Alignment should account for long instruction names

Usually, the assembly output is aligned using spaces on the first argument, e.g.:

        mov     rcx, qword ptr [r8 + rax + 32]
        imul    rcx, qword ptr [rdx + rax + 32]
        add     rcx, qword ptr [rsi + rax + 32]

However, when the compilers generate instructions with very long names, the alignment is not pushed rightwards to accommodate for these, and we get, for example:

        vmovsd  xmm0, qword ptr [rdx + rax + 32] # xmm0 = mem[0],zero
        vmovsd  xmm1, qword ptr [rcx + rax + 32] # xmm1 = mem[0],zero
        vfmadd213sd     xmm1, xmm0, qword ptr [rsi + rax + 32]
        vmovsd  qword ptr [rdi + rax + 32], xmm1

which is kind of ugly and less readable. Now, assuming the user has not enabled \s+, I believe the arguments should be aligned by ensuring there’s enough space.

[mattgodbolt]

I agree it’d be lovely to fix this. GCC and clang only put a single tab between instructions and their operands, which means as soon as an instruction is 8 or more characters long, the indent doesn’t work. We can of course parse the instructions, and re-flow them, but then all instructions would be super-indented to account for a single very long instructions. I think this is why the compilers stick to a single tab character, instead of using two or more. I guess we could treat tabs as being (say) 12 characters long which I think covers all current x86 instructions…

[RubenRBS]

Thanks for taking the time to check this issue, let us know if you find any other problem where we can help 😃