okay, I didn't know that, thanks for the background. I wonder why the assembler would not optimize this though.

I noticed that I then could also shave of one byte more by using lea esi, [rel msg] instead of lea rsi, [rel msg].