我试图测量我的电脑执行不同操作所需的估计周期长度,所以我执行相同的一个100K次,并计算平均值.
我使用 loop unwinding更准确一点：我在每次迭代中执行10个基本操作,并将我的索引增加10个,从而减少循环操作.
我使用 loop unwinding更准确一点：我在每次迭代中执行10个基本操作,并将我的索引增加10个,从而减少循环操作.
没有一个这对我的问题非常重要：编译器有没有办法理解我几次做相同的操作,只执行一次？
这是我的循环：
for (i=0; i<iterations; i+=LOOP_FACTOR)
{
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
}
另外,我不知道是否重要 – 我正在使用Eclipse.我认为这可能是重要的,因为有不同的编译器.
解决方法
在没有优化的GCC中,它被编译为：
(gdb) disas main Dump of assembler code for function main: 0x00000000004004e4 <+0>: push rbp 0x00000000004004e5 <+1>: mov rbp,rsp 0x00000000004004e8 <+4>: mov DWORD PTR [rip+0x200482],0x0 # 0x600974 <i> 0x00000000004004f2 <+14>: jmp 0x400567 <main+131> 0x00000000004004f4 <+16>: mov DWORD PTR [rip+0x200472],0xffffffff # 0x600970 <result> 0x00000000004004fe <+26>: mov DWORD PTR [rip+0x200468],0xffffffff # 0x600970 <result> 0x0000000000400508 <+36>: mov DWORD PTR [rip+0x20045e],0xffffffff # 0x600970 <result> 0x0000000000400512 <+46>: mov DWORD PTR [rip+0x200454],0xffffffff # 0x600970 <result> 0x000000000040051c <+56>: mov DWORD PTR [rip+0x20044a],0xffffffff # 0x600970 <result> 0x0000000000400526 <+66>: mov DWORD PTR [rip+0x200440],0xffffffff # 0x600970 <result> 0x0000000000400530 <+76>: mov DWORD PTR [rip+0x200436],0xffffffff # 0x600970 <result> 0x000000000040053a <+86>: mov DWORD PTR [rip+0x20042c],0xffffffff # 0x600970 <result> 0x0000000000400544 <+96>: mov DWORD PTR [rip+0x200422],0xffffffff # 0x600970 <result> 0x000000000040054e <+106>: mov DWORD PTR [rip+0x200418],0xffffffff # 0x600970 <result> 0x0000000000400558 <+116>: mov eax,DWORD PTR [rip+0x200416] # 0x600974 <i> 0x000000000040055e <+122>: add eax,0x1 0x0000000000400561 <+125>: mov DWORD PTR [rip+0x20040d],eax # 0x600974 <i> 0x0000000000400567 <+131>: mov eax,DWORD PTR [rip+0x200407] # 0x600974 <i> 0x000000000040056d <+137>: cmp eax,0x3e7 0x0000000000400572 <+142>: jle 0x4004f4 <main+16> 0x0000000000400574 <+144>: mov eax,DWORD PTR [rip+0x2003f6] # 0x600970 <result> 0x000000000040057a <+150>: mov esi,eax 0x000000000040057c <+152>: mov edi,0x40067c 0x0000000000400581 <+157>: mov eax,0x0 0x0000000000400586 <+162>: call 0x4003e0 <printf@plt> 0x000000000040058b <+167>: pop rbp 0x000000000040058c <+168>: ret
但是如果你运行基本的优化(gcc -O),那么它被缩写为一个写：
Dump of assembler code for function main: 0x00000000004004e4 <+0>: sub rsp,0x8 0x00000000004004e8 <+4>: mov eax,0x3e8 0x00000000004004ed <+9>: sub eax,0x1 0x00000000004004f0 <+12>: jne 0x4004ed <main+9> 0x00000000004004f2 <+14>: mov DWORD PTR [rip+0x2003fc],0xffffffff # 0x6008f8 <result> 0x00000000004004fc <+24>: mov DWORD PTR [rip+0x2003f6],0x3e8 # 0x6008fc <i> 0x0000000000400506 <+34>: mov esi,0xffffffff 0x000000000040050b <+39>: mov edi,0x40060c 0x0000000000400510 <+44>: mov eax,0x0 0x0000000000400515 <+49>: call 0x4003e0 <printf@plt> 0x000000000040051a <+54>: add rsp,0x8 0x000000000040051e <+58>: ret
我的测试代码是：
#define TIMES 1000
int result,i;
int main() {
for (i=0; i<TIMES; i++)
{
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
result = -1;
}
printf("%d",result);
}
