C++ 连接字符串的最快方法(最佳方法)

标签 c++ string performance concatenation

我正在制作一个必须提供非常大的报告的应用程序,我遇到了过程中最慢的部分是当它必须创建最终的 JSON 字符串时,所以我尝试了不同的方法来加快速度,看看哪个是最快的

我从对字符串的 +=+ 操作开始。 然后我用 += 改变了所有的 + 并且它改进了一点。 然后我尝试了一些更大胆的方法,它使用 char * buffer 并自己管理连接,甚至制作了我自己的 INTFLOAT 到字符串功能。 它缩短了 30% 的时间,但这个过程仍然需要大约 15 段的大报告,我想将时间减少到最少,所以我会在这里粘贴我的代码,看看是否有任何关于如何改进性能的建议.

char * itoa2(UINT num, char * str)
{
    if (num == 0)
    {
        *str++ = '0';
        return str;
    }

    // Process individual digits
    UCHAR nsiz = (UCHAR)log10(num) + 1;
    str += nsiz;
    while (num != 0)
    {
        *str-- = (num % 10) + '0';
        num = num/10;
    }
    return str + nsiz + 1;
}

char * ftoa2(float num, char * str) {
    int intg = (UINT)num;
    int dec = (UINT)((num - intg) * 1000);



    // Process individual digits
    UCHAR nsiz = (UCHAR)log10(intg) + 1;
    str += nsiz - 1;
    if (dec > 0)
    {
        nsiz += (UCHAR)log10(dec) + 2;
        str += nsiz - 1;
        while (dec != 0) {
            *str-- = (dec % 10) + '0';
            dec = dec / 10;
        }
        *str-- = '.';
    }

    if (intg == 0)
    {
        *str-- = '0';
    } else {
        while (intg != 0) {
            *str-- = (intg % 10) + '0';
            intg = intg / 10;
        }
    }


    return str + nsiz + 1;
}

void Formatter::jsonFormat2() {

    char * result = new char[1024 * 1024 * 1024]; //TODO: Use small buffers instead of this

    char * pos;

    string tmp;

    pos = result;

    (*pos++) = '[';

    const char * key;
    int keyPos;
    const Reporter::Entry * entry;
    set<Field *, Field::cmpOrder> * grouped = params->getGrouped();

    bool byDay, byMonth, byYear;
    time_t ts = 0;
    struct tm * lt;

    char type;

    FieldData::fieldVariant fkey(0, nullptr);
    FieldData::fieldVariant * fvar;

    byDay = params->getByDay();
    byMonth = params->getByMonth();
    byYear = params->getByYear();


    for (auto& it : *report) {
        key = it.first;
        entry = it.second;
        keyPos = 0;

        (*pos++) = '{';

        if (byDay || byMonth || byYear) {
            ts = *(time_t *)key;
            keyPos += sizeof(time_t);
            lt = gmtime(&ts);
        }

        if (byDay) {
            memcpy(pos, "\"day\":\"", 7);
            pos+=7;
            pos = itoa2(lt->tm_year + 1900, pos);
            (*pos++) = '-';
            if (lt->tm_mon+1 < 10) (*pos++) = '0';
            pos = itoa2(lt->tm_mon+1, pos);
            (*pos++) = '-';
            if (lt->tm_mday < 10) (*pos++) = '0';
            pos = itoa2(lt->tm_mday, pos);
            (*pos++) = '\"'; (*pos++) = ',';

            memcpy(pos, "\"wday\":\"", 8);
            pos+=8;

            memcpy(pos, &weekDayC[lt->tm_wday * 3], 3);
            pos+=3;

            (*pos++) = '\"'; (*pos++) = ',';
        }
        if (byMonth) {
            memcpy(pos, "\"month\":\"", 9);
            pos+=9;
            pos = itoa2(lt->tm_year + 1900, pos);
            (*pos++) = '-';
            if (lt->tm_mon+1 < 10) (*pos++) = '0';
            pos = itoa2(lt->tm_mon, pos);
            (*pos++) = '\"'; (*pos++) = ',';
        }
        if (byYear) {
            memcpy(pos, "\"year\":\"", 8);
            pos+=9;
            pos = itoa2(lt->tm_year + 1900, pos);
            (*pos++) = '\"'; (*pos++) = ',';
        }


        for (auto& field : *grouped) {
            (*pos++) = '\"';

            tmp = field->getName();
            memcpy(pos, tmp.c_str(), tmp.length());
            pos+=tmp.length();
            memcpy(pos, "\":\"", 3);
            pos+=3;

            type = field->getType();
            if (type == INT8) {
                pos = itoa2(*(key + keyPos), pos);
            } else if (type == INT16) {
                pos = itoa2(*(USHORT *)(key + keyPos), pos);
            } else if (type == INT32) {
                pos = itoa2(*(UINT *)(key + keyPos), pos);
            } else if (type == INT64) {
                pos = itoa2(*(ULONG *)(key + keyPos), pos);
            } else if (type == INT128) {
                pos = itoa2(*(UXLONG *)(key + keyPos), pos);
            } else if (type == CHAR2) {
                memcpy(pos, key + keyPos, 2);
                pos+=2;
            }

            (*pos++) = '\"'; (*pos++) = ',';


            for (auto& fieldData : *field->getFieldData()) {
                fkey.type = field->getType();

                if (fkey.type == CHAR2) {
                    fkey.data = new string((char *) (key + keyPos), 2);
                } else
                    fkey.data = (void *)key + keyPos;

                fvar = fieldData->find(&fkey);

                (*pos++) = '\"';

                tmp = fieldData->colName;
                memcpy(pos, tmp.c_str(), tmp.length());
                pos+=tmp.length();
                memcpy(pos, "\":\"", 3);
                pos+=3;
                if (fvar != nullptr) tmp = toString(fvar); else tmp = "null";
                memcpy(pos, tmp.c_str(), tmp.length());
                pos+=tmp.length();
                memcpy(pos, "\",", 2);
                pos+=2;

            }
            keyPos += field->getBytes();
        }

        memcpy(pos, "\"imps\":\"", 8);
        pos+=8;
        pos = itoa2(entry->imps, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"impsn\":\"", 9);
        pos+=9;
        pos = itoa2(entry->impsn, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"impsu\":\"", 9);
        pos+=9;
        pos = itoa2(entry->impsu, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"clicks\":\"", 10);
        pos+=10;
        pos = itoa2(entry->clicks, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"clicksu\":\"", 11);
        pos+=11;
        pos = itoa2(entry->clicksu, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"revenue\":\"", 11);
        pos+=11;
        pos = ftoa2(entry->revenue, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"revenue_2\":\"", 13);
        pos+=13;
        pos = ftoa2(entry->revenue_2, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"revenue_3\":\"", 13);
        pos+=13;
        pos = ftoa2(entry->revenue_3, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"convs\":\"", 9);
        pos+=9;
        pos = itoa2(entry->convs, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"convs_2\":\"", 11);
        pos+=11;
        pos = itoa2(entry->convs_2, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        memcpy(pos, "\"convs_3\":\"", 11);
        pos+=11;
        pos = itoa2(entry->convs_3, pos);
        (*pos++) = '\"'; (*pos++) = ',';

        (*pos++) = '}'; (*pos++) = ',';
        pos+=3;
    }
    pos--;

    (*pos++) = ']';
    (*pos++) = '\0';

    if (params->getVerbose()) cout << string(result, pos - result);
}

最佳答案

同样的方法,Formatter::jsonFormat2() , 你有

char * result = new char[1024 * 1024 * 1024];

在顶部,

if (params->getVerbose()) cout << string(result, pos - result);

在底部。

这意味着,您不仅每次调用此方法都会有 1GB 的内存泄漏,如果getVerbose(),您还白做了很多工作。恰好返回 false(您可以在一开始就检查一下)。

此外,如果您打算将所有内容写入std::cout (或与此相关的任何流),考虑使用流插入语法(<< 运算符)直接编写各种片段。

关于C++ 连接字符串的最快方法(最佳方法),我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/32705130/

相关文章:

c++ - 从文件中读取整数,中间有一个字符串

Java-Chat-App::分析用户输入

c# - 由同一个 Linq 和 Lambda 表达式生成的两个不同的 SQL 语句

C++:跨迭代重用的浮点值

c++ - 如何使用 glReadPixels 读取 float 据

string - 如何以尽可能小的可打印方式表示 32 字节的二进制数据

c++ - curl 字符串追加问题,curl_formadd C++ 读取带有额外字符的变量但直接工作正常 CURLFORM_COPYNAME, "key"[?]

ios - iOS 开发框架基准

performance - 访问一个字节比访问一个比特更快吗?为什么?

C++ 减少冗余