glibc代码学习(2)：strcpy源代码(2)

本文的copyleft归[e-mail]所有，使用GPL发布，可以自由拷贝，转载。但转载请保持文档的完整性，注明原作者及原链接，严禁用于任何商业用途。
作者：[e-mail]
博客：linuxfocus.blog.chinaunix.net 

上一篇博文，发现glibc中的strcpy的效率居然比我写出的例子要差，这实在让我感到惊奇。下面看看为什么glibc中的实现，效率会低呢？让我们反汇编两个实现的代码

第一个strcpy，是我写的例子。

char* my_strcpy1(char *dest, const char *src)
{
    char *d = dest;
    register char c;

    do {
        c = *src++;
        *d++ = c;
    } while ('\0' != c);

    return dest;
}
它对应的汇编代码如下：

Dump of assembler code for function my_strcpy1:
0x08048394 <+0>:     push   %ebp
0x08048395 <+1>:     mov    %esp,%ebp
0x08048397 <+3>:     push   %ebx
0x08048398 <+4>:     sub    $0x10,%esp
0x0804839b <+7>:     mov    0x8(%ebp),%eax
0x0804839e <+10>:    mov    %eax,-0x8(%ebp)
0x080483a1 <+13>:    mov    0xc(%ebp),%eax
0x080483a4 <+16>:    movzbl (%eax),%ebx
0x080483a7 <+19>:    addl   $0x1,0xc(%ebp)
0x080483ab <+23>:    mov    -0x8(%ebp),%eax
0x080483ae <+26>:    mov    %bl,(%eax)
0x080483b0 <+28>:    addl   $0x1,-0x8(%ebp)
0x080483b4 <+32>:    test   %bl,%bl
0x080483b6 <+34>:    jne    0x80483a1
0x080483b8 <+36>:    mov    0x8(%ebp),%eax
0x080483bb <+39>:    add    $0x10,%esp
0x080483be <+42>:    pop    %ebx
0x080483bf <+43>:    pop    %ebp
0x080483c0 <+44>:    ret
End of assembler dump.

红色部分的汇编代码为do {} while循环代码。

glibc中的strcpy的代码如下：

/* Copy SRC to DEST. */
char *
my_strcpy2 (dest, src)
     char *dest;
     const char *src;
{
  register char c;
  char * s = (char *)src;
  const int off = dest - s - 1;

  do
    {
      c = *s++;
      s[off] = c;
    }
  while (c != '\0');

  return dest;
}

对应的汇编代码如下：

Dump of assembler code for function my_strcpy2:
0x080483c1 <+0>:     push   %ebp
0x080483c2 <+1>:     mov    %esp,%ebp
0x080483c4 <+3>:     push   %ebx
0x080483c5 <+4>:     sub    $0x10,%esp
0x080483c8 <+7>:     mov    0xc(%ebp),%eax
0x080483cb <+10>:    mov    %eax,-0xc(%ebp)
0x080483ce <+13>:    mov    0x8(%ebp),%edx
0x080483d1 <+16>:    mov    -0xc(%ebp),%eax
0x080483d4 <+19>:    mov    %edx,%ecx
0x080483d6 <+21>:    sub    %eax,%ecx
0x080483d8 <+23>:    mov    %ecx,%eax
0x080483da <+25>:    sub    $0x1,%eax
0x080483dd <+28>:    mov    %eax,-0x8(%ebp)
0x080483e0 <+31>:    mov    -0xc(%ebp),%eax
0x080483e3 <+34>:    movzbl (%eax),%ebx
0x080483e6 <+37>:    addl   $0x1,-0xc(%ebp)
0x080483ea <+41>:    mov    -0x8(%ebp),%eax
0x080483ed <+44>:    add    -0xc(%ebp),%eax
0x080483f0 <+47>:    mov    %bl,(%eax)
0x080483f2 <+49>:    test   %bl,%bl
0x080483f4 <+51>:    jne    0x80483e0
0x080483f6 <+53>:    mov    0x8(%ebp),%eax
0x080483f9 <+56>:    add    $0x10,%esp
0x080483fc <+59>:    pop    %ebx
0x080483fd <+60>:    pop    %ebp
0x080483fe <+61>:    ret
End of assembler dump.

这里的红色部分同样是对应的do{}while循环代码。

两个实现对应的汇编代码基本相似，那么是否由循环前面的代码引起的呢。my_strcpy2使用了offset，所以多了一些mov和sub操作。我再次更改了代码，在my_strcpy2中不再计算offset。

#include <stdio.h>
#include <stdlib.h>


char* my_strcpy1(char *dest, const char *src)
{
    char *d = dest;
    register char c;

    do {
        c = *src++;
        *d++ = c;
    } while ('\0' != c);

    return dest;
}


int off;

/* Copy SRC to DEST. */
char *
my_strcpy2 (dest, src)
     char *dest;
     const char *src;
{
  register char c;
  char * s = (char *)src;

  do
    {
      c = *s++;
      s[off] = c;
    }
  while (c != '\0');

  return dest;
}

int main()
{
    const char *str1 = "test1";
    char buf[100];

    off = buf-str1-1;

    int i;
    for (i = 0; i < 10000000; ++i) {
        my_strcpy1(buf, str1);
    }

    return 0;
}

通过使用一个off的全局变量，来省得my_strcpy2的offset的计算。但是结果仍然是my_strcpy1效率跟高。

my_strcpy1的时间约为0.147s，而my_strcpy2的时间为0.220s。再次查看汇编

(gdb) disassemble my_strcpy1
Dump of assembler code for function my_strcpy1:
0x08048394 <+0>:     push   %ebp
0x08048395 <+1>:     mov    %esp,%ebp
0x08048397 <+3>:     push   %ebx
0x08048398 <+4>:     sub    $0x10,%esp
0x0804839b <+7>:     mov    0x8(%ebp),%eax
0x0804839e <+10>:    mov    %eax,-0x8(%ebp)
0x080483a1 <+13>:    mov    0xc(%ebp),%eax
0x080483a4 <+16>:    movzbl (%eax),%ebx
0x080483a7 <+19>:    addl   $0x1,0xc(%ebp)
0x080483ab <+23>:    mov    -0x8(%ebp),%eax
0x080483ae <+26>:    mov    %bl,(%eax)
0x080483b0 <+28>:    addl   $0x1,-0x8(%ebp)
0x080483b4 <+32>:    test   %bl,%bl
0x080483b6 <+34>:    jne    0x80483a1
0x080483b8 <+36>:    mov    0x8(%ebp),%eax
0x080483bb <+39>:    add    $0x10,%esp
0x080483be <+42>:    pop    %ebx
0x080483bf <+43>:    pop    %ebp
0x080483c0 <+44>:    ret
End of assembler dump.


(gdb) disassemble my_strcpy2
Dump of assembler code for function my_strcpy2:
0x080483c1 <+0>:     push   %ebp
0x080483c2 <+1>:     mov    %esp,%ebp
0x080483c4 <+3>:     push   %ebx
0x080483c5 <+4>:     sub    $0x10,%esp
0x080483c8 <+7>:     mov    0xc(%ebp),%eax
0x080483cb <+10>:    mov    %eax,-0x8(%ebp)
0x080483ce <+13>:    mov    -0x8(%ebp),%eax
0x080483d1 <+16>:    movzbl (%eax),%ebx
0x080483d4 <+19>:    addl   $0x1,-0x8(%ebp)
0x080483d8 <+23>:    mov    0x80496bc,%eax
0x080483dd <+28>:    add    -0x8(%ebp),%eax
0x080483e0 <+31>:    mov    %bl,(%eax)
0x080483e2 <+33>:    test   %bl,%bl
0x080483e4 <+35>:    jne    0x80483ce
0x080483e6 <+37>:    mov    0x8(%ebp),%eax
0x080483e9 <+40>:    add    $0x10,%esp
0x080483ec <+43>:    pop    %ebx
0x080483ed <+44>:    pop    %ebp
0x080483ee <+45>:    ret
End of assembler dump.

现在效率仍然有区别，那么看来还是循环处出的问题。时间又晚了，下次再继续研究。