretty cool. It would be interesting to know how performance compares with these optimizations, a

Cool! Didn't mean this to sound negative, I think the post is pretty cool. It would be interesting to know how performance compares with these optimizations, a cursory look makes it seem like it does similar things as you did manually (although no build ID for gcc6 pre-release in Matt Godbolt's (otherwise absolutely amazing) tool doesn't really help with reproducibility)