用两个一维数组交叉想减为例:
long long *pDataStop1 = new long long[iSize1];
long long *pDataStop2 = new long long[iSize2];
1.申请内存指针用操作,数据的画,自己导入模拟数据
int m_iMaxDiff;
QMap<long long, int> m_mStatisTau;
int iSize1 = 7000;
int iSize2 = 7000;
long long *pDataStop1 = new long long[iSize1];
long long *pDataStop2 = new long long[iSize2];
long long iStop2;
long long iValue;
for(int j = 0; j < iSize1; j += 4) {
for(int k = 0; k < iSize2; k++) {
iStop2 = pDataStop2[k];
iValue = pDataStop1[j] - iStop2;
if(std::abs(iValue) <= m_iMaxDiff) {
m_mStatisTau[iValue] += 1;
}
iValue = pDataStop1[j + 1] - iStop2;
if(std::abs(iValue) <= m_iMaxDiff) {
m_mStatisTau[iValue] += 1;
}
iValue = pDataStop1[j + 2] - iStop2;
if(std::abs(iValue) <= m_iMaxDiff) {
m_mStatisTau[iValue] += 1;
}
iValue = pDataStop1[j + 3] - iStop2;
if(std::abs(iValue) <= m_iMaxDiff) {
m_mStatisTau[iValue] += 1;
}
}
}
这个执行完,大概需要310ms
2.std::transform,内部用的多线程,c++17才能使用,使用std::execution::par_unseq项是最快的
for(int i = 0; i < iSize; i++) {
int iSize1 = vDataStop2[i].size();
std::vector<long long> vOut(iSize1);
for(int j = 0; j < vDataStop1[i].size(); j++) {
QVector<long long> vStop1(iSize1);
std::fill(vStop1.begin(), vStop1.end(), vDataStop1[i][j]);
std::transform(std::execution::par_unseq, vDataStop2[i].begin(),
vDataStop2[i].end(), vStop1.begin(), vOut.begin(), CalculatePoor);
}
}
这个使用vector大概是490ms,使用指针大概是410ms,可以参考我上一个博客,写的比较详细
3.AVX256指令集
__m256i m1, m2;
long long re[4];
for(int j = 0; j < iSize1; j += 4) {
m1 = _mm256_set_epi64x(pDataStop1[j],pDataStop1[j+1],pDataStop1[j+2], pDataStop1[j+3]);
for(int k = 0; k < iSize2; k++) {
m2 = _mm256_set_epi64x(pDataStop2[k], pDataStop2[k], pDataStop2[k], pDataStop2[k]);
__m256i l1 = _mm256_sub_epi64(m1, m2);
re[3] = l1.m256i_i64[0];
re[2] = l1.m256i_i64[1];
re[1] = l1.m256i_i64[2];
re[0] = l1.m256i_i64[3];
if(std::abs(re[3]) <= m_iMaxDiff) {
m_mStatisTau[re[3]] += 1;
}
if(std::abs(re[2]) <= m_iMaxDiff) {
m_mStatisTau[re[2]] += 1;
}
if(std::abs(re[1]) <= m_iMaxDiff) {
m_mStatisTau[re[1]] += 1;
}
if(std::abs(re[0]) <= m_iMaxDiff) {
m_mStatisTau[re[0]] += 1;
}
}
}
这个执行时间大概320ms