I just rebuilt with less aggressive optimization flags.   

Instead of -march=native -msse2 -mfpmath=sse -O3 -ffast-math -funroll-loops
-fomit-frame-pointer -floop-interchange -floop-strip-mine -floop-block

I tried -O2 -fomit-frame-pointer

It now compiles, so I guess that this is a compiler regression.