There is %zN construct, which prints opcode suffix for operand N. I think it is supported by clang. If you use it, you would avoid passing the instruction mnemonic to macros. Also, I suspect that it would work without suffix at all.

With clang 4.0 I get "error: invalid operand in inline asm," though it works with gcc 4.9. But indeed, it looks like the suffixes aren't needed anyway.