Thanks :) That's where I was trying to get to. I did some smaller performance comparisons against C (with -O0), where I was at about 90% speed. But there is a lot of performance to gain, if I optimize the resulting assembly. There are lots of cases where I push and pop directly afterwards because of the general expression code generation (no real knowledge of broader context). So I expect that to help a lot regarding performance. Also things like jump tables for simple switches are on the table.

Yes, I thought about going for LLVM or another representation but decided to do it once myself (no given performance optimizations or the like) with room for improvement.