My C compiler, Lance, now has SLP autovectorization.
Here it showcases being able to vectorize code that is out of loops. (through GDB disassembly)
Currently the runtime overhead is a little bit high, but it is able to beat scalar code in a comfortable amount of cases. The journey to reduce overhead was the hardest part about this for sure. Loop autovec had much less overhead relatively, but it also couldn't apply to nearly as much code as SLP autovec can.