clangs inline assembly sucks, compared to gcc.
Show this thread
-
Another thing is, memory constraints are always spilled on the stack. GCC can correctly keep them in registers and never touch the stack. That's implemented by the LLVM guys because "it's easier". Or plain unimplemented features, like return values in rdx:rax.
I personally don't care much for the x86 specific stuff (other than that I like my programs to run fast :), so I wasn't even aware of features like "=@ccz" to access ZF. Don't need sth like that on RISC-V. :) But spilling stuff unnecessarily on the stack is certainly bad. Ouch.
