ポインタが指し示すデータを一時変数に受けていたせいで最適化が効かなくなった例 ref: http://qiita.com/abedominal/items/f188737285c8d4e71a03

After.cpp

push    ebp
mov ebp, esp
mov edx, DWORD PTR _b$[ebp]
mov ecx, DWORD PTR _r$[ebp]
movzx   eax, WORD PTR [edx+8]
mov WORD PTR [ecx], ax
movzx   eax, WORD PTR [edx+40]
mov WORD PTR [ecx+2], ax
movzx   eax, WORD PTR [edx+72]
mov WORD PTR [ecx+4], ax
pop ebp
ret 8

Before.asm

push    ebp
mov ebp, esp
push    esi
mov esi, DWORD PTR _r$[ebp]
push    edi
mov edi, DWORD PTR _b$[ebp]
movzx   ecx, WORD PTR [edi]
movzx   eax, WORD PTR [edi+8]
movzx   edx, WORD PTR [edi+2]
mov WORD PTR [esi], ax
movzx   eax, WORD PTR [ecx+edi+8]
mov WORD PTR [esi+2], ax
lea eax, DWORD PTR [ecx+edx]
movzx   eax, WORD PTR [eax+edi+8]
pop edi
mov WORD PTR [esi+4], ax
pop esi
pop ebp
ret 8

Before.cpp

class BufferAnalyzeDirect
{
public:
    static const size_t msgOfs_ = offsetof(Header, messageByte);
    static const size_t optOfs_ = offsetof(Header, optionByte);
    void operator()(uint8_t* b, Result & r)
    {
        uint16_t msgByte = *reinterpret_cast<uint16_t*>(&b[msgOfs_]);
        uint16_t optByte = *reinterpret_cast<uint16_t*>(&b[optOfs_]);
        r.requestStatus = *reinterpret_cast<uint16_t*>(&b[4 + 4]);
        r.optionCode = *reinterpret_cast<uint16_t*>(&b[4 + 4 + msgByte]);
        r.count = *reinterpret_cast<uint16_t*>(&b[4 + 4 + msgByte + optByte]);
    }
};

file2.txt

Before:      0.198654sec
After:       0.096776sec