2.2.1 需要用汇编来排错的常见情况

 int hgt=4;
int wid=7;
for (i=0; i<hgt; i++)
for (j=0; j<wid; j++)
A[i*wid+j] = exp(-(i*i+j*j));

int hgt=4;
int wid=7;
24:       for (i=0; i<hgt; i++)
0040107A   xor         ebp,ebp
0040107C   lea         edi,[esp+10h]
25:        for (j=0; j<wid; j++)
26:            A[i*wid+j] = exp(-(i*i+j*j));
00401080   mov         ebx,ebp
00401082   xor         esi,esi
// The result of i*i is saved in ebx
00401084   imul        ebx,ebp
00401087   mov         eax,esi
// Only one imul occurs in every inner loop (j*j)
00401089   imul        eax,esi     
// Use the saved i*i in ebx directly. !!Optimized!!
0040108C   add         eax,ebx  
0040108E   neg         eax   
00401090   push        eax
00401091   call        @ILT+0(exp) (00401005)
00401096   add         esp,4   
// Save the result back to A[]. The addr of current offset in A[] is saved in edi
00401099   mov         dword ptr [edi],eax 
0040109B   inc         esi
// Simply add edi by 4. Does not calculate with i*wid. Imul is never used. !!Optimized!!
0040109C   add         edi,4
0040109F   cmp         esi,7
004010A2   jl          main+17h (00401087)
004010A4   inc         ebp
004010A5   cmp         ebp,4
004010A8   jl          main+10h (00401080)

// The following code crashes/abnormal in release build 
when "whole program optimizations /GL"
// is set. The bug is fixed in VS2005
#include <string>
#pragma warning( push )
#pragma warning( disable : 4702 ) // unreachable code in <vector>
#include <vector>
#pragma warning( pop )
#include <algorithm>
#include <iostream>
//vcsig
// T = float, U = std::cstring
template <typename T, typename U>  T func_template( const U & u )
{
std::cout<<u<<std::endl;
const char* str=u.c_str();
printf(str);
return static_cast<T>(0);
}
void crash_in_release()
{
std::vector<std::string>  vStr;
vStr.push_back("1.0");
vStr.push_back("0.0");
vStr.push_back("4.4");
std::vector<float>  vDest( vStr.size(), 0.0 );
std::vector<std::string>::iterator _First=vStr.begin();
std::vector<std::string>::iterator _Last=vStr.end();
std::vector<float>::iterator _Dest=vDest.begin();
std::transform( _First,_Last,_Dest, func_template<float,std::string> );  
_First=vStr.begin();
_Last=vStr.end();
_Dest=vDest.begin();

for (; _First != _Last; ++_First, ++_Dest)
*_Dest =  func_template<float,std::string>(*_First);  
}

int main(int, char*)
{

getchar();
crash_in_release();
return 0;
}

1. 取消precompiled header。
2. 编译选项是: /O2 /GL /D "WIN32" /D "NDEBUG" /D "_CONSOLE" 
/D "_MBCS" /FD /EHsc /ML /GS /Fo"Release/" /Fd"Release/vc70.pdb" 
/W4 /nologo /c /Wp64 /Zi /TP。

*_Dest = _Func(*_First);

EAX = 0012FEA8 EBX = 0037138C ECX = 003712BC EDX = 00371338 ESI 
= 00371338 EDI = 003712B0 EIP = 00402228 ESP = 0012FE70 EBP = 0012FEA8 EFL = 00000297 
388: *_Dest = _Func(*_First);
00402228 push esi
00402229 call dword ptr [esp+28h] 
0040222D fstp dword ptr [edi] 

EAX = 003712B0 EBX = 00371338 ECX = 003712BC EDX = 00000000 
ESI = 00371338 EDI = 0037138C EIP = 00401242 ESP = 0012FE98 EBP = 003712B0 EFL = 00000297
37: *_Dest = func_template<float,std::string>(*_First);
00401240 mov ebx,esi 
00401242 call func_template <float,std::basic_string<char,
std::char_traits<char>,std::allocator<char> > > (4021A0h) 
00401247 fstp dword ptr [ebp] 

004021A0 push esi
004021A1 push edi 
16:  std::cout<<u<<std::endl;
004021A2 push ebx
004021A3 push offset std::cout (414170h) 
004021A8 call std::operator<<<char,std::char_traits<char>,std::allocator<char> > (402280h)

PRB: Application Center Cluster Members Are Automatically Synchronized After Rebooting
http://support.microsoft.com/kb/282278/en-us