第十三號艦隊
0%

goto and computed goto

Posted on Edited on

在網路上看到computed goto，才知道這不在C/C++標準裡面，只是gcc/clang中的extension，難怪我以前沒聽過
goto就不說了，goto fail太有名了，C/C++都已經建議少用goto，就介紹變種的computed goto

Example

gcc稱computed goto為"label as values"，使用label當作pointer，然後用goto jump到執行的地方

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
#include <stdio.h>

int main() {
    void *labels[] = {&&label1, &&label2, &&label3, &&label4};
    int i = 1;

    goto *labels[i];

label1:
    printf("This is label 1\n");
    goto end;

label2:
    printf("This is label 2\n");
    goto end;

label3:
    printf("This is label 3\n");
    goto end;
    
label4:
	printf("Invalid label\n");
	goto end;

end:
    printf("End of program\n");
    return 0;
}

Without computed gogo

將前面的範例，用符合標準的方式寫一次，能用的工具通常就是switch

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
#include <stdio.h>

int main() {
    int i = 1;

    switch (i) {
        case 0:
            printf("This is label 1\n");
            break;
        case 1:
            printf("This is label 2\n");
            break;
        case 2:
            printf("This is label 3\n");
            break;
        default:
            printf("Invalid label\n");
            break;
    }

    printf("End of program\n");
    return 0;
}

Generated assembly code

先看Switch的版本

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
main: # @main
  push rbp
  mov rbp, rsp
  sub rsp, 48
  mov dword ptr [rbp - 4], 0
  mov dword ptr [rbp - 8], 1
  mov eax, dword ptr [rbp - 8]
  test eax, eax
  mov dword ptr [rbp - 12], eax # 4-byte Spill
  je .LBB0_1
  jmp .LBB0_6
.LBB0_6:
  mov eax, dword ptr [rbp - 12] # 4-byte Reload
  sub eax, 1
  mov dword ptr [rbp - 16], eax # 4-byte Spill
  je .LBB0_2
  jmp .LBB0_7
.LBB0_7:
  mov eax, dword ptr [rbp - 12] # 4-byte Reload
  sub eax, 2
  mov dword ptr [rbp - 20], eax # 4-byte Spill
  je .LBB0_3
  jmp .LBB0_4
.LBB0_1:
  movabs rdi, offset .L.str
  mov al, 0
  call printf
  mov dword ptr [rbp - 24], eax # 4-byte Spill
  jmp .LBB0_5
.LBB0_2:
  movabs rdi, offset .L.str.1
  mov al, 0
  call printf
  mov dword ptr [rbp - 28], eax # 4-byte Spill
  jmp .LBB0_5
.LBB0_3:
  movabs rdi, offset .L.str.2
  mov al, 0
  call printf
  mov dword ptr [rbp - 32], eax # 4-byte Spill
  jmp .LBB0_5
.LBB0_4:
  movabs rdi, offset .L.str.3
  mov al, 0
  call printf
  mov dword ptr [rbp - 36], eax # 4-byte Spill
.LBB0_5:
  movabs rdi, offset .L.str.4
  mov al, 0
  call printf
  xor ecx, ecx
  mov dword ptr [rbp - 40], eax # 4-byte Spill
  mov eax, ecx
  add rsp, 48
  pop rbp
  ret
.L.str:
  .asciz "This is label 1\n"

.L.str.1:
  .asciz "This is label 2\n"

.L.str.2:
  .asciz "This is label 3\n"

.L.str.3:
  .asciz "Invalid label\n"

.L.str.4:
  .asciz "End of program\n"

LBB0_1到.LBB0_4分別對應四個switch case，main的前半段和LBB0_6和LBB0_7來決定跳到哪個label去
如果用computed goto的話

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
main: # @main
  push rbp
  mov rbp, rsp
  sub rsp, 96
  mov dword ptr [rbp - 4], 0
  mov rax, qword ptr [.L__const.main.labels]
  mov qword ptr [rbp - 48], rax
  mov rax, qword ptr [.L__const.main.labels+8]
  mov qword ptr [rbp - 40], rax
  mov rax, qword ptr [.L__const.main.labels+16]
  mov qword ptr [rbp - 32], rax
  mov rax, qword ptr [.L__const.main.labels+24]
  mov qword ptr [rbp - 24], rax
  mov dword ptr [rbp - 52], 1
  movsxd rax, dword ptr [rbp - 52]
  mov rax, qword ptr [rbp + 8*rax - 48]
  mov qword ptr [rbp - 64], rax # 8-byte Spill
  jmp .LBB0_6
.Ltmp1: # Block address taken
  movabs rdi, offset .L.str
  mov al, 0
  call printf
  mov dword ptr [rbp - 68], eax # 4-byte Spill
  jmp .LBB0_5
.Ltmp2: # Block address taken
  movabs rdi, offset .L.str.1
  mov al, 0
  call printf
  mov dword ptr [rbp - 72], eax # 4-byte Spill
  jmp .LBB0_5
.Ltmp3: # Block address taken
  movabs rdi, offset .L.str.2
  mov al, 0
  call printf
  mov dword ptr [rbp - 76], eax # 4-byte Spill
  jmp .LBB0_5
.Ltmp4: # Block address taken
  movabs rdi, offset .L.str.3
  mov al, 0
  call printf
  mov dword ptr [rbp - 80], eax # 4-byte Spill
.LBB0_5:
  movabs rdi, offset .L.str.4
  mov al, 0
  call printf
  xor ecx, ecx
  mov dword ptr [rbp - 84], eax # 4-byte Spill
  mov eax, ecx
  add rsp, 96
  pop rbp
  ret
.LBB0_6:
  mov rax, qword ptr [rbp - 64] # 8-byte Reload
  jmp rax
.L__const.main.labels:
  .quad .Ltmp1
  .quad .Ltmp2
  .quad .Ltmp3
  .quad .Ltmp4

.L.str:
  .asciz "This is label 1\n"

.L.str.1:
  .asciz "This is label 2\n"

.L.str.2:
  .asciz "This is label 3\n"

.L.str.3:
  .asciz "Invalid label\n"

.L.str.4:
  .asciz "End of program\n"

這裡的Ltmp1到Ltmp4一樣是對應四個case，而main的前半段就計算該跳到哪個label，在

1
2
3
.LBB0_6:
  mov rax, qword ptr [rbp - 64] # 8-byte Reload
  jmp rax

直接跳走了，根據benchmark的結果，這樣的作法比swithc快上不少，所以很多emulator和interpreter都採用這種作法

Computed goto and C++

有個很陰險的地方

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
#include <iostream>

class Test {
public:
    Test() { std::cout << "Constructor called\n"; }
    ~Test() { std::cout << "Destructor called\n"; }
};

int main() {
    void* labels[] = {&&label1, &&label2};
    int i = 1;

    {
        Test t;
        goto *labels[i];
    }

label1:
    std::cout << "This is label 1\n";
    goto end;

label2:
    std::cout << "This is label 2\n";
    goto end;

end:
    std::cout << "End of program\n";
    return 0;
}

在這種情況下，Test的destructor不起作用
如果改成

1
2
3
4
{
    Test t;
    goto label2;
}

逕行了，不過這通常不是我們要的，可能要依需求修改程式