switch文のケースに使われているリテラル値で最適化はしてくれるのか？ ref: http://qiita.com/abedominal/items/5f4eb9750737aff76efe

file0.cpp

//maskの右辺値はコンパイル時に決められるから計算コスト不要であることを期待
class Select0
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        uint32_t mask = 0;
        switch (bitPos)
        {
        case 1:  mask = 1<<0; break;
        case 2:  mask = 1<<1; break;
        case 3:  mask = 1<<2; break;
        case 4:  mask = 1<<3; break;
        case 5:  mask = 1<<4; break;
        case 6:  mask = 1<<5; break;
        case 7:  mask = 1<<6; break;
        case 8:  mask = 1<<7; break;
        default: return false;
        }
        return (dword & mask) != 0;
    }
};
//Select0の派生。Select0が最適化されていたらこれと同じ速度になるはず。
class Select00
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        switch (bitPos)
        {
        case 1:  return (dword & (1 << 0)) != 0;
        case 2:  return (dword & (1 << 1)) != 0;
        case 3:  return (dword & (1 << 2)) != 0;
        case 4:  return (dword & (1 << 3)) != 0;
        case 5:  return (dword & (1 << 4)) != 0;
        case 6:  return (dword & (1 << 5)) != 0;
        case 7:  return (dword & (1 << 6)) != 0;
        case 8:  return (dword & (1 << 7)) != 0;
        default: return false;
        }
    }
};
//Select0の派生。Select0が最適化されていたらこれと同じ速度になるはず。
class Select01
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        switch (bitPos)
        {
        case 1:  return (dword & (1  )) != 0;
        case 2:  return (dword & (2  )) != 0;
        case 3:  return (dword & (4  )) != 0;
        case 4:  return (dword & (8  )) != 0;
        case 5:  return (dword & (16 )) != 0;
        case 6:  return (dword & (32 )) != 0;
        case 7:  return (dword & (64 )) != 0;
        case 8:  return (dword & (128)) != 0;
        default: return false;
        }
    }
};
//最適化が賢ければSelect0はこう書いても同じになるはず。
class Select1
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        switch (bitPos)
        {
        case 1: case 2: case 3: case 4: case 5: case 6: case 7: case 8:
        {
            const uint32_t mask = 1 << (bitPos - 1);
            return (dword & mask) != 0;
        }
        default: return false;
        }
    }
};
//if文で普通に処理するパターン
class Select2
{

public:
    bool operator()(int bitPos, uint32_t dword)
    {
        if (bitPos < 1 || 8 < bitPos)
        {
            return false;
        }
        return ((dword >> (bitPos - 1)) & 1) == 1;
    }

};
//if文で範囲外をはじいてからテーブル処理するパターン。テーブル内のシフト演算はコンパイル時に確定するはず。
class Select30
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        if (bitPos < 1 || 8 < bitPos)
        {
            return false;
        }
        static const uint32_t mask[] = { 0, 1, 1 << 1, 1 << 2, 1 << 3, 1 << 4, 1 << 5, 1 << 6, 1 << 7};
        return (dword & mask[bitPos]) != 0;
    }

};
//Select30の亜種。範囲外はすべて0番目の配列で処理
class Select31
{
public:
    bool operator()(int bitPos, uint32_t dword)
    {
        static const uint32_t mask[] = { 0, 1, 1 << 1, 1 << 2, 1 << 3, 1 << 4, 1 << 5, 1 << 6, 1 << 7 };
        return (dword & mask[8<bitPos ? 0 : bitPos]) != 0;
    }

};

file1.cpp

#include <iostream>
#include "Timer.hpp"

template<typename T>
class Measure
{
    static const uint64_t rep = 100000000;
    int count_;
public:
    Measure() :count_()
    {}
    int count()
    {
        return count_;
    }
    double execute(uint32_t dword)
    {
        Timer<TimerType::HighAccuracy> timer;
        for (uint64_t i = 0; i < rep; ++i)
        {
            if (T()(i % 8 + 1, dword))
            {
                ++count_;
            }
        }
        return timer.elapsed();
    }

};
#include <iomanip>
#define measure(n,dw) \
{ \
    Measure<Select##n> a; \
    volatile double r = a.execute(dw); \
    int c = a.count(); \
    cout << #n":\t" << /*fixed << setprecision(6) << */ r << ", " << c << endl; \
}
//setprecisionをつけるとなぜかSelect2が遅くなる
int main(int argv, char* argc[])
{
    static const int testNum = 2;
    //最適化効いたらいやなので標準入力から32という数値を放り込む
    uint32_t in;
    cin >> in;

    volatile uint32_t dw = in;

    measure(0, dw);
    measure(00, dw);
    measure(01, dw);
    measure(1, dw);
    measure(2, dw);
    measure(30, dw);
    measure(31, dw);

    return 0;
}

file1.txt

0:      0.276666, 12500000
00:     0.216566, 12500000
01:     0.213629, 12500000
1:      0.143589, 12500000
2:      0.113139, 12500000
30:     0.115278, 12500000
31:     0.144108, 12500000
続行するには何かキーを押してください . . .

file2.txt

0:      0.276666, 12500000
00:     0.216566, 12500000
01:     0.213629, 12500000
1:      0.143589, 12500000
2:      0.113139, 12500000
30:     0.115278, 12500000
31:     0.144108, 12500000
続行するには何かキーを押してください . . .

file3.txt

$LL4@execute:
    mov eax, ecx
    and eax, 7
    cmp eax, 7
    ja  SHORT $LN3@execute
    jmp DWORD PTR $LN45@execute[eax*4]
$LN23@execute:
    mov al, bl
    and eax, -255               ; ffffff01H
    jmp SHORT $LN24@execute
$LN22@execute:
    mov eax, ebx
    shr eax, 1
    jmp SHORT $LN40@execute
$LN21@execute:
    mov eax, ebx
    shr eax, 2
    jmp SHORT $LN40@execute
$LN20@execute:
    mov eax, ebx
    shr eax, 3
    jmp SHORT $LN40@execute
$LN19@execute:
    mov eax, ebx
    shr eax, 4
    jmp SHORT $LN40@execute
$LN18@execute:
    mov eax, ebx
    shr eax, 5
    jmp SHORT $LN40@execute
$LN17@execute:
    mov eax, ebx
    shr eax, 6
    jmp SHORT $LN40@execute
$LN16@execute:
    mov eax, ebx
    shr eax, 7
$LN40@execute:
    and al, 1
$LN24@execute:
    test    al, al
    je  SHORT $LN3@execute; Line 23
    inc DWORD PTR [edi]

file4.txt

_TEXT   SEGMENT
_dword$ = 8                     ; size = 4
$T1 = 12                        ; size = 4
??RSelect1@@QAE_NHI@Z PROC              ; Select1::operator(), COMDAT
    push    ebp
    mov ebp, esp
    cmp DWORD PTR $T1[ebp], 8
    jle SHORT $LN2@operator
    xor al, al
    pop ebp
    ret 8
$LN2@operator:
    dec ecx
    mov eax, 1
    shl eax, cl
    and eax, DWORD PTR _dword$[ebp]
    neg eax
    sbb eax, eax
    neg eax
    pop ebp
    ret 8
??RSelect1@@QAE_NHI@Z ENDP              ; Select1::operator()
_TEXT   ENDS

file5.txt

    cmp ecx, 7
    ja  SHORT $LN3@execute
    mov eax, ebx
    shr eax, cl
    and al, 1