Showing raw machine code along with disassembled code

[bicycle1885]

This pull request implements a new option to code_native to print raw machine code along with disassembled code. If the keyword argument rawcode is set to true (false by default), raw machine code in hexadecimal format is printed as a comment just before each disassembled instruction line. This kind of information may sometimes be useful when the user is interested in code size or alignment of a tight loop (see #39108 (comment), for example).

Here is an example. You can see L16 is aligned with a nop instruction.

julia> function collatz(n)
           s = 0
           while n != 1
               if iseven(n)
                   n = n ÷ 2
               else
                   n = 3n + 1
               end
               s += 1
           end
           return s
       end
collatz (generic function with 1 method)

julia> code_native(collatz, (Int,), syntax = :intel, binary = true)
        .text
; code origin: 00007f4736da69a0, code size: 64
; ┌ @ REPL[1]:1 within `collatz'
; 69a0: 31 c0
        xor     eax, eax
; │ @ REPL[1]:3 within `collatz'
; │┌ @ operators.jl:264 within `!='
; ││┌ @ promotion.jl:409 within `=='
; 69a2: 48 83 ff 01
        cmp     rdi, 1
; │└└
; 69a6: 75 16
        jne     L30
; │ @ REPL[1]:11 within `collatz'
L8:
; 69a8: c3
        ret
; 69a9: 0f 1f 80 00 00 00 00
        nop     dword ptr [rax]
; │ @ REPL[1]:7 within `collatz'
; │┌ @ int.jl:87 within `+'
L16:
; 69b0: 48 8d 7c 7f 01
        lea     rdi, [rdi + 2*rdi + 1]
; │└
; │ @ REPL[1]:9 within `collatz'
; │┌ @ int.jl:87 within `+'
L21:
; 69b5: 48 ff c0
        inc     rax
; │└
; │ @ REPL[1]:3 within `collatz'
; │┌ @ operators.jl:264 within `!='
; ││┌ @ promotion.jl:409 within `=='
; 69b8: 48 83 ff 01
        cmp     rdi, 1
; │└└
; 69bc: 74 ea
        je      L8
; │ @ REPL[1]:4 within `collatz'
L30:
; 69be: 40 f6 c7 01
        test    dil, 1
; 69c2: 75 ec
        jne     L16
; │ @ REPL[1]:5 within `collatz'
; │┌ @ int.jl:263 within `div'
; 69c4: 48 89 f9
        mov     rcx, rdi
; 69c7: 48 c1 e9 3f
        shr     rcx, 63
; 69cb: 48 01 f9
        add     rcx, rdi
; 69ce: 48 d1 f9
        sar     rcx
; 69d1: 48 89 cf
        mov     rdi, rcx
; │└
; 69d4: eb df
        jmp     L21
; 69d6: 66 2e 0f 1f 84 00 00 00 00 00
        nop     word ptr cs:[rax + rax]
; └

Closes #39108.

[yuyichao]

Note that for AArch32 and AArch64 (and possibly other ISA with fixed size instructions, not sure about Thumb....) at least gdb and objdump uses the convention of printing out the instruction in 32bit hex instead of bytes. It'll be nice if this can be consistent with that.

[bicycle1885]

Thank you. I've updated the printer to follow the convention but I cannot test it because I have no access to ARM machines.

[vtjnash]

I wonder if this is MCAsmInfo::getMinInstAlignment https://llvm.org/doxygen/classllvm_1_1MCAsmInfo.html#a2c92e93d2452cf0ad4bc91c46685425b