用sse4.1指令集加速ceil( )函数

C/C++标准库的ceil()函数，能否加速一下？

在OpenCV中，看到了cvCeil()函数，它是用sse2加速的。

其实现在我用的PC，avx2都有支持了，sse、avx系列是递增式支持的，用sse4.1来优化一下，应该还是能普遍使用的。

本质上大多数人写的大多数程序，都是调用现成的包/API，无非是调包侠也有鄙视链而已。

今天做一回SSE4.1的调包侠，让原本100ms水平的计算过程，优化到25ms，比clang-8.0（实际上是它的标准库）里的ceil()再快5ms：

#include <iostream>
#include <cmath>
#include <sstream>
#include <chrono>

//sse2
#if defined _MSC_VER && defined _M_X64 || (defined __GNUC__ && defined __SSE2__&& !defined __APPLE__)
#include <emmintrin.h>
#endif

//sse 4.1
#include <smmintrin.h>


//sse2 optimized
inline int cvCeil(double value)
{
#if defined _MSC_VER && defined _M_X64 || (defined __GNUC__ && defined __SSE2__&& !defined __APPLE__)
    __m128d t = _mm_set_sd( value );
    int i = _mm_cvtsd_si32(t);
    return i + _mm_movemask_pd(_mm_cmplt_sd(_mm_cvtsi32_sd(t,i), t));
#elif defined __GNUC__
    int i = (int)value;
    return i + (i < value);
#else
    int i = cvRound(value);
    float diff = (float)(i - value);
    return i + (diff < 0);
#endif
}


//sse4 optimized
inline int myCeil(double value)
{
#if defined _MSC_VER && defined _M_X64 || (defined __GNUC__ && defined __SSE2__&& !defined __APPLE__)

    __m128d val = _mm_set_sd(value);
    __m128d dst;
    _mm_round_sd(dst, val, _MM_FROUND_CEIL);
    return _mm_cvtsd_si32(dst);


    //__m128d t = _mm_set_sd( value );
    //int i = _mm_cvtsd_si32(t);
    //return i + _mm_movemask_pd(_mm_cmplt_sd(_mm_cvtsi32_sd(t,i), t));
#elif defined __GNUC__
    int i = (int)value;
    return i + (i < value);
#else
    int i = cvRound(value);
    float diff = (float)(i - value);
    return i + (diff < 0);
#endif
}



template<typename T, typename P>
std::string toString(std::chrono::duration<T,P> dt)
{
    std::ostringstream str;
    using namespace std::chrono;
    str << duration_cast<microseconds>(dt).count()*1e-3 << " ms";
    return str.str();
}

int main () {
    volatile double x = 34.234;
    volatile double y1, y2, y3;
    const int MAX_ITER=100000000;
    const auto t0 = std::chrono::steady_clock::now();

    for(int i=0; i<MAX_ITER; i++) {
        y1 = std::ceil(x);
    }
    const auto t1 = std::chrono::steady_clock::now();

    for(int i=0; i<MAX_ITER; i++) {
        y2 = cvCeil(x);
    }
    const auto t2 = std::chrono::steady_clock::now();

    for(int i=0; i<MAX_ITER; i++) {
        y3 = myCeil(x);
    }
    const auto t3 = std::chrono::steady_clock::now();



    std::cout << "std::ceil: " << toString(t1-t0) << "\n";
    std::cout << "cvCeil   : " << toString(t2-t1) << "\n";
    std::cout << "myCeil   : " << toString(t3-t2) << "\n";
    std::cout << "y1=" << y1 << ", y2=" << y2 << ", y3=" << y3 << std::endl;

    return 0;
}

编译指令：

clang++-8 main8.cpp -O3 -o a8 -std=c++11 -msse4

运行输出：

std::ceil: 30.347 ms
cvCeil   : 106.99 ms
myCeil   : 25.318 ms
y1=35, y2=35, y3=35

refs:
is cvCeil() faster than standard library?
MSDN - _mm_round_sd

用sse4.1指令集加速ceil( )函数

猜你喜欢